diff options
Diffstat (limited to 'sys/nfsclient')
| -rw-r--r-- | sys/nfsclient/nfs.h | 296 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_bio.c | 1801 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_kdtrace.c | 542 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_krpc.c | 881 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_nfsiod.c | 338 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_node.c | 280 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_subs.c | 1147 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_vfsops.c | 1579 | ||||
| -rw-r--r-- | sys/nfsclient/nfs_vnops.c | 3523 | ||||
| -rw-r--r-- | sys/nfsclient/nfsargs.h | 102 | ||||
| -rw-r--r-- | sys/nfsclient/nfsm_subs.h | 206 | ||||
| -rw-r--r-- | sys/nfsclient/nfsmount.h | 132 | ||||
| -rw-r--r-- | sys/nfsclient/nfsnode.h | 212 | ||||
| -rw-r--r-- | sys/nfsclient/nfsstats.h | 69 | ||||
| -rw-r--r-- | sys/nfsclient/nlminfo.h | 42 |
15 files changed, 11150 insertions, 0 deletions
diff --git a/sys/nfsclient/nfs.h b/sys/nfsclient/nfs.h new file mode 100644 index 0000000..22a291a --- /dev/null +++ b/sys/nfsclient/nfs.h @@ -0,0 +1,296 @@ +/*- + * Copyright (c) 1989, 1993, 1995 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs.h 8.4 (Berkeley) 5/1/95 + * $FreeBSD$ + */ + +#ifndef _NFSCLIENT_NFS_H_ +#define _NFSCLIENT_NFS_H_ + +#ifdef _KERNEL +#include "opt_nfs.h" +#endif + +#include <nfsclient/nfsargs.h> + +/* + * Tunable constants for nfs + */ + +#define NFS_TICKINTVL 10 /* Desired time for a tick (msec) */ +#define NFS_HZ (hz / nfs_ticks) /* Ticks/sec */ +#define NFS_TIMEO (1 * NFS_HZ) /* Default timeout = 1 second */ +#define NFS_MINTIMEO (1 * NFS_HZ) /* Min timeout to use */ +#define NFS_MAXTIMEO (60 * NFS_HZ) /* Max timeout to backoff to */ +#define NFS_MINIDEMTIMEO (5 * NFS_HZ) /* Min timeout for non-idempotent ops*/ +#define NFS_MAXREXMIT 100 /* Stop counting after this many */ +#define NFS_RETRANS 10 /* Num of retrans for UDP soft mounts */ +#define NFS_RETRANS_TCP 2 /* Num of retrans for TCP soft mounts */ +#define NFS_MAXGRPS 16 /* Max. size of groups list */ +#ifndef NFS_MINATTRTIMO +#define NFS_MINATTRTIMO 3 /* VREG attrib cache timeout in sec */ +#endif +#ifndef NFS_MAXATTRTIMO +#define NFS_MAXATTRTIMO 60 +#endif +#ifndef NFS_MINDIRATTRTIMO +#define NFS_MINDIRATTRTIMO 3 /* VDIR attrib cache timeout in sec */ +#endif +#ifndef NFS_MAXDIRATTRTIMO +#define NFS_MAXDIRATTRTIMO 60 +#endif +#ifndef NFS_ACCESSCACHESIZE +#define NFS_ACCESSCACHESIZE 8 /* Per-node access cache entries */ +#endif +#define NFS_WSIZE 8192 /* Def. write data size <= 8192 */ +#define NFS_RSIZE 8192 /* Def. read data size <= 8192 */ +#define NFS_READDIRSIZE 8192 /* Def. readdir size */ +#define NFS_DEFRAHEAD 1 /* Def. read ahead # blocks */ +#define NFS_MAXRAHEAD 4 /* Max. read ahead # blocks */ +#define NFS_MAXASYNCDAEMON 64 /* Max. number async_daemons runnable */ +#define NFS_DIRBLKSIZ 4096 /* Must be a multiple of DIRBLKSIZ */ +#ifdef _KERNEL +#define DIRBLKSIZ 512 /* XXX we used to use ufs's DIRBLKSIZ */ +#endif +#define NFS_MAXDEADTHRESH 9 /* How long till we say 'server not responding' */ + +/* + * Oddballs + */ +#define NFS_CMPFH(n, f, s) \ + ((n)->n_fhsize == (s) && !bcmp((caddr_t)(n)->n_fhp, (caddr_t)(f), (s))) +#define NFS_ISV3(v) (VFSTONFS((v)->v_mount)->nm_flag & NFSMNT_NFSV3) +#define NFS_ISV4(v) (VFSTONFS((v)->v_mount)->nm_flag & NFSMNT_NFSV4) + +#define NFSSTA_HASWRITEVERF 0x00040000 /* Has write verifier for V3 */ +#define NFSSTA_GOTFSINFO 0x00100000 /* Got the V3 fsinfo */ +#define NFSSTA_SNDLOCK 0x01000000 /* Send socket lock */ +#define NFSSTA_WANTSND 0x02000000 /* Want above */ +#define NFSSTA_TIMEO 0x10000000 /* Experiencing a timeout */ +#define NFSSTA_LOCKTIMEO 0x20000000 /* Experiencing a lockd timeout */ + + +/* + * XXX to allow amd to include nfs.h without nfsproto.h + */ +#ifdef NFS_NPROCS +#include <nfsclient/nfsstats.h> +#endif + +/* + * vfs.oldnfs sysctl(3) identifiers + */ +#define NFS_NFSSTATS 1 /* struct: struct nfsstats */ + +#ifdef _KERNEL + +#ifdef MALLOC_DECLARE +MALLOC_DECLARE(M_NFSREQ); +MALLOC_DECLARE(M_NFSDIROFF); +MALLOC_DECLARE(M_NFSBIGFH); +MALLOC_DECLARE(M_NFSHASH); +MALLOC_DECLARE(M_NFSDIRECTIO); +#endif + +extern struct uma_zone *nfsmount_zone; + +extern struct nfsstats nfsstats; +extern struct mtx nfs_iod_mtx; +extern struct task nfs_nfsiodnew_task; + +extern int nfs_numasync; +extern unsigned int nfs_iodmax; +extern int nfs_pbuf_freecnt; +extern int nfs_ticks; + +/* Data constants in XDR form */ +extern u_int32_t nfs_true, nfs_false, nfs_xdrneg1; +extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers; +extern u_int32_t rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr; + +extern int nfsv3_procid[NFS_NPROCS]; + +/* + * Socket errors ignored for connectionless sockets?? + * For now, ignore them all + */ +#define NFSIGNORE_SOERROR(s, e) \ + ((e) != EINTR && (e) != EIO && \ + (e) != ERESTART && (e) != EWOULDBLOCK && \ + ((s) & PR_CONNREQUIRED) == 0) + +struct nfsmount; + +struct buf; +struct socket; +struct uio; +struct vattr; + +/* + * Pointers to ops that differ from v3 to v4 + */ +struct nfs_rpcops { + int (*nr_readrpc)(struct vnode *vp, struct uio *uiop, + struct ucred *cred); + int (*nr_writerpc)(struct vnode *vp, struct uio *uiop, + struct ucred *cred, int *iomode, int *must_commit); + int (*nr_writebp)(struct buf *bp, int force, struct thread *td); + int (*nr_readlinkrpc)(struct vnode *vp, struct uio *uiop, + struct ucred *cred); + void (*nr_invaldir)(struct vnode *vp); + int (*nr_commit)(struct vnode *vp, u_quad_t offset, int cnt, + struct ucred *cred, struct thread *td); +}; + +/* + * Defines for WebNFS + */ + +#define WEBNFS_ESC_CHAR '%' +#define WEBNFS_SPECCHAR_START 0x80 + +#define WEBNFS_NATIVE_CHAR 0x80 +/* + * .. + * Possibly more here in the future. + */ + +/* + * Macro for converting escape characters in WebNFS pathnames. + * Should really be in libkern. + */ + +#define HEXTOC(c) \ + ((c) >= 'a' ? ((c) - ('a' - 10)) : \ + ((c) >= 'A' ? ((c) - ('A' - 10)) : ((c) - '0'))) +#define HEXSTRTOI(p) \ + ((HEXTOC(p[0]) << 4) + HEXTOC(p[1])) + +/* nfs_sigintr() helper, when 'rep' has all we need */ +#define NFS_SIGREP(rep) nfs_sigintr((rep)->r_nmp, (rep), (rep)->r_td) + +#ifdef NFS_DEBUG + +extern int nfs_debug; +#define NFS_DEBUG_ASYNCIO 1 /* asynchronous i/o */ +#define NFS_DEBUG_WG 2 /* server write gathering */ +#define NFS_DEBUG_RC 4 /* server request caching */ + +#define NFS_DPF(cat, args) \ + do { \ + if (nfs_debug & NFS_DEBUG_##cat) printf args; \ + } while (0) + +#else + +#define NFS_DPF(cat, args) + +#endif + +/* + * On fast networks, the estimator will try to reduce the + * timeout lower than the latency of the server's disks, + * which results in too many timeouts, so cap the lower + * bound. + */ +#define NFS_MINRTO (NFS_HZ >> 2) + +/* + * Keep the RTO from increasing to unreasonably large values + * when a server is not responding. + */ +#define NFS_MAXRTO (20 * NFS_HZ) + +enum nfs_rto_timer_t { + NFS_DEFAULT_TIMER, + NFS_GETATTR_TIMER, + NFS_LOOKUP_TIMER, + NFS_READ_TIMER, + NFS_WRITE_TIMER, +}; +#define NFS_MAX_TIMER (NFS_WRITE_TIMER) + +#define NFS_INITRTT (NFS_HZ << 3) + +vfs_init_t nfs_init; +vfs_uninit_t nfs_uninit; +int nfs_mountroot(struct mount *mp); + +void nfs_purgecache(struct vnode *); +int nfs_vinvalbuf(struct vnode *, int, struct thread *, int); +int nfs_readrpc(struct vnode *, struct uio *, struct ucred *); +int nfs_writerpc(struct vnode *, struct uio *, struct ucred *, int *, + int *); +int nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, + struct ucred *cred, struct thread *td); +int nfs_readdirrpc(struct vnode *, struct uio *, struct ucred *); +void nfs_nfsiodnew(void); +void nfs_nfsiodnew_tq(__unused void *, int); +int nfs_asyncio(struct nfsmount *, struct buf *, struct ucred *, struct thread *); +int nfs_doio(struct vnode *, struct buf *, struct ucred *, struct thread *); +void nfs_doio_directwrite (struct buf *); +int nfs_readlinkrpc(struct vnode *, struct uio *, struct ucred *); +int nfs_sigintr(struct nfsmount *, struct thread *); +int nfs_readdirplusrpc(struct vnode *, struct uio *, struct ucred *); +int nfs_request(struct vnode *, struct mbuf *, int, struct thread *, + struct ucred *, struct mbuf **, struct mbuf **, caddr_t *); +int nfs_loadattrcache(struct vnode **, struct mbuf **, caddr_t *, + struct vattr *, int); +int nfsm_mbuftouio(struct mbuf **, struct uio *, int, caddr_t *); +void nfs_nhinit(void); +void nfs_nhuninit(void); +int nfs_nmcancelreqs(struct nfsmount *); +void nfs_timer(void*); + +int nfs_connect(struct nfsmount *); +void nfs_disconnect(struct nfsmount *); +void nfs_safedisconnect(struct nfsmount *); +int nfs_getattrcache(struct vnode *, struct vattr *); +int nfs_iosize(struct nfsmount *nmp); +int nfsm_strtmbuf(struct mbuf **, char **, const char *, long); +int nfs_bioread(struct vnode *, struct uio *, int, struct ucred *); +int nfsm_uiotombuf(struct uio *, struct mbuf **, int, caddr_t *); +void nfs_clearcommit(struct mount *); +int nfs_writebp(struct buf *, int, struct thread *); +int nfs_fsinfo(struct nfsmount *, struct vnode *, struct ucred *, + struct thread *); +int nfs_meta_setsize (struct vnode *, struct ucred *, + struct thread *, u_quad_t); + +void nfs_set_sigmask(struct thread *td, sigset_t *oldset); +void nfs_restore_sigmask(struct thread *td, sigset_t *set); +int nfs_msleep(struct thread *td, void *ident, struct mtx *mtx, + int priority, char *wmesg, int timo); + +#endif /* _KERNEL */ + +#endif diff --git a/sys/nfsclient/nfs_bio.c b/sys/nfsclient/nfs_bio.c new file mode 100644 index 0000000..d564c08 --- /dev/null +++ b/sys/nfsclient/nfs_bio.c @@ -0,0 +1,1801 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include "opt_kdtrace.h" + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/bio.h> +#include <sys/buf.h> +#include <sys/kernel.h> +#include <sys/mbuf.h> +#include <sys/mount.h> +#include <sys/proc.h> +#include <sys/vmmeter.h> +#include <sys/vnode.h> + +#include <vm/vm.h> +#include <vm/vm_extern.h> +#include <vm/vm_page.h> +#include <vm/vm_object.h> +#include <vm/vm_pager.h> +#include <vm/vnode_pager.h> + +#include <nfs/nfsproto.h> +#include <nfsclient/nfs.h> +#include <nfsclient/nfsmount.h> +#include <nfsclient/nfsnode.h> +#include <nfs/nfs_kdtrace.h> + +static struct buf *nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size, + struct thread *td); +static int nfs_directio_write(struct vnode *vp, struct uio *uiop, + struct ucred *cred, int ioflag); + +extern int nfs_directio_enable; +extern int nfs_directio_allow_mmap; + +/* + * Vnode op for VM getpages. + */ +int +nfs_getpages(struct vop_getpages_args *ap) +{ + int i, error, nextoff, size, toff, count, npages; + struct uio uio; + struct iovec iov; + vm_offset_t kva; + struct buf *bp; + struct vnode *vp; + struct thread *td; + struct ucred *cred; + struct nfsmount *nmp; + vm_object_t object; + vm_page_t *pages; + struct nfsnode *np; + + vp = ap->a_vp; + np = VTONFS(vp); + td = curthread; /* XXX */ + cred = curthread->td_ucred; /* XXX */ + nmp = VFSTONFS(vp->v_mount); + pages = ap->a_m; + count = ap->a_count; + + if ((object = vp->v_object) == NULL) { + nfs_printf("nfs_getpages: called with non-merged cache vnode??\n"); + return (VM_PAGER_ERROR); + } + + if (nfs_directio_enable && !nfs_directio_allow_mmap) { + mtx_lock(&np->n_mtx); + if ((np->n_flag & NNONCACHE) && (vp->v_type == VREG)) { + mtx_unlock(&np->n_mtx); + nfs_printf("nfs_getpages: called on non-cacheable vnode??\n"); + return (VM_PAGER_ERROR); + } else + mtx_unlock(&np->n_mtx); + } + + mtx_lock(&nmp->nm_mtx); + if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && + (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { + mtx_unlock(&nmp->nm_mtx); + /* We'll never get here for v4, because we always have fsinfo */ + (void)nfs_fsinfo(nmp, vp, cred, td); + } else + mtx_unlock(&nmp->nm_mtx); + + npages = btoc(count); + + /* + * If the requested page is partially valid, just return it and + * allow the pager to zero-out the blanks. Partially valid pages + * can only occur at the file EOF. + */ + VM_OBJECT_LOCK(object); + if (pages[ap->a_reqpage]->valid != 0) { + for (i = 0; i < npages; ++i) { + if (i != ap->a_reqpage) { + vm_page_lock(pages[i]); + vm_page_free(pages[i]); + vm_page_unlock(pages[i]); + } + } + VM_OBJECT_UNLOCK(object); + return (0); + } + VM_OBJECT_UNLOCK(object); + + /* + * We use only the kva address for the buffer, but this is extremely + * convienient and fast. + */ + bp = getpbuf(&nfs_pbuf_freecnt); + + kva = (vm_offset_t) bp->b_data; + pmap_qenter(kva, pages, npages); + PCPU_INC(cnt.v_vnodein); + PCPU_ADD(cnt.v_vnodepgsin, npages); + + iov.iov_base = (caddr_t) kva; + iov.iov_len = count; + uio.uio_iov = &iov; + uio.uio_iovcnt = 1; + uio.uio_offset = IDX_TO_OFF(pages[0]->pindex); + uio.uio_resid = count; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_READ; + uio.uio_td = td; + + error = (nmp->nm_rpcops->nr_readrpc)(vp, &uio, cred); + pmap_qremove(kva, npages); + + relpbuf(bp, &nfs_pbuf_freecnt); + + if (error && (uio.uio_resid == count)) { + nfs_printf("nfs_getpages: error %d\n", error); + VM_OBJECT_LOCK(object); + for (i = 0; i < npages; ++i) { + if (i != ap->a_reqpage) { + vm_page_lock(pages[i]); + vm_page_free(pages[i]); + vm_page_unlock(pages[i]); + } + } + VM_OBJECT_UNLOCK(object); + return (VM_PAGER_ERROR); + } + + /* + * Calculate the number of bytes read and validate only that number + * of bytes. Note that due to pending writes, size may be 0. This + * does not mean that the remaining data is invalid! + */ + + size = count - uio.uio_resid; + VM_OBJECT_LOCK(object); + for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { + vm_page_t m; + nextoff = toff + PAGE_SIZE; + m = pages[i]; + + if (nextoff <= size) { + /* + * Read operation filled an entire page + */ + m->valid = VM_PAGE_BITS_ALL; + KASSERT(m->dirty == 0, + ("nfs_getpages: page %p is dirty", m)); + } else if (size > toff) { + /* + * Read operation filled a partial page. + */ + m->valid = 0; + vm_page_set_valid_range(m, 0, size - toff); + KASSERT(m->dirty == 0, + ("nfs_getpages: page %p is dirty", m)); + } else { + /* + * Read operation was short. If no error occured + * we may have hit a zero-fill section. We simply + * leave valid set to 0. + */ + ; + } + if (i != ap->a_reqpage) { + /* + * Whether or not to leave the page activated is up in + * the air, but we should put the page on a page queue + * somewhere (it already is in the object). Result: + * It appears that emperical results show that + * deactivating pages is best. + */ + + /* + * Just in case someone was asking for this page we + * now tell them that it is ok to use. + */ + if (!error) { + if (m->oflags & VPO_WANTED) { + vm_page_lock(m); + vm_page_activate(m); + vm_page_unlock(m); + } else { + vm_page_lock(m); + vm_page_deactivate(m); + vm_page_unlock(m); + } + vm_page_wakeup(m); + } else { + vm_page_lock(m); + vm_page_free(m); + vm_page_unlock(m); + } + } + } + VM_OBJECT_UNLOCK(object); + return (0); +} + +/* + * Vnode op for VM putpages. + */ +int +nfs_putpages(struct vop_putpages_args *ap) +{ + struct uio uio; + struct iovec iov; + vm_offset_t kva; + struct buf *bp; + int iomode, must_commit, i, error, npages, count; + off_t offset; + int *rtvals; + struct vnode *vp; + struct thread *td; + struct ucred *cred; + struct nfsmount *nmp; + struct nfsnode *np; + vm_page_t *pages; + + vp = ap->a_vp; + np = VTONFS(vp); + td = curthread; /* XXX */ + cred = curthread->td_ucred; /* XXX */ + nmp = VFSTONFS(vp->v_mount); + pages = ap->a_m; + count = ap->a_count; + rtvals = ap->a_rtvals; + npages = btoc(count); + offset = IDX_TO_OFF(pages[0]->pindex); + + mtx_lock(&nmp->nm_mtx); + if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && + (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { + mtx_unlock(&nmp->nm_mtx); + (void)nfs_fsinfo(nmp, vp, cred, td); + } else + mtx_unlock(&nmp->nm_mtx); + + mtx_lock(&np->n_mtx); + if (nfs_directio_enable && !nfs_directio_allow_mmap && + (np->n_flag & NNONCACHE) && (vp->v_type == VREG)) { + mtx_unlock(&np->n_mtx); + nfs_printf("nfs_putpages: called on noncache-able vnode??\n"); + mtx_lock(&np->n_mtx); + } + + for (i = 0; i < npages; i++) + rtvals[i] = VM_PAGER_ERROR; + + /* + * When putting pages, do not extend file past EOF. + */ + if (offset + count > np->n_size) { + count = np->n_size - offset; + if (count < 0) + count = 0; + } + mtx_unlock(&np->n_mtx); + + /* + * We use only the kva address for the buffer, but this is extremely + * convienient and fast. + */ + bp = getpbuf(&nfs_pbuf_freecnt); + + kva = (vm_offset_t) bp->b_data; + pmap_qenter(kva, pages, npages); + PCPU_INC(cnt.v_vnodeout); + PCPU_ADD(cnt.v_vnodepgsout, count); + + iov.iov_base = (caddr_t) kva; + iov.iov_len = count; + uio.uio_iov = &iov; + uio.uio_iovcnt = 1; + uio.uio_offset = offset; + uio.uio_resid = count; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_WRITE; + uio.uio_td = td; + + if ((ap->a_sync & VM_PAGER_PUT_SYNC) == 0) + iomode = NFSV3WRITE_UNSTABLE; + else + iomode = NFSV3WRITE_FILESYNC; + + error = (nmp->nm_rpcops->nr_writerpc)(vp, &uio, cred, &iomode, &must_commit); + + pmap_qremove(kva, npages); + relpbuf(bp, &nfs_pbuf_freecnt); + + if (!error) { + vnode_pager_undirty_pages(pages, rtvals, count - uio.uio_resid); + if (must_commit) { + nfs_clearcommit(vp->v_mount); + } + } + return rtvals[0]; +} + +/* + * For nfs, cache consistency can only be maintained approximately. + * Although RFC1094 does not specify the criteria, the following is + * believed to be compatible with the reference port. + * For nfs: + * If the file's modify time on the server has changed since the + * last read rpc or you have written to the file, + * you may have lost data cache consistency with the + * server, so flush all of the file's data out of the cache. + * Then force a getattr rpc to ensure that you have up to date + * attributes. + * NB: This implies that cache data can be read when up to + * NFS_ATTRTIMEO seconds out of date. If you find that you need current + * attributes this could be forced by setting n_attrstamp to 0 before + * the VOP_GETATTR() call. + */ +static inline int +nfs_bioread_check_cons(struct vnode *vp, struct thread *td, struct ucred *cred) +{ + int error = 0; + struct vattr vattr; + struct nfsnode *np = VTONFS(vp); + int old_lock; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + + /* + * Grab the exclusive lock before checking whether the cache is + * consistent. + * XXX - We can make this cheaper later (by acquiring cheaper locks). + * But for now, this suffices. + */ + old_lock = nfs_upgrade_vnlock(vp); + if (vp->v_iflag & VI_DOOMED) { + nfs_downgrade_vnlock(vp, old_lock); + return (EBADF); + } + + mtx_lock(&np->n_mtx); + if (np->n_flag & NMODIFIED) { + mtx_unlock(&np->n_mtx); + if (vp->v_type != VREG) { + if (vp->v_type != VDIR) + panic("nfs: bioread, not dir"); + (nmp->nm_rpcops->nr_invaldir)(vp); + error = nfs_vinvalbuf(vp, V_SAVE, td, 1); + if (error) + goto out; + } + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + error = VOP_GETATTR(vp, &vattr, cred); + if (error) + goto out; + mtx_lock(&np->n_mtx); + np->n_mtime = vattr.va_mtime; + mtx_unlock(&np->n_mtx); + } else { + mtx_unlock(&np->n_mtx); + error = VOP_GETATTR(vp, &vattr, cred); + if (error) + return (error); + mtx_lock(&np->n_mtx); + if ((np->n_flag & NSIZECHANGED) + || (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime))) { + mtx_unlock(&np->n_mtx); + if (vp->v_type == VDIR) + (nmp->nm_rpcops->nr_invaldir)(vp); + error = nfs_vinvalbuf(vp, V_SAVE, td, 1); + if (error) + goto out; + mtx_lock(&np->n_mtx); + np->n_mtime = vattr.va_mtime; + np->n_flag &= ~NSIZECHANGED; + } + mtx_unlock(&np->n_mtx); + } +out: + nfs_downgrade_vnlock(vp, old_lock); + return error; +} + +/* + * Vnode op for read using bio + */ +int +nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred) +{ + struct nfsnode *np = VTONFS(vp); + int biosize, i; + struct buf *bp, *rabp; + struct thread *td; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + daddr_t lbn, rabn; + off_t end; + int bcount; + int seqcount; + int nra, error = 0, n = 0, on = 0; + + KASSERT(uio->uio_rw == UIO_READ, ("nfs_read mode")); + if (uio->uio_resid == 0) + return (0); + if (uio->uio_offset < 0) /* XXX VDIR cookies can be negative */ + return (EINVAL); + td = uio->uio_td; + + mtx_lock(&nmp->nm_mtx); + if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && + (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { + mtx_unlock(&nmp->nm_mtx); + (void)nfs_fsinfo(nmp, vp, cred, td); + } else + mtx_unlock(&nmp->nm_mtx); + + end = uio->uio_offset + uio->uio_resid; + if (vp->v_type != VDIR && + (end > nmp->nm_maxfilesize || end < uio->uio_offset)) + return (EFBIG); + + if (nfs_directio_enable && (ioflag & IO_DIRECT) && (vp->v_type == VREG)) + /* No caching/ no readaheads. Just read data into the user buffer */ + return nfs_readrpc(vp, uio, cred); + + biosize = vp->v_bufobj.bo_bsize; + seqcount = (int)((off_t)(ioflag >> IO_SEQSHIFT) * biosize / BKVASIZE); + + error = nfs_bioread_check_cons(vp, td, cred); + if (error) + return error; + + do { + u_quad_t nsize; + + mtx_lock(&np->n_mtx); + nsize = np->n_size; + mtx_unlock(&np->n_mtx); + + switch (vp->v_type) { + case VREG: + nfsstats.biocache_reads++; + lbn = uio->uio_offset / biosize; + on = uio->uio_offset & (biosize - 1); + + /* + * Start the read ahead(s), as required. + */ + if (nmp->nm_readahead > 0) { + for (nra = 0; nra < nmp->nm_readahead && nra < seqcount && + (off_t)(lbn + 1 + nra) * biosize < nsize; nra++) { + rabn = lbn + 1 + nra; + if (incore(&vp->v_bufobj, rabn) == NULL) { + rabp = nfs_getcacheblk(vp, rabn, biosize, td); + if (!rabp) { + error = nfs_sigintr(nmp, td); + return (error ? error : EINTR); + } + if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { + rabp->b_flags |= B_ASYNC; + rabp->b_iocmd = BIO_READ; + vfs_busy_pages(rabp, 0); + if (nfs_asyncio(nmp, rabp, cred, td)) { + rabp->b_flags |= B_INVAL; + rabp->b_ioflags |= BIO_ERROR; + vfs_unbusy_pages(rabp); + brelse(rabp); + break; + } + } else { + brelse(rabp); + } + } + } + } + + /* Note that bcount is *not* DEV_BSIZE aligned. */ + bcount = biosize; + if ((off_t)lbn * biosize >= nsize) { + bcount = 0; + } else if ((off_t)(lbn + 1) * biosize > nsize) { + bcount = nsize - (off_t)lbn * biosize; + } + bp = nfs_getcacheblk(vp, lbn, bcount, td); + + if (!bp) { + error = nfs_sigintr(nmp, td); + return (error ? error : EINTR); + } + + /* + * If B_CACHE is not set, we must issue the read. If this + * fails, we return an error. + */ + + if ((bp->b_flags & B_CACHE) == 0) { + bp->b_iocmd = BIO_READ; + vfs_busy_pages(bp, 0); + error = nfs_doio(vp, bp, cred, td); + if (error) { + brelse(bp); + return (error); + } + } + + /* + * on is the offset into the current bp. Figure out how many + * bytes we can copy out of the bp. Note that bcount is + * NOT DEV_BSIZE aligned. + * + * Then figure out how many bytes we can copy into the uio. + */ + + n = 0; + if (on < bcount) + n = min((unsigned)(bcount - on), uio->uio_resid); + break; + case VLNK: + nfsstats.biocache_readlinks++; + bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, td); + if (!bp) { + error = nfs_sigintr(nmp, td); + return (error ? error : EINTR); + } + if ((bp->b_flags & B_CACHE) == 0) { + bp->b_iocmd = BIO_READ; + vfs_busy_pages(bp, 0); + error = nfs_doio(vp, bp, cred, td); + if (error) { + bp->b_ioflags |= BIO_ERROR; + brelse(bp); + return (error); + } + } + n = min(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); + on = 0; + break; + case VDIR: + nfsstats.biocache_readdirs++; + if (np->n_direofoffset + && uio->uio_offset >= np->n_direofoffset) { + return (0); + } + lbn = (uoff_t)uio->uio_offset / NFS_DIRBLKSIZ; + on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); + bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, td); + if (!bp) { + error = nfs_sigintr(nmp, td); + return (error ? error : EINTR); + } + if ((bp->b_flags & B_CACHE) == 0) { + bp->b_iocmd = BIO_READ; + vfs_busy_pages(bp, 0); + error = nfs_doio(vp, bp, cred, td); + if (error) { + brelse(bp); + } + while (error == NFSERR_BAD_COOKIE) { + (nmp->nm_rpcops->nr_invaldir)(vp); + error = nfs_vinvalbuf(vp, 0, td, 1); + /* + * Yuck! The directory has been modified on the + * server. The only way to get the block is by + * reading from the beginning to get all the + * offset cookies. + * + * Leave the last bp intact unless there is an error. + * Loop back up to the while if the error is another + * NFSERR_BAD_COOKIE (double yuch!). + */ + for (i = 0; i <= lbn && !error; i++) { + if (np->n_direofoffset + && (i * NFS_DIRBLKSIZ) >= np->n_direofoffset) + return (0); + bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, td); + if (!bp) { + error = nfs_sigintr(nmp, td); + return (error ? error : EINTR); + } + if ((bp->b_flags & B_CACHE) == 0) { + bp->b_iocmd = BIO_READ; + vfs_busy_pages(bp, 0); + error = nfs_doio(vp, bp, cred, td); + /* + * no error + B_INVAL == directory EOF, + * use the block. + */ + if (error == 0 && (bp->b_flags & B_INVAL)) + break; + } + /* + * An error will throw away the block and the + * for loop will break out. If no error and this + * is not the block we want, we throw away the + * block and go for the next one via the for loop. + */ + if (error || i < lbn) + brelse(bp); + } + } + /* + * The above while is repeated if we hit another cookie + * error. If we hit an error and it wasn't a cookie error, + * we give up. + */ + if (error) + return (error); + } + + /* + * If not eof and read aheads are enabled, start one. + * (You need the current block first, so that you have the + * directory offset cookie of the next block.) + */ + if (nmp->nm_readahead > 0 && + (bp->b_flags & B_INVAL) == 0 && + (np->n_direofoffset == 0 || + (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && + incore(&vp->v_bufobj, lbn + 1) == NULL) { + rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, td); + if (rabp) { + if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { + rabp->b_flags |= B_ASYNC; + rabp->b_iocmd = BIO_READ; + vfs_busy_pages(rabp, 0); + if (nfs_asyncio(nmp, rabp, cred, td)) { + rabp->b_flags |= B_INVAL; + rabp->b_ioflags |= BIO_ERROR; + vfs_unbusy_pages(rabp); + brelse(rabp); + } + } else { + brelse(rabp); + } + } + } + /* + * Unlike VREG files, whos buffer size ( bp->b_bcount ) is + * chopped for the EOF condition, we cannot tell how large + * NFS directories are going to be until we hit EOF. So + * an NFS directory buffer is *not* chopped to its EOF. Now, + * it just so happens that b_resid will effectively chop it + * to EOF. *BUT* this information is lost if the buffer goes + * away and is reconstituted into a B_CACHE state ( due to + * being VMIO ) later. So we keep track of the directory eof + * in np->n_direofoffset and chop it off as an extra step + * right here. + */ + n = lmin(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); + if (np->n_direofoffset && n > np->n_direofoffset - uio->uio_offset) + n = np->n_direofoffset - uio->uio_offset; + break; + default: + nfs_printf(" nfs_bioread: type %x unexpected\n", vp->v_type); + bp = NULL; + break; + }; + + if (n > 0) { + error = uiomove(bp->b_data + on, (int)n, uio); + } + if (vp->v_type == VLNK) + n = 0; + if (bp != NULL) + brelse(bp); + } while (error == 0 && uio->uio_resid > 0 && n > 0); + return (error); +} + +/* + * The NFS write path cannot handle iovecs with len > 1. So we need to + * break up iovecs accordingly (restricting them to wsize). + * For the SYNC case, we can do this with 1 copy (user buffer -> mbuf). + * For the ASYNC case, 2 copies are needed. The first a copy from the + * user buffer to a staging buffer and then a second copy from the staging + * buffer to mbufs. This can be optimized by copying from the user buffer + * directly into mbufs and passing the chain down, but that requires a + * fair amount of re-working of the relevant codepaths (and can be done + * later). + */ +static int +nfs_directio_write(vp, uiop, cred, ioflag) + struct vnode *vp; + struct uio *uiop; + struct ucred *cred; + int ioflag; +{ + int error; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + struct thread *td = uiop->uio_td; + int size; + int wsize; + + mtx_lock(&nmp->nm_mtx); + wsize = nmp->nm_wsize; + mtx_unlock(&nmp->nm_mtx); + if (ioflag & IO_SYNC) { + int iomode, must_commit; + struct uio uio; + struct iovec iov; +do_sync: + while (uiop->uio_resid > 0) { + size = min(uiop->uio_resid, wsize); + size = min(uiop->uio_iov->iov_len, size); + iov.iov_base = uiop->uio_iov->iov_base; + iov.iov_len = size; + uio.uio_iov = &iov; + uio.uio_iovcnt = 1; + uio.uio_offset = uiop->uio_offset; + uio.uio_resid = size; + uio.uio_segflg = UIO_USERSPACE; + uio.uio_rw = UIO_WRITE; + uio.uio_td = td; + iomode = NFSV3WRITE_FILESYNC; + error = (nmp->nm_rpcops->nr_writerpc)(vp, &uio, cred, + &iomode, &must_commit); + KASSERT((must_commit == 0), + ("nfs_directio_write: Did not commit write")); + if (error) + return (error); + uiop->uio_offset += size; + uiop->uio_resid -= size; + if (uiop->uio_iov->iov_len <= size) { + uiop->uio_iovcnt--; + uiop->uio_iov++; + } else { + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + size; + uiop->uio_iov->iov_len -= size; + } + } + } else { + struct uio *t_uio; + struct iovec *t_iov; + struct buf *bp; + + /* + * Break up the write into blocksize chunks and hand these + * over to nfsiod's for write back. + * Unfortunately, this incurs a copy of the data. Since + * the user could modify the buffer before the write is + * initiated. + * + * The obvious optimization here is that one of the 2 copies + * in the async write path can be eliminated by copying the + * data here directly into mbufs and passing the mbuf chain + * down. But that will require a fair amount of re-working + * of the code and can be done if there's enough interest + * in NFS directio access. + */ + while (uiop->uio_resid > 0) { + size = min(uiop->uio_resid, wsize); + size = min(uiop->uio_iov->iov_len, size); + bp = getpbuf(&nfs_pbuf_freecnt); + t_uio = malloc(sizeof(struct uio), M_NFSDIRECTIO, M_WAITOK); + t_iov = malloc(sizeof(struct iovec), M_NFSDIRECTIO, M_WAITOK); + t_iov->iov_base = malloc(size, M_NFSDIRECTIO, M_WAITOK); + t_iov->iov_len = size; + t_uio->uio_iov = t_iov; + t_uio->uio_iovcnt = 1; + t_uio->uio_offset = uiop->uio_offset; + t_uio->uio_resid = size; + t_uio->uio_segflg = UIO_SYSSPACE; + t_uio->uio_rw = UIO_WRITE; + t_uio->uio_td = td; + bcopy(uiop->uio_iov->iov_base, t_iov->iov_base, size); + bp->b_flags |= B_DIRECT; + bp->b_iocmd = BIO_WRITE; + if (cred != NOCRED) { + crhold(cred); + bp->b_wcred = cred; + } else + bp->b_wcred = NOCRED; + bp->b_caller1 = (void *)t_uio; + bp->b_vp = vp; + error = nfs_asyncio(nmp, bp, NOCRED, td); + if (error) { + free(t_iov->iov_base, M_NFSDIRECTIO); + free(t_iov, M_NFSDIRECTIO); + free(t_uio, M_NFSDIRECTIO); + bp->b_vp = NULL; + relpbuf(bp, &nfs_pbuf_freecnt); + if (error == EINTR) + return (error); + goto do_sync; + } + uiop->uio_offset += size; + uiop->uio_resid -= size; + if (uiop->uio_iov->iov_len <= size) { + uiop->uio_iovcnt--; + uiop->uio_iov++; + } else { + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + size; + uiop->uio_iov->iov_len -= size; + } + } + } + return (0); +} + +/* + * Vnode op for write using bio + */ +int +nfs_write(struct vop_write_args *ap) +{ + int biosize; + struct uio *uio = ap->a_uio; + struct thread *td = uio->uio_td; + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + struct ucred *cred = ap->a_cred; + int ioflag = ap->a_ioflag; + struct buf *bp; + struct vattr vattr; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + daddr_t lbn; + off_t end; + int bcount; + int n, on, error = 0; + + KASSERT(uio->uio_rw == UIO_WRITE, ("nfs_write mode")); + KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread, + ("nfs_write proc")); + if (vp->v_type != VREG) + return (EIO); + mtx_lock(&np->n_mtx); + if (np->n_flag & NWRITEERR) { + np->n_flag &= ~NWRITEERR; + mtx_unlock(&np->n_mtx); + return (np->n_error); + } else + mtx_unlock(&np->n_mtx); + mtx_lock(&nmp->nm_mtx); + if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && + (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { + mtx_unlock(&nmp->nm_mtx); + (void)nfs_fsinfo(nmp, vp, cred, td); + } else + mtx_unlock(&nmp->nm_mtx); + + /* + * Synchronously flush pending buffers if we are in synchronous + * mode or if we are appending. + */ + if (ioflag & (IO_APPEND | IO_SYNC)) { + mtx_lock(&np->n_mtx); + if (np->n_flag & NMODIFIED) { + mtx_unlock(&np->n_mtx); +#ifdef notyet /* Needs matching nonblock semantics elsewhere, too. */ + /* + * Require non-blocking, synchronous writes to + * dirty files to inform the program it needs + * to fsync(2) explicitly. + */ + if (ioflag & IO_NDELAY) + return (EAGAIN); +#endif +flush_and_restart: + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + error = nfs_vinvalbuf(vp, V_SAVE, td, 1); + if (error) + return (error); + } else + mtx_unlock(&np->n_mtx); + } + + /* + * If IO_APPEND then load uio_offset. We restart here if we cannot + * get the append lock. + */ + if (ioflag & IO_APPEND) { + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + error = VOP_GETATTR(vp, &vattr, cred); + if (error) + return (error); + mtx_lock(&np->n_mtx); + uio->uio_offset = np->n_size; + mtx_unlock(&np->n_mtx); + } + + if (uio->uio_offset < 0) + return (EINVAL); + end = uio->uio_offset + uio->uio_resid; + if (end > nmp->nm_maxfilesize || end < uio->uio_offset) + return (EFBIG); + if (uio->uio_resid == 0) + return (0); + + if (nfs_directio_enable && (ioflag & IO_DIRECT) && vp->v_type == VREG) + return nfs_directio_write(vp, uio, cred, ioflag); + + /* + * Maybe this should be above the vnode op call, but so long as + * file servers have no limits, i don't think it matters + */ + if (vn_rlimit_fsize(vp, uio, td)) + return (EFBIG); + + biosize = vp->v_bufobj.bo_bsize; + /* + * Find all of this file's B_NEEDCOMMIT buffers. If our writes + * would exceed the local maximum per-file write commit size when + * combined with those, we must decide whether to flush, + * go synchronous, or return error. We don't bother checking + * IO_UNIT -- we just make all writes atomic anyway, as there's + * no point optimizing for something that really won't ever happen. + */ + if (!(ioflag & IO_SYNC)) { + int nflag; + + mtx_lock(&np->n_mtx); + nflag = np->n_flag; + mtx_unlock(&np->n_mtx); + int needrestart = 0; + if (nmp->nm_wcommitsize < uio->uio_resid) { + /* + * If this request could not possibly be completed + * without exceeding the maximum outstanding write + * commit size, see if we can convert it into a + * synchronous write operation. + */ + if (ioflag & IO_NDELAY) + return (EAGAIN); + ioflag |= IO_SYNC; + if (nflag & NMODIFIED) + needrestart = 1; + } else if (nflag & NMODIFIED) { + int wouldcommit = 0; + BO_LOCK(&vp->v_bufobj); + if (vp->v_bufobj.bo_dirty.bv_cnt != 0) { + TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, + b_bobufs) { + if (bp->b_flags & B_NEEDCOMMIT) + wouldcommit += bp->b_bcount; + } + } + BO_UNLOCK(&vp->v_bufobj); + /* + * Since we're not operating synchronously and + * bypassing the buffer cache, we are in a commit + * and holding all of these buffers whether + * transmitted or not. If not limited, this + * will lead to the buffer cache deadlocking, + * as no one else can flush our uncommitted buffers. + */ + wouldcommit += uio->uio_resid; + /* + * If we would initially exceed the maximum + * outstanding write commit size, flush and restart. + */ + if (wouldcommit > nmp->nm_wcommitsize) + needrestart = 1; + } + if (needrestart) + goto flush_and_restart; + } + + do { + nfsstats.biocache_writes++; + lbn = uio->uio_offset / biosize; + on = uio->uio_offset & (biosize-1); + n = min((unsigned)(biosize - on), uio->uio_resid); +again: + /* + * Handle direct append and file extension cases, calculate + * unaligned buffer size. + */ + mtx_lock(&np->n_mtx); + if (uio->uio_offset == np->n_size && n) { + mtx_unlock(&np->n_mtx); + /* + * Get the buffer (in its pre-append state to maintain + * B_CACHE if it was previously set). Resize the + * nfsnode after we have locked the buffer to prevent + * readers from reading garbage. + */ + bcount = on; + bp = nfs_getcacheblk(vp, lbn, bcount, td); + + if (bp != NULL) { + long save; + + mtx_lock(&np->n_mtx); + np->n_size = uio->uio_offset + n; + np->n_flag |= NMODIFIED; + vnode_pager_setsize(vp, np->n_size); + mtx_unlock(&np->n_mtx); + + save = bp->b_flags & B_CACHE; + bcount += n; + allocbuf(bp, bcount); + bp->b_flags |= save; + } + } else { + /* + * Obtain the locked cache block first, and then + * adjust the file's size as appropriate. + */ + bcount = on + n; + if ((off_t)lbn * biosize + bcount < np->n_size) { + if ((off_t)(lbn + 1) * biosize < np->n_size) + bcount = biosize; + else + bcount = np->n_size - (off_t)lbn * biosize; + } + mtx_unlock(&np->n_mtx); + bp = nfs_getcacheblk(vp, lbn, bcount, td); + mtx_lock(&np->n_mtx); + if (uio->uio_offset + n > np->n_size) { + np->n_size = uio->uio_offset + n; + np->n_flag |= NMODIFIED; + vnode_pager_setsize(vp, np->n_size); + } + mtx_unlock(&np->n_mtx); + } + + if (!bp) { + error = nfs_sigintr(nmp, td); + if (!error) + error = EINTR; + break; + } + + /* + * Issue a READ if B_CACHE is not set. In special-append + * mode, B_CACHE is based on the buffer prior to the write + * op and is typically set, avoiding the read. If a read + * is required in special append mode, the server will + * probably send us a short-read since we extended the file + * on our end, resulting in b_resid == 0 and, thusly, + * B_CACHE getting set. + * + * We can also avoid issuing the read if the write covers + * the entire buffer. We have to make sure the buffer state + * is reasonable in this case since we will not be initiating + * I/O. See the comments in kern/vfs_bio.c's getblk() for + * more information. + * + * B_CACHE may also be set due to the buffer being cached + * normally. + */ + + if (on == 0 && n == bcount) { + bp->b_flags |= B_CACHE; + bp->b_flags &= ~B_INVAL; + bp->b_ioflags &= ~BIO_ERROR; + } + + if ((bp->b_flags & B_CACHE) == 0) { + bp->b_iocmd = BIO_READ; + vfs_busy_pages(bp, 0); + error = nfs_doio(vp, bp, cred, td); + if (error) { + brelse(bp); + break; + } + } + if (bp->b_wcred == NOCRED) + bp->b_wcred = crhold(cred); + mtx_lock(&np->n_mtx); + np->n_flag |= NMODIFIED; + mtx_unlock(&np->n_mtx); + + /* + * If dirtyend exceeds file size, chop it down. This should + * not normally occur but there is an append race where it + * might occur XXX, so we log it. + * + * If the chopping creates a reverse-indexed or degenerate + * situation with dirtyoff/end, we 0 both of them. + */ + + if (bp->b_dirtyend > bcount) { + nfs_printf("NFS append race @%lx:%d\n", + (long)bp->b_blkno * DEV_BSIZE, + bp->b_dirtyend - bcount); + bp->b_dirtyend = bcount; + } + + if (bp->b_dirtyoff >= bp->b_dirtyend) + bp->b_dirtyoff = bp->b_dirtyend = 0; + + /* + * If the new write will leave a contiguous dirty + * area, just update the b_dirtyoff and b_dirtyend, + * otherwise force a write rpc of the old dirty area. + * + * While it is possible to merge discontiguous writes due to + * our having a B_CACHE buffer ( and thus valid read data + * for the hole), we don't because it could lead to + * significant cache coherency problems with multiple clients, + * especially if locking is implemented later on. + * + * as an optimization we could theoretically maintain + * a linked list of discontinuous areas, but we would still + * have to commit them separately so there isn't much + * advantage to it except perhaps a bit of asynchronization. + */ + + if (bp->b_dirtyend > 0 && + (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { + if (bwrite(bp) == EINTR) { + error = EINTR; + break; + } + goto again; + } + + error = uiomove((char *)bp->b_data + on, n, uio); + + /* + * Since this block is being modified, it must be written + * again and not just committed. Since write clustering does + * not work for the stage 1 data write, only the stage 2 + * commit rpc, we have to clear B_CLUSTEROK as well. + */ + bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); + + if (error) { + bp->b_ioflags |= BIO_ERROR; + brelse(bp); + break; + } + + /* + * Only update dirtyoff/dirtyend if not a degenerate + * condition. + */ + if (n) { + if (bp->b_dirtyend > 0) { + bp->b_dirtyoff = min(on, bp->b_dirtyoff); + bp->b_dirtyend = max((on + n), bp->b_dirtyend); + } else { + bp->b_dirtyoff = on; + bp->b_dirtyend = on + n; + } + vfs_bio_set_valid(bp, on, n); + } + + /* + * If IO_SYNC do bwrite(). + * + * IO_INVAL appears to be unused. The idea appears to be + * to turn off caching in this case. Very odd. XXX + */ + if ((ioflag & IO_SYNC)) { + if (ioflag & IO_INVAL) + bp->b_flags |= B_NOCACHE; + error = bwrite(bp); + if (error) + break; + } else if ((n + on) == biosize) { + bp->b_flags |= B_ASYNC; + (void) (nmp->nm_rpcops->nr_writebp)(bp, 0, NULL); + } else { + bdwrite(bp); + } + } while (uio->uio_resid > 0 && n > 0); + + return (error); +} + +/* + * Get an nfs cache block. + * + * Allocate a new one if the block isn't currently in the cache + * and return the block marked busy. If the calling process is + * interrupted by a signal for an interruptible mount point, return + * NULL. + * + * The caller must carefully deal with the possible B_INVAL state of + * the buffer. nfs_doio() clears B_INVAL (and nfs_asyncio() clears it + * indirectly), so synchronous reads can be issued without worrying about + * the B_INVAL state. We have to be a little more careful when dealing + * with writes (see comments in nfs_write()) when extending a file past + * its EOF. + */ +static struct buf * +nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size, struct thread *td) +{ + struct buf *bp; + struct mount *mp; + struct nfsmount *nmp; + + mp = vp->v_mount; + nmp = VFSTONFS(mp); + + if (nmp->nm_flag & NFSMNT_INT) { + sigset_t oldset; + + nfs_set_sigmask(td, &oldset); + bp = getblk(vp, bn, size, NFS_PCATCH, 0, 0); + nfs_restore_sigmask(td, &oldset); + while (bp == NULL) { + if (nfs_sigintr(nmp, td)) + return (NULL); + bp = getblk(vp, bn, size, 0, 2 * hz, 0); + } + } else { + bp = getblk(vp, bn, size, 0, 0, 0); + } + + if (vp->v_type == VREG) + bp->b_blkno = bn * (vp->v_bufobj.bo_bsize / DEV_BSIZE); + return (bp); +} + +/* + * Flush and invalidate all dirty buffers. If another process is already + * doing the flush, just wait for completion. + */ +int +nfs_vinvalbuf(struct vnode *vp, int flags, struct thread *td, int intrflg) +{ + struct nfsnode *np = VTONFS(vp); + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + int error = 0, slpflag, slptimeo; + int old_lock = 0; + + ASSERT_VOP_LOCKED(vp, "nfs_vinvalbuf"); + + if ((nmp->nm_flag & NFSMNT_INT) == 0) + intrflg = 0; + if (intrflg) { + slpflag = NFS_PCATCH; + slptimeo = 2 * hz; + } else { + slpflag = 0; + slptimeo = 0; + } + + old_lock = nfs_upgrade_vnlock(vp); + if (vp->v_iflag & VI_DOOMED) { + /* + * Since vgonel() uses the generic vinvalbuf() to flush + * dirty buffers and it does not call this function, it + * is safe to just return OK when VI_DOOMED is set. + */ + nfs_downgrade_vnlock(vp, old_lock); + return (0); + } + + /* + * Now, flush as required. + */ + if ((flags & V_SAVE) && (vp->v_bufobj.bo_object != NULL)) { + VM_OBJECT_LOCK(vp->v_bufobj.bo_object); + vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC); + VM_OBJECT_UNLOCK(vp->v_bufobj.bo_object); + /* + * If the page clean was interrupted, fail the invalidation. + * Not doing so, we run the risk of losing dirty pages in the + * vinvalbuf() call below. + */ + if (intrflg && (error = nfs_sigintr(nmp, td))) + goto out; + } + + error = vinvalbuf(vp, flags, slpflag, 0); + while (error) { + if (intrflg && (error = nfs_sigintr(nmp, td))) + goto out; + error = vinvalbuf(vp, flags, 0, slptimeo); + } + mtx_lock(&np->n_mtx); + if (np->n_directio_asyncwr == 0) + np->n_flag &= ~NMODIFIED; + mtx_unlock(&np->n_mtx); +out: + nfs_downgrade_vnlock(vp, old_lock); + return error; +} + +/* + * Initiate asynchronous I/O. Return an error if no nfsiods are available. + * This is mainly to avoid queueing async I/O requests when the nfsiods + * are all hung on a dead server. + * + * Note: nfs_asyncio() does not clear (BIO_ERROR|B_INVAL) but when the bp + * is eventually dequeued by the async daemon, nfs_doio() *will*. + */ +int +nfs_asyncio(struct nfsmount *nmp, struct buf *bp, struct ucred *cred, struct thread *td) +{ + int iod; + int gotiod; + int slpflag = 0; + int slptimeo = 0; + int error, error2; + + /* + * Commits are usually short and sweet so lets save some cpu and + * leave the async daemons for more important rpc's (such as reads + * and writes). + */ + mtx_lock(&nfs_iod_mtx); + if (bp->b_iocmd == BIO_WRITE && (bp->b_flags & B_NEEDCOMMIT) && + (nmp->nm_bufqiods > nfs_numasync / 2)) { + mtx_unlock(&nfs_iod_mtx); + return(EIO); + } +again: + if (nmp->nm_flag & NFSMNT_INT) + slpflag = NFS_PCATCH; + gotiod = FALSE; + + /* + * Find a free iod to process this request. + */ + for (iod = 0; iod < nfs_numasync; iod++) + if (nfs_iodwant[iod] == NFSIOD_AVAILABLE) { + gotiod = TRUE; + break; + } + + /* + * Try to create one if none are free. + */ + if (!gotiod) + nfs_nfsiodnew(); + else { + /* + * Found one, so wake it up and tell it which + * mount to process. + */ + NFS_DPF(ASYNCIO, ("nfs_asyncio: waking iod %d for mount %p\n", + iod, nmp)); + nfs_iodwant[iod] = NFSIOD_NOT_AVAILABLE; + nfs_iodmount[iod] = nmp; + nmp->nm_bufqiods++; + wakeup(&nfs_iodwant[iod]); + } + + /* + * If none are free, we may already have an iod working on this mount + * point. If so, it will process our request. + */ + if (!gotiod) { + if (nmp->nm_bufqiods > 0) { + NFS_DPF(ASYNCIO, + ("nfs_asyncio: %d iods are already processing mount %p\n", + nmp->nm_bufqiods, nmp)); + gotiod = TRUE; + } + } + + /* + * If we have an iod which can process the request, then queue + * the buffer. + */ + if (gotiod) { + /* + * Ensure that the queue never grows too large. We still want + * to asynchronize so we block rather then return EIO. + */ + while (nmp->nm_bufqlen >= 2 * nfs_numasync) { + NFS_DPF(ASYNCIO, + ("nfs_asyncio: waiting for mount %p queue to drain\n", nmp)); + nmp->nm_bufqwant = TRUE; + error = nfs_msleep(td, &nmp->nm_bufq, &nfs_iod_mtx, + slpflag | PRIBIO, + "nfsaio", slptimeo); + if (error) { + error2 = nfs_sigintr(nmp, td); + if (error2) { + mtx_unlock(&nfs_iod_mtx); + return (error2); + } + if (slpflag == NFS_PCATCH) { + slpflag = 0; + slptimeo = 2 * hz; + } + } + /* + * We might have lost our iod while sleeping, + * so check and loop if nescessary. + */ + goto again; + } + + /* We might have lost our nfsiod */ + if (nmp->nm_bufqiods == 0) { + NFS_DPF(ASYNCIO, +("nfs_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); + goto again; + } + + if (bp->b_iocmd == BIO_READ) { + if (bp->b_rcred == NOCRED && cred != NOCRED) + bp->b_rcred = crhold(cred); + } else { + if (bp->b_wcred == NOCRED && cred != NOCRED) + bp->b_wcred = crhold(cred); + } + + if (bp->b_flags & B_REMFREE) + bremfreef(bp); + BUF_KERNPROC(bp); + TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); + nmp->nm_bufqlen++; + if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) { + mtx_lock(&(VTONFS(bp->b_vp))->n_mtx); + VTONFS(bp->b_vp)->n_flag |= NMODIFIED; + VTONFS(bp->b_vp)->n_directio_asyncwr++; + mtx_unlock(&(VTONFS(bp->b_vp))->n_mtx); + } + mtx_unlock(&nfs_iod_mtx); + return (0); + } + + mtx_unlock(&nfs_iod_mtx); + + /* + * All the iods are busy on other mounts, so return EIO to + * force the caller to process the i/o synchronously. + */ + NFS_DPF(ASYNCIO, ("nfs_asyncio: no iods available, i/o is synchronous\n")); + return (EIO); +} + +void +nfs_doio_directwrite(struct buf *bp) +{ + int iomode, must_commit; + struct uio *uiop = (struct uio *)bp->b_caller1; + char *iov_base = uiop->uio_iov->iov_base; + struct nfsmount *nmp = VFSTONFS(bp->b_vp->v_mount); + + iomode = NFSV3WRITE_FILESYNC; + uiop->uio_td = NULL; /* NULL since we're in nfsiod */ + (nmp->nm_rpcops->nr_writerpc)(bp->b_vp, uiop, bp->b_wcred, &iomode, &must_commit); + KASSERT((must_commit == 0), ("nfs_doio_directwrite: Did not commit write")); + free(iov_base, M_NFSDIRECTIO); + free(uiop->uio_iov, M_NFSDIRECTIO); + free(uiop, M_NFSDIRECTIO); + if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) { + struct nfsnode *np = VTONFS(bp->b_vp); + mtx_lock(&np->n_mtx); + np->n_directio_asyncwr--; + if (np->n_directio_asyncwr == 0) { + VTONFS(bp->b_vp)->n_flag &= ~NMODIFIED; + if ((np->n_flag & NFSYNCWAIT)) { + np->n_flag &= ~NFSYNCWAIT; + wakeup((caddr_t)&np->n_directio_asyncwr); + } + } + mtx_unlock(&np->n_mtx); + } + bp->b_vp = NULL; + relpbuf(bp, &nfs_pbuf_freecnt); +} + +/* + * Do an I/O operation to/from a cache block. This may be called + * synchronously or from an nfsiod. + */ +int +nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td) +{ + struct uio *uiop; + struct nfsnode *np; + struct nfsmount *nmp; + int error = 0, iomode, must_commit = 0; + struct uio uio; + struct iovec io; + struct proc *p = td ? td->td_proc : NULL; + uint8_t iocmd; + + np = VTONFS(vp); + nmp = VFSTONFS(vp->v_mount); + uiop = &uio; + uiop->uio_iov = &io; + uiop->uio_iovcnt = 1; + uiop->uio_segflg = UIO_SYSSPACE; + uiop->uio_td = td; + + /* + * clear BIO_ERROR and B_INVAL state prior to initiating the I/O. We + * do this here so we do not have to do it in all the code that + * calls us. + */ + bp->b_flags &= ~B_INVAL; + bp->b_ioflags &= ~BIO_ERROR; + + KASSERT(!(bp->b_flags & B_DONE), ("nfs_doio: bp %p already marked done", bp)); + iocmd = bp->b_iocmd; + if (iocmd == BIO_READ) { + io.iov_len = uiop->uio_resid = bp->b_bcount; + io.iov_base = bp->b_data; + uiop->uio_rw = UIO_READ; + + switch (vp->v_type) { + case VREG: + uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; + nfsstats.read_bios++; + error = (nmp->nm_rpcops->nr_readrpc)(vp, uiop, cr); + + if (!error) { + if (uiop->uio_resid) { + /* + * If we had a short read with no error, we must have + * hit a file hole. We should zero-fill the remainder. + * This can also occur if the server hits the file EOF. + * + * Holes used to be able to occur due to pending + * writes, but that is not possible any longer. + */ + int nread = bp->b_bcount - uiop->uio_resid; + int left = uiop->uio_resid; + + if (left > 0) + bzero((char *)bp->b_data + nread, left); + uiop->uio_resid = 0; + } + } + /* ASSERT_VOP_LOCKED(vp, "nfs_doio"); */ + if (p && (vp->v_vflag & VV_TEXT)) { + mtx_lock(&np->n_mtx); + if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &np->n_vattr.va_mtime)) { + mtx_unlock(&np->n_mtx); + PROC_LOCK(p); + killproc(p, "text file modification"); + PROC_UNLOCK(p); + } else + mtx_unlock(&np->n_mtx); + } + break; + case VLNK: + uiop->uio_offset = (off_t)0; + nfsstats.readlink_bios++; + error = (nmp->nm_rpcops->nr_readlinkrpc)(vp, uiop, cr); + break; + case VDIR: + nfsstats.readdir_bios++; + uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; + if ((nmp->nm_flag & NFSMNT_RDIRPLUS) != 0) { + error = nfs_readdirplusrpc(vp, uiop, cr); + if (error == NFSERR_NOTSUPP) + nmp->nm_flag &= ~NFSMNT_RDIRPLUS; + } + if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) + error = nfs_readdirrpc(vp, uiop, cr); + /* + * end-of-directory sets B_INVAL but does not generate an + * error. + */ + if (error == 0 && uiop->uio_resid == bp->b_bcount) + bp->b_flags |= B_INVAL; + break; + default: + nfs_printf("nfs_doio: type %x unexpected\n", vp->v_type); + break; + }; + if (error) { + bp->b_ioflags |= BIO_ERROR; + bp->b_error = error; + } + } else { + /* + * If we only need to commit, try to commit + */ + if (bp->b_flags & B_NEEDCOMMIT) { + int retv; + off_t off; + + off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; + retv = (nmp->nm_rpcops->nr_commit)( + vp, off, bp->b_dirtyend-bp->b_dirtyoff, + bp->b_wcred, td); + if (retv == 0) { + bp->b_dirtyoff = bp->b_dirtyend = 0; + bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); + bp->b_resid = 0; + bufdone(bp); + return (0); + } + if (retv == NFSERR_STALEWRITEVERF) { + nfs_clearcommit(vp->v_mount); + } + } + + /* + * Setup for actual write + */ + mtx_lock(&np->n_mtx); + if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size) + bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE; + mtx_unlock(&np->n_mtx); + + if (bp->b_dirtyend > bp->b_dirtyoff) { + io.iov_len = uiop->uio_resid = bp->b_dirtyend + - bp->b_dirtyoff; + uiop->uio_offset = (off_t)bp->b_blkno * DEV_BSIZE + + bp->b_dirtyoff; + io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; + uiop->uio_rw = UIO_WRITE; + nfsstats.write_bios++; + + if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE | B_CLUSTER)) == B_ASYNC) + iomode = NFSV3WRITE_UNSTABLE; + else + iomode = NFSV3WRITE_FILESYNC; + + error = (nmp->nm_rpcops->nr_writerpc)(vp, uiop, cr, &iomode, &must_commit); + + /* + * When setting B_NEEDCOMMIT also set B_CLUSTEROK to try + * to cluster the buffers needing commit. This will allow + * the system to submit a single commit rpc for the whole + * cluster. We can do this even if the buffer is not 100% + * dirty (relative to the NFS blocksize), so we optimize the + * append-to-file-case. + * + * (when clearing B_NEEDCOMMIT, B_CLUSTEROK must also be + * cleared because write clustering only works for commit + * rpc's, not for the data portion of the write). + */ + + if (!error && iomode == NFSV3WRITE_UNSTABLE) { + bp->b_flags |= B_NEEDCOMMIT; + if (bp->b_dirtyoff == 0 + && bp->b_dirtyend == bp->b_bcount) + bp->b_flags |= B_CLUSTEROK; + } else { + bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); + } + + /* + * For an interrupted write, the buffer is still valid + * and the write hasn't been pushed to the server yet, + * so we can't set BIO_ERROR and report the interruption + * by setting B_EINTR. For the B_ASYNC case, B_EINTR + * is not relevant, so the rpc attempt is essentially + * a noop. For the case of a V3 write rpc not being + * committed to stable storage, the block is still + * dirty and requires either a commit rpc or another + * write rpc with iomode == NFSV3WRITE_FILESYNC before + * the block is reused. This is indicated by setting + * the B_DELWRI and B_NEEDCOMMIT flags. + * + * If the buffer is marked B_PAGING, it does not reside on + * the vp's paging queues so we cannot call bdirty(). The + * bp in this case is not an NFS cache block so we should + * be safe. XXX + * + * The logic below breaks up errors into recoverable and + * unrecoverable. For the former, we clear B_INVAL|B_NOCACHE + * and keep the buffer around for potential write retries. + * For the latter (eg ESTALE), we toss the buffer away (B_INVAL) + * and save the error in the nfsnode. This is less than ideal + * but necessary. Keeping such buffers around could potentially + * cause buffer exhaustion eventually (they can never be written + * out, so will get constantly be re-dirtied). It also causes + * all sorts of vfs panics. For non-recoverable write errors, + * also invalidate the attrcache, so we'll be forced to go over + * the wire for this object, returning an error to user on next + * call (most of the time). + */ + if (error == EINTR || error == EIO || error == ETIMEDOUT + || (!error && (bp->b_flags & B_NEEDCOMMIT))) { + int s; + + s = splbio(); + bp->b_flags &= ~(B_INVAL|B_NOCACHE); + if ((bp->b_flags & B_PAGING) == 0) { + bdirty(bp); + bp->b_flags &= ~B_DONE; + } + if (error && (bp->b_flags & B_ASYNC) == 0) + bp->b_flags |= B_EINTR; + splx(s); + } else { + if (error) { + bp->b_ioflags |= BIO_ERROR; + bp->b_flags |= B_INVAL; + bp->b_error = np->n_error = error; + mtx_lock(&np->n_mtx); + np->n_flag |= NWRITEERR; + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + mtx_unlock(&np->n_mtx); + } + bp->b_dirtyoff = bp->b_dirtyend = 0; + } + } else { + bp->b_resid = 0; + bufdone(bp); + return (0); + } + } + bp->b_resid = uiop->uio_resid; + if (must_commit) + nfs_clearcommit(vp->v_mount); + bufdone(bp); + return (error); +} + +/* + * Used to aid in handling ftruncate() operations on the NFS client side. + * Truncation creates a number of special problems for NFS. We have to + * throw away VM pages and buffer cache buffers that are beyond EOF, and + * we have to properly handle VM pages or (potentially dirty) buffers + * that straddle the truncation point. + */ + +int +nfs_meta_setsize(struct vnode *vp, struct ucred *cred, struct thread *td, u_quad_t nsize) +{ + struct nfsnode *np = VTONFS(vp); + u_quad_t tsize; + int biosize = vp->v_bufobj.bo_bsize; + int error = 0; + + mtx_lock(&np->n_mtx); + tsize = np->n_size; + np->n_size = nsize; + mtx_unlock(&np->n_mtx); + + if (nsize < tsize) { + struct buf *bp; + daddr_t lbn; + int bufsize; + + /* + * vtruncbuf() doesn't get the buffer overlapping the + * truncation point. We may have a B_DELWRI and/or B_CACHE + * buffer that now needs to be truncated. + */ + error = vtruncbuf(vp, cred, td, nsize, biosize); + lbn = nsize / biosize; + bufsize = nsize & (biosize - 1); + bp = nfs_getcacheblk(vp, lbn, bufsize, td); + if (!bp) + return EINTR; + if (bp->b_dirtyoff > bp->b_bcount) + bp->b_dirtyoff = bp->b_bcount; + if (bp->b_dirtyend > bp->b_bcount) + bp->b_dirtyend = bp->b_bcount; + bp->b_flags |= B_RELBUF; /* don't leave garbage around */ + brelse(bp); + } else { + vnode_pager_setsize(vp, nsize); + } + return(error); +} + diff --git a/sys/nfsclient/nfs_kdtrace.c b/sys/nfsclient/nfs_kdtrace.c new file mode 100644 index 0000000..429dbc3 --- /dev/null +++ b/sys/nfsclient/nfs_kdtrace.c @@ -0,0 +1,542 @@ +/*- + * Copyright (c) 2009 Robert N. M. Watson + * All rights reserved. + * + * This software was developed at the University of Cambridge Computer + * Laboratory with support from a grant from Google, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/malloc.h> +#include <sys/module.h> + +#include <sys/dtrace.h> +#include <sys/dtrace_bsd.h> + +#include <nfs/nfsproto.h> + +/* + * dtnfsclient is a DTrace provider that tracks the intent to perform RPCs + * in the NFS client, as well as acess to and maintenance of the access and + * attribute caches. This is not quite the same as RPCs, because NFS may + * issue multiple RPC transactions in the event that authentication fails, + * there's a jukebox error, or none at all if the access or attribute cache + * hits. However, it cleanly represents the logical layer between RPC + * transmission and vnode/vfs operations, providing access to state linking + * the two. + */ + +static int dtnfsclient_unload(void); +static void dtnfsclient_getargdesc(void *, dtrace_id_t, void *, + dtrace_argdesc_t *); +static void dtnfsclient_provide(void *, dtrace_probedesc_t *); +static void dtnfsclient_destroy(void *, dtrace_id_t, void *); +static void dtnfsclient_enable(void *, dtrace_id_t, void *); +static void dtnfsclient_disable(void *, dtrace_id_t, void *); +static void dtnfsclient_load(void *); + +static dtrace_pattr_t dtnfsclient_attr = { +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +}; + +/* + * Description of NFSv3 and (optional) NFSv2 probes for a procedure. + */ +struct dtnfsclient_rpc { + char *nr_v3_name; + char *nr_v2_name; /* Or NULL if none. */ + + /* + * IDs for the start and done cases, for both NFSv2 and NFSv3. + */ + uint32_t nr_v2_id_start, nr_v2_id_done; + uint32_t nr_v3_id_start, nr_v3_id_done; +}; + +/* + * This table is indexed by NFSv3 procedure number, but also used for NFSv2 + * procedure names. + */ +static struct dtnfsclient_rpc dtnfsclient_rpcs[NFS_NPROCS] = { + { "null", "null" }, + { "getattr", "getattr" }, + { "setattr", "setattr" }, + { "lookup", "lookup" }, + { "access" }, + { "readlink", "readlink" }, + { "read", "read" }, + { "write", "write" }, + { "create", "create" }, + { "mkdir", "mkdir" }, + { "symlink", "symlink" }, + { "mknod" }, + { "remove", "remove" }, + { "rmdir", "rmdir" }, + { "rename", "rename" }, + { "link", "link" }, + { "readdir", "readdir" }, + { "readdirplus" }, + { "fsstat", "statfs" }, + { "fsinfo" }, + { "pathconf" }, + { "commit" }, + { "noop" }, +}; + +/* + * Module name strings. + */ +static char *dtnfsclient_accesscache_str = "accesscache"; +static char *dtnfsclient_attrcache_str = "attrcache"; +static char *dtnfsclient_nfs2_str = "nfs2"; +static char *dtnfsclient_nfs3_str = "nfs3"; + +/* + * Function name strings. + */ +static char *dtnfsclient_flush_str = "flush"; +static char *dtnfsclient_load_str = "load"; +static char *dtnfsclient_get_str = "get"; + +/* + * Name strings. + */ +static char *dtnfsclient_done_str = "done"; +static char *dtnfsclient_hit_str = "hit"; +static char *dtnfsclient_miss_str = "miss"; +static char *dtnfsclient_start_str = "start"; + +static dtrace_pops_t dtnfsclient_pops = { + dtnfsclient_provide, + NULL, + dtnfsclient_enable, + dtnfsclient_disable, + NULL, + NULL, + dtnfsclient_getargdesc, + NULL, + NULL, + dtnfsclient_destroy +}; + +static dtrace_provider_id_t dtnfsclient_id; + +/* + * Most probes are generated from the above RPC table, but for access and + * attribute caches, we have specific IDs we recognize and handle specially + * in various spots. + */ +extern uint32_t nfsclient_accesscache_flush_done_id; +extern uint32_t nfsclient_accesscache_get_hit_id; +extern uint32_t nfsclient_accesscache_get_miss_id; +extern uint32_t nfsclient_accesscache_load_done_id; + +extern uint32_t nfsclient_attrcache_flush_done_id; +extern uint32_t nfsclient_attrcache_get_hit_id; +extern uint32_t nfsclient_attrcache_get_miss_id; +extern uint32_t nfsclient_attrcache_load_done_id; + +/* + * When tracing on a procedure is enabled, the DTrace ID for an RPC event is + * stored in one of these two NFS client-allocated arrays; 0 indicates that + * the event is not being traced so probes should not be called. + * + * For simplicity, we allocate both v2 and v3 arrays as NFS_NPROCS, and the + * v2 array is simply sparse. + */ +extern uint32_t nfsclient_nfs2_start_probes[NFS_NPROCS]; +extern uint32_t nfsclient_nfs2_done_probes[NFS_NPROCS]; + +extern uint32_t nfsclient_nfs3_start_probes[NFS_NPROCS]; +extern uint32_t nfsclient_nfs3_done_probes[NFS_NPROCS]; + +/* + * Look up a DTrace probe ID to see if it's associated with a "done" event -- + * if so, we will return a fourth argument type of "int". + */ +static int +dtnfs23_isdoneprobe(dtrace_id_t id) +{ + int i; + + for (i = 0; i < NFS_NPROCS; i++) { + if (dtnfsclient_rpcs[i].nr_v3_id_done == id || + dtnfsclient_rpcs[i].nr_v2_id_done == id) + return (1); + } + return (0); +} + +static void +dtnfsclient_getargdesc(void *arg, dtrace_id_t id, void *parg, + dtrace_argdesc_t *desc) +{ + const char *p = NULL; + + if (id == nfsclient_accesscache_flush_done_id || + id == nfsclient_attrcache_flush_done_id || + id == nfsclient_attrcache_get_miss_id) { + switch (desc->dtargd_ndx) { + case 0: + p = "struct vnode *"; + break; + default: + desc->dtargd_ndx = DTRACE_ARGNONE; + break; + } + } else if (id == nfsclient_accesscache_get_hit_id || + id == nfsclient_accesscache_get_miss_id) { + switch (desc->dtargd_ndx) { + case 0: + p = "struct vnode *"; + break; + case 1: + p = "uid_t"; + break; + case 2: + p = "uint32_t"; + break; + default: + desc->dtargd_ndx = DTRACE_ARGNONE; + break; + } + } else if (id == nfsclient_accesscache_load_done_id) { + switch (desc->dtargd_ndx) { + case 0: + p = "struct vnode *"; + break; + case 1: + p = "uid_t"; + break; + case 2: + p = "uint32_t"; + break; + case 3: + p = "int"; + break; + default: + desc->dtargd_ndx = DTRACE_ARGNONE; + break; + } + } else if (id == nfsclient_attrcache_get_hit_id) { + switch (desc->dtargd_ndx) { + case 0: + p = "struct vnode *"; + break; + case 1: + p = "struct vattr *"; + break; + default: + desc->dtargd_ndx = DTRACE_ARGNONE; + break; + } + } else if (id == nfsclient_attrcache_load_done_id) { + switch (desc->dtargd_ndx) { + case 0: + p = "struct vnode *"; + break; + case 1: + p = "struct vattr *"; + break; + case 2: + p = "int"; + break; + default: + desc->dtargd_ndx = DTRACE_ARGNONE; + break; + } + } else { + switch (desc->dtargd_ndx) { + case 0: + p = "struct vnode *"; + break; + case 1: + p = "struct mbuf *"; + break; + case 2: + p = "struct ucred *"; + break; + case 3: + p = "int"; + break; + case 4: + if (dtnfs23_isdoneprobe(id)) { + p = "int"; + break; + } + /* FALLSTHROUGH */ + default: + desc->dtargd_ndx = DTRACE_ARGNONE; + break; + } + } + if (p != NULL) + strlcpy(desc->dtargd_native, p, sizeof(desc->dtargd_native)); +} + +static void +dtnfsclient_provide(void *arg, dtrace_probedesc_t *desc) +{ + int i; + + if (desc != NULL) + return; + + /* + * Register access cache probes. + */ + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_flush_str, dtnfsclient_done_str) == 0) { + nfsclient_accesscache_flush_done_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_flush_str, dtnfsclient_done_str, 0, NULL); + } + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_get_str, dtnfsclient_hit_str) == 0) { + nfsclient_accesscache_get_hit_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_get_str, dtnfsclient_hit_str, 0, NULL); + } + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_get_str, dtnfsclient_miss_str) == 0) { + nfsclient_accesscache_get_miss_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_get_str, dtnfsclient_miss_str, 0, NULL); + } + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_load_str, dtnfsclient_done_str) == 0) { + nfsclient_accesscache_load_done_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_accesscache_str, + dtnfsclient_load_str, dtnfsclient_done_str, 0, NULL); + } + + /* + * Register attribute cache probes. + */ + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_flush_str, dtnfsclient_done_str) == 0) { + nfsclient_attrcache_flush_done_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_flush_str, dtnfsclient_done_str, 0, NULL); + } + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_get_str, dtnfsclient_hit_str) == 0) { + nfsclient_attrcache_get_hit_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_get_str, dtnfsclient_hit_str, 0, NULL); + } + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_get_str, dtnfsclient_miss_str) == 0) { + nfsclient_attrcache_get_miss_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_get_str, dtnfsclient_miss_str, 0, NULL); + } + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_load_str, dtnfsclient_done_str) == 0) { + nfsclient_attrcache_load_done_id = dtrace_probe_create( + dtnfsclient_id, dtnfsclient_attrcache_str, + dtnfsclient_load_str, dtnfsclient_done_str, 0, NULL); + } + + /* + * Register NFSv2 RPC procedures; note sparseness check for each slot + * in the NFSv3 procnum-indexed array. + */ + for (i = 0; i < NFS_NPROCS; i++) { + if (dtnfsclient_rpcs[i].nr_v2_name != NULL && + dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_nfs2_str, + dtnfsclient_rpcs[i].nr_v2_name, dtnfsclient_start_str) == + 0) { + dtnfsclient_rpcs[i].nr_v2_id_start = + dtrace_probe_create(dtnfsclient_id, + dtnfsclient_nfs2_str, + dtnfsclient_rpcs[i].nr_v2_name, + dtnfsclient_start_str, 0, + &nfsclient_nfs2_start_probes[i]); + } + if (dtnfsclient_rpcs[i].nr_v2_name != NULL && + dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_nfs2_str, + dtnfsclient_rpcs[i].nr_v2_name, dtnfsclient_done_str) == + 0) { + dtnfsclient_rpcs[i].nr_v2_id_done = + dtrace_probe_create(dtnfsclient_id, + dtnfsclient_nfs2_str, + dtnfsclient_rpcs[i].nr_v2_name, + dtnfsclient_done_str, 0, + &nfsclient_nfs2_done_probes[i]); + } + } + + /* + * Register NFSv3 RPC procedures. + */ + for (i = 0; i < NFS_NPROCS; i++) { + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_nfs3_str, + dtnfsclient_rpcs[i].nr_v3_name, dtnfsclient_start_str) == + 0) { + dtnfsclient_rpcs[i].nr_v3_id_start = + dtrace_probe_create(dtnfsclient_id, + dtnfsclient_nfs3_str, + dtnfsclient_rpcs[i].nr_v3_name, + dtnfsclient_start_str, 0, + &nfsclient_nfs3_start_probes[i]); + } + if (dtrace_probe_lookup(dtnfsclient_id, dtnfsclient_nfs3_str, + dtnfsclient_rpcs[i].nr_v3_name, dtnfsclient_done_str) == + 0) { + dtnfsclient_rpcs[i].nr_v3_id_done = + dtrace_probe_create(dtnfsclient_id, + dtnfsclient_nfs3_str, + dtnfsclient_rpcs[i].nr_v3_name, + dtnfsclient_done_str, 0, + &nfsclient_nfs3_done_probes[i]); + } + } +} + +static void +dtnfsclient_destroy(void *arg, dtrace_id_t id, void *parg) +{ +} + +static void +dtnfsclient_enable(void *arg, dtrace_id_t id, void *parg) +{ + uint32_t *p = parg; + void *f = dtrace_probe; + + if (id == nfsclient_accesscache_flush_done_id) + dtrace_nfsclient_accesscache_flush_done_probe = f; + else if (id == nfsclient_accesscache_get_hit_id) + dtrace_nfsclient_accesscache_get_hit_probe = f; + else if (id == nfsclient_accesscache_get_miss_id) + dtrace_nfsclient_accesscache_get_miss_probe = f; + else if (id == nfsclient_accesscache_load_done_id) + dtrace_nfsclient_accesscache_load_done_probe = f; + else if (id == nfsclient_attrcache_flush_done_id) + dtrace_nfsclient_attrcache_flush_done_probe = f; + else if (id == nfsclient_attrcache_get_hit_id) + dtrace_nfsclient_attrcache_get_hit_probe = f; + else if (id == nfsclient_attrcache_get_miss_id) + dtrace_nfsclient_attrcache_get_miss_probe = f; + else if (id == nfsclient_attrcache_load_done_id) + dtrace_nfsclient_attrcache_load_done_probe = f; + else + *p = id; +} + +static void +dtnfsclient_disable(void *arg, dtrace_id_t id, void *parg) +{ + uint32_t *p = parg; + + if (id == nfsclient_accesscache_flush_done_id) + dtrace_nfsclient_accesscache_flush_done_probe = NULL; + else if (id == nfsclient_accesscache_get_hit_id) + dtrace_nfsclient_accesscache_get_hit_probe = NULL; + else if (id == nfsclient_accesscache_get_miss_id) + dtrace_nfsclient_accesscache_get_miss_probe = NULL; + else if (id == nfsclient_accesscache_load_done_id) + dtrace_nfsclient_accesscache_load_done_probe = NULL; + else if (id == nfsclient_attrcache_flush_done_id) + dtrace_nfsclient_attrcache_flush_done_probe = NULL; + else if (id == nfsclient_attrcache_get_hit_id) + dtrace_nfsclient_attrcache_get_hit_probe = NULL; + else if (id == nfsclient_attrcache_get_miss_id) + dtrace_nfsclient_attrcache_get_miss_probe = NULL; + else if (id == nfsclient_attrcache_load_done_id) + dtrace_nfsclient_attrcache_load_done_probe = NULL; + else + *p = 0; +} + +static void +dtnfsclient_load(void *dummy) +{ + + if (dtrace_register("nfsclient", &dtnfsclient_attr, + DTRACE_PRIV_USER, NULL, &dtnfsclient_pops, NULL, + &dtnfsclient_id) != 0) + return; + + dtrace_nfsclient_nfs23_start_probe = + (dtrace_nfsclient_nfs23_start_probe_func_t)dtrace_probe; + dtrace_nfsclient_nfs23_done_probe = + (dtrace_nfsclient_nfs23_done_probe_func_t)dtrace_probe; +} + + +static int +dtnfsclient_unload() +{ + + dtrace_nfsclient_nfs23_start_probe = NULL; + dtrace_nfsclient_nfs23_done_probe = NULL; + + return (dtrace_unregister(dtnfsclient_id)); +} + +static int +dtnfsclient_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + } + + return (error); +} + +SYSINIT(dtnfsclient_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, + dtnfsclient_load, NULL); +SYSUNINIT(dtnfsclient_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, + dtnfsclient_unload, NULL); + +DEV_MODULE(dtnfsclient, dtnfsclient_modevent, NULL); +MODULE_VERSION(dtnfsclient, 1); +MODULE_DEPEND(dtnfsclient, dtrace, 1, 1, 1); +MODULE_DEPEND(dtnfsclient, opensolaris, 1, 1, 1); +MODULE_DEPEND(dtnfsclient, oldnfs, 1, 1, 1); diff --git a/sys/nfsclient/nfs_krpc.c b/sys/nfsclient/nfs_krpc.c new file mode 100644 index 0000000..0389d29 --- /dev/null +++ b/sys/nfsclient/nfs_krpc.c @@ -0,0 +1,881 @@ +/*- + * Copyright (c) 1989, 1991, 1993, 1995 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +/* + * Socket operations for use by nfs + */ + +#include "opt_inet6.h" +#include "opt_kdtrace.h" +#include "opt_kgssapi.h" + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/limits.h> +#include <sys/lock.h> +#include <sys/malloc.h> +#include <sys/mbuf.h> +#include <sys/mount.h> +#include <sys/mutex.h> +#include <sys/proc.h> +#include <sys/signalvar.h> +#include <sys/syscallsubr.h> +#include <sys/sysctl.h> +#include <sys/syslog.h> +#include <sys/vnode.h> + +#include <rpc/rpc.h> + +#include <nfs/nfsproto.h> +#include <nfsclient/nfs.h> +#include <nfs/xdr_subs.h> +#include <nfsclient/nfsm_subs.h> +#include <nfsclient/nfsmount.h> +#include <nfsclient/nfsnode.h> + +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> + +dtrace_nfsclient_nfs23_start_probe_func_t + dtrace_nfsclient_nfs23_start_probe; + +dtrace_nfsclient_nfs23_done_probe_func_t + dtrace_nfsclient_nfs23_done_probe; + +/* + * Registered probes by RPC type. + */ +uint32_t nfsclient_nfs2_start_probes[NFS_NPROCS]; +uint32_t nfsclient_nfs2_done_probes[NFS_NPROCS]; + +uint32_t nfsclient_nfs3_start_probes[NFS_NPROCS]; +uint32_t nfsclient_nfs3_done_probes[NFS_NPROCS]; +#endif + +static int nfs_bufpackets = 4; +static int nfs_reconnects; +static int nfs3_jukebox_delay = 10; +static int nfs_skip_wcc_data_onerr = 1; +static int fake_wchan; + +SYSCTL_DECL(_vfs_oldnfs); + +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, + "Buffer reservation size 2 < x < 64"); +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0, + "Number of times the nfs client has had to reconnect"); +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, + &nfs3_jukebox_delay, 0, + "Number of seconds to delay a retry after receiving EJUKEBOX"); +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, + &nfs_skip_wcc_data_onerr, 0, + "Disable weak cache consistency checking when server returns an error"); + +static void nfs_down(struct nfsmount *, struct thread *, const char *, + int, int); +static void nfs_up(struct nfsmount *, struct thread *, const char *, + int, int); +static int nfs_msg(struct thread *, const char *, const char *, int); + +extern int nfsv2_procid[]; + +struct nfs_cached_auth { + int ca_refs; /* refcount, including 1 from the cache */ + uid_t ca_uid; /* uid that corresponds to this auth */ + AUTH *ca_auth; /* RPC auth handle */ +}; + +/* + * RTT estimator + */ + +static enum nfs_rto_timer_t nfs_proct[NFS_NPROCS] = { + NFS_DEFAULT_TIMER, /* NULL */ + NFS_GETATTR_TIMER, /* GETATTR */ + NFS_DEFAULT_TIMER, /* SETATTR */ + NFS_LOOKUP_TIMER, /* LOOKUP */ + NFS_GETATTR_TIMER, /* ACCESS */ + NFS_READ_TIMER, /* READLINK */ + NFS_READ_TIMER, /* READ */ + NFS_WRITE_TIMER, /* WRITE */ + NFS_DEFAULT_TIMER, /* CREATE */ + NFS_DEFAULT_TIMER, /* MKDIR */ + NFS_DEFAULT_TIMER, /* SYMLINK */ + NFS_DEFAULT_TIMER, /* MKNOD */ + NFS_DEFAULT_TIMER, /* REMOVE */ + NFS_DEFAULT_TIMER, /* RMDIR */ + NFS_DEFAULT_TIMER, /* RENAME */ + NFS_DEFAULT_TIMER, /* LINK */ + NFS_READ_TIMER, /* READDIR */ + NFS_READ_TIMER, /* READDIRPLUS */ + NFS_DEFAULT_TIMER, /* FSSTAT */ + NFS_DEFAULT_TIMER, /* FSINFO */ + NFS_DEFAULT_TIMER, /* PATHCONF */ + NFS_DEFAULT_TIMER, /* COMMIT */ + NFS_DEFAULT_TIMER, /* NOOP */ +}; + +/* + * Choose the correct RTT timer for this NFS procedure. + */ +static inline enum nfs_rto_timer_t +nfs_rto_timer(u_int32_t procnum) +{ + + return (nfs_proct[procnum]); +} + +/* + * Initialize the RTT estimator state for a new mount point. + */ +static void +nfs_init_rtt(struct nfsmount *nmp) +{ + int i; + + for (i = 0; i < NFS_MAX_TIMER; i++) { + nmp->nm_timers[i].rt_srtt = hz; + nmp->nm_timers[i].rt_deviate = 0; + nmp->nm_timers[i].rt_rtxcur = hz; + } +} + +/* + * Initialize sockets and congestion for a new NFS connection. + * We do not free the sockaddr if error. + */ +int +nfs_connect(struct nfsmount *nmp) +{ + int rcvreserve, sndreserve; + int pktscale; + struct sockaddr *saddr; + struct ucred *origcred; + struct thread *td = curthread; + CLIENT *client; + struct netconfig *nconf; + rpcvers_t vers; + int one = 1, retries; + struct timeval timo; + + /* + * We need to establish the socket using the credentials of + * the mountpoint. Some parts of this process (such as + * sobind() and soconnect()) will use the curent thread's + * credential instead of the socket credential. To work + * around this, temporarily change the current thread's + * credential to that of the mountpoint. + * + * XXX: It would be better to explicitly pass the correct + * credential to sobind() and soconnect(). + */ + origcred = td->td_ucred; + td->td_ucred = nmp->nm_mountp->mnt_cred; + saddr = nmp->nm_nam; + + vers = NFS_VER2; + if (nmp->nm_flag & NFSMNT_NFSV3) + vers = NFS_VER3; + else if (nmp->nm_flag & NFSMNT_NFSV4) + vers = NFS_VER4; + if (saddr->sa_family == AF_INET) + if (nmp->nm_sotype == SOCK_DGRAM) + nconf = getnetconfigent("udp"); + else + nconf = getnetconfigent("tcp"); + else + if (nmp->nm_sotype == SOCK_DGRAM) + nconf = getnetconfigent("udp6"); + else + nconf = getnetconfigent("tcp6"); + + /* + * Get buffer reservation size from sysctl, but impose reasonable + * limits. + */ + pktscale = nfs_bufpackets; + if (pktscale < 2) + pktscale = 2; + if (pktscale > 64) + pktscale = 64; + mtx_lock(&nmp->nm_mtx); + if (nmp->nm_sotype == SOCK_DGRAM) { + sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale; + rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + + NFS_MAXPKTHDR) * pktscale; + } else if (nmp->nm_sotype == SOCK_SEQPACKET) { + sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale; + rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + + NFS_MAXPKTHDR) * pktscale; + } else { + if (nmp->nm_sotype != SOCK_STREAM) + panic("nfscon sotype"); + sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + + sizeof (u_int32_t)) * pktscale; + rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + + sizeof (u_int32_t)) * pktscale; + } + mtx_unlock(&nmp->nm_mtx); + + client = clnt_reconnect_create(nconf, saddr, NFS_PROG, vers, + sndreserve, rcvreserve); + CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq"); + if (nmp->nm_flag & NFSMNT_INT) + CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one); + if (nmp->nm_flag & NFSMNT_RESVPORT) + CLNT_CONTROL(client, CLSET_PRIVPORT, &one); + if ((nmp->nm_flag & NFSMNT_SOFT) != 0) { + if (nmp->nm_sotype == SOCK_DGRAM) + /* + * For UDP, the large timeout for a reconnect will + * be set to "nm_retry * nm_timeo / 2", so we only + * want to do 2 reconnect timeout retries. + */ + retries = 2; + else + retries = nmp->nm_retry; + } else + retries = INT_MAX; + CLNT_CONTROL(client, CLSET_RETRIES, &retries); + + /* + * For UDP, there are 2 timeouts: + * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer + * that does a retransmit of an RPC request using the same socket + * and xid. This is what you normally want to do, since NFS + * servers depend on "same xid" for their Duplicate Request Cache. + * - timeout specified in CLNT_CALL_MBUF(), which specifies when + * retransmits on the same socket should fail and a fresh socket + * created. Each of these timeouts counts as one CLSET_RETRIES, + * as set above. + * Set the initial retransmit timeout for UDP. This timeout doesn't + * exist for TCP and the following call just fails, which is ok. + */ + timo.tv_sec = nmp->nm_timeo / NFS_HZ; + timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ; + CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo); + + mtx_lock(&nmp->nm_mtx); + if (nmp->nm_client) { + /* + * Someone else already connected. + */ + CLNT_RELEASE(client); + } else + nmp->nm_client = client; + + /* + * Protocols that do not require connections may be optionally left + * unconnected for servers that reply from a port other than NFS_PORT. + */ + if (!(nmp->nm_flag & NFSMNT_NOCONN)) { + mtx_unlock(&nmp->nm_mtx); + CLNT_CONTROL(client, CLSET_CONNECT, &one); + } else + mtx_unlock(&nmp->nm_mtx); + + /* Restore current thread's credentials. */ + td->td_ucred = origcred; + + mtx_lock(&nmp->nm_mtx); + /* Initialize other non-zero congestion variables. */ + nfs_init_rtt(nmp); + mtx_unlock(&nmp->nm_mtx); + return (0); +} + +/* + * NFS disconnect. Clean up and unlink. + */ +void +nfs_disconnect(struct nfsmount *nmp) +{ + CLIENT *client; + + mtx_lock(&nmp->nm_mtx); + if (nmp->nm_client) { + client = nmp->nm_client; + nmp->nm_client = NULL; + mtx_unlock(&nmp->nm_mtx); + rpc_gss_secpurge_call(client); + CLNT_CLOSE(client); + CLNT_RELEASE(client); + } else + mtx_unlock(&nmp->nm_mtx); +} + +void +nfs_safedisconnect(struct nfsmount *nmp) +{ + + nfs_disconnect(nmp); +} + +static AUTH * +nfs_getauth(struct nfsmount *nmp, struct ucred *cred) +{ + rpc_gss_service_t svc; + AUTH *auth; + + switch (nmp->nm_secflavor) { + case RPCSEC_GSS_KRB5: + case RPCSEC_GSS_KRB5I: + case RPCSEC_GSS_KRB5P: + if (!nmp->nm_mech_oid) + if (!rpc_gss_mech_to_oid_call("kerberosv5", + &nmp->nm_mech_oid)) + return (NULL); + if (nmp->nm_secflavor == RPCSEC_GSS_KRB5) + svc = rpc_gss_svc_none; + else if (nmp->nm_secflavor == RPCSEC_GSS_KRB5I) + svc = rpc_gss_svc_integrity; + else + svc = rpc_gss_svc_privacy; + auth = rpc_gss_secfind_call(nmp->nm_client, cred, + nmp->nm_principal, nmp->nm_mech_oid, svc); + if (auth) + return (auth); + /* fallthrough */ + case AUTH_SYS: + default: + return (authunix_create(cred)); + + } +} + +/* + * Callback from the RPC code to generate up/down notifications. + */ + +struct nfs_feedback_arg { + struct nfsmount *nf_mount; + int nf_lastmsg; /* last tprintf */ + int nf_tprintfmsg; + struct thread *nf_td; +}; + +static void +nfs_feedback(int type, int proc, void *arg) +{ + struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg; + struct nfsmount *nmp = nf->nf_mount; + struct timeval now; + + getmicrouptime(&now); + + switch (type) { + case FEEDBACK_REXMIT2: + case FEEDBACK_RECONNECT: + if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) { + nfs_down(nmp, nf->nf_td, + "not responding", 0, NFSSTA_TIMEO); + nf->nf_tprintfmsg = TRUE; + nf->nf_lastmsg = now.tv_sec; + } + break; + + case FEEDBACK_OK: + nfs_up(nf->nf_mount, nf->nf_td, + "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg); + break; + } +} + +/* + * nfs_request - goes something like this + * - fill in request struct + * - links it into list + * - calls nfs_send() for first transmit + * - calls nfs_receive() to get reply + * - break down rpc header and return with nfs reply pointed to + * by mrep or error + * nb: always frees up mreq mbuf list + */ +int +nfs_request(struct vnode *vp, struct mbuf *mreq, int procnum, + struct thread *td, struct ucred *cred, struct mbuf **mrp, + struct mbuf **mdp, caddr_t *dposp) +{ + struct mbuf *mrep; + u_int32_t *tl; + struct nfsmount *nmp; + struct mbuf *md; + time_t waituntil; + caddr_t dpos; + int error = 0, timeo; + struct timeval now; + AUTH *auth = NULL; + enum nfs_rto_timer_t timer; + struct nfs_feedback_arg nf; + struct rpc_callextra ext; + enum clnt_stat stat; + struct timeval timo; + + /* Reject requests while attempting a forced unmount. */ + if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) { + m_freem(mreq); + return (ESTALE); + } + nmp = VFSTONFS(vp->v_mount); + bzero(&nf, sizeof(struct nfs_feedback_arg)); + nf.nf_mount = nmp; + nf.nf_td = td; + getmicrouptime(&now); + nf.nf_lastmsg = now.tv_sec - + ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay)); + + /* + * XXX if not already connected call nfs_connect now. Longer + * term, change nfs_mount to call nfs_connect unconditionally + * and let clnt_reconnect_create handle reconnects. + */ + if (!nmp->nm_client) + nfs_connect(nmp); + + auth = nfs_getauth(nmp, cred); + if (!auth) { + m_freem(mreq); + return (EACCES); + } + bzero(&ext, sizeof(ext)); + ext.rc_auth = auth; + + ext.rc_feedback = nfs_feedback; + ext.rc_feedback_arg = &nf; + + /* + * Use a conservative timeout for RPCs other than getattr, + * lookup, read or write. The justification for doing "other" + * this way is that these RPCs happen so infrequently that + * timer est. would probably be stale. Also, since many of + * these RPCs are non-idempotent, a conservative timeout is + * desired. + */ + timer = nfs_rto_timer(procnum); + if (timer != NFS_DEFAULT_TIMER) + ext.rc_timers = &nmp->nm_timers[timer - 1]; + else + ext.rc_timers = NULL; + +#ifdef KDTRACE_HOOKS + if (dtrace_nfsclient_nfs23_start_probe != NULL) { + uint32_t probe_id; + int probe_procnum; + + if (nmp->nm_flag & NFSMNT_NFSV3) { + probe_id = nfsclient_nfs3_start_probes[procnum]; + probe_procnum = procnum; + } else { + probe_id = nfsclient_nfs2_start_probes[procnum]; + probe_procnum = nfsv2_procid[procnum]; + } + if (probe_id != 0) + (dtrace_nfsclient_nfs23_start_probe)(probe_id, vp, + mreq, cred, probe_procnum); + } +#endif + + nfsstats.rpcrequests++; +tryagain: + /* + * This timeout specifies when a new socket should be created, + * along with new xid values. For UDP, this should be done + * infrequently, since retransmits of RPC requests should normally + * use the same xid. + */ + if (nmp->nm_sotype == SOCK_DGRAM) { + if ((nmp->nm_flag & NFSMNT_SOFT) != 0) { + /* + * CLSET_RETRIES is set to 2, so this should be half + * of the total timeout required. + */ + timeo = nmp->nm_retry * nmp->nm_timeo / 2; + if (timeo < 1) + timeo = 1; + timo.tv_sec = timeo / NFS_HZ; + timo.tv_usec = (timeo % NFS_HZ) * 1000000 / NFS_HZ; + } else { + /* For UDP hard mounts, use a large value. */ + timo.tv_sec = NFS_MAXTIMEO / NFS_HZ; + timo.tv_usec = 0; + } + } else { + timo.tv_sec = nmp->nm_timeo / NFS_HZ; + timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ; + } + mrep = NULL; + stat = CLNT_CALL_MBUF(nmp->nm_client, &ext, + (nmp->nm_flag & NFSMNT_NFSV3) ? procnum : nfsv2_procid[procnum], + mreq, &mrep, timo); + + /* + * If there was a successful reply and a tprintf msg. + * tprintf a response. + */ + if (stat == RPC_SUCCESS) + error = 0; + else if (stat == RPC_TIMEDOUT) + error = ETIMEDOUT; + else if (stat == RPC_VERSMISMATCH) + error = EOPNOTSUPP; + else if (stat == RPC_PROGVERSMISMATCH) + error = EPROTONOSUPPORT; + else + error = EACCES; + if (error) + goto nfsmout; + + KASSERT(mrep != NULL, ("mrep shouldn't be NULL if no error\n")); + + /* + * Search for any mbufs that are not a multiple of 4 bytes long + * or with m_data not longword aligned. + * These could cause pointer alignment problems, so copy them to + * well aligned mbufs. + */ + error = nfs_realign(&mrep, M_DONTWAIT); + if (error == ENOMEM) { + m_freem(mrep); + AUTH_DESTROY(auth); + return (error); + } + + md = mrep; + dpos = mtod(mrep, caddr_t); + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + if (*tl != 0) { + error = fxdr_unsigned(int, *tl); + if ((nmp->nm_flag & NFSMNT_NFSV3) && + error == NFSERR_TRYLATER) { + m_freem(mrep); + error = 0; + waituntil = time_second + nfs3_jukebox_delay; + while (time_second < waituntil) + (void)tsleep(&fake_wchan, PSOCK, "nqnfstry", + hz); + goto tryagain; + } + /* + * Make sure NFSERR_RETERR isn't bogusly set by a server + * such as amd. (No actual NFS error has bit 31 set.) + */ + error &= ~NFSERR_RETERR; + + /* + * If the File Handle was stale, invalidate the lookup + * cache, just in case. + */ + if (error == ESTALE) + nfs_purgecache(vp); + /* + * Skip wcc data on NFS errors for now. NetApp filers + * return corrupt postop attrs in the wcc data for NFS + * err EROFS. Not sure if they could return corrupt + * postop attrs for others errors. + */ + if ((nmp->nm_flag & NFSMNT_NFSV3) && + !nfs_skip_wcc_data_onerr) { + *mrp = mrep; + *mdp = md; + *dposp = dpos; + error |= NFSERR_RETERR; + } else + m_freem(mrep); + goto nfsmout; + } + +#ifdef KDTRACE_HOOKS + if (dtrace_nfsclient_nfs23_done_probe != NULL) { + uint32_t probe_id; + int probe_procnum; + + if (nmp->nm_flag & NFSMNT_NFSV3) { + probe_id = nfsclient_nfs3_done_probes[procnum]; + probe_procnum = procnum; + } else { + probe_id = nfsclient_nfs2_done_probes[procnum]; + probe_procnum = (nmp->nm_flag & NFSMNT_NFSV3) ? + procnum : nfsv2_procid[procnum]; + } + if (probe_id != 0) + (dtrace_nfsclient_nfs23_done_probe)(probe_id, vp, + mreq, cred, probe_procnum, 0); + } +#endif + m_freem(mreq); + *mrp = mrep; + *mdp = md; + *dposp = dpos; + AUTH_DESTROY(auth); + return (0); + +nfsmout: +#ifdef KDTRACE_HOOKS + if (dtrace_nfsclient_nfs23_done_probe != NULL) { + uint32_t probe_id; + int probe_procnum; + + if (nmp->nm_flag & NFSMNT_NFSV3) { + probe_id = nfsclient_nfs3_done_probes[procnum]; + probe_procnum = procnum; + } else { + probe_id = nfsclient_nfs2_done_probes[procnum]; + probe_procnum = (nmp->nm_flag & NFSMNT_NFSV3) ? + procnum : nfsv2_procid[procnum]; + } + if (probe_id != 0) + (dtrace_nfsclient_nfs23_done_probe)(probe_id, vp, + mreq, cred, probe_procnum, error); + } +#endif + m_freem(mreq); + if (auth) + AUTH_DESTROY(auth); + return (error); +} + +/* + * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and + * wait for all requests to complete. This is used by forced unmounts + * to terminate any outstanding RPCs. + */ +int +nfs_nmcancelreqs(struct nfsmount *nmp) +{ + + if (nmp->nm_client) + CLNT_CLOSE(nmp->nm_client); + return (0); +} + +/* + * Any signal that can interrupt an NFS operation in an intr mount + * should be added to this set. SIGSTOP and SIGKILL cannot be masked. + */ +int nfs_sig_set[] = { + SIGINT, + SIGTERM, + SIGHUP, + SIGKILL, + SIGSTOP, + SIGQUIT +}; + +/* + * Check to see if one of the signals in our subset is pending on + * the process (in an intr mount). + */ +static int +nfs_sig_pending(sigset_t set) +{ + int i; + + for (i = 0 ; i < sizeof(nfs_sig_set)/sizeof(int) ; i++) + if (SIGISMEMBER(set, nfs_sig_set[i])) + return (1); + return (0); +} + +/* + * The set/restore sigmask functions are used to (temporarily) overwrite + * the process p_sigmask during an RPC call (for example). These are also + * used in other places in the NFS client that might tsleep(). + */ +void +nfs_set_sigmask(struct thread *td, sigset_t *oldset) +{ + sigset_t newset; + int i; + struct proc *p; + + SIGFILLSET(newset); + if (td == NULL) + td = curthread; /* XXX */ + p = td->td_proc; + /* Remove the NFS set of signals from newset. */ + PROC_LOCK(p); + mtx_lock(&p->p_sigacts->ps_mtx); + for (i = 0 ; i < sizeof(nfs_sig_set)/sizeof(int) ; i++) { + /* + * But make sure we leave the ones already masked + * by the process, i.e. remove the signal from the + * temporary signalmask only if it wasn't already + * in p_sigmask. + */ + if (!SIGISMEMBER(td->td_sigmask, nfs_sig_set[i]) && + !SIGISMEMBER(p->p_sigacts->ps_sigignore, nfs_sig_set[i])) + SIGDELSET(newset, nfs_sig_set[i]); + } + mtx_unlock(&p->p_sigacts->ps_mtx); + PROC_UNLOCK(p); + kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, 0); +} + +void +nfs_restore_sigmask(struct thread *td, sigset_t *set) +{ + if (td == NULL) + td = curthread; /* XXX */ + kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0); +} + +/* + * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the + * old one after msleep() returns. + */ +int +nfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, + char *wmesg, int timo) +{ + sigset_t oldset; + int error; + struct proc *p; + + if ((priority & PCATCH) == 0) + return msleep(ident, mtx, priority, wmesg, timo); + if (td == NULL) + td = curthread; /* XXX */ + nfs_set_sigmask(td, &oldset); + error = msleep(ident, mtx, priority, wmesg, timo); + nfs_restore_sigmask(td, &oldset); + p = td->td_proc; + return (error); +} + +/* + * Test for a termination condition pending on the process. + * This is used for NFSMNT_INT mounts. + */ +int +nfs_sigintr(struct nfsmount *nmp, struct thread *td) +{ + struct proc *p; + sigset_t tmpset; + + /* Terminate all requests while attempting a forced unmount. */ + if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) + return (EIO); + if (!(nmp->nm_flag & NFSMNT_INT)) + return (0); + if (td == NULL) + return (0); + p = td->td_proc; + PROC_LOCK(p); + tmpset = p->p_siglist; + SIGSETOR(tmpset, td->td_siglist); + SIGSETNAND(tmpset, td->td_sigmask); + mtx_lock(&p->p_sigacts->ps_mtx); + SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore); + mtx_unlock(&p->p_sigacts->ps_mtx); + if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist)) + && nfs_sig_pending(tmpset)) { + PROC_UNLOCK(p); + return (EINTR); + } + PROC_UNLOCK(p); + return (0); +} + +static int +nfs_msg(struct thread *td, const char *server, const char *msg, int error) +{ + struct proc *p; + + p = td ? td->td_proc : NULL; + if (error) + tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server, + msg, error); + else + tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg); + return (0); +} + +static void +nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg, + int error, int flags) +{ + if (nmp == NULL) + return; + mtx_lock(&nmp->nm_mtx); + if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) { + nmp->nm_state |= NFSSTA_TIMEO; + mtx_unlock(&nmp->nm_mtx); + vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, + VQ_NOTRESP, 0); + } else + mtx_unlock(&nmp->nm_mtx); + mtx_lock(&nmp->nm_mtx); + if ((flags & NFSSTA_LOCKTIMEO) && + !(nmp->nm_state & NFSSTA_LOCKTIMEO)) { + nmp->nm_state |= NFSSTA_LOCKTIMEO; + mtx_unlock(&nmp->nm_mtx); + vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, + VQ_NOTRESPLOCK, 0); + } else + mtx_unlock(&nmp->nm_mtx); + nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error); +} + +static void +nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg, + int flags, int tprintfmsg) +{ + if (nmp == NULL) + return; + if (tprintfmsg) + nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0); + + mtx_lock(&nmp->nm_mtx); + if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) { + nmp->nm_state &= ~NFSSTA_TIMEO; + mtx_unlock(&nmp->nm_mtx); + vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, + VQ_NOTRESP, 1); + } else + mtx_unlock(&nmp->nm_mtx); + + mtx_lock(&nmp->nm_mtx); + if ((flags & NFSSTA_LOCKTIMEO) && + (nmp->nm_state & NFSSTA_LOCKTIMEO)) { + nmp->nm_state &= ~NFSSTA_LOCKTIMEO; + mtx_unlock(&nmp->nm_mtx); + vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, + VQ_NOTRESPLOCK, 1); + } else + mtx_unlock(&nmp->nm_mtx); +} diff --git a/sys/nfsclient/nfs_nfsiod.c b/sys/nfsclient/nfs_nfsiod.c new file mode 100644 index 0000000..d34c205 --- /dev/null +++ b/sys/nfsclient/nfs_nfsiod.c @@ -0,0 +1,338 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/sysproto.h> +#include <sys/kernel.h> +#include <sys/sysctl.h> +#include <sys/file.h> +#include <sys/filedesc.h> +#include <sys/vnode.h> +#include <sys/malloc.h> +#include <sys/mount.h> +#include <sys/proc.h> +#include <sys/bio.h> +#include <sys/buf.h> +#include <sys/mbuf.h> +#include <sys/socket.h> +#include <sys/socketvar.h> +#include <sys/domain.h> +#include <sys/protosw.h> +#include <sys/namei.h> +#include <sys/unistd.h> +#include <sys/kthread.h> +#include <sys/fcntl.h> +#include <sys/lockf.h> +#include <sys/mutex.h> +#include <sys/taskqueue.h> + +#include <netinet/in.h> +#include <netinet/tcp.h> + +#include <nfs/xdr_subs.h> +#include <nfs/nfsproto.h> +#include <nfsclient/nfs.h> +#include <nfsclient/nfsm_subs.h> +#include <nfsclient/nfsmount.h> +#include <nfsclient/nfsnode.h> +#include <nfs/nfs_lock.h> + +static MALLOC_DEFINE(M_NFSSVC, "nfsclient_srvsock", "Nfs server structure"); + +static void nfssvc_iod(void *); + +static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; + +SYSCTL_DECL(_vfs_oldnfs); + +/* Maximum number of seconds a nfsiod kthread will sleep before exiting */ +static unsigned int nfs_iodmaxidle = 120; +SYSCTL_UINT(_vfs_oldnfs, OID_AUTO, iodmaxidle, CTLFLAG_RW, &nfs_iodmaxidle, 0, + "Max number of seconds an nfsiod kthread will sleep before exiting"); + +/* Maximum number of nfsiod kthreads */ +unsigned int nfs_iodmax = 20; + +/* Minimum number of nfsiod kthreads to keep as spares */ +static unsigned int nfs_iodmin = 0; + +static int nfs_nfsiodnew_sync(void); + +static int +sysctl_iodmin(SYSCTL_HANDLER_ARGS) +{ + int error, i; + int newmin; + + newmin = nfs_iodmin; + error = sysctl_handle_int(oidp, &newmin, 0, req); + if (error || (req->newptr == NULL)) + return (error); + mtx_lock(&nfs_iod_mtx); + if (newmin > nfs_iodmax) { + error = EINVAL; + goto out; + } + nfs_iodmin = newmin; + if (nfs_numasync >= nfs_iodmin) + goto out; + /* + * If the current number of nfsiod is lower + * than the new minimum, create some more. + */ + for (i = nfs_iodmin - nfs_numasync; i > 0; i--) + nfs_nfsiodnew_sync(); +out: + mtx_unlock(&nfs_iod_mtx); + return (0); +} +SYSCTL_PROC(_vfs_oldnfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0, + sizeof (nfs_iodmin), sysctl_iodmin, "IU", + "Min number of nfsiod kthreads to keep as spares"); + + +static int +sysctl_iodmax(SYSCTL_HANDLER_ARGS) +{ + int error, i; + int iod, newmax; + + newmax = nfs_iodmax; + error = sysctl_handle_int(oidp, &newmax, 0, req); + if (error || (req->newptr == NULL)) + return (error); + if (newmax > NFS_MAXASYNCDAEMON) + return (EINVAL); + mtx_lock(&nfs_iod_mtx); + nfs_iodmax = newmax; + if (nfs_numasync <= nfs_iodmax) + goto out; + /* + * If there are some asleep nfsiods that should + * exit, wakeup() them so that they check nfs_iodmax + * and exit. Those who are active will exit as + * soon as they finish I/O. + */ + iod = nfs_numasync - 1; + for (i = 0; i < nfs_numasync - nfs_iodmax; i++) { + if (nfs_iodwant[iod] == NFSIOD_AVAILABLE) + wakeup(&nfs_iodwant[iod]); + iod--; + } +out: + mtx_unlock(&nfs_iod_mtx); + return (0); +} +SYSCTL_PROC(_vfs_oldnfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0, + sizeof (nfs_iodmax), sysctl_iodmax, "IU", + "Max number of nfsiod kthreads"); + +static int +nfs_nfsiodnew_sync(void) +{ + int error, i; + + mtx_assert(&nfs_iod_mtx, MA_OWNED); + for (i = 0; i < nfs_iodmax; i++) { + if (nfs_asyncdaemon[i] == 0) { + nfs_asyncdaemon[i] = 1; + break; + } + } + if (i == nfs_iodmax) + return (0); + mtx_unlock(&nfs_iod_mtx); + error = kproc_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, + RFHIGHPID, 0, "nfsiod %d", i); + mtx_lock(&nfs_iod_mtx); + if (error == 0) { + nfs_numasync++; + nfs_iodwant[i] = NFSIOD_AVAILABLE; + } else + nfs_asyncdaemon[i] = 0; + return (error); +} + +void +nfs_nfsiodnew_tq(__unused void *arg, int pending) +{ + + mtx_lock(&nfs_iod_mtx); + while (pending > 0) { + pending--; + nfs_nfsiodnew_sync(); + } + mtx_unlock(&nfs_iod_mtx); +} + +void +nfs_nfsiodnew(void) +{ + + mtx_assert(&nfs_iod_mtx, MA_OWNED); + taskqueue_enqueue(taskqueue_thread, &nfs_nfsiodnew_task); +} + +static void +nfsiod_setup(void *dummy) +{ + int error; + + TUNABLE_INT_FETCH("vfs.oldnfs.iodmin", &nfs_iodmin); + mtx_lock(&nfs_iod_mtx); + /* Silently limit the start number of nfsiod's */ + if (nfs_iodmin > NFS_MAXASYNCDAEMON) + nfs_iodmin = NFS_MAXASYNCDAEMON; + + while (nfs_numasync < nfs_iodmin) { + error = nfs_nfsiodnew_sync(); + if (error == -1) + panic("nfsiod_setup: nfs_nfsiodnew failed"); + } + mtx_unlock(&nfs_iod_mtx); +} +SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL); + +static int nfs_defect = 0; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, + "Allow nfsiods to migrate serving different mounts"); + +/* + * Asynchronous I/O daemons for client nfs. + * They do read-ahead and write-behind operations on the block I/O cache. + * Returns if we hit the timeout defined by the iodmaxidle sysctl. + */ +static void +nfssvc_iod(void *instance) +{ + struct buf *bp; + struct nfsmount *nmp; + int myiod, timo; + int error = 0; + + mtx_lock(&nfs_iod_mtx); + myiod = (int *)instance - nfs_asyncdaemon; + /* + * Main loop + */ + for (;;) { + while (((nmp = nfs_iodmount[myiod]) == NULL) + || !TAILQ_FIRST(&nmp->nm_bufq)) { + if (myiod >= nfs_iodmax) + goto finish; + if (nmp) + nmp->nm_bufqiods--; + if (nfs_iodwant[myiod] == NFSIOD_NOT_AVAILABLE) + nfs_iodwant[myiod] = NFSIOD_AVAILABLE; + nfs_iodmount[myiod] = NULL; + /* + * Always keep at least nfs_iodmin kthreads. + */ + timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz; + error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH, + "-", timo); + if (error) { + nmp = nfs_iodmount[myiod]; + /* + * Rechecking the nm_bufq closes a rare race where the + * nfsiod is woken up at the exact time the idle timeout + * fires + */ + if (nmp && TAILQ_FIRST(&nmp->nm_bufq)) + error = 0; + break; + } + } + if (error) + break; + while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { + int giant_locked = 0; + + /* Take one off the front of the list */ + TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); + nmp->nm_bufqlen--; + if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { + nmp->nm_bufqwant = 0; + wakeup(&nmp->nm_bufq); + } + mtx_unlock(&nfs_iod_mtx); + if (NFS_ISV4(bp->b_vp)) { + giant_locked = 1; + mtx_lock(&Giant); + } + if (bp->b_flags & B_DIRECT) { + KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set")); + (void)nfs_doio_directwrite(bp); + } else { + if (bp->b_iocmd == BIO_READ) + (void) nfs_doio(bp->b_vp, bp, bp->b_rcred, NULL); + else + (void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL); + } + if (giant_locked) + mtx_unlock(&Giant); + mtx_lock(&nfs_iod_mtx); + /* + * If there are more than one iod on this mount, then defect + * so that the iods can be shared out fairly between the mounts + */ + if (nfs_defect && nmp->nm_bufqiods > 1) { + NFS_DPF(ASYNCIO, + ("nfssvc_iod: iod %d defecting from mount %p\n", + myiod, nmp)); + nfs_iodmount[myiod] = NULL; + nmp->nm_bufqiods--; + break; + } + } + } +finish: + nfs_asyncdaemon[myiod] = 0; + if (nmp) + nmp->nm_bufqiods--; + nfs_iodwant[myiod] = NFSIOD_NOT_AVAILABLE; + nfs_iodmount[myiod] = NULL; + /* Someone may be waiting for the last nfsiod to terminate. */ + if (--nfs_numasync == 0) + wakeup(&nfs_numasync); + mtx_unlock(&nfs_iod_mtx); + if ((error == 0) || (error == EWOULDBLOCK)) + kproc_exit(0); + /* Abnormal termination */ + kproc_exit(1); +} diff --git a/sys/nfsclient/nfs_node.c b/sys/nfsclient/nfs_node.c new file mode 100644 index 0000000..afe3341 --- /dev/null +++ b/sys/nfsclient/nfs_node.c @@ -0,0 +1,280 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs_node.c 8.6 (Berkeley) 5/22/95 + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/fcntl.h> +#include <sys/fnv_hash.h> +#include <sys/lock.h> +#include <sys/malloc.h> +#include <sys/mbuf.h> +#include <sys/mount.h> +#include <sys/namei.h> +#include <sys/proc.h> +#include <sys/socket.h> +#include <sys/sysctl.h> +#include <sys/taskqueue.h> +#include <sys/vnode.h> + +#include <vm/uma.h> + +#include <nfs/nfsproto.h> +#include <nfs/nfs_lock.h> +#include <nfsclient/nfs.h> +#include <nfsclient/nfsnode.h> +#include <nfsclient/nfsmount.h> + +static uma_zone_t nfsnode_zone; + +static void nfs_freesillyrename(void *arg, __unused int pending); + +#define TRUE 1 +#define FALSE 0 + +void +nfs_nhinit(void) +{ + + nfsnode_zone = uma_zcreate("NFSNODE", sizeof(struct nfsnode), NULL, + NULL, NULL, NULL, UMA_ALIGN_PTR, 0); +} + +void +nfs_nhuninit(void) +{ + uma_zdestroy(nfsnode_zone); +} + +struct nfs_vncmp { + int fhsize; + void *fh; +}; + +static int +nfs_vncmpf(struct vnode *vp, void *arg) +{ + struct nfs_vncmp *a; + struct nfsnode *np; + + a = arg; + np = VTONFS(vp); + return (bcmp(a->fh, np->n_fhp, a->fhsize)); +} + +/* + * Look up a vnode/nfsnode by file handle. + * Callers must check for mount points!! + * In all cases, a pointer to a + * nfsnode structure is returned. + */ +int +nfs_nget(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp, int flags) +{ + struct thread *td = curthread; /* XXX */ + struct nfsnode *np; + struct vnode *vp; + struct vnode *nvp; + int error; + u_int hash; + struct nfsmount *nmp; + struct nfs_vncmp ncmp; + + nmp = VFSTONFS(mntp); + *npp = NULL; + + hash = fnv_32_buf(fhp->fh_bytes, fhsize, FNV1_32_INIT); + ncmp.fhsize = fhsize; + ncmp.fh = fhp; + + error = vfs_hash_get(mntp, hash, flags, + td, &nvp, nfs_vncmpf, &ncmp); + if (error) + return (error); + if (nvp != NULL) { + *npp = VTONFS(nvp); + return (0); + } + + /* + * Allocate before getnewvnode since doing so afterward + * might cause a bogus v_data pointer to get dereferenced + * elsewhere if zalloc should block. + */ + np = uma_zalloc(nfsnode_zone, M_WAITOK | M_ZERO); + + error = getnewvnode("nfs", mntp, &nfs_vnodeops, &nvp); + if (error) { + uma_zfree(nfsnode_zone, np); + return (error); + } + vp = nvp; + vp->v_bufobj.bo_ops = &buf_ops_nfs; + vp->v_data = np; + np->n_vnode = vp; + /* + * Initialize the mutex even if the vnode is going to be a loser. + * This simplifies the logic in reclaim, which can then unconditionally + * destroy the mutex (in the case of the loser, or if hash_insert happened + * to return an error no special casing is needed). + */ + mtx_init(&np->n_mtx, "NFSnode lock", NULL, MTX_DEF); + /* + * NFS supports recursive and shared locking. + */ + lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); + VN_LOCK_AREC(vp); + VN_LOCK_ASHARE(vp); + if (fhsize > NFS_SMALLFH) { + np->n_fhp = malloc(fhsize, M_NFSBIGFH, M_WAITOK); + } else + np->n_fhp = &np->n_fh; + bcopy((caddr_t)fhp, (caddr_t)np->n_fhp, fhsize); + np->n_fhsize = fhsize; + error = insmntque(vp, mntp); + if (error != 0) { + *npp = NULL; + if (np->n_fhsize > NFS_SMALLFH) { + free((caddr_t)np->n_fhp, M_NFSBIGFH); + } + mtx_destroy(&np->n_mtx); + uma_zfree(nfsnode_zone, np); + return (error); + } + error = vfs_hash_insert(vp, hash, flags, + td, &nvp, nfs_vncmpf, &ncmp); + if (error) + return (error); + if (nvp != NULL) { + *npp = VTONFS(nvp); + /* vfs_hash_insert() vput()'s the losing vnode */ + return (0); + } + *npp = np; + + return (0); +} + +/* + * Do the vrele(sp->s_dvp) as a separate task in order to avoid a + * deadlock because of a LOR when vrele() locks the directory vnode. + */ +static void +nfs_freesillyrename(void *arg, __unused int pending) +{ + struct sillyrename *sp; + + sp = arg; + vrele(sp->s_dvp); + free(sp, M_NFSREQ); +} + +int +nfs_inactive(struct vop_inactive_args *ap) +{ + struct nfsnode *np; + struct sillyrename *sp; + struct thread *td = curthread; /* XXX */ + + np = VTONFS(ap->a_vp); + mtx_lock(&np->n_mtx); + if (ap->a_vp->v_type != VDIR) { + sp = np->n_sillyrename; + np->n_sillyrename = NULL; + } else + sp = NULL; + if (sp) { + mtx_unlock(&np->n_mtx); + (void)nfs_vinvalbuf(ap->a_vp, 0, td, 1); + /* + * Remove the silly file that was rename'd earlier + */ + (sp->s_removeit)(sp); + crfree(sp->s_cred); + TASK_INIT(&sp->s_task, 0, nfs_freesillyrename, sp); + taskqueue_enqueue(taskqueue_thread, &sp->s_task); + mtx_lock(&np->n_mtx); + } + np->n_flag &= NMODIFIED; + mtx_unlock(&np->n_mtx); + return (0); +} + +/* + * Reclaim an nfsnode so that it can be used for other purposes. + */ +int +nfs_reclaim(struct vop_reclaim_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + struct nfsdmap *dp, *dp2; + + /* + * If the NLM is running, give it a chance to abort pending + * locks. + */ + if (nfs_reclaim_p) + nfs_reclaim_p(ap); + + /* + * Destroy the vm object and flush associated pages. + */ + vnode_destroy_vobject(vp); + + vfs_hash_remove(vp); + + /* + * Free up any directory cookie structures and + * large file handle structures that might be associated with + * this nfs node. + */ + if (vp->v_type == VDIR) { + dp = LIST_FIRST(&np->n_cookies); + while (dp) { + dp2 = dp; + dp = LIST_NEXT(dp, ndm_list); + free((caddr_t)dp2, M_NFSDIROFF); + } + } + if (np->n_fhsize > NFS_SMALLFH) { + free((caddr_t)np->n_fhp, M_NFSBIGFH); + } + mtx_destroy(&np->n_mtx); + uma_zfree(nfsnode_zone, vp->v_data); + vp->v_data = NULL; + return (0); +} diff --git a/sys/nfsclient/nfs_subs.c b/sys/nfsclient/nfs_subs.c new file mode 100644 index 0000000..d00a025 --- /dev/null +++ b/sys/nfsclient/nfs_subs.c @@ -0,0 +1,1147 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95 + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +/* + * These functions support the macros and help fiddle mbuf chains for + * the nfs op functions. They do things like create the rpc header and + * copy data between mbuf chains and uio lists. + */ + +#include "opt_kdtrace.h" + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/bio.h> +#include <sys/buf.h> +#include <sys/proc.h> +#include <sys/mount.h> +#include <sys/vnode.h> +#include <sys/namei.h> +#include <sys/mbuf.h> +#include <sys/socket.h> +#include <sys/stat.h> +#include <sys/malloc.h> +#include <sys/sysent.h> +#include <sys/syscall.h> +#include <sys/sysproto.h> +#include <sys/taskqueue.h> + +#include <vm/vm.h> +#include <vm/vm_object.h> +#include <vm/vm_extern.h> +#include <vm/uma.h> + +#include <nfs/nfsproto.h> +#include <nfsclient/nfs.h> +#include <nfsclient/nfsnode.h> +#include <nfs/nfs_kdtrace.h> +#include <nfs/xdr_subs.h> +#include <nfsclient/nfsm_subs.h> +#include <nfsclient/nfsmount.h> + +#include <netinet/in.h> + +/* + * Note that stdarg.h and the ANSI style va_start macro is used for both + * ANSI and traditional C compilers. + */ +#include <machine/stdarg.h> + +#ifdef KDTRACE_HOOKS +dtrace_nfsclient_attrcache_flush_probe_func_t + dtrace_nfsclient_attrcache_flush_done_probe; +uint32_t nfsclient_attrcache_flush_done_id; + +dtrace_nfsclient_attrcache_get_hit_probe_func_t + dtrace_nfsclient_attrcache_get_hit_probe; +uint32_t nfsclient_attrcache_get_hit_id; + +dtrace_nfsclient_attrcache_get_miss_probe_func_t + dtrace_nfsclient_attrcache_get_miss_probe; +uint32_t nfsclient_attrcache_get_miss_id; + +dtrace_nfsclient_attrcache_load_probe_func_t + dtrace_nfsclient_attrcache_load_done_probe; +uint32_t nfsclient_attrcache_load_done_id; +#endif /* !KDTRACE_HOOKS */ + +/* + * Data items converted to xdr at startup, since they are constant + * This is kinda hokey, but may save a little time doing byte swaps + */ +u_int32_t nfs_xdrneg1; +u_int32_t nfs_true, nfs_false; + +/* And other global data */ +static u_int32_t nfs_xid = 0; +static enum vtype nv2tov_type[8]= { + VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON +}; + +int nfs_ticks; +int nfs_pbuf_freecnt = -1; /* start out unlimited */ + +struct nfs_bufq nfs_bufq; +static struct mtx nfs_xid_mtx; +struct task nfs_nfsiodnew_task; + +/* + * and the reverse mapping from generic to Version 2 procedure numbers + */ +int nfsv2_procid[NFS_NPROCS] = { + NFSV2PROC_NULL, + NFSV2PROC_GETATTR, + NFSV2PROC_SETATTR, + NFSV2PROC_LOOKUP, + NFSV2PROC_NOOP, + NFSV2PROC_READLINK, + NFSV2PROC_READ, + NFSV2PROC_WRITE, + NFSV2PROC_CREATE, + NFSV2PROC_MKDIR, + NFSV2PROC_SYMLINK, + NFSV2PROC_CREATE, + NFSV2PROC_REMOVE, + NFSV2PROC_RMDIR, + NFSV2PROC_RENAME, + NFSV2PROC_LINK, + NFSV2PROC_READDIR, + NFSV2PROC_NOOP, + NFSV2PROC_STATFS, + NFSV2PROC_NOOP, + NFSV2PROC_NOOP, + NFSV2PROC_NOOP, + NFSV2PROC_NOOP, +}; + +LIST_HEAD(nfsnodehashhead, nfsnode); + +u_int32_t +nfs_xid_gen(void) +{ + uint32_t xid; + + mtx_lock(&nfs_xid_mtx); + + /* Get a pretty random xid to start with */ + if (!nfs_xid) + nfs_xid = random(); + /* + * Skip zero xid if it should ever happen. + */ + if (++nfs_xid == 0) + nfs_xid++; + xid = nfs_xid; + mtx_unlock(&nfs_xid_mtx); + return xid; +} + +/* + * Create the header for an rpc request packet + * The hsiz is the size of the rest of the nfs request header. + * (just used to decide if a cluster is a good idea) + */ +struct mbuf * +nfsm_reqhead(struct vnode *vp, u_long procid, int hsiz) +{ + struct mbuf *mb; + + MGET(mb, M_WAIT, MT_DATA); + if (hsiz >= MINCLSIZE) + MCLGET(mb, M_WAIT); + mb->m_len = 0; + return (mb); +} + +/* + * copies a uio scatter/gather list to an mbuf chain. + * NOTE: can ony handle iovcnt == 1 + */ +int +nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, caddr_t *bpos) +{ + char *uiocp; + struct mbuf *mp, *mp2; + int xfer, left, mlen; + int uiosiz, clflg, rem; + char *cp; + + KASSERT(uiop->uio_iovcnt == 1, ("nfsm_uiotombuf: iovcnt != 1")); + + if (siz > MLEN) /* or should it >= MCLBYTES ?? */ + clflg = 1; + else + clflg = 0; + rem = nfsm_rndup(siz)-siz; + mp = mp2 = *mq; + while (siz > 0) { + left = uiop->uio_iov->iov_len; + uiocp = uiop->uio_iov->iov_base; + if (left > siz) + left = siz; + uiosiz = left; + while (left > 0) { + mlen = M_TRAILINGSPACE(mp); + if (mlen == 0) { + MGET(mp, M_WAIT, MT_DATA); + if (clflg) + MCLGET(mp, M_WAIT); + mp->m_len = 0; + mp2->m_next = mp; + mp2 = mp; + mlen = M_TRAILINGSPACE(mp); + } + xfer = (left > mlen) ? mlen : left; +#ifdef notdef + /* Not Yet.. */ + if (uiop->uio_iov->iov_op != NULL) + (*(uiop->uio_iov->iov_op)) + (uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); + else +#endif + if (uiop->uio_segflg == UIO_SYSSPACE) + bcopy(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); + else + copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); + mp->m_len += xfer; + left -= xfer; + uiocp += xfer; + uiop->uio_offset += xfer; + uiop->uio_resid -= xfer; + } + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + uiosiz; + uiop->uio_iov->iov_len -= uiosiz; + siz -= uiosiz; + } + if (rem > 0) { + if (rem > M_TRAILINGSPACE(mp)) { + MGET(mp, M_WAIT, MT_DATA); + mp->m_len = 0; + mp2->m_next = mp; + } + cp = mtod(mp, caddr_t)+mp->m_len; + for (left = 0; left < rem; left++) + *cp++ = '\0'; + mp->m_len += rem; + *bpos = cp; + } else + *bpos = mtod(mp, caddr_t)+mp->m_len; + *mq = mp; + return (0); +} + +/* + * Copy a string into mbufs for the hard cases... + */ +int +nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz) +{ + struct mbuf *m1 = NULL, *m2; + long left, xfer, len, tlen; + u_int32_t *tl; + int putsize; + + putsize = 1; + m2 = *mb; + left = M_TRAILINGSPACE(m2); + if (left > 0) { + tl = ((u_int32_t *)(*bpos)); + *tl++ = txdr_unsigned(siz); + putsize = 0; + left -= NFSX_UNSIGNED; + m2->m_len += NFSX_UNSIGNED; + if (left > 0) { + bcopy(cp, (caddr_t) tl, left); + siz -= left; + cp += left; + m2->m_len += left; + left = 0; + } + } + /* Loop around adding mbufs */ + while (siz > 0) { + MGET(m1, M_WAIT, MT_DATA); + if (siz > MLEN) + MCLGET(m1, M_WAIT); + m1->m_len = NFSMSIZ(m1); + m2->m_next = m1; + m2 = m1; + tl = mtod(m1, u_int32_t *); + tlen = 0; + if (putsize) { + *tl++ = txdr_unsigned(siz); + m1->m_len -= NFSX_UNSIGNED; + tlen = NFSX_UNSIGNED; + putsize = 0; + } + if (siz < m1->m_len) { + len = nfsm_rndup(siz); + xfer = siz; + if (xfer < len) + *(tl+(xfer>>2)) = 0; + } else { + xfer = len = m1->m_len; + } + bcopy(cp, (caddr_t) tl, xfer); + m1->m_len = len+tlen; + siz -= xfer; + cp += xfer; + } + *mb = m1; + *bpos = mtod(m1, caddr_t)+m1->m_len; + return (0); +} + +/* + * Called once to initialize data structures... + */ +int +nfs_init(struct vfsconf *vfsp) +{ + int i; + + nfsmount_zone = uma_zcreate("NFSMOUNT", sizeof(struct nfsmount), + NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); + nfs_true = txdr_unsigned(TRUE); + nfs_false = txdr_unsigned(FALSE); + nfs_xdrneg1 = txdr_unsigned(-1); + nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; + if (nfs_ticks < 1) + nfs_ticks = 1; + /* Ensure async daemons disabled */ + for (i = 0; i < NFS_MAXASYNCDAEMON; i++) { + nfs_iodwant[i] = NFSIOD_NOT_AVAILABLE; + nfs_iodmount[i] = NULL; + } + nfs_nhinit(); /* Init the nfsnode table */ + + /* + * Initialize reply list and start timer + */ + mtx_init(&nfs_iod_mtx, "NFS iod lock", NULL, MTX_DEF); + mtx_init(&nfs_xid_mtx, "NFS xid lock", NULL, MTX_DEF); + TASK_INIT(&nfs_nfsiodnew_task, 0, nfs_nfsiodnew_tq, NULL); + + nfs_pbuf_freecnt = nswbuf / 2 + 1; + + return (0); +} + +int +nfs_uninit(struct vfsconf *vfsp) +{ + int i; + + /* + * Tell all nfsiod processes to exit. Clear nfs_iodmax, and wakeup + * any sleeping nfsiods so they check nfs_iodmax and exit. + * Drain nfsiodnew task before we wait for them to finish. + */ + mtx_lock(&nfs_iod_mtx); + nfs_iodmax = 0; + mtx_unlock(&nfs_iod_mtx); + taskqueue_drain(taskqueue_thread, &nfs_nfsiodnew_task); + mtx_lock(&nfs_iod_mtx); + for (i = 0; i < nfs_numasync; i++) + if (nfs_iodwant[i] == NFSIOD_AVAILABLE) + wakeup(&nfs_iodwant[i]); + /* The last nfsiod to exit will wake us up when nfs_numasync hits 0 */ + while (nfs_numasync) + msleep(&nfs_numasync, &nfs_iod_mtx, PWAIT, "ioddie", 0); + mtx_unlock(&nfs_iod_mtx); + nfs_nhuninit(); + uma_zdestroy(nfsmount_zone); + return (0); +} + +void +nfs_dircookie_lock(struct nfsnode *np) +{ + mtx_lock(&np->n_mtx); + while (np->n_flag & NDIRCOOKIELK) + (void) msleep(&np->n_flag, &np->n_mtx, PZERO, "nfsdirlk", 0); + np->n_flag |= NDIRCOOKIELK; + mtx_unlock(&np->n_mtx); +} + +void +nfs_dircookie_unlock(struct nfsnode *np) +{ + mtx_lock(&np->n_mtx); + np->n_flag &= ~NDIRCOOKIELK; + wakeup(&np->n_flag); + mtx_unlock(&np->n_mtx); +} + +int +nfs_upgrade_vnlock(struct vnode *vp) +{ + int old_lock; + + ASSERT_VOP_LOCKED(vp, "nfs_upgrade_vnlock"); + old_lock = VOP_ISLOCKED(vp); + if (old_lock != LK_EXCLUSIVE) { + KASSERT(old_lock == LK_SHARED, + ("nfs_upgrade_vnlock: wrong old_lock %d", old_lock)); + /* Upgrade to exclusive lock, this might block */ + vn_lock(vp, LK_UPGRADE | LK_RETRY); + } + return (old_lock); +} + +void +nfs_downgrade_vnlock(struct vnode *vp, int old_lock) +{ + if (old_lock != LK_EXCLUSIVE) { + KASSERT(old_lock == LK_SHARED, ("wrong old_lock %d", old_lock)); + /* Downgrade from exclusive lock. */ + vn_lock(vp, LK_DOWNGRADE | LK_RETRY); + } +} + +void +nfs_printf(const char *fmt, ...) +{ + va_list ap; + + mtx_lock(&Giant); + va_start(ap, fmt); + vprintf(fmt, ap); + va_end(ap); + mtx_unlock(&Giant); +} + +/* + * Attribute cache routines. + * nfs_loadattrcache() - loads or updates the cache contents from attributes + * that are on the mbuf list + * nfs_getattrcache() - returns valid attributes if found in cache, returns + * error otherwise + */ + +/* + * Load the attribute cache (that lives in the nfsnode entry) with + * the values on the mbuf list and + * Iff vap not NULL + * copy the attributes to *vaper + */ +int +nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp, + struct vattr *vaper, int dontshrink) +{ + struct vnode *vp = *vpp; + struct vattr *vap; + struct nfs_fattr *fp; + struct nfsnode *np = NULL; + int32_t t1; + caddr_t cp2; + int rdev; + struct mbuf *md; + enum vtype vtyp; + u_short vmode; + struct timespec mtime, mtime_save; + int v3 = NFS_ISV3(vp); + int error = 0; + + md = *mdp; + t1 = (mtod(md, caddr_t) + md->m_len) - *dposp; + cp2 = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, M_WAIT); + if (cp2 == NULL) { + error = EBADRPC; + goto out; + } + fp = (struct nfs_fattr *)cp2; + if (v3) { + vtyp = nfsv3tov_type(fp->fa_type); + vmode = fxdr_unsigned(u_short, fp->fa_mode); + rdev = makedev(fxdr_unsigned(int, fp->fa3_rdev.specdata1), + fxdr_unsigned(int, fp->fa3_rdev.specdata2)); + fxdr_nfsv3time(&fp->fa3_mtime, &mtime); + } else { + vtyp = nfsv2tov_type(fp->fa_type); + vmode = fxdr_unsigned(u_short, fp->fa_mode); + /* + * XXX + * + * The duplicate information returned in fa_type and fa_mode + * is an ambiguity in the NFS version 2 protocol. + * + * VREG should be taken literally as a regular file. If a + * server intents to return some type information differently + * in the upper bits of the mode field (e.g. for sockets, or + * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we + * leave the examination of the mode bits even in the VREG + * case to avoid breakage for bogus servers, but we make sure + * that there are actually type bits set in the upper part of + * fa_mode (and failing that, trust the va_type field). + * + * NFSv3 cleared the issue, and requires fa_mode to not + * contain any type information (while also introduing sockets + * and FIFOs for fa_type). + */ + if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0)) + vtyp = IFTOVT(vmode); + rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); + fxdr_nfsv2time(&fp->fa2_mtime, &mtime); + + /* + * Really ugly NFSv2 kludge. + */ + if (vtyp == VCHR && rdev == 0xffffffff) + vtyp = VFIFO; + } + + /* + * If v_type == VNON it is a new node, so fill in the v_type, + * n_mtime fields. Check to see if it represents a special + * device, and if so, check for a possible alias. Once the + * correct vnode has been obtained, fill in the rest of the + * information. + */ + np = VTONFS(vp); + mtx_lock(&np->n_mtx); + if (vp->v_type != vtyp) { + vp->v_type = vtyp; + if (vp->v_type == VFIFO) + vp->v_op = &nfs_fifoops; + np->n_mtime = mtime; + } + vap = &np->n_vattr; + vap->va_type = vtyp; + vap->va_mode = (vmode & 07777); + vap->va_rdev = rdev; + mtime_save = vap->va_mtime; + vap->va_mtime = mtime; + vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; + if (v3) { + vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); + vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid); + vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid); + vap->va_size = fxdr_hyper(&fp->fa3_size); + vap->va_blocksize = NFS_FABLKSIZE; + vap->va_bytes = fxdr_hyper(&fp->fa3_used); + vap->va_fileid = fxdr_unsigned(int32_t, + fp->fa3_fileid.nfsuquad[1]); + fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); + fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime); + vap->va_flags = 0; + vap->va_filerev = 0; + } else { + vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); + vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid); + vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid); + vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); + vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize); + vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks) + * NFS_FABLKSIZE; + vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); + fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); + vap->va_flags = 0; + vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t, + fp->fa2_ctime.nfsv2_sec); + vap->va_ctime.tv_nsec = 0; + vap->va_gen = fxdr_unsigned(u_int32_t, fp->fa2_ctime.nfsv2_usec); + vap->va_filerev = 0; + } + np->n_attrstamp = time_second; + if (vap->va_size != np->n_size) { + if (vap->va_type == VREG) { + if (dontshrink && vap->va_size < np->n_size) { + /* + * We've been told not to shrink the file; + * zero np->n_attrstamp to indicate that + * the attributes are stale. + */ + vap->va_size = np->n_size; + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + } else if (np->n_flag & NMODIFIED) { + /* + * We've modified the file: Use the larger + * of our size, and the server's size. + */ + if (vap->va_size < np->n_size) { + vap->va_size = np->n_size; + } else { + np->n_size = vap->va_size; + np->n_flag |= NSIZECHANGED; + } + } else { + np->n_size = vap->va_size; + np->n_flag |= NSIZECHANGED; + } + vnode_pager_setsize(vp, np->n_size); + } else { + np->n_size = vap->va_size; + } + } + /* + * The following checks are added to prevent a race between (say) + * a READDIR+ and a WRITE. + * READDIR+, WRITE requests sent out. + * READDIR+ resp, WRITE resp received on client. + * However, the WRITE resp was handled before the READDIR+ resp + * causing the post op attrs from the write to be loaded first + * and the attrs from the READDIR+ to be loaded later. If this + * happens, we have stale attrs loaded into the attrcache. + * We detect this by for the mtime moving back. We invalidate the + * attrcache when this happens. + */ + if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { + /* Size changed or mtime went backwards */ + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + } + if (vaper != NULL) { + bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap)); + if (np->n_flag & NCHG) { + if (np->n_flag & NACC) + vaper->va_atime = np->n_atim; + if (np->n_flag & NUPD) + vaper->va_mtime = np->n_mtim; + } + } + +#ifdef KDTRACE_HOOKS + if (np->n_attrstamp != 0) + KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, &np->n_vattr, 0); +#endif + mtx_unlock(&np->n_mtx); +out: +#ifdef KDTRACE_HOOKS + if (error) + KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, NULL, error); +#endif + return (error); +} + +#ifdef NFS_ACDEBUG +#include <sys/sysctl.h> +SYSCTL_DECL(_vfs_oldnfs); +static int nfs_acdebug; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, + "Toggle acdebug (attribute cache debug) flag"); +#endif + +/* + * Check the time stamp + * If the cache is valid, copy contents to *vap and return 0 + * otherwise return an error + */ +int +nfs_getattrcache(struct vnode *vp, struct vattr *vaper) +{ + struct nfsnode *np; + struct vattr *vap; + struct nfsmount *nmp; + int timeo; + + np = VTONFS(vp); + vap = &np->n_vattr; + nmp = VFSTONFS(vp->v_mount); +#ifdef NFS_ACDEBUG + mtx_lock(&Giant); /* nfs_printf() */ +#endif + mtx_lock(&np->n_mtx); + /* XXX n_mtime doesn't seem to be updated on a miss-and-reload */ + timeo = (time_second - np->n_mtime.tv_sec) / 10; + +#ifdef NFS_ACDEBUG + if (nfs_acdebug>1) + nfs_printf("nfs_getattrcache: initial timeo = %d\n", timeo); +#endif + + if (vap->va_type == VDIR) { + if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin) + timeo = nmp->nm_acdirmin; + else if (timeo > nmp->nm_acdirmax) + timeo = nmp->nm_acdirmax; + } else { + if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin) + timeo = nmp->nm_acregmin; + else if (timeo > nmp->nm_acregmax) + timeo = nmp->nm_acregmax; + } + +#ifdef NFS_ACDEBUG + if (nfs_acdebug > 2) + nfs_printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n", + nmp->nm_acregmin, nmp->nm_acregmax, + nmp->nm_acdirmin, nmp->nm_acdirmax); + + if (nfs_acdebug) + nfs_printf("nfs_getattrcache: age = %d; final timeo = %d\n", + (time_second - np->n_attrstamp), timeo); +#endif + + if ((time_second - np->n_attrstamp) >= timeo) { + nfsstats.attrcache_misses++; + mtx_unlock(&np->n_mtx); +#ifdef NFS_ACDEBUG + mtx_unlock(&Giant); /* nfs_printf() */ +#endif + KDTRACE_NFS_ATTRCACHE_GET_MISS(vp); + return (ENOENT); + } + nfsstats.attrcache_hits++; + if (vap->va_size != np->n_size) { + if (vap->va_type == VREG) { + if (np->n_flag & NMODIFIED) { + if (vap->va_size < np->n_size) + vap->va_size = np->n_size; + else + np->n_size = vap->va_size; + } else { + np->n_size = vap->va_size; + } + vnode_pager_setsize(vp, np->n_size); + } else { + np->n_size = vap->va_size; + } + } + bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr)); + if (np->n_flag & NCHG) { + if (np->n_flag & NACC) + vaper->va_atime = np->n_atim; + if (np->n_flag & NUPD) + vaper->va_mtime = np->n_mtim; + } + mtx_unlock(&np->n_mtx); +#ifdef NFS_ACDEBUG + mtx_unlock(&Giant); /* nfs_printf() */ +#endif + KDTRACE_NFS_ATTRCACHE_GET_HIT(vp, vap); + return (0); +} + +/* + * Purge all cached information about an NFS vnode including name + * cache entries, the attribute cache, and the access cache. This is + * called when an NFS request for a node fails with a stale + * filehandle. + */ +void +nfs_purgecache(struct vnode *vp) +{ + struct nfsnode *np; + int i; + + np = VTONFS(vp); + cache_purge(vp); + mtx_lock(&np->n_mtx); + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + for (i = 0; i < NFS_ACCESSCACHESIZE; i++) + np->n_accesscache[i].stamp = 0; + KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); + mtx_unlock(&np->n_mtx); +} + +static nfsuint64 nfs_nullcookie = { { 0, 0 } }; +/* + * This function finds the directory cookie that corresponds to the + * logical byte offset given. + */ +nfsuint64 * +nfs_getcookie(struct nfsnode *np, off_t off, int add) +{ + struct nfsdmap *dp, *dp2; + int pos; + nfsuint64 *retval = NULL; + + pos = (uoff_t)off / NFS_DIRBLKSIZ; + if (pos == 0 || off < 0) { + KASSERT(!add, ("nfs getcookie add at <= 0")); + return (&nfs_nullcookie); + } + pos--; + dp = LIST_FIRST(&np->n_cookies); + if (!dp) { + if (add) { + dp = malloc(sizeof (struct nfsdmap), + M_NFSDIROFF, M_WAITOK); + dp->ndm_eocookie = 0; + LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list); + } else + goto out; + } + while (pos >= NFSNUMCOOKIES) { + pos -= NFSNUMCOOKIES; + if (LIST_NEXT(dp, ndm_list)) { + if (!add && dp->ndm_eocookie < NFSNUMCOOKIES && + pos >= dp->ndm_eocookie) + goto out; + dp = LIST_NEXT(dp, ndm_list); + } else if (add) { + dp2 = malloc(sizeof (struct nfsdmap), + M_NFSDIROFF, M_WAITOK); + dp2->ndm_eocookie = 0; + LIST_INSERT_AFTER(dp, dp2, ndm_list); + dp = dp2; + } else + goto out; + } + if (pos >= dp->ndm_eocookie) { + if (add) + dp->ndm_eocookie = pos + 1; + else + goto out; + } + retval = &dp->ndm_cookies[pos]; +out: + return (retval); +} + +/* + * Invalidate cached directory information, except for the actual directory + * blocks (which are invalidated separately). + * Done mainly to avoid the use of stale offset cookies. + */ +void +nfs_invaldir(struct vnode *vp) +{ + struct nfsnode *np = VTONFS(vp); + + KASSERT(vp->v_type == VDIR, ("nfs: invaldir not dir")); + nfs_dircookie_lock(np); + np->n_direofoffset = 0; + np->n_cookieverf.nfsuquad[0] = 0; + np->n_cookieverf.nfsuquad[1] = 0; + if (LIST_FIRST(&np->n_cookies)) + LIST_FIRST(&np->n_cookies)->ndm_eocookie = 0; + nfs_dircookie_unlock(np); +} + +/* + * The write verifier has changed (probably due to a server reboot), so all + * B_NEEDCOMMIT blocks will have to be written again. Since they are on the + * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT + * and B_CLUSTEROK flags. Once done the new write verifier can be set for the + * mount point. + * + * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data + * writes are not clusterable. + */ +void +nfs_clearcommit(struct mount *mp) +{ + struct vnode *vp, *nvp; + struct buf *bp, *nbp; + struct bufobj *bo; + + MNT_ILOCK(mp); + MNT_VNODE_FOREACH(vp, mp, nvp) { + bo = &vp->v_bufobj; + VI_LOCK(vp); + if (vp->v_iflag & VI_DOOMED) { + VI_UNLOCK(vp); + continue; + } + vholdl(vp); + VI_UNLOCK(vp); + MNT_IUNLOCK(mp); + BO_LOCK(bo); + TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { + if (!BUF_ISLOCKED(bp) && + (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) + == (B_DELWRI | B_NEEDCOMMIT)) + bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); + } + BO_UNLOCK(bo); + vdrop(vp); + MNT_ILOCK(mp); + } + MNT_IUNLOCK(mp); +} + +/* + * Helper functions for former macros. Some of these should be + * moved to their callers. + */ + +int +nfsm_mtofh_xx(struct vnode *d, struct vnode **v, int v3, int *f, + struct mbuf **md, caddr_t *dpos) +{ + struct nfsnode *ttnp; + struct vnode *ttvp; + nfsfh_t *ttfhp; + u_int32_t *tl; + int ttfhsize; + int t1; + + if (v3) { + tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); + if (tl == NULL) + return EBADRPC; + *f = fxdr_unsigned(int, *tl); + } else + *f = 1; + if (*f) { + t1 = nfsm_getfh_xx(&ttfhp, &ttfhsize, (v3), md, dpos); + if (t1 != 0) + return t1; + t1 = nfs_nget(d->v_mount, ttfhp, ttfhsize, &ttnp, LK_EXCLUSIVE); + if (t1 != 0) + return t1; + *v = NFSTOV(ttnp); + } + if (v3) { + tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); + if (tl == NULL) + return EBADRPC; + if (*f) + *f = fxdr_unsigned(int, *tl); + else if (fxdr_unsigned(int, *tl)) + nfsm_adv_xx(NFSX_V3FATTR, md, dpos); + } + if (*f) { + ttvp = *v; + t1 = nfs_loadattrcache(&ttvp, md, dpos, NULL, 0); + if (t1) + return t1; + *v = ttvp; + } + return 0; +} + +int +nfsm_getfh_xx(nfsfh_t **f, int *s, int v3, struct mbuf **md, caddr_t *dpos) +{ + u_int32_t *tl; + + if (v3) { + tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); + if (tl == NULL) + return EBADRPC; + *s = fxdr_unsigned(int, *tl); + if (*s <= 0 || *s > NFSX_V3FHMAX) + return EBADRPC; + } else + *s = NFSX_V2FH; + *f = nfsm_dissect_xx(nfsm_rndup(*s), md, dpos); + if (*f == NULL) + return EBADRPC; + else + return 0; +} + + +int +nfsm_loadattr_xx(struct vnode **v, struct vattr *va, struct mbuf **md, + caddr_t *dpos) +{ + int t1; + + struct vnode *ttvp = *v; + t1 = nfs_loadattrcache(&ttvp, md, dpos, va, 0); + if (t1 != 0) + return t1; + *v = ttvp; + return 0; +} + +int +nfsm_postop_attr_xx(struct vnode **v, int *f, struct vattr *va, + struct mbuf **md, caddr_t *dpos) +{ + u_int32_t *tl; + int t1; + + struct vnode *ttvp = *v; + tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); + if (tl == NULL) + return EBADRPC; + *f = fxdr_unsigned(int, *tl); + if (*f != 0) { + t1 = nfs_loadattrcache(&ttvp, md, dpos, va, 1); + if (t1 != 0) { + *f = 0; + return t1; + } + *v = ttvp; + } + return 0; +} + +int +nfsm_wcc_data_xx(struct vnode **v, int *f, struct mbuf **md, caddr_t *dpos) +{ + u_int32_t *tl; + int ttattrf, ttretf = 0; + int t1; + + tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos); + if (tl == NULL) + return EBADRPC; + if (*tl == nfs_true) { + tl = nfsm_dissect_xx(6 * NFSX_UNSIGNED, md, dpos); + if (tl == NULL) + return EBADRPC; + mtx_lock(&(VTONFS(*v))->n_mtx); + if (*f) + ttretf = (VTONFS(*v)->n_mtime.tv_sec == fxdr_unsigned(u_int32_t, *(tl + 2)) && + VTONFS(*v)->n_mtime.tv_nsec == fxdr_unsigned(u_int32_t, *(tl + 3))); + mtx_unlock(&(VTONFS(*v))->n_mtx); + } + t1 = nfsm_postop_attr_xx(v, &ttattrf, NULL, md, dpos); + if (t1) + return t1; + if (*f) + *f = ttretf; + else + *f = ttattrf; + return 0; +} + +int +nfsm_strtom_xx(const char *a, int s, int m, struct mbuf **mb, caddr_t *bpos) +{ + u_int32_t *tl; + int t1; + + if (s > m) + return ENAMETOOLONG; + t1 = nfsm_rndup(s) + NFSX_UNSIGNED; + if (t1 <= M_TRAILINGSPACE(*mb)) { + tl = nfsm_build_xx(t1, mb, bpos); + *tl++ = txdr_unsigned(s); + *(tl + ((t1 >> 2) - 2)) = 0; + bcopy(a, tl, s); + } else { + t1 = nfsm_strtmbuf(mb, bpos, a, s); + if (t1 != 0) + return t1; + } + return 0; +} + +int +nfsm_fhtom_xx(struct vnode *v, int v3, struct mbuf **mb, caddr_t *bpos) +{ + u_int32_t *tl; + int t1; + caddr_t cp; + + if (v3) { + t1 = nfsm_rndup(VTONFS(v)->n_fhsize) + NFSX_UNSIGNED; + if (t1 < M_TRAILINGSPACE(*mb)) { + tl = nfsm_build_xx(t1, mb, bpos); + *tl++ = txdr_unsigned(VTONFS(v)->n_fhsize); + *(tl + ((t1 >> 2) - 2)) = 0; + bcopy(VTONFS(v)->n_fhp, tl, VTONFS(v)->n_fhsize); + } else { + t1 = nfsm_strtmbuf(mb, bpos, + (const char *)VTONFS(v)->n_fhp, + VTONFS(v)->n_fhsize); + if (t1 != 0) + return t1; + } + } else { + cp = nfsm_build_xx(NFSX_V2FH, mb, bpos); + bcopy(VTONFS(v)->n_fhp, cp, NFSX_V2FH); + } + return 0; +} + +void +nfsm_v3attrbuild_xx(struct vattr *va, int full, struct mbuf **mb, + caddr_t *bpos) +{ + u_int32_t *tl; + + if (va->va_mode != (mode_t)VNOVAL) { + tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos); + *tl++ = nfs_true; + *tl = txdr_unsigned(va->va_mode); + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = nfs_false; + } + if (full && va->va_uid != (uid_t)VNOVAL) { + tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos); + *tl++ = nfs_true; + *tl = txdr_unsigned(va->va_uid); + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = nfs_false; + } + if (full && va->va_gid != (gid_t)VNOVAL) { + tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos); + *tl++ = nfs_true; + *tl = txdr_unsigned(va->va_gid); + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = nfs_false; + } + if (full && va->va_size != VNOVAL) { + tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos); + *tl++ = nfs_true; + txdr_hyper(va->va_size, tl); + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = nfs_false; + } + if (va->va_atime.tv_sec != VNOVAL) { + if (va->va_atime.tv_sec != time_second) { + tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos); + *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); + txdr_nfsv3time(&va->va_atime, tl); + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); + } + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); + } + if (va->va_mtime.tv_sec != VNOVAL) { + if (va->va_mtime.tv_sec != time_second) { + tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos); + *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); + txdr_nfsv3time(&va->va_mtime, tl); + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); + } + } else { + tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos); + *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); + } +} diff --git a/sys/nfsclient/nfs_vfsops.c b/sys/nfsclient/nfs_vfsops.c new file mode 100644 index 0000000..58384e8 --- /dev/null +++ b/sys/nfsclient/nfs_vfsops.c @@ -0,0 +1,1579 @@ +/*- + * Copyright (c) 1989, 1993, 1995 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs_vfsops.c 8.12 (Berkeley) 5/20/95 + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + + +#include "opt_bootp.h" +#include "opt_nfsroot.h" + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/bio.h> +#include <sys/buf.h> +#include <sys/jail.h> +#include <sys/limits.h> +#include <sys/lock.h> +#include <sys/malloc.h> +#include <sys/mbuf.h> +#include <sys/module.h> +#include <sys/mount.h> +#include <sys/proc.h> +#include <sys/socket.h> +#include <sys/socketvar.h> +#include <sys/sockio.h> +#include <sys/sysctl.h> +#include <sys/syslog.h> +#include <sys/vnode.h> +#include <sys/signalvar.h> + +#include <vm/vm.h> +#include <vm/vm_extern.h> +#include <vm/uma.h> + +#include <net/if.h> +#include <net/route.h> +#include <net/vnet.h> + +#include <netinet/in.h> + +#include <rpc/rpc.h> + +#include <nfs/nfsproto.h> +#include <nfsclient/nfs.h> +#include <nfsclient/nfsnode.h> +#include <nfsclient/nfsmount.h> +#include <nfs/xdr_subs.h> +#include <nfsclient/nfsm_subs.h> +#include <nfs/nfsdiskless.h> + +FEATURE(nfsclient, "NFS client"); + +MALLOC_DEFINE(M_NFSREQ, "nfsclient_req", "NFS request header"); +MALLOC_DEFINE(M_NFSBIGFH, "nfsclient_bigfh", "NFS version 3 file handle"); +MALLOC_DEFINE(M_NFSDIROFF, "nfsclient_diroff", "NFS directory offset data"); +MALLOC_DEFINE(M_NFSHASH, "nfsclient_hash", "NFS hash tables"); +MALLOC_DEFINE(M_NFSDIRECTIO, "nfsclient_directio", "NFS Direct IO async write state"); + +uma_zone_t nfsmount_zone; + +struct nfsstats nfsstats; + +SYSCTL_NODE(_vfs, OID_AUTO, oldnfs, CTLFLAG_RW, 0, "Old NFS filesystem"); +SYSCTL_STRUCT(_vfs_oldnfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RW, + &nfsstats, nfsstats, "S,nfsstats"); +static int nfs_ip_paranoia = 1; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, + &nfs_ip_paranoia, 0, + "Disallow accepting replies from IPs which differ from those sent"); +#ifdef NFS_DEBUG +int nfs_debug; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, + "Toggle debug flag"); +#endif +static int nfs_tprintf_initial_delay = NFS_TPRINTF_INITIAL_DELAY; +SYSCTL_INT(_vfs_oldnfs, NFS_TPRINTF_INITIAL_DELAY, + downdelayinitial, CTLFLAG_RW, &nfs_tprintf_initial_delay, 0, + "Delay before printing \"nfs server not responding\" messages"); +/* how long between console messages "nfs server foo not responding" */ +static int nfs_tprintf_delay = NFS_TPRINTF_DELAY; +SYSCTL_INT(_vfs_oldnfs, NFS_TPRINTF_DELAY, + downdelayinterval, CTLFLAG_RW, &nfs_tprintf_delay, 0, + "Delay between printing \"nfs server not responding\" messages"); + +static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp, + struct nfs_args *argp, const char *hostname); +static int mountnfs(struct nfs_args *, struct mount *, + struct sockaddr *, char *, struct vnode **, + struct ucred *cred, int, int); +static void nfs_getnlminfo(struct vnode *, uint8_t *, size_t *, + struct sockaddr_storage *, int *, off_t *, + struct timeval *); +static vfs_mount_t nfs_mount; +static vfs_cmount_t nfs_cmount; +static vfs_unmount_t nfs_unmount; +static vfs_root_t nfs_root; +static vfs_statfs_t nfs_statfs; +static vfs_sync_t nfs_sync; +static vfs_sysctl_t nfs_sysctl; + +static int fake_wchan; + +/* + * nfs vfs operations. + */ +static struct vfsops nfs_vfsops = { + .vfs_init = nfs_init, + .vfs_mount = nfs_mount, + .vfs_cmount = nfs_cmount, + .vfs_root = nfs_root, + .vfs_statfs = nfs_statfs, + .vfs_sync = nfs_sync, + .vfs_uninit = nfs_uninit, + .vfs_unmount = nfs_unmount, + .vfs_sysctl = nfs_sysctl, +}; +VFS_SET(nfs_vfsops, oldnfs, VFCF_NETWORK); + +/* So that loader and kldload(2) can find us, wherever we are.. */ +MODULE_VERSION(oldnfs, 1); +MODULE_DEPEND(oldnfs, krpc, 1, 1, 1); +#ifdef KGSSAPI +MODULE_DEPEND(oldnfs, kgssapi, 1, 1, 1); +#endif +MODULE_DEPEND(oldnfs, nfs_common, 1, 1, 1); +MODULE_DEPEND(oldnfs, nfslock, 1, 1, 1); + +static struct nfs_rpcops nfs_rpcops = { + nfs_readrpc, + nfs_writerpc, + nfs_writebp, + nfs_readlinkrpc, + nfs_invaldir, + nfs_commit, +}; + +/* + * This structure is now defined in sys/nfs/nfs_diskless.c so that it + * can be shared by both NFS clients. It is declared here so that it + * will be defined for kernels built without NFS_ROOT, although it + * isn't used in that case. + */ +#ifndef NFS_ROOT +struct nfs_diskless nfs_diskless = { { { 0 } } }; +struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; +int nfs_diskless_valid = 0; +#endif + +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, diskless_valid, CTLFLAG_RD, + &nfs_diskless_valid, 0, + "Has the diskless struct been filled correctly"); + +SYSCTL_STRING(_vfs_oldnfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, + nfsv3_diskless.root_hostnam, 0, "Path to nfs root"); + +SYSCTL_OPAQUE(_vfs_oldnfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, + &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr, + "%Ssockaddr_in", "Diskless root nfs address"); + + +void nfsargs_ntoh(struct nfs_args *); +static int nfs_mountdiskless(char *, + struct sockaddr_in *, struct nfs_args *, + struct thread *, struct vnode **, struct mount *); +static void nfs_convert_diskless(void); +static void nfs_convert_oargs(struct nfs_args *args, + struct onfs_args *oargs); + +int +nfs_iosize(struct nfsmount *nmp) +{ + int iosize; + + /* + * Calculate the size used for io buffers. Use the larger + * of the two sizes to minimise nfs requests but make sure + * that it is at least one VM page to avoid wasting buffer + * space. + */ + iosize = imax(nmp->nm_rsize, nmp->nm_wsize); + iosize = imax(iosize, PAGE_SIZE); + return (iosize); +} + +static void +nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs) +{ + + args->version = NFS_ARGSVERSION; + args->addr = oargs->addr; + args->addrlen = oargs->addrlen; + args->sotype = oargs->sotype; + args->proto = oargs->proto; + args->fh = oargs->fh; + args->fhsize = oargs->fhsize; + args->flags = oargs->flags; + args->wsize = oargs->wsize; + args->rsize = oargs->rsize; + args->readdirsize = oargs->readdirsize; + args->timeo = oargs->timeo; + args->retrans = oargs->retrans; + args->maxgrouplist = oargs->maxgrouplist; + args->readahead = oargs->readahead; + args->deadthresh = oargs->deadthresh; + args->hostname = oargs->hostname; +} + +static void +nfs_convert_diskless(void) +{ + + bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, + sizeof(struct ifaliasreq)); + bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, + sizeof(struct sockaddr_in)); + nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args); + if (nfsv3_diskless.root_args.flags & NFSMNT_NFSV3) { + nfsv3_diskless.root_fhsize = NFSX_V3FH; + bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V3FH); + } else { + nfsv3_diskless.root_fhsize = NFSX_V2FH; + bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH); + } + bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, + sizeof(struct sockaddr_in)); + bcopy(nfs_diskless.root_hostnam, nfsv3_diskless.root_hostnam, MNAMELEN); + nfsv3_diskless.root_time = nfs_diskless.root_time; + bcopy(nfs_diskless.my_hostnam, nfsv3_diskless.my_hostnam, + MAXHOSTNAMELEN); + nfs_diskless_valid = 3; +} + +/* + * nfs statfs call + */ +static int +nfs_statfs(struct mount *mp, struct statfs *sbp) +{ + struct vnode *vp; + struct thread *td; + struct nfs_statfs *sfp; + caddr_t bpos, dpos; + struct nfsmount *nmp = VFSTONFS(mp); + int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr; + struct mbuf *mreq, *mrep, *md, *mb; + struct nfsnode *np; + u_quad_t tquad; + + td = curthread; +#ifndef nolint + sfp = NULL; +#endif + error = vfs_busy(mp, MBF_NOWAIT); + if (error) + return (error); + error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE); + if (error) { + vfs_unbusy(mp); + return (error); + } + vp = NFSTOV(np); + mtx_lock(&nmp->nm_mtx); + if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { + mtx_unlock(&nmp->nm_mtx); + (void)nfs_fsinfo(nmp, vp, td->td_ucred, td); + } else + mtx_unlock(&nmp->nm_mtx); + nfsstats.rpccnt[NFSPROC_FSSTAT]++; + mreq = nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + nfsm_request(vp, NFSPROC_FSSTAT, td, td->td_ucred); + if (v3) + nfsm_postop_attr(vp, retattr); + if (error) { + if (mrep != NULL) + m_freem(mrep); + goto nfsmout; + } + sfp = nfsm_dissect(struct nfs_statfs *, NFSX_STATFS(v3)); + mtx_lock(&nmp->nm_mtx); + sbp->f_iosize = nfs_iosize(nmp); + mtx_unlock(&nmp->nm_mtx); + if (v3) { + sbp->f_bsize = NFS_FABLKSIZE; + tquad = fxdr_hyper(&sfp->sf_tbytes); + sbp->f_blocks = tquad / NFS_FABLKSIZE; + tquad = fxdr_hyper(&sfp->sf_fbytes); + sbp->f_bfree = tquad / NFS_FABLKSIZE; + tquad = fxdr_hyper(&sfp->sf_abytes); + sbp->f_bavail = tquad / NFS_FABLKSIZE; + sbp->f_files = (fxdr_unsigned(int32_t, + sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff); + sbp->f_ffree = (fxdr_unsigned(int32_t, + sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff); + } else { + sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); + sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); + sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); + sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); + sbp->f_files = 0; + sbp->f_ffree = 0; + } + m_freem(mrep); +nfsmout: + vput(vp); + vfs_unbusy(mp); + return (error); +} + +/* + * nfs version 3 fsinfo rpc call + */ +int +nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred, + struct thread *td) +{ + struct nfsv3_fsinfo *fsp; + u_int32_t pref, max; + caddr_t bpos, dpos; + int error = 0, retattr; + struct mbuf *mreq, *mrep, *md, *mb; + u_int64_t maxfsize; + + nfsstats.rpccnt[NFSPROC_FSINFO]++; + mreq = nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, 1); + nfsm_request(vp, NFSPROC_FSINFO, td, cred); + nfsm_postop_attr(vp, retattr); + if (!error) { + fsp = nfsm_dissect(struct nfsv3_fsinfo *, NFSX_V3FSINFO); + pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); + mtx_lock(&nmp->nm_mtx); + if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) + nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & + ~(NFS_FABLKSIZE - 1); + max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); + if (max < nmp->nm_wsize && max > 0) { + nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); + if (nmp->nm_wsize == 0) + nmp->nm_wsize = max; + } + pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); + if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) + nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & + ~(NFS_FABLKSIZE - 1); + max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); + if (max < nmp->nm_rsize && max > 0) { + nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); + if (nmp->nm_rsize == 0) + nmp->nm_rsize = max; + } + pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); + if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ) + nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) & + ~(NFS_DIRBLKSIZ - 1); + if (max < nmp->nm_readdirsize && max > 0) { + nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1); + if (nmp->nm_readdirsize == 0) + nmp->nm_readdirsize = max; + } + maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); + if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) + nmp->nm_maxfilesize = maxfsize; + nmp->nm_mountp->mnt_stat.f_iosize = nfs_iosize(nmp); + nmp->nm_state |= NFSSTA_GOTFSINFO; + mtx_unlock(&nmp->nm_mtx); + } + m_freem(mrep); +nfsmout: + return (error); +} + +/* + * Mount a remote root fs via. nfs. This depends on the info in the + * nfs_diskless structure that has been filled in properly by some primary + * bootstrap. + * It goes something like this: + * - do enough of "ifconfig" by calling ifioctl() so that the system + * can talk to the server + * - If nfs_diskless.mygateway is filled in, use that address as + * a default gateway. + * - build the rootfs mount point and call mountnfs() to do the rest. + * + * It is assumed to be safe to read, modify, and write the nfsv3_diskless + * structure, as well as other global NFS client variables here, as + * nfs_mountroot() will be called once in the boot before any other NFS + * client activity occurs. + */ +int +nfs_mountroot(struct mount *mp) +{ + struct thread *td = curthread; + struct nfsv3_diskless *nd = &nfsv3_diskless; + struct socket *so; + struct vnode *vp; + struct ifreq ir; + int error; + u_long l; + char buf[128]; + char *cp; + + +#if defined(BOOTP_NFSROOT) && defined(BOOTP) + bootpc_init(); /* use bootp to get nfs_diskless filled in */ +#elif defined(NFS_ROOT) + nfs_setup_diskless(); +#endif + + if (nfs_diskless_valid == 0) { + return (-1); + } + if (nfs_diskless_valid == 1) + nfs_convert_diskless(); + + /* + * XXX splnet, so networks will receive... + */ + splnet(); + + /* + * Do enough of ifconfig(8) so that the critical net interface can + * talk to the server. + */ + error = socreate(nd->myif.ifra_addr.sa_family, &so, nd->root_args.sotype, 0, + td->td_ucred, td); + if (error) + panic("nfs_mountroot: socreate(%04x): %d", + nd->myif.ifra_addr.sa_family, error); + +#if 0 /* XXX Bad idea */ + /* + * We might not have been told the right interface, so we pass + * over the first ten interfaces of the same kind, until we get + * one of them configured. + */ + + for (i = strlen(nd->myif.ifra_name) - 1; + nd->myif.ifra_name[i] >= '0' && + nd->myif.ifra_name[i] <= '9'; + nd->myif.ifra_name[i] ++) { + error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); + if(!error) + break; + } +#endif + + error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); + if (error) + panic("nfs_mountroot: SIOCAIFADDR: %d", error); + + if ((cp = getenv("boot.netif.mtu")) != NULL) { + ir.ifr_mtu = strtol(cp, NULL, 10); + bcopy(nd->myif.ifra_name, ir.ifr_name, IFNAMSIZ); + freeenv(cp); + error = ifioctl(so, SIOCSIFMTU, (caddr_t)&ir, td); + if (error) + printf("nfs_mountroot: SIOCSIFMTU: %d", error); + } + soclose(so); + + /* + * If the gateway field is filled in, set it as the default route. + * Note that pxeboot will set a default route of 0 if the route + * is not set by the DHCP server. Check also for a value of 0 + * to avoid panicking inappropriately in that situation. + */ + if (nd->mygateway.sin_len != 0 && + nd->mygateway.sin_addr.s_addr != 0) { + struct sockaddr_in mask, sin; + + bzero((caddr_t)&mask, sizeof(mask)); + sin = mask; + sin.sin_family = AF_INET; + sin.sin_len = sizeof(sin); + /* XXX MRT use table 0 for this sort of thing */ + CURVNET_SET(TD_TO_VNET(td)); + error = rtrequest(RTM_ADD, (struct sockaddr *)&sin, + (struct sockaddr *)&nd->mygateway, + (struct sockaddr *)&mask, + RTF_UP | RTF_GATEWAY, NULL); + CURVNET_RESTORE(); + if (error) + panic("nfs_mountroot: RTM_ADD: %d", error); + } + + /* + * Create the rootfs mount point. + */ + nd->root_args.fh = nd->root_fh; + nd->root_args.fhsize = nd->root_fhsize; + l = ntohl(nd->root_saddr.sin_addr.s_addr); + snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", + (l >> 24) & 0xff, (l >> 16) & 0xff, + (l >> 8) & 0xff, (l >> 0) & 0xff, nd->root_hostnam); + printf("NFS ROOT: %s\n", buf); + nd->root_args.hostname = buf; + if ((error = nfs_mountdiskless(buf, + &nd->root_saddr, &nd->root_args, td, &vp, mp)) != 0) { + return (error); + } + + /* + * This is not really an nfs issue, but it is much easier to + * set hostname here and then let the "/etc/rc.xxx" files + * mount the right /var based upon its preset value. + */ + mtx_lock(&prison0.pr_mtx); + strlcpy(prison0.pr_hostname, nd->my_hostnam, + sizeof (prison0.pr_hostname)); + mtx_unlock(&prison0.pr_mtx); + inittodr(ntohl(nd->root_time)); + return (0); +} + +/* + * Internal version of mount system call for diskless setup. + */ +static int +nfs_mountdiskless(char *path, + struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, + struct vnode **vpp, struct mount *mp) +{ + struct sockaddr *nam; + int error; + + nam = sodupsockaddr((struct sockaddr *)sin, M_WAITOK); + if ((error = mountnfs(args, mp, nam, path, vpp, td->td_ucred, + NFS_DEFAULT_NAMETIMEO, NFS_DEFAULT_NEGNAMETIMEO)) != 0) { + printf("nfs_mountroot: mount %s on /: %d\n", path, error); + return (error); + } + return (0); +} + +static int +nfs_sec_name_to_num(char *sec) +{ + if (!strcmp(sec, "krb5")) + return (RPCSEC_GSS_KRB5); + if (!strcmp(sec, "krb5i")) + return (RPCSEC_GSS_KRB5I); + if (!strcmp(sec, "krb5p")) + return (RPCSEC_GSS_KRB5P); + if (!strcmp(sec, "sys")) + return (AUTH_SYS); + /* + * Userland should validate the string but we will try and + * cope with unexpected values. + */ + return (AUTH_SYS); +} + +static void +nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp, + const char *hostname) +{ + int s; + int adjsock; + int maxio; + char *p; + char *secname; + char *principal; + + s = splnet(); + + /* + * Set read-only flag if requested; otherwise, clear it if this is + * an update. If this is not an update, then either the read-only + * flag is already clear, or this is a root mount and it was set + * intentionally at some previous point. + */ + if (vfs_getopt(mp->mnt_optnew, "ro", NULL, NULL) == 0) { + MNT_ILOCK(mp); + mp->mnt_flag |= MNT_RDONLY; + MNT_IUNLOCK(mp); + } else if (mp->mnt_flag & MNT_UPDATE) { + MNT_ILOCK(mp); + mp->mnt_flag &= ~MNT_RDONLY; + MNT_IUNLOCK(mp); + } + + /* + * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes + * no sense in that context. Also, set up appropriate retransmit + * and soft timeout behavior. + */ + if (argp->sotype == SOCK_STREAM) { + nmp->nm_flag &= ~NFSMNT_NOCONN; + nmp->nm_flag |= NFSMNT_DUMBTIMR; + nmp->nm_timeo = NFS_MAXTIMEO; + nmp->nm_retry = NFS_RETRANS_TCP; + } + + /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */ + if ((argp->flags & NFSMNT_NFSV3) == 0) + nmp->nm_flag &= ~NFSMNT_RDIRPLUS; + + /* Re-bind if rsrvd port requested and wasn't on one */ + adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) + && (argp->flags & NFSMNT_RESVPORT); + /* Also re-bind if we're switching to/from a connected UDP socket */ + adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != + (argp->flags & NFSMNT_NOCONN)); + + /* Update flags atomically. Don't change the lock bits. */ + nmp->nm_flag = argp->flags | nmp->nm_flag; + splx(s); + + if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { + nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; + if (nmp->nm_timeo < NFS_MINTIMEO) + nmp->nm_timeo = NFS_MINTIMEO; + else if (nmp->nm_timeo > NFS_MAXTIMEO) + nmp->nm_timeo = NFS_MAXTIMEO; + } + + if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { + nmp->nm_retry = argp->retrans; + if (nmp->nm_retry > NFS_MAXREXMIT) + nmp->nm_retry = NFS_MAXREXMIT; + } + + if (argp->flags & NFSMNT_NFSV3) { + if (argp->sotype == SOCK_DGRAM) + maxio = NFS_MAXDGRAMDATA; + else + maxio = NFS_MAXDATA; + } else + maxio = NFS_V2MAXDATA; + + if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { + nmp->nm_wsize = argp->wsize; + /* Round down to multiple of blocksize */ + nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); + if (nmp->nm_wsize <= 0) + nmp->nm_wsize = NFS_FABLKSIZE; + } + if (nmp->nm_wsize > maxio) + nmp->nm_wsize = maxio; + if (nmp->nm_wsize > MAXBSIZE) + nmp->nm_wsize = MAXBSIZE; + + if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { + nmp->nm_rsize = argp->rsize; + /* Round down to multiple of blocksize */ + nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); + if (nmp->nm_rsize <= 0) + nmp->nm_rsize = NFS_FABLKSIZE; + } + if (nmp->nm_rsize > maxio) + nmp->nm_rsize = maxio; + if (nmp->nm_rsize > MAXBSIZE) + nmp->nm_rsize = MAXBSIZE; + + if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { + nmp->nm_readdirsize = argp->readdirsize; + } + if (nmp->nm_readdirsize > maxio) + nmp->nm_readdirsize = maxio; + if (nmp->nm_readdirsize > nmp->nm_rsize) + nmp->nm_readdirsize = nmp->nm_rsize; + + if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) + nmp->nm_acregmin = argp->acregmin; + else + nmp->nm_acregmin = NFS_MINATTRTIMO; + if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) + nmp->nm_acregmax = argp->acregmax; + else + nmp->nm_acregmax = NFS_MAXATTRTIMO; + if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) + nmp->nm_acdirmin = argp->acdirmin; + else + nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; + if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) + nmp->nm_acdirmax = argp->acdirmax; + else + nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; + if (nmp->nm_acdirmin > nmp->nm_acdirmax) + nmp->nm_acdirmin = nmp->nm_acdirmax; + if (nmp->nm_acregmin > nmp->nm_acregmax) + nmp->nm_acregmin = nmp->nm_acregmax; + + if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) { + if (argp->maxgrouplist <= NFS_MAXGRPS) + nmp->nm_numgrps = argp->maxgrouplist; + else + nmp->nm_numgrps = NFS_MAXGRPS; + } + if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { + if (argp->readahead <= NFS_MAXRAHEAD) + nmp->nm_readahead = argp->readahead; + else + nmp->nm_readahead = NFS_MAXRAHEAD; + } + if ((argp->flags & NFSMNT_WCOMMITSIZE) && argp->wcommitsize >= 0) { + if (argp->wcommitsize < nmp->nm_wsize) + nmp->nm_wcommitsize = nmp->nm_wsize; + else + nmp->nm_wcommitsize = argp->wcommitsize; + } + if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 0) { + if (argp->deadthresh <= NFS_MAXDEADTHRESH) + nmp->nm_deadthresh = argp->deadthresh; + else + nmp->nm_deadthresh = NFS_MAXDEADTHRESH; + } + + adjsock |= ((nmp->nm_sotype != argp->sotype) || + (nmp->nm_soproto != argp->proto)); + nmp->nm_sotype = argp->sotype; + nmp->nm_soproto = argp->proto; + + if (nmp->nm_client && adjsock) { + nfs_safedisconnect(nmp); + if (nmp->nm_sotype == SOCK_DGRAM) + while (nfs_connect(nmp)) { + printf("nfs_args: retrying connect\n"); + (void) tsleep(&fake_wchan, PSOCK, "nfscon", hz); + } + } + + if (hostname) { + strlcpy(nmp->nm_hostname, hostname, + sizeof(nmp->nm_hostname)); + p = strchr(nmp->nm_hostname, ':'); + if (p) + *p = '\0'; + } + + if (vfs_getopt(mp->mnt_optnew, "sec", + (void **) &secname, NULL) == 0) { + nmp->nm_secflavor = nfs_sec_name_to_num(secname); + } else { + nmp->nm_secflavor = AUTH_SYS; + } + + if (vfs_getopt(mp->mnt_optnew, "principal", + (void **) &principal, NULL) == 0) { + strlcpy(nmp->nm_principal, principal, + sizeof(nmp->nm_principal)); + } else { + snprintf(nmp->nm_principal, sizeof(nmp->nm_principal), + "nfs@%s", nmp->nm_hostname); + } +} + +static const char *nfs_opts[] = { "from", "nfs_args", + "noatime", "noexec", "suiddir", "nosuid", "nosymfollow", "union", + "noclusterr", "noclusterw", "multilabel", "acls", "force", "update", + "async", "dumbtimer", "noconn", "nolockd", "intr", "rdirplus", "resvport", + "readahead", "readdirsize", "soft", "hard", "mntudp", "tcp", "udp", + "wsize", "rsize", "retrans", "acregmin", "acregmax", "acdirmin", + "acdirmax", "deadthresh", "hostname", "timeout", "addr", "fh", "nfsv3", + "sec", "maxgroups", "principal", "negnametimeo", "nocto", "wcommitsize", + "nametimeo", + NULL }; + +/* + * VFS Operations. + * + * mount system call + * It seems a bit dumb to copyinstr() the host and path here and then + * bcopy() them in mountnfs(), but I wanted to detect errors before + * doing the sockargs() call because sockargs() allocates an mbuf and + * an error after that means that I have to release the mbuf. + */ +/* ARGSUSED */ +static int +nfs_mount(struct mount *mp) +{ + struct nfs_args args = { + .version = NFS_ARGSVERSION, + .addr = NULL, + .addrlen = sizeof (struct sockaddr_in), + .sotype = SOCK_STREAM, + .proto = 0, + .fh = NULL, + .fhsize = 0, + .flags = NFSMNT_RESVPORT, + .wsize = NFS_WSIZE, + .rsize = NFS_RSIZE, + .readdirsize = NFS_READDIRSIZE, + .timeo = 10, + .retrans = NFS_RETRANS, + .maxgrouplist = NFS_MAXGRPS, + .readahead = NFS_DEFRAHEAD, + .wcommitsize = 0, /* was: NQ_DEFLEASE */ + .deadthresh = NFS_MAXDEADTHRESH, /* was: NQ_DEADTHRESH */ + .hostname = NULL, + /* args version 4 */ + .acregmin = NFS_MINATTRTIMO, + .acregmax = NFS_MAXATTRTIMO, + .acdirmin = NFS_MINDIRATTRTIMO, + .acdirmax = NFS_MAXDIRATTRTIMO, + }; + int error, ret, has_nfs_args_opt; + int has_addr_opt, has_fh_opt, has_hostname_opt; + struct sockaddr *nam; + struct vnode *vp; + char hst[MNAMELEN]; + size_t len; + u_char nfh[NFSX_V3FHMAX]; + char *opt; + int nametimeo = NFS_DEFAULT_NAMETIMEO; + int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO; + + has_nfs_args_opt = 0; + has_addr_opt = 0; + has_fh_opt = 0; + has_hostname_opt = 0; + + if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) { + error = EINVAL; + goto out; + } + + if ((mp->mnt_flag & (MNT_ROOTFS | MNT_UPDATE)) == MNT_ROOTFS) { + error = nfs_mountroot(mp); + goto out; + } + + /* + * The old mount_nfs program passed the struct nfs_args + * from userspace to kernel. The new mount_nfs program + * passes string options via nmount() from userspace to kernel + * and we populate the struct nfs_args in the kernel. + */ + if (vfs_getopt(mp->mnt_optnew, "nfs_args", NULL, NULL) == 0) { + error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args, + sizeof args); + if (error) + goto out; + + if (args.version != NFS_ARGSVERSION) { + error = EPROGMISMATCH; + goto out; + } + has_nfs_args_opt = 1; + } + + if (vfs_getopt(mp->mnt_optnew, "dumbtimer", NULL, NULL) == 0) + args.flags |= NFSMNT_DUMBTIMR; + if (vfs_getopt(mp->mnt_optnew, "noconn", NULL, NULL) == 0) + args.flags |= NFSMNT_NOCONN; + if (vfs_getopt(mp->mnt_optnew, "conn", NULL, NULL) == 0) + args.flags |= NFSMNT_NOCONN; + if (vfs_getopt(mp->mnt_optnew, "nolockd", NULL, NULL) == 0) + args.flags |= NFSMNT_NOLOCKD; + if (vfs_getopt(mp->mnt_optnew, "lockd", NULL, NULL) == 0) + args.flags &= ~NFSMNT_NOLOCKD; + if (vfs_getopt(mp->mnt_optnew, "intr", NULL, NULL) == 0) + args.flags |= NFSMNT_INT; + if (vfs_getopt(mp->mnt_optnew, "rdirplus", NULL, NULL) == 0) + args.flags |= NFSMNT_RDIRPLUS; + if (vfs_getopt(mp->mnt_optnew, "resvport", NULL, NULL) == 0) + args.flags |= NFSMNT_RESVPORT; + if (vfs_getopt(mp->mnt_optnew, "noresvport", NULL, NULL) == 0) + args.flags &= ~NFSMNT_RESVPORT; + if (vfs_getopt(mp->mnt_optnew, "soft", NULL, NULL) == 0) + args.flags |= NFSMNT_SOFT; + if (vfs_getopt(mp->mnt_optnew, "hard", NULL, NULL) == 0) + args.flags &= ~NFSMNT_SOFT; + if (vfs_getopt(mp->mnt_optnew, "mntudp", NULL, NULL) == 0) + args.sotype = SOCK_DGRAM; + if (vfs_getopt(mp->mnt_optnew, "udp", NULL, NULL) == 0) + args.sotype = SOCK_DGRAM; + if (vfs_getopt(mp->mnt_optnew, "tcp", NULL, NULL) == 0) + args.sotype = SOCK_STREAM; + if (vfs_getopt(mp->mnt_optnew, "nfsv3", NULL, NULL) == 0) + args.flags |= NFSMNT_NFSV3; + if (vfs_getopt(mp->mnt_optnew, "nocto", NULL, NULL) == 0) + args.flags |= NFSMNT_NOCTO; + if (vfs_getopt(mp->mnt_optnew, "readdirsize", (void **)&opt, NULL) == 0) { + if (opt == NULL) { + vfs_mount_error(mp, "illegal readdirsize"); + error = EINVAL; + goto out; + } + ret = sscanf(opt, "%d", &args.readdirsize); + if (ret != 1 || args.readdirsize <= 0) { + vfs_mount_error(mp, "illegal readdirsize: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_READDIRSIZE; + } + if (vfs_getopt(mp->mnt_optnew, "readahead", (void **)&opt, NULL) == 0) { + if (opt == NULL) { + vfs_mount_error(mp, "illegal readahead"); + error = EINVAL; + goto out; + } + ret = sscanf(opt, "%d", &args.readahead); + if (ret != 1 || args.readahead <= 0) { + vfs_mount_error(mp, "illegal readahead: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_READAHEAD; + } + if (vfs_getopt(mp->mnt_optnew, "wsize", (void **)&opt, NULL) == 0) { + if (opt == NULL) { + vfs_mount_error(mp, "illegal wsize"); + error = EINVAL; + goto out; + } + ret = sscanf(opt, "%d", &args.wsize); + if (ret != 1 || args.wsize <= 0) { + vfs_mount_error(mp, "illegal wsize: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_WSIZE; + } + if (vfs_getopt(mp->mnt_optnew, "rsize", (void **)&opt, NULL) == 0) { + if (opt == NULL) { + vfs_mount_error(mp, "illegal rsize"); + error = EINVAL; + goto out; + } + ret = sscanf(opt, "%d", &args.rsize); + if (ret != 1 || args.rsize <= 0) { + vfs_mount_error(mp, "illegal wsize: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_RSIZE; + } + if (vfs_getopt(mp->mnt_optnew, "retrans", (void **)&opt, NULL) == 0) { + if (opt == NULL) { + vfs_mount_error(mp, "illegal retrans"); + error = EINVAL; + goto out; + } + ret = sscanf(opt, "%d", &args.retrans); + if (ret != 1 || args.retrans <= 0) { + vfs_mount_error(mp, "illegal retrans: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_RETRANS; + } + if (vfs_getopt(mp->mnt_optnew, "acregmin", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.acregmin); + if (ret != 1 || args.acregmin < 0) { + vfs_mount_error(mp, "illegal acregmin: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_ACREGMIN; + } + if (vfs_getopt(mp->mnt_optnew, "acregmax", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.acregmax); + if (ret != 1 || args.acregmax < 0) { + vfs_mount_error(mp, "illegal acregmax: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_ACREGMAX; + } + if (vfs_getopt(mp->mnt_optnew, "acdirmin", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.acdirmin); + if (ret != 1 || args.acdirmin < 0) { + vfs_mount_error(mp, "illegal acdirmin: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_ACDIRMIN; + } + if (vfs_getopt(mp->mnt_optnew, "acdirmax", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.acdirmax); + if (ret != 1 || args.acdirmax < 0) { + vfs_mount_error(mp, "illegal acdirmax: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_ACDIRMAX; + } + if (vfs_getopt(mp->mnt_optnew, "wcommitsize", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.wcommitsize); + if (ret != 1 || args.wcommitsize < 0) { + vfs_mount_error(mp, "illegal wcommitsize: %s", opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_WCOMMITSIZE; + } + if (vfs_getopt(mp->mnt_optnew, "deadthresh", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.deadthresh); + if (ret != 1 || args.deadthresh <= 0) { + vfs_mount_error(mp, "illegal deadthresh: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_DEADTHRESH; + } + if (vfs_getopt(mp->mnt_optnew, "timeout", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.timeo); + if (ret != 1 || args.timeo <= 0) { + vfs_mount_error(mp, "illegal timeout: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_TIMEO; + } + if (vfs_getopt(mp->mnt_optnew, "maxgroups", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &args.maxgrouplist); + if (ret != 1 || args.maxgrouplist <= 0) { + vfs_mount_error(mp, "illegal maxgroups: %s", + opt); + error = EINVAL; + goto out; + } + args.flags |= NFSMNT_MAXGRPS; + } + if (vfs_getopt(mp->mnt_optnew, "nametimeo", (void **)&opt, NULL) == 0) { + ret = sscanf(opt, "%d", &nametimeo); + if (ret != 1 || nametimeo < 0) { + vfs_mount_error(mp, "illegal nametimeo: %s", opt); + error = EINVAL; + goto out; + } + } + if (vfs_getopt(mp->mnt_optnew, "negnametimeo", (void **)&opt, NULL) + == 0) { + ret = sscanf(opt, "%d", &negnametimeo); + if (ret != 1 || negnametimeo < 0) { + vfs_mount_error(mp, "illegal negnametimeo: %s", + opt); + error = EINVAL; + goto out; + } + } + if (vfs_getopt(mp->mnt_optnew, "addr", (void **)&args.addr, + &args.addrlen) == 0) { + has_addr_opt = 1; + if (args.addrlen > SOCK_MAXADDRLEN) { + error = ENAMETOOLONG; + goto out; + } + nam = malloc(args.addrlen, M_SONAME, + M_WAITOK); + bcopy(args.addr, nam, args.addrlen); + nam->sa_len = args.addrlen; + } + if (vfs_getopt(mp->mnt_optnew, "fh", (void **)&args.fh, + &args.fhsize) == 0) { + has_fh_opt = 1; + } + if (vfs_getopt(mp->mnt_optnew, "hostname", (void **)&args.hostname, + NULL) == 0) { + has_hostname_opt = 1; + } + if (args.hostname == NULL) { + vfs_mount_error(mp, "Invalid hostname"); + error = EINVAL; + goto out; + } + if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) { + vfs_mount_error(mp, "Bad file handle"); + error = EINVAL; + goto out; + } + + if (mp->mnt_flag & MNT_UPDATE) { + struct nfsmount *nmp = VFSTONFS(mp); + + if (nmp == NULL) { + error = EIO; + goto out; + } + + /* + * If a change from TCP->UDP is done and there are thread(s) + * that have I/O RPC(s) in progress with a tranfer size + * greater than NFS_MAXDGRAMDATA, those thread(s) will be + * hung, retrying the RPC(s) forever. Usually these threads + * will be seen doing an uninterruptible sleep on wait channel + * "newnfsreq" (truncated to "newnfsre" by procstat). + */ + if (args.sotype == SOCK_DGRAM && nmp->nm_sotype == SOCK_STREAM) + tprintf(curthread->td_proc, LOG_WARNING, + "Warning: mount -u that changes TCP->UDP can result in hung threads\n"); + + /* + * When doing an update, we can't change from or to + * v3, switch lockd strategies or change cookie translation + */ + args.flags = (args.flags & + ~(NFSMNT_NFSV3 | NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)) | + (nmp->nm_flag & + (NFSMNT_NFSV3 | NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)); + nfs_decode_args(mp, nmp, &args, NULL); + goto out; + } + + /* + * Make the nfs_ip_paranoia sysctl serve as the default connection + * or no-connection mode for those protocols that support + * no-connection mode (the flag will be cleared later for protocols + * that do not support no-connection mode). This will allow a client + * to receive replies from a different IP then the request was + * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), + * not 0. + */ + if (nfs_ip_paranoia == 0) + args.flags |= NFSMNT_NOCONN; + + if (has_nfs_args_opt) { + /* + * In the 'nfs_args' case, the pointers in the args + * structure are in userland - we copy them in here. + */ + if (!has_fh_opt) { + error = copyin((caddr_t)args.fh, (caddr_t)nfh, + args.fhsize); + if (error) { + goto out; + } + args.fh = nfh; + } + if (!has_hostname_opt) { + error = copyinstr(args.hostname, hst, MNAMELEN-1, &len); + if (error) { + goto out; + } + bzero(&hst[len], MNAMELEN - len); + args.hostname = hst; + } + if (!has_addr_opt) { + /* sockargs() call must be after above copyin() calls */ + error = getsockaddr(&nam, (caddr_t)args.addr, + args.addrlen); + if (error) { + goto out; + } + } + } else if (has_addr_opt == 0) { + vfs_mount_error(mp, "No server address"); + error = EINVAL; + goto out; + } + error = mountnfs(&args, mp, nam, args.hostname, &vp, + curthread->td_ucred, nametimeo, negnametimeo); +out: + if (!error) { + MNT_ILOCK(mp); + mp->mnt_kern_flag |= (MNTK_MPSAFE|MNTK_LOOKUP_SHARED); + MNT_IUNLOCK(mp); + } + return (error); +} + + +/* + * VFS Operations. + * + * mount system call + * It seems a bit dumb to copyinstr() the host and path here and then + * bcopy() them in mountnfs(), but I wanted to detect errors before + * doing the sockargs() call because sockargs() allocates an mbuf and + * an error after that means that I have to release the mbuf. + */ +/* ARGSUSED */ +static int +nfs_cmount(struct mntarg *ma, void *data, uint64_t flags) +{ + int error; + struct nfs_args args; + + error = copyin(data, &args, sizeof (struct nfs_args)); + if (error) + return error; + + ma = mount_arg(ma, "nfs_args", &args, sizeof args); + + error = kernel_mount(ma, flags); + return (error); +} + +/* + * Common code for mount and mountroot + */ +static int +mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, + char *hst, struct vnode **vpp, struct ucred *cred, int nametimeo, + int negnametimeo) +{ + struct nfsmount *nmp; + struct nfsnode *np; + int error; + struct vattr attrs; + + if (mp->mnt_flag & MNT_UPDATE) { + nmp = VFSTONFS(mp); + printf("%s: MNT_UPDATE is no longer handled here\n", __func__); + free(nam, M_SONAME); + return (0); + } else { + nmp = uma_zalloc(nfsmount_zone, M_WAITOK); + bzero((caddr_t)nmp, sizeof (struct nfsmount)); + TAILQ_INIT(&nmp->nm_bufq); + mp->mnt_data = nmp; + nmp->nm_getinfo = nfs_getnlminfo; + nmp->nm_vinvalbuf = nfs_vinvalbuf; + } + vfs_getnewfsid(mp); + nmp->nm_mountp = mp; + mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF); + + /* + * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too + * high, depending on whether we end up with negative offsets in + * the client or server somewhere. 2GB-1 may be safer. + * + * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum + * that we can handle until we find out otherwise. + */ + if ((argp->flags & NFSMNT_NFSV3) == 0) + nmp->nm_maxfilesize = 0xffffffffLL; + else + nmp->nm_maxfilesize = OFF_MAX; + + nmp->nm_timeo = NFS_TIMEO; + nmp->nm_retry = NFS_RETRANS; + if ((argp->flags & NFSMNT_NFSV3) && argp->sotype == SOCK_STREAM) { + nmp->nm_wsize = nmp->nm_rsize = NFS_MAXDATA; + } else { + nmp->nm_wsize = NFS_WSIZE; + nmp->nm_rsize = NFS_RSIZE; + } + nmp->nm_wcommitsize = hibufspace / (desiredvnodes / 1000); + nmp->nm_readdirsize = NFS_READDIRSIZE; + nmp->nm_numgrps = NFS_MAXGRPS; + nmp->nm_readahead = NFS_DEFRAHEAD; + nmp->nm_deadthresh = NFS_MAXDEADTHRESH; + nmp->nm_nametimeo = nametimeo; + nmp->nm_negnametimeo = negnametimeo; + nmp->nm_tprintf_delay = nfs_tprintf_delay; + if (nmp->nm_tprintf_delay < 0) + nmp->nm_tprintf_delay = 0; + nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay; + if (nmp->nm_tprintf_initial_delay < 0) + nmp->nm_tprintf_initial_delay = 0; + nmp->nm_fhsize = argp->fhsize; + bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); + bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); + nmp->nm_nam = nam; + /* Set up the sockets and per-host congestion */ + nmp->nm_sotype = argp->sotype; + nmp->nm_soproto = argp->proto; + nmp->nm_rpcops = &nfs_rpcops; + + nfs_decode_args(mp, nmp, argp, hst); + + /* + * For Connection based sockets (TCP,...) defer the connect until + * the first request, in case the server is not responding. + */ + if (nmp->nm_sotype == SOCK_DGRAM && + (error = nfs_connect(nmp))) + goto bad; + + /* + * This is silly, but it has to be set so that vinifod() works. + * We do not want to do an nfs_statfs() here since we can get + * stuck on a dead server and we are holding a lock on the mount + * point. + */ + mtx_lock(&nmp->nm_mtx); + mp->mnt_stat.f_iosize = nfs_iosize(nmp); + mtx_unlock(&nmp->nm_mtx); + /* + * A reference count is needed on the nfsnode representing the + * remote root. If this object is not persistent, then backward + * traversals of the mount point (i.e. "..") will not work if + * the nfsnode gets flushed out of the cache. Ufs does not have + * this problem, because one can identify root inodes by their + * number == ROOTINO (2). + */ + error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE); + if (error) + goto bad; + *vpp = NFSTOV(np); + + /* + * Get file attributes and transfer parameters for the + * mountpoint. This has the side effect of filling in + * (*vpp)->v_type with the correct value. + */ + if (argp->flags & NFSMNT_NFSV3) + nfs_fsinfo(nmp, *vpp, curthread->td_ucred, curthread); + else + VOP_GETATTR(*vpp, &attrs, curthread->td_ucred); + + /* + * Lose the lock but keep the ref. + */ + VOP_UNLOCK(*vpp, 0); + + return (0); +bad: + nfs_disconnect(nmp); + mtx_destroy(&nmp->nm_mtx); + uma_zfree(nfsmount_zone, nmp); + free(nam, M_SONAME); + return (error); +} + +/* + * unmount system call + */ +static int +nfs_unmount(struct mount *mp, int mntflags) +{ + struct nfsmount *nmp; + int error, flags = 0; + + if (mntflags & MNT_FORCE) + flags |= FORCECLOSE; + nmp = VFSTONFS(mp); + /* + * Goes something like this.. + * - Call vflush() to clear out vnodes for this filesystem + * - Close the socket + * - Free up the data structures + */ + /* In the forced case, cancel any outstanding requests. */ + if (flags & FORCECLOSE) { + error = nfs_nmcancelreqs(nmp); + if (error) + goto out; + } + /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ + error = vflush(mp, 1, flags, curthread); + if (error) + goto out; + + /* + * We are now committed to the unmount. + */ + nfs_disconnect(nmp); + free(nmp->nm_nam, M_SONAME); + + mtx_destroy(&nmp->nm_mtx); + uma_zfree(nfsmount_zone, nmp); +out: + return (error); +} + +/* + * Return root of a filesystem + */ +static int +nfs_root(struct mount *mp, int flags, struct vnode **vpp) +{ + struct vnode *vp; + struct nfsmount *nmp; + struct nfsnode *np; + int error; + + nmp = VFSTONFS(mp); + error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, flags); + if (error) + return error; + vp = NFSTOV(np); + /* + * Get transfer parameters and attributes for root vnode once. + */ + mtx_lock(&nmp->nm_mtx); + if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0 && + (nmp->nm_flag & NFSMNT_NFSV3)) { + mtx_unlock(&nmp->nm_mtx); + nfs_fsinfo(nmp, vp, curthread->td_ucred, curthread); + } else + mtx_unlock(&nmp->nm_mtx); + if (vp->v_type == VNON) + vp->v_type = VDIR; + vp->v_vflag |= VV_ROOT; + *vpp = vp; + return (0); +} + +/* + * Flush out the buffer cache + */ +/* ARGSUSED */ +static int +nfs_sync(struct mount *mp, int waitfor) +{ + struct vnode *vp, *mvp; + struct thread *td; + int error, allerror = 0; + + td = curthread; + + MNT_ILOCK(mp); + /* + * If a forced dismount is in progress, return from here so that + * the umount(2) syscall doesn't get stuck in VFS_SYNC() before + * calling VFS_UNMOUNT(). + */ + if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + MNT_IUNLOCK(mp); + return (EBADF); + } + + /* + * Force stale buffer cache information to be flushed. + */ +loop: + MNT_VNODE_FOREACH(vp, mp, mvp) { + VI_LOCK(vp); + MNT_IUNLOCK(mp); + /* XXX Racy bv_cnt check. */ + if (VOP_ISLOCKED(vp) || vp->v_bufobj.bo_dirty.bv_cnt == 0 || + waitfor == MNT_LAZY) { + VI_UNLOCK(vp); + MNT_ILOCK(mp); + continue; + } + if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { + MNT_ILOCK(mp); + MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); + goto loop; + } + error = VOP_FSYNC(vp, waitfor, td); + if (error) + allerror = error; + VOP_UNLOCK(vp, 0); + vrele(vp); + + MNT_ILOCK(mp); + } + MNT_IUNLOCK(mp); + return (allerror); +} + +static int +nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req) +{ + struct nfsmount *nmp = VFSTONFS(mp); + struct vfsquery vq; + int error; + + bzero(&vq, sizeof(vq)); + switch (op) { +#if 0 + case VFS_CTL_NOLOCKS: + val = (nmp->nm_flag & NFSMNT_NOLOCKS) ? 1 : 0; + if (req->oldptr != NULL) { + error = SYSCTL_OUT(req, &val, sizeof(val)); + if (error) + return (error); + } + if (req->newptr != NULL) { + error = SYSCTL_IN(req, &val, sizeof(val)); + if (error) + return (error); + if (val) + nmp->nm_flag |= NFSMNT_NOLOCKS; + else + nmp->nm_flag &= ~NFSMNT_NOLOCKS; + } + break; +#endif + case VFS_CTL_QUERY: + mtx_lock(&nmp->nm_mtx); + if (nmp->nm_state & NFSSTA_TIMEO) + vq.vq_flags |= VQ_NOTRESP; + mtx_unlock(&nmp->nm_mtx); +#if 0 + if (!(nmp->nm_flag & NFSMNT_NOLOCKS) && + (nmp->nm_state & NFSSTA_LOCKTIMEO)) + vq.vq_flags |= VQ_NOTRESPLOCK; +#endif + error = SYSCTL_OUT(req, &vq, sizeof(vq)); + break; + case VFS_CTL_TIMEO: + if (req->oldptr != NULL) { + error = SYSCTL_OUT(req, &nmp->nm_tprintf_initial_delay, + sizeof(nmp->nm_tprintf_initial_delay)); + if (error) + return (error); + } + if (req->newptr != NULL) { + error = vfs_suser(mp, req->td); + if (error) + return (error); + error = SYSCTL_IN(req, &nmp->nm_tprintf_initial_delay, + sizeof(nmp->nm_tprintf_initial_delay)); + if (error) + return (error); + if (nmp->nm_tprintf_initial_delay < 0) + nmp->nm_tprintf_initial_delay = 0; + } + break; + default: + return (ENOTSUP); + } + return (0); +} + +/* + * Extract the information needed by the nlm from the nfs vnode. + */ +static void +nfs_getnlminfo(struct vnode *vp, uint8_t *fhp, size_t *fhlenp, + struct sockaddr_storage *sp, int *is_v3p, off_t *sizep, + struct timeval *timeop) +{ + struct nfsmount *nmp; + struct nfsnode *np = VTONFS(vp); + + nmp = VFSTONFS(vp->v_mount); + if (fhlenp != NULL) + *fhlenp = (size_t)np->n_fhsize; + if (fhp != NULL) + bcopy(np->n_fhp, fhp, np->n_fhsize); + if (sp != NULL) + bcopy(nmp->nm_nam, sp, min(nmp->nm_nam->sa_len, sizeof(*sp))); + if (is_v3p != NULL) + *is_v3p = NFS_ISV3(vp); + if (sizep != NULL) + *sizep = np->n_size; + if (timeop != NULL) { + timeop->tv_sec = nmp->nm_timeo / NFS_HZ; + timeop->tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ); + } +} + diff --git a/sys/nfsclient/nfs_vnops.c b/sys/nfsclient/nfs_vnops.c new file mode 100644 index 0000000..241af51 --- /dev/null +++ b/sys/nfsclient/nfs_vnops.c @@ -0,0 +1,3523 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +/* + * vnode op calls for Sun NFS version 2 and 3 + */ + +#include "opt_inet.h" +#include "opt_kdtrace.h" + +#include <sys/param.h> +#include <sys/kernel.h> +#include <sys/systm.h> +#include <sys/resourcevar.h> +#include <sys/proc.h> +#include <sys/mount.h> +#include <sys/bio.h> +#include <sys/buf.h> +#include <sys/jail.h> +#include <sys/malloc.h> +#include <sys/mbuf.h> +#include <sys/namei.h> +#include <sys/socket.h> +#include <sys/vnode.h> +#include <sys/dirent.h> +#include <sys/fcntl.h> +#include <sys/lockf.h> +#include <sys/stat.h> +#include <sys/sysctl.h> +#include <sys/signalvar.h> + +#include <vm/vm.h> +#include <vm/vm_extern.h> +#include <vm/vm_object.h> + +#include <fs/fifofs/fifo.h> + +#include <nfs/nfsproto.h> +#include <nfsclient/nfs.h> +#include <nfsclient/nfsnode.h> +#include <nfsclient/nfsmount.h> +#include <nfs/nfs_kdtrace.h> +#include <nfs/nfs_lock.h> +#include <nfs/xdr_subs.h> +#include <nfsclient/nfsm_subs.h> + +#include <net/if.h> +#include <netinet/in.h> +#include <netinet/in_var.h> + +#include <machine/stdarg.h> + +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> + +dtrace_nfsclient_accesscache_flush_probe_func_t + dtrace_nfsclient_accesscache_flush_done_probe; +uint32_t nfsclient_accesscache_flush_done_id; + +dtrace_nfsclient_accesscache_get_probe_func_t + dtrace_nfsclient_accesscache_get_hit_probe, + dtrace_nfsclient_accesscache_get_miss_probe; +uint32_t nfsclient_accesscache_get_hit_id; +uint32_t nfsclient_accesscache_get_miss_id; + +dtrace_nfsclient_accesscache_load_probe_func_t + dtrace_nfsclient_accesscache_load_done_probe; +uint32_t nfsclient_accesscache_load_done_id; +#endif /* !KDTRACE_HOOKS */ + +/* Defs */ +#define TRUE 1 +#define FALSE 0 + +/* + * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these + * calls are not in getblk() and brelse() so that they would not be necessary + * here. + */ +#ifndef B_VMIO +#define vfs_busy_pages(bp, f) +#endif + +static vop_read_t nfsfifo_read; +static vop_write_t nfsfifo_write; +static vop_close_t nfsfifo_close; +static int nfs_flush(struct vnode *, int, int); +static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *); +static vop_lookup_t nfs_lookup; +static vop_create_t nfs_create; +static vop_mknod_t nfs_mknod; +static vop_open_t nfs_open; +static vop_close_t nfs_close; +static vop_access_t nfs_access; +static vop_getattr_t nfs_getattr; +static vop_setattr_t nfs_setattr; +static vop_read_t nfs_read; +static vop_fsync_t nfs_fsync; +static vop_remove_t nfs_remove; +static vop_link_t nfs_link; +static vop_rename_t nfs_rename; +static vop_mkdir_t nfs_mkdir; +static vop_rmdir_t nfs_rmdir; +static vop_symlink_t nfs_symlink; +static vop_readdir_t nfs_readdir; +static vop_strategy_t nfs_strategy; +static int nfs_lookitup(struct vnode *, const char *, int, + struct ucred *, struct thread *, struct nfsnode **); +static int nfs_sillyrename(struct vnode *, struct vnode *, + struct componentname *); +static vop_access_t nfsspec_access; +static vop_readlink_t nfs_readlink; +static vop_print_t nfs_print; +static vop_advlock_t nfs_advlock; +static vop_advlockasync_t nfs_advlockasync; + +/* + * Global vfs data structures for nfs + */ +struct vop_vector nfs_vnodeops = { + .vop_default = &default_vnodeops, + .vop_access = nfs_access, + .vop_advlock = nfs_advlock, + .vop_advlockasync = nfs_advlockasync, + .vop_close = nfs_close, + .vop_create = nfs_create, + .vop_fsync = nfs_fsync, + .vop_getattr = nfs_getattr, + .vop_getpages = nfs_getpages, + .vop_putpages = nfs_putpages, + .vop_inactive = nfs_inactive, + .vop_link = nfs_link, + .vop_lookup = nfs_lookup, + .vop_mkdir = nfs_mkdir, + .vop_mknod = nfs_mknod, + .vop_open = nfs_open, + .vop_print = nfs_print, + .vop_read = nfs_read, + .vop_readdir = nfs_readdir, + .vop_readlink = nfs_readlink, + .vop_reclaim = nfs_reclaim, + .vop_remove = nfs_remove, + .vop_rename = nfs_rename, + .vop_rmdir = nfs_rmdir, + .vop_setattr = nfs_setattr, + .vop_strategy = nfs_strategy, + .vop_symlink = nfs_symlink, + .vop_write = nfs_write, +}; + +struct vop_vector nfs_fifoops = { + .vop_default = &fifo_specops, + .vop_access = nfsspec_access, + .vop_close = nfsfifo_close, + .vop_fsync = nfs_fsync, + .vop_getattr = nfs_getattr, + .vop_inactive = nfs_inactive, + .vop_print = nfs_print, + .vop_read = nfsfifo_read, + .vop_reclaim = nfs_reclaim, + .vop_setattr = nfs_setattr, + .vop_write = nfsfifo_write, +}; + +static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, + struct componentname *cnp, struct vattr *vap); +static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen, + struct ucred *cred, struct thread *td); +static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, + int fnamelen, struct vnode *tdvp, + const char *tnameptr, int tnamelen, + struct ucred *cred, struct thread *td); +static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp, + struct sillyrename *sp); + +/* + * Global variables + */ +struct mtx nfs_iod_mtx; +enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON]; +struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; +int nfs_numasync = 0; +#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1)) + +SYSCTL_DECL(_vfs_oldnfs); + +static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, + &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); + +static int nfs_prime_access_cache = 0; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, + &nfs_prime_access_cache, 0, + "Prime NFS ACCESS cache when fetching attributes"); + +static int nfsv3_commit_on_close = 0; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW, + &nfsv3_commit_on_close, 0, "write+commit on close, else only write"); + +static int nfs_clean_pages_on_close = 1; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, + &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); + +int nfs_directio_enable = 0; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, + &nfs_directio_enable, 0, "Enable NFS directio"); + +/* + * This sysctl allows other processes to mmap a file that has been opened + * O_DIRECT by a process. In general, having processes mmap the file while + * Direct IO is in progress can lead to Data Inconsistencies. But, we allow + * this by default to prevent DoS attacks - to prevent a malicious user from + * opening up files O_DIRECT preventing other users from mmap'ing these + * files. "Protected" environments where stricter consistency guarantees are + * required can disable this knob. The process that opened the file O_DIRECT + * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not + * meaningful. + */ +int nfs_directio_allow_mmap = 1; +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, + &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); + +#if 0 +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD, + &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count"); + +SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD, + &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count"); +#endif + +#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \ + | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \ + | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP) + +/* + * SMP Locking Note : + * The list of locks after the description of the lock is the ordering + * of other locks acquired with the lock held. + * np->n_mtx : Protects the fields in the nfsnode. + VM Object Lock + VI_MTX (acquired indirectly) + * nmp->nm_mtx : Protects the fields in the nfsmount. + rep->r_mtx + * nfs_iod_mtx : Global lock, protects shared nfsiod state. + * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. + nmp->nm_mtx + rep->r_mtx + * rep->r_mtx : Protects the fields in an nfsreq. + */ + +static int +nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td, + struct ucred *cred, uint32_t *retmode) +{ + const int v3 = 1; + u_int32_t *tl; + int error = 0, attrflag, i, lrupos; + + struct mbuf *mreq, *mrep, *md, *mb; + caddr_t bpos, dpos; + u_int32_t rmode; + struct nfsnode *np = VTONFS(vp); + + nfsstats.rpccnt[NFSPROC_ACCESS]++; + mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); + *tl = txdr_unsigned(wmode); + nfsm_request(vp, NFSPROC_ACCESS, td, cred); + nfsm_postop_attr(vp, attrflag); + if (!error) { + lrupos = 0; + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + rmode = fxdr_unsigned(u_int32_t, *tl); + mtx_lock(&np->n_mtx); + for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { + if (np->n_accesscache[i].uid == cred->cr_uid) { + np->n_accesscache[i].mode = rmode; + np->n_accesscache[i].stamp = time_second; + break; + } + if (i > 0 && np->n_accesscache[i].stamp < + np->n_accesscache[lrupos].stamp) + lrupos = i; + } + if (i == NFS_ACCESSCACHESIZE) { + np->n_accesscache[lrupos].uid = cred->cr_uid; + np->n_accesscache[lrupos].mode = rmode; + np->n_accesscache[lrupos].stamp = time_second; + } + mtx_unlock(&np->n_mtx); + if (retmode != NULL) + *retmode = rmode; + KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); + } + m_freem(mrep); +nfsmout: +#ifdef KDTRACE_HOOKS + if (error) { + KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, + error); + } +#endif + return (error); +} + +/* + * nfs access vnode op. + * For nfs version 2, just return ok. File accesses may fail later. + * For nfs version 3, use the access rpc to check accessibility. If file modes + * are changed on the server, accesses might still fail later. + */ +static int +nfs_access(struct vop_access_args *ap) +{ + struct vnode *vp = ap->a_vp; + int error = 0, i, gotahit; + u_int32_t mode, rmode, wmode; + int v3 = NFS_ISV3(vp); + struct nfsnode *np = VTONFS(vp); + + /* + * Disallow write attempts on filesystems mounted read-only; + * unless the file is a socket, fifo, or a block or character + * device resident on the filesystem. + */ + if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { + switch (vp->v_type) { + case VREG: + case VDIR: + case VLNK: + return (EROFS); + default: + break; + } + } + /* + * For nfs v3, check to see if we have done this recently, and if + * so return our cached result instead of making an ACCESS call. + * If not, do an access rpc, otherwise you are stuck emulating + * ufs_access() locally using the vattr. This may not be correct, + * since the server may apply other access criteria such as + * client uid-->server uid mapping that we do not know about. + */ + if (v3) { + if (ap->a_accmode & VREAD) + mode = NFSV3ACCESS_READ; + else + mode = 0; + if (vp->v_type != VDIR) { + if (ap->a_accmode & VWRITE) + mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND); + if (ap->a_accmode & VEXEC) + mode |= NFSV3ACCESS_EXECUTE; + } else { + if (ap->a_accmode & VWRITE) + mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND | + NFSV3ACCESS_DELETE); + if (ap->a_accmode & VEXEC) + mode |= NFSV3ACCESS_LOOKUP; + } + /* XXX safety belt, only make blanket request if caching */ + if (nfsaccess_cache_timeout > 0) { + wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY | + NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE | + NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP; + } else { + wmode = mode; + } + + /* + * Does our cached result allow us to give a definite yes to + * this request? + */ + gotahit = 0; + mtx_lock(&np->n_mtx); + for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { + if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { + if (time_second < (np->n_accesscache[i].stamp + + nfsaccess_cache_timeout) && + (np->n_accesscache[i].mode & mode) == mode) { + nfsstats.accesscache_hits++; + gotahit = 1; + } + break; + } + } + mtx_unlock(&np->n_mtx); +#ifdef KDTRACE_HOOKS + if (gotahit) + KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, + ap->a_cred->cr_uid, mode); + else + KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, + ap->a_cred->cr_uid, mode); +#endif + if (gotahit == 0) { + /* + * Either a no, or a don't know. Go to the wire. + */ + nfsstats.accesscache_misses++; + error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred, + &rmode); + if (!error) { + if ((rmode & mode) != mode) + error = EACCES; + } + } + return (error); + } else { + if ((error = nfsspec_access(ap)) != 0) { + return (error); + } + /* + * Attempt to prevent a mapped root from accessing a file + * which it shouldn't. We try to read a byte from the file + * if the user is root and the file is not zero length. + * After calling nfsspec_access, we should have the correct + * file size cached. + */ + mtx_lock(&np->n_mtx); + if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) + && VTONFS(vp)->n_size > 0) { + struct iovec aiov; + struct uio auio; + char buf[1]; + + mtx_unlock(&np->n_mtx); + aiov.iov_base = buf; + aiov.iov_len = 1; + auio.uio_iov = &aiov; + auio.uio_iovcnt = 1; + auio.uio_offset = 0; + auio.uio_resid = 1; + auio.uio_segflg = UIO_SYSSPACE; + auio.uio_rw = UIO_READ; + auio.uio_td = ap->a_td; + + if (vp->v_type == VREG) + error = nfs_readrpc(vp, &auio, ap->a_cred); + else if (vp->v_type == VDIR) { + char* bp; + bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); + aiov.iov_base = bp; + aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; + error = nfs_readdirrpc(vp, &auio, ap->a_cred); + free(bp, M_TEMP); + } else if (vp->v_type == VLNK) + error = nfs_readlinkrpc(vp, &auio, ap->a_cred); + else + error = EACCES; + } else + mtx_unlock(&np->n_mtx); + return (error); + } +} + +int nfs_otw_getattr_avoid = 0; + +/* + * nfs open vnode op + * Check to see if the type is ok + * and that deletion is not in progress. + * For paged in text files, you will need to flush the page cache + * if consistency is lost. + */ +/* ARGSUSED */ +static int +nfs_open(struct vop_open_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + struct vattr vattr; + int error; + int fmode = ap->a_mode; + + if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) + return (EOPNOTSUPP); + + /* + * Get a valid lease. If cached data is stale, flush it. + */ + mtx_lock(&np->n_mtx); + if (np->n_flag & NMODIFIED) { + mtx_unlock(&np->n_mtx); + error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); + if (error == EINTR || error == EIO) + return (error); + mtx_lock(&np->n_mtx); + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + if (vp->v_type == VDIR) + np->n_direofoffset = 0; + mtx_unlock(&np->n_mtx); + error = VOP_GETATTR(vp, &vattr, ap->a_cred); + if (error) + return (error); + mtx_lock(&np->n_mtx); + np->n_mtime = vattr.va_mtime; + } else { + mtx_unlock(&np->n_mtx); + error = VOP_GETATTR(vp, &vattr, ap->a_cred); + if (error) + return (error); + mtx_lock(&np->n_mtx); + if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { + if (vp->v_type == VDIR) + np->n_direofoffset = 0; + mtx_unlock(&np->n_mtx); + error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); + if (error == EINTR || error == EIO) { + return (error); + } + mtx_lock(&np->n_mtx); + np->n_mtime = vattr.va_mtime; + } + } + /* + * If the object has >= 1 O_DIRECT active opens, we disable caching. + */ + if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { + if (np->n_directio_opens == 0) { + mtx_unlock(&np->n_mtx); + error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); + if (error) + return (error); + mtx_lock(&np->n_mtx); + np->n_flag |= NNONCACHE; + } + np->n_directio_opens++; + } + mtx_unlock(&np->n_mtx); + vnode_create_vobject(vp, vattr.va_size, ap->a_td); + return (0); +} + +/* + * nfs close vnode op + * What an NFS client should do upon close after writing is a debatable issue. + * Most NFS clients push delayed writes to the server upon close, basically for + * two reasons: + * 1 - So that any write errors may be reported back to the client process + * doing the close system call. By far the two most likely errors are + * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. + * 2 - To put a worst case upper bound on cache inconsistency between + * multiple clients for the file. + * There is also a consistency problem for Version 2 of the protocol w.r.t. + * not being able to tell if other clients are writing a file concurrently, + * since there is no way of knowing if the changed modify time in the reply + * is only due to the write for this client. + * (NFS Version 3 provides weak cache consistency data in the reply that + * should be sufficient to detect and handle this case.) + * + * The current code does the following: + * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers + * for NFS Version 3 - flush dirty buffers to the server but don't invalidate + * or commit them (this satisfies 1 and 2 except for the + * case where the server crashes after this close but + * before the commit RPC, which is felt to be "good + * enough". Changing the last argument to nfs_flush() to + * a 1 would force a commit operation, if it is felt a + * commit is necessary now. + */ +/* ARGSUSED */ +static int +nfs_close(struct vop_close_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + int error = 0; + int fmode = ap->a_fflag; + + if (vp->v_type == VREG) { + /* + * Examine and clean dirty pages, regardless of NMODIFIED. + * This closes a major hole in close-to-open consistency. + * We want to push out all dirty pages (and buffers) on + * close, regardless of whether they were dirtied by + * mmap'ed writes or via write(). + */ + if (nfs_clean_pages_on_close && vp->v_object) { + VM_OBJECT_LOCK(vp->v_object); + vm_object_page_clean(vp->v_object, 0, 0, 0); + VM_OBJECT_UNLOCK(vp->v_object); + } + mtx_lock(&np->n_mtx); + if (np->n_flag & NMODIFIED) { + mtx_unlock(&np->n_mtx); + if (NFS_ISV3(vp)) { + /* + * Under NFSv3 we have dirty buffers to dispose of. We + * must flush them to the NFS server. We have the option + * of waiting all the way through the commit rpc or just + * waiting for the initial write. The default is to only + * wait through the initial write so the data is in the + * server's cache, which is roughly similar to the state + * a standard disk subsystem leaves the file in on close(). + * + * We cannot clear the NMODIFIED bit in np->n_flag due to + * potential races with other processes, and certainly + * cannot clear it if we don't commit. + */ + int cm = nfsv3_commit_on_close ? 1 : 0; + error = nfs_flush(vp, MNT_WAIT, cm); + /* np->n_flag &= ~NMODIFIED; */ + } else + error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1); + mtx_lock(&np->n_mtx); + } + if (np->n_flag & NWRITEERR) { + np->n_flag &= ~NWRITEERR; + error = np->n_error; + } + mtx_unlock(&np->n_mtx); + } + if (nfs_directio_enable) + KASSERT((np->n_directio_asyncwr == 0), + ("nfs_close: dirty unflushed (%d) directio buffers\n", + np->n_directio_asyncwr)); + if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { + mtx_lock(&np->n_mtx); + KASSERT((np->n_directio_opens > 0), + ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); + np->n_directio_opens--; + if (np->n_directio_opens == 0) + np->n_flag &= ~NNONCACHE; + mtx_unlock(&np->n_mtx); + } + return (error); +} + +/* + * nfs getattr call from vfs. + */ +static int +nfs_getattr(struct vop_getattr_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + struct thread *td = curthread; + struct vattr *vap = ap->a_vap; + struct vattr vattr; + caddr_t bpos, dpos; + int error = 0; + struct mbuf *mreq, *mrep, *md, *mb; + int v3 = NFS_ISV3(vp); + + /* + * Update local times for special files. + */ + mtx_lock(&np->n_mtx); + if (np->n_flag & (NACC | NUPD)) + np->n_flag |= NCHG; + mtx_unlock(&np->n_mtx); + /* + * First look in the cache. + */ + if (nfs_getattrcache(vp, &vattr) == 0) + goto nfsmout; + if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) { + nfsstats.accesscache_misses++; + nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL); + if (nfs_getattrcache(vp, &vattr) == 0) + goto nfsmout; + } + nfsstats.rpccnt[NFSPROC_GETATTR]++; + mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred); + if (!error) { + nfsm_loadattr(vp, &vattr); + } + m_freem(mrep); +nfsmout: + vap->va_type = vattr.va_type; + vap->va_mode = vattr.va_mode; + vap->va_nlink = vattr.va_nlink; + vap->va_uid = vattr.va_uid; + vap->va_gid = vattr.va_gid; + vap->va_fsid = vattr.va_fsid; + vap->va_fileid = vattr.va_fileid; + vap->va_size = vattr.va_size; + vap->va_blocksize = vattr.va_blocksize; + vap->va_atime = vattr.va_atime; + vap->va_mtime = vattr.va_mtime; + vap->va_ctime = vattr.va_ctime; + vap->va_gen = vattr.va_gen; + vap->va_flags = vattr.va_flags; + vap->va_rdev = vattr.va_rdev; + vap->va_bytes = vattr.va_bytes; + vap->va_filerev = vattr.va_filerev; + + return (error); +} + +/* + * nfs setattr call. + */ +static int +nfs_setattr(struct vop_setattr_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + struct vattr *vap = ap->a_vap; + struct thread *td = curthread; + int error = 0; + u_quad_t tsize; + +#ifndef nolint + tsize = (u_quad_t)0; +#endif + + /* + * Setting of flags is not supported. + */ + if (vap->va_flags != VNOVAL) + return (EOPNOTSUPP); + + /* + * Disallow write attempts if the filesystem is mounted read-only. + */ + if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || + vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || + vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && + (vp->v_mount->mnt_flag & MNT_RDONLY)) { + error = EROFS; + goto out; + } + if (vap->va_size != VNOVAL) { + switch (vp->v_type) { + case VDIR: + return (EISDIR); + case VCHR: + case VBLK: + case VSOCK: + case VFIFO: + if (vap->va_mtime.tv_sec == VNOVAL && + vap->va_atime.tv_sec == VNOVAL && + vap->va_mode == (mode_t)VNOVAL && + vap->va_uid == (uid_t)VNOVAL && + vap->va_gid == (gid_t)VNOVAL) + return (0); + vap->va_size = VNOVAL; + break; + default: + /* + * Disallow write attempts if the filesystem is + * mounted read-only. + */ + if (vp->v_mount->mnt_flag & MNT_RDONLY) + return (EROFS); + /* + * We run vnode_pager_setsize() early (why?), + * we must set np->n_size now to avoid vinvalbuf + * V_SAVE races that might setsize a lower + * value. + */ + mtx_lock(&np->n_mtx); + tsize = np->n_size; + mtx_unlock(&np->n_mtx); + error = nfs_meta_setsize(vp, ap->a_cred, td, + vap->va_size); + mtx_lock(&np->n_mtx); + if (np->n_flag & NMODIFIED) { + tsize = np->n_size; + mtx_unlock(&np->n_mtx); + if (vap->va_size == 0) + error = nfs_vinvalbuf(vp, 0, td, 1); + else + error = nfs_vinvalbuf(vp, V_SAVE, td, 1); + if (error) { + vnode_pager_setsize(vp, tsize); + goto out; + } + } else + mtx_unlock(&np->n_mtx); + /* + * np->n_size has already been set to vap->va_size + * in nfs_meta_setsize(). We must set it again since + * nfs_loadattrcache() could be called through + * nfs_meta_setsize() and could modify np->n_size. + */ + mtx_lock(&np->n_mtx); + np->n_vattr.va_size = np->n_size = vap->va_size; + mtx_unlock(&np->n_mtx); + }; + } else { + mtx_lock(&np->n_mtx); + if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && + (np->n_flag & NMODIFIED) && vp->v_type == VREG) { + mtx_unlock(&np->n_mtx); + if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 && + (error == EINTR || error == EIO)) + return error; + } else + mtx_unlock(&np->n_mtx); + } + error = nfs_setattrrpc(vp, vap, ap->a_cred); + if (error && vap->va_size != VNOVAL) { + mtx_lock(&np->n_mtx); + np->n_size = np->n_vattr.va_size = tsize; + vnode_pager_setsize(vp, tsize); + mtx_unlock(&np->n_mtx); + } +out: + return (error); +} + +/* + * Do an nfs setattr rpc. + */ +static int +nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred) +{ + struct nfsv2_sattr *sp; + struct nfsnode *np = VTONFS(vp); + caddr_t bpos, dpos; + u_int32_t *tl; + int error = 0, i, wccflag = NFSV3_WCCRATTR; + struct mbuf *mreq, *mrep, *md, *mb; + int v3 = NFS_ISV3(vp); + + nfsstats.rpccnt[NFSPROC_SETATTR]++; + mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + if (v3) { + nfsm_v3attrbuild(vap, TRUE); + tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); + *tl = nfs_false; + } else { + sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); + if (vap->va_mode == (mode_t)VNOVAL) + sp->sa_mode = nfs_xdrneg1; + else + sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode); + if (vap->va_uid == (uid_t)VNOVAL) + sp->sa_uid = nfs_xdrneg1; + else + sp->sa_uid = txdr_unsigned(vap->va_uid); + if (vap->va_gid == (gid_t)VNOVAL) + sp->sa_gid = nfs_xdrneg1; + else + sp->sa_gid = txdr_unsigned(vap->va_gid); + sp->sa_size = txdr_unsigned(vap->va_size); + txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); + txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); + } + nfsm_request(vp, NFSPROC_SETATTR, curthread, cred); + if (v3) { + mtx_lock(&np->n_mtx); + for (i = 0; i < NFS_ACCESSCACHESIZE; i++) + np->n_accesscache[i].stamp = 0; + mtx_unlock(&np->n_mtx); + KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); + nfsm_wcc_data(vp, wccflag); + } else + nfsm_loadattr(vp, NULL); + m_freem(mrep); +nfsmout: + return (error); +} + +/* + * nfs lookup call, one step at a time... + * First look in cache + * If not found, unlock the directory nfsnode and do the rpc + */ +static int +nfs_lookup(struct vop_lookup_args *ap) +{ + struct componentname *cnp = ap->a_cnp; + struct vnode *dvp = ap->a_dvp; + struct vnode **vpp = ap->a_vpp; + struct mount *mp = dvp->v_mount; + struct vattr vattr; + struct timespec nctime; + int flags = cnp->cn_flags; + struct vnode *newvp; + struct nfsmount *nmp; + caddr_t bpos, dpos; + struct mbuf *mreq, *mrep, *md, *mb; + long len; + nfsfh_t *fhp; + struct nfsnode *np, *newnp; + int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks; + int v3 = NFS_ISV3(dvp); + struct thread *td = cnp->cn_thread; + + *vpp = NULLVP; + if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && + (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) + return (EROFS); + if (dvp->v_type != VDIR) + return (ENOTDIR); + nmp = VFSTONFS(mp); + np = VTONFS(dvp); + if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) { + *vpp = NULLVP; + return (error); + } + error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); + if (error > 0 && error != ENOENT) + return (error); + if (error == -1) { + /* + * Lookups of "." are special and always return the + * current directory. cache_lookup() already handles + * associated locking bookkeeping, etc. + */ + if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { + /* XXX: Is this really correct? */ + if (cnp->cn_nameiop != LOOKUP && + (flags & ISLASTCN)) + cnp->cn_flags |= SAVENAME; + return (0); + } + + /* + * We only accept a positive hit in the cache if the + * change time of the file matches our cached copy. + * Otherwise, we discard the cache entry and fallback + * to doing a lookup RPC. We also only trust cache + * entries for less than nm_nametimeo seconds. + * + * To better handle stale file handles and attributes, + * clear the attribute cache of this node if it is a + * leaf component, part of an open() call, and not + * locally modified before fetching the attributes. + * This should allow stale file handles to be detected + * here where we can fall back to a LOOKUP RPC to + * recover rather than having nfs_open() detect the + * stale file handle and failing open(2) with ESTALE. + */ + newvp = *vpp; + newnp = VTONFS(newvp); + if (!(nmp->nm_flag & NFSMNT_NOCTO) && + (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && + !(newnp->n_flag & NMODIFIED)) { + mtx_lock(&newnp->n_mtx); + newnp->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); + mtx_unlock(&newnp->n_mtx); + } + if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && + VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && + timespeccmp(&vattr.va_ctime, &nctime, ==)) { + nfsstats.lookupcache_hits++; + if (cnp->cn_nameiop != LOOKUP && + (flags & ISLASTCN)) + cnp->cn_flags |= SAVENAME; + return (0); + } + cache_purge(newvp); + if (dvp != newvp) + vput(newvp); + else + vrele(newvp); + *vpp = NULLVP; + } else if (error == ENOENT) { + if (dvp->v_iflag & VI_DOOMED) + return (ENOENT); + /* + * We only accept a negative hit in the cache if the + * modification time of the parent directory matches + * the cached copy in the name cache entry. + * Otherwise, we discard all of the negative cache + * entries for this directory. We also only trust + * negative cache entries for up to nm_negnametimeo + * seconds. + */ + if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && + VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && + timespeccmp(&vattr.va_mtime, &nctime, ==)) { + nfsstats.lookupcache_hits++; + return (ENOENT); + } + cache_purge_negative(dvp); + } + + attrflag = dattrflag = 0; + error = 0; + newvp = NULLVP; + nfsstats.lookupcache_misses++; + nfsstats.rpccnt[NFSPROC_LOOKUP]++; + len = cnp->cn_namelen; + mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, + NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); + nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred); + if (error) { + if (v3) { + nfsm_postop_attr_va(dvp, dattrflag, &vattr); + m_freem(mrep); + } + goto nfsmout; + } + nfsm_getfh(fhp, fhsize, v3); + + /* + * Handle RENAME case... + */ + if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { + if (NFS_CMPFH(np, fhp, fhsize)) { + m_freem(mrep); + return (EISDIR); + } + error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE); + if (error) { + m_freem(mrep); + return (error); + } + newvp = NFSTOV(np); + if (v3) { + nfsm_postop_attr(newvp, attrflag); + nfsm_postop_attr(dvp, attrflag); + } else + nfsm_loadattr(newvp, NULL); + *vpp = newvp; + m_freem(mrep); + cnp->cn_flags |= SAVENAME; + return (0); + } + + if (flags & ISDOTDOT) { + ltype = VOP_ISLOCKED(dvp); + error = vfs_busy(mp, MBF_NOWAIT); + if (error != 0) { + vfs_ref(mp); + VOP_UNLOCK(dvp, 0); + error = vfs_busy(mp, 0); + vn_lock(dvp, ltype | LK_RETRY); + vfs_rel(mp); + if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { + vfs_unbusy(mp); + error = ENOENT; + } + if (error != 0) { + m_freem(mrep); + return (error); + } + } + VOP_UNLOCK(dvp, 0); + error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags); + if (error == 0) + newvp = NFSTOV(np); + vfs_unbusy(mp); + if (newvp != dvp) + vn_lock(dvp, ltype | LK_RETRY); + if (dvp->v_iflag & VI_DOOMED) { + if (error == 0) { + if (newvp == dvp) + vrele(newvp); + else + vput(newvp); + } + error = ENOENT; + } + if (error) { + m_freem(mrep); + return (error); + } + } else if (NFS_CMPFH(np, fhp, fhsize)) { + VREF(dvp); + newvp = dvp; + } else { + error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags); + if (error) { + m_freem(mrep); + return (error); + } + newvp = NFSTOV(np); + + /* + * Flush the attribute cache when opening a leaf node + * to ensure that fresh attributes are fetched in + * nfs_open() if we are unable to fetch attributes + * from the LOOKUP reply. + */ + if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && + !(np->n_flag & NMODIFIED)) { + mtx_lock(&np->n_mtx); + np->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); + mtx_unlock(&np->n_mtx); + } + } + if (v3) { + nfsm_postop_attr_va(newvp, attrflag, &vattr); + nfsm_postop_attr(dvp, dattrflag); + } else { + nfsm_loadattr(newvp, &vattr); + attrflag = 1; + } + if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) + cnp->cn_flags |= SAVENAME; + if ((cnp->cn_flags & MAKEENTRY) && + (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && attrflag) { + cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime); + } + *vpp = newvp; + m_freem(mrep); +nfsmout: + if (error) { + if (newvp != NULLVP) { + vput(newvp); + *vpp = NULLVP; + } + + if (error != ENOENT) + goto done; + + /* The requested file was not found. */ + if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && + (flags & ISLASTCN)) { + /* + * XXX: UFS does a full VOP_ACCESS(dvp, + * VWRITE) here instead of just checking + * MNT_RDONLY. + */ + if (mp->mnt_flag & MNT_RDONLY) + return (EROFS); + cnp->cn_flags |= SAVENAME; + return (EJUSTRETURN); + } + + if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE && + dattrflag) { + /* + * Cache the modification time of the parent + * directory from the post-op attributes in + * the name cache entry. The negative cache + * entry will be ignored once the directory + * has changed. Don't bother adding the entry + * if the directory has already changed. + */ + mtx_lock(&np->n_mtx); + if (timespeccmp(&np->n_vattr.va_mtime, + &vattr.va_mtime, ==)) { + mtx_unlock(&np->n_mtx); + cache_enter_time(dvp, NULL, cnp, + &vattr.va_mtime); + } else + mtx_unlock(&np->n_mtx); + } + return (ENOENT); + } +done: + return (error); +} + +/* + * nfs read call. + * Just call nfs_bioread() to do the work. + */ +static int +nfs_read(struct vop_read_args *ap) +{ + struct vnode *vp = ap->a_vp; + + switch (vp->v_type) { + case VREG: + return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); + case VDIR: + return (EISDIR); + default: + return (EOPNOTSUPP); + } +} + +/* + * nfs readlink call + */ +static int +nfs_readlink(struct vop_readlink_args *ap) +{ + struct vnode *vp = ap->a_vp; + + if (vp->v_type != VLNK) + return (EINVAL); + return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred)); +} + +/* + * Do a readlink rpc. + * Called by nfs_doio() from below the buffer cache. + */ +int +nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) +{ + caddr_t bpos, dpos; + int error = 0, len, attrflag; + struct mbuf *mreq, *mrep, *md, *mb; + int v3 = NFS_ISV3(vp); + + nfsstats.rpccnt[NFSPROC_READLINK]++; + mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred); + if (v3) + nfsm_postop_attr(vp, attrflag); + if (!error) { + nfsm_strsiz(len, NFS_MAXPATHLEN); + if (len == NFS_MAXPATHLEN) { + struct nfsnode *np = VTONFS(vp); + mtx_lock(&np->n_mtx); + if (np->n_size && np->n_size < NFS_MAXPATHLEN) + len = np->n_size; + mtx_unlock(&np->n_mtx); + } + nfsm_mtouio(uiop, len); + } + m_freem(mrep); +nfsmout: + return (error); +} + +/* + * nfs read rpc call + * Ditto above + */ +int +nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) +{ + u_int32_t *tl; + caddr_t bpos, dpos; + struct mbuf *mreq, *mrep, *md, *mb; + struct nfsmount *nmp; + off_t end; + int error = 0, len, retlen, tsiz, eof, attrflag; + int v3 = NFS_ISV3(vp); + int rsize; + +#ifndef nolint + eof = 0; +#endif + nmp = VFSTONFS(vp->v_mount); + tsiz = uiop->uio_resid; + mtx_lock(&nmp->nm_mtx); + end = uiop->uio_offset + tsiz; + if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) { + mtx_unlock(&nmp->nm_mtx); + return (EFBIG); + } + rsize = nmp->nm_rsize; + mtx_unlock(&nmp->nm_mtx); + while (tsiz > 0) { + nfsstats.rpccnt[NFSPROC_READ]++; + len = (tsiz > rsize) ? rsize : tsiz; + mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3); + if (v3) { + txdr_hyper(uiop->uio_offset, tl); + *(tl + 2) = txdr_unsigned(len); + } else { + *tl++ = txdr_unsigned(uiop->uio_offset); + *tl++ = txdr_unsigned(len); + *tl = 0; + } + nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred); + if (v3) { + nfsm_postop_attr(vp, attrflag); + if (error) { + m_freem(mrep); + goto nfsmout; + } + tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED); + eof = fxdr_unsigned(int, *(tl + 1)); + } else { + nfsm_loadattr(vp, NULL); + } + nfsm_strsiz(retlen, rsize); + nfsm_mtouio(uiop, retlen); + m_freem(mrep); + tsiz -= retlen; + if (v3) { + if (eof || retlen == 0) { + tsiz = 0; + } + } else if (retlen < len) { + tsiz = 0; + } + } +nfsmout: + return (error); +} + +/* + * nfs write call + */ +int +nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, + int *iomode, int *must_commit) +{ + u_int32_t *tl; + int32_t backup; + caddr_t bpos, dpos; + struct mbuf *mreq, *mrep, *md, *mb; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + off_t end; + int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit; + int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC; + int wsize; + + KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1")); + *must_commit = 0; + tsiz = uiop->uio_resid; + mtx_lock(&nmp->nm_mtx); + end = uiop->uio_offset + tsiz; + if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) { + mtx_unlock(&nmp->nm_mtx); + return (EFBIG); + } + wsize = nmp->nm_wsize; + mtx_unlock(&nmp->nm_mtx); + while (tsiz > 0) { + nfsstats.rpccnt[NFSPROC_WRITE]++; + len = (tsiz > wsize) ? wsize : tsiz; + mreq = nfsm_reqhead(vp, NFSPROC_WRITE, + NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + if (v3) { + tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); + txdr_hyper(uiop->uio_offset, tl); + tl += 2; + *tl++ = txdr_unsigned(len); + *tl++ = txdr_unsigned(*iomode); + *tl = txdr_unsigned(len); + } else { + u_int32_t x; + + tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED); + /* Set both "begin" and "current" to non-garbage. */ + x = txdr_unsigned((u_int32_t)uiop->uio_offset); + *tl++ = x; /* "begin offset" */ + *tl++ = x; /* "current offset" */ + x = txdr_unsigned(len); + *tl++ = x; /* total to this offset */ + *tl = x; /* size of this write */ + } + nfsm_uiotom(uiop, len); + nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred); + if (v3) { + wccflag = NFSV3_WCCCHK; + nfsm_wcc_data(vp, wccflag); + if (!error) { + tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED + + NFSX_V3WRITEVERF); + rlen = fxdr_unsigned(int, *tl++); + if (rlen == 0) { + error = NFSERR_IO; + m_freem(mrep); + break; + } else if (rlen < len) { + backup = len - rlen; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base - + backup; + uiop->uio_iov->iov_len += backup; + uiop->uio_offset -= backup; + uiop->uio_resid += backup; + len = rlen; + } + commit = fxdr_unsigned(int, *tl++); + + /* + * Return the lowest committment level + * obtained by any of the RPCs. + */ + if (committed == NFSV3WRITE_FILESYNC) + committed = commit; + else if (committed == NFSV3WRITE_DATASYNC && + commit == NFSV3WRITE_UNSTABLE) + committed = commit; + mtx_lock(&nmp->nm_mtx); + if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){ + bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, + NFSX_V3WRITEVERF); + nmp->nm_state |= NFSSTA_HASWRITEVERF; + } else if (bcmp((caddr_t)tl, + (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) { + *must_commit = 1; + bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, + NFSX_V3WRITEVERF); + } + mtx_unlock(&nmp->nm_mtx); + } + } else { + nfsm_loadattr(vp, NULL); + } + if (wccflag) { + mtx_lock(&(VTONFS(vp))->n_mtx); + VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime; + mtx_unlock(&(VTONFS(vp))->n_mtx); + } + m_freem(mrep); + if (error) + break; + tsiz -= len; + } +nfsmout: + if (DOINGASYNC(vp)) + committed = NFSV3WRITE_FILESYNC; + *iomode = committed; + if (error) + uiop->uio_resid = tsiz; + return (error); +} + +/* + * nfs mknod rpc + * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the + * mode set to specify the file type and the size field for rdev. + */ +static int +nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, + struct vattr *vap) +{ + struct nfsv2_sattr *sp; + u_int32_t *tl; + struct vnode *newvp = NULL; + struct nfsnode *np = NULL; + struct vattr vattr; + caddr_t bpos, dpos; + int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0; + struct mbuf *mreq, *mrep, *md, *mb; + u_int32_t rdev; + int v3 = NFS_ISV3(dvp); + + if (vap->va_type == VCHR || vap->va_type == VBLK) + rdev = txdr_unsigned(vap->va_rdev); + else if (vap->va_type == VFIFO || vap->va_type == VSOCK) + rdev = nfs_xdrneg1; + else { + return (EOPNOTSUPP); + } + if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) + return (error); + nfsstats.rpccnt[NFSPROC_MKNOD]++; + mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED + + + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); + if (v3) { + tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); + *tl++ = vtonfsv3_type(vap->va_type); + nfsm_v3attrbuild(vap, FALSE); + if (vap->va_type == VCHR || vap->va_type == VBLK) { + tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); + *tl++ = txdr_unsigned(major(vap->va_rdev)); + *tl = txdr_unsigned(minor(vap->va_rdev)); + } + } else { + sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); + sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); + sp->sa_uid = nfs_xdrneg1; + sp->sa_gid = nfs_xdrneg1; + sp->sa_size = rdev; + txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); + txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); + } + nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred); + if (!error) { + nfsm_mtofh(dvp, newvp, v3, gotvp); + if (!gotvp) { + if (newvp) { + vput(newvp); + newvp = NULL; + } + error = nfs_lookitup(dvp, cnp->cn_nameptr, + cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); + if (!error) + newvp = NFSTOV(np); + } + } + if (v3) + nfsm_wcc_data(dvp, wccflag); + m_freem(mrep); +nfsmout: + if (error) { + if (newvp) + vput(newvp); + } else { + *vpp = newvp; + } + mtx_lock(&(VTONFS(dvp))->n_mtx); + VTONFS(dvp)->n_flag |= NMODIFIED; + if (!wccflag) { + VTONFS(dvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); + } + mtx_unlock(&(VTONFS(dvp))->n_mtx); + return (error); +} + +/* + * nfs mknod vop + * just call nfs_mknodrpc() to do the work. + */ +/* ARGSUSED */ +static int +nfs_mknod(struct vop_mknod_args *ap) +{ + return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); +} + +static u_long create_verf; +/* + * nfs file create call + */ +static int +nfs_create(struct vop_create_args *ap) +{ + struct vnode *dvp = ap->a_dvp; + struct vattr *vap = ap->a_vap; + struct componentname *cnp = ap->a_cnp; + struct nfsv2_sattr *sp; + u_int32_t *tl; + struct nfsnode *np = NULL; + struct vnode *newvp = NULL; + caddr_t bpos, dpos; + int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0; + struct mbuf *mreq, *mrep, *md, *mb; + struct vattr vattr; + int v3 = NFS_ISV3(dvp); + + /* + * Oops, not for me.. + */ + if (vap->va_type == VSOCK) { + error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap); + return (error); + } + + if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) { + return (error); + } + if (vap->va_vaflags & VA_EXCLUSIVE) + fmode |= O_EXCL; +again: + nfsstats.rpccnt[NFSPROC_CREATE]++; + mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED + + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); + if (v3) { + tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED); + if (fmode & O_EXCL) { + *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE); + tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF); +#ifdef INET + CURVNET_SET(CRED_TO_VNET(cnp->cn_cred)); + IN_IFADDR_RLOCK(); + if (!TAILQ_EMPTY(&V_in_ifaddrhead)) + *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr; + else +#endif + *tl++ = create_verf; +#ifdef INET + IN_IFADDR_RUNLOCK(); + CURVNET_RESTORE(); +#endif + *tl = ++create_verf; + } else { + *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED); + nfsm_v3attrbuild(vap, FALSE); + } + } else { + sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); + sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); + sp->sa_uid = nfs_xdrneg1; + sp->sa_gid = nfs_xdrneg1; + sp->sa_size = 0; + txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); + txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); + } + nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred); + if (!error) { + nfsm_mtofh(dvp, newvp, v3, gotvp); + if (!gotvp) { + if (newvp) { + vput(newvp); + newvp = NULL; + } + error = nfs_lookitup(dvp, cnp->cn_nameptr, + cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); + if (!error) + newvp = NFSTOV(np); + } + } + if (v3) + nfsm_wcc_data(dvp, wccflag); + m_freem(mrep); +nfsmout: + if (error) { + if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { + fmode &= ~O_EXCL; + goto again; + } + if (newvp) + vput(newvp); + } else if (v3 && (fmode & O_EXCL)) { + /* + * We are normally called with only a partially initialized + * VAP. Since the NFSv3 spec says that server may use the + * file attributes to store the verifier, the spec requires + * us to do a SETATTR RPC. FreeBSD servers store the verifier + * in atime, but we can't really assume that all servers will + * so we ensure that our SETATTR sets both atime and mtime. + */ + if (vap->va_mtime.tv_sec == VNOVAL) + vfs_timestamp(&vap->va_mtime); + if (vap->va_atime.tv_sec == VNOVAL) + vap->va_atime = vap->va_mtime; + error = nfs_setattrrpc(newvp, vap, cnp->cn_cred); + if (error) + vput(newvp); + } + if (!error) { + *ap->a_vpp = newvp; + } + mtx_lock(&(VTONFS(dvp))->n_mtx); + VTONFS(dvp)->n_flag |= NMODIFIED; + if (!wccflag) { + VTONFS(dvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); + } + mtx_unlock(&(VTONFS(dvp))->n_mtx); + return (error); +} + +/* + * nfs file remove call + * To try and make nfs semantics closer to ufs semantics, a file that has + * other processes using the vnode is renamed instead of removed and then + * removed later on the last close. + * - If v_usecount > 1 + * If a rename is not already in the works + * call nfs_sillyrename() to set it up + * else + * do the remove rpc + */ +static int +nfs_remove(struct vop_remove_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct vnode *dvp = ap->a_dvp; + struct componentname *cnp = ap->a_cnp; + struct nfsnode *np = VTONFS(vp); + int error = 0; + struct vattr vattr; + + KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); + KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); + if (vp->v_type == VDIR) + error = EPERM; + else if (vrefcnt(vp) == 1 || (np->n_sillyrename && + !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) { + /* + * Purge the name cache so that the chance of a lookup for + * the name succeeding while the remove is in progress is + * minimized. Without node locking it can still happen, such + * that an I/O op returns ESTALE, but since you get this if + * another host removes the file.. + */ + cache_purge(vp); + /* + * throw away biocache buffers, mainly to avoid + * unnecessary delayed writes later. + */ + error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1); + /* Do the rpc */ + if (error != EINTR && error != EIO) + error = nfs_removerpc(dvp, cnp->cn_nameptr, + cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); + /* + * Kludge City: If the first reply to the remove rpc is lost.. + * the reply to the retransmitted request will be ENOENT + * since the file was in fact removed + * Therefore, we cheat and return success. + */ + if (error == ENOENT) + error = 0; + } else if (!np->n_sillyrename) + error = nfs_sillyrename(dvp, vp, cnp); + mtx_lock(&np->n_mtx); + np->n_attrstamp = 0; + mtx_unlock(&np->n_mtx); + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + return (error); +} + +/* + * nfs file remove rpc called from nfs_inactive + */ +int +nfs_removeit(struct sillyrename *sp) +{ + /* + * Make sure that the directory vnode is still valid. + * XXX we should lock sp->s_dvp here. + */ + if (sp->s_dvp->v_type == VBAD) + return (0); + return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred, + NULL)); +} + +/* + * Nfs remove rpc, called from nfs_remove() and nfs_removeit(). + */ +static int +nfs_removerpc(struct vnode *dvp, const char *name, int namelen, + struct ucred *cred, struct thread *td) +{ + caddr_t bpos, dpos; + int error = 0, wccflag = NFSV3_WCCRATTR; + struct mbuf *mreq, *mrep, *md, *mb; + int v3 = NFS_ISV3(dvp); + + nfsstats.rpccnt[NFSPROC_REMOVE]++; + mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE, + NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(name, namelen, NFS_MAXNAMLEN); + nfsm_request(dvp, NFSPROC_REMOVE, td, cred); + if (v3) + nfsm_wcc_data(dvp, wccflag); + m_freem(mrep); +nfsmout: + mtx_lock(&(VTONFS(dvp))->n_mtx); + VTONFS(dvp)->n_flag |= NMODIFIED; + if (!wccflag) { + VTONFS(dvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); + } + mtx_unlock(&(VTONFS(dvp))->n_mtx); + return (error); +} + +/* + * nfs file rename call + */ +static int +nfs_rename(struct vop_rename_args *ap) +{ + struct vnode *fvp = ap->a_fvp; + struct vnode *tvp = ap->a_tvp; + struct vnode *fdvp = ap->a_fdvp; + struct vnode *tdvp = ap->a_tdvp; + struct componentname *tcnp = ap->a_tcnp; + struct componentname *fcnp = ap->a_fcnp; + int error; + + KASSERT((tcnp->cn_flags & HASBUF) != 0 && + (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); + /* Check for cross-device rename */ + if ((fvp->v_mount != tdvp->v_mount) || + (tvp && (fvp->v_mount != tvp->v_mount))) { + error = EXDEV; + goto out; + } + + if (fvp == tvp) { + nfs_printf("nfs_rename: fvp == tvp (can't happen)\n"); + error = 0; + goto out; + } + if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) + goto out; + + /* + * We have to flush B_DELWRI data prior to renaming + * the file. If we don't, the delayed-write buffers + * can be flushed out later after the file has gone stale + * under NFSV3. NFSV2 does not have this problem because + * ( as far as I can tell ) it flushes dirty buffers more + * often. + * + * Skip the rename operation if the fsync fails, this can happen + * due to the server's volume being full, when we pushed out data + * that was written back to our cache earlier. Not checking for + * this condition can result in potential (silent) data loss. + */ + error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); + VOP_UNLOCK(fvp, 0); + if (!error && tvp) + error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); + if (error) + goto out; + + /* + * If the tvp exists and is in use, sillyrename it before doing the + * rename of the new file over it. + * XXX Can't sillyrename a directory. + */ + if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && + tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { + vput(tvp); + tvp = NULL; + } + + error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen, + tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, + tcnp->cn_thread); + + if (fvp->v_type == VDIR) { + if (tvp != NULL && tvp->v_type == VDIR) + cache_purge(tdvp); + cache_purge(fdvp); + } + +out: + if (tdvp == tvp) + vrele(tdvp); + else + vput(tdvp); + if (tvp) + vput(tvp); + vrele(fdvp); + vrele(fvp); + /* + * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. + */ + if (error == ENOENT) + error = 0; + return (error); +} + +/* + * nfs file rename rpc called from nfs_remove() above + */ +static int +nfs_renameit(struct vnode *sdvp, struct componentname *scnp, + struct sillyrename *sp) +{ + + return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp, + sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread)); +} + +/* + * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). + */ +static int +nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, + struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred, + struct thread *td) +{ + caddr_t bpos, dpos; + int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR; + struct mbuf *mreq, *mrep, *md, *mb; + int v3 = NFS_ISV3(fdvp); + + nfsstats.rpccnt[NFSPROC_RENAME]++; + mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME, + (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) + + nfsm_rndup(tnamelen)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(fdvp, v3); + nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN); + nfsm_fhtom(tdvp, v3); + nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN); + nfsm_request(fdvp, NFSPROC_RENAME, td, cred); + if (v3) { + nfsm_wcc_data(fdvp, fwccflag); + nfsm_wcc_data(tdvp, twccflag); + } + m_freem(mrep); +nfsmout: + mtx_lock(&(VTONFS(fdvp))->n_mtx); + VTONFS(fdvp)->n_flag |= NMODIFIED; + mtx_unlock(&(VTONFS(fdvp))->n_mtx); + mtx_lock(&(VTONFS(tdvp))->n_mtx); + VTONFS(tdvp)->n_flag |= NMODIFIED; + mtx_unlock(&(VTONFS(tdvp))->n_mtx); + if (!fwccflag) { + VTONFS(fdvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); + } + if (!twccflag) { + VTONFS(tdvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); + } + return (error); +} + +/* + * nfs hard link create call + */ +static int +nfs_link(struct vop_link_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct vnode *tdvp = ap->a_tdvp; + struct componentname *cnp = ap->a_cnp; + caddr_t bpos, dpos; + int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0; + struct mbuf *mreq, *mrep, *md, *mb; + int v3; + + if (vp->v_mount != tdvp->v_mount) { + return (EXDEV); + } + + /* + * Push all writes to the server, so that the attribute cache + * doesn't get "out of sync" with the server. + * XXX There should be a better way! + */ + VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); + + v3 = NFS_ISV3(vp); + nfsstats.rpccnt[NFSPROC_LINK]++; + mreq = nfsm_reqhead(vp, NFSPROC_LINK, + NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + nfsm_fhtom(tdvp, v3); + nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); + nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred); + if (v3) { + nfsm_postop_attr(vp, attrflag); + nfsm_wcc_data(tdvp, wccflag); + } + m_freem(mrep); +nfsmout: + mtx_lock(&(VTONFS(tdvp))->n_mtx); + VTONFS(tdvp)->n_flag |= NMODIFIED; + mtx_unlock(&(VTONFS(tdvp))->n_mtx); + if (!attrflag) { + VTONFS(vp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); + } + if (!wccflag) { + VTONFS(tdvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); + } + return (error); +} + +/* + * nfs symbolic link create call + */ +static int +nfs_symlink(struct vop_symlink_args *ap) +{ + struct vnode *dvp = ap->a_dvp; + struct vattr *vap = ap->a_vap; + struct componentname *cnp = ap->a_cnp; + struct nfsv2_sattr *sp; + caddr_t bpos, dpos; + int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp; + struct mbuf *mreq, *mrep, *md, *mb; + struct vnode *newvp = NULL; + int v3 = NFS_ISV3(dvp); + + nfsstats.rpccnt[NFSPROC_SYMLINK]++; + slen = strlen(ap->a_target); + mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED + + nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); + if (v3) { + nfsm_v3attrbuild(vap, FALSE); + } + nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN); + if (!v3) { + sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); + sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode); + sp->sa_uid = nfs_xdrneg1; + sp->sa_gid = nfs_xdrneg1; + sp->sa_size = nfs_xdrneg1; + txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); + txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); + } + + /* + * Issue the NFS request and get the rpc response. + * + * Only NFSv3 responses returning an error of 0 actually return + * a file handle that can be converted into newvp without having + * to do an extra lookup rpc. + */ + nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred); + if (v3) { + if (error == 0) + nfsm_mtofh(dvp, newvp, v3, gotvp); + nfsm_wcc_data(dvp, wccflag); + } + + /* + * out code jumps -> here, mrep is also freed. + */ + + m_freem(mrep); +nfsmout: + + /* + * If we do not have an error and we could not extract the newvp from + * the response due to the request being NFSv2, we have to do a + * lookup in order to obtain a newvp to return. + */ + if (error == 0 && newvp == NULL) { + struct nfsnode *np = NULL; + + error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, + cnp->cn_cred, cnp->cn_thread, &np); + if (!error) + newvp = NFSTOV(np); + } + if (error) { + if (newvp) + vput(newvp); + } else { + *ap->a_vpp = newvp; + } + mtx_lock(&(VTONFS(dvp))->n_mtx); + VTONFS(dvp)->n_flag |= NMODIFIED; + mtx_unlock(&(VTONFS(dvp))->n_mtx); + if (!wccflag) { + VTONFS(dvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); + } + return (error); +} + +/* + * nfs make dir call + */ +static int +nfs_mkdir(struct vop_mkdir_args *ap) +{ + struct vnode *dvp = ap->a_dvp; + struct vattr *vap = ap->a_vap; + struct componentname *cnp = ap->a_cnp; + struct nfsv2_sattr *sp; + int len; + struct nfsnode *np = NULL; + struct vnode *newvp = NULL; + caddr_t bpos, dpos; + int error = 0, wccflag = NFSV3_WCCRATTR; + int gotvp = 0; + struct mbuf *mreq, *mrep, *md, *mb; + struct vattr vattr; + int v3 = NFS_ISV3(dvp); + + if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) + return (error); + len = cnp->cn_namelen; + nfsstats.rpccnt[NFSPROC_MKDIR]++; + mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR, + NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN); + if (v3) { + nfsm_v3attrbuild(vap, FALSE); + } else { + sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR); + sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode); + sp->sa_uid = nfs_xdrneg1; + sp->sa_gid = nfs_xdrneg1; + sp->sa_size = nfs_xdrneg1; + txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); + txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); + } + nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred); + if (!error) + nfsm_mtofh(dvp, newvp, v3, gotvp); + if (v3) + nfsm_wcc_data(dvp, wccflag); + m_freem(mrep); +nfsmout: + mtx_lock(&(VTONFS(dvp))->n_mtx); + VTONFS(dvp)->n_flag |= NMODIFIED; + mtx_unlock(&(VTONFS(dvp))->n_mtx); + if (!wccflag) { + VTONFS(dvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); + } + if (error == 0 && newvp == NULL) { + error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred, + cnp->cn_thread, &np); + if (!error) { + newvp = NFSTOV(np); + if (newvp->v_type != VDIR) + error = EEXIST; + } + } + if (error) { + if (newvp) + vput(newvp); + } else + *ap->a_vpp = newvp; + return (error); +} + +/* + * nfs remove directory call + */ +static int +nfs_rmdir(struct vop_rmdir_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct vnode *dvp = ap->a_dvp; + struct componentname *cnp = ap->a_cnp; + caddr_t bpos, dpos; + int error = 0, wccflag = NFSV3_WCCRATTR; + struct mbuf *mreq, *mrep, *md, *mb; + int v3 = NFS_ISV3(dvp); + + if (dvp == vp) + return (EINVAL); + nfsstats.rpccnt[NFSPROC_RMDIR]++; + mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR, + NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN); + nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred); + if (v3) + nfsm_wcc_data(dvp, wccflag); + m_freem(mrep); +nfsmout: + mtx_lock(&(VTONFS(dvp))->n_mtx); + VTONFS(dvp)->n_flag |= NMODIFIED; + mtx_unlock(&(VTONFS(dvp))->n_mtx); + if (!wccflag) { + VTONFS(dvp)->n_attrstamp = 0; + KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); + } + cache_purge(dvp); + cache_purge(vp); + /* + * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. + */ + if (error == ENOENT) + error = 0; + return (error); +} + +/* + * nfs readdir call + */ +static int +nfs_readdir(struct vop_readdir_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + struct uio *uio = ap->a_uio; + int tresid, error = 0; + struct vattr vattr; + + if (vp->v_type != VDIR) + return(EPERM); + + /* + * First, check for hit on the EOF offset cache + */ + if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && + (np->n_flag & NMODIFIED) == 0) { + if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { + mtx_lock(&np->n_mtx); + if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { + mtx_unlock(&np->n_mtx); + nfsstats.direofcache_hits++; + goto out; + } else + mtx_unlock(&np->n_mtx); + } + } + + /* + * Call nfs_bioread() to do the real work. + */ + tresid = uio->uio_resid; + error = nfs_bioread(vp, uio, 0, ap->a_cred); + + if (!error && uio->uio_resid == tresid) { + nfsstats.direofcache_misses++; + } +out: + return (error); +} + +/* + * Readdir rpc call. + * Called from below the buffer cache by nfs_doio(). + */ +int +nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) +{ + int len, left; + struct dirent *dp = NULL; + u_int32_t *tl; + caddr_t cp; + nfsuint64 *cookiep; + caddr_t bpos, dpos; + struct mbuf *mreq, *mrep, *md, *mb; + nfsuint64 cookie; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + struct nfsnode *dnp = VTONFS(vp); + u_quad_t fileno; + int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; + int attrflag; + int v3 = NFS_ISV3(vp); + + KASSERT(uiop->uio_iovcnt == 1 && + (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && + (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, + ("nfs readdirrpc bad uio")); + + /* + * If there is no cookie, assume directory was stale. + */ + nfs_dircookie_lock(dnp); + cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); + if (cookiep) { + cookie = *cookiep; + nfs_dircookie_unlock(dnp); + } else { + nfs_dircookie_unlock(dnp); + return (NFSERR_BAD_COOKIE); + } + + /* + * Loop around doing readdir rpc's of size nm_readdirsize + * truncated to a multiple of DIRBLKSIZ. + * The stopping criteria is EOF or buffer full. + */ + while (more_dirs && bigenough) { + nfsstats.rpccnt[NFSPROC_READDIR]++; + mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) + + NFSX_READDIR(v3)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, v3); + if (v3) { + tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED); + *tl++ = cookie.nfsuquad[0]; + *tl++ = cookie.nfsuquad[1]; + mtx_lock(&dnp->n_mtx); + *tl++ = dnp->n_cookieverf.nfsuquad[0]; + *tl++ = dnp->n_cookieverf.nfsuquad[1]; + mtx_unlock(&dnp->n_mtx); + } else { + tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED); + *tl++ = cookie.nfsuquad[0]; + } + *tl = txdr_unsigned(nmp->nm_readdirsize); + nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred); + if (v3) { + nfsm_postop_attr(vp, attrflag); + if (!error) { + tl = nfsm_dissect(u_int32_t *, + 2 * NFSX_UNSIGNED); + mtx_lock(&dnp->n_mtx); + dnp->n_cookieverf.nfsuquad[0] = *tl++; + dnp->n_cookieverf.nfsuquad[1] = *tl; + mtx_unlock(&dnp->n_mtx); + } else { + m_freem(mrep); + goto nfsmout; + } + } + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + more_dirs = fxdr_unsigned(int, *tl); + + /* loop thru the dir entries, doctoring them to 4bsd form */ + while (more_dirs && bigenough) { + if (v3) { + tl = nfsm_dissect(u_int32_t *, + 3 * NFSX_UNSIGNED); + fileno = fxdr_hyper(tl); + len = fxdr_unsigned(int, *(tl + 2)); + } else { + tl = nfsm_dissect(u_int32_t *, + 2 * NFSX_UNSIGNED); + fileno = fxdr_unsigned(u_quad_t, *tl++); + len = fxdr_unsigned(int, *tl); + } + if (len <= 0 || len > NFS_MAXNAMLEN) { + error = EBADRPC; + m_freem(mrep); + goto nfsmout; + } + tlen = nfsm_rndup(len); + if (tlen == len) + tlen += 4; /* To ensure null termination */ + left = DIRBLKSIZ - blksiz; + if ((tlen + DIRHDSIZ) > left) { + dp->d_reclen += left; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + left; + uiop->uio_iov->iov_len -= left; + uiop->uio_offset += left; + uiop->uio_resid -= left; + blksiz = 0; + } + if ((tlen + DIRHDSIZ) > uiop->uio_resid) + bigenough = 0; + if (bigenough) { + dp = (struct dirent *)uiop->uio_iov->iov_base; + dp->d_fileno = (int)fileno; + dp->d_namlen = len; + dp->d_reclen = tlen + DIRHDSIZ; + dp->d_type = DT_UNKNOWN; + blksiz += dp->d_reclen; + if (blksiz == DIRBLKSIZ) + blksiz = 0; + uiop->uio_offset += DIRHDSIZ; + uiop->uio_resid -= DIRHDSIZ; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + DIRHDSIZ; + uiop->uio_iov->iov_len -= DIRHDSIZ; + nfsm_mtouio(uiop, len); + cp = uiop->uio_iov->iov_base; + tlen -= len; + *cp = '\0'; /* null terminate */ + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + tlen; + uiop->uio_iov->iov_len -= tlen; + uiop->uio_offset += tlen; + uiop->uio_resid -= tlen; + } else + nfsm_adv(nfsm_rndup(len)); + if (v3) { + tl = nfsm_dissect(u_int32_t *, + 3 * NFSX_UNSIGNED); + } else { + tl = nfsm_dissect(u_int32_t *, + 2 * NFSX_UNSIGNED); + } + if (bigenough) { + cookie.nfsuquad[0] = *tl++; + if (v3) + cookie.nfsuquad[1] = *tl++; + } else if (v3) + tl += 2; + else + tl++; + more_dirs = fxdr_unsigned(int, *tl); + } + /* + * If at end of rpc data, get the eof boolean + */ + if (!more_dirs) { + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + more_dirs = (fxdr_unsigned(int, *tl) == 0); + } + m_freem(mrep); + } + /* + * Fill last record, iff any, out to a multiple of DIRBLKSIZ + * by increasing d_reclen for the last record. + */ + if (blksiz > 0) { + left = DIRBLKSIZ - blksiz; + dp->d_reclen += left; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + left; + uiop->uio_iov->iov_len -= left; + uiop->uio_offset += left; + uiop->uio_resid -= left; + } + + /* + * We are now either at the end of the directory or have filled the + * block. + */ + if (bigenough) + dnp->n_direofoffset = uiop->uio_offset; + else { + if (uiop->uio_resid > 0) + nfs_printf("EEK! readdirrpc resid > 0\n"); + nfs_dircookie_lock(dnp); + cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); + *cookiep = cookie; + nfs_dircookie_unlock(dnp); + } +nfsmout: + return (error); +} + +/* + * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc(). + */ +int +nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) +{ + int len, left; + struct dirent *dp; + u_int32_t *tl; + caddr_t cp; + struct vnode *newvp; + nfsuint64 *cookiep; + caddr_t bpos, dpos, dpossav1, dpossav2; + struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2; + struct nameidata nami, *ndp = &nami; + struct componentname *cnp = &ndp->ni_cnd; + nfsuint64 cookie; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + struct nfsnode *dnp = VTONFS(vp), *np; + struct vattr vattr; + nfsfh_t *fhp; + u_quad_t fileno; + int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i; + int attrflag, fhsize; + +#ifndef nolint + dp = NULL; +#endif + KASSERT(uiop->uio_iovcnt == 1 && + (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && + (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, + ("nfs readdirplusrpc bad uio")); + ndp->ni_dvp = vp; + newvp = NULLVP; + + /* + * If there is no cookie, assume directory was stale. + */ + nfs_dircookie_lock(dnp); + cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0); + if (cookiep) { + cookie = *cookiep; + nfs_dircookie_unlock(dnp); + } else { + nfs_dircookie_unlock(dnp); + return (NFSERR_BAD_COOKIE); + } + /* + * Loop around doing readdir rpc's of size nm_readdirsize + * truncated to a multiple of DIRBLKSIZ. + * The stopping criteria is EOF or buffer full. + */ + while (more_dirs && bigenough) { + nfsstats.rpccnt[NFSPROC_READDIRPLUS]++; + mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS, + NFSX_FH(1) + 6 * NFSX_UNSIGNED); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, 1); + tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED); + *tl++ = cookie.nfsuquad[0]; + *tl++ = cookie.nfsuquad[1]; + mtx_lock(&dnp->n_mtx); + *tl++ = dnp->n_cookieverf.nfsuquad[0]; + *tl++ = dnp->n_cookieverf.nfsuquad[1]; + mtx_unlock(&dnp->n_mtx); + *tl++ = txdr_unsigned(nmp->nm_readdirsize); + *tl = txdr_unsigned(nmp->nm_rsize); + nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred); + nfsm_postop_attr(vp, attrflag); + if (error) { + m_freem(mrep); + goto nfsmout; + } + tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); + mtx_lock(&dnp->n_mtx); + dnp->n_cookieverf.nfsuquad[0] = *tl++; + dnp->n_cookieverf.nfsuquad[1] = *tl++; + mtx_unlock(&dnp->n_mtx); + more_dirs = fxdr_unsigned(int, *tl); + + /* loop thru the dir entries, doctoring them to 4bsd form */ + while (more_dirs && bigenough) { + tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); + fileno = fxdr_hyper(tl); + len = fxdr_unsigned(int, *(tl + 2)); + if (len <= 0 || len > NFS_MAXNAMLEN) { + error = EBADRPC; + m_freem(mrep); + goto nfsmout; + } + tlen = nfsm_rndup(len); + if (tlen == len) + tlen += 4; /* To ensure null termination*/ + left = DIRBLKSIZ - blksiz; + if ((tlen + DIRHDSIZ) > left) { + dp->d_reclen += left; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + left; + uiop->uio_iov->iov_len -= left; + uiop->uio_offset += left; + uiop->uio_resid -= left; + blksiz = 0; + } + if ((tlen + DIRHDSIZ) > uiop->uio_resid) + bigenough = 0; + if (bigenough) { + dp = (struct dirent *)uiop->uio_iov->iov_base; + dp->d_fileno = (int)fileno; + dp->d_namlen = len; + dp->d_reclen = tlen + DIRHDSIZ; + dp->d_type = DT_UNKNOWN; + blksiz += dp->d_reclen; + if (blksiz == DIRBLKSIZ) + blksiz = 0; + uiop->uio_offset += DIRHDSIZ; + uiop->uio_resid -= DIRHDSIZ; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + DIRHDSIZ; + uiop->uio_iov->iov_len -= DIRHDSIZ; + cnp->cn_nameptr = uiop->uio_iov->iov_base; + cnp->cn_namelen = len; + nfsm_mtouio(uiop, len); + cp = uiop->uio_iov->iov_base; + tlen -= len; + *cp = '\0'; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + tlen; + uiop->uio_iov->iov_len -= tlen; + uiop->uio_offset += tlen; + uiop->uio_resid -= tlen; + } else + nfsm_adv(nfsm_rndup(len)); + tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED); + if (bigenough) { + cookie.nfsuquad[0] = *tl++; + cookie.nfsuquad[1] = *tl++; + } else + tl += 2; + + /* + * Since the attributes are before the file handle + * (sigh), we must skip over the attributes and then + * come back and get them. + */ + attrflag = fxdr_unsigned(int, *tl); + if (attrflag) { + dpossav1 = dpos; + mdsav1 = md; + nfsm_adv(NFSX_V3FATTR); + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + doit = fxdr_unsigned(int, *tl); + /* + * Skip loading the attrs for "..". There's a + * race between loading the attrs here and + * lookups that look for the directory currently + * being read (in the parent). We try to acquire + * the exclusive lock on ".." here, owning the + * lock on the directory being read. Lookup will + * hold the lock on ".." and try to acquire the + * lock on the directory being read. + * + * There are other ways of fixing this, one would + * be to do a trylock on the ".." vnode and skip + * loading the attrs on ".." if it happens to be + * locked by another process. But skipping the + * attrload on ".." seems the easiest option. + */ + if (strcmp(dp->d_name, "..") == 0) { + doit = 0; + /* + * We've already skipped over the attrs, + * skip over the filehandle. And store d_type + * as VDIR. + */ + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + i = fxdr_unsigned(int, *tl); + nfsm_adv(nfsm_rndup(i)); + dp->d_type = IFTODT(VTTOIF(VDIR)); + } + if (doit) { + nfsm_getfh(fhp, fhsize, 1); + if (NFS_CMPFH(dnp, fhp, fhsize)) { + VREF(vp); + newvp = vp; + np = dnp; + } else { + error = nfs_nget(vp->v_mount, fhp, + fhsize, &np, LK_EXCLUSIVE); + if (error) + doit = 0; + else + newvp = NFSTOV(np); + } + } + if (doit && bigenough) { + dpossav2 = dpos; + dpos = dpossav1; + mdsav2 = md; + md = mdsav1; + nfsm_loadattr(newvp, &vattr); + dpos = dpossav2; + md = mdsav2; + dp->d_type = IFTODT(VTTOIF(vattr.va_type)); + ndp->ni_vp = newvp; + cache_enter_time(ndp->ni_dvp, ndp->ni_vp, cnp, + &vattr.va_ctime); + } + } else { + /* Just skip over the file handle */ + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + i = fxdr_unsigned(int, *tl); + if (i) { + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + fhsize = fxdr_unsigned(int, *tl); + nfsm_adv(nfsm_rndup(fhsize)); + } + } + if (newvp != NULLVP) { + if (newvp == vp) + vrele(newvp); + else + vput(newvp); + newvp = NULLVP; + } + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + more_dirs = fxdr_unsigned(int, *tl); + } + /* + * If at end of rpc data, get the eof boolean + */ + if (!more_dirs) { + tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED); + more_dirs = (fxdr_unsigned(int, *tl) == 0); + } + m_freem(mrep); + } + /* + * Fill last record, iff any, out to a multiple of DIRBLKSIZ + * by increasing d_reclen for the last record. + */ + if (blksiz > 0) { + left = DIRBLKSIZ - blksiz; + dp->d_reclen += left; + uiop->uio_iov->iov_base = + (char *)uiop->uio_iov->iov_base + left; + uiop->uio_iov->iov_len -= left; + uiop->uio_offset += left; + uiop->uio_resid -= left; + } + + /* + * We are now either at the end of the directory or have filled the + * block. + */ + if (bigenough) + dnp->n_direofoffset = uiop->uio_offset; + else { + if (uiop->uio_resid > 0) + nfs_printf("EEK! readdirplusrpc resid > 0\n"); + nfs_dircookie_lock(dnp); + cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1); + *cookiep = cookie; + nfs_dircookie_unlock(dnp); + } +nfsmout: + if (newvp != NULLVP) { + if (newvp == vp) + vrele(newvp); + else + vput(newvp); + newvp = NULLVP; + } + return (error); +} + +/* + * Silly rename. To make the NFS filesystem that is stateless look a little + * more like the "ufs" a remove of an active vnode is translated to a rename + * to a funny looking filename that is removed by nfs_inactive on the + * nfsnode. There is the potential for another process on a different client + * to create the same funny name between the nfs_lookitup() fails and the + * nfs_rename() completes, but... + */ +static int +nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) +{ + struct sillyrename *sp; + struct nfsnode *np; + int error; + short pid; + unsigned int lticks; + + cache_purge(dvp); + np = VTONFS(vp); + KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); + sp = malloc(sizeof (struct sillyrename), + M_NFSREQ, M_WAITOK); + sp->s_cred = crhold(cnp->cn_cred); + sp->s_dvp = dvp; + sp->s_removeit = nfs_removeit; + VREF(dvp); + + /* + * Fudge together a funny name. + * Changing the format of the funny name to accomodate more + * sillynames per directory. + * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is + * CPU ticks since boot. + */ + pid = cnp->cn_thread->td_proc->p_pid; + lticks = (unsigned int)ticks; + for ( ; ; ) { + sp->s_namlen = sprintf(sp->s_name, + ".nfs.%08x.%04x4.4", lticks, + pid); + if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, + cnp->cn_thread, NULL)) + break; + lticks++; + } + error = nfs_renameit(dvp, cnp, sp); + if (error) + goto bad; + error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, + cnp->cn_thread, &np); + np->n_sillyrename = sp; + return (0); +bad: + vrele(sp->s_dvp); + crfree(sp->s_cred); + free((caddr_t)sp, M_NFSREQ); + return (error); +} + +/* + * Look up a file name and optionally either update the file handle or + * allocate an nfsnode, depending on the value of npp. + * npp == NULL --> just do the lookup + * *npp == NULL --> allocate a new nfsnode and make sure attributes are + * handled too + * *npp != NULL --> update the file handle in the vnode + */ +static int +nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred, + struct thread *td, struct nfsnode **npp) +{ + struct vnode *newvp = NULL; + struct nfsnode *np, *dnp = VTONFS(dvp); + caddr_t bpos, dpos; + int error = 0, fhlen, attrflag; + struct mbuf *mreq, *mrep, *md, *mb; + nfsfh_t *nfhp; + int v3 = NFS_ISV3(dvp); + + nfsstats.rpccnt[NFSPROC_LOOKUP]++; + mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP, + NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(dvp, v3); + nfsm_strtom(name, len, NFS_MAXNAMLEN); + nfsm_request(dvp, NFSPROC_LOOKUP, td, cred); + if (npp && !error) { + nfsm_getfh(nfhp, fhlen, v3); + if (*npp) { + np = *npp; + if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) { + free((caddr_t)np->n_fhp, M_NFSBIGFH); + np->n_fhp = &np->n_fh; + } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH) + np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK); + bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen); + np->n_fhsize = fhlen; + newvp = NFSTOV(np); + } else if (NFS_CMPFH(dnp, nfhp, fhlen)) { + VREF(dvp); + newvp = dvp; + } else { + error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE); + if (error) { + m_freem(mrep); + return (error); + } + newvp = NFSTOV(np); + } + if (v3) { + nfsm_postop_attr(newvp, attrflag); + if (!attrflag && *npp == NULL) { + m_freem(mrep); + if (newvp == dvp) + vrele(newvp); + else + vput(newvp); + return (ENOENT); + } + } else + nfsm_loadattr(newvp, NULL); + } + m_freem(mrep); +nfsmout: + if (npp && *npp == NULL) { + if (error) { + if (newvp) { + if (newvp == dvp) + vrele(newvp); + else + vput(newvp); + } + } else + *npp = np; + } + return (error); +} + +/* + * Nfs Version 3 commit rpc + */ +int +nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, + struct thread *td) +{ + u_int32_t *tl; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + caddr_t bpos, dpos; + int error = 0, wccflag = NFSV3_WCCRATTR; + struct mbuf *mreq, *mrep, *md, *mb; + + mtx_lock(&nmp->nm_mtx); + if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { + mtx_unlock(&nmp->nm_mtx); + return (0); + } + mtx_unlock(&nmp->nm_mtx); + nfsstats.rpccnt[NFSPROC_COMMIT]++; + mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1)); + mb = mreq; + bpos = mtod(mb, caddr_t); + nfsm_fhtom(vp, 1); + tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED); + txdr_hyper(offset, tl); + tl += 2; + *tl = txdr_unsigned(cnt); + nfsm_request(vp, NFSPROC_COMMIT, td, cred); + nfsm_wcc_data(vp, wccflag); + if (!error) { + tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF); + if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl, + NFSX_V3WRITEVERF)) { + bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf, + NFSX_V3WRITEVERF); + error = NFSERR_STALEWRITEVERF; + } + } + m_freem(mrep); +nfsmout: + return (error); +} + +/* + * Strategy routine. + * For async requests when nfsiod(s) are running, queue the request by + * calling nfs_asyncio(), otherwise just all nfs_doio() to do the + * request. + */ +static int +nfs_strategy(struct vop_strategy_args *ap) +{ + struct buf *bp = ap->a_bp; + struct ucred *cr; + + KASSERT(!(bp->b_flags & B_DONE), + ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); + BUF_ASSERT_HELD(bp); + + if (bp->b_iocmd == BIO_READ) + cr = bp->b_rcred; + else + cr = bp->b_wcred; + + /* + * If the op is asynchronous and an i/o daemon is waiting + * queue the request, wake it up and wait for completion + * otherwise just do it ourselves. + */ + if ((bp->b_flags & B_ASYNC) == 0 || + nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread)) + (void)nfs_doio(ap->a_vp, bp, cr, curthread); + return (0); +} + +/* + * fsync vnode op. Just call nfs_flush() with commit == 1. + */ +/* ARGSUSED */ +static int +nfs_fsync(struct vop_fsync_args *ap) +{ + + return (nfs_flush(ap->a_vp, ap->a_waitfor, 1)); +} + +/* + * Flush all the blocks associated with a vnode. + * Walk through the buffer pool and push any dirty pages + * associated with the vnode. + */ +static int +nfs_flush(struct vnode *vp, int waitfor, int commit) +{ + struct nfsnode *np = VTONFS(vp); + struct buf *bp; + int i; + struct buf *nbp; + struct nfsmount *nmp = VFSTONFS(vp->v_mount); + int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; + int passone = 1; + u_quad_t off, endoff, toff; + struct ucred* wcred = NULL; + struct buf **bvec = NULL; + struct bufobj *bo; + struct thread *td = curthread; +#ifndef NFS_COMMITBVECSIZ +#define NFS_COMMITBVECSIZ 20 +#endif + struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; + int bvecsize = 0, bveccount; + + if (nmp->nm_flag & NFSMNT_INT) + slpflag = NFS_PCATCH; + if (!commit) + passone = 0; + bo = &vp->v_bufobj; + /* + * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the + * server, but has not been committed to stable storage on the server + * yet. On the first pass, the byte range is worked out and the commit + * rpc is done. On the second pass, nfs_writebp() is called to do the + * job. + */ +again: + off = (u_quad_t)-1; + endoff = 0; + bvecpos = 0; + if (NFS_ISV3(vp) && commit) { + if (bvec != NULL && bvec != bvec_on_stack) + free(bvec, M_TEMP); + /* + * Count up how many buffers waiting for a commit. + */ + bveccount = 0; + BO_LOCK(bo); + TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { + if (!BUF_ISLOCKED(bp) && + (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) + == (B_DELWRI | B_NEEDCOMMIT)) + bveccount++; + } + /* + * Allocate space to remember the list of bufs to commit. It is + * important to use M_NOWAIT here to avoid a race with nfs_write. + * If we can't get memory (for whatever reason), we will end up + * committing the buffers one-by-one in the loop below. + */ + if (bveccount > NFS_COMMITBVECSIZ) { + /* + * Release the vnode interlock to avoid a lock + * order reversal. + */ + BO_UNLOCK(bo); + bvec = (struct buf **) + malloc(bveccount * sizeof(struct buf *), + M_TEMP, M_NOWAIT); + BO_LOCK(bo); + if (bvec == NULL) { + bvec = bvec_on_stack; + bvecsize = NFS_COMMITBVECSIZ; + } else + bvecsize = bveccount; + } else { + bvec = bvec_on_stack; + bvecsize = NFS_COMMITBVECSIZ; + } + TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { + if (bvecpos >= bvecsize) + break; + if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { + nbp = TAILQ_NEXT(bp, b_bobufs); + continue; + } + if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != + (B_DELWRI | B_NEEDCOMMIT)) { + BUF_UNLOCK(bp); + nbp = TAILQ_NEXT(bp, b_bobufs); + continue; + } + BO_UNLOCK(bo); + bremfree(bp); + /* + * Work out if all buffers are using the same cred + * so we can deal with them all with one commit. + * + * NOTE: we are not clearing B_DONE here, so we have + * to do it later on in this routine if we intend to + * initiate I/O on the bp. + * + * Note: to avoid loopback deadlocks, we do not + * assign b_runningbufspace. + */ + if (wcred == NULL) + wcred = bp->b_wcred; + else if (wcred != bp->b_wcred) + wcred = NOCRED; + vfs_busy_pages(bp, 1); + + BO_LOCK(bo); + /* + * bp is protected by being locked, but nbp is not + * and vfs_busy_pages() may sleep. We have to + * recalculate nbp. + */ + nbp = TAILQ_NEXT(bp, b_bobufs); + + /* + * A list of these buffers is kept so that the + * second loop knows which buffers have actually + * been committed. This is necessary, since there + * may be a race between the commit rpc and new + * uncommitted writes on the file. + */ + bvec[bvecpos++] = bp; + toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + + bp->b_dirtyoff; + if (toff < off) + off = toff; + toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); + if (toff > endoff) + endoff = toff; + } + BO_UNLOCK(bo); + } + if (bvecpos > 0) { + /* + * Commit data on the server, as required. + * If all bufs are using the same wcred, then use that with + * one call for all of them, otherwise commit each one + * separately. + */ + if (wcred != NOCRED) + retv = nfs_commit(vp, off, (int)(endoff - off), + wcred, td); + else { + retv = 0; + for (i = 0; i < bvecpos; i++) { + off_t off, size; + bp = bvec[i]; + off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + + bp->b_dirtyoff; + size = (u_quad_t)(bp->b_dirtyend + - bp->b_dirtyoff); + retv = nfs_commit(vp, off, (int)size, + bp->b_wcred, td); + if (retv) break; + } + } + + if (retv == NFSERR_STALEWRITEVERF) + nfs_clearcommit(vp->v_mount); + + /* + * Now, either mark the blocks I/O done or mark the + * blocks dirty, depending on whether the commit + * succeeded. + */ + for (i = 0; i < bvecpos; i++) { + bp = bvec[i]; + bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); + if (retv) { + /* + * Error, leave B_DELWRI intact + */ + vfs_unbusy_pages(bp); + brelse(bp); + } else { + /* + * Success, remove B_DELWRI ( bundirty() ). + * + * b_dirtyoff/b_dirtyend seem to be NFS + * specific. We should probably move that + * into bundirty(). XXX + */ + bufobj_wref(bo); + bp->b_flags |= B_ASYNC; + bundirty(bp); + bp->b_flags &= ~B_DONE; + bp->b_ioflags &= ~BIO_ERROR; + bp->b_dirtyoff = bp->b_dirtyend = 0; + bufdone(bp); + } + } + } + + /* + * Start/do any write(s) that are required. + */ +loop: + BO_LOCK(bo); + TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { + if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { + if (waitfor != MNT_WAIT || passone) + continue; + + error = BUF_TIMELOCK(bp, + LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, + BO_MTX(bo), "nfsfsync", slpflag, slptimeo); + if (error == 0) { + BUF_UNLOCK(bp); + goto loop; + } + if (error == ENOLCK) { + error = 0; + goto loop; + } + if (nfs_sigintr(nmp, td)) { + error = EINTR; + goto done; + } + if (slpflag & PCATCH) { + slpflag = 0; + slptimeo = 2 * hz; + } + goto loop; + } + if ((bp->b_flags & B_DELWRI) == 0) + panic("nfs_fsync: not dirty"); + if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { + BUF_UNLOCK(bp); + continue; + } + BO_UNLOCK(bo); + bremfree(bp); + if (passone || !commit) + bp->b_flags |= B_ASYNC; + else + bp->b_flags |= B_ASYNC; + bwrite(bp); + if (nfs_sigintr(nmp, td)) { + error = EINTR; + goto done; + } + goto loop; + } + if (passone) { + passone = 0; + BO_UNLOCK(bo); + goto again; + } + if (waitfor == MNT_WAIT) { + while (bo->bo_numoutput) { + error = bufobj_wwait(bo, slpflag, slptimeo); + if (error) { + BO_UNLOCK(bo); + error = nfs_sigintr(nmp, td); + if (error) + goto done; + if (slpflag & PCATCH) { + slpflag = 0; + slptimeo = 2 * hz; + } + BO_LOCK(bo); + } + } + if (bo->bo_dirty.bv_cnt != 0 && commit) { + BO_UNLOCK(bo); + goto loop; + } + /* + * Wait for all the async IO requests to drain + */ + BO_UNLOCK(bo); + mtx_lock(&np->n_mtx); + while (np->n_directio_asyncwr > 0) { + np->n_flag |= NFSYNCWAIT; + error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr, + &np->n_mtx, slpflag | (PRIBIO + 1), + "nfsfsync", 0); + if (error) { + if (nfs_sigintr(nmp, td)) { + mtx_unlock(&np->n_mtx); + error = EINTR; + goto done; + } + } + } + mtx_unlock(&np->n_mtx); + } else + BO_UNLOCK(bo); + mtx_lock(&np->n_mtx); + if (np->n_flag & NWRITEERR) { + error = np->n_error; + np->n_flag &= ~NWRITEERR; + } + if (commit && bo->bo_dirty.bv_cnt == 0 && + bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) + np->n_flag &= ~NMODIFIED; + mtx_unlock(&np->n_mtx); +done: + if (bvec != NULL && bvec != bvec_on_stack) + free(bvec, M_TEMP); + return (error); +} + +/* + * NFS advisory byte-level locks. + */ +static int +nfs_advlock(struct vop_advlock_args *ap) +{ + struct vnode *vp = ap->a_vp; + u_quad_t size; + int error; + + error = vn_lock(vp, LK_SHARED); + if (error) + return (error); + if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { + size = VTONFS(vp)->n_size; + VOP_UNLOCK(vp, 0); + error = lf_advlock(ap, &(vp->v_lockf), size); + } else { + if (nfs_advlock_p) + error = nfs_advlock_p(ap); + else + error = ENOLCK; + } + + return (error); +} + +/* + * NFS advisory byte-level locks. + */ +static int +nfs_advlockasync(struct vop_advlockasync_args *ap) +{ + struct vnode *vp = ap->a_vp; + u_quad_t size; + int error; + + error = vn_lock(vp, LK_SHARED); + if (error) + return (error); + if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { + size = VTONFS(vp)->n_size; + VOP_UNLOCK(vp, 0); + error = lf_advlockasync(ap, &(vp->v_lockf), size); + } else { + VOP_UNLOCK(vp, 0); + error = EOPNOTSUPP; + } + return (error); +} + +/* + * Print out the contents of an nfsnode. + */ +static int +nfs_print(struct vop_print_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + + nfs_printf("\tfileid %ld fsid 0x%x", + np->n_vattr.va_fileid, np->n_vattr.va_fsid); + if (vp->v_type == VFIFO) + fifo_printinfo(vp); + printf("\n"); + return (0); +} + +/* + * This is the "real" nfs::bwrite(struct buf*). + * We set B_CACHE if this is a VMIO buffer. + */ +int +nfs_writebp(struct buf *bp, int force __unused, struct thread *td) +{ + int s; + int oldflags = bp->b_flags; +#if 0 + int retv = 1; + off_t off; +#endif + + BUF_ASSERT_HELD(bp); + + if (bp->b_flags & B_INVAL) { + brelse(bp); + return(0); + } + + bp->b_flags |= B_CACHE; + + /* + * Undirty the bp. We will redirty it later if the I/O fails. + */ + + s = splbio(); + bundirty(bp); + bp->b_flags &= ~B_DONE; + bp->b_ioflags &= ~BIO_ERROR; + bp->b_iocmd = BIO_WRITE; + + bufobj_wref(bp->b_bufobj); + curthread->td_ru.ru_oublock++; + splx(s); + + /* + * Note: to avoid loopback deadlocks, we do not + * assign b_runningbufspace. + */ + vfs_busy_pages(bp, 1); + + BUF_KERNPROC(bp); + bp->b_iooffset = dbtob(bp->b_blkno); + bstrategy(bp); + + if( (oldflags & B_ASYNC) == 0) { + int rtval = bufwait(bp); + + if (oldflags & B_DELWRI) { + s = splbio(); + reassignbuf(bp); + splx(s); + } + brelse(bp); + return (rtval); + } + + return (0); +} + +/* + * nfs special file access vnode op. + * Essentially just get vattr and then imitate iaccess() since the device is + * local to the client. + */ +static int +nfsspec_access(struct vop_access_args *ap) +{ + struct vattr *vap; + struct ucred *cred = ap->a_cred; + struct vnode *vp = ap->a_vp; + accmode_t accmode = ap->a_accmode; + struct vattr vattr; + int error; + + /* + * Disallow write attempts on filesystems mounted read-only; + * unless the file is a socket, fifo, or a block or character + * device resident on the filesystem. + */ + if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { + switch (vp->v_type) { + case VREG: + case VDIR: + case VLNK: + return (EROFS); + default: + break; + } + } + vap = &vattr; + error = VOP_GETATTR(vp, vap, cred); + if (error) + goto out; + error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, + accmode, cred, NULL); +out: + return error; +} + +/* + * Read wrapper for fifos. + */ +static int +nfsfifo_read(struct vop_read_args *ap) +{ + struct nfsnode *np = VTONFS(ap->a_vp); + int error; + + /* + * Set access flag. + */ + mtx_lock(&np->n_mtx); + np->n_flag |= NACC; + getnanotime(&np->n_atim); + mtx_unlock(&np->n_mtx); + error = fifo_specops.vop_read(ap); + return error; +} + +/* + * Write wrapper for fifos. + */ +static int +nfsfifo_write(struct vop_write_args *ap) +{ + struct nfsnode *np = VTONFS(ap->a_vp); + + /* + * Set update flag. + */ + mtx_lock(&np->n_mtx); + np->n_flag |= NUPD; + getnanotime(&np->n_mtim); + mtx_unlock(&np->n_mtx); + return(fifo_specops.vop_write(ap)); +} + +/* + * Close wrapper for fifos. + * + * Update the times on the nfsnode then do fifo close. + */ +static int +nfsfifo_close(struct vop_close_args *ap) +{ + struct vnode *vp = ap->a_vp; + struct nfsnode *np = VTONFS(vp); + struct vattr vattr; + struct timespec ts; + + mtx_lock(&np->n_mtx); + if (np->n_flag & (NACC | NUPD)) { + getnanotime(&ts); + if (np->n_flag & NACC) + np->n_atim = ts; + if (np->n_flag & NUPD) + np->n_mtim = ts; + np->n_flag |= NCHG; + if (vrefcnt(vp) == 1 && + (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { + VATTR_NULL(&vattr); + if (np->n_flag & NACC) + vattr.va_atime = np->n_atim; + if (np->n_flag & NUPD) + vattr.va_mtime = np->n_mtim; + mtx_unlock(&np->n_mtx); + (void)VOP_SETATTR(vp, &vattr, ap->a_cred); + goto out; + } + } + mtx_unlock(&np->n_mtx); +out: + return (fifo_specops.vop_close(ap)); +} + +/* + * Just call nfs_writebp() with the force argument set to 1. + * + * NOTE: B_DONE may or may not be set in a_bp on call. + */ +static int +nfs_bwrite(struct buf *bp) +{ + + return (nfs_writebp(bp, 1, curthread)); +} + +struct buf_ops buf_ops_nfs = { + .bop_name = "buf_ops_nfs", + .bop_write = nfs_bwrite, + .bop_strategy = bufstrategy, + .bop_sync = bufsync, + .bop_bdflush = bufbdflush, +}; diff --git a/sys/nfsclient/nfsargs.h b/sys/nfsclient/nfsargs.h new file mode 100644 index 0000000..f71a555 --- /dev/null +++ b/sys/nfsclient/nfsargs.h @@ -0,0 +1,102 @@ +/*- + * Copyright (c) 1989, 1993, 1995 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs.h 8.4 (Berkeley) 5/1/95 + * $FreeBSD$ + */ + +#ifndef _NFSCLIENT_NFSARGS_H_ +#define _NFSCLIENT_NFSARGS_H_ + +/* + * Arguments to mount NFS + */ +#define NFS_ARGSVERSION 3 /* change when nfs_args changes */ +struct nfs_args { + int version; /* args structure version number */ + struct sockaddr *addr; /* file server address */ + int addrlen; /* length of address */ + int sotype; /* Socket type */ + int proto; /* and Protocol */ + u_char *fh; /* File handle to be mounted */ + int fhsize; /* Size, in bytes, of fh */ + int flags; /* flags */ + int wsize; /* write size in bytes */ + int rsize; /* read size in bytes */ + int readdirsize; /* readdir size in bytes */ + int timeo; /* initial timeout in .1 secs */ + int retrans; /* times to retry send */ + int maxgrouplist; /* Max. size of group list */ + int readahead; /* # of blocks to readahead */ + int wcommitsize; /* Max. write commit size in bytes */ + int deadthresh; /* Retrans threshold */ + char *hostname; /* server's name */ + int acregmin; /* cache attrs for reg files min time */ + int acregmax; /* cache attrs for reg files max time */ + int acdirmin; /* cache attrs for dirs min time */ + int acdirmax; /* cache attrs for dirs max time */ +}; + +/* + * NFS mount option flags + */ +#define NFSMNT_SOFT 0x00000001 /* soft mount (hard is default) */ +#define NFSMNT_WSIZE 0x00000002 /* set write size */ +#define NFSMNT_RSIZE 0x00000004 /* set read size */ +#define NFSMNT_TIMEO 0x00000008 /* set initial timeout */ +#define NFSMNT_RETRANS 0x00000010 /* set number of request retries */ +#define NFSMNT_MAXGRPS 0x00000020 /* set maximum grouplist size */ +#define NFSMNT_INT 0x00000040 /* allow interrupts on hard mount */ +#define NFSMNT_NOCONN 0x00000080 /* Don't Connect the socket */ +/* 0x100 free, was NFSMNT_NQNFS */ +#define NFSMNT_NFSV3 0x00000200 /* Use NFS Version 3 protocol */ +#define NFSMNT_KERB 0x00000400 /* Use RPCSEC_GSS/Krb5 */ +#define NFSMNT_DUMBTIMR 0x00000800 /* Don't estimate rtt dynamically */ +#define NFSMNT_WCOMMITSIZE 0x00001000 /* set max write commit size */ +#define NFSMNT_READAHEAD 0x00002000 /* set read ahead */ +#define NFSMNT_DEADTHRESH 0x00004000 /* set dead server retry thresh */ +#define NFSMNT_RESVPORT 0x00008000 /* Allocate a reserved port */ +#define NFSMNT_RDIRPLUS 0x00010000 /* Use Readdirplus for V3 */ +#define NFSMNT_READDIRSIZE 0x00020000 /* Set readdir size */ +#define NFSMNT_ACREGMIN 0x00040000 +#define NFSMNT_ACREGMAX 0x00080000 +#define NFSMNT_ACDIRMIN 0x00100000 +#define NFSMNT_ACDIRMAX 0x00200000 +#define NFSMNT_NOLOCKD 0x00400000 /* Locks are local */ +#define NFSMNT_NFSV4 0x00800000 /* Use NFS Version 4 protocol */ +#define NFSMNT_HASWRITEVERF 0x01000000 /* NFSv4 Write verifier */ +#define NFSMNT_INTEGRITY 0x02000000 /* Use integrity with RPCSEC_GSS */ +#define NFSMNT_PRIVACY 0x04000000 /* Use privacy with RPCSEC_GSS */ +#define NFSMNT_ALLGSSNAME 0x08000000 /* Use principal for all accesses */ +#define NFSMNT_STRICT3530 0x10000000 /* Adhere strictly to RFC3530 */ +#define NFSMNT_NOCTO 0x20000000 /* Don't flush attrcache on open */ + +#endif diff --git a/sys/nfsclient/nfsm_subs.h b/sys/nfsclient/nfsm_subs.h new file mode 100644 index 0000000..c12badd --- /dev/null +++ b/sys/nfsclient/nfsm_subs.h @@ -0,0 +1,206 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfsm_subs.h 8.2 (Berkeley) 3/30/95 + * $FreeBSD$ + */ + +#ifndef _NFSCLIENT_NFSM_SUBS_H_ +#define _NFSCLIENT_NFSM_SUBS_H_ + +#include <nfs/nfs_common.h> + +#define nfsv2tov_type(a) nv2tov_type[fxdr_unsigned(u_int32_t,(a))&0x7] + +struct ucred; +struct vnode; + +/* + * These macros do strange and peculiar things to mbuf chains for + * the assistance of the nfs code. To attempt to use them for any + * other purpose will be dangerous. (they make weird assumptions) + */ + +/* + * First define what the actual subs. return + */ +u_int32_t nfs_xid_gen(void); +struct mbuf *nfsm_reqhead(struct vnode *vp, u_long procid, int hsiz); + +#define M_HASCL(m) ((m)->m_flags & M_EXT) +#define NFSMINOFF(m) \ + do { \ + if (M_HASCL(m)) \ + (m)->m_data = (m)->m_ext.ext_buf; \ + else if ((m)->m_flags & M_PKTHDR) \ + (m)->m_data = (m)->m_pktdat; \ + else \ + (m)->m_data = (m)->m_dat; \ + } while (0) +#define NFSMSIZ(m) ((M_HASCL(m))?MCLBYTES: \ + (((m)->m_flags & M_PKTHDR)?MHLEN:MLEN)) + +/* + * Now for the macros that do the simple stuff and call the functions + * for the hard stuff. + * These macros use several vars. declared in nfsm_reqhead and these + * vars. must not be used elsewhere unless you are careful not to corrupt + * them. The vars. starting with pN and tN (N=1,2,3,..) are temporaries + * that may be used so long as the value is not expected to retained + * after a macro. + * I know, this is kind of dorkey, but it makes the actual op functions + * fairly clean and deals with the mess caused by the xdr discriminating + * unions. + */ + + +/* *********************************** */ +/* Request generation phase macros */ + +int nfsm_fhtom_xx(struct vnode *v, int v3, struct mbuf **mb, + caddr_t *bpos); +void nfsm_v3attrbuild_xx(struct vattr *va, int full, struct mbuf **mb, + caddr_t *bpos); +int nfsm_strtom_xx(const char *a, int s, int m, struct mbuf **mb, + caddr_t *bpos); + +#define nfsm_bcheck(t1, mreq) \ +do { \ + if (t1) { \ + error = t1; \ + m_freem(mreq); \ + goto nfsmout; \ + } \ +} while (0) + +#define nfsm_fhtom(v, v3) \ +do { \ + int32_t t1; \ + t1 = nfsm_fhtom_xx((v), (v3), &mb, &bpos); \ + nfsm_bcheck(t1, mreq); \ +} while (0) + +/* If full is true, set all fields, otherwise just set mode and time fields */ +#define nfsm_v3attrbuild(a, full) \ + nfsm_v3attrbuild_xx(a, full, &mb, &bpos) + +#define nfsm_uiotom(p, s) \ +do { \ + int t1; \ + t1 = nfsm_uiotombuf((p), &mb, (s), &bpos); \ + nfsm_bcheck(t1, mreq); \ +} while (0) + +#define nfsm_strtom(a, s, m) \ +do { \ + int t1; \ + t1 = nfsm_strtom_xx((a), (s), (m), &mb, &bpos); \ + nfsm_bcheck(t1, mreq); \ +} while (0) + +/* *********************************** */ +/* Send the request */ + +#define nfsm_request(v, t, p, c) \ +do { \ + sigset_t oldset; \ + nfs_set_sigmask(p, &oldset); \ + error = nfs_request((v), mreq, (t), (p), (c), &mrep, &md, &dpos); \ + nfs_restore_sigmask(p, &oldset); \ + if (error != 0) { \ + if (error & NFSERR_RETERR) \ + error &= ~NFSERR_RETERR; \ + else \ + goto nfsmout; \ + } \ +} while (0) + +/* *********************************** */ +/* Reply interpretation phase macros */ + +int nfsm_mtofh_xx(struct vnode *d, struct vnode **v, int v3, int *f, + struct mbuf **md, caddr_t *dpos); +int nfsm_getfh_xx(nfsfh_t **f, int *s, int v3, struct mbuf **md, + caddr_t *dpos); +int nfsm_loadattr_xx(struct vnode **v, struct vattr *va, struct mbuf **md, + caddr_t *dpos); +int nfsm_postop_attr_xx(struct vnode **v, int *f, struct vattr *va, + struct mbuf **md, caddr_t *dpos); +int nfsm_wcc_data_xx(struct vnode **v, int *f, struct mbuf **md, + caddr_t *dpos); + +#define nfsm_mtofh(d, v, v3, f) \ +do { \ + int32_t t1; \ + t1 = nfsm_mtofh_xx((d), &(v), (v3), &(f), &md, &dpos); \ + nfsm_dcheck(t1, mrep); \ +} while (0) + +#define nfsm_getfh(f, s, v3) \ +do { \ + int32_t t1; \ + t1 = nfsm_getfh_xx(&(f), &(s), (v3), &md, &dpos); \ + nfsm_dcheck(t1, mrep); \ +} while (0) + +#define nfsm_loadattr(v, a) \ +do { \ + int32_t t1; \ + t1 = nfsm_loadattr_xx(&v, a, &md, &dpos); \ + nfsm_dcheck(t1, mrep); \ +} while (0) + +#define nfsm_postop_attr(v, f) \ +do { \ + int32_t t1; \ + t1 = nfsm_postop_attr_xx(&v, &f, NULL, &md, &dpos); \ + nfsm_dcheck(t1, mrep); \ +} while (0) + +#define nfsm_postop_attr_va(v, f, va) \ +do { \ + int32_t t1; \ + t1 = nfsm_postop_attr_xx(&v, &f, va, &md, &dpos); \ + nfsm_dcheck(t1, mrep); \ +} while (0) + +/* Used as (f) for nfsm_wcc_data() */ +#define NFSV3_WCCRATTR 0 +#define NFSV3_WCCCHK 1 + +#define nfsm_wcc_data(v, f) \ +do { \ + int32_t t1; \ + t1 = nfsm_wcc_data_xx(&v, &f, &md, &dpos); \ + nfsm_dcheck(t1, mrep); \ +} while (0) + +#endif diff --git a/sys/nfsclient/nfsmount.h b/sys/nfsclient/nfsmount.h new file mode 100644 index 0000000..77171a1 --- /dev/null +++ b/sys/nfsclient/nfsmount.h @@ -0,0 +1,132 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfsmount.h 8.3 (Berkeley) 3/30/95 + * $FreeBSD$ + */ + +#ifndef _NFSCLIENT_NFSMOUNT_H_ +#define _NFSCLIENT_NFSMOUNT_H_ + +#include <sys/socket.h> + +#include <nfs/nfs_mountcommon.h> + +#include <rpc/types.h> +#include <rpc/auth.h> +#include <rpc/clnt.h> +#include <rpc/rpcsec_gss.h> + +/* + * Mount structure. + * One allocated on every NFS mount. + * Holds NFS specific information for mount. + */ +struct nfsmount { + struct nfsmount_common nm_com; /* Common fields for nlm */ + int nm_numgrps; /* Max. size of groupslist */ + u_char nm_fh[NFSX_V4FH]; /* File handle of root dir */ + int nm_fhsize; /* Size of root file handle */ + int nm_sotype; /* Type of socket */ + int nm_soproto; /* and protocol */ + int nm_soflags; /* pr_flags for socket protocol */ + struct sockaddr *nm_nam; /* Addr of server */ + int nm_deadthresh; /* Threshold of timeouts-->dead server*/ + int nm_rsize; /* Max size of read rpc */ + int nm_wsize; /* Max size of write rpc */ + int nm_readdirsize; /* Size of a readdir rpc */ + int nm_readahead; /* Num. of blocks to readahead */ + int nm_wcommitsize; /* Max size of commit for write */ + int nm_acdirmin; /* Directory attr cache min lifetime */ + int nm_acdirmax; /* Directory attr cache max lifetime */ + int nm_acregmin; /* Reg file attr cache min lifetime */ + int nm_acregmax; /* Reg file attr cache max lifetime */ + u_char nm_verf[NFSX_V3WRITEVERF]; /* V3 write verifier */ + TAILQ_HEAD(, buf) nm_bufq; /* async io buffer queue */ + short nm_bufqlen; /* number of buffers in queue */ + short nm_bufqwant; /* process wants to add to the queue */ + int nm_bufqiods; /* number of iods processing queue */ + u_int64_t nm_maxfilesize; /* maximum file size */ + struct nfs_rpcops *nm_rpcops; + int nm_tprintf_initial_delay; /* initial delay */ + int nm_tprintf_delay; /* interval for messages */ + int nm_secflavor; /* auth flavor to use for rpc */ + struct __rpc_client *nm_client; + struct rpc_timers nm_timers[NFS_MAX_TIMER]; /* RTT Timers for rpcs */ + char nm_principal[MNAMELEN]; /* GSS-API principal of server */ + gss_OID nm_mech_oid; /* OID of selected GSS-API mechanism */ + int nm_nametimeo; /* timeout for +ve entries (sec) */ + int nm_negnametimeo; /* timeout for -ve entries (sec) */ + + /* NFSv4 */ + uint64_t nm_clientid; + fsid_t nm_fsid; + u_int nm_lease_time; + time_t nm_last_renewal; +}; + +#define nm_mtx nm_com.nmcom_mtx +#define nm_flag nm_com.nmcom_flag +#define nm_state nm_com.nmcom_state +#define nm_mountp nm_com.nmcom_mountp +#define nm_timeo nm_com.nmcom_timeo +#define nm_retry nm_com.nmcom_retry +#define nm_hostname nm_com.nmcom_hostname +#define nm_getinfo nm_com.nmcom_getinfo +#define nm_vinvalbuf nm_com.nmcom_vinvalbuf + +#if defined(_KERNEL) +/* + * Convert mount ptr to nfsmount ptr. + */ +#define VFSTONFS(mp) ((struct nfsmount *)((mp)->mnt_data)) + +#ifndef NFS_TPRINTF_INITIAL_DELAY +#define NFS_TPRINTF_INITIAL_DELAY 12 +#endif + +#ifndef NFS_TPRINTF_DELAY +#define NFS_TPRINTF_DELAY 30 +#endif + +#ifndef NFS_DEFAULT_NAMETIMEO +#define NFS_DEFAULT_NAMETIMEO 60 +#endif + +#ifndef NFS_DEFAULT_NEGNAMETIMEO +#define NFS_DEFAULT_NEGNAMETIMEO 60 +#endif + +#define NFS_PCATCH (PCATCH | PBDRY) + +#endif + +#endif diff --git a/sys/nfsclient/nfsnode.h b/sys/nfsclient/nfsnode.h new file mode 100644 index 0000000..60e5613 --- /dev/null +++ b/sys/nfsclient/nfsnode.h @@ -0,0 +1,212 @@ +/*- + * Copyright (c) 1989, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfsnode.h 8.9 (Berkeley) 5/14/95 + * $FreeBSD$ + */ + +#ifndef _NFSCLIENT_NFSNODE_H_ +#define _NFSCLIENT_NFSNODE_H_ + +#include <sys/_task.h> +#if !defined(_NFSCLIENT_NFS_H_) && !defined(_KERNEL) +#include <nfs/nfs.h> +#endif + +/* + * Silly rename structure that hangs off the nfsnode until the name + * can be removed by nfs_inactive() + */ +struct sillyrename { + struct task s_task; + struct ucred *s_cred; + struct vnode *s_dvp; + int (*s_removeit)(struct sillyrename *sp); + long s_namlen; + char s_name[32]; +}; + +/* + * This structure is used to save the logical directory offset to + * NFS cookie mappings. + * The mappings are stored in a list headed + * by n_cookies, as required. + * There is one mapping for each NFS_DIRBLKSIZ bytes of directory information + * stored in increasing logical offset byte order. + */ +#define NFSNUMCOOKIES 31 + +struct nfsdmap { + LIST_ENTRY(nfsdmap) ndm_list; + int ndm_eocookie; + union { + nfsuint64 ndmu3_cookies[NFSNUMCOOKIES]; + uint64_t ndmu4_cookies[NFSNUMCOOKIES]; + } ndm_un1; +}; + +#define ndm_cookies ndm_un1.ndmu3_cookies +#define ndm4_cookies ndm_un1.ndmu4_cookies + +struct nfs_accesscache { + u_int32_t mode; /* ACCESS mode cache */ + uid_t uid; /* credentials having mode */ + time_t stamp; /* mode cache timestamp */ +}; + +/* + * The nfsnode is the nfs equivalent to ufs's inode. Any similarity + * is purely coincidental. + * There is a unique nfsnode allocated for each active file, + * each current directory, each mounted-on file, text file, and the root. + * An nfsnode is 'named' by its file handle. (nget/nfs_node.c) + * If this structure exceeds 256 bytes (it is currently 256 using 4.4BSD-Lite + * type definitions), file handles of > 32 bytes should probably be split out + * into a separate MALLOC()'d data structure. (Reduce the size of nfsfh_t by + * changing the definition in nfsproto.h of NFS_SMALLFH.) + * NB: Hopefully the current order of the fields is such that everything will + * be well aligned and, therefore, tightly packed. + */ +struct nfsnode { + struct mtx n_mtx; /* Protects all of these members */ + u_quad_t n_size; /* Current size of file */ + u_quad_t n_brev; /* Modify rev when cached */ + u_quad_t n_lrev; /* Modify rev for lease */ + struct vattr n_vattr; /* Vnode attribute cache */ + time_t n_attrstamp; /* Attr. cache timestamp */ + struct nfs_accesscache n_accesscache[NFS_ACCESSCACHESIZE]; + struct timespec n_mtime; /* Prev modify time. */ + nfsfh_t *n_fhp; /* NFS File Handle */ + struct vnode *n_vnode; /* associated vnode */ + struct vnode *n_dvp; /* parent vnode */ + int n_error; /* Save write error value */ + union { + struct timespec nf_atim; /* Special file times */ + nfsuint64 nd_cookieverf; /* Cookie verifier (dir only) */ + u_char nd4_cookieverf[NFSX_V4VERF]; + } n_un1; + union { + struct timespec nf_mtim; + off_t nd_direof; /* Dir. EOF offset cache */ + } n_un2; + union { + struct sillyrename *nf_silly; /* Ptr to silly rename struct */ + LIST_HEAD(, nfsdmap) nd_cook; /* cookies */ + } n_un3; + short n_fhsize; /* size in bytes, of fh */ + short n_flag; /* Flag for locking.. */ + nfsfh_t n_fh; /* Small File Handle */ + u_char *n_name; /* leaf name, for v4 OPEN op */ + uint32_t n_namelen; + int n_directio_opens; + int n_directio_asyncwr; +}; + +#define n_atim n_un1.nf_atim +#define n_mtim n_un2.nf_mtim +#define n_sillyrename n_un3.nf_silly +#define n_cookieverf n_un1.nd_cookieverf +#define n4_cookieverf n_un1.nd4_cookieverf +#define n_direofoffset n_un2.nd_direof +#define n_cookies n_un3.nd_cook + +/* + * Flags for n_flag + */ +#define NFSYNCWAIT 0x0002 /* fsync waiting for all directio async writes + to drain */ +#define NMODIFIED 0x0004 /* Might have a modified buffer in bio */ +#define NWRITEERR 0x0008 /* Flag write errors so close will know */ +/* 0x20, 0x40, 0x80 free */ +#define NACC 0x0100 /* Special file accessed */ +#define NUPD 0x0200 /* Special file updated */ +#define NCHG 0x0400 /* Special file times changed */ +#define NCREATED 0x0800 /* Opened by nfs_create() */ +#define NTRUNCATE 0x1000 /* Opened by nfs_setattr() */ +#define NSIZECHANGED 0x2000 /* File size has changed: need cache inval */ +#define NNONCACHE 0x4000 /* Node marked as noncacheable */ +#define NDIRCOOKIELK 0x8000 /* Lock to serialize access to directory cookies */ + +/* + * Convert between nfsnode pointers and vnode pointers + */ +#define VTONFS(vp) ((struct nfsnode *)(vp)->v_data) +#define NFSTOV(np) ((struct vnode *)(np)->n_vnode) + +#define NFS_TIMESPEC_COMPARE(T1, T2) (((T1)->tv_sec != (T2)->tv_sec) || ((T1)->tv_nsec != (T2)->tv_nsec)) + +/* + * NFS iod threads can be in one of these two states once spawned. + * NFSIOD_NOT_AVAILABLE - Cannot be assigned an I/O operation at this time. + * NFSIOD_AVAILABLE - Available to be assigned an I/O operation. + */ +enum nfsiod_state { + NFSIOD_NOT_AVAILABLE = 0, + NFSIOD_AVAILABLE = 1, +}; + +/* + * Queue head for nfsiod's + */ +extern TAILQ_HEAD(nfs_bufq, buf) nfs_bufq; +extern enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON]; +extern struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON]; + +#if defined(_KERNEL) + +extern struct vop_vector nfs_fifoops; +extern struct vop_vector nfs_vnodeops; +extern struct buf_ops buf_ops_nfs; + +/* + * Prototypes for NFS vnode operations + */ +int nfs_getpages(struct vop_getpages_args *); +int nfs_putpages(struct vop_putpages_args *); +int nfs_write(struct vop_write_args *); +int nfs_inactive(struct vop_inactive_args *); +int nfs_reclaim(struct vop_reclaim_args *); + +/* other stuff */ +int nfs_removeit(struct sillyrename *); +int nfs_nget(struct mount *, nfsfh_t *, int, struct nfsnode **, int flags); +nfsuint64 *nfs_getcookie(struct nfsnode *, off_t, int); +void nfs_invaldir(struct vnode *); +int nfs_upgrade_vnlock(struct vnode *vp); +void nfs_downgrade_vnlock(struct vnode *vp, int old_lock); +void nfs_printf(const char *fmt, ...); + +void nfs_dircookie_lock(struct nfsnode *np); +void nfs_dircookie_unlock(struct nfsnode *np); + +#endif /* _KERNEL */ + +#endif diff --git a/sys/nfsclient/nfsstats.h b/sys/nfsclient/nfsstats.h new file mode 100644 index 0000000..8028429 --- /dev/null +++ b/sys/nfsclient/nfsstats.h @@ -0,0 +1,69 @@ +/*- + * Copyright (c) 1989, 1993, 1995 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Rick Macklem at The University of Guelph. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)nfs.h 8.4 (Berkeley) 5/1/95 + * $FreeBSD$ + */ + +#ifndef _NFSCLIENT_NFSSTATS_H_ +#define _NFSCLIENT_NFSSTATS_H_ + +/* + * Stats structure + */ +struct nfsstats { + int attrcache_hits; + int attrcache_misses; + int lookupcache_hits; + int lookupcache_misses; + int direofcache_hits; + int direofcache_misses; + int accesscache_hits; + int accesscache_misses; + int biocache_reads; + int read_bios; + int read_physios; + int biocache_writes; + int write_bios; + int write_physios; + int biocache_readlinks; + int readlink_bios; + int biocache_readdirs; + int readdir_bios; + int rpcretries; + int rpcrequests; + int rpctimeouts; + int rpcunexpected; + int rpcinvalid; + int rpccnt[NFS_NPROCS]; +}; + +#endif diff --git a/sys/nfsclient/nlminfo.h b/sys/nfsclient/nlminfo.h new file mode 100644 index 0000000..7d3e0ff --- /dev/null +++ b/sys/nfsclient/nlminfo.h @@ -0,0 +1,42 @@ +/*- + * Copyright (c) 1998 Berkeley Software Design, Inc. All rights reserved. + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Berkeley Software Design Inc's name may not be used to endorse or + * promote products derived from this software without specific prior + * written permission. + * + * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * from BSDI nlminfo.h,v 2.1 1998/03/18 01:30:38 don Exp + * $FreeBSD$ + */ + +/* + * Misc NLM informationi, some needed for the master lockd process, and some + * needed by every process doing nlm based locking. + */ +struct nlminfo { + /* these are used by any process doing nlm locking */ + int msg_seq; /* sequence counter for lock requests */ + int retcode; /* return code for lock requests */ + int set_getlk_pid; + int getlk_pid; + struct timeval pid_start; /* process starting time */ +}; |
