diff options
Diffstat (limited to 'sys/fs/tmpfs')
| -rw-r--r-- | sys/fs/tmpfs/tmpfs.h | 289 | ||||
| -rw-r--r-- | sys/fs/tmpfs/tmpfs_fifoops.c | 28 | ||||
| -rw-r--r-- | sys/fs/tmpfs/tmpfs_subr.c | 220 | ||||
| -rw-r--r-- | sys/fs/tmpfs/tmpfs_vfsops.c | 108 | ||||
| -rw-r--r-- | sys/fs/tmpfs/tmpfs_vnops.c | 257 | ||||
| -rw-r--r-- | sys/fs/tmpfs/tmpfs_vnops.h | 1 |
6 files changed, 570 insertions, 333 deletions
diff --git a/sys/fs/tmpfs/tmpfs.h b/sys/fs/tmpfs/tmpfs.h index b077489..37e5bbb 100644 --- a/sys/fs/tmpfs/tmpfs.h +++ b/sys/fs/tmpfs/tmpfs.h @@ -80,8 +80,10 @@ struct tmpfs_dirent { uint32_t td_hash; u_int td_namelen; - /* Pointer to the node this entry refers to. In case this field - * is NULL, the node is a whiteout. */ + /* + * Pointer to the node this entry refers to. In case this field + * is NULL, the node is a whiteout. + */ struct tmpfs_node * td_node; union { @@ -94,21 +96,24 @@ struct tmpfs_dirent { } ud; }; -/* A directory in tmpfs holds a list of directory entries, which in - * turn point to other files (which can be directories themselves). +/* + * A directory in tmpfs holds a collection of directory entries, which + * in turn point to other files (which can be directories themselves). * - * In tmpfs, this list is managed by a RB-Tree, whose head is defined by - * the struct tmpfs_dir type. + * In tmpfs, this collection is managed by a RB-Tree, whose head is + * defined by the struct tmpfs_dir type. * * It is important to notice that directories do not have entries for . and * .. as other file systems do. These can be generated when requested * based on information available by other means, such as the pointer to * the node itself in the former case or the pointer to the parent directory * in the latter case. This is done to simplify tmpfs's code and, more - * importantly, to remove redundancy. */ + * importantly, to remove redundancy. + */ RB_HEAD(tmpfs_dir, tmpfs_dirent); -/* Each entry in a directory has a cookie that identifies it. Cookies +/* + * Each entry in a directory has a cookie that identifies it. Cookies * supersede offsets within directories because, given how tmpfs stores * directories in memory, there is no such thing as an offset. * @@ -139,51 +144,67 @@ RB_HEAD(tmpfs_dir, tmpfs_dirent); * a particular type. The code must be careful to only access those * attributes that are actually allowed by the node's type. * - * * Below is the key of locks used to protected the fields in the following * structures. - * + * (v) vnode lock in exclusive mode + * (vi) vnode lock in exclusive mode, or vnode lock in shared vnode and + * tn_interlock + * (i) tn_interlock + * (m) tmpfs_mount tm_allnode_lock + * (c) stable after creation */ struct tmpfs_node { - /* Doubly-linked list entry which links all existing nodes for a - * single file system. This is provided to ease the removal of - * all nodes during the unmount operation. */ - LIST_ENTRY(tmpfs_node) tn_entries; + /* + * Doubly-linked list entry which links all existing nodes for + * a single file system. This is provided to ease the removal + * of all nodes during the unmount operation, and to support + * the implementation of VOP_VNTOCNP(). tn_attached is false + * when the node is removed from list and unlocked. + */ + LIST_ENTRY(tmpfs_node) tn_entries; /* (m) */ + bool tn_attached; /* (m) */ - /* The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', + /* + * The node's type. Any of 'VBLK', 'VCHR', 'VDIR', 'VFIFO', * 'VLNK', 'VREG' and 'VSOCK' is allowed. The usage of vnode * types instead of a custom enumeration is to make things simpler - * and faster, as we do not need to convert between two types. */ - enum vtype tn_type; + * and faster, as we do not need to convert between two types. + */ + enum vtype tn_type; /* (c) */ /* Node identifier. */ - ino_t tn_id; + ino_t tn_id; /* (c) */ - /* Node's internal status. This is used by several file system + /* + * Node's internal status. This is used by several file system * operations to do modifications to the node in a delayed - * fashion. */ - int tn_status; + * fashion. + */ + int tn_status; /* (vi) */ #define TMPFS_NODE_ACCESSED (1 << 1) #define TMPFS_NODE_MODIFIED (1 << 2) #define TMPFS_NODE_CHANGED (1 << 3) - /* The node size. It does not necessarily match the real amount - * of memory consumed by it. */ - off_t tn_size; + /* + * The node size. It does not necessarily match the real amount + * of memory consumed by it. + */ + off_t tn_size; /* (v) */ /* Generic node attributes. */ - uid_t tn_uid; - gid_t tn_gid; - mode_t tn_mode; - u_long tn_flags; - nlink_t tn_links; - struct timespec tn_atime; - struct timespec tn_mtime; - struct timespec tn_ctime; - struct timespec tn_birthtime; - unsigned long tn_gen; - - /* As there is a single vnode for each active file within the + uid_t tn_uid; /* (v) */ + gid_t tn_gid; /* (v) */ + mode_t tn_mode; /* (v) */ + u_long tn_flags; /* (v) */ + nlink_t tn_links; /* (v) */ + struct timespec tn_atime; /* (vi) */ + struct timespec tn_mtime; /* (vi) */ + struct timespec tn_ctime; /* (vi) */ + struct timespec tn_birthtime; /* (v) */ + unsigned long tn_gen; /* (c) */ + + /* + * As there is a single vnode for each active file within the * system, care has to be taken to avoid allocating more than one * vnode per file. In order to do this, a bidirectional association * is kept between vnodes and nodes. @@ -196,74 +217,84 @@ struct tmpfs_node { * tn_vnode. * * May be NULL when the node is unused (that is, no vnode has been - * allocated for it or it has been reclaimed). */ - struct vnode * tn_vnode; + * allocated for it or it has been reclaimed). + */ + struct vnode * tn_vnode; /* (i) */ - /* interlock to protect tn_vpstate */ + /* + * Interlock to protect tn_vpstate, and tn_status under shared + * vnode lock. + */ struct mtx tn_interlock; - /* Identify if current node has vnode assiocate with + /* + * Identify if current node has vnode assiocate with * or allocating vnode. */ - int tn_vpstate; + int tn_vpstate; /* (i) */ + + /* Transient refcounter on this node. */ + u_int tn_refcount; /* (m) + (i) */ /* misc data field for different tn_type node */ union { /* Valid when tn_type == VBLK || tn_type == VCHR. */ - dev_t tn_rdev; + dev_t tn_rdev; /* (c) */ /* Valid when tn_type == VDIR. */ struct tn_dir { - /* Pointer to the parent directory. The root + /* + * Pointer to the parent directory. The root * directory has a pointer to itself in this field; - * this property identifies the root node. */ + * this property identifies the root node. + */ struct tmpfs_node * tn_parent; - /* Head of a tree that links the contents of - * the directory together. */ + /* + * Head of a tree that links the contents of + * the directory together. + */ struct tmpfs_dir tn_dirhead; - /* Head of a list the contains fake directory entries + /* + * Head of a list the contains fake directory entries * heads, i.e. entries with TMPFS_DIRCOOKIE_DUPHEAD - * flag. */ + * flag. + */ struct tmpfs_dir_duphead tn_dupindex; - /* Number and pointer of the first directory entry + /* + * Number and pointer of the first directory entry * returned by the readdir operation if it were * called again to continue reading data from the * same directory as before. This is used to speed * up reads of long directories, assuming that no * more than one read is in progress at a given time. - * Otherwise, these values are discarded. */ + * Otherwise, these values are discarded. + */ off_t tn_readdir_lastn; struct tmpfs_dirent * tn_readdir_lastp; } tn_dir; /* Valid when tn_type == VLNK. */ /* The link's target, allocated from a string pool. */ - char * tn_link; + char * tn_link; /* (c) */ /* Valid when tn_type == VREG. */ struct tn_reg { - /* The contents of regular files stored in a tmpfs - * file system are represented by a single anonymous - * memory object (aobj, for short). The aobj provides - * direct access to any position within the file, - * because its contents are always mapped in a - * contiguous region of virtual memory. It is a task - * of the memory management subsystem (see uvm(9)) to - * issue the required page ins or page outs whenever - * a position within the file is accessed. */ - vm_object_t tn_aobj; - - }tn_reg; - - /* Valid when tn_type = VFIFO */ - struct tn_fifo { - fo_rdwr_t *tn_fo_read; - fo_rdwr_t *tn_fo_write; - }tn_fifo; - }tn_spec; + /* + * The contents of regular files stored in a + * tmpfs file system are represented by a + * single anonymous memory object (aobj, for + * short). The aobj provides direct access to + * any position within the file. It is a task + * of the memory management subsystem to issue + * the required page ins or page outs whenever + * a position within the file is accessed. + */ + vm_object_t tn_aobj; /* (c) */ + } tn_reg; + } tn_spec; /* (v) */ }; LIST_HEAD(tmpfs_node_list, tmpfs_node); @@ -281,21 +312,12 @@ LIST_HEAD(tmpfs_node_list, tmpfs_node); #ifdef INVARIANTS #define TMPFS_ASSERT_LOCKED(node) do { \ - MPASS(node != NULL); \ - MPASS(node->tn_vnode != NULL); \ - if (!VOP_ISLOCKED(node->tn_vnode) && \ - !mtx_owned(TMPFS_NODE_MTX(node))) \ - panic("tmpfs: node is not locked: %p", node); \ - } while (0) -#define TMPFS_ASSERT_ELOCKED(node) do { \ MPASS((node) != NULL); \ MPASS((node)->tn_vnode != NULL); \ - mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED); \ - ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs"); \ + ASSERT_VOP_LOCKED((node)->tn_vnode, "tmpfs assert"); \ } while (0) #else #define TMPFS_ASSERT_LOCKED(node) (void)0 -#define TMPFS_ASSERT_ELOCKED(node) (void)0 #endif #define TMPFS_VNODE_ALLOCATING 1 @@ -307,26 +329,32 @@ LIST_HEAD(tmpfs_node_list, tmpfs_node); * Internal representation of a tmpfs mount point. */ struct tmpfs_mount { - /* Maximum number of memory pages available for use by the file + /* + * Maximum number of memory pages available for use by the file * system, set during mount time. This variable must never be * used directly as it may be bigger than the current amount of - * free memory; in the extreme case, it will hold the SIZE_MAX - * value. */ - size_t tm_pages_max; + * free memory; in the extreme case, it will hold the ULONG_MAX + * value. + */ + u_long tm_pages_max; /* Number of pages in use by the file system. */ - size_t tm_pages_used; + u_long tm_pages_used; - /* Pointer to the node representing the root directory of this - * file system. */ + /* + * Pointer to the node representing the root directory of this + * file system. + */ struct tmpfs_node * tm_root; - /* Maximum number of possible nodes for this file system; set + /* + * Maximum number of possible nodes for this file system; set * during mount time. We need a hard limit on the maximum number * of nodes to avoid allocating too much of them; their objects * cannot be released until the file system is unmounted. * Otherwise, we could easily run out of memory by creating lots - * of empty files and then simply removing them. */ + * of empty files and then simply removing them. + */ ino_t tm_nodes_max; /* unrhdr used to allocate inode numbers */ @@ -335,38 +363,33 @@ struct tmpfs_mount { /* Number of nodes currently that are in use. */ ino_t tm_nodes_inuse; + /* Refcounter on this struct tmpfs_mount. */ + uint64_t tm_refcount; + /* maximum representable file size */ u_int64_t tm_maxfilesize; - /* Nodes are organized in two different lists. The used list - * contains all nodes that are currently used by the file system; - * i.e., they refer to existing files. The available list contains - * all nodes that are currently available for use by new files. - * Nodes must be kept in this list (instead of deleting them) - * because we need to keep track of their generation number (tn_gen - * field). - * - * Note that nodes are lazily allocated: if the available list is - * empty and we have enough space to create more nodes, they will be - * created and inserted in the used list. Once these are released, - * they will go into the available list, remaining alive until the - * file system is unmounted. */ + /* + * The used list contains all nodes that are currently used by + * the file system; i.e., they refer to existing files. + */ struct tmpfs_node_list tm_nodes_used; - /* All node lock to protect the node list and tmp_pages_used */ - struct mtx allnode_lock; + /* All node lock to protect the node list and tmp_pages_used. */ + struct mtx tm_allnode_lock; - /* Pools used to store file system meta data. These are not shared - * across several instances of tmpfs for the reasons described in - * tmpfs_pool.c. */ + /* Zones used to store file system meta data, per tmpfs mount. */ uma_zone_t tm_dirent_pool; uma_zone_t tm_node_pool; /* Read-only status. */ - int tm_ronly; + bool tm_ronly; + /* Do not use namecache. */ + bool tm_nonc; }; -#define TMPFS_LOCK(tm) mtx_lock(&(tm)->allnode_lock) -#define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->allnode_lock) +#define TMPFS_LOCK(tm) mtx_lock(&(tm)->tm_allnode_lock) +#define TMPFS_UNLOCK(tm) mtx_unlock(&(tm)->tm_allnode_lock) +#define TMPFS_MP_ASSERT_LOCKED(tm) mtx_assert(&(tm)->tm_allnode_lock, MA_OWNED) /* * This structure maps a file identifier to a tmpfs node. Used by the @@ -379,15 +402,24 @@ struct tmpfs_fid { unsigned long tf_gen; }; +struct tmpfs_dir_cursor { + struct tmpfs_dirent *tdc_current; + struct tmpfs_dirent *tdc_tree; +}; + #ifdef _KERNEL /* * Prototypes for tmpfs_subr.c. */ +void tmpfs_ref_node(struct tmpfs_node *node); +void tmpfs_ref_node_locked(struct tmpfs_node *node); int tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *, enum vtype, uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *, char *, dev_t, struct tmpfs_node **); void tmpfs_free_node(struct tmpfs_mount *, struct tmpfs_node *); +bool tmpfs_free_node_locked(struct tmpfs_mount *, struct tmpfs_node *, bool); +void tmpfs_free_tmp(struct tmpfs_mount *); int tmpfs_alloc_dirent(struct tmpfs_mount *, struct tmpfs_node *, const char *, u_int, struct tmpfs_dirent **); void tmpfs_free_dirent(struct tmpfs_mount *, struct tmpfs_dirent *); @@ -420,8 +452,13 @@ int tmpfs_chtimes(struct vnode *, struct vattr *, struct ucred *cred, void tmpfs_itimes(struct vnode *, const struct timespec *, const struct timespec *); +void tmpfs_set_status(struct tmpfs_node *node, int status); void tmpfs_update(struct vnode *); int tmpfs_truncate(struct vnode *, off_t); +struct tmpfs_dirent *tmpfs_dir_first(struct tmpfs_node *dnode, + struct tmpfs_dir_cursor *dc); +struct tmpfs_dirent *tmpfs_dir_next(struct tmpfs_node *dnode, + struct tmpfs_dir_cursor *dc); /* * Convenience macros to simplify some logical expressions. @@ -447,10 +484,6 @@ int tmpfs_truncate(struct vnode *, off_t); } while (0) /* - * Memory management stuff. - */ - -/* * Amount of memory pages to reserve for the system (e.g., to not use by * tmpfs). */ @@ -467,37 +500,41 @@ size_t tmpfs_pages_used(struct tmpfs_mount *tmp); * specific ones. */ -static inline -struct tmpfs_mount * +static inline struct tmpfs_mount * VFS_TO_TMPFS(struct mount *mp) { struct tmpfs_mount *tmp; - MPASS((mp) != NULL && (mp)->mnt_data != NULL); - tmp = (struct tmpfs_mount *)(mp)->mnt_data; - return tmp; + MPASS(mp != NULL && mp->mnt_data != NULL); + tmp = (struct tmpfs_mount *)mp->mnt_data; + return (tmp); } -static inline -struct tmpfs_node * +static inline struct tmpfs_node * VP_TO_TMPFS_NODE(struct vnode *vp) { struct tmpfs_node *node; - MPASS((vp) != NULL && (vp)->v_data != NULL); + MPASS(vp != NULL && vp->v_data != NULL); node = (struct tmpfs_node *)vp->v_data; - return node; + return (node); } -static inline -struct tmpfs_node * +static inline struct tmpfs_node * VP_TO_TMPFS_DIR(struct vnode *vp) { struct tmpfs_node *node; node = VP_TO_TMPFS_NODE(vp); TMPFS_VALIDATE_DIR(node); - return node; + return (node); +} + +static inline bool +tmpfs_use_nc(struct vnode *vp) +{ + + return (!(VFS_TO_TMPFS(vp->v_mount)->tm_nonc)); } #endif /* _FS_TMPFS_TMPFS_H_ */ diff --git a/sys/fs/tmpfs/tmpfs_fifoops.c b/sys/fs/tmpfs/tmpfs_fifoops.c index 89ebe85..f1743db 100644 --- a/sys/fs/tmpfs/tmpfs_fifoops.c +++ b/sys/fs/tmpfs/tmpfs_fifoops.c @@ -49,35 +49,14 @@ #include <fs/tmpfs/tmpfs_vnops.h> static int -tmpfs_fifo_kqfilter(struct vop_kqfilter_args *ap) -{ - struct vnode *vp; - struct tmpfs_node *node; - - vp = ap->a_vp; - node = VP_TO_TMPFS_NODE(vp); - - switch (ap->a_kn->kn_filter){ - case EVFILT_READ: - node->tn_status |= TMPFS_NODE_ACCESSED; - break; - case EVFILT_WRITE: - node->tn_status |= TMPFS_NODE_MODIFIED; - break; - } - - return fifo_specops.vop_kqfilter(ap); -} - -static int tmpfs_fifo_close(struct vop_close_args *v) { struct tmpfs_node *node; - node = VP_TO_TMPFS_NODE(v->a_vp); - node->tn_status |= TMPFS_NODE_ACCESSED; + node = VP_TO_TMPFS_NODE(v->a_vp); + tmpfs_set_status(node, TMPFS_NODE_ACCESSED); tmpfs_update(v->a_vp); - return fifo_specops.vop_close(v); + return (fifo_specops.vop_close(v)); } /* @@ -90,6 +69,5 @@ struct vop_vector tmpfs_fifoop_entries = { .vop_access = tmpfs_access, .vop_getattr = tmpfs_getattr, .vop_setattr = tmpfs_setattr, - .vop_kqfilter = tmpfs_fifo_kqfilter, }; diff --git a/sys/fs/tmpfs/tmpfs_subr.c b/sys/fs/tmpfs/tmpfs_subr.c index 9781b2c..816b3cc 100644 --- a/sys/fs/tmpfs/tmpfs_subr.c +++ b/sys/fs/tmpfs/tmpfs_subr.c @@ -61,11 +61,6 @@ __FBSDID("$FreeBSD$"); #include <fs/tmpfs/tmpfs_fifoops.h> #include <fs/tmpfs/tmpfs_vnops.h> -struct tmpfs_dir_cursor { - struct tmpfs_dirent *tdc_current; - struct tmpfs_dirent *tdc_tree; -}; - SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system"); static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED; @@ -129,13 +124,33 @@ tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages) if (tmpfs_mem_avail() < req_pages) return (0); - if (tmp->tm_pages_max != SIZE_MAX && + if (tmp->tm_pages_max != ULONG_MAX && tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp)) return (0); return (1); } +void +tmpfs_ref_node(struct tmpfs_node *node) +{ + + TMPFS_NODE_LOCK(node); + tmpfs_ref_node_locked(node); + TMPFS_NODE_UNLOCK(node); +} + +void +tmpfs_ref_node_locked(struct tmpfs_node *node) +{ + + TMPFS_NODE_ASSERT_LOCKED(node); + KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node)); + KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node, + node->tn_refcount)); + node->tn_refcount++; +} + /* * Allocates a new node of type 'type' inside the 'tmp' mount point, with * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', @@ -198,8 +213,8 @@ tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type, return (EBUSY); } - nnode = (struct tmpfs_node *)uma_zalloc_arg( - tmp->tm_node_pool, tmp, M_WAITOK); + nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp, + M_WAITOK); /* Generic initialization. */ nnode->tn_type = type; @@ -210,6 +225,7 @@ tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type, nnode->tn_gid = gid; nnode->tn_mode = mode; nnode->tn_id = alloc_unr(tmp->tm_ino_unr); + nnode->tn_refcount = 1; /* Type-specific initialization. */ switch (nnode->tn_type) { @@ -257,58 +273,65 @@ tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type, break; default: - panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); + panic("tmpfs_alloc_node: type %p %d", nnode, + (int)nnode->tn_type); } TMPFS_LOCK(tmp); LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); + nnode->tn_attached = true; tmp->tm_nodes_inuse++; + tmp->tm_refcount++; TMPFS_UNLOCK(tmp); *node = nnode; - return 0; + return (0); } /* * Destroys the node pointed to by node from the file system 'tmp'. - * If the node does not belong to the given mount point, the results are - * unpredicted. - * - * If the node references a directory; no entries are allowed because - * their removal could need a recursive algorithm, something forbidden in - * kernel space. Furthermore, there is not need to provide such - * functionality (recursive removal) because the only primitives offered - * to the user are the removal of empty directories and the deletion of - * individual files. - * - * Note that nodes are not really deleted; in fact, when a node has been - * allocated, it cannot be deleted during the whole life of the file - * system. Instead, they are moved to the available list and remain there - * until reused. + * If the node references a directory, no entries are allowed. */ void tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) { + + TMPFS_LOCK(tmp); + TMPFS_NODE_LOCK(node); + if (!tmpfs_free_node_locked(tmp, node, false)) { + TMPFS_NODE_UNLOCK(node); + TMPFS_UNLOCK(tmp); + } +} + +bool +tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node, + bool detach) +{ vm_object_t uobj; + TMPFS_MP_ASSERT_LOCKED(tmp); + TMPFS_NODE_ASSERT_LOCKED(node); + KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node)); + + node->tn_refcount--; + if (node->tn_attached && (detach || node->tn_refcount == 0)) { + MPASS(tmp->tm_nodes_inuse > 0); + tmp->tm_nodes_inuse--; + LIST_REMOVE(node, tn_entries); + node->tn_attached = false; + } + if (node->tn_refcount > 0) + return (false); + #ifdef INVARIANTS - TMPFS_NODE_LOCK(node); MPASS(node->tn_vnode == NULL); MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); - TMPFS_NODE_UNLOCK(node); #endif - - TMPFS_LOCK(tmp); - LIST_REMOVE(node, tn_entries); - tmp->tm_nodes_inuse--; + TMPFS_NODE_UNLOCK(node); TMPFS_UNLOCK(tmp); switch (node->tn_type) { - case VNON: - /* Do not do anything. VNON is provided to let the - * allocation routine clean itself easily by avoiding - * duplicating code in it. */ - /* FALLTHROUGH */ case VBLK: /* FALLTHROUGH */ case VCHR: @@ -327,9 +350,7 @@ tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) case VREG: uobj = node->tn_reg.tn_aobj; if (uobj != NULL) { - TMPFS_LOCK(tmp); - tmp->tm_pages_used -= uobj->size; - TMPFS_UNLOCK(tmp); + atomic_subtract_long(&tmp->tm_pages_used, uobj->size); KASSERT((uobj->flags & OBJ_TMPFS) == 0, ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj)); vm_object_deallocate(uobj); @@ -342,6 +363,9 @@ tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) free_unr(tmp->tm_ino_unr, node->tn_id); uma_zfree(tmp->tm_node_pool, node); + TMPFS_LOCK(tmp); + tmpfs_free_tmp(tmp); + return (true); } static __inline uint32_t @@ -487,13 +511,16 @@ tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, struct vnode **vpp) { struct vnode *vp; + struct tmpfs_mount *tm; vm_object_t object; int error; error = 0; -loop: + tm = VFS_TO_TMPFS(mp); TMPFS_NODE_LOCK(node); -loop1: + tmpfs_ref_node_locked(node); +loop: + TMPFS_NODE_ASSERT_LOCKED(node); if ((vp = node->tn_vnode) != NULL) { MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); VI_LOCK(vp); @@ -513,12 +540,14 @@ loop1: msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 0, "tmpfsE", 0); } - goto loop1; + goto loop; } TMPFS_NODE_UNLOCK(node); error = vget(vp, lkflag | LK_INTERLOCK, curthread); - if (error == ENOENT) + if (error == ENOENT) { + TMPFS_NODE_LOCK(node); goto loop; + } if (error != 0) { vp = NULL; goto out; @@ -530,6 +559,7 @@ loop1: */ if (node->tn_vnode == NULL || node->tn_vnode != vp) { vput(vp); + TMPFS_NODE_LOCK(node); goto loop; } @@ -551,11 +581,9 @@ loop1: if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { node->tn_vpstate |= TMPFS_VNODE_WANT; error = msleep((caddr_t) &node->tn_vpstate, - TMPFS_NODE_MTX(node), PDROP | PCATCH, - "tmpfs_alloc_vp", 0); - if (error) - return error; - + TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0); + if (error != 0) + goto out; goto loop; } else node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; @@ -563,7 +591,8 @@ loop1: TMPFS_NODE_UNLOCK(node); /* Get a new vnode and associate it with our node. */ - error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); + error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ? + &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp); if (error != 0) goto unlock; MPASS(vp != NULL); @@ -611,7 +640,7 @@ loop1: VN_LOCK_ASHARE(vp); error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); - if (error) + if (error != 0) vp = NULL; unlock: @@ -629,18 +658,19 @@ unlock: TMPFS_NODE_UNLOCK(node); out: - *vpp = vp; + if (error == 0) { + *vpp = vp; #ifdef INVARIANTS - if (error == 0) { MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); TMPFS_NODE_LOCK(node); MPASS(*vpp == node->tn_vnode); TMPFS_NODE_UNLOCK(node); - } #endif + } + tmpfs_free_node(tm, node); - return error; + return (error); } /* @@ -683,7 +713,7 @@ tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, struct tmpfs_node *node; struct tmpfs_node *parent; - MPASS(VOP_ISLOCKED(dvp)); + ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file"); MPASS(cnp->cn_flags & HASBUF); tmp = VFS_TO_TMPFS(dvp->v_mount); @@ -708,8 +738,8 @@ tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, /* Allocate a node that represents the new file. */ error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type, - cnp->cn_cred->cr_uid, - dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node); + cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent, + target, vap->va_rdev, &node); if (error != 0) return (error); @@ -738,7 +768,7 @@ tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, return (0); } -static struct tmpfs_dirent * +struct tmpfs_dirent * tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) { struct tmpfs_dirent *de; @@ -752,7 +782,7 @@ tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) return (dc->tdc_current); } -static struct tmpfs_dirent * +struct tmpfs_dirent * tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) { struct tmpfs_dirent *de; @@ -1092,9 +1122,9 @@ tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) else error = uiomove(&dent, dent.d_reclen, uio); - node->tn_status |= TMPFS_NODE_ACCESSED; + tmpfs_set_status(node, TMPFS_NODE_ACCESSED); - return error; + return (error); } /* @@ -1117,9 +1147,8 @@ tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) * Return ENOENT if the current node is already removed. */ TMPFS_ASSERT_LOCKED(node); - if (node->tn_dir.tn_parent == NULL) { + if (node->tn_dir.tn_parent == NULL) return (ENOENT); - } TMPFS_NODE_LOCK(node->tn_dir.tn_parent); dent.d_fileno = node->tn_dir.tn_parent->tn_id; @@ -1137,9 +1166,9 @@ tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) else error = uiomove(&dent, dent.d_reclen, uio); - node->tn_status |= TMPFS_NODE_ACCESSED; + tmpfs_set_status(node, TMPFS_NODE_ACCESSED); - return error; + return (error); } /* @@ -1282,7 +1311,7 @@ tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, node->tn_dir.tn_readdir_lastn = off; node->tn_dir.tn_readdir_lastp = de; - node->tn_status |= TMPFS_NODE_ACCESSED; + tmpfs_set_status(node, TMPFS_NODE_ACCESSED); return error; } @@ -1413,9 +1442,7 @@ retry: uobj->size = newpages; VM_OBJECT_WUNLOCK(uobj); - TMPFS_LOCK(tmp); - tmp->tm_pages_used += (newpages - oldpages); - TMPFS_UNLOCK(tmp); + atomic_add_long(&tmp->tm_pages_used, newpages - oldpages); node->tn_size = newsize; return (0); @@ -1458,7 +1485,7 @@ tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, int error; struct tmpfs_node *node; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chflags"); node = VP_TO_TMPFS_NODE(vp); @@ -1498,9 +1525,9 @@ tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, node->tn_flags = flags; node->tn_status |= TMPFS_NODE_CHANGED; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chflags2"); - return 0; + return (0); } /* @@ -1514,7 +1541,7 @@ tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) int error; struct tmpfs_node *node; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chmod"); node = VP_TO_TMPFS_NODE(vp); @@ -1554,9 +1581,9 @@ tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) node->tn_status |= TMPFS_NODE_CHANGED; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chmod2"); - return 0; + return (0); } /* @@ -1575,7 +1602,7 @@ tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, uid_t ouid; gid_t ogid; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chown"); node = VP_TO_TMPFS_NODE(vp); @@ -1625,9 +1652,9 @@ tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, node->tn_mode &= ~(S_ISUID | S_ISGID); } - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chown2"); - return 0; + return (0); } /* @@ -1642,7 +1669,7 @@ tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, int error; struct tmpfs_node *node; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chsize"); node = VP_TO_TMPFS_NODE(vp); @@ -1680,9 +1707,9 @@ tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents * for us, as will update tn_status; no need to do that here. */ - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chsize2"); - return error; + return (error); } /* @@ -1697,7 +1724,7 @@ tmpfs_chtimes(struct vnode *vp, struct vattr *vap, int error; struct tmpfs_node *node; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chtimes"); node = VP_TO_TMPFS_NODE(vp); @@ -1728,9 +1755,20 @@ tmpfs_chtimes(struct vnode *vp, struct vattr *vap, if (vap->va_birthtime.tv_nsec != VNOVAL && vap->va_birthtime.tv_nsec != VNOVAL) node->tn_birthtime = vap->va_birthtime; - MPASS(VOP_ISLOCKED(vp)); + ASSERT_VOP_ELOCKED(vp, "chtimes2"); - return 0; + return (0); +} + +void +tmpfs_set_status(struct tmpfs_node *node, int status) +{ + + if ((node->tn_status & status) == status) + return; + TMPFS_NODE_LOCK(node); + node->tn_status |= status; + TMPFS_NODE_UNLOCK(node); } /* Sync timestamps */ @@ -1741,6 +1779,7 @@ tmpfs_itimes(struct vnode *vp, const struct timespec *acc, struct tmpfs_node *node; struct timespec now; + ASSERT_VOP_LOCKED(vp, "tmpfs_itimes"); node = VP_TO_TMPFS_NODE(vp); if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | @@ -1748,6 +1787,7 @@ tmpfs_itimes(struct vnode *vp, const struct timespec *acc, return; vfs_timestamp(&now); + TMPFS_NODE_LOCK(node); if (node->tn_status & TMPFS_NODE_ACCESSED) { if (acc == NULL) acc = &now; @@ -1758,11 +1798,12 @@ tmpfs_itimes(struct vnode *vp, const struct timespec *acc, mod = &now; node->tn_mtime = *mod; } - if (node->tn_status & TMPFS_NODE_CHANGED) { + if (node->tn_status & TMPFS_NODE_CHANGED) node->tn_ctime = now; - } - node->tn_status &= - ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); + node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | + TMPFS_NODE_CHANGED); + TMPFS_NODE_UNLOCK(node); + } void @@ -1794,14 +1835,13 @@ tmpfs_truncate(struct vnode *vp, off_t length) return (EFBIG); error = tmpfs_reg_resize(vp, length, FALSE); - if (error == 0) { + if (error == 0) node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; - } out: tmpfs_update(vp); - return error; + return (error); } static __inline int diff --git a/sys/fs/tmpfs/tmpfs_vfsops.c b/sys/fs/tmpfs/tmpfs_vfsops.c index 7af143b..65a2a12 100644 --- a/sys/fs/tmpfs/tmpfs_vfsops.c +++ b/sys/fs/tmpfs/tmpfs_vfsops.c @@ -78,7 +78,7 @@ static int tmpfs_statfs(struct mount *, struct statfs *); static const char *tmpfs_opts[] = { "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export", - "union", NULL + "union", "nonc", NULL }; static const char *tmpfs_updateopts[] = { @@ -137,6 +137,7 @@ tmpfs_mount(struct mount *mp) struct tmpfs_node *root; struct thread *td = curthread; int error; + bool nonc; /* Size counters. */ u_quad_t pages; off_t nodes_max, size_max, maxfilesize; @@ -185,11 +186,12 @@ tmpfs_mount(struct mount *mp) size_max = 0; if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0) maxfilesize = 0; + nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0; /* Do not allow mounts if we do not have enough memory to preserve * the minimum reserved pages. */ if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED) - return ENOSPC; + return (ENOSPC); /* Get the maximum number of memory pages this file system is * allowed to use, based on the maximum size the user passed in @@ -218,37 +220,35 @@ tmpfs_mount(struct mount *mp) tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount), M_TMPFSMNT, M_WAITOK | M_ZERO); - mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF); + mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF); tmp->tm_nodes_max = nodes_max; tmp->tm_nodes_inuse = 0; + tmp->tm_refcount = 1; tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX; LIST_INIT(&tmp->tm_nodes_used); tmp->tm_pages_max = pages; tmp->tm_pages_used = 0; - tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock); + tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->tm_allnode_lock); tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent", - sizeof(struct tmpfs_dirent), - NULL, NULL, NULL, NULL, + sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); tmp->tm_node_pool = uma_zcreate("TMPFS node", - sizeof(struct tmpfs_node), - tmpfs_node_ctor, tmpfs_node_dtor, - tmpfs_node_init, tmpfs_node_fini, - UMA_ALIGN_PTR, 0); + sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor, + tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0); tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0; + tmp->tm_nonc = nonc; /* Allocate the root node. */ - error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, - root_gid, root_mode & ALLPERMS, NULL, NULL, - VNOVAL, &root); + error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid, + root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root); if (error != 0 || root == NULL) { - uma_zdestroy(tmp->tm_node_pool); - uma_zdestroy(tmp->tm_dirent_pool); - delete_unrhdr(tmp->tm_ino_unr); - free(tmp, M_TMPFSMNT); - return error; + uma_zdestroy(tmp->tm_node_pool); + uma_zdestroy(tmp->tm_dirent_pool); + delete_unrhdr(tmp->tm_ino_unr); + free(tmp, M_TMPFSMNT); + return (error); } KASSERT(root->tn_id == 2, ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id)); @@ -306,25 +306,17 @@ tmpfs_unmount(struct mount *mp, int mntflags) TMPFS_LOCK(tmp); while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) { - TMPFS_UNLOCK(tmp); + TMPFS_NODE_LOCK(node); if (node->tn_type == VDIR) tmpfs_dir_destroy(tmp, node); - tmpfs_free_node(tmp, node); - TMPFS_LOCK(tmp); + if (tmpfs_free_node_locked(tmp, node, true)) + TMPFS_LOCK(tmp); + else + TMPFS_NODE_UNLOCK(node); } - TMPFS_UNLOCK(tmp); - - uma_zdestroy(tmp->tm_dirent_pool); - uma_zdestroy(tmp->tm_node_pool); - delete_unrhdr(tmp->tm_ino_unr); - mtx_destroy(&tmp->allnode_lock); - MPASS(tmp->tm_pages_used == 0); - MPASS(tmp->tm_nodes_inuse == 0); - - /* Throw away the tmpfs_mount structure. */ - free(mp->mnt_data, M_TMPFSMNT); mp->mnt_data = NULL; + tmpfs_free_tmp(tmp); vfs_write_resume(mp, VR_START_WRITE); MNT_ILOCK(mp); @@ -334,52 +326,74 @@ tmpfs_unmount(struct mount *mp, int mntflags) return (0); } +void +tmpfs_free_tmp(struct tmpfs_mount *tmp) +{ + + MPASS(tmp->tm_refcount > 0); + tmp->tm_refcount--; + if (tmp->tm_refcount > 0) { + TMPFS_UNLOCK(tmp); + return; + } + TMPFS_UNLOCK(tmp); + + uma_zdestroy(tmp->tm_dirent_pool); + uma_zdestroy(tmp->tm_node_pool); + delete_unrhdr(tmp->tm_ino_unr); + + mtx_destroy(&tmp->tm_allnode_lock); + MPASS(tmp->tm_pages_used == 0); + MPASS(tmp->tm_nodes_inuse == 0); + + free(tmp, M_TMPFSMNT); +} + static int tmpfs_root(struct mount *mp, int flags, struct vnode **vpp) { int error; - error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp); - if (!error) + error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp); + if (error == 0) (*vpp)->v_vflag |= VV_ROOT; - - return error; + return (error); } static int tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp) { - boolean_t found; struct tmpfs_fid *tfhp; struct tmpfs_mount *tmp; struct tmpfs_node *node; + int error; tmp = VFS_TO_TMPFS(mp); tfhp = (struct tmpfs_fid *)fhp; if (tfhp->tf_len != sizeof(struct tmpfs_fid)) - return EINVAL; + return (EINVAL); if (tfhp->tf_id >= tmp->tm_nodes_max) - return EINVAL; - - found = FALSE; + return (EINVAL); TMPFS_LOCK(tmp); LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { if (node->tn_id == tfhp->tf_id && node->tn_gen == tfhp->tf_gen) { - found = TRUE; + tmpfs_ref_node(node); break; } } TMPFS_UNLOCK(tmp); - if (found) - return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp)); - - return (EINVAL); + if (node != NULL) { + error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp); + tmpfs_free_node(tmp, node); + } else + error = EINVAL; + return (error); } /* ARGSUSED2 */ @@ -395,7 +409,7 @@ tmpfs_statfs(struct mount *mp, struct statfs *sbp) sbp->f_bsize = PAGE_SIZE; used = tmpfs_pages_used(tmp); - if (tmp->tm_pages_max != SIZE_MAX) + if (tmp->tm_pages_max != ULONG_MAX) sbp->f_blocks = tmp->tm_pages_max; else sbp->f_blocks = used + tmpfs_mem_avail(); diff --git a/sys/fs/tmpfs/tmpfs_vnops.c b/sys/fs/tmpfs/tmpfs_vnops.c index f01c8be..6d6982f 100644 --- a/sys/fs/tmpfs/tmpfs_vnops.c +++ b/sys/fs/tmpfs/tmpfs_vnops.c @@ -76,13 +76,11 @@ tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags, } static int -tmpfs_lookup(struct vop_cachedlookup_args *v) +tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { - struct vnode *dvp = v->a_dvp; - struct vnode **vpp = v->a_vpp; - struct componentname *cnp = v->a_cnp; struct tmpfs_dirent *de; - struct tmpfs_node *dnode; + struct tmpfs_node *dnode, *pnode; + struct tmpfs_mount *tm; int error; dnode = VP_TO_TMPFS_DIR(dvp); @@ -104,8 +102,12 @@ tmpfs_lookup(struct vop_cachedlookup_args *v) goto out; } if (cnp->cn_flags & ISDOTDOT) { + tm = VFS_TO_TMPFS(dvp->v_mount); + pnode = dnode->tn_dir.tn_parent; + tmpfs_ref_node(pnode); error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc, - dnode->tn_dir.tn_parent, cnp->cn_lkflags, vpp); + pnode, cnp->cn_lkflags, vpp); + tmpfs_free_node(tm, pnode); if (error != 0) goto out; } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { @@ -117,10 +119,12 @@ tmpfs_lookup(struct vop_cachedlookup_args *v) if (de != NULL && de->td_node == NULL) cnp->cn_flags |= ISWHITEOUT; if (de == NULL || de->td_node == NULL) { - /* The entry was not found in the directory. + /* + * The entry was not found in the directory. * This is OK if we are creating or renaming an * entry and are working on the last component of - * the path name. */ + * the path name. + */ if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == CREATE || \ cnp->cn_nameiop == RENAME || @@ -132,8 +136,10 @@ tmpfs_lookup(struct vop_cachedlookup_args *v) if (error != 0) goto out; - /* Keep the component name in the buffer for - * future uses. */ + /* + * Keep the component name in the buffer for + * future uses. + */ cnp->cn_flags |= SAVENAME; error = EJUSTRETURN; @@ -142,14 +148,18 @@ tmpfs_lookup(struct vop_cachedlookup_args *v) } else { struct tmpfs_node *tnode; - /* The entry was found, so get its associated - * tmpfs_node. */ + /* + * The entry was found, so get its associated + * tmpfs_node. + */ tnode = de->td_node; - /* If we are not at the last path component and + /* + * If we are not at the last path component and * found a non-directory or non-link entry (which * may itself be pointing to a directory), raise - * an error. */ + * an error. + */ if ((tnode->tn_type != VDIR && tnode->tn_type != VLNK) && !(cnp->cn_flags & ISLASTCN)) { @@ -157,9 +167,11 @@ tmpfs_lookup(struct vop_cachedlookup_args *v) goto out; } - /* If we are deleting or renaming the entry, keep + /* + * If we are deleting or renaming the entry, keep * track of its tmpfs_dirent so that it can be - * easily deleted later. */ + * easily deleted later. + */ if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { @@ -175,8 +187,9 @@ tmpfs_lookup(struct vop_cachedlookup_args *v) goto out; if ((dnode->tn_mode & S_ISTXT) && - VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) && - VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) { + VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, + cnp->cn_thread) && VOP_ACCESS(*vpp, VADMIN, + cnp->cn_cred, cnp->cn_thread)) { error = EPERM; vput(*vpp); *vpp = NULL; @@ -192,18 +205,36 @@ tmpfs_lookup(struct vop_cachedlookup_args *v) } } - /* Store the result of this lookup in the cache. Avoid this if the + /* + * Store the result of this lookup in the cache. Avoid this if the * request was for creation, as it does not improve timings on - * emprical tests. */ - if ((cnp->cn_flags & MAKEENTRY) != 0) + * emprical tests. + */ + if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) cache_enter(dvp, *vpp, cnp); out: - /* If there were no errors, *vpp cannot be null and it must be - * locked. */ + /* + * If there were no errors, *vpp cannot be null and it must be + * locked. + */ MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp))); - return error; + return (error); +} + +static int +tmpfs_cached_lookup(struct vop_cachedlookup_args *v) +{ + + return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp)); +} + +static int +tmpfs_lookup(struct vop_lookup_args *v) +{ + + return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp)); } static int @@ -218,7 +249,7 @@ tmpfs_create(struct vop_create_args *v) MPASS(vap->va_type == VREG || vap->va_type == VSOCK); error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); - if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0) + if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) cache_enter(dvp, *vpp, cnp); return (error); } @@ -441,7 +472,7 @@ tmpfs_read(struct vop_read_args *v) if (uio->uio_offset < 0) return (EINVAL); node = VP_TO_TMPFS_NODE(vp); - node->tn_status |= TMPFS_NODE_ACCESSED; + tmpfs_set_status(node, TMPFS_NODE_ACCESSED); return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio)); } @@ -989,10 +1020,12 @@ tmpfs_rename(struct vop_rename_args *v) tmpfs_dir_attach(tdvp, de); - cache_purge(fvp); - if (tvp != NULL) - cache_purge(tvp); - cache_purge_negative(tdvp); + if (tmpfs_use_nc(fvp)) { + cache_purge(fvp); + if (tvp != NULL) + cache_purge(tvp); + cache_purge_negative(tdvp); + } error = 0; @@ -1078,8 +1111,8 @@ tmpfs_rmdir(struct vop_rmdir_args *v) v->a_cnp->cn_namelen)); /* Check flags to see if we are allowed to remove the directory. */ - if (dnode->tn_flags & APPEND - || node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) { + if ((dnode->tn_flags & APPEND) != 0 || + (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) != 0) { error = EPERM; goto out; } @@ -1092,23 +1125,23 @@ tmpfs_rmdir(struct vop_rmdir_args *v) /* No vnode should be allocated for this entry from this point */ TMPFS_NODE_LOCK(node); - TMPFS_ASSERT_ELOCKED(node); node->tn_links--; node->tn_dir.tn_parent = NULL; - node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ + node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; TMPFS_NODE_UNLOCK(node); TMPFS_NODE_LOCK(dnode); - TMPFS_ASSERT_ELOCKED(dnode); dnode->tn_links--; - dnode->tn_status |= TMPFS_NODE_ACCESSED | \ - TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; + dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | + TMPFS_NODE_MODIFIED; TMPFS_NODE_UNLOCK(dnode); - cache_purge(dvp); - cache_purge(vp); + if (tmpfs_use_nc(dvp)) { + cache_purge(dvp); + cache_purge(vp); + } /* Free the directory entry we just deleted. Note that the node * referred by it will not be removed until the vnode is really @@ -1216,9 +1249,9 @@ tmpfs_readlink(struct vop_readlink_args *v) error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid), uio); - node->tn_status |= TMPFS_NODE_ACCESSED; + tmpfs_set_status(node, TMPFS_NODE_ACCESSED); - return error; + return (error); } static int @@ -1252,10 +1285,10 @@ tmpfs_reclaim(struct vop_reclaim_args *v) else vnode_destroy_vobject(vp); vp->v_object = NULL; - cache_purge(vp); + if (tmpfs_use_nc(vp)) + cache_purge(vp); TMPFS_NODE_LOCK(node); - TMPFS_ASSERT_ELOCKED(node); tmpfs_free_vp(vp); /* If the node referenced by this vnode was deleted by the user, @@ -1385,13 +1418,139 @@ tmpfs_whiteout(struct vop_whiteout_args *ap) } } +static int +tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp, + struct tmpfs_dirent **pde) +{ + struct tmpfs_dir_cursor dc; + struct tmpfs_dirent *de; + + for (de = tmpfs_dir_first(tnp, &dc); de != NULL; + de = tmpfs_dir_next(tnp, &dc)) { + if (de->td_node == tn) { + *pde = de; + return (0); + } + } + return (ENOENT); +} + +static int +tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn, + struct tmpfs_node *tnp, char *buf, int *buflen, struct vnode **dvp) +{ + struct tmpfs_dirent *de; + int error, i; + + error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED, + dvp); + if (error != 0) + return (error); + error = tmpfs_vptocnp_dir(tn, tnp, &de); + if (error == 0) { + i = *buflen; + i -= de->td_namelen; + if (i < 0) { + error = ENOMEM; + } else { + bcopy(de->ud.td_name, buf + i, de->td_namelen); + *buflen = i; + } + } + if (error == 0) { + if (vp != *dvp) + VOP_UNLOCK(*dvp, 0); + } else { + if (vp != *dvp) + vput(*dvp); + else + vrele(vp); + } + return (error); +} + +static int +tmpfs_vptocnp(struct vop_vptocnp_args *ap) +{ + struct vnode *vp, **dvp; + struct tmpfs_node *tn, *tnp, *tnp1; + struct tmpfs_dirent *de; + struct tmpfs_mount *tm; + char *buf; + int *buflen; + int error; + + vp = ap->a_vp; + dvp = ap->a_vpp; + buf = ap->a_buf; + buflen = ap->a_buflen; + + tm = VFS_TO_TMPFS(vp->v_mount); + tn = VP_TO_TMPFS_NODE(vp); + if (tn->tn_type == VDIR) { + tnp = tn->tn_dir.tn_parent; + if (tnp == NULL) + return (ENOENT); + tmpfs_ref_node(tnp); + error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf, + buflen, dvp); + tmpfs_free_node(tm, tnp); + return (error); + } +restart: + TMPFS_LOCK(tm); + LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) { + if (tnp->tn_type != VDIR) + continue; + TMPFS_NODE_LOCK(tnp); + tmpfs_ref_node_locked(tnp); + + /* + * tn_vnode cannot be instantiated while we hold the + * node lock, so the directory cannot be changed while + * we iterate over it. Do this to avoid instantiating + * vnode for directories which cannot point to our + * node. + */ + error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp, + &de) : 0; + + if (error == 0) { + TMPFS_NODE_UNLOCK(tnp); + TMPFS_UNLOCK(tm); + error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen, + dvp); + if (error == 0) { + tmpfs_free_node(tm, tnp); + return (0); + } + if ((vp->v_iflag & VI_DOOMED) != 0) { + tmpfs_free_node(tm, tnp); + return (ENOENT); + } + TMPFS_LOCK(tm); + TMPFS_NODE_LOCK(tnp); + } + if (tmpfs_free_node_locked(tm, tnp, false)) { + goto restart; + } else { + KASSERT(tnp->tn_refcount > 0, + ("node %p refcount zero", tnp)); + tnp1 = LIST_NEXT(tnp, tn_entries); + TMPFS_NODE_UNLOCK(tnp); + } + } + TMPFS_UNLOCK(tm); + return (ENOENT); +} + /* - * vnode operations vector used for files stored in a tmpfs file system. + * Vnode operations vector used for files stored in a tmpfs file system. */ struct vop_vector tmpfs_vnodeop_entries = { .vop_default = &default_vnodeops, .vop_lookup = vfs_cache_lookup, - .vop_cachedlookup = tmpfs_lookup, + .vop_cachedlookup = tmpfs_cached_lookup, .vop_create = tmpfs_create, .vop_mknod = tmpfs_mknod, .vop_open = tmpfs_open, @@ -1417,5 +1576,13 @@ struct vop_vector tmpfs_vnodeop_entries = { .vop_vptofh = tmpfs_vptofh, .vop_whiteout = tmpfs_whiteout, .vop_bmap = VOP_EOPNOTSUPP, + .vop_vptocnp = tmpfs_vptocnp, }; +/* + * Same vector for mounts which do not use namecache. + */ +struct vop_vector tmpfs_vnodeop_nonc_entries = { + .vop_default = &tmpfs_vnodeop_entries, + .vop_lookup = tmpfs_lookup, +}; diff --git a/sys/fs/tmpfs/tmpfs_vnops.h b/sys/fs/tmpfs/tmpfs_vnops.h index 1e06c13..db614a05 100644 --- a/sys/fs/tmpfs/tmpfs_vnops.h +++ b/sys/fs/tmpfs/tmpfs_vnops.h @@ -44,6 +44,7 @@ */ extern struct vop_vector tmpfs_vnodeop_entries; +extern struct vop_vector tmpfs_vnodeop_nonc_entries; vop_access_t tmpfs_access; vop_getattr_t tmpfs_getattr; |
