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 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
392 } else if (S_ISDIR(inode->i_mode)) {
393 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
394 inode->i_fop = &nfs_dir_operations;
395 inode->i_data.a_ops = &nfs_dir_aops;
396 /* Deal with crossing mountpoints */
397 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
398 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
399 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
400 inode->i_op = &nfs_referral_inode_operations;
402 inode->i_op = &nfs_mountpoint_inode_operations;
404 inode->i_flags |= S_AUTOMOUNT;
406 } else if (S_ISLNK(inode->i_mode))
407 inode->i_op = &nfs_symlink_inode_operations;
409 init_special_inode(inode, inode->i_mode, fattr->rdev);
411 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
412 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
413 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
414 inode->i_version = 0;
417 inode->i_uid = make_kuid(&init_user_ns, -2);
418 inode->i_gid = make_kgid(&init_user_ns, -2);
420 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
424 nfsi->read_cache_jiffies = fattr->time_start;
425 nfsi->attr_gencount = fattr->gencount;
426 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
427 inode->i_atime = fattr->atime;
428 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
429 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
430 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
431 inode->i_mtime = fattr->mtime;
432 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
433 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
434 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
435 inode->i_ctime = fattr->ctime;
436 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
437 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
438 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
439 inode->i_version = fattr->change_attr;
440 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
441 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
442 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
443 inode->i_size = nfs_size_to_loff_t(fattr->size);
445 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
446 | NFS_INO_REVAL_PAGECACHE);
447 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
448 set_nlink(inode, fattr->nlink);
449 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
450 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
451 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
452 inode->i_uid = fattr->uid;
453 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
454 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
455 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
456 inode->i_gid = fattr->gid;
457 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
458 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
459 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
460 inode->i_blocks = fattr->du.nfs2.blocks;
461 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
463 * report the blocks in 512byte units
465 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
468 nfs_setsecurity(inode, fattr, label);
470 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
471 nfsi->attrtimeo_timestamp = now;
472 nfsi->access_cache = RB_ROOT;
474 nfs_fscache_init_inode(inode);
476 unlock_new_inode(inode);
478 nfs_refresh_inode(inode, fattr);
479 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
481 (unsigned long long)NFS_FILEID(inode),
482 nfs_display_fhandle_hash(fh),
483 atomic_read(&inode->i_count));
489 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
492 EXPORT_SYMBOL_GPL(nfs_fhget);
494 #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)
497 nfs_setattr(struct dentry *dentry, struct iattr *attr)
499 struct inode *inode = dentry->d_inode;
500 struct nfs_fattr *fattr;
503 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
505 /* skip mode change if it's just for clearing setuid/setgid */
506 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
507 attr->ia_valid &= ~ATTR_MODE;
509 if (attr->ia_valid & ATTR_SIZE) {
510 BUG_ON(!S_ISREG(inode->i_mode));
512 if (attr->ia_size == i_size_read(inode))
513 attr->ia_valid &= ~ATTR_SIZE;
516 /* Optimization: if the end result is no change, don't RPC */
517 attr->ia_valid &= NFS_VALID_ATTRS;
518 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
521 trace_nfs_setattr_enter(inode);
523 /* Write all dirty data */
524 if (S_ISREG(inode->i_mode)) {
525 nfs_inode_dio_wait(inode);
529 fattr = nfs_alloc_fattr();
533 * Return any delegations if we're going to change ACLs
535 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
536 NFS_PROTO(inode)->return_delegation(inode);
537 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
539 error = nfs_refresh_inode(inode, fattr);
540 nfs_free_fattr(fattr);
542 trace_nfs_setattr_exit(inode, error);
545 EXPORT_SYMBOL_GPL(nfs_setattr);
548 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
549 * @inode: inode of the file used
550 * @offset: file offset to start truncating
552 * This is a copy of the common vmtruncate, but with the locking
553 * corrected to take into account the fact that NFS requires
554 * inode->i_size to be updated under the inode->i_lock.
556 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
560 err = inode_newsize_ok(inode, offset);
564 spin_lock(&inode->i_lock);
565 i_size_write(inode, offset);
568 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
569 spin_unlock(&inode->i_lock);
571 truncate_pagecache(inode, offset);
577 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
578 * @inode: pointer to struct inode
579 * @attr: pointer to struct iattr
581 * Note: we do this in the *proc.c in order to ensure that
582 * it works for things like exclusive creates too.
584 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
586 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
587 spin_lock(&inode->i_lock);
588 if ((attr->ia_valid & ATTR_MODE) != 0) {
589 int mode = attr->ia_mode & S_IALLUGO;
590 mode |= inode->i_mode & ~S_IALLUGO;
591 inode->i_mode = mode;
593 if ((attr->ia_valid & ATTR_UID) != 0)
594 inode->i_uid = attr->ia_uid;
595 if ((attr->ia_valid & ATTR_GID) != 0)
596 inode->i_gid = attr->ia_gid;
597 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
598 | NFS_INO_INVALID_ACL);
599 spin_unlock(&inode->i_lock);
601 if ((attr->ia_valid & ATTR_SIZE) != 0) {
602 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
603 nfs_vmtruncate(inode, attr->ia_size);
606 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
608 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
610 struct dentry *parent;
612 parent = dget_parent(dentry);
613 nfs_force_use_readdirplus(parent->d_inode);
617 static bool nfs_need_revalidate_inode(struct inode *inode)
619 if (NFS_I(inode)->cache_validity &
620 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
622 if (nfs_attribute_cache_expired(inode))
627 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
629 struct inode *inode = dentry->d_inode;
630 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
633 trace_nfs_getattr_enter(inode);
634 /* Flush out writes to the server in order to update c/mtime. */
635 if (S_ISREG(inode->i_mode)) {
636 nfs_inode_dio_wait(inode);
637 err = filemap_write_and_wait(inode->i_mapping);
643 * We may force a getattr if the user cares about atime.
645 * Note that we only have to check the vfsmount flags here:
646 * - NFS always sets S_NOATIME by so checking it would give a
648 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
649 * no point in checking those.
651 if ((mnt->mnt_flags & MNT_NOATIME) ||
652 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
655 if (need_atime || nfs_need_revalidate_inode(inode)) {
656 struct nfs_server *server = NFS_SERVER(inode);
658 if (server->caps & NFS_CAP_READDIRPLUS)
659 nfs_request_parent_use_readdirplus(dentry);
660 err = __nfs_revalidate_inode(server, inode);
663 generic_fillattr(inode, stat);
664 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
667 trace_nfs_getattr_exit(inode, err);
670 EXPORT_SYMBOL_GPL(nfs_getattr);
672 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
674 atomic_set(&l_ctx->count, 1);
675 l_ctx->lockowner.l_owner = current->files;
676 l_ctx->lockowner.l_pid = current->tgid;
677 INIT_LIST_HEAD(&l_ctx->list);
678 nfs_iocounter_init(&l_ctx->io_count);
681 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
683 struct nfs_lock_context *head = &ctx->lock_context;
684 struct nfs_lock_context *pos = head;
687 if (pos->lockowner.l_owner != current->files)
689 if (pos->lockowner.l_pid != current->tgid)
691 atomic_inc(&pos->count);
693 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
697 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
699 struct nfs_lock_context *res, *new = NULL;
700 struct inode *inode = ctx->dentry->d_inode;
702 spin_lock(&inode->i_lock);
703 res = __nfs_find_lock_context(ctx);
705 spin_unlock(&inode->i_lock);
706 new = kmalloc(sizeof(*new), GFP_KERNEL);
708 return ERR_PTR(-ENOMEM);
709 nfs_init_lock_context(new);
710 spin_lock(&inode->i_lock);
711 res = __nfs_find_lock_context(ctx);
713 list_add_tail(&new->list, &ctx->lock_context.list);
714 new->open_context = ctx;
719 spin_unlock(&inode->i_lock);
723 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
725 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
727 struct nfs_open_context *ctx = l_ctx->open_context;
728 struct inode *inode = ctx->dentry->d_inode;
730 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
732 list_del(&l_ctx->list);
733 spin_unlock(&inode->i_lock);
736 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
739 * nfs_close_context - Common close_context() routine NFSv2/v3
740 * @ctx: pointer to context
741 * @is_sync: is this a synchronous close
743 * always ensure that the attributes are up to date if we're mounted
744 * with close-to-open semantics
746 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
749 struct nfs_server *server;
751 if (!(ctx->mode & FMODE_WRITE))
755 inode = ctx->dentry->d_inode;
756 if (!list_empty(&NFS_I(inode)->open_files))
758 server = NFS_SERVER(inode);
759 if (server->flags & NFS_MOUNT_NOCTO)
761 nfs_revalidate_inode(server, inode);
763 EXPORT_SYMBOL_GPL(nfs_close_context);
765 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
767 struct nfs_open_context *ctx;
768 struct rpc_cred *cred = rpc_lookup_cred();
770 return ERR_CAST(cred);
772 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
775 return ERR_PTR(-ENOMEM);
777 nfs_sb_active(dentry->d_sb);
778 ctx->dentry = dget(dentry);
784 nfs_init_lock_context(&ctx->lock_context);
785 ctx->lock_context.open_context = ctx;
786 INIT_LIST_HEAD(&ctx->list);
787 ctx->mdsthreshold = NULL;
790 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
792 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
795 atomic_inc(&ctx->lock_context.count);
798 EXPORT_SYMBOL_GPL(get_nfs_open_context);
800 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
802 struct inode *inode = ctx->dentry->d_inode;
803 struct super_block *sb = ctx->dentry->d_sb;
805 if (!list_empty(&ctx->list)) {
806 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
808 list_del(&ctx->list);
809 spin_unlock(&inode->i_lock);
810 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
813 NFS_PROTO(inode)->close_context(ctx, is_sync);
814 if (ctx->cred != NULL)
815 put_rpccred(ctx->cred);
818 kfree(ctx->mdsthreshold);
822 void put_nfs_open_context(struct nfs_open_context *ctx)
824 __put_nfs_open_context(ctx, 0);
826 EXPORT_SYMBOL_GPL(put_nfs_open_context);
829 * Ensure that mmap has a recent RPC credential for use when writing out
832 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
834 struct inode *inode = ctx->dentry->d_inode;
835 struct nfs_inode *nfsi = NFS_I(inode);
837 spin_lock(&inode->i_lock);
838 list_add(&ctx->list, &nfsi->open_files);
839 spin_unlock(&inode->i_lock);
841 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
843 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
845 filp->private_data = get_nfs_open_context(ctx);
846 if (list_empty(&ctx->list))
847 nfs_inode_attach_open_context(ctx);
849 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
852 * Given an inode, search for an open context with the desired characteristics
854 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
856 struct nfs_inode *nfsi = NFS_I(inode);
857 struct nfs_open_context *pos, *ctx = NULL;
859 spin_lock(&inode->i_lock);
860 list_for_each_entry(pos, &nfsi->open_files, list) {
861 if (cred != NULL && pos->cred != cred)
863 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
865 ctx = get_nfs_open_context(pos);
868 spin_unlock(&inode->i_lock);
872 static void nfs_file_clear_open_context(struct file *filp)
874 struct nfs_open_context *ctx = nfs_file_open_context(filp);
877 struct inode *inode = ctx->dentry->d_inode;
879 filp->private_data = NULL;
880 spin_lock(&inode->i_lock);
881 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
882 spin_unlock(&inode->i_lock);
883 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
888 * These allocate and release file read/write context information.
890 int nfs_open(struct inode *inode, struct file *filp)
892 struct nfs_open_context *ctx;
894 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
897 nfs_file_set_open_context(filp, ctx);
898 put_nfs_open_context(ctx);
899 nfs_fscache_open_file(inode, filp);
903 int nfs_release(struct inode *inode, struct file *filp)
905 nfs_file_clear_open_context(filp);
910 * This function is called whenever some part of NFS notices that
911 * the cached attributes have to be refreshed.
914 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
916 int status = -ESTALE;
917 struct nfs4_label *label = NULL;
918 struct nfs_fattr *fattr = NULL;
919 struct nfs_inode *nfsi = NFS_I(inode);
921 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
922 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
924 trace_nfs_revalidate_inode_enter(inode);
926 if (is_bad_inode(inode))
928 if (NFS_STALE(inode))
932 fattr = nfs_alloc_fattr();
936 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
938 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
940 status = PTR_ERR(label);
944 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
946 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
948 (unsigned long long)NFS_FILEID(inode), status);
949 if (status == -ESTALE) {
950 nfs_zap_caches(inode);
951 if (!S_ISDIR(inode->i_mode))
952 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
957 status = nfs_refresh_inode(inode, fattr);
959 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
961 (unsigned long long)NFS_FILEID(inode), status);
965 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
966 nfs_zap_acl_cache(inode);
968 nfs_setsecurity(inode, fattr, label);
970 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
972 (unsigned long long)NFS_FILEID(inode));
975 nfs4_label_free(label);
977 nfs_free_fattr(fattr);
978 trace_nfs_revalidate_inode_exit(inode, status);
982 int nfs_attribute_timeout(struct inode *inode)
984 struct nfs_inode *nfsi = NFS_I(inode);
986 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
989 int nfs_attribute_cache_expired(struct inode *inode)
991 if (nfs_have_delegated_attributes(inode))
993 return nfs_attribute_timeout(inode);
997 * nfs_revalidate_inode - Revalidate the inode attributes
998 * @server - pointer to nfs_server struct
999 * @inode - pointer to inode struct
1001 * Updates inode attribute information by retrieving the data from the server.
1003 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1005 if (!nfs_need_revalidate_inode(inode))
1006 return NFS_STALE(inode) ? -ESTALE : 0;
1007 return __nfs_revalidate_inode(server, inode);
1009 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1011 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1013 if (!(NFS_I(inode)->cache_validity &
1014 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1015 && !nfs_attribute_cache_expired(inode))
1016 return NFS_STALE(inode) ? -ESTALE : 0;
1020 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1022 struct nfs_inode *nfsi = NFS_I(inode);
1025 if (mapping->nrpages != 0) {
1026 if (S_ISREG(inode->i_mode)) {
1027 ret = nfs_sync_mapping(mapping);
1031 ret = invalidate_inode_pages2(mapping);
1035 if (S_ISDIR(inode->i_mode)) {
1036 spin_lock(&inode->i_lock);
1037 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1038 spin_unlock(&inode->i_lock);
1040 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1041 nfs_fscache_wait_on_invalidate(inode);
1043 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1045 (unsigned long long)NFS_FILEID(inode));
1049 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1051 if (nfs_have_delegated_attributes(inode))
1053 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1054 || nfs_attribute_timeout(inode)
1055 || NFS_STALE(inode);
1059 * nfs_revalidate_mapping - Revalidate the pagecache
1060 * @inode - pointer to host inode
1061 * @mapping - pointer to mapping
1063 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1065 struct nfs_inode *nfsi = NFS_I(inode);
1066 unsigned long *bitlock = &nfsi->flags;
1069 /* swapfiles are not supposed to be shared. */
1070 if (IS_SWAPFILE(inode))
1073 if (nfs_mapping_need_revalidate_inode(inode)) {
1074 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1080 * We must clear NFS_INO_INVALID_DATA first to ensure that
1081 * invalidations that come in while we're shooting down the mappings
1082 * are respected. But, that leaves a race window where one revalidator
1083 * can clear the flag, and then another checks it before the mapping
1084 * gets invalidated. Fix that by serializing access to this part of
1087 * At the same time, we need to allow other tasks to see whether we
1088 * might be in the middle of invalidating the pages, so we only set
1089 * the bit lock here if it looks like we're going to be doing that.
1092 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1093 nfs_wait_bit_killable, TASK_KILLABLE);
1096 spin_lock(&inode->i_lock);
1097 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1098 spin_unlock(&inode->i_lock);
1101 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1103 spin_unlock(&inode->i_lock);
1107 set_bit(NFS_INO_INVALIDATING, bitlock);
1109 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1110 spin_unlock(&inode->i_lock);
1111 trace_nfs_invalidate_mapping_enter(inode);
1112 ret = nfs_invalidate_mapping(inode, mapping);
1113 trace_nfs_invalidate_mapping_exit(inode, ret);
1115 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1116 smp_mb__after_atomic();
1117 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1122 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1124 struct nfs_inode *nfsi = NFS_I(inode);
1125 unsigned long ret = 0;
1127 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1128 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1129 && inode->i_version == fattr->pre_change_attr) {
1130 inode->i_version = fattr->change_attr;
1131 if (S_ISDIR(inode->i_mode))
1132 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1133 ret |= NFS_INO_INVALID_ATTR;
1135 /* If we have atomic WCC data, we may update some attributes */
1136 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1137 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1138 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1139 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1140 ret |= NFS_INO_INVALID_ATTR;
1143 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1144 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1145 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1146 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1147 if (S_ISDIR(inode->i_mode))
1148 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1149 ret |= NFS_INO_INVALID_ATTR;
1151 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1152 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1153 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1154 && nfsi->nrequests == 0) {
1155 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1156 ret |= NFS_INO_INVALID_ATTR;
1163 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1164 * @inode - pointer to inode
1165 * @fattr - updated attributes
1167 * Verifies the attribute cache. If we have just changed the attributes,
1168 * so that fattr carries weak cache consistency data, then it may
1169 * also update the ctime/mtime/change_attribute.
1171 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1173 struct nfs_inode *nfsi = NFS_I(inode);
1174 loff_t cur_size, new_isize;
1175 unsigned long invalid = 0;
1178 if (nfs_have_delegated_attributes(inode))
1180 /* Has the inode gone and changed behind our back? */
1181 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1183 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1186 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1187 inode->i_version != fattr->change_attr)
1188 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1190 /* Verify a few of the more important attributes */
1191 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1192 invalid |= NFS_INO_INVALID_ATTR;
1194 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1195 cur_size = i_size_read(inode);
1196 new_isize = nfs_size_to_loff_t(fattr->size);
1197 if (cur_size != new_isize && nfsi->nrequests == 0)
1198 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1201 /* Have any file permissions changed? */
1202 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1203 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1204 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1205 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1206 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1207 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1209 /* Has the link count changed? */
1210 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1211 invalid |= NFS_INO_INVALID_ATTR;
1213 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1214 invalid |= NFS_INO_INVALID_ATIME;
1217 nfs_set_cache_invalid(inode, invalid);
1219 nfsi->read_cache_jiffies = fattr->time_start;
1223 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1225 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1227 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1230 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1232 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1234 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1237 static atomic_long_t nfs_attr_generation_counter;
1239 static unsigned long nfs_read_attr_generation_counter(void)
1241 return atomic_long_read(&nfs_attr_generation_counter);
1244 unsigned long nfs_inc_attr_generation_counter(void)
1246 return atomic_long_inc_return(&nfs_attr_generation_counter);
1249 void nfs_fattr_init(struct nfs_fattr *fattr)
1252 fattr->time_start = jiffies;
1253 fattr->gencount = nfs_inc_attr_generation_counter();
1254 fattr->owner_name = NULL;
1255 fattr->group_name = NULL;
1257 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1259 struct nfs_fattr *nfs_alloc_fattr(void)
1261 struct nfs_fattr *fattr;
1263 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1265 nfs_fattr_init(fattr);
1268 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1270 struct nfs_fh *nfs_alloc_fhandle(void)
1274 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1279 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1283 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1284 * in the same way that wireshark does
1288 * For debugging only.
1290 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1292 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1293 * not on the result */
1294 return nfs_fhandle_hash(fh);
1296 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1299 * _nfs_display_fhandle - display an NFS file handle on the console
1301 * @fh: file handle to display
1302 * @caption: display caption
1304 * For debugging only.
1306 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1310 if (fh == NULL || fh->size == 0) {
1311 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1315 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1316 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1317 for (i = 0; i < fh->size; i += 16) {
1318 __be32 *pos = (__be32 *)&fh->data[i];
1320 switch ((fh->size - i - 1) >> 2) {
1322 printk(KERN_DEFAULT " %08x\n",
1326 printk(KERN_DEFAULT " %08x %08x\n",
1327 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1330 printk(KERN_DEFAULT " %08x %08x %08x\n",
1331 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1332 be32_to_cpup(pos + 2));
1335 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1336 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1337 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1341 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1345 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1346 * @inode - pointer to inode
1347 * @fattr - attributes
1349 * Attempt to divine whether or not an RPC call reply carrying stale
1350 * attributes got scheduled after another call carrying updated ones.
1352 * To do so, the function first assumes that a more recent ctime means
1353 * that the attributes in fattr are newer, however it also attempt to
1354 * catch the case where ctime either didn't change, or went backwards
1355 * (if someone reset the clock on the server) by looking at whether
1356 * or not this RPC call was started after the inode was last updated.
1357 * Note also the check for wraparound of 'attr_gencount'
1359 * The function returns 'true' if it thinks the attributes in 'fattr' are
1360 * more recent than the ones cached in the inode.
1363 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1365 const struct nfs_inode *nfsi = NFS_I(inode);
1367 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1368 nfs_ctime_need_update(inode, fattr) ||
1369 nfs_size_need_update(inode, fattr) ||
1370 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1374 * Don't trust the change_attribute, mtime, ctime or size if
1375 * a pnfs LAYOUTCOMMIT is outstanding
1377 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1378 struct nfs_fattr *fattr)
1380 if (pnfs_layoutcommit_outstanding(inode))
1381 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1382 NFS_ATTR_FATTR_MTIME |
1383 NFS_ATTR_FATTR_CTIME |
1384 NFS_ATTR_FATTR_SIZE);
1387 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1391 trace_nfs_refresh_inode_enter(inode);
1393 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1395 if (nfs_inode_attrs_need_update(inode, fattr))
1396 ret = nfs_update_inode(inode, fattr);
1398 ret = nfs_check_inode_attributes(inode, fattr);
1400 trace_nfs_refresh_inode_exit(inode, ret);
1405 * nfs_refresh_inode - try to update the inode attribute cache
1406 * @inode - pointer to inode
1407 * @fattr - updated attributes
1409 * Check that an RPC call that returned attributes has not overlapped with
1410 * other recent updates of the inode metadata, then decide whether it is
1411 * safe to do a full update of the inode attributes, or whether just to
1412 * call nfs_check_inode_attributes.
1414 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1418 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1420 spin_lock(&inode->i_lock);
1421 status = nfs_refresh_inode_locked(inode, fattr);
1422 spin_unlock(&inode->i_lock);
1426 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1428 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1430 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1432 if (S_ISDIR(inode->i_mode))
1433 invalid |= NFS_INO_INVALID_DATA;
1434 nfs_set_cache_invalid(inode, invalid);
1435 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1437 return nfs_refresh_inode_locked(inode, fattr);
1441 * nfs_post_op_update_inode - try to update the inode attribute cache
1442 * @inode - pointer to inode
1443 * @fattr - updated attributes
1445 * After an operation that has changed the inode metadata, mark the
1446 * attribute cache as being invalid, then try to update it.
1448 * NB: if the server didn't return any post op attributes, this
1449 * function will force the retrieval of attributes before the next
1450 * NFS request. Thus it should be used only for operations that
1451 * are expected to change one or more attributes, to avoid
1452 * unnecessary NFS requests and trips through nfs_update_inode().
1454 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1458 spin_lock(&inode->i_lock);
1459 status = nfs_post_op_update_inode_locked(inode, fattr);
1460 spin_unlock(&inode->i_lock);
1464 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1467 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1468 * @inode - pointer to inode
1469 * @fattr - updated attributes
1471 * After an operation that has changed the inode metadata, mark the
1472 * attribute cache as being invalid, then try to update it. Fake up
1473 * weak cache consistency data, if none exist.
1475 * This function is mainly designed to be used by the ->write_done() functions.
1477 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1481 spin_lock(&inode->i_lock);
1482 /* Don't do a WCC update if these attributes are already stale */
1483 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1484 !nfs_inode_attrs_need_update(inode, fattr)) {
1485 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1486 | NFS_ATTR_FATTR_PRESIZE
1487 | NFS_ATTR_FATTR_PREMTIME
1488 | NFS_ATTR_FATTR_PRECTIME);
1491 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1492 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1493 fattr->pre_change_attr = inode->i_version;
1494 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1496 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1497 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1498 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1499 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1501 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1502 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1503 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1504 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1506 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1507 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1508 fattr->pre_size = i_size_read(inode);
1509 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1512 status = nfs_post_op_update_inode_locked(inode, fattr);
1513 spin_unlock(&inode->i_lock);
1516 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1519 * Many nfs protocol calls return the new file attributes after
1520 * an operation. Here we update the inode to reflect the state
1521 * of the server's inode.
1523 * This is a bit tricky because we have to make sure all dirty pages
1524 * have been sent off to the server before calling invalidate_inode_pages.
1525 * To make sure no other process adds more write requests while we try
1526 * our best to flush them, we make them sleep during the attribute refresh.
1528 * A very similar scenario holds for the dir cache.
1530 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1532 struct nfs_server *server;
1533 struct nfs_inode *nfsi = NFS_I(inode);
1534 loff_t cur_isize, new_isize;
1535 unsigned long invalid = 0;
1536 unsigned long now = jiffies;
1537 unsigned long save_cache_validity;
1539 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1540 __func__, inode->i_sb->s_id, inode->i_ino,
1541 nfs_display_fhandle_hash(NFS_FH(inode)),
1542 atomic_read(&inode->i_count), fattr->valid);
1544 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1545 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1546 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1547 NFS_SERVER(inode)->nfs_client->cl_hostname,
1548 inode->i_sb->s_id, (long long)nfsi->fileid,
1549 (long long)fattr->fileid);
1554 * Make sure the inode's type hasn't changed.
1556 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1558 * Big trouble! The inode has become a different object.
1560 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1561 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1565 server = NFS_SERVER(inode);
1566 /* Update the fsid? */
1567 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1568 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1569 !IS_AUTOMOUNT(inode))
1570 server->fsid = fattr->fsid;
1573 * Update the read time so we don't revalidate too often.
1575 nfsi->read_cache_jiffies = fattr->time_start;
1577 save_cache_validity = nfsi->cache_validity;
1578 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1579 | NFS_INO_INVALID_ATIME
1580 | NFS_INO_REVAL_FORCED
1581 | NFS_INO_REVAL_PAGECACHE);
1583 /* Do atomic weak cache consistency updates */
1584 invalid |= nfs_wcc_update_inode(inode, fattr);
1586 /* More cache consistency checks */
1587 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1588 if (inode->i_version != fattr->change_attr) {
1589 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1590 inode->i_sb->s_id, inode->i_ino);
1591 invalid |= NFS_INO_INVALID_ATTR
1592 | NFS_INO_INVALID_DATA
1593 | NFS_INO_INVALID_ACCESS
1594 | NFS_INO_INVALID_ACL
1595 | NFS_INO_REVAL_PAGECACHE;
1596 if (S_ISDIR(inode->i_mode))
1597 nfs_force_lookup_revalidate(inode);
1598 inode->i_version = fattr->change_attr;
1600 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1601 nfsi->cache_validity |= save_cache_validity;
1603 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1604 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1605 } else if (server->caps & NFS_CAP_MTIME)
1606 nfsi->cache_validity |= save_cache_validity &
1607 (NFS_INO_INVALID_ATTR
1608 | NFS_INO_REVAL_FORCED);
1610 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1611 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1612 } else if (server->caps & NFS_CAP_CTIME)
1613 nfsi->cache_validity |= save_cache_validity &
1614 (NFS_INO_INVALID_ATTR
1615 | NFS_INO_REVAL_FORCED);
1617 /* Check if our cached file size is stale */
1618 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1619 new_isize = nfs_size_to_loff_t(fattr->size);
1620 cur_isize = i_size_read(inode);
1621 if (new_isize != cur_isize) {
1622 /* Do we perhaps have any outstanding writes, or has
1623 * the file grown beyond our last write? */
1624 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1625 i_size_write(inode, new_isize);
1626 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1627 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1629 dprintk("NFS: isize change on server for file %s/%ld "
1633 (long long)cur_isize,
1634 (long long)new_isize);
1637 nfsi->cache_validity |= save_cache_validity &
1638 (NFS_INO_INVALID_ATTR
1639 | NFS_INO_REVAL_PAGECACHE
1640 | NFS_INO_REVAL_FORCED);
1643 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1644 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1645 else if (server->caps & NFS_CAP_ATIME)
1646 nfsi->cache_validity |= save_cache_validity &
1647 (NFS_INO_INVALID_ATIME
1648 | NFS_INO_REVAL_FORCED);
1650 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1651 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1652 umode_t newmode = inode->i_mode & S_IFMT;
1653 newmode |= fattr->mode & S_IALLUGO;
1654 inode->i_mode = newmode;
1655 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1657 } else if (server->caps & NFS_CAP_MODE)
1658 nfsi->cache_validity |= save_cache_validity &
1659 (NFS_INO_INVALID_ATTR
1660 | NFS_INO_INVALID_ACCESS
1661 | NFS_INO_INVALID_ACL
1662 | NFS_INO_REVAL_FORCED);
1664 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1665 if (!uid_eq(inode->i_uid, fattr->uid)) {
1666 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1667 inode->i_uid = fattr->uid;
1669 } else if (server->caps & NFS_CAP_OWNER)
1670 nfsi->cache_validity |= save_cache_validity &
1671 (NFS_INO_INVALID_ATTR
1672 | NFS_INO_INVALID_ACCESS
1673 | NFS_INO_INVALID_ACL
1674 | NFS_INO_REVAL_FORCED);
1676 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1677 if (!gid_eq(inode->i_gid, fattr->gid)) {
1678 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1679 inode->i_gid = fattr->gid;
1681 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1682 nfsi->cache_validity |= save_cache_validity &
1683 (NFS_INO_INVALID_ATTR
1684 | NFS_INO_INVALID_ACCESS
1685 | NFS_INO_INVALID_ACL
1686 | NFS_INO_REVAL_FORCED);
1688 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1689 if (inode->i_nlink != fattr->nlink) {
1690 invalid |= NFS_INO_INVALID_ATTR;
1691 if (S_ISDIR(inode->i_mode))
1692 invalid |= NFS_INO_INVALID_DATA;
1693 set_nlink(inode, fattr->nlink);
1695 } else if (server->caps & NFS_CAP_NLINK)
1696 nfsi->cache_validity |= save_cache_validity &
1697 (NFS_INO_INVALID_ATTR
1698 | NFS_INO_REVAL_FORCED);
1700 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1702 * report the blocks in 512byte units
1704 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1706 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1707 inode->i_blocks = fattr->du.nfs2.blocks;
1709 /* Update attrtimeo value if we're out of the unstable period */
1710 if (invalid & NFS_INO_INVALID_ATTR) {
1711 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1712 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1713 nfsi->attrtimeo_timestamp = now;
1714 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1716 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1717 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1718 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1719 nfsi->attrtimeo_timestamp = now;
1722 invalid &= ~NFS_INO_INVALID_ATTR;
1723 /* Don't invalidate the data if we were to blame */
1724 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1725 || S_ISLNK(inode->i_mode)))
1726 invalid &= ~NFS_INO_INVALID_DATA;
1727 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1728 (save_cache_validity & NFS_INO_REVAL_FORCED))
1729 nfs_set_cache_invalid(inode, invalid);
1734 * No need to worry about unhashing the dentry, as the
1735 * lookup validation will know that the inode is bad.
1736 * (But we fall through to invalidate the caches.)
1738 nfs_invalidate_inode(inode);
1742 struct inode *nfs_alloc_inode(struct super_block *sb)
1744 struct nfs_inode *nfsi;
1745 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1749 nfsi->cache_validity = 0UL;
1750 #if IS_ENABLED(CONFIG_NFS_V4)
1751 nfsi->nfs4_acl = NULL;
1752 #endif /* CONFIG_NFS_V4 */
1753 return &nfsi->vfs_inode;
1755 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1757 static void nfs_i_callback(struct rcu_head *head)
1759 struct inode *inode = container_of(head, struct inode, i_rcu);
1760 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1763 void nfs_destroy_inode(struct inode *inode)
1765 call_rcu(&inode->i_rcu, nfs_i_callback);
1767 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1769 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1771 #if IS_ENABLED(CONFIG_NFS_V4)
1772 INIT_LIST_HEAD(&nfsi->open_states);
1773 nfsi->delegation = NULL;
1774 nfsi->delegation_state = 0;
1775 init_rwsem(&nfsi->rwsem);
1776 nfsi->layout = NULL;
1780 static void init_once(void *foo)
1782 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1784 inode_init_once(&nfsi->vfs_inode);
1785 INIT_LIST_HEAD(&nfsi->open_files);
1786 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1787 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1788 INIT_LIST_HEAD(&nfsi->commit_info.list);
1789 nfsi->nrequests = 0;
1790 nfsi->commit_info.ncommit = 0;
1791 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1792 atomic_set(&nfsi->silly_count, 1);
1793 INIT_HLIST_HEAD(&nfsi->silly_list);
1794 init_waitqueue_head(&nfsi->waitqueue);
1795 nfs4_init_once(nfsi);
1798 static int __init nfs_init_inodecache(void)
1800 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1801 sizeof(struct nfs_inode),
1802 0, (SLAB_RECLAIM_ACCOUNT|
1805 if (nfs_inode_cachep == NULL)
1811 static void nfs_destroy_inodecache(void)
1814 * Make sure all delayed rcu free inodes are flushed before we
1818 kmem_cache_destroy(nfs_inode_cachep);
1821 struct workqueue_struct *nfsiod_workqueue;
1822 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1825 * start up the nfsiod workqueue
1827 static int nfsiod_start(void)
1829 struct workqueue_struct *wq;
1830 dprintk("RPC: creating workqueue nfsiod\n");
1831 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1834 nfsiod_workqueue = wq;
1839 * Destroy the nfsiod workqueue
1841 static void nfsiod_stop(void)
1843 struct workqueue_struct *wq;
1845 wq = nfsiod_workqueue;
1848 nfsiod_workqueue = NULL;
1849 destroy_workqueue(wq);
1853 EXPORT_SYMBOL_GPL(nfs_net_id);
1855 static int nfs_net_init(struct net *net)
1857 nfs_clients_init(net);
1858 return nfs_fs_proc_net_init(net);
1861 static void nfs_net_exit(struct net *net)
1863 nfs_fs_proc_net_exit(net);
1864 nfs_cleanup_cb_ident_idr(net);
1867 static struct pernet_operations nfs_net_ops = {
1868 .init = nfs_net_init,
1869 .exit = nfs_net_exit,
1871 .size = sizeof(struct nfs_net),
1877 static int __init init_nfs_fs(void)
1881 err = register_pernet_subsys(&nfs_net_ops);
1885 err = nfs_fscache_register();
1889 err = nfsiod_start();
1893 err = nfs_fs_proc_init();
1897 err = nfs_init_nfspagecache();
1901 err = nfs_init_inodecache();
1905 err = nfs_init_readpagecache();
1909 err = nfs_init_writepagecache();
1913 err = nfs_init_directcache();
1917 #ifdef CONFIG_PROC_FS
1918 rpc_proc_register(&init_net, &nfs_rpcstat);
1920 if ((err = register_nfs_fs()) != 0)
1925 #ifdef CONFIG_PROC_FS
1926 rpc_proc_unregister(&init_net, "nfs");
1928 nfs_destroy_directcache();
1930 nfs_destroy_writepagecache();
1932 nfs_destroy_readpagecache();
1934 nfs_destroy_inodecache();
1936 nfs_destroy_nfspagecache();
1942 nfs_fscache_unregister();
1944 unregister_pernet_subsys(&nfs_net_ops);
1949 static void __exit exit_nfs_fs(void)
1951 nfs_destroy_directcache();
1952 nfs_destroy_writepagecache();
1953 nfs_destroy_readpagecache();
1954 nfs_destroy_inodecache();
1955 nfs_destroy_nfspagecache();
1956 nfs_fscache_unregister();
1957 unregister_pernet_subsys(&nfs_net_ops);
1958 #ifdef CONFIG_PROC_FS
1959 rpc_proc_unregister(&init_net, "nfs");
1961 unregister_nfs_fs();
1966 /* Not quite true; I just maintain it */
1967 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1968 MODULE_LICENSE("GPL");
1969 module_param(enable_ino64, bool, 0644);
1971 module_init(init_nfs_fs)
1972 module_exit(exit_nfs_fs)