diff options
author | dyson <dyson@FreeBSD.org> | 1998-03-07 21:37:31 +0000 |
---|---|---|
committer | dyson <dyson@FreeBSD.org> | 1998-03-07 21:37:31 +0000 |
commit | 8ceb6160f494e2331b3f2e241e09d92673e397af (patch) | |
tree | 5030aec9050b0e765c5aea8634ba2de5ad3306e2 | |
parent | ab602aeb2614330963a1cee7162c9b6b22b9b9d9 (diff) | |
download | FreeBSD-src-8ceb6160f494e2331b3f2e241e09d92673e397af.zip FreeBSD-src-8ceb6160f494e2331b3f2e241e09d92673e397af.tar.gz |
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
36 files changed, 766 insertions, 426 deletions
diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c index 7ad037f..f2c0fe4 100644 --- a/sys/amd64/amd64/pmap.c +++ b/sys/amd64/amd64/pmap.c @@ -39,7 +39,7 @@ * SUCH DAMAGE. * * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91 - * $Id: pmap.c,v 1.186 1998/02/12 22:00:01 bde Exp $ + * $Id: pmap.c,v 1.187 1998/03/01 04:18:54 dyson Exp $ */ /* @@ -2383,7 +2383,6 @@ retry: continue; } if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && - (p->busy == 0) && (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { if ((p->queue - p->pc) == PQ_CACHE) vm_page_deactivate(p); @@ -2404,7 +2403,6 @@ retry: p = vm_page_lookup(object, tmpidx + pindex); if (p && ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && - (p->busy == 0) && (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { if ((p->queue - p->pc) == PQ_CACHE) vm_page_deactivate(p); diff --git a/sys/amd64/include/smp.h b/sys/amd64/include/smp.h index d40a933..b10930a 100644 --- a/sys/amd64/include/smp.h +++ b/sys/amd64/include/smp.h @@ -6,7 +6,7 @@ * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp * ---------------------------------------------------------------------------- * - * $Id: smp.h,v 1.38 1998/03/03 22:56:26 tegge Exp $ + * $Id: smp.h,v 1.39 1998/03/05 21:45:44 tegge Exp $ * */ @@ -67,11 +67,11 @@ extern u_int mpintr_lock; /* functions in mplock.s */ void get_mplock __P((void)); void rel_mplock __P((void)); -int try_mplock __P((void)); +int try_mplock __P((void)); #ifdef RECURSIVE_MPINTRLOCK void get_mpintrlock __P((void)); void rel_mpintrlock __P((void)); -int try_mpintrlock __P((void)); +int try_mpintrlock __P((void)); #endif /* RECURSIVE_MPINTRLOCK */ /* global data in apic_vector.s */ diff --git a/sys/fs/specfs/spec_vnops.c b/sys/fs/specfs/spec_vnops.c index 50a70dc..c058665 100644 --- a/sys/fs/specfs/spec_vnops.c +++ b/sys/fs/specfs/spec_vnops.c @@ -31,7 +31,7 @@ * SUCH DAMAGE. * * @(#)spec_vnops.c 8.14 (Berkeley) 5/21/95 - * $Id: spec_vnops.c,v 1.56 1998/02/06 12:13:43 eivind Exp $ + * $Id: spec_vnops.c,v 1.57 1998/03/04 06:44:59 dyson Exp $ */ #include <sys/param.h> @@ -723,8 +723,10 @@ spec_getpages(ap) daddr_t blkno; struct buf *bp; vm_ooffset_t offset; + int toff, nextoff, nread; struct vnode *vp = ap->a_vp; int blksiz; + int gotreqpage; error = 0; pcount = round_page(ap->a_count) / PAGE_SIZE; @@ -788,8 +790,6 @@ spec_getpages(ap) /* Do the input. */ VOP_STRATEGY(bp); - if (bp->b_flags & B_ASYNC) - return (VM_PAGER_PEND); s = splbio(); @@ -799,12 +799,19 @@ spec_getpages(ap) splx(s); - if ((bp->b_flags & B_ERROR) != 0) - error = EIO; + if ((bp->b_flags & B_ERROR) != 0) { + if (bp->b_error) + error = bp->b_error; + else + error = EIO; + } + + nread = size - bp->b_resid; - if (!error && ap->a_count != pcount * PAGE_SIZE) - bzero((caddr_t)kva + ap->a_count, - PAGE_SIZE * pcount - ap->a_count); + if (nread < ap->a_count) { + bzero((caddr_t)kva + nread, + ap->a_count - nread); + } pmap_qremove(kva, pcount); /* @@ -812,36 +819,53 @@ spec_getpages(ap) */ relpbuf(bp); - for (i = 0; i < pcount; i++) { - ap->a_m[i]->dirty = 0; - ap->a_m[i]->valid = VM_PAGE_BITS_ALL; - ap->a_m[i]->flags &= ~PG_ZERO; - 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. - */ + gotreqpage = 0; + for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) { + vm_page_t m; + nextoff = toff + PAGE_SIZE; + m = ap->a_m[i]; + + m->flags &= ~PG_ZERO; + if (nextoff <= nread) { + m->valid = VM_PAGE_BITS_ALL; + m->dirty = 0; + } else if (toff < nread) { + int nvalid = ((nread + DEV_BSIZE - 1) - toff) & ~(DEV_BSIZE - 1); + vm_page_set_validclean(m, 0, nvalid); + } else { + m->valid = 0; + m->dirty = 0; + } + + if (i != ap->a_reqpage) { /* * Just in case someone was asking for this page we * now tell them that it is ok to use. */ - if (!error) { - if (ap->a_m[i]->flags & PG_WANTED) - vm_page_activate(ap->a_m[i]); - else - vm_page_deactivate(ap->a_m[i]); - PAGE_WAKEUP(ap->a_m[i]); - } else - vnode_pager_freepage(ap->a_m[i]); + if (!error || (m->valid == VM_PAGE_BITS_ALL)) { + if (m->valid) { + if (m->flags & PG_WANTED) { + vm_page_activate(m); + } else { + vm_page_deactivate(m); + } + PAGE_WAKEUP(m); + } else { + vm_page_free(m); + } + } else { + vm_page_free(m); + } + } else if (m->valid) { + gotreqpage = 1; } } - if (error) - printf("spec_getpages: I/O read error\n"); - return (error ? VM_PAGER_ERROR : VM_PAGER_OK); + if (!gotreqpage) { + printf("spec_getpages: I/O read failure: (code=%d)\n", error); + return VM_PAGER_ERROR; + } + return VM_PAGER_OK; } /* ARGSUSED */ diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c index 7ad037f..f2c0fe4 100644 --- a/sys/i386/i386/pmap.c +++ b/sys/i386/i386/pmap.c @@ -39,7 +39,7 @@ * SUCH DAMAGE. * * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91 - * $Id: pmap.c,v 1.186 1998/02/12 22:00:01 bde Exp $ + * $Id: pmap.c,v 1.187 1998/03/01 04:18:54 dyson Exp $ */ /* @@ -2383,7 +2383,6 @@ retry: continue; } if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && - (p->busy == 0) && (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { if ((p->queue - p->pc) == PQ_CACHE) vm_page_deactivate(p); @@ -2404,7 +2403,6 @@ retry: p = vm_page_lookup(object, tmpidx + pindex); if (p && ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && - (p->busy == 0) && (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { if ((p->queue - p->pc) == PQ_CACHE) vm_page_deactivate(p); diff --git a/sys/i386/include/smp.h b/sys/i386/include/smp.h index d40a933..b10930a 100644 --- a/sys/i386/include/smp.h +++ b/sys/i386/include/smp.h @@ -6,7 +6,7 @@ * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp * ---------------------------------------------------------------------------- * - * $Id: smp.h,v 1.38 1998/03/03 22:56:26 tegge Exp $ + * $Id: smp.h,v 1.39 1998/03/05 21:45:44 tegge Exp $ * */ @@ -67,11 +67,11 @@ extern u_int mpintr_lock; /* functions in mplock.s */ void get_mplock __P((void)); void rel_mplock __P((void)); -int try_mplock __P((void)); +int try_mplock __P((void)); #ifdef RECURSIVE_MPINTRLOCK void get_mpintrlock __P((void)); void rel_mpintrlock __P((void)); -int try_mpintrlock __P((void)); +int try_mpintrlock __P((void)); #endif /* RECURSIVE_MPINTRLOCK */ /* global data in apic_vector.s */ diff --git a/sys/kern/kern_exec.c b/sys/kern/kern_exec.c index fcfe4e2..64fea59 100644 --- a/sys/kern/kern_exec.c +++ b/sys/kern/kern_exec.c @@ -23,7 +23,7 @@ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * - * $Id: kern_exec.c,v 1.78 1998/02/25 13:08:06 bde Exp $ + * $Id: kern_exec.c,v 1.79 1998/03/02 05:47:55 peter Exp $ */ #include <sys/param.h> @@ -389,8 +389,7 @@ exec_map_first_page(imgp) if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) { vm_page_protect(ma[0], VM_PROT_NONE); - vm_page_deactivate(ma[0]); - PAGE_WAKEUP(ma[0]); + vm_page_free(ma[0]); splx(s); return EIO; } diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c index d6d02b6d..4c09e1d 100644 --- a/sys/kern/vfs_bio.c +++ b/sys/kern/vfs_bio.c @@ -11,7 +11,7 @@ * 2. Absolutely no warranty of function or purpose is made by the author * John S. Dyson. * - * $Id: vfs_bio.c,v 1.152 1998/03/01 04:18:42 dyson Exp $ + * $Id: vfs_bio.c,v 1.153 1998/03/04 03:17:30 dyson Exp $ */ /* @@ -138,7 +138,7 @@ SYSCTL_INT(_vfs, OID_AUTO, kvafreespace, CTLFLAG_RD, &kvafreespace, 0, ""); static LIST_HEAD(bufhashhdr, buf) bufhashtbl[BUFHSZ], invalhash; -static TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES]; +struct bqueues bufqueues[BUFFER_QUEUES] = {0}; extern int vm_swap_size; @@ -520,7 +520,7 @@ brelse(struct buf * bp) relpbuf(bp); return; } - /* anyone need a "free" block? */ + s = splbio(); /* anyone need this block? */ @@ -538,10 +538,11 @@ brelse(struct buf * bp) if (bp->b_flags & B_DELWRI) --numdirtybuffers; bp->b_flags &= ~(B_DELWRI | B_CACHE); - if (((bp->b_flags & B_VMIO) == 0) && bp->b_vp) { + if ((bp->b_flags & B_VMIO) == 0) { if (bp->b_bufsize) allocbuf(bp, 0); - brelvp(bp); + if (bp->b_vp) + brelvp(bp); } } @@ -571,78 +572,72 @@ brelse(struct buf * bp) && bp->b_validend == bp->b_bufsize)) #endif ) { - vm_ooffset_t foff; - vm_object_t obj; - int i, resid; + + int i, j, resid; vm_page_t m; + off_t foff; + vm_pindex_t poff; + vm_object_t obj; struct vnode *vp; - int iototal = bp->b_bufsize; + int blksize; vp = bp->b_vp; -#if !defined(MAX_PERF) - if (!vp) - panic("brelse: missing vp"); -#endif + if (vp->v_type == VBLK) + blksize = DEV_BSIZE; + else + blksize = vp->v_mount->mnt_stat.f_iosize; - if (bp->b_npages) { - vm_pindex_t poff; - obj = (vm_object_t) vp->v_object; - if (vp->v_type == VBLK) - foff = ((vm_ooffset_t) bp->b_lblkno) << DEV_BSHIFT; - else - foff = (vm_ooffset_t) vp->v_mount->mnt_stat.f_iosize * bp->b_lblkno; - poff = OFF_TO_IDX(foff); - for (i = 0; i < bp->b_npages; i++) { - m = bp->b_pages[i]; - if (m == bogus_page) { - m = vm_page_lookup(obj, poff + i); + resid = bp->b_bufsize; + foff = -1LL; + + for (i = 0; i < bp->b_npages; i++) { + m = bp->b_pages[i]; + if (m == bogus_page) { + + obj = (vm_object_t) vp->v_object; + + foff = (off_t) bp->b_lblkno * blksize; + poff = OFF_TO_IDX(foff); + + for (j = i; j < bp->b_npages; j++) { + m = bp->b_pages[j]; + if (m == bogus_page) { + m = vm_page_lookup(obj, poff + j); #if !defined(MAX_PERF) - if (!m) { - panic("brelse: page missing\n"); - } -#endif - bp->b_pages[i] = m; - pmap_qenter(trunc_page(bp->b_data), - bp->b_pages, bp->b_npages); - } - resid = IDX_TO_OFF(m->pindex+1) - foff; - if (resid > iototal) - resid = iototal; - if (resid > 0) { - /* - * Don't invalidate the page if the local machine has already - * modified it. This is the lesser of two evils, and should - * be fixed. - */ - if (bp->b_flags & (B_NOCACHE | B_ERROR)) { - vm_page_test_dirty(m); - if (m->dirty == 0) { - vm_page_set_invalid(m, (vm_offset_t) foff, resid); - if (m->valid == 0) - vm_page_protect(m, VM_PROT_NONE); - } - } - if (resid >= PAGE_SIZE) { - if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { - bp->b_flags |= B_INVAL; - } - } else { - if (!vm_page_is_valid(m, - (((vm_offset_t) bp->b_data) & PAGE_MASK), resid)) { - bp->b_flags |= B_INVAL; + if (!m) { + panic("brelse: page missing\n"); } +#endif + bp->b_pages[j] = m; } } - foff += resid; - iototal -= resid; + + if ((bp->b_flags & B_INVAL) == 0) { + pmap_qenter(trunc_page(bp->b_data), bp->b_pages, bp->b_npages); + } + break; + } + if (bp->b_flags & (B_NOCACHE|B_ERROR)) { + if ((blksize & PAGE_MASK) == 0) { + vm_page_set_invalid(m, 0, resid); + } else { + if (foff == -1LL) + foff = (off_t) bp->b_lblkno * blksize; + vm_page_set_invalid(m, (vm_offset_t) foff, resid); + } } + resid -= PAGE_SIZE; } + if (bp->b_flags & (B_INVAL | B_RELBUF)) vfs_vmio_release(bp); + } else if (bp->b_flags & B_VMIO) { + if (bp->b_flags & (B_INVAL | B_RELBUF)) vfs_vmio_release(bp); + } #if !defined(MAX_PERF) @@ -755,6 +750,7 @@ vfs_vmio_release(bp) m = bp->b_pages[i]; bp->b_pages[i] = NULL; vm_page_unwire(m); + /* * We don't mess with busy pages, it is * the responsibility of the process that @@ -765,11 +761,6 @@ vfs_vmio_release(bp) if (m->wire_count == 0) { - if (m->flags & PG_WANTED) { - m->flags &= ~PG_WANTED; - wakeup(m); - } - /* * If this is an async free -- we cannot place * pages onto the cache queue. If it is an @@ -895,33 +886,6 @@ vfs_bio_awrite(struct buf * bp) return nwritten; } } -#if 0 - else if ((vp->v_flag & VOBJBUF) && (vp->v_type == VBLK) && - ((size = bp->b_bufsize) >= PAGE_SIZE)) { - maxcl = MAXPHYS / size; - for (i = 1; i < maxcl; i++) { - if ((bpa = gbincore(vp, lblkno + i)) && - ((bpa->b_flags & (B_BUSY | B_DELWRI | B_CLUSTEROK | B_INVAL)) == - (B_DELWRI | B_CLUSTEROK)) && - (bpa->b_bufsize == size)) { - if (bpa->b_blkno != - bp->b_blkno + ((i * size) >> DEV_BSHIFT)) - break; - } else { - break; - } - } - ncl = i; - /* - * this is a possible cluster write - */ - if (ncl != 1) { - nwritten = cluster_wbuild(vp, size, lblkno, ncl); - splx(s); - return nwritten; - } - } -#endif bremfree(bp); splx(s); @@ -1362,7 +1326,7 @@ struct buf * getblk(struct vnode * vp, daddr_t blkno, int size, int slpflag, int slptimeo) { struct buf *bp; - int s; + int i, s; struct bufhashhdr *bh; int maxsize; int generation; @@ -1474,9 +1438,10 @@ loop1: } else { bp->b_flags &= ~B_VMIO; } - splx(s); allocbuf(bp, size); + + splx(s); #ifdef PC98 /* * 1024byte/sector support @@ -1660,6 +1625,8 @@ allocbuf(struct buf * bp, int size) int pageindex, curbpnpages; struct vnode *vp; int bsize; + int orig_validoff = bp->b_validoff; + int orig_validend = bp->b_validend; vp = bp->b_vp; @@ -1676,8 +1643,9 @@ allocbuf(struct buf * bp, int size) off = (vm_ooffset_t) bp->b_lblkno * bsize; curbpnpages = bp->b_npages; doretry: + bp->b_validoff = orig_validoff; + bp->b_validend = orig_validend; bp->b_flags |= B_CACHE; - bp->b_validoff = bp->b_validend = 0; for (toff = 0; toff < newbsize; toff += tinc) { int bytesinpage; @@ -1705,14 +1673,11 @@ allocbuf(struct buf * bp, int size) vm_pageout_deficit += (desiredpages - bp->b_npages); goto doretry; } - /* - * Normally it is unwise to clear PG_BUSY without - * PAGE_WAKEUP -- but it is okay here, as there is - * no chance for blocking between here and vm_page_alloc - */ - m->flags &= ~PG_BUSY; + vm_page_wire(m); + m->flags &= ~PG_BUSY; bp->b_flags &= ~B_CACHE; + } else if (m->flags & PG_BUSY) { s = splvm(); if (m->flags & PG_BUSY) { @@ -1935,14 +1900,13 @@ biodone(register struct buf * bp) #endif panic("biodone: page busy < 0\n"); } - m->flags |= PG_BUSY; - --m->busy; - PAGE_WAKEUP(m); + PAGE_BWAKEUP(m); --obj->paging_in_progress; foff += resid; iosize -= resid; } - if (obj && obj->paging_in_progress == 0 && + if (obj && + (obj->paging_in_progress == 0) && (obj->flags & OBJ_PIPWNT)) { obj->flags &= ~OBJ_PIPWNT; wakeup(obj); @@ -2038,9 +2002,7 @@ vfs_unbusy_pages(struct buf * bp) pmap_qenter(trunc_page(bp->b_data), bp->b_pages, bp->b_npages); } --obj->paging_in_progress; - m->flags |= PG_BUSY; - --m->busy; - PAGE_WAKEUP(m); + PAGE_BWAKEUP(m); } if (obj->paging_in_progress == 0 && (obj->flags & OBJ_PIPWNT)) { @@ -2271,6 +2233,7 @@ tryagain: goto tryagain; } vm_page_wire(p); + p->valid = VM_PAGE_BITS_ALL; pmap_kenter(pg, VM_PAGE_TO_PHYS(p)); bp->b_pages[index] = p; PAGE_WAKEUP(p); diff --git a/sys/kern/vfs_cluster.c b/sys/kern/vfs_cluster.c index 220f760..7f477bf 100644 --- a/sys/kern/vfs_cluster.c +++ b/sys/kern/vfs_cluster.c @@ -33,7 +33,7 @@ * SUCH DAMAGE. * * @(#)vfs_cluster.c 8.7 (Berkeley) 2/13/94 - * $Id: vfs_cluster.c,v 1.54 1998/02/04 22:32:39 eivind Exp $ + * $Id: vfs_cluster.c,v 1.55 1998/02/06 12:13:30 eivind Exp $ */ #include "opt_debug_cluster.h" @@ -150,17 +150,12 @@ cluster_read(vp, filesize, lblkno, size, cred, totread, seqcount, bpp) (i == (maxra - 1))) tbp->b_flags |= B_RAM; -#if 0 - if (tbp->b_usecount == 0) { - /* - * Make sure that the soon-to-be used readaheads - * are still there. The getblk/bqrelse pair will - * boost the priority of the buffer. - */ - tbp = getblk(vp, lblkno+i, size, 0, 0); - bqrelse(tbp); + if ((tbp->b_usecount < 5) && + ((tbp->b_flags & B_BUSY) == 0) && + (tbp->b_qindex == QUEUE_LRU)) { + TAILQ_REMOVE(&bufqueues[QUEUE_LRU], tbp, b_freelist); + TAILQ_INSERT_TAIL(&bufqueues[QUEUE_LRU], tbp, b_freelist); } -#endif } splx(s); if (i >= maxra) { @@ -215,7 +210,6 @@ single_block_read: * if we have been doing sequential I/O, then do some read-ahead */ rbp = NULL; - /* if (seqcount && (lblkno < (origblkno + maxra))) { */ if (seqcount && (lblkno < (origblkno + seqcount))) { /* * we now build the read-ahead buffer if it is desirable. diff --git a/sys/kern/vfs_default.c b/sys/kern/vfs_default.c index 1aae395..33a2450 100644 --- a/sys/kern/vfs_default.c +++ b/sys/kern/vfs_default.c @@ -342,7 +342,7 @@ vop_sharedlock(ap) return (0); MALLOC(vp->v_vnlock, struct lock *, sizeof(struct lock), M_VNODE, M_WAITOK); - lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); + lockinit(vp->v_vnlock, PVFS, "vnlock", 0, LK_NOPAUSE); } switch (flags & LK_TYPE_MASK) { case LK_DRAIN: @@ -410,7 +410,7 @@ vop_nolock(ap) return (0); MALLOC(vp->v_vnlock, struct lock *, sizeof(struct lock), M_VNODE, M_WAITOK); - lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); + lockinit(vp->v_vnlock, PVFS, "vnlock", 0, LK_NOPAUSE); } switch (flags & LK_TYPE_MASK) { case LK_DRAIN: diff --git a/sys/kern/vfs_export.c b/sys/kern/vfs_export.c index 052d122..5d27cf5 100644 --- a/sys/kern/vfs_export.c +++ b/sys/kern/vfs_export.c @@ -36,7 +36,7 @@ * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 - * $Id: vfs_subr.c,v 1.135 1998/03/01 04:18:44 dyson Exp $ + * $Id: vfs_subr.c,v 1.136 1998/03/01 23:07:45 dyson Exp $ */ /* @@ -181,7 +181,7 @@ vfs_busy(mp, flags, interlkp, p) } return (ENOENT); } - lkflags = LK_SHARED; + lkflags = LK_SHARED | LK_NOPAUSE; if (interlkp) lkflags |= LK_INTERLOCK; if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) @@ -224,7 +224,7 @@ vfs_rootmountalloc(fstypename, devname, mpp) return (ENODEV); mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); bzero((char *)mp, (u_long)sizeof(struct mount)); - lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); + lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); (void)vfs_busy(mp, LK_NOWAIT, 0, p); LIST_INIT(&mp->mnt_vnodelist); mp->mnt_vfc = vfsp; diff --git a/sys/kern/vfs_extattr.c b/sys/kern/vfs_extattr.c index b8fba2c..596de95 100644 --- a/sys/kern/vfs_extattr.c +++ b/sys/kern/vfs_extattr.c @@ -36,7 +36,7 @@ * SUCH DAMAGE. * * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 - * $Id: vfs_syscalls.c,v 1.92 1998/02/08 01:41:33 dyson Exp $ + * $Id: vfs_syscalls.c,v 1.93 1998/02/15 04:17:09 dyson Exp $ */ /* For 4.3 integer FS ID compatibility */ @@ -244,7 +244,7 @@ mount(p, uap) mp = (struct mount *)malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); bzero((char *)mp, (u_long)sizeof(struct mount)); - lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); + lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); (void)vfs_busy(mp, LK_NOWAIT, 0, p); mp->mnt_op = vfsp->vfc_vfsops; mp->mnt_vfc = vfsp; @@ -2158,9 +2158,9 @@ fsync(p, uap) if (error = getvnode(p->p_fd, SCARG(uap, fd), &fp)) return (error); vp = (struct vnode *)fp->f_data; - if ((error = vn_lock(vp, LK_EXCLUSIVE|LK_RETRY, p)) == NULL) { + if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p)) == NULL) { if (vp->v_object) { - vm_object_page_clean(vp->v_object, 0, 0 ,0); + vm_object_page_clean(vp->v_object, 0, 0, FALSE); } error = VOP_FSYNC(vp, fp->f_cred, (vp->v_mount && (vp->v_mount->mnt_flag & MNT_ASYNC)) ? diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c index 052d122..5d27cf5 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -36,7 +36,7 @@ * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 - * $Id: vfs_subr.c,v 1.135 1998/03/01 04:18:44 dyson Exp $ + * $Id: vfs_subr.c,v 1.136 1998/03/01 23:07:45 dyson Exp $ */ /* @@ -181,7 +181,7 @@ vfs_busy(mp, flags, interlkp, p) } return (ENOENT); } - lkflags = LK_SHARED; + lkflags = LK_SHARED | LK_NOPAUSE; if (interlkp) lkflags |= LK_INTERLOCK; if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) @@ -224,7 +224,7 @@ vfs_rootmountalloc(fstypename, devname, mpp) return (ENODEV); mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); bzero((char *)mp, (u_long)sizeof(struct mount)); - lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); + lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); (void)vfs_busy(mp, LK_NOWAIT, 0, p); LIST_INIT(&mp->mnt_vnodelist); mp->mnt_vfc = vfsp; diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c index b8fba2c..596de95 100644 --- a/sys/kern/vfs_syscalls.c +++ b/sys/kern/vfs_syscalls.c @@ -36,7 +36,7 @@ * SUCH DAMAGE. * * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 - * $Id: vfs_syscalls.c,v 1.92 1998/02/08 01:41:33 dyson Exp $ + * $Id: vfs_syscalls.c,v 1.93 1998/02/15 04:17:09 dyson Exp $ */ /* For 4.3 integer FS ID compatibility */ @@ -244,7 +244,7 @@ mount(p, uap) mp = (struct mount *)malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); bzero((char *)mp, (u_long)sizeof(struct mount)); - lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); + lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); (void)vfs_busy(mp, LK_NOWAIT, 0, p); mp->mnt_op = vfsp->vfc_vfsops; mp->mnt_vfc = vfsp; @@ -2158,9 +2158,9 @@ fsync(p, uap) if (error = getvnode(p->p_fd, SCARG(uap, fd), &fp)) return (error); vp = (struct vnode *)fp->f_data; - if ((error = vn_lock(vp, LK_EXCLUSIVE|LK_RETRY, p)) == NULL) { + if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p)) == NULL) { if (vp->v_object) { - vm_object_page_clean(vp->v_object, 0, 0 ,0); + vm_object_page_clean(vp->v_object, 0, 0, FALSE); } error = VOP_FSYNC(vp, fp->f_cred, (vp->v_mount && (vp->v_mount->mnt_flag & MNT_ASYNC)) ? diff --git a/sys/miscfs/specfs/spec_vnops.c b/sys/miscfs/specfs/spec_vnops.c index 50a70dc..c058665 100644 --- a/sys/miscfs/specfs/spec_vnops.c +++ b/sys/miscfs/specfs/spec_vnops.c @@ -31,7 +31,7 @@ * SUCH DAMAGE. * * @(#)spec_vnops.c 8.14 (Berkeley) 5/21/95 - * $Id: spec_vnops.c,v 1.56 1998/02/06 12:13:43 eivind Exp $ + * $Id: spec_vnops.c,v 1.57 1998/03/04 06:44:59 dyson Exp $ */ #include <sys/param.h> @@ -723,8 +723,10 @@ spec_getpages(ap) daddr_t blkno; struct buf *bp; vm_ooffset_t offset; + int toff, nextoff, nread; struct vnode *vp = ap->a_vp; int blksiz; + int gotreqpage; error = 0; pcount = round_page(ap->a_count) / PAGE_SIZE; @@ -788,8 +790,6 @@ spec_getpages(ap) /* Do the input. */ VOP_STRATEGY(bp); - if (bp->b_flags & B_ASYNC) - return (VM_PAGER_PEND); s = splbio(); @@ -799,12 +799,19 @@ spec_getpages(ap) splx(s); - if ((bp->b_flags & B_ERROR) != 0) - error = EIO; + if ((bp->b_flags & B_ERROR) != 0) { + if (bp->b_error) + error = bp->b_error; + else + error = EIO; + } + + nread = size - bp->b_resid; - if (!error && ap->a_count != pcount * PAGE_SIZE) - bzero((caddr_t)kva + ap->a_count, - PAGE_SIZE * pcount - ap->a_count); + if (nread < ap->a_count) { + bzero((caddr_t)kva + nread, + ap->a_count - nread); + } pmap_qremove(kva, pcount); /* @@ -812,36 +819,53 @@ spec_getpages(ap) */ relpbuf(bp); - for (i = 0; i < pcount; i++) { - ap->a_m[i]->dirty = 0; - ap->a_m[i]->valid = VM_PAGE_BITS_ALL; - ap->a_m[i]->flags &= ~PG_ZERO; - 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. - */ + gotreqpage = 0; + for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) { + vm_page_t m; + nextoff = toff + PAGE_SIZE; + m = ap->a_m[i]; + + m->flags &= ~PG_ZERO; + if (nextoff <= nread) { + m->valid = VM_PAGE_BITS_ALL; + m->dirty = 0; + } else if (toff < nread) { + int nvalid = ((nread + DEV_BSIZE - 1) - toff) & ~(DEV_BSIZE - 1); + vm_page_set_validclean(m, 0, nvalid); + } else { + m->valid = 0; + m->dirty = 0; + } + + if (i != ap->a_reqpage) { /* * Just in case someone was asking for this page we * now tell them that it is ok to use. */ - if (!error) { - if (ap->a_m[i]->flags & PG_WANTED) - vm_page_activate(ap->a_m[i]); - else - vm_page_deactivate(ap->a_m[i]); - PAGE_WAKEUP(ap->a_m[i]); - } else - vnode_pager_freepage(ap->a_m[i]); + if (!error || (m->valid == VM_PAGE_BITS_ALL)) { + if (m->valid) { + if (m->flags & PG_WANTED) { + vm_page_activate(m); + } else { + vm_page_deactivate(m); + } + PAGE_WAKEUP(m); + } else { + vm_page_free(m); + } + } else { + vm_page_free(m); + } + } else if (m->valid) { + gotreqpage = 1; } } - if (error) - printf("spec_getpages: I/O read error\n"); - return (error ? VM_PAGER_ERROR : VM_PAGER_OK); + if (!gotreqpage) { + printf("spec_getpages: I/O read failure: (code=%d)\n", error); + return VM_PAGER_ERROR; + } + return VM_PAGER_OK; } /* ARGSUSED */ diff --git a/sys/nfs/nfs_bio.c b/sys/nfs/nfs_bio.c index a03b91b..be2b423 100644 --- a/sys/nfs/nfs_bio.c +++ b/sys/nfs/nfs_bio.c @@ -34,7 +34,7 @@ * SUCH DAMAGE. * * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 - * $Id: nfs_bio.c,v 1.50 1998/02/06 12:13:55 eivind Exp $ + * $Id: nfs_bio.c,v 1.51 1998/03/06 09:46:43 msmith Exp $ */ @@ -65,6 +65,7 @@ static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, struct proc *p)); +static void nfs_prot_buf __P((struct buf *bp, int off, int n)); extern int nfs_numasync; extern struct nfsstats nfsstats; @@ -76,58 +77,153 @@ int nfs_getpages(ap) struct vop_getpages_args *ap; { - int i, pcount, error; + int i, error, nextoff, size, toff, npages; struct uio uio; struct iovec iov; vm_page_t m; vm_offset_t kva; + struct buf *bp; if ((ap->a_vp->v_object) == NULL) { printf("nfs_getpages: called with non-merged cache vnode??\n"); return EOPNOTSUPP; } - m = ap->a_m[ap->a_reqpage]; - kva = vm_pager_map_page(m); + /* + * We use only the kva address for the buffer, but this is extremely + * convienient and fast. + */ + bp = getpbuf(); + + npages = btoc(ap->a_count); + kva = (vm_offset_t) bp->b_data; + pmap_qenter(kva, ap->a_m, npages); iov.iov_base = (caddr_t) kva; - iov.iov_len = PAGE_SIZE; + iov.iov_len = ap->a_count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; - uio.uio_offset = IDX_TO_OFF(m->pindex); - uio.uio_resid = PAGE_SIZE; + uio.uio_offset = IDX_TO_OFF(ap->a_m[0]->pindex); + uio.uio_resid = ap->a_count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_procp = curproc; error = nfs_readrpc(ap->a_vp, &uio, curproc->p_ucred); - vm_pager_unmap_page(kva); + pmap_qremove(kva, npages); + + relpbuf(bp); + + if (error && (uio.uio_resid == ap->a_count)) + return VM_PAGER_ERROR; + + size = ap->a_count - uio.uio_resid; - pcount = round_page(ap->a_count) / PAGE_SIZE; - for (i = 0; i < pcount; i++) { + for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { + vm_page_t m; + nextoff = toff + PAGE_SIZE; + m = ap->a_m[i]; + + m->flags &= ~PG_ZERO; + + if (nextoff <= size) { + m->valid = VM_PAGE_BITS_ALL; + m->dirty = 0; + } else { + int nvalid = ((size + DEV_BSIZE - 1) - toff) & ~(DEV_BSIZE - 1); + vm_page_set_validclean(m, 0, nvalid); + } + if (i != ap->a_reqpage) { - vnode_pager_freepage(ap->a_m[i]); + /* + * 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->flags & PG_WANTED) + vm_page_activate(m); + else + vm_page_deactivate(m); + PAGE_WAKEUP(m); + } else { + vnode_pager_freepage(m); + } } } - - if (error && (uio.uio_resid == PAGE_SIZE)) - return VM_PAGER_ERROR; return 0; } - /* - * put page routine - * - * XXX By default, wimp out... note that a_offset is ignored (and always - * XXX has been). + * Vnode op for VM putpages. */ int nfs_putpages(ap) struct vop_putpages_args *ap; { - return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count, - ap->a_sync, ap->a_rtvals); + struct uio uio; + struct iovec iov; + vm_page_t m; + vm_offset_t kva; + struct buf *bp; + int iomode, must_commit, i, error, npages; + int *rtvals; + + rtvals = ap->a_rtvals; + + npages = btoc(ap->a_count); + + for (i = 0; i < npages; i++) { + rtvals[i] = VM_PAGER_AGAIN; + } + + /* + * We use only the kva address for the buffer, but this is extremely + * convienient and fast. + */ + bp = getpbuf(); + + kva = (vm_offset_t) bp->b_data; + pmap_qenter(kva, ap->a_m, npages); + + iov.iov_base = (caddr_t) kva; + iov.iov_len = ap->a_count; + uio.uio_iov = &iov; + uio.uio_iovcnt = 1; + uio.uio_offset = IDX_TO_OFF(ap->a_m[0]->pindex); + uio.uio_resid = ap->a_count; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_WRITE; + uio.uio_procp = curproc; + + if ((ap->a_sync & VM_PAGER_PUT_SYNC) == 0) + iomode = NFSV3WRITE_UNSTABLE; + else + iomode = NFSV3WRITE_FILESYNC; + + error = nfs_writerpc(ap->a_vp, &uio, + curproc->p_ucred, &iomode, &must_commit); + + pmap_qremove(kva, npages); + relpbuf(bp); + + if (!error) { + int nwritten = round_page(ap->a_count - uio.uio_resid) / PAGE_SIZE; + for (i = 0; i < nwritten; i++) { + rtvals[i] = VM_PAGER_OK; + ap->a_m[i]->dirty = 0; + } + if (must_commit) + nfs_clearcommit(ap->a_vp->v_mount); + } + return ap->a_rtvals[0]; } /* @@ -464,7 +560,7 @@ again: }; if (n > 0) { - error = uiomove(bp->b_data + on, (int)n, uio); + error = uiomove(bp->b_data + on, (int)n, uio); } switch (vp->v_type) { case VREG: @@ -484,6 +580,24 @@ again: return (error); } +static void +nfs_prot_buf(bp, off, n) + struct buf *bp; + int off; + int n; +{ + int pindex, boff, end; + + if ((bp->b_flags & B_VMIO) == 0) + return; + + end = round_page(off + n); + for (boff = trunc_page(off); boff < end; boff += PAGE_SIZE) { + pindex = boff >> PAGE_SHIFT; + vm_page_protect(bp->b_pages[pindex], VM_PROT_NONE); + } +} + /* * Vnode op for write using bio */ @@ -648,12 +762,19 @@ again: goto again; } } + error = uiomove((char *)bp->b_data + on, n, uio); if (error) { bp->b_flags |= B_ERROR; brelse(bp); return (error); } + + /* + * This will keep the buffer and mmaped regions more coherent. + */ + nfs_prot_buf(bp, on, n); + if (bp->b_dirtyend > 0) { bp->b_dirtyoff = min(on, bp->b_dirtyoff); bp->b_dirtyend = max((on + n), bp->b_dirtyend); @@ -681,6 +802,8 @@ again: */ if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { bp->b_proc = p; + if (ioflag & IO_INVAL) + bp->b_flags |= B_INVAL; error = VOP_BWRITE(bp); if (error) return (error); diff --git a/sys/nfs/nfs_vnops.c b/sys/nfs/nfs_vnops.c index 60b8460..1021f47 100644 --- a/sys/nfs/nfs_vnops.c +++ b/sys/nfs/nfs_vnops.c @@ -34,7 +34,7 @@ * SUCH DAMAGE. * * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 - * $Id: nfs_vnops.c,v 1.78 1998/02/06 12:13:58 eivind Exp $ + * $Id: nfs_vnops.c,v 1.79 1998/03/06 09:46:48 msmith Exp $ */ @@ -3000,6 +3000,9 @@ nfs_writebp(bp, force) if(!(bp->b_flags & B_BUSY)) panic("bwrite: buffer is not busy???"); + if (bp->b_flags & B_INVAL) + bp->b_flags |= B_INVAL | B_NOCACHE; + if (bp->b_flags & B_DELWRI) { --numdirtybuffers; if (needsbuffer) @@ -3045,6 +3048,7 @@ nfs_writebp(bp, force) if (oldflags & B_DELWRI) { reassignbuf(bp, bp->b_vp); } + brelse(bp); return (rtval); } diff --git a/sys/nfsclient/nfs_bio.c b/sys/nfsclient/nfs_bio.c index a03b91b..be2b423 100644 --- a/sys/nfsclient/nfs_bio.c +++ b/sys/nfsclient/nfs_bio.c @@ -34,7 +34,7 @@ * SUCH DAMAGE. * * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 - * $Id: nfs_bio.c,v 1.50 1998/02/06 12:13:55 eivind Exp $ + * $Id: nfs_bio.c,v 1.51 1998/03/06 09:46:43 msmith Exp $ */ @@ -65,6 +65,7 @@ static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size, struct proc *p)); +static void nfs_prot_buf __P((struct buf *bp, int off, int n)); extern int nfs_numasync; extern struct nfsstats nfsstats; @@ -76,58 +77,153 @@ int nfs_getpages(ap) struct vop_getpages_args *ap; { - int i, pcount, error; + int i, error, nextoff, size, toff, npages; struct uio uio; struct iovec iov; vm_page_t m; vm_offset_t kva; + struct buf *bp; if ((ap->a_vp->v_object) == NULL) { printf("nfs_getpages: called with non-merged cache vnode??\n"); return EOPNOTSUPP; } - m = ap->a_m[ap->a_reqpage]; - kva = vm_pager_map_page(m); + /* + * We use only the kva address for the buffer, but this is extremely + * convienient and fast. + */ + bp = getpbuf(); + + npages = btoc(ap->a_count); + kva = (vm_offset_t) bp->b_data; + pmap_qenter(kva, ap->a_m, npages); iov.iov_base = (caddr_t) kva; - iov.iov_len = PAGE_SIZE; + iov.iov_len = ap->a_count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; - uio.uio_offset = IDX_TO_OFF(m->pindex); - uio.uio_resid = PAGE_SIZE; + uio.uio_offset = IDX_TO_OFF(ap->a_m[0]->pindex); + uio.uio_resid = ap->a_count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_procp = curproc; error = nfs_readrpc(ap->a_vp, &uio, curproc->p_ucred); - vm_pager_unmap_page(kva); + pmap_qremove(kva, npages); + + relpbuf(bp); + + if (error && (uio.uio_resid == ap->a_count)) + return VM_PAGER_ERROR; + + size = ap->a_count - uio.uio_resid; - pcount = round_page(ap->a_count) / PAGE_SIZE; - for (i = 0; i < pcount; i++) { + for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { + vm_page_t m; + nextoff = toff + PAGE_SIZE; + m = ap->a_m[i]; + + m->flags &= ~PG_ZERO; + + if (nextoff <= size) { + m->valid = VM_PAGE_BITS_ALL; + m->dirty = 0; + } else { + int nvalid = ((size + DEV_BSIZE - 1) - toff) & ~(DEV_BSIZE - 1); + vm_page_set_validclean(m, 0, nvalid); + } + if (i != ap->a_reqpage) { - vnode_pager_freepage(ap->a_m[i]); + /* + * 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->flags & PG_WANTED) + vm_page_activate(m); + else + vm_page_deactivate(m); + PAGE_WAKEUP(m); + } else { + vnode_pager_freepage(m); + } } } - - if (error && (uio.uio_resid == PAGE_SIZE)) - return VM_PAGER_ERROR; return 0; } - /* - * put page routine - * - * XXX By default, wimp out... note that a_offset is ignored (and always - * XXX has been). + * Vnode op for VM putpages. */ int nfs_putpages(ap) struct vop_putpages_args *ap; { - return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count, - ap->a_sync, ap->a_rtvals); + struct uio uio; + struct iovec iov; + vm_page_t m; + vm_offset_t kva; + struct buf *bp; + int iomode, must_commit, i, error, npages; + int *rtvals; + + rtvals = ap->a_rtvals; + + npages = btoc(ap->a_count); + + for (i = 0; i < npages; i++) { + rtvals[i] = VM_PAGER_AGAIN; + } + + /* + * We use only the kva address for the buffer, but this is extremely + * convienient and fast. + */ + bp = getpbuf(); + + kva = (vm_offset_t) bp->b_data; + pmap_qenter(kva, ap->a_m, npages); + + iov.iov_base = (caddr_t) kva; + iov.iov_len = ap->a_count; + uio.uio_iov = &iov; + uio.uio_iovcnt = 1; + uio.uio_offset = IDX_TO_OFF(ap->a_m[0]->pindex); + uio.uio_resid = ap->a_count; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_WRITE; + uio.uio_procp = curproc; + + if ((ap->a_sync & VM_PAGER_PUT_SYNC) == 0) + iomode = NFSV3WRITE_UNSTABLE; + else + iomode = NFSV3WRITE_FILESYNC; + + error = nfs_writerpc(ap->a_vp, &uio, + curproc->p_ucred, &iomode, &must_commit); + + pmap_qremove(kva, npages); + relpbuf(bp); + + if (!error) { + int nwritten = round_page(ap->a_count - uio.uio_resid) / PAGE_SIZE; + for (i = 0; i < nwritten; i++) { + rtvals[i] = VM_PAGER_OK; + ap->a_m[i]->dirty = 0; + } + if (must_commit) + nfs_clearcommit(ap->a_vp->v_mount); + } + return ap->a_rtvals[0]; } /* @@ -464,7 +560,7 @@ again: }; if (n > 0) { - error = uiomove(bp->b_data + on, (int)n, uio); + error = uiomove(bp->b_data + on, (int)n, uio); } switch (vp->v_type) { case VREG: @@ -484,6 +580,24 @@ again: return (error); } +static void +nfs_prot_buf(bp, off, n) + struct buf *bp; + int off; + int n; +{ + int pindex, boff, end; + + if ((bp->b_flags & B_VMIO) == 0) + return; + + end = round_page(off + n); + for (boff = trunc_page(off); boff < end; boff += PAGE_SIZE) { + pindex = boff >> PAGE_SHIFT; + vm_page_protect(bp->b_pages[pindex], VM_PROT_NONE); + } +} + /* * Vnode op for write using bio */ @@ -648,12 +762,19 @@ again: goto again; } } + error = uiomove((char *)bp->b_data + on, n, uio); if (error) { bp->b_flags |= B_ERROR; brelse(bp); return (error); } + + /* + * This will keep the buffer and mmaped regions more coherent. + */ + nfs_prot_buf(bp, on, n); + if (bp->b_dirtyend > 0) { bp->b_dirtyoff = min(on, bp->b_dirtyoff); bp->b_dirtyend = max((on + n), bp->b_dirtyend); @@ -681,6 +802,8 @@ again: */ if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) { bp->b_proc = p; + if (ioflag & IO_INVAL) + bp->b_flags |= B_INVAL; error = VOP_BWRITE(bp); if (error) return (error); diff --git a/sys/nfsclient/nfs_vnops.c b/sys/nfsclient/nfs_vnops.c index 60b8460..1021f47 100644 --- a/sys/nfsclient/nfs_vnops.c +++ b/sys/nfsclient/nfs_vnops.c @@ -34,7 +34,7 @@ * SUCH DAMAGE. * * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95 - * $Id: nfs_vnops.c,v 1.78 1998/02/06 12:13:58 eivind Exp $ + * $Id: nfs_vnops.c,v 1.79 1998/03/06 09:46:48 msmith Exp $ */ @@ -3000,6 +3000,9 @@ nfs_writebp(bp, force) if(!(bp->b_flags & B_BUSY)) panic("bwrite: buffer is not busy???"); + if (bp->b_flags & B_INVAL) + bp->b_flags |= B_INVAL | B_NOCACHE; + if (bp->b_flags & B_DELWRI) { --numdirtybuffers; if (needsbuffer) @@ -3045,6 +3048,7 @@ nfs_writebp(bp, force) if (oldflags & B_DELWRI) { reassignbuf(bp, bp->b_vp); } + brelse(bp); return (rtval); } diff --git a/sys/sys/bio.h b/sys/sys/bio.h index 1cad39a..47117ed 100644 --- a/sys/sys/bio.h +++ b/sys/sys/bio.h @@ -36,7 +36,7 @@ * SUCH DAMAGE. * * @(#)buf.h 8.9 (Berkeley) 3/30/95 - * $Id: buf.h,v 1.44 1997/12/02 21:07:14 phk Exp $ + * $Id: buf.h,v 1.45 1998/01/22 17:30:10 dyson Exp $ */ #ifndef _SYS_BUF_H_ @@ -253,6 +253,7 @@ extern struct buf *swbuf; /* Swap I/O buffer headers. */ extern int nswbuf; /* Number of swap I/O buffer headers. */ extern int needsbuffer, numdirtybuffers; extern TAILQ_HEAD(swqueue, buf) bswlist; +extern TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES]; void bufinit __P((void)); void bremfree __P((struct buf *)); diff --git a/sys/sys/buf.h b/sys/sys/buf.h index 1cad39a..47117ed 100644 --- a/sys/sys/buf.h +++ b/sys/sys/buf.h @@ -36,7 +36,7 @@ * SUCH DAMAGE. * * @(#)buf.h 8.9 (Berkeley) 3/30/95 - * $Id: buf.h,v 1.44 1997/12/02 21:07:14 phk Exp $ + * $Id: buf.h,v 1.45 1998/01/22 17:30:10 dyson Exp $ */ #ifndef _SYS_BUF_H_ @@ -253,6 +253,7 @@ extern struct buf *swbuf; /* Swap I/O buffer headers. */ extern int nswbuf; /* Number of swap I/O buffer headers. */ extern int needsbuffer, numdirtybuffers; extern TAILQ_HEAD(swqueue, buf) bswlist; +extern TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES]; void bufinit __P((void)); void bremfree __P((struct buf *)); diff --git a/sys/sys/smp.h b/sys/sys/smp.h index d40a933..b10930a 100644 --- a/sys/sys/smp.h +++ b/sys/sys/smp.h @@ -6,7 +6,7 @@ * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp * ---------------------------------------------------------------------------- * - * $Id: smp.h,v 1.38 1998/03/03 22:56:26 tegge Exp $ + * $Id: smp.h,v 1.39 1998/03/05 21:45:44 tegge Exp $ * */ @@ -67,11 +67,11 @@ extern u_int mpintr_lock; /* functions in mplock.s */ void get_mplock __P((void)); void rel_mplock __P((void)); -int try_mplock __P((void)); +int try_mplock __P((void)); #ifdef RECURSIVE_MPINTRLOCK void get_mpintrlock __P((void)); void rel_mpintrlock __P((void)); -int try_mpintrlock __P((void)); +int try_mpintrlock __P((void)); #endif /* RECURSIVE_MPINTRLOCK */ /* global data in apic_vector.s */ diff --git a/sys/sys/vnode.h b/sys/sys/vnode.h index 7f2bd50..aed0b49 100644 --- a/sys/sys/vnode.h +++ b/sys/sys/vnode.h @@ -31,7 +31,7 @@ * SUCH DAMAGE. * * @(#)vnode.h 8.7 (Berkeley) 2/4/94 - * $Id: vnode.h,v 1.65 1998/01/17 09:16:39 dyson Exp $ + * $Id: vnode.h,v 1.66 1998/01/24 02:01:31 dyson Exp $ */ #ifndef _SYS_VNODE_H_ @@ -197,6 +197,7 @@ struct vattr { #define IO_NODELOCKED 0x08 /* underlying node already locked */ #define IO_NDELAY 0x10 /* FNDELAY flag set in file table */ #define IO_VMIO 0x20 /* data already in VMIO space */ +#define IO_INVAL 0x40 /* invalidate after I/O */ /* * Modes. Some values same as Ixxx entries from inode.h for now. diff --git a/sys/ufs/ffs/ffs_inode.c b/sys/ufs/ffs/ffs_inode.c index b86c43e..cf1c043 100644 --- a/sys/ufs/ffs/ffs_inode.c +++ b/sys/ufs/ffs/ffs_inode.c @@ -31,7 +31,7 @@ * SUCH DAMAGE. * * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95 - * $Id: ffs_inode.c,v 1.33 1998/02/04 22:33:31 eivind Exp $ + * $Id: ffs_inode.c,v 1.34 1998/02/06 12:14:14 eivind Exp $ */ #include "opt_quota.h" @@ -204,6 +204,7 @@ ffs_truncate(vp, length, flags, cred, p) * value of osize is 0, length will be at least 1. */ if (osize < length) { + vnode_pager_setsize(ovp, length); offset = blkoff(fs, length - 1); lbn = lblkno(fs, length - 1); aflags = B_CLRBUF; @@ -214,7 +215,6 @@ ffs_truncate(vp, length, flags, cred, p) if (error) return (error); oip->i_size = length; - vnode_pager_setsize(ovp, length); if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; if (aflags & B_SYNC) @@ -257,7 +257,6 @@ ffs_truncate(vp, length, flags, cred, p) else bawrite(bp); } - vnode_pager_setsize(ovp, length); /* * Calculate index into inode's block list of * last direct and indirect blocks (if any) @@ -298,6 +297,7 @@ ffs_truncate(vp, length, flags, cred, p) oip->i_size = osize; vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA; allerror = vinvalbuf(ovp, vflags, cred, p, 0, 0); + vnode_pager_setsize(ovp, length); /* * Indirect blocks first. diff --git a/sys/ufs/ffs/ffs_vfsops.c b/sys/ufs/ffs/ffs_vfsops.c index cd65a91..edfe0e6 100644 --- a/sys/ufs/ffs/ffs_vfsops.c +++ b/sys/ufs/ffs/ffs_vfsops.c @@ -31,7 +31,7 @@ * SUCH DAMAGE. * * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95 - * $Id: ffs_vfsops.c,v 1.73 1998/03/01 22:46:46 msmith Exp $ + * $Id: ffs_vfsops.c,v 1.74 1998/03/07 14:59:44 bde Exp $ */ #include "opt_quota.h" @@ -559,10 +559,7 @@ ffs_mountfs(devvp, mp, p, malloctype) if (error) return (error); ncount = vcount(devvp); -/* - if (devvp->v_object) - ncount -= 1; -*/ + if (ncount > 1 && devvp != rootvp) return (EBUSY); if (error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0)) @@ -983,8 +980,9 @@ ffs_vget(mp, ino, vpp) ump = VFSTOUFS(mp); dev = ump->um_dev; restart: - if ((*vpp = ufs_ihashget(dev, ino)) != NULL) + if ((*vpp = ufs_ihashget(dev, ino)) != NULL) { return (0); + } /* * Lock out the creation of new entries in the FFS hash table in diff --git a/sys/ufs/ufs/ufs_readwrite.c b/sys/ufs/ufs/ufs_readwrite.c index 19f7280..6abb130 100644 --- a/sys/ufs/ufs/ufs_readwrite.c +++ b/sys/ufs/ufs/ufs_readwrite.c @@ -31,7 +31,7 @@ * SUCH DAMAGE. * * @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 - * $Id: ufs_readwrite.c,v 1.42 1998/02/05 03:32:33 dyson Exp $ + * $Id: ufs_readwrite.c,v 1.43 1998/02/26 06:39:50 msmith Exp $ */ #define BLKSIZE(a, b, c) blksize(a, b, c) @@ -74,6 +74,7 @@ READ(ap) int error; u_short mode; int seqcount; + int ioflag; vm_object_t object; vp = ap->a_vp; @@ -81,6 +82,7 @@ READ(ap) ip = VTOI(vp); mode = ip->i_mode; uio = ap->a_uio; + ioflag = ap->a_ioflag; #ifdef DIAGNOSTIC if (uio->uio_rw != UIO_READ) @@ -106,7 +108,7 @@ READ(ap) if (object) vm_object_reference(object); #if 1 - if ((vfs_ioopt > 1) && object) { + if ((ioflag & IO_VMIO) == 0 && (vfs_ioopt > 1) && object) { int nread, toread; toread = uio->uio_resid; if (toread > bytesinfile) @@ -128,7 +130,7 @@ READ(ap) if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) break; #if 1 - if ((vfs_ioopt > 1) && object) { + if ((ioflag & IO_VMIO) == 0 && (vfs_ioopt > 1) && object) { int nread, toread; toread = uio->uio_resid; if (toread > bytesinfile) @@ -208,10 +210,23 @@ READ(ap) if (error) break; - bqrelse(bp); + if (ioflag & IO_VMIO) { + bp->b_flags |= B_RELBUF; + brelse(bp); + } else { + bqrelse(bp); + } + } + + if (bp != NULL) { + if (ioflag & IO_VMIO) { + bp->b_flags |= B_RELBUF; + brelse(bp); + } else { + bqrelse(bp); + } } - if (bp != NULL) - bqrelse(bp); + if (object) vm_object_vndeallocate(object); if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) @@ -397,8 +412,10 @@ ffs_getpages(ap) { off_t foff, physoffset; int i, size, bsize; - struct vnode *dp; + struct vnode *dp, *vp; vm_object_t obj; + vm_pindex_t pindex, firstindex; + vm_page_t m, mreq; int bbackwards, bforwards; int pbackwards, pforwards; int firstpage; @@ -411,56 +428,92 @@ ffs_getpages(ap) pcount = round_page(ap->a_count) / PAGE_SIZE; + mreq = ap->a_m[ap->a_reqpage]; + firstindex = ap->a_m[0]->pindex; + /* * if ANY DEV_BSIZE blocks are valid on a large filesystem block * then, the entire page is valid -- */ - if (ap->a_m[ap->a_reqpage]->valid) { - ap->a_m[ap->a_reqpage]->valid = VM_PAGE_BITS_ALL; + if (mreq->valid) { + mreq->valid = VM_PAGE_BITS_ALL; for (i = 0; i < pcount; i++) { - if (i != ap->a_reqpage) - vnode_pager_freepage(ap->a_m[i]); + if (i != ap->a_reqpage) { + vm_page_free(ap->a_m[i]); + } } return VM_PAGER_OK; } - obj = ap->a_m[ap->a_reqpage]->object; - bsize = ap->a_vp->v_mount->mnt_stat.f_iosize; - - if (obj->behavior == OBJ_SEQUENTIAL) { + vp = ap->a_vp; + obj = vp->v_object; + bsize = vp->v_mount->mnt_stat.f_iosize; + pindex = mreq->pindex; + foff = IDX_TO_OFF(pindex) /* + ap->a_offset should be zero */; + + if (firstindex == 0) + vp->v_lastr = 0; + + if ((obj->behavior != OBJ_RANDOM) && + ((firstindex != 0) && (firstindex <= vp->v_lastr) && + ((firstindex + pcount) > vp->v_lastr)) || + (obj->behavior == OBJ_SEQUENTIAL)) { struct uio auio; struct iovec aiov; int error; - vm_page_t m; for (i = 0; i < pcount; i++) { - if (i != ap->a_reqpage) { - vnode_pager_freepage(ap->a_m[i]); - } + m = ap->a_m[i]; + vm_page_activate(m); + m->busy++; + m->flags &= ~PG_BUSY; } - m = ap->a_m[ap->a_reqpage]; - - m->busy++; - m->flags &= ~PG_BUSY; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = 0; aiov.iov_len = MAXBSIZE; auio.uio_resid = MAXBSIZE; - auio.uio_offset = IDX_TO_OFF(m->pindex); + auio.uio_offset = foff; auio.uio_segflg = UIO_NOCOPY; auio.uio_rw = UIO_READ; auio.uio_procp = curproc; - error = VOP_READ(ap->a_vp, &auio, - ((MAXBSIZE / bsize) << 16), curproc->p_ucred); + error = VOP_READ(vp, &auio, + IO_VMIO | ((MAXBSIZE / bsize) << 16), curproc->p_ucred); - m->flags |= PG_BUSY; - m->busy--; + for (i = 0; i < pcount; i++) { + m = ap->a_m[i]; + m->busy--; + + if ((m != mreq) && (m->wire_count == 0) && (m->hold_count == 0) && + (m->valid == 0) && (m->busy == 0) && + (m->flags & PG_BUSY) == 0) { + m->flags |= PG_BUSY; + vm_page_free(m); + } else if (m == mreq) { + while (m->flags & PG_BUSY) { + vm_page_sleep(m, "ffspwt", NULL); + } + m->flags |= PG_BUSY; + vp->v_lastr = m->pindex + 1; + } else { + if (m->wire_count == 0) { + if (m->busy || (m->flags & PG_MAPPED) || + (m->flags & (PG_WANTED | PG_BUSY)) == PG_WANTED) { + vm_page_activate(m); + } else { + vm_page_deactivate(m); + } + } + vp->v_lastr = m->pindex + 1; + } + } - if (error && (auio.uio_resid == MAXBSIZE)) + if (mreq->valid == 0) return VM_PAGER_ERROR; - return 0; + + mreq->valid = VM_PAGE_BITS_ALL; + return VM_PAGER_OK; } /* @@ -468,21 +521,20 @@ ffs_getpages(ap) * reqlblkno is the logical block that contains the page * poff is the index of the page into the logical block */ - foff = IDX_TO_OFF(ap->a_m[ap->a_reqpage]->pindex) + ap->a_offset; reqlblkno = foff / bsize; poff = (foff % bsize) / PAGE_SIZE; - if ( VOP_BMAP( ap->a_vp, reqlblkno, &dp, &reqblkno, + if ( VOP_BMAP( vp, reqlblkno, &dp, &reqblkno, &bforwards, &bbackwards) || (reqblkno == -1)) { for(i = 0; i < pcount; i++) { if (i != ap->a_reqpage) - vnode_pager_freepage(ap->a_m[i]); + vm_page_free(ap->a_m[i]); } if (reqblkno == -1) { - if ((ap->a_m[ap->a_reqpage]->flags & PG_ZERO) == 0) - vm_page_zero_fill(ap->a_m[ap->a_reqpage]); - ap->a_m[ap->a_reqpage]->dirty = 0; - ap->a_m[ap->a_reqpage]->valid = VM_PAGE_BITS_ALL; + if ((mreq->flags & PG_ZERO) == 0) + vm_page_zero_fill(mreq); + mreq->dirty = 0; + mreq->valid = VM_PAGE_BITS_ALL; return VM_PAGER_OK; } else { return VM_PAGER_ERROR; @@ -502,7 +554,7 @@ ffs_getpages(ap) if (ap->a_reqpage > pbackwards) { firstpage = ap->a_reqpage - pbackwards; for(i=0;i<firstpage;i++) - vnode_pager_freepage(ap->a_m[i]); + vm_page_free(ap->a_m[i]); } /* @@ -513,7 +565,7 @@ ffs_getpages(ap) bforwards * pagesperblock; if (pforwards < (pcount - (ap->a_reqpage + 1))) { for( i = ap->a_reqpage + pforwards + 1; i < pcount; i++) - vnode_pager_freepage(ap->a_m[i]); + vm_page_free(ap->a_m[i]); pcount = ap->a_reqpage + pforwards + 1; } @@ -529,11 +581,13 @@ ffs_getpages(ap) */ size = pcount * PAGE_SIZE; + vp->v_lastr = mreq->pindex + pcount; + if ((IDX_TO_OFF(ap->a_m[firstpage]->pindex) + size) > - ((vm_object_t) ap->a_vp->v_object)->un_pager.vnp.vnp_size) - size = ((vm_object_t) ap->a_vp->v_object)->un_pager.vnp.vnp_size - IDX_TO_OFF(ap->a_m[firstpage]->pindex); + obj->un_pager.vnp.vnp_size) + size = obj->un_pager.vnp.vnp_size - foff; - physoffset -= IDX_TO_OFF(ap->a_m[ap->a_reqpage]->pindex); + physoffset -= foff; rtval = VOP_GETPAGES(dp, &ap->a_m[firstpage], size, (ap->a_reqpage - firstpage), physoffset); diff --git a/sys/vm/swap_pager.c b/sys/vm/swap_pager.c index 828eab9..53e0318 100644 --- a/sys/vm/swap_pager.c +++ b/sys/vm/swap_pager.c @@ -39,7 +39,7 @@ * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$ * * @(#)swap_pager.c 8.9 (Berkeley) 3/21/94 - * $Id: swap_pager.c,v 1.90 1998/02/25 03:55:47 dyson Exp $ + * $Id: swap_pager.c,v 1.91 1998/03/01 04:18:14 dyson Exp $ */ /* @@ -1578,9 +1578,7 @@ swap_pager_finish(spc) printf("swap_pager_finish: I/O error, clean of page %lx failed\n", (u_long) VM_PAGE_TO_PHYS(ma[i])); ma[i]->dirty = VM_PAGE_BITS_ALL; - ma[i]->flags |= PG_BUSY; - ma[i]->busy--; - PAGE_WAKEUP(ma[i]); + PAGE_BWAKEUP(ma[i]); } object->paging_in_progress -= spc->spc_count; @@ -1651,9 +1649,7 @@ swap_pager_iodone(bp) /* * we wakeup any processes that are waiting on these pages. */ - ma[i]->flags |= PG_BUSY; - ma[i]->busy--; - PAGE_WAKEUP(ma[i]); + PAGE_BWAKEUP(ma[i]); } } diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c index 930d687..7036e5f 100644 --- a/sys/vm/vm_map.c +++ b/sys/vm/vm_map.c @@ -61,7 +61,7 @@ * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_map.c,v 1.116 1998/02/23 08:22:33 dyson Exp $ + * $Id: vm_map.c,v 1.117 1998/02/25 03:55:49 dyson Exp $ */ /* @@ -286,7 +286,7 @@ vm_map_init(map, min, max) map->first_free = &map->header; map->hint = &map->header; map->timestamp = 0; - lockinit(&map->lock, PVM, "thrd_sleep", 0, 0); + lockinit(&map->lock, PVM, "thrd_sleep", 0, LK_NOPAUSE); } /* @@ -1665,12 +1665,15 @@ vm_map_clean(map, start, end, syncio, invalidate) * idea. */ if (current->protection & VM_PROT_WRITE) { + int flags; if (object->type == OBJT_VNODE) vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); + flags = (syncio || invalidate) ? OBJPC_SYNC : 0; + flags |= invalidate ? OBJPC_INVAL : 0; vm_object_page_clean(object, OFF_TO_IDX(offset), OFF_TO_IDX(offset + size + PAGE_MASK), - (syncio||invalidate)?1:0); + flags); if (invalidate) vm_object_page_remove(object, OFF_TO_IDX(offset), diff --git a/sys/vm/vm_mmap.c b/sys/vm/vm_mmap.c index 3184054..dc41acb 100644 --- a/sys/vm/vm_mmap.c +++ b/sys/vm/vm_mmap.c @@ -38,7 +38,7 @@ * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ * * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94 - * $Id: vm_mmap.c,v 1.72 1998/02/04 22:33:48 eivind Exp $ + * $Id: vm_mmap.c,v 1.73 1998/02/06 12:14:25 eivind Exp $ */ /* @@ -642,13 +642,6 @@ mincore(p, uap) vm_map_lock(map); - /* - * Not needed here - */ -#if 0 - VM_MAP_RANGE_CHECK(map, addr, end); -#endif - if (!vm_map_lookup_entry(map, addr, &entry)) entry = entry->next; diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c index fd63b8d..bb2549c 100644 --- a/sys/vm/vm_object.c +++ b/sys/vm/vm_object.c @@ -61,7 +61,7 @@ * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_object.c,v 1.114 1998/02/25 03:55:50 dyson Exp $ + * $Id: vm_object.c,v 1.115 1998/03/01 04:18:22 dyson Exp $ */ /* @@ -431,7 +431,7 @@ vm_object_terminate(object) /* * Clean pages and flush buffers. */ - vm_object_page_clean(object, 0, 0, TRUE); + vm_object_page_clean(object, 0, 0, OBJPC_SYNC); vp = (struct vnode *) object->handle; vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0); @@ -499,11 +499,11 @@ vm_object_dispose(object) */ void -vm_object_page_clean(object, start, end, syncio) +vm_object_page_clean(object, start, end, flags) vm_object_t object; vm_pindex_t start; vm_pindex_t end; - boolean_t syncio; + int flags; { register vm_page_t p, np, tp; register vm_offset_t tstart, tend; @@ -515,6 +515,7 @@ vm_object_page_clean(object, start, end, syncio) int chkb; int maxb; int i; + int pagerflags; vm_page_t maf[vm_pageout_page_count]; vm_page_t mab[vm_pageout_page_count]; vm_page_t ma[vm_pageout_page_count]; @@ -525,6 +526,9 @@ vm_object_page_clean(object, start, end, syncio) (object->flags & OBJ_MIGHTBEDIRTY) == 0) return; + pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) ? VM_PAGER_PUT_SYNC : 0; + pagerflags |= (flags & OBJPC_INVAL) ? VM_PAGER_PUT_INVAL : 0; + vp = object->handle; object->flags |= OBJ_CLEANING; @@ -628,29 +632,23 @@ rescan: for(i=0;i<maxb;i++) { int index = (maxb - i) - 1; ma[index] = mab[i]; - ma[index]->flags |= PG_BUSY; ma[index]->flags &= ~PG_CLEANCHK; - vm_page_protect(ma[index], VM_PROT_READ); } - vm_page_protect(p, VM_PROT_READ); - p->flags |= PG_BUSY; p->flags &= ~PG_CLEANCHK; ma[maxb] = p; for(i=0;i<maxf;i++) { int index = (maxb + i) + 1; ma[index] = maf[i]; - ma[index]->flags |= PG_BUSY; ma[index]->flags &= ~PG_CLEANCHK; - vm_page_protect(ma[index], VM_PROT_READ); } runlen = maxb + maxf + 1; splx(s); - vm_pageout_flush(ma, runlen, 0); + vm_pageout_flush(ma, runlen, pagerflags); if (object->generation != curgeneration) goto rescan; } - VOP_FSYNC(vp, NULL, syncio, curproc); + VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?1:0, curproc); object->flags &= ~OBJ_CLEANING; return; @@ -1314,7 +1312,7 @@ again: if (vm_page_sleep(p, "vmopar", &p->busy)) goto again; - if (clean_only) { + if (clean_only && p->valid) { vm_page_test_dirty(p); if (p->valid & p->dirty) continue; @@ -1342,9 +1340,9 @@ again: * interrupt -- minimize the spl transitions */ if (vm_page_sleep(p, "vmopar", &p->busy)) - goto again; + goto again; - if (clean_only) { + if (clean_only && p->valid) { vm_page_test_dirty(p); if (p->valid & p->dirty) { start += 1; diff --git a/sys/vm/vm_object.h b/sys/vm/vm_object.h index 76cb4d2..5d83269 100644 --- a/sys/vm/vm_object.h +++ b/sys/vm/vm_object.h @@ -61,7 +61,7 @@ * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_object.h,v 1.45 1998/02/05 03:32:45 dyson Exp $ + * $Id: vm_object.h,v 1.46 1998/02/25 03:55:52 dyson Exp $ */ /* @@ -142,6 +142,9 @@ struct vm_object { #ifdef KERNEL +#define OBJPC_SYNC 0x1 /* sync I/O */ +#define OBJPC_INVAL 0x2 /* invalidate */ + TAILQ_HEAD(object_q, vm_object); extern struct object_q vm_object_list; /* list of allocated objects */ diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c index 439fc3d..f896666 100644 --- a/sys/vm/vm_page.c +++ b/sys/vm/vm_page.c @@ -34,7 +34,7 @@ * SUCH DAMAGE. * * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91 - * $Id: vm_page.c,v 1.93 1998/02/09 06:11:32 eivind Exp $ + * $Id: vm_page.c,v 1.94 1998/03/01 04:18:24 dyson Exp $ */ /* @@ -88,6 +88,7 @@ static void vm_page_queue_init __P((void)); static vm_page_t vm_page_select_free __P((vm_object_t object, vm_pindex_t pindex, int prefqueue)); +static vm_page_t vm_page_select_cache __P((vm_object_t, vm_pindex_t)); /* * Associated with page of user-allocatable memory is a @@ -685,6 +686,36 @@ vm_page_select(object, pindex, basequeue) } /* + * Find a page on the cache queue with color optimization. As pages + * might be found, but not applicable, they are deactivated. This + * keeps us from using potentially busy cached pages. + */ +vm_page_t +vm_page_select_cache(object, pindex) + vm_object_t object; + vm_pindex_t pindex; +{ + vm_page_t m; + + while (TRUE) { +#if PQ_L2_SIZE > 1 + int index; + index = (pindex + object->pg_color) & PQ_L2_MASK; + m = vm_page_list_find(PQ_CACHE, index); + +#else + m = TAILQ_FIRST(vm_page_queues[PQ_CACHE].pl); +#endif + if (m && ((m->flags & PG_BUSY) || m->busy || + m->hold_count || m->wire_count)) { + vm_page_deactivate(m); + continue; + } + return m; + } +} + +/* * Find a free or zero page, with specified preference. */ static vm_page_t @@ -825,7 +856,7 @@ vm_page_alloc(object, pindex, page_req) panic("vm_page_alloc(NORMAL): missing page on free queue\n"); #endif } else { - m = vm_page_select(object, pindex, PQ_CACHE); + m = vm_page_select_cache(object, pindex); if (m == NULL) { splx(s); #if defined(DIAGNOSTIC) @@ -847,7 +878,7 @@ vm_page_alloc(object, pindex, page_req) panic("vm_page_alloc(ZERO): missing page on free queue\n"); #endif } else { - m = vm_page_select(object, pindex, PQ_CACHE); + m = vm_page_select_cache(object, pindex); if (m == NULL) { splx(s); #if defined(DIAGNOSTIC) @@ -871,7 +902,7 @@ vm_page_alloc(object, pindex, page_req) panic("vm_page_alloc(SYSTEM): missing page on free queue\n"); #endif } else { - m = vm_page_select(object, pindex, PQ_CACHE); + m = vm_page_select_cache(object, pindex); if (m == NULL) { splx(s); #if defined(DIAGNOSTIC) @@ -986,18 +1017,18 @@ vm_wait() int vm_page_sleep(vm_page_t m, char *msg, char *busy) { vm_object_t object = m->object; - int generation = object->generation; + int slept = 0; if ((busy && *busy) || (m->flags & PG_BUSY)) { int s; s = splvm(); if ((busy && *busy) || (m->flags & PG_BUSY)) { m->flags |= PG_WANTED; - tsleep(m, PVM, msg, 800); + tsleep(m, PVM, msg, 0); + slept = 1; } splx(s); } - return ((generation != object->generation) || (busy && *busy) || - (m->flags & PG_BUSY)); + return slept; } /* @@ -1540,13 +1571,11 @@ again1: if (m->dirty) { if (m->object->type == OBJT_VNODE) { vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); - vm_object_page_clean(m->object, 0, 0, TRUE); + vm_object_page_clean(m->object, 0, 0, OBJPC_SYNC); VOP_UNLOCK(m->object->handle, 0, curproc); goto again1; } else if (m->object->type == OBJT_SWAP || m->object->type == OBJT_DEFAULT) { - m->flags |= PG_BUSY; - vm_page_protect(m, VM_PROT_NONE); vm_pageout_flush(&m, 1, 0); goto again1; } @@ -1570,13 +1599,11 @@ again1: if (m->dirty) { if (m->object->type == OBJT_VNODE) { vn_lock(m->object->handle, LK_EXCLUSIVE | LK_RETRY, curproc); - vm_object_page_clean(m->object, 0, 0, TRUE); + vm_object_page_clean(m->object, 0, 0, OBJPC_SYNC); VOP_UNLOCK(m->object->handle, 0, curproc); goto again1; } else if (m->object->type == OBJT_SWAP || m->object->type == OBJT_DEFAULT) { - m->flags |= PG_BUSY; - vm_page_protect(m, VM_PROT_NONE); vm_pageout_flush(&m, 1, 0); goto again1; } diff --git a/sys/vm/vm_page.h b/sys/vm/vm_page.h index b9241dc..c80bfdd 100644 --- a/sys/vm/vm_page.h +++ b/sys/vm/vm_page.h @@ -61,7 +61,7 @@ * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_page.h,v 1.36 1998/02/05 03:32:47 dyson Exp $ + * $Id: vm_page.h,v 1.37 1998/03/01 04:18:26 dyson Exp $ */ /* @@ -276,6 +276,16 @@ extern vm_offset_t last_phys_addr; /* physical address for last_page */ } \ } +#define PAGE_BWAKEUP(m) { \ + (m)->busy--; \ + if ((((m)->flags & (PG_WANTED | PG_BUSY)) == PG_WANTED) && \ + ((m)->busy == 0)) { \ + (m)->flags &= ~PG_WANTED; \ + wakeup((m)); \ + } \ +} + + #if PAGE_SIZE == 4096 #define VM_PAGE_BITS_ALL 0xff #endif @@ -350,11 +360,11 @@ vm_page_protect(vm_page_t mem, int prot) { if (prot == VM_PROT_NONE) { if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) { - pmap_page_protect(VM_PAGE_TO_PHYS(mem), prot); + pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_NONE); mem->flags &= ~(PG_WRITEABLE|PG_MAPPED); } } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) { - pmap_page_protect(VM_PAGE_TO_PHYS(mem), prot); + pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_READ); mem->flags &= ~PG_WRITEABLE; } } diff --git a/sys/vm/vm_pageout.c b/sys/vm/vm_pageout.c index b614c3d..3eff5ed 100644 --- a/sys/vm/vm_pageout.c +++ b/sys/vm/vm_pageout.c @@ -65,7 +65,7 @@ * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_pageout.c,v 1.116 1998/02/24 10:16:23 dyson Exp $ + * $Id: vm_pageout.c,v 1.117 1998/03/01 04:18:28 dyson Exp $ */ /* @@ -100,7 +100,7 @@ /* the kernel process "vm_pageout"*/ static void vm_pageout __P((void)); -static int vm_pageout_clean __P((vm_page_t, int)); +static int vm_pageout_clean __P((vm_page_t)); static int vm_pageout_scan __P((void)); static int vm_pageout_free_page_calc __P((vm_size_t count)); struct proc *pageproc; @@ -218,9 +218,8 @@ void pmap_collect(void); * move!) */ static int -vm_pageout_clean(m, sync) +vm_pageout_clean(m) vm_page_t m; - int sync; { register vm_object_t object; vm_page_t mc[2*vm_pageout_page_count]; @@ -234,22 +233,21 @@ vm_pageout_clean(m, sync) * If not OBJT_SWAP, additional memory may be needed to do the pageout. * Try to avoid the deadlock. */ - if ((sync != VM_PAGEOUT_FORCE) && - (object->type == OBJT_DEFAULT) && + if ((object->type == OBJT_DEFAULT) && ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)) return 0; /* * Don't mess with the page if it's busy. */ - if ((!sync && m->hold_count != 0) || + if ((m->hold_count != 0) || ((m->busy != 0) || (m->flags & PG_BUSY))) return 0; /* * Try collapsing before it's too late. */ - if (!sync && object->backing_object) { + if (object->backing_object) { vm_object_collapse(object); } @@ -295,8 +293,7 @@ vm_pageout_clean(m, sync) } vm_page_test_dirty(p); if ((p->dirty & p->valid) != 0 && - ((p->queue == PQ_INACTIVE) || - (sync == VM_PAGEOUT_FORCE)) && + (p->queue == PQ_INACTIVE) && (p->wire_count == 0) && (p->hold_count == 0)) { mc[vm_pageout_page_count + i] = p; @@ -330,8 +327,7 @@ do_backward: } vm_page_test_dirty(p); if ((p->dirty & p->valid) != 0 && - ((p->queue == PQ_INACTIVE) || - (sync == VM_PAGEOUT_FORCE)) && + (p->queue == PQ_INACTIVE) && (p->wire_count == 0) && (p->hold_count == 0)) { mc[vm_pageout_page_count - i] = p; @@ -351,30 +347,30 @@ do_backward: /* * we allow reads during pageouts... */ - for (i = page_base; i < (page_base + pageout_count); i++) { - mc[i]->busy++; - vm_page_protect(mc[i], VM_PROT_READ); - } - - return vm_pageout_flush(&mc[page_base], pageout_count, sync); + return vm_pageout_flush(&mc[page_base], pageout_count, 0); } int -vm_pageout_flush(mc, count, sync) +vm_pageout_flush(mc, count, flags) vm_page_t *mc; int count; - int sync; + int flags; { register vm_object_t object; int pageout_status[count]; int numpagedout = 0; int i; + for (i = 0; i < count; i++) { + mc[i]->busy++; + vm_page_protect(mc[i], VM_PROT_READ); + } + object = mc[0]->object; object->paging_in_progress += count; vm_pager_put_pages(object, mc, count, - ((sync || (object == kernel_object)) ? TRUE : FALSE), + (flags | ((object == kernel_object) ? OBJPC_SYNC : 0)), pageout_status); for (i = 0; i < count; i++) { @@ -417,9 +413,7 @@ vm_pageout_flush(mc, count, sync) */ if (pageout_status[i] != VM_PAGER_PEND) { vm_object_pip_wakeup(object); - mt->flags |= PG_BUSY; - mt->busy--; - PAGE_WAKEUP(mt); + PAGE_BWAKEUP(mt); } } return numpagedout; @@ -840,7 +834,7 @@ rescan0: * laundry. If it is still in the laundry, then we * start the cleaning operation. */ - written = vm_pageout_clean(m, 0); + written = vm_pageout_clean(m); if (vp) vput(vp); diff --git a/sys/vm/vm_pager.c b/sys/vm/vm_pager.c index bdb3dae..0f11408 100644 --- a/sys/vm/vm_pager.c +++ b/sys/vm/vm_pager.c @@ -61,7 +61,7 @@ * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * - * $Id: vm_pager.c,v 1.34 1998/02/06 12:14:29 eivind Exp $ + * $Id: vm_pager.c,v 1.35 1998/02/23 08:22:40 dyson Exp $ */ /* @@ -183,14 +183,14 @@ vm_pager_get_pages(object, m, count, reqpage) } int -vm_pager_put_pages(object, m, count, sync, rtvals) +vm_pager_put_pages(object, m, count, flags, rtvals) vm_object_t object; vm_page_t *m; int count; - boolean_t sync; + int flags; int *rtvals; { - return ((*pagertab[object->type]->pgo_putpages)(object, m, count, sync, rtvals)); + return ((*pagertab[object->type]->pgo_putpages)(object, m, count, flags, rtvals)); } boolean_t diff --git a/sys/vm/vm_pager.h b/sys/vm/vm_pager.h index ec21453..fb385fa 100644 --- a/sys/vm/vm_pager.h +++ b/sys/vm/vm_pager.h @@ -36,7 +36,7 @@ * SUCH DAMAGE. * * @(#)vm_pager.h 8.4 (Berkeley) 1/12/94 - * $Id: vm_pager.h,v 1.14 1997/10/12 20:26:32 phk Exp $ + * $Id: vm_pager.h,v 1.15 1998/02/03 22:19:35 bde Exp $ */ /* @@ -55,7 +55,7 @@ struct pagerops { vm_object_t (*pgo_alloc) __P((void *, vm_size_t, vm_prot_t, vm_ooffset_t)); /* Allocate pager. */ void (*pgo_dealloc) __P((vm_object_t)); /* Disassociate. */ int (*pgo_getpages) __P((vm_object_t, vm_page_t *, int, int)); /* Get (read) page. */ - int (*pgo_putpages) __P((vm_object_t, vm_page_t *, int, boolean_t, int *)); /* Put (write) page. */ + int (*pgo_putpages) __P((vm_object_t, vm_page_t *, int, int, int *)); /* Put (write) page. */ boolean_t (*pgo_haspage) __P((vm_object_t, vm_pindex_t, int *, int *)); /* Does pager have page? */ void (*pgo_sync) __P((void)); }; @@ -76,6 +76,9 @@ struct pagerops { #define VM_PAGER_ERROR 4 #define VM_PAGER_AGAIN 5 +#define VM_PAGER_PUT_SYNC 0x1 +#define VM_PAGER_PUT_INVAL 0x2 + #ifdef KERNEL #ifdef MALLOC_DECLARE diff --git a/sys/vm/vnode_pager.c b/sys/vm/vnode_pager.c index 33c69a3..bedffdc 100644 --- a/sys/vm/vnode_pager.c +++ b/sys/vm/vnode_pager.c @@ -38,7 +38,7 @@ * SUCH DAMAGE. * * from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91 - * $Id: vnode_pager.c,v 1.87 1998/02/26 06:39:58 msmith Exp $ + * $Id: vnode_pager.c,v 1.88 1998/03/01 04:18:31 dyson Exp $ */ /* @@ -557,7 +557,7 @@ vnode_pager_generic_getpages(vp, m, bytecount, reqpage) { vm_object_t object; vm_offset_t kva; - off_t foff; + off_t foff, tfoff, nextoff; int i, size, bsize, first, firstaddr; struct vnode *dp; int runpg; @@ -749,11 +749,22 @@ vnode_pager_generic_getpages(vp, m, bytecount, reqpage) */ relpbuf(bp); - for (i = 0; i < count; i++) { - pmap_clear_modify(VM_PAGE_TO_PHYS(m[i])); - m[i]->dirty = 0; - m[i]->valid = VM_PAGE_BITS_ALL; - m[i]->flags &= ~PG_ZERO; + for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) { + vm_page_t mt; + + nextoff = tfoff + PAGE_SIZE; + mt = m[i]; + + if (nextoff <= size) { + mt->valid = VM_PAGE_BITS_ALL; + mt->dirty = 0; + pmap_clear_modify(VM_PAGE_TO_PHYS(mt)); + } else { + int nvalid = ((size + DEV_BSIZE - 1) - tfoff) & ~(DEV_BSIZE - 1); + vm_page_set_validclean(mt, 0, nvalid); + } + + mt->flags &= ~PG_ZERO; if (i != reqpage) { /* @@ -769,13 +780,13 @@ vnode_pager_generic_getpages(vp, m, bytecount, reqpage) * now tell them that it is ok to use */ if (!error) { - if (m[i]->flags & PG_WANTED) - vm_page_activate(m[i]); + if (mt->flags & PG_WANTED) + vm_page_activate(mt); else - vm_page_deactivate(m[i]); - PAGE_WAKEUP(m[i]); + vm_page_deactivate(mt); + PAGE_WAKEUP(mt); } else { - vnode_pager_freepage(m[i]); + vnode_pager_freepage(mt); } } } @@ -814,11 +825,11 @@ vnode_pager_putpages(object, m, count, sync, rtvals) * own vnodes if they fail to implement VOP_GETPAGES. */ int -vnode_pager_generic_putpages(vp, m, bytecount, sync, rtvals) +vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals) struct vnode *vp; vm_page_t *m; int bytecount; - boolean_t sync; + int flags; int *rtvals; { int i; @@ -830,6 +841,7 @@ vnode_pager_generic_putpages(vp, m, bytecount, sync, rtvals) struct uio auio; struct iovec aiov; int error; + int ioflags; object = vp->v_object; count = bytecount / PAGE_SIZE; @@ -838,7 +850,8 @@ vnode_pager_generic_putpages(vp, m, bytecount, sync, rtvals) rtvals[i] = VM_PAGER_AGAIN; if ((int) m[0]->pindex < 0) { - printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n", m[0]->pindex, m[0]->dirty); + printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n", + m[0]->pindex, m[0]->dirty); rtvals[0] = VM_PAGER_BAD; return VM_PAGER_BAD; } @@ -857,21 +870,12 @@ vnode_pager_generic_putpages(vp, m, bytecount, sync, rtvals) for (i = ncount; i < count; i++) { rtvals[i] = VM_PAGER_BAD; } -#ifdef BOGUS - if (ncount == 0) { - printf("vnode_pager_putpages: write past end of file: %d, %lu\n", - poffset, - (unsigned long) object->un_pager.vnp.vnp_size); - return rtvals[0]; - } -#endif } } - for (i = 0; i < count; i++) { - m[i]->busy++; - m[i]->flags &= ~PG_BUSY; - } + ioflags = IO_VMIO; + ioflags |= (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ? IO_SYNC: 0; + ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; aiov.iov_base = (caddr_t) 0; aiov.iov_len = maxsize; @@ -882,7 +886,7 @@ vnode_pager_generic_putpages(vp, m, bytecount, sync, rtvals) auio.uio_rw = UIO_WRITE; auio.uio_resid = maxsize; auio.uio_procp = (struct proc *) 0; - error = VOP_WRITE(vp, &auio, IO_VMIO|(sync?IO_SYNC:0), curproc->p_ucred); + error = VOP_WRITE(vp, &auio, ioflags, curproc->p_ucred); cnt.v_vnodeout++; cnt.v_vnodepgsout += ncount; |