4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
42 #include <asm/uaccess.h>
46 #include "delegation.h"
56 #define NFSDBG_FACILITY NFSDBG_VFS
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 static struct kmem_cache * nfs_inode_cachep;
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 return nfs_fileid_to_ino_t(fattr->fileid);
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
78 int nfs_wait_bit_killable(struct wait_bit_key *key)
80 if (fatal_signal_pending(current))
82 freezable_schedule_unsafe();
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
94 u64 nfs_compat_user_ino64(u64 fileid)
105 if (sizeof(ino) < sizeof(fileid))
106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
110 int nfs_drop_inode(struct inode *inode)
112 return NFS_STALE(inode) || generic_drop_inode(inode);
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
116 void nfs_clear_inode(struct inode *inode)
119 * The following should never happen...
121 WARN_ON_ONCE(nfs_have_writebacks(inode));
122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 nfs_zap_acl_cache(inode);
124 nfs_access_zap_cache(inode);
125 nfs_fscache_clear_inode(inode);
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
129 void nfs_evict_inode(struct inode *inode)
131 truncate_inode_pages_final(&inode->i_data);
133 nfs_clear_inode(inode);
137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
139 int nfs_sync_mapping(struct address_space *mapping)
143 if (mapping->nrpages != 0) {
144 unmap_mapping_range(mapping, 0, 0, 0);
145 ret = nfs_wb_all(mapping->host);
150 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
152 struct nfs_inode *nfsi = NFS_I(inode);
154 if (inode->i_mapping->nrpages == 0)
155 flags &= ~NFS_INO_INVALID_DATA;
156 nfsi->cache_validity |= flags;
157 if (flags & NFS_INO_INVALID_DATA)
158 nfs_fscache_invalidate(inode);
162 * Invalidate the local caches
164 static void nfs_zap_caches_locked(struct inode *inode)
166 struct nfs_inode *nfsi = NFS_I(inode);
167 int mode = inode->i_mode;
169 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
171 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
172 nfsi->attrtimeo_timestamp = jiffies;
174 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
175 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
176 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
177 | NFS_INO_INVALID_DATA
178 | NFS_INO_INVALID_ACCESS
179 | NFS_INO_INVALID_ACL
180 | NFS_INO_REVAL_PAGECACHE);
182 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
183 | NFS_INO_INVALID_ACCESS
184 | NFS_INO_INVALID_ACL
185 | NFS_INO_REVAL_PAGECACHE);
186 nfs_zap_label_cache_locked(nfsi);
189 void nfs_zap_caches(struct inode *inode)
191 spin_lock(&inode->i_lock);
192 nfs_zap_caches_locked(inode);
193 spin_unlock(&inode->i_lock);
195 EXPORT_SYMBOL_GPL(nfs_zap_caches);
197 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
199 if (mapping->nrpages != 0) {
200 spin_lock(&inode->i_lock);
201 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
202 spin_unlock(&inode->i_lock);
206 void nfs_zap_acl_cache(struct inode *inode)
208 void (*clear_acl_cache)(struct inode *);
210 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
211 if (clear_acl_cache != NULL)
212 clear_acl_cache(inode);
213 spin_lock(&inode->i_lock);
214 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
215 spin_unlock(&inode->i_lock);
217 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
219 void nfs_invalidate_atime(struct inode *inode)
221 spin_lock(&inode->i_lock);
222 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
223 spin_unlock(&inode->i_lock);
225 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
228 * Invalidate, but do not unhash, the inode.
229 * NB: must be called with inode->i_lock held!
231 static void nfs_invalidate_inode(struct inode *inode)
233 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
234 nfs_zap_caches_locked(inode);
237 struct nfs_find_desc {
239 struct nfs_fattr *fattr;
243 * In NFSv3 we can have 64bit inode numbers. In order to support
244 * this, and re-exported directories (also seen in NFSv2)
245 * we are forced to allow 2 different inodes to have the same
249 nfs_find_actor(struct inode *inode, void *opaque)
251 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
252 struct nfs_fh *fh = desc->fh;
253 struct nfs_fattr *fattr = desc->fattr;
255 if (NFS_FILEID(inode) != fattr->fileid)
257 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
259 if (nfs_compare_fh(NFS_FH(inode), fh))
261 if (is_bad_inode(inode) || NFS_STALE(inode))
267 nfs_init_locked(struct inode *inode, void *opaque)
269 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
270 struct nfs_fattr *fattr = desc->fattr;
272 set_nfs_fileid(inode, fattr->fileid);
273 nfs_copy_fh(NFS_FH(inode), desc->fh);
277 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
278 static void nfs_clear_label_invalid(struct inode *inode)
280 spin_lock(&inode->i_lock);
281 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
282 spin_unlock(&inode->i_lock);
285 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
286 struct nfs4_label *label)
293 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
294 error = security_inode_notifysecctx(inode, label->label,
297 printk(KERN_ERR "%s() %s %d "
298 "security_inode_notifysecctx() %d\n",
300 (char *)label->label,
302 nfs_clear_label_invalid(inode);
306 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
308 struct nfs4_label *label = NULL;
309 int minor_version = server->nfs_client->cl_minorversion;
311 if (minor_version < 2)
314 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
317 label = kzalloc(sizeof(struct nfs4_label), flags);
319 return ERR_PTR(-ENOMEM);
321 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
322 if (label->label == NULL) {
324 return ERR_PTR(-ENOMEM);
326 label->len = NFS4_MAXLABELLEN;
330 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
332 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
333 struct nfs4_label *label)
337 EXPORT_SYMBOL_GPL(nfs_setsecurity);
340 * This is our front-end to iget that looks up inodes by file handle
341 * instead of inode number.
344 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
346 struct nfs_find_desc desc = {
350 struct inode *inode = ERR_PTR(-ENOENT);
353 nfs_attr_check_mountpoint(sb, fattr);
355 if (nfs_attr_use_mounted_on_fileid(fattr))
356 fattr->fileid = fattr->mounted_on_fileid;
357 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
359 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
362 hash = nfs_fattr_to_ino_t(fattr);
364 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
366 inode = ERR_PTR(-ENOMEM);
370 if (inode->i_state & I_NEW) {
371 struct nfs_inode *nfsi = NFS_I(inode);
372 unsigned long now = jiffies;
374 /* We set i_ino for the few things that still rely on it,
378 /* We can't support update_atime(), since the server will reset it */
379 inode->i_flags |= S_NOATIME|S_NOCMTIME;
380 inode->i_mode = fattr->mode;
381 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
382 && nfs_server_capable(inode, NFS_CAP_MODE))
383 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
384 /* Why so? Because we want revalidate for devices/FIFOs, and
385 * that's precisely what we have in nfs_file_inode_operations.
387 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
388 if (S_ISREG(inode->i_mode)) {
389 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
390 inode->i_data.a_ops = &nfs_file_aops;
391 } else if (S_ISDIR(inode->i_mode)) {
392 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
393 inode->i_fop = &nfs_dir_operations;
394 inode->i_data.a_ops = &nfs_dir_aops;
395 /* Deal with crossing mountpoints */
396 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
397 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
398 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
399 inode->i_op = &nfs_referral_inode_operations;
401 inode->i_op = &nfs_mountpoint_inode_operations;
403 inode->i_flags |= S_AUTOMOUNT;
405 } else if (S_ISLNK(inode->i_mode))
406 inode->i_op = &nfs_symlink_inode_operations;
408 init_special_inode(inode, inode->i_mode, fattr->rdev);
410 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
411 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
412 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
413 inode->i_version = 0;
416 inode->i_uid = make_kuid(&init_user_ns, -2);
417 inode->i_gid = make_kgid(&init_user_ns, -2);
419 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
423 nfsi->read_cache_jiffies = fattr->time_start;
424 nfsi->attr_gencount = fattr->gencount;
425 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
426 inode->i_atime = fattr->atime;
427 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
428 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
429 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
430 inode->i_mtime = fattr->mtime;
431 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
432 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
433 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
434 inode->i_ctime = fattr->ctime;
435 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
436 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
437 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
438 inode->i_version = fattr->change_attr;
439 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
440 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
441 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
442 inode->i_size = nfs_size_to_loff_t(fattr->size);
444 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
445 | NFS_INO_REVAL_PAGECACHE);
446 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
447 set_nlink(inode, fattr->nlink);
448 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
449 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
450 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
451 inode->i_uid = fattr->uid;
452 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
453 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
454 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
455 inode->i_gid = fattr->gid;
456 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
457 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
458 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
459 inode->i_blocks = fattr->du.nfs2.blocks;
460 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
462 * report the blocks in 512byte units
464 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
467 nfs_setsecurity(inode, fattr, label);
469 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
470 nfsi->attrtimeo_timestamp = now;
471 nfsi->access_cache = RB_ROOT;
473 nfs_fscache_init_inode(inode);
475 unlock_new_inode(inode);
477 nfs_refresh_inode(inode, fattr);
478 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
480 (unsigned long long)NFS_FILEID(inode),
481 nfs_display_fhandle_hash(fh),
482 atomic_read(&inode->i_count));
488 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
491 EXPORT_SYMBOL_GPL(nfs_fhget);
493 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
496 nfs_setattr(struct dentry *dentry, struct iattr *attr)
498 struct inode *inode = dentry->d_inode;
499 struct nfs_fattr *fattr;
502 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
504 /* skip mode change if it's just for clearing setuid/setgid */
505 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
506 attr->ia_valid &= ~ATTR_MODE;
508 if (attr->ia_valid & ATTR_SIZE) {
511 BUG_ON(!S_ISREG(inode->i_mode));
513 i_size = i_size_read(inode);
514 if (attr->ia_size == i_size)
515 attr->ia_valid &= ~ATTR_SIZE;
516 else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
520 /* Optimization: if the end result is no change, don't RPC */
521 attr->ia_valid &= NFS_VALID_ATTRS;
522 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
525 trace_nfs_setattr_enter(inode);
527 /* Write all dirty data */
528 if (S_ISREG(inode->i_mode)) {
529 nfs_inode_dio_wait(inode);
533 fattr = nfs_alloc_fattr();
537 * Return any delegations if we're going to change ACLs
539 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
540 NFS_PROTO(inode)->return_delegation(inode);
541 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
543 error = nfs_refresh_inode(inode, fattr);
544 nfs_free_fattr(fattr);
546 trace_nfs_setattr_exit(inode, error);
549 EXPORT_SYMBOL_GPL(nfs_setattr);
552 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
553 * @inode: inode of the file used
554 * @offset: file offset to start truncating
556 * This is a copy of the common vmtruncate, but with the locking
557 * corrected to take into account the fact that NFS requires
558 * inode->i_size to be updated under the inode->i_lock.
559 * Note: must be called with inode->i_lock held!
561 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
565 err = inode_newsize_ok(inode, offset);
569 i_size_write(inode, offset);
572 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
574 spin_unlock(&inode->i_lock);
575 truncate_pagecache(inode, offset);
576 spin_lock(&inode->i_lock);
582 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
583 * @inode: pointer to struct inode
584 * @attr: pointer to struct iattr
586 * Note: we do this in the *proc.c in order to ensure that
587 * it works for things like exclusive creates too.
589 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
590 struct nfs_fattr *fattr)
592 /* Barrier: bump the attribute generation count. */
593 nfs_fattr_set_barrier(fattr);
595 spin_lock(&inode->i_lock);
596 NFS_I(inode)->attr_gencount = fattr->gencount;
597 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
598 if ((attr->ia_valid & ATTR_MODE) != 0) {
599 int mode = attr->ia_mode & S_IALLUGO;
600 mode |= inode->i_mode & ~S_IALLUGO;
601 inode->i_mode = mode;
603 if ((attr->ia_valid & ATTR_UID) != 0)
604 inode->i_uid = attr->ia_uid;
605 if ((attr->ia_valid & ATTR_GID) != 0)
606 inode->i_gid = attr->ia_gid;
607 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
608 | NFS_INO_INVALID_ACL);
610 if ((attr->ia_valid & ATTR_SIZE) != 0) {
611 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
612 nfs_vmtruncate(inode, attr->ia_size);
614 nfs_update_inode(inode, fattr);
615 spin_unlock(&inode->i_lock);
617 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
619 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
621 struct dentry *parent;
623 parent = dget_parent(dentry);
624 nfs_force_use_readdirplus(parent->d_inode);
628 static bool nfs_need_revalidate_inode(struct inode *inode)
630 if (NFS_I(inode)->cache_validity &
631 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
633 if (nfs_attribute_cache_expired(inode))
638 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
640 struct inode *inode = dentry->d_inode;
641 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
644 trace_nfs_getattr_enter(inode);
645 /* Flush out writes to the server in order to update c/mtime. */
646 if (S_ISREG(inode->i_mode)) {
647 nfs_inode_dio_wait(inode);
648 err = filemap_write_and_wait(inode->i_mapping);
654 * We may force a getattr if the user cares about atime.
656 * Note that we only have to check the vfsmount flags here:
657 * - NFS always sets S_NOATIME by so checking it would give a
659 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
660 * no point in checking those.
662 if ((mnt->mnt_flags & MNT_NOATIME) ||
663 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
666 if (need_atime || nfs_need_revalidate_inode(inode)) {
667 struct nfs_server *server = NFS_SERVER(inode);
669 if (server->caps & NFS_CAP_READDIRPLUS)
670 nfs_request_parent_use_readdirplus(dentry);
671 err = __nfs_revalidate_inode(server, inode);
674 generic_fillattr(inode, stat);
675 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
678 trace_nfs_getattr_exit(inode, err);
681 EXPORT_SYMBOL_GPL(nfs_getattr);
683 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
685 atomic_set(&l_ctx->count, 1);
686 l_ctx->lockowner.l_owner = current->files;
687 l_ctx->lockowner.l_pid = current->tgid;
688 INIT_LIST_HEAD(&l_ctx->list);
689 nfs_iocounter_init(&l_ctx->io_count);
692 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
694 struct nfs_lock_context *head = &ctx->lock_context;
695 struct nfs_lock_context *pos = head;
698 if (pos->lockowner.l_owner != current->files)
700 if (pos->lockowner.l_pid != current->tgid)
702 atomic_inc(&pos->count);
704 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
708 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
710 struct nfs_lock_context *res, *new = NULL;
711 struct inode *inode = ctx->dentry->d_inode;
713 spin_lock(&inode->i_lock);
714 res = __nfs_find_lock_context(ctx);
716 spin_unlock(&inode->i_lock);
717 new = kmalloc(sizeof(*new), GFP_KERNEL);
719 return ERR_PTR(-ENOMEM);
720 nfs_init_lock_context(new);
721 spin_lock(&inode->i_lock);
722 res = __nfs_find_lock_context(ctx);
724 list_add_tail(&new->list, &ctx->lock_context.list);
725 new->open_context = ctx;
730 spin_unlock(&inode->i_lock);
734 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
736 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
738 struct nfs_open_context *ctx = l_ctx->open_context;
739 struct inode *inode = ctx->dentry->d_inode;
741 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
743 list_del(&l_ctx->list);
744 spin_unlock(&inode->i_lock);
747 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
750 * nfs_close_context - Common close_context() routine NFSv2/v3
751 * @ctx: pointer to context
752 * @is_sync: is this a synchronous close
754 * always ensure that the attributes are up to date if we're mounted
755 * with close-to-open semantics
757 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
760 struct nfs_server *server;
762 if (!(ctx->mode & FMODE_WRITE))
766 inode = ctx->dentry->d_inode;
767 if (!list_empty(&NFS_I(inode)->open_files))
769 server = NFS_SERVER(inode);
770 if (server->flags & NFS_MOUNT_NOCTO)
772 nfs_revalidate_inode(server, inode);
774 EXPORT_SYMBOL_GPL(nfs_close_context);
776 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
778 struct nfs_open_context *ctx;
779 struct rpc_cred *cred = rpc_lookup_cred();
781 return ERR_CAST(cred);
783 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
786 return ERR_PTR(-ENOMEM);
788 nfs_sb_active(dentry->d_sb);
789 ctx->dentry = dget(dentry);
795 nfs_init_lock_context(&ctx->lock_context);
796 ctx->lock_context.open_context = ctx;
797 INIT_LIST_HEAD(&ctx->list);
798 ctx->mdsthreshold = NULL;
801 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
803 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
806 atomic_inc(&ctx->lock_context.count);
809 EXPORT_SYMBOL_GPL(get_nfs_open_context);
811 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
813 struct inode *inode = ctx->dentry->d_inode;
814 struct super_block *sb = ctx->dentry->d_sb;
816 if (!list_empty(&ctx->list)) {
817 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
819 list_del(&ctx->list);
820 spin_unlock(&inode->i_lock);
821 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
824 NFS_PROTO(inode)->close_context(ctx, is_sync);
825 if (ctx->cred != NULL)
826 put_rpccred(ctx->cred);
829 kfree(ctx->mdsthreshold);
833 void put_nfs_open_context(struct nfs_open_context *ctx)
835 __put_nfs_open_context(ctx, 0);
837 EXPORT_SYMBOL_GPL(put_nfs_open_context);
840 * Ensure that mmap has a recent RPC credential for use when writing out
843 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
845 struct inode *inode = ctx->dentry->d_inode;
846 struct nfs_inode *nfsi = NFS_I(inode);
848 spin_lock(&inode->i_lock);
849 list_add(&ctx->list, &nfsi->open_files);
850 spin_unlock(&inode->i_lock);
852 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
854 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
856 filp->private_data = get_nfs_open_context(ctx);
857 if (list_empty(&ctx->list))
858 nfs_inode_attach_open_context(ctx);
860 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
863 * Given an inode, search for an open context with the desired characteristics
865 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
867 struct nfs_inode *nfsi = NFS_I(inode);
868 struct nfs_open_context *pos, *ctx = NULL;
870 spin_lock(&inode->i_lock);
871 list_for_each_entry(pos, &nfsi->open_files, list) {
872 if (cred != NULL && pos->cred != cred)
874 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
876 ctx = get_nfs_open_context(pos);
879 spin_unlock(&inode->i_lock);
883 static void nfs_file_clear_open_context(struct file *filp)
885 struct nfs_open_context *ctx = nfs_file_open_context(filp);
888 struct inode *inode = ctx->dentry->d_inode;
890 filp->private_data = NULL;
891 spin_lock(&inode->i_lock);
892 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
893 spin_unlock(&inode->i_lock);
894 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
899 * These allocate and release file read/write context information.
901 int nfs_open(struct inode *inode, struct file *filp)
903 struct nfs_open_context *ctx;
905 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
908 nfs_file_set_open_context(filp, ctx);
909 put_nfs_open_context(ctx);
910 nfs_fscache_open_file(inode, filp);
914 int nfs_release(struct inode *inode, struct file *filp)
916 nfs_file_clear_open_context(filp);
921 * This function is called whenever some part of NFS notices that
922 * the cached attributes have to be refreshed.
925 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
927 int status = -ESTALE;
928 struct nfs4_label *label = NULL;
929 struct nfs_fattr *fattr = NULL;
930 struct nfs_inode *nfsi = NFS_I(inode);
932 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
933 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
935 trace_nfs_revalidate_inode_enter(inode);
937 if (is_bad_inode(inode))
939 if (NFS_STALE(inode))
943 fattr = nfs_alloc_fattr();
947 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
949 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
951 status = PTR_ERR(label);
955 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
957 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
959 (unsigned long long)NFS_FILEID(inode), status);
960 if (status == -ESTALE) {
961 nfs_zap_caches(inode);
962 if (!S_ISDIR(inode->i_mode))
963 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
968 status = nfs_refresh_inode(inode, fattr);
970 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
972 (unsigned long long)NFS_FILEID(inode), status);
976 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
977 nfs_zap_acl_cache(inode);
979 nfs_setsecurity(inode, fattr, label);
981 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
983 (unsigned long long)NFS_FILEID(inode));
986 nfs4_label_free(label);
988 nfs_free_fattr(fattr);
989 trace_nfs_revalidate_inode_exit(inode, status);
993 int nfs_attribute_timeout(struct inode *inode)
995 struct nfs_inode *nfsi = NFS_I(inode);
997 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1000 int nfs_attribute_cache_expired(struct inode *inode)
1002 if (nfs_have_delegated_attributes(inode))
1004 return nfs_attribute_timeout(inode);
1008 * nfs_revalidate_inode - Revalidate the inode attributes
1009 * @server - pointer to nfs_server struct
1010 * @inode - pointer to inode struct
1012 * Updates inode attribute information by retrieving the data from the server.
1014 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1016 if (!nfs_need_revalidate_inode(inode))
1017 return NFS_STALE(inode) ? -ESTALE : 0;
1018 return __nfs_revalidate_inode(server, inode);
1020 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1022 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1024 if (!(NFS_I(inode)->cache_validity &
1025 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1026 && !nfs_attribute_cache_expired(inode))
1027 return NFS_STALE(inode) ? -ESTALE : 0;
1031 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1033 struct nfs_inode *nfsi = NFS_I(inode);
1036 if (mapping->nrpages != 0) {
1037 if (S_ISREG(inode->i_mode)) {
1038 ret = nfs_sync_mapping(mapping);
1042 ret = invalidate_inode_pages2(mapping);
1046 if (S_ISDIR(inode->i_mode)) {
1047 spin_lock(&inode->i_lock);
1048 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1049 spin_unlock(&inode->i_lock);
1051 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1052 nfs_fscache_wait_on_invalidate(inode);
1054 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1056 (unsigned long long)NFS_FILEID(inode));
1060 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1062 if (nfs_have_delegated_attributes(inode))
1064 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1065 || nfs_attribute_timeout(inode)
1066 || NFS_STALE(inode);
1070 * nfs_revalidate_mapping - Revalidate the pagecache
1071 * @inode - pointer to host inode
1072 * @mapping - pointer to mapping
1074 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1076 struct nfs_inode *nfsi = NFS_I(inode);
1077 unsigned long *bitlock = &nfsi->flags;
1080 /* swapfiles are not supposed to be shared. */
1081 if (IS_SWAPFILE(inode))
1084 if (nfs_mapping_need_revalidate_inode(inode)) {
1085 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1091 * We must clear NFS_INO_INVALID_DATA first to ensure that
1092 * invalidations that come in while we're shooting down the mappings
1093 * are respected. But, that leaves a race window where one revalidator
1094 * can clear the flag, and then another checks it before the mapping
1095 * gets invalidated. Fix that by serializing access to this part of
1098 * At the same time, we need to allow other tasks to see whether we
1099 * might be in the middle of invalidating the pages, so we only set
1100 * the bit lock here if it looks like we're going to be doing that.
1103 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1104 nfs_wait_bit_killable, TASK_KILLABLE);
1107 spin_lock(&inode->i_lock);
1108 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1109 spin_unlock(&inode->i_lock);
1112 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1114 spin_unlock(&inode->i_lock);
1118 set_bit(NFS_INO_INVALIDATING, bitlock);
1120 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1121 spin_unlock(&inode->i_lock);
1122 trace_nfs_invalidate_mapping_enter(inode);
1123 ret = nfs_invalidate_mapping(inode, mapping);
1124 trace_nfs_invalidate_mapping_exit(inode, ret);
1126 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1127 smp_mb__after_atomic();
1128 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1133 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1135 struct nfs_inode *nfsi = NFS_I(inode);
1136 unsigned long ret = 0;
1138 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1139 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1140 && inode->i_version == fattr->pre_change_attr) {
1141 inode->i_version = fattr->change_attr;
1142 if (S_ISDIR(inode->i_mode))
1143 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1144 ret |= NFS_INO_INVALID_ATTR;
1146 /* If we have atomic WCC data, we may update some attributes */
1147 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1148 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1149 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1150 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1151 ret |= NFS_INO_INVALID_ATTR;
1154 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1155 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1156 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1157 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1158 if (S_ISDIR(inode->i_mode))
1159 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1160 ret |= NFS_INO_INVALID_ATTR;
1162 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1163 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1164 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1165 && nfsi->nrequests == 0) {
1166 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1167 ret |= NFS_INO_INVALID_ATTR;
1174 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1175 * @inode - pointer to inode
1176 * @fattr - updated attributes
1178 * Verifies the attribute cache. If we have just changed the attributes,
1179 * so that fattr carries weak cache consistency data, then it may
1180 * also update the ctime/mtime/change_attribute.
1182 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1184 struct nfs_inode *nfsi = NFS_I(inode);
1185 loff_t cur_size, new_isize;
1186 unsigned long invalid = 0;
1189 if (nfs_have_delegated_attributes(inode))
1191 /* Has the inode gone and changed behind our back? */
1192 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1194 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1197 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1198 inode->i_version != fattr->change_attr)
1199 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1201 /* Verify a few of the more important attributes */
1202 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1203 invalid |= NFS_INO_INVALID_ATTR;
1205 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1206 cur_size = i_size_read(inode);
1207 new_isize = nfs_size_to_loff_t(fattr->size);
1208 if (cur_size != new_isize && nfsi->nrequests == 0)
1209 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1212 /* Have any file permissions changed? */
1213 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1214 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1215 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1216 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1217 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1218 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1220 /* Has the link count changed? */
1221 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1222 invalid |= NFS_INO_INVALID_ATTR;
1224 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1225 invalid |= NFS_INO_INVALID_ATIME;
1228 nfs_set_cache_invalid(inode, invalid);
1230 nfsi->read_cache_jiffies = fattr->time_start;
1234 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1236 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1238 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1241 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1243 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1245 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1248 static atomic_long_t nfs_attr_generation_counter;
1250 static unsigned long nfs_read_attr_generation_counter(void)
1252 return atomic_long_read(&nfs_attr_generation_counter);
1255 unsigned long nfs_inc_attr_generation_counter(void)
1257 return atomic_long_inc_return(&nfs_attr_generation_counter);
1260 void nfs_fattr_init(struct nfs_fattr *fattr)
1263 fattr->time_start = jiffies;
1264 fattr->gencount = nfs_inc_attr_generation_counter();
1265 fattr->owner_name = NULL;
1266 fattr->group_name = NULL;
1268 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1271 * nfs_fattr_set_barrier
1272 * @fattr: attributes
1274 * Used to set a barrier after an attribute was updated. This
1275 * barrier ensures that older attributes from RPC calls that may
1276 * have raced with our update cannot clobber these new values.
1277 * Note that you are still responsible for ensuring that other
1278 * operations which change the attribute on the server do not
1281 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1283 fattr->gencount = nfs_inc_attr_generation_counter();
1286 struct nfs_fattr *nfs_alloc_fattr(void)
1288 struct nfs_fattr *fattr;
1290 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1292 nfs_fattr_init(fattr);
1295 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1297 struct nfs_fh *nfs_alloc_fhandle(void)
1301 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1306 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1310 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1311 * in the same way that wireshark does
1315 * For debugging only.
1317 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1319 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1320 * not on the result */
1321 return nfs_fhandle_hash(fh);
1323 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1326 * _nfs_display_fhandle - display an NFS file handle on the console
1328 * @fh: file handle to display
1329 * @caption: display caption
1331 * For debugging only.
1333 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1337 if (fh == NULL || fh->size == 0) {
1338 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1342 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1343 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1344 for (i = 0; i < fh->size; i += 16) {
1345 __be32 *pos = (__be32 *)&fh->data[i];
1347 switch ((fh->size - i - 1) >> 2) {
1349 printk(KERN_DEFAULT " %08x\n",
1353 printk(KERN_DEFAULT " %08x %08x\n",
1354 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1357 printk(KERN_DEFAULT " %08x %08x %08x\n",
1358 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1359 be32_to_cpup(pos + 2));
1362 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1363 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1364 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1368 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1372 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1373 * @inode - pointer to inode
1374 * @fattr - attributes
1376 * Attempt to divine whether or not an RPC call reply carrying stale
1377 * attributes got scheduled after another call carrying updated ones.
1379 * To do so, the function first assumes that a more recent ctime means
1380 * that the attributes in fattr are newer, however it also attempt to
1381 * catch the case where ctime either didn't change, or went backwards
1382 * (if someone reset the clock on the server) by looking at whether
1383 * or not this RPC call was started after the inode was last updated.
1384 * Note also the check for wraparound of 'attr_gencount'
1386 * The function returns 'true' if it thinks the attributes in 'fattr' are
1387 * more recent than the ones cached in the inode.
1390 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1392 const struct nfs_inode *nfsi = NFS_I(inode);
1394 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1395 nfs_ctime_need_update(inode, fattr) ||
1396 nfs_size_need_update(inode, fattr) ||
1397 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1401 * Don't trust the change_attribute, mtime, ctime or size if
1402 * a pnfs LAYOUTCOMMIT is outstanding
1404 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1405 struct nfs_fattr *fattr)
1407 if (pnfs_layoutcommit_outstanding(inode))
1408 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1409 NFS_ATTR_FATTR_MTIME |
1410 NFS_ATTR_FATTR_CTIME |
1411 NFS_ATTR_FATTR_SIZE);
1414 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1418 trace_nfs_refresh_inode_enter(inode);
1420 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1422 if (nfs_inode_attrs_need_update(inode, fattr))
1423 ret = nfs_update_inode(inode, fattr);
1425 ret = nfs_check_inode_attributes(inode, fattr);
1427 trace_nfs_refresh_inode_exit(inode, ret);
1432 * nfs_refresh_inode - try to update the inode attribute cache
1433 * @inode - pointer to inode
1434 * @fattr - updated attributes
1436 * Check that an RPC call that returned attributes has not overlapped with
1437 * other recent updates of the inode metadata, then decide whether it is
1438 * safe to do a full update of the inode attributes, or whether just to
1439 * call nfs_check_inode_attributes.
1441 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1445 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1447 spin_lock(&inode->i_lock);
1448 status = nfs_refresh_inode_locked(inode, fattr);
1449 spin_unlock(&inode->i_lock);
1453 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1455 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1457 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1459 if (S_ISDIR(inode->i_mode))
1460 invalid |= NFS_INO_INVALID_DATA;
1461 nfs_set_cache_invalid(inode, invalid);
1462 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1464 return nfs_refresh_inode_locked(inode, fattr);
1468 * nfs_post_op_update_inode - try to update the inode attribute cache
1469 * @inode - pointer to inode
1470 * @fattr - updated attributes
1472 * After an operation that has changed the inode metadata, mark the
1473 * attribute cache as being invalid, then try to update it.
1475 * NB: if the server didn't return any post op attributes, this
1476 * function will force the retrieval of attributes before the next
1477 * NFS request. Thus it should be used only for operations that
1478 * are expected to change one or more attributes, to avoid
1479 * unnecessary NFS requests and trips through nfs_update_inode().
1481 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1485 spin_lock(&inode->i_lock);
1486 status = nfs_post_op_update_inode_locked(inode, fattr);
1487 spin_unlock(&inode->i_lock);
1491 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1494 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1495 * @inode - pointer to inode
1496 * @fattr - updated attributes
1498 * After an operation that has changed the inode metadata, mark the
1499 * attribute cache as being invalid, then try to update it. Fake up
1500 * weak cache consistency data, if none exist.
1502 * This function is mainly designed to be used by the ->write_done() functions.
1504 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1508 spin_lock(&inode->i_lock);
1509 /* Don't do a WCC update if these attributes are already stale */
1510 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1511 !nfs_inode_attrs_need_update(inode, fattr)) {
1512 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1513 | NFS_ATTR_FATTR_PRESIZE
1514 | NFS_ATTR_FATTR_PREMTIME
1515 | NFS_ATTR_FATTR_PRECTIME);
1518 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1519 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1520 fattr->pre_change_attr = inode->i_version;
1521 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1523 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1524 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1525 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1526 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1528 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1529 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1530 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1531 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1533 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1534 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1535 fattr->pre_size = i_size_read(inode);
1536 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1539 status = nfs_post_op_update_inode_locked(inode, fattr);
1540 spin_unlock(&inode->i_lock);
1543 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1546 * Many nfs protocol calls return the new file attributes after
1547 * an operation. Here we update the inode to reflect the state
1548 * of the server's inode.
1550 * This is a bit tricky because we have to make sure all dirty pages
1551 * have been sent off to the server before calling invalidate_inode_pages.
1552 * To make sure no other process adds more write requests while we try
1553 * our best to flush them, we make them sleep during the attribute refresh.
1555 * A very similar scenario holds for the dir cache.
1557 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1559 struct nfs_server *server;
1560 struct nfs_inode *nfsi = NFS_I(inode);
1561 loff_t cur_isize, new_isize;
1562 unsigned long invalid = 0;
1563 unsigned long now = jiffies;
1564 unsigned long save_cache_validity;
1566 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1567 __func__, inode->i_sb->s_id, inode->i_ino,
1568 nfs_display_fhandle_hash(NFS_FH(inode)),
1569 atomic_read(&inode->i_count), fattr->valid);
1571 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1572 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1573 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1574 NFS_SERVER(inode)->nfs_client->cl_hostname,
1575 inode->i_sb->s_id, (long long)nfsi->fileid,
1576 (long long)fattr->fileid);
1581 * Make sure the inode's type hasn't changed.
1583 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1585 * Big trouble! The inode has become a different object.
1587 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1588 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1592 server = NFS_SERVER(inode);
1593 /* Update the fsid? */
1594 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1595 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1596 !IS_AUTOMOUNT(inode))
1597 server->fsid = fattr->fsid;
1600 * Update the read time so we don't revalidate too often.
1602 nfsi->read_cache_jiffies = fattr->time_start;
1604 save_cache_validity = nfsi->cache_validity;
1605 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1606 | NFS_INO_INVALID_ATIME
1607 | NFS_INO_REVAL_FORCED
1608 | NFS_INO_REVAL_PAGECACHE);
1610 /* Do atomic weak cache consistency updates */
1611 invalid |= nfs_wcc_update_inode(inode, fattr);
1613 /* More cache consistency checks */
1614 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1615 if (inode->i_version != fattr->change_attr) {
1616 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1617 inode->i_sb->s_id, inode->i_ino);
1618 invalid |= NFS_INO_INVALID_ATTR
1619 | NFS_INO_INVALID_DATA
1620 | NFS_INO_INVALID_ACCESS
1621 | NFS_INO_INVALID_ACL
1622 | NFS_INO_REVAL_PAGECACHE;
1623 if (S_ISDIR(inode->i_mode))
1624 nfs_force_lookup_revalidate(inode);
1625 inode->i_version = fattr->change_attr;
1627 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1628 nfsi->cache_validity |= save_cache_validity;
1630 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1631 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1632 } else if (server->caps & NFS_CAP_MTIME)
1633 nfsi->cache_validity |= save_cache_validity &
1634 (NFS_INO_INVALID_ATTR
1635 | NFS_INO_REVAL_FORCED);
1637 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1638 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1639 } else if (server->caps & NFS_CAP_CTIME)
1640 nfsi->cache_validity |= save_cache_validity &
1641 (NFS_INO_INVALID_ATTR
1642 | NFS_INO_REVAL_FORCED);
1644 /* Check if our cached file size is stale */
1645 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1646 new_isize = nfs_size_to_loff_t(fattr->size);
1647 cur_isize = i_size_read(inode);
1648 if (new_isize != cur_isize) {
1649 /* Do we perhaps have any outstanding writes, or has
1650 * the file grown beyond our last write? */
1651 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1652 i_size_write(inode, new_isize);
1653 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1654 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1656 dprintk("NFS: isize change on server for file %s/%ld "
1660 (long long)cur_isize,
1661 (long long)new_isize);
1664 nfsi->cache_validity |= save_cache_validity &
1665 (NFS_INO_INVALID_ATTR
1666 | NFS_INO_REVAL_PAGECACHE
1667 | NFS_INO_REVAL_FORCED);
1670 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1671 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1672 else if (server->caps & NFS_CAP_ATIME)
1673 nfsi->cache_validity |= save_cache_validity &
1674 (NFS_INO_INVALID_ATIME
1675 | NFS_INO_REVAL_FORCED);
1677 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1678 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1679 umode_t newmode = inode->i_mode & S_IFMT;
1680 newmode |= fattr->mode & S_IALLUGO;
1681 inode->i_mode = newmode;
1682 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1684 } else if (server->caps & NFS_CAP_MODE)
1685 nfsi->cache_validity |= save_cache_validity &
1686 (NFS_INO_INVALID_ATTR
1687 | NFS_INO_INVALID_ACCESS
1688 | NFS_INO_INVALID_ACL
1689 | NFS_INO_REVAL_FORCED);
1691 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1692 if (!uid_eq(inode->i_uid, fattr->uid)) {
1693 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1694 inode->i_uid = fattr->uid;
1696 } else if (server->caps & NFS_CAP_OWNER)
1697 nfsi->cache_validity |= save_cache_validity &
1698 (NFS_INO_INVALID_ATTR
1699 | NFS_INO_INVALID_ACCESS
1700 | NFS_INO_INVALID_ACL
1701 | NFS_INO_REVAL_FORCED);
1703 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1704 if (!gid_eq(inode->i_gid, fattr->gid)) {
1705 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1706 inode->i_gid = fattr->gid;
1708 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1709 nfsi->cache_validity |= save_cache_validity &
1710 (NFS_INO_INVALID_ATTR
1711 | NFS_INO_INVALID_ACCESS
1712 | NFS_INO_INVALID_ACL
1713 | NFS_INO_REVAL_FORCED);
1715 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1716 if (inode->i_nlink != fattr->nlink) {
1717 invalid |= NFS_INO_INVALID_ATTR;
1718 if (S_ISDIR(inode->i_mode))
1719 invalid |= NFS_INO_INVALID_DATA;
1720 set_nlink(inode, fattr->nlink);
1722 } else if (server->caps & NFS_CAP_NLINK)
1723 nfsi->cache_validity |= save_cache_validity &
1724 (NFS_INO_INVALID_ATTR
1725 | NFS_INO_REVAL_FORCED);
1727 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1729 * report the blocks in 512byte units
1731 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1733 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1734 inode->i_blocks = fattr->du.nfs2.blocks;
1736 /* Update attrtimeo value if we're out of the unstable period */
1737 if (invalid & NFS_INO_INVALID_ATTR) {
1738 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1739 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1740 nfsi->attrtimeo_timestamp = now;
1741 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1743 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1744 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1745 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1746 nfsi->attrtimeo_timestamp = now;
1749 invalid &= ~NFS_INO_INVALID_ATTR;
1750 /* Don't invalidate the data if we were to blame */
1751 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1752 || S_ISLNK(inode->i_mode)))
1753 invalid &= ~NFS_INO_INVALID_DATA;
1754 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1755 (save_cache_validity & NFS_INO_REVAL_FORCED))
1756 nfs_set_cache_invalid(inode, invalid);
1761 * No need to worry about unhashing the dentry, as the
1762 * lookup validation will know that the inode is bad.
1763 * (But we fall through to invalidate the caches.)
1765 nfs_invalidate_inode(inode);
1769 struct inode *nfs_alloc_inode(struct super_block *sb)
1771 struct nfs_inode *nfsi;
1772 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1776 nfsi->cache_validity = 0UL;
1777 #if IS_ENABLED(CONFIG_NFS_V4)
1778 nfsi->nfs4_acl = NULL;
1779 #endif /* CONFIG_NFS_V4 */
1780 return &nfsi->vfs_inode;
1782 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1784 static void nfs_i_callback(struct rcu_head *head)
1786 struct inode *inode = container_of(head, struct inode, i_rcu);
1787 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1790 void nfs_destroy_inode(struct inode *inode)
1792 call_rcu(&inode->i_rcu, nfs_i_callback);
1794 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1796 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1798 #if IS_ENABLED(CONFIG_NFS_V4)
1799 INIT_LIST_HEAD(&nfsi->open_states);
1800 nfsi->delegation = NULL;
1801 init_rwsem(&nfsi->rwsem);
1802 nfsi->layout = NULL;
1806 static void init_once(void *foo)
1808 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1810 inode_init_once(&nfsi->vfs_inode);
1811 INIT_LIST_HEAD(&nfsi->open_files);
1812 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1813 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1814 INIT_LIST_HEAD(&nfsi->commit_info.list);
1815 nfsi->nrequests = 0;
1816 nfsi->commit_info.ncommit = 0;
1817 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1818 atomic_set(&nfsi->silly_count, 1);
1819 INIT_HLIST_HEAD(&nfsi->silly_list);
1820 init_waitqueue_head(&nfsi->waitqueue);
1821 nfs4_init_once(nfsi);
1824 static int __init nfs_init_inodecache(void)
1826 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1827 sizeof(struct nfs_inode),
1828 0, (SLAB_RECLAIM_ACCOUNT|
1831 if (nfs_inode_cachep == NULL)
1837 static void nfs_destroy_inodecache(void)
1840 * Make sure all delayed rcu free inodes are flushed before we
1844 kmem_cache_destroy(nfs_inode_cachep);
1847 struct workqueue_struct *nfsiod_workqueue;
1848 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1851 * start up the nfsiod workqueue
1853 static int nfsiod_start(void)
1855 struct workqueue_struct *wq;
1856 dprintk("RPC: creating workqueue nfsiod\n");
1857 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1860 nfsiod_workqueue = wq;
1865 * Destroy the nfsiod workqueue
1867 static void nfsiod_stop(void)
1869 struct workqueue_struct *wq;
1871 wq = nfsiod_workqueue;
1874 nfsiod_workqueue = NULL;
1875 destroy_workqueue(wq);
1879 EXPORT_SYMBOL_GPL(nfs_net_id);
1881 static int nfs_net_init(struct net *net)
1883 nfs_clients_init(net);
1884 return nfs_fs_proc_net_init(net);
1887 static void nfs_net_exit(struct net *net)
1889 nfs_fs_proc_net_exit(net);
1890 nfs_cleanup_cb_ident_idr(net);
1893 static struct pernet_operations nfs_net_ops = {
1894 .init = nfs_net_init,
1895 .exit = nfs_net_exit,
1897 .size = sizeof(struct nfs_net),
1903 static int __init init_nfs_fs(void)
1907 err = register_pernet_subsys(&nfs_net_ops);
1911 err = nfs_fscache_register();
1915 err = nfsiod_start();
1919 err = nfs_fs_proc_init();
1923 err = nfs_init_nfspagecache();
1927 err = nfs_init_inodecache();
1931 err = nfs_init_readpagecache();
1935 err = nfs_init_writepagecache();
1939 err = nfs_init_directcache();
1943 #ifdef CONFIG_PROC_FS
1944 rpc_proc_register(&init_net, &nfs_rpcstat);
1946 if ((err = register_nfs_fs()) != 0)
1951 #ifdef CONFIG_PROC_FS
1952 rpc_proc_unregister(&init_net, "nfs");
1954 nfs_destroy_directcache();
1956 nfs_destroy_writepagecache();
1958 nfs_destroy_readpagecache();
1960 nfs_destroy_inodecache();
1962 nfs_destroy_nfspagecache();
1968 nfs_fscache_unregister();
1970 unregister_pernet_subsys(&nfs_net_ops);
1975 static void __exit exit_nfs_fs(void)
1977 nfs_destroy_directcache();
1978 nfs_destroy_writepagecache();
1979 nfs_destroy_readpagecache();
1980 nfs_destroy_inodecache();
1981 nfs_destroy_nfspagecache();
1982 nfs_fscache_unregister();
1983 unregister_pernet_subsys(&nfs_net_ops);
1984 #ifdef CONFIG_PROC_FS
1985 rpc_proc_unregister(&init_net, "nfs");
1987 unregister_nfs_fs();
1992 /* Not quite true; I just maintain it */
1993 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1994 MODULE_LICENSE("GPL");
1995 module_param(enable_ino64, bool, 0644);
1997 module_init(init_nfs_fs)
1998 module_exit(exit_nfs_fs)