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(void *word)
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(&inode->i_data, 0);
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);
151 * Invalidate the local caches
153 static void nfs_zap_caches_locked(struct inode *inode)
155 struct nfs_inode *nfsi = NFS_I(inode);
156 int mode = inode->i_mode;
158 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
160 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
161 nfsi->attrtimeo_timestamp = jiffies;
163 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
164 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
165 nfs_fscache_invalidate(inode);
166 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
167 | NFS_INO_INVALID_LABEL
168 | NFS_INO_INVALID_DATA
169 | NFS_INO_INVALID_ACCESS
170 | NFS_INO_INVALID_ACL
171 | NFS_INO_REVAL_PAGECACHE;
173 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
174 | NFS_INO_INVALID_LABEL
175 | NFS_INO_INVALID_ACCESS
176 | NFS_INO_INVALID_ACL
177 | NFS_INO_REVAL_PAGECACHE;
180 void nfs_zap_caches(struct inode *inode)
182 spin_lock(&inode->i_lock);
183 nfs_zap_caches_locked(inode);
184 spin_unlock(&inode->i_lock);
187 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
189 if (mapping->nrpages != 0) {
190 spin_lock(&inode->i_lock);
191 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
192 nfs_fscache_invalidate(inode);
193 spin_unlock(&inode->i_lock);
197 void nfs_zap_acl_cache(struct inode *inode)
199 void (*clear_acl_cache)(struct inode *);
201 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
202 if (clear_acl_cache != NULL)
203 clear_acl_cache(inode);
204 spin_lock(&inode->i_lock);
205 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
206 spin_unlock(&inode->i_lock);
208 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
210 void nfs_invalidate_atime(struct inode *inode)
212 spin_lock(&inode->i_lock);
213 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
214 spin_unlock(&inode->i_lock);
216 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
219 * Invalidate, but do not unhash, the inode.
220 * NB: must be called with inode->i_lock held!
222 static void nfs_invalidate_inode(struct inode *inode)
224 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
225 nfs_zap_caches_locked(inode);
228 struct nfs_find_desc {
230 struct nfs_fattr *fattr;
234 * In NFSv3 we can have 64bit inode numbers. In order to support
235 * this, and re-exported directories (also seen in NFSv2)
236 * we are forced to allow 2 different inodes to have the same
240 nfs_find_actor(struct inode *inode, void *opaque)
242 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
243 struct nfs_fh *fh = desc->fh;
244 struct nfs_fattr *fattr = desc->fattr;
246 if (NFS_FILEID(inode) != fattr->fileid)
248 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
250 if (nfs_compare_fh(NFS_FH(inode), fh))
252 if (is_bad_inode(inode) || NFS_STALE(inode))
258 nfs_init_locked(struct inode *inode, void *opaque)
260 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
261 struct nfs_fattr *fattr = desc->fattr;
263 set_nfs_fileid(inode, fattr->fileid);
264 nfs_copy_fh(NFS_FH(inode), desc->fh);
268 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
269 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
270 struct nfs4_label *label)
277 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
278 error = security_inode_notifysecctx(inode, label->label,
281 printk(KERN_ERR "%s() %s %d "
282 "security_inode_notifysecctx() %d\n",
284 (char *)label->label,
289 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
291 struct nfs4_label *label = NULL;
292 int minor_version = server->nfs_client->cl_minorversion;
294 if (minor_version < 2)
297 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
300 label = kzalloc(sizeof(struct nfs4_label), flags);
302 return ERR_PTR(-ENOMEM);
304 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
305 if (label->label == NULL) {
307 return ERR_PTR(-ENOMEM);
309 label->len = NFS4_MAXLABELLEN;
313 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
315 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
316 struct nfs4_label *label)
320 EXPORT_SYMBOL_GPL(nfs_setsecurity);
323 * This is our front-end to iget that looks up inodes by file handle
324 * instead of inode number.
327 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
329 struct nfs_find_desc desc = {
333 struct inode *inode = ERR_PTR(-ENOENT);
336 nfs_attr_check_mountpoint(sb, fattr);
338 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
339 !nfs_attr_use_mounted_on_fileid(fattr))
341 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
344 hash = nfs_fattr_to_ino_t(fattr);
346 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
348 inode = ERR_PTR(-ENOMEM);
352 if (inode->i_state & I_NEW) {
353 struct nfs_inode *nfsi = NFS_I(inode);
354 unsigned long now = jiffies;
356 /* We set i_ino for the few things that still rely on it,
360 /* We can't support update_atime(), since the server will reset it */
361 inode->i_flags |= S_NOATIME|S_NOCMTIME;
362 inode->i_mode = fattr->mode;
363 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
364 && nfs_server_capable(inode, NFS_CAP_MODE))
365 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
366 /* Why so? Because we want revalidate for devices/FIFOs, and
367 * that's precisely what we have in nfs_file_inode_operations.
369 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
370 if (S_ISREG(inode->i_mode)) {
371 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
372 inode->i_data.a_ops = &nfs_file_aops;
373 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
374 } else if (S_ISDIR(inode->i_mode)) {
375 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
376 inode->i_fop = &nfs_dir_operations;
377 inode->i_data.a_ops = &nfs_dir_aops;
378 /* Deal with crossing mountpoints */
379 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
380 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
381 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
382 inode->i_op = &nfs_referral_inode_operations;
384 inode->i_op = &nfs_mountpoint_inode_operations;
386 inode->i_flags |= S_AUTOMOUNT;
388 } else if (S_ISLNK(inode->i_mode))
389 inode->i_op = &nfs_symlink_inode_operations;
391 init_special_inode(inode, inode->i_mode, fattr->rdev);
393 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
394 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
395 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
396 inode->i_version = 0;
399 inode->i_uid = make_kuid(&init_user_ns, -2);
400 inode->i_gid = make_kgid(&init_user_ns, -2);
402 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
406 nfsi->read_cache_jiffies = fattr->time_start;
407 nfsi->attr_gencount = fattr->gencount;
408 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
409 inode->i_atime = fattr->atime;
410 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
411 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
412 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
413 inode->i_mtime = fattr->mtime;
414 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
415 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
416 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
417 inode->i_ctime = fattr->ctime;
418 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
419 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
420 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
421 inode->i_version = fattr->change_attr;
422 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
423 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
424 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
425 inode->i_size = nfs_size_to_loff_t(fattr->size);
427 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
428 | NFS_INO_REVAL_PAGECACHE;
429 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
430 set_nlink(inode, fattr->nlink);
431 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
432 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
433 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
434 inode->i_uid = fattr->uid;
435 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
436 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
437 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
438 inode->i_gid = fattr->gid;
439 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
440 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
441 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
442 inode->i_blocks = fattr->du.nfs2.blocks;
443 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
445 * report the blocks in 512byte units
447 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
450 nfs_setsecurity(inode, fattr, label);
452 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
453 nfsi->attrtimeo_timestamp = now;
454 nfsi->access_cache = RB_ROOT;
456 nfs_fscache_init_inode(inode);
458 unlock_new_inode(inode);
460 nfs_refresh_inode(inode, fattr);
461 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
463 (unsigned long long)NFS_FILEID(inode),
464 nfs_display_fhandle_hash(fh),
465 atomic_read(&inode->i_count));
471 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
474 EXPORT_SYMBOL_GPL(nfs_fhget);
476 #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)
479 nfs_setattr(struct dentry *dentry, struct iattr *attr)
481 struct inode *inode = dentry->d_inode;
482 struct nfs_fattr *fattr;
485 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
487 /* skip mode change if it's just for clearing setuid/setgid */
488 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
489 attr->ia_valid &= ~ATTR_MODE;
491 if (attr->ia_valid & ATTR_SIZE) {
492 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
493 attr->ia_valid &= ~ATTR_SIZE;
496 /* Optimization: if the end result is no change, don't RPC */
497 attr->ia_valid &= NFS_VALID_ATTRS;
498 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
501 trace_nfs_setattr_enter(inode);
503 /* Write all dirty data */
504 if (S_ISREG(inode->i_mode)) {
505 nfs_inode_dio_wait(inode);
509 fattr = nfs_alloc_fattr();
513 * Return any delegations if we're going to change ACLs
515 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
516 NFS_PROTO(inode)->return_delegation(inode);
517 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
519 error = nfs_refresh_inode(inode, fattr);
520 nfs_free_fattr(fattr);
522 trace_nfs_setattr_exit(inode, error);
525 EXPORT_SYMBOL_GPL(nfs_setattr);
528 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
529 * @inode: inode of the file used
530 * @offset: file offset to start truncating
532 * This is a copy of the common vmtruncate, but with the locking
533 * corrected to take into account the fact that NFS requires
534 * inode->i_size to be updated under the inode->i_lock.
536 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
540 err = inode_newsize_ok(inode, offset);
544 spin_lock(&inode->i_lock);
545 i_size_write(inode, offset);
546 spin_unlock(&inode->i_lock);
548 truncate_pagecache(inode, offset);
554 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
555 * @inode: pointer to struct inode
556 * @attr: pointer to struct iattr
558 * Note: we do this in the *proc.c in order to ensure that
559 * it works for things like exclusive creates too.
561 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
563 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
564 spin_lock(&inode->i_lock);
565 if ((attr->ia_valid & ATTR_MODE) != 0) {
566 int mode = attr->ia_mode & S_IALLUGO;
567 mode |= inode->i_mode & ~S_IALLUGO;
568 inode->i_mode = mode;
570 if ((attr->ia_valid & ATTR_UID) != 0)
571 inode->i_uid = attr->ia_uid;
572 if ((attr->ia_valid & ATTR_GID) != 0)
573 inode->i_gid = attr->ia_gid;
574 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
575 spin_unlock(&inode->i_lock);
577 if ((attr->ia_valid & ATTR_SIZE) != 0) {
578 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
579 nfs_vmtruncate(inode, attr->ia_size);
582 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
584 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
586 struct inode *inode = dentry->d_inode;
587 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
590 trace_nfs_getattr_enter(inode);
591 /* Flush out writes to the server in order to update c/mtime. */
592 if (S_ISREG(inode->i_mode)) {
593 nfs_inode_dio_wait(inode);
594 err = filemap_write_and_wait(inode->i_mapping);
600 * We may force a getattr if the user cares about atime.
602 * Note that we only have to check the vfsmount flags here:
603 * - NFS always sets S_NOATIME by so checking it would give a
605 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
606 * no point in checking those.
608 if ((mnt->mnt_flags & MNT_NOATIME) ||
609 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
613 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
615 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
617 generic_fillattr(inode, stat);
618 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
621 trace_nfs_getattr_exit(inode, err);
624 EXPORT_SYMBOL_GPL(nfs_getattr);
626 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
628 atomic_set(&l_ctx->count, 1);
629 l_ctx->lockowner.l_owner = current->files;
630 l_ctx->lockowner.l_pid = current->tgid;
631 INIT_LIST_HEAD(&l_ctx->list);
632 nfs_iocounter_init(&l_ctx->io_count);
635 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
637 struct nfs_lock_context *head = &ctx->lock_context;
638 struct nfs_lock_context *pos = head;
641 if (pos->lockowner.l_owner != current->files)
643 if (pos->lockowner.l_pid != current->tgid)
645 atomic_inc(&pos->count);
647 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
651 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
653 struct nfs_lock_context *res, *new = NULL;
654 struct inode *inode = ctx->dentry->d_inode;
656 spin_lock(&inode->i_lock);
657 res = __nfs_find_lock_context(ctx);
659 spin_unlock(&inode->i_lock);
660 new = kmalloc(sizeof(*new), GFP_KERNEL);
662 return ERR_PTR(-ENOMEM);
663 nfs_init_lock_context(new);
664 spin_lock(&inode->i_lock);
665 res = __nfs_find_lock_context(ctx);
667 list_add_tail(&new->list, &ctx->lock_context.list);
668 new->open_context = ctx;
673 spin_unlock(&inode->i_lock);
678 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
680 struct nfs_open_context *ctx = l_ctx->open_context;
681 struct inode *inode = ctx->dentry->d_inode;
683 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
685 list_del(&l_ctx->list);
686 spin_unlock(&inode->i_lock);
691 * nfs_close_context - Common close_context() routine NFSv2/v3
692 * @ctx: pointer to context
693 * @is_sync: is this a synchronous close
695 * always ensure that the attributes are up to date if we're mounted
696 * with close-to-open semantics
698 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
701 struct nfs_server *server;
703 if (!(ctx->mode & FMODE_WRITE))
707 inode = ctx->dentry->d_inode;
708 if (!list_empty(&NFS_I(inode)->open_files))
710 server = NFS_SERVER(inode);
711 if (server->flags & NFS_MOUNT_NOCTO)
713 nfs_revalidate_inode(server, inode);
715 EXPORT_SYMBOL_GPL(nfs_close_context);
717 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
719 struct nfs_open_context *ctx;
720 struct rpc_cred *cred = rpc_lookup_cred();
722 return ERR_CAST(cred);
724 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
727 return ERR_PTR(-ENOMEM);
729 nfs_sb_active(dentry->d_sb);
730 ctx->dentry = dget(dentry);
736 nfs_init_lock_context(&ctx->lock_context);
737 ctx->lock_context.open_context = ctx;
738 INIT_LIST_HEAD(&ctx->list);
739 ctx->mdsthreshold = NULL;
742 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
744 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
747 atomic_inc(&ctx->lock_context.count);
750 EXPORT_SYMBOL_GPL(get_nfs_open_context);
752 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
754 struct inode *inode = ctx->dentry->d_inode;
755 struct super_block *sb = ctx->dentry->d_sb;
757 if (!list_empty(&ctx->list)) {
758 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
760 list_del(&ctx->list);
761 spin_unlock(&inode->i_lock);
762 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
765 NFS_PROTO(inode)->close_context(ctx, is_sync);
766 if (ctx->cred != NULL)
767 put_rpccred(ctx->cred);
770 kfree(ctx->mdsthreshold);
774 void put_nfs_open_context(struct nfs_open_context *ctx)
776 __put_nfs_open_context(ctx, 0);
778 EXPORT_SYMBOL_GPL(put_nfs_open_context);
781 * Ensure that mmap has a recent RPC credential for use when writing out
784 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
786 struct inode *inode = ctx->dentry->d_inode;
787 struct nfs_inode *nfsi = NFS_I(inode);
789 spin_lock(&inode->i_lock);
790 list_add(&ctx->list, &nfsi->open_files);
791 spin_unlock(&inode->i_lock);
793 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
795 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
797 filp->private_data = get_nfs_open_context(ctx);
798 if (list_empty(&ctx->list))
799 nfs_inode_attach_open_context(ctx);
801 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
804 * Given an inode, search for an open context with the desired characteristics
806 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
808 struct nfs_inode *nfsi = NFS_I(inode);
809 struct nfs_open_context *pos, *ctx = NULL;
811 spin_lock(&inode->i_lock);
812 list_for_each_entry(pos, &nfsi->open_files, list) {
813 if (cred != NULL && pos->cred != cred)
815 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
817 ctx = get_nfs_open_context(pos);
820 spin_unlock(&inode->i_lock);
824 static void nfs_file_clear_open_context(struct file *filp)
826 struct nfs_open_context *ctx = nfs_file_open_context(filp);
829 struct inode *inode = ctx->dentry->d_inode;
831 filp->private_data = NULL;
832 spin_lock(&inode->i_lock);
833 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
834 spin_unlock(&inode->i_lock);
835 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
840 * These allocate and release file read/write context information.
842 int nfs_open(struct inode *inode, struct file *filp)
844 struct nfs_open_context *ctx;
846 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
849 nfs_file_set_open_context(filp, ctx);
850 put_nfs_open_context(ctx);
851 nfs_fscache_open_file(inode, filp);
855 int nfs_release(struct inode *inode, struct file *filp)
857 nfs_file_clear_open_context(filp);
862 * This function is called whenever some part of NFS notices that
863 * the cached attributes have to be refreshed.
866 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
868 int status = -ESTALE;
869 struct nfs4_label *label = NULL;
870 struct nfs_fattr *fattr = NULL;
871 struct nfs_inode *nfsi = NFS_I(inode);
873 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
874 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
876 trace_nfs_revalidate_inode_enter(inode);
878 if (is_bad_inode(inode))
880 if (NFS_STALE(inode))
884 fattr = nfs_alloc_fattr();
888 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
890 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
892 status = PTR_ERR(label);
896 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
898 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
900 (unsigned long long)NFS_FILEID(inode), status);
901 if (status == -ESTALE) {
902 nfs_zap_caches(inode);
903 if (!S_ISDIR(inode->i_mode))
904 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
909 status = nfs_refresh_inode(inode, fattr);
911 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
913 (unsigned long long)NFS_FILEID(inode), status);
917 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
918 nfs_zap_acl_cache(inode);
920 nfs_setsecurity(inode, fattr, label);
922 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
924 (unsigned long long)NFS_FILEID(inode));
927 nfs4_label_free(label);
929 nfs_free_fattr(fattr);
930 trace_nfs_revalidate_inode_exit(inode, status);
934 int nfs_attribute_timeout(struct inode *inode)
936 struct nfs_inode *nfsi = NFS_I(inode);
938 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
941 int nfs_attribute_cache_expired(struct inode *inode)
943 if (nfs_have_delegated_attributes(inode))
945 return nfs_attribute_timeout(inode);
949 * nfs_revalidate_inode - Revalidate the inode attributes
950 * @server - pointer to nfs_server struct
951 * @inode - pointer to inode struct
953 * Updates inode attribute information by retrieving the data from the server.
955 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
957 if (!(NFS_I(inode)->cache_validity &
958 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
959 && !nfs_attribute_cache_expired(inode))
960 return NFS_STALE(inode) ? -ESTALE : 0;
961 return __nfs_revalidate_inode(server, inode);
963 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
965 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
967 struct nfs_inode *nfsi = NFS_I(inode);
970 if (mapping->nrpages != 0) {
971 if (S_ISREG(inode->i_mode)) {
972 ret = nfs_sync_mapping(mapping);
976 ret = invalidate_inode_pages2(mapping);
980 spin_lock(&inode->i_lock);
981 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
982 if (S_ISDIR(inode->i_mode))
983 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
984 spin_unlock(&inode->i_lock);
985 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
986 nfs_fscache_wait_on_invalidate(inode);
988 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
990 (unsigned long long)NFS_FILEID(inode));
994 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
996 if (nfs_have_delegated_attributes(inode))
998 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
999 || nfs_attribute_timeout(inode)
1000 || NFS_STALE(inode);
1004 * nfs_revalidate_mapping - Revalidate the pagecache
1005 * @inode - pointer to host inode
1006 * @mapping - pointer to mapping
1008 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1010 struct nfs_inode *nfsi = NFS_I(inode);
1013 /* swapfiles are not supposed to be shared. */
1014 if (IS_SWAPFILE(inode))
1017 if (nfs_mapping_need_revalidate_inode(inode)) {
1018 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1022 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1023 trace_nfs_invalidate_mapping_enter(inode);
1024 ret = nfs_invalidate_mapping(inode, mapping);
1025 trace_nfs_invalidate_mapping_exit(inode, ret);
1032 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1034 struct nfs_inode *nfsi = NFS_I(inode);
1035 unsigned long ret = 0;
1037 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1038 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1039 && inode->i_version == fattr->pre_change_attr) {
1040 inode->i_version = fattr->change_attr;
1041 if (S_ISDIR(inode->i_mode))
1042 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1043 ret |= NFS_INO_INVALID_ATTR;
1045 /* If we have atomic WCC data, we may update some attributes */
1046 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1047 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1048 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1049 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1050 ret |= NFS_INO_INVALID_ATTR;
1053 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1054 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1055 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1056 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1057 if (S_ISDIR(inode->i_mode))
1058 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1059 ret |= NFS_INO_INVALID_ATTR;
1061 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1062 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1063 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1064 && nfsi->npages == 0) {
1065 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1066 ret |= NFS_INO_INVALID_ATTR;
1069 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1070 nfs_fscache_invalidate(inode);
1076 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1077 * @inode - pointer to inode
1078 * @fattr - updated attributes
1080 * Verifies the attribute cache. If we have just changed the attributes,
1081 * so that fattr carries weak cache consistency data, then it may
1082 * also update the ctime/mtime/change_attribute.
1084 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1086 struct nfs_inode *nfsi = NFS_I(inode);
1087 loff_t cur_size, new_isize;
1088 unsigned long invalid = 0;
1091 if (nfs_have_delegated_attributes(inode))
1093 /* Has the inode gone and changed behind our back? */
1094 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1096 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1099 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1100 inode->i_version != fattr->change_attr)
1101 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1103 /* Verify a few of the more important attributes */
1104 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1105 invalid |= NFS_INO_INVALID_ATTR;
1107 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1108 cur_size = i_size_read(inode);
1109 new_isize = nfs_size_to_loff_t(fattr->size);
1110 if (cur_size != new_isize && nfsi->npages == 0)
1111 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1114 /* Have any file permissions changed? */
1115 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1116 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1117 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1118 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1119 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1120 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1122 /* Has the link count changed? */
1123 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1124 invalid |= NFS_INO_INVALID_ATTR;
1126 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1127 invalid |= NFS_INO_INVALID_ATIME;
1130 nfsi->cache_validity |= invalid;
1132 nfsi->read_cache_jiffies = fattr->time_start;
1136 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1138 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1140 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1143 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1145 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1147 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1150 static atomic_long_t nfs_attr_generation_counter;
1152 static unsigned long nfs_read_attr_generation_counter(void)
1154 return atomic_long_read(&nfs_attr_generation_counter);
1157 unsigned long nfs_inc_attr_generation_counter(void)
1159 return atomic_long_inc_return(&nfs_attr_generation_counter);
1162 void nfs_fattr_init(struct nfs_fattr *fattr)
1165 fattr->time_start = jiffies;
1166 fattr->gencount = nfs_inc_attr_generation_counter();
1167 fattr->owner_name = NULL;
1168 fattr->group_name = NULL;
1170 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1172 struct nfs_fattr *nfs_alloc_fattr(void)
1174 struct nfs_fattr *fattr;
1176 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1178 nfs_fattr_init(fattr);
1181 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1183 struct nfs_fh *nfs_alloc_fhandle(void)
1187 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1192 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1196 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1197 * in the same way that wireshark does
1201 * For debugging only.
1203 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1205 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1206 * not on the result */
1207 return nfs_fhandle_hash(fh);
1209 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1212 * _nfs_display_fhandle - display an NFS file handle on the console
1214 * @fh: file handle to display
1215 * @caption: display caption
1217 * For debugging only.
1219 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1223 if (fh == NULL || fh->size == 0) {
1224 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1228 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1229 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1230 for (i = 0; i < fh->size; i += 16) {
1231 __be32 *pos = (__be32 *)&fh->data[i];
1233 switch ((fh->size - i - 1) >> 2) {
1235 printk(KERN_DEFAULT " %08x\n",
1239 printk(KERN_DEFAULT " %08x %08x\n",
1240 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1243 printk(KERN_DEFAULT " %08x %08x %08x\n",
1244 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1245 be32_to_cpup(pos + 2));
1248 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1249 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1250 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1254 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1258 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1259 * @inode - pointer to inode
1260 * @fattr - attributes
1262 * Attempt to divine whether or not an RPC call reply carrying stale
1263 * attributes got scheduled after another call carrying updated ones.
1265 * To do so, the function first assumes that a more recent ctime means
1266 * that the attributes in fattr are newer, however it also attempt to
1267 * catch the case where ctime either didn't change, or went backwards
1268 * (if someone reset the clock on the server) by looking at whether
1269 * or not this RPC call was started after the inode was last updated.
1270 * Note also the check for wraparound of 'attr_gencount'
1272 * The function returns 'true' if it thinks the attributes in 'fattr' are
1273 * more recent than the ones cached in the inode.
1276 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1278 const struct nfs_inode *nfsi = NFS_I(inode);
1280 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1281 nfs_ctime_need_update(inode, fattr) ||
1282 nfs_size_need_update(inode, fattr) ||
1283 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1287 * Don't trust the change_attribute, mtime, ctime or size if
1288 * a pnfs LAYOUTCOMMIT is outstanding
1290 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1291 struct nfs_fattr *fattr)
1293 if (pnfs_layoutcommit_outstanding(inode))
1294 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1295 NFS_ATTR_FATTR_MTIME |
1296 NFS_ATTR_FATTR_CTIME |
1297 NFS_ATTR_FATTR_SIZE);
1300 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1304 trace_nfs_refresh_inode_enter(inode);
1306 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1308 if (nfs_inode_attrs_need_update(inode, fattr))
1309 ret = nfs_update_inode(inode, fattr);
1311 ret = nfs_check_inode_attributes(inode, fattr);
1313 trace_nfs_refresh_inode_exit(inode, ret);
1318 * nfs_refresh_inode - try to update the inode attribute cache
1319 * @inode - pointer to inode
1320 * @fattr - updated attributes
1322 * Check that an RPC call that returned attributes has not overlapped with
1323 * other recent updates of the inode metadata, then decide whether it is
1324 * safe to do a full update of the inode attributes, or whether just to
1325 * call nfs_check_inode_attributes.
1327 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1331 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1333 spin_lock(&inode->i_lock);
1334 status = nfs_refresh_inode_locked(inode, fattr);
1335 spin_unlock(&inode->i_lock);
1339 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1341 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1343 struct nfs_inode *nfsi = NFS_I(inode);
1345 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1346 if (S_ISDIR(inode->i_mode)) {
1347 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1348 nfs_fscache_invalidate(inode);
1350 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1352 return nfs_refresh_inode_locked(inode, fattr);
1356 * nfs_post_op_update_inode - try to update the inode attribute cache
1357 * @inode - pointer to inode
1358 * @fattr - updated attributes
1360 * After an operation that has changed the inode metadata, mark the
1361 * attribute cache as being invalid, then try to update it.
1363 * NB: if the server didn't return any post op attributes, this
1364 * function will force the retrieval of attributes before the next
1365 * NFS request. Thus it should be used only for operations that
1366 * are expected to change one or more attributes, to avoid
1367 * unnecessary NFS requests and trips through nfs_update_inode().
1369 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1373 spin_lock(&inode->i_lock);
1374 status = nfs_post_op_update_inode_locked(inode, fattr);
1375 spin_unlock(&inode->i_lock);
1379 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1382 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1383 * @inode - pointer to inode
1384 * @fattr - updated attributes
1386 * After an operation that has changed the inode metadata, mark the
1387 * attribute cache as being invalid, then try to update it. Fake up
1388 * weak cache consistency data, if none exist.
1390 * This function is mainly designed to be used by the ->write_done() functions.
1392 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1396 spin_lock(&inode->i_lock);
1397 /* Don't do a WCC update if these attributes are already stale */
1398 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1399 !nfs_inode_attrs_need_update(inode, fattr)) {
1400 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1401 | NFS_ATTR_FATTR_PRESIZE
1402 | NFS_ATTR_FATTR_PREMTIME
1403 | NFS_ATTR_FATTR_PRECTIME);
1406 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1407 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1408 fattr->pre_change_attr = inode->i_version;
1409 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1411 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1412 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1413 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1414 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1416 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1417 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1418 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1419 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1421 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1422 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1423 fattr->pre_size = i_size_read(inode);
1424 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1427 status = nfs_post_op_update_inode_locked(inode, fattr);
1428 spin_unlock(&inode->i_lock);
1431 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1434 * Many nfs protocol calls return the new file attributes after
1435 * an operation. Here we update the inode to reflect the state
1436 * of the server's inode.
1438 * This is a bit tricky because we have to make sure all dirty pages
1439 * have been sent off to the server before calling invalidate_inode_pages.
1440 * To make sure no other process adds more write requests while we try
1441 * our best to flush them, we make them sleep during the attribute refresh.
1443 * A very similar scenario holds for the dir cache.
1445 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1447 struct nfs_server *server;
1448 struct nfs_inode *nfsi = NFS_I(inode);
1449 loff_t cur_isize, new_isize;
1450 unsigned long invalid = 0;
1451 unsigned long now = jiffies;
1452 unsigned long save_cache_validity;
1454 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1455 __func__, inode->i_sb->s_id, inode->i_ino,
1456 nfs_display_fhandle_hash(NFS_FH(inode)),
1457 atomic_read(&inode->i_count), fattr->valid);
1459 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1460 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1461 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1462 NFS_SERVER(inode)->nfs_client->cl_hostname,
1463 inode->i_sb->s_id, (long long)nfsi->fileid,
1464 (long long)fattr->fileid);
1469 * Make sure the inode's type hasn't changed.
1471 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1473 * Big trouble! The inode has become a different object.
1475 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1476 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1480 server = NFS_SERVER(inode);
1481 /* Update the fsid? */
1482 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1483 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1484 !IS_AUTOMOUNT(inode))
1485 server->fsid = fattr->fsid;
1488 * Update the read time so we don't revalidate too often.
1490 nfsi->read_cache_jiffies = fattr->time_start;
1492 save_cache_validity = nfsi->cache_validity;
1493 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1494 | NFS_INO_INVALID_ATIME
1495 | NFS_INO_REVAL_FORCED
1496 | NFS_INO_REVAL_PAGECACHE);
1498 /* Do atomic weak cache consistency updates */
1499 invalid |= nfs_wcc_update_inode(inode, fattr);
1501 /* More cache consistency checks */
1502 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1503 if (inode->i_version != fattr->change_attr) {
1504 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1505 inode->i_sb->s_id, inode->i_ino);
1506 invalid |= NFS_INO_INVALID_ATTR
1507 | NFS_INO_INVALID_DATA
1508 | NFS_INO_INVALID_ACCESS
1509 | NFS_INO_INVALID_ACL
1510 | NFS_INO_REVAL_PAGECACHE;
1511 if (S_ISDIR(inode->i_mode))
1512 nfs_force_lookup_revalidate(inode);
1513 inode->i_version = fattr->change_attr;
1515 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1516 invalid |= save_cache_validity;
1518 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1519 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1520 } else if (server->caps & NFS_CAP_MTIME)
1521 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1522 | NFS_INO_REVAL_FORCED);
1524 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1525 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1526 } else if (server->caps & NFS_CAP_CTIME)
1527 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1528 | NFS_INO_REVAL_FORCED);
1530 /* Check if our cached file size is stale */
1531 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1532 new_isize = nfs_size_to_loff_t(fattr->size);
1533 cur_isize = i_size_read(inode);
1534 if (new_isize != cur_isize) {
1535 /* Do we perhaps have any outstanding writes, or has
1536 * the file grown beyond our last write? */
1537 if ((nfsi->npages == 0) || new_isize > cur_isize) {
1538 i_size_write(inode, new_isize);
1539 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1541 dprintk("NFS: isize change on server for file %s/%ld "
1545 (long long)cur_isize,
1546 (long long)new_isize);
1549 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1550 | NFS_INO_REVAL_PAGECACHE
1551 | NFS_INO_REVAL_FORCED);
1554 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1555 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1556 else if (server->caps & NFS_CAP_ATIME)
1557 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1558 | NFS_INO_REVAL_FORCED);
1560 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1561 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1562 umode_t newmode = inode->i_mode & S_IFMT;
1563 newmode |= fattr->mode & S_IALLUGO;
1564 inode->i_mode = newmode;
1565 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1567 } else if (server->caps & NFS_CAP_MODE)
1568 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1569 | NFS_INO_INVALID_ACCESS
1570 | NFS_INO_INVALID_ACL
1571 | NFS_INO_REVAL_FORCED);
1573 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1574 if (!uid_eq(inode->i_uid, fattr->uid)) {
1575 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1576 inode->i_uid = fattr->uid;
1578 } else if (server->caps & NFS_CAP_OWNER)
1579 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1580 | NFS_INO_INVALID_ACCESS
1581 | NFS_INO_INVALID_ACL
1582 | NFS_INO_REVAL_FORCED);
1584 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1585 if (!gid_eq(inode->i_gid, fattr->gid)) {
1586 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1587 inode->i_gid = fattr->gid;
1589 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1590 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1591 | NFS_INO_INVALID_ACCESS
1592 | NFS_INO_INVALID_ACL
1593 | NFS_INO_REVAL_FORCED);
1595 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1596 if (inode->i_nlink != fattr->nlink) {
1597 invalid |= NFS_INO_INVALID_ATTR;
1598 if (S_ISDIR(inode->i_mode))
1599 invalid |= NFS_INO_INVALID_DATA;
1600 set_nlink(inode, fattr->nlink);
1602 } else if (server->caps & NFS_CAP_NLINK)
1603 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1604 | NFS_INO_REVAL_FORCED);
1606 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1608 * report the blocks in 512byte units
1610 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1612 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1613 inode->i_blocks = fattr->du.nfs2.blocks;
1615 /* Update attrtimeo value if we're out of the unstable period */
1616 if (invalid & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) {
1617 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1618 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1619 nfsi->attrtimeo_timestamp = now;
1620 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1622 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1623 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1624 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1625 nfsi->attrtimeo_timestamp = now;
1628 invalid &= ~NFS_INO_INVALID_ATTR;
1629 invalid &= ~NFS_INO_INVALID_LABEL;
1630 /* Don't invalidate the data if we were to blame */
1631 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1632 || S_ISLNK(inode->i_mode)))
1633 invalid &= ~NFS_INO_INVALID_DATA;
1634 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1635 (save_cache_validity & NFS_INO_REVAL_FORCED))
1636 nfsi->cache_validity |= invalid;
1638 if (invalid & NFS_INO_INVALID_DATA)
1639 nfs_fscache_invalidate(inode);
1644 * No need to worry about unhashing the dentry, as the
1645 * lookup validation will know that the inode is bad.
1646 * (But we fall through to invalidate the caches.)
1648 nfs_invalidate_inode(inode);
1652 struct inode *nfs_alloc_inode(struct super_block *sb)
1654 struct nfs_inode *nfsi;
1655 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1659 nfsi->cache_validity = 0UL;
1660 #if IS_ENABLED(CONFIG_NFS_V4)
1661 nfsi->nfs4_acl = NULL;
1662 #endif /* CONFIG_NFS_V4 */
1663 return &nfsi->vfs_inode;
1665 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1667 static void nfs_i_callback(struct rcu_head *head)
1669 struct inode *inode = container_of(head, struct inode, i_rcu);
1670 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1673 void nfs_destroy_inode(struct inode *inode)
1675 call_rcu(&inode->i_rcu, nfs_i_callback);
1677 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1679 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1681 #if IS_ENABLED(CONFIG_NFS_V4)
1682 INIT_LIST_HEAD(&nfsi->open_states);
1683 nfsi->delegation = NULL;
1684 nfsi->delegation_state = 0;
1685 init_rwsem(&nfsi->rwsem);
1686 nfsi->layout = NULL;
1690 static void init_once(void *foo)
1692 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1694 inode_init_once(&nfsi->vfs_inode);
1695 INIT_LIST_HEAD(&nfsi->open_files);
1696 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1697 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1698 INIT_LIST_HEAD(&nfsi->commit_info.list);
1700 nfsi->commit_info.ncommit = 0;
1701 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1702 atomic_set(&nfsi->silly_count, 1);
1703 INIT_HLIST_HEAD(&nfsi->silly_list);
1704 init_waitqueue_head(&nfsi->waitqueue);
1705 nfs4_init_once(nfsi);
1708 static int __init nfs_init_inodecache(void)
1710 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1711 sizeof(struct nfs_inode),
1712 0, (SLAB_RECLAIM_ACCOUNT|
1715 if (nfs_inode_cachep == NULL)
1721 static void nfs_destroy_inodecache(void)
1724 * Make sure all delayed rcu free inodes are flushed before we
1728 kmem_cache_destroy(nfs_inode_cachep);
1731 struct workqueue_struct *nfsiod_workqueue;
1732 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1735 * start up the nfsiod workqueue
1737 static int nfsiod_start(void)
1739 struct workqueue_struct *wq;
1740 dprintk("RPC: creating workqueue nfsiod\n");
1741 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1744 nfsiod_workqueue = wq;
1749 * Destroy the nfsiod workqueue
1751 static void nfsiod_stop(void)
1753 struct workqueue_struct *wq;
1755 wq = nfsiod_workqueue;
1758 nfsiod_workqueue = NULL;
1759 destroy_workqueue(wq);
1763 EXPORT_SYMBOL_GPL(nfs_net_id);
1765 static int nfs_net_init(struct net *net)
1767 nfs_clients_init(net);
1771 static void nfs_net_exit(struct net *net)
1773 nfs_cleanup_cb_ident_idr(net);
1776 static struct pernet_operations nfs_net_ops = {
1777 .init = nfs_net_init,
1778 .exit = nfs_net_exit,
1780 .size = sizeof(struct nfs_net),
1786 static int __init init_nfs_fs(void)
1790 err = register_pernet_subsys(&nfs_net_ops);
1794 err = nfs_fscache_register();
1798 err = nfsiod_start();
1802 err = nfs_fs_proc_init();
1806 err = nfs_init_nfspagecache();
1810 err = nfs_init_inodecache();
1814 err = nfs_init_readpagecache();
1818 err = nfs_init_writepagecache();
1822 err = nfs_init_directcache();
1826 #ifdef CONFIG_PROC_FS
1827 rpc_proc_register(&init_net, &nfs_rpcstat);
1829 if ((err = register_nfs_fs()) != 0)
1834 #ifdef CONFIG_PROC_FS
1835 rpc_proc_unregister(&init_net, "nfs");
1837 nfs_destroy_directcache();
1839 nfs_destroy_writepagecache();
1841 nfs_destroy_readpagecache();
1843 nfs_destroy_inodecache();
1845 nfs_destroy_nfspagecache();
1851 nfs_fscache_unregister();
1853 unregister_pernet_subsys(&nfs_net_ops);
1858 static void __exit exit_nfs_fs(void)
1860 nfs_destroy_directcache();
1861 nfs_destroy_writepagecache();
1862 nfs_destroy_readpagecache();
1863 nfs_destroy_inodecache();
1864 nfs_destroy_nfspagecache();
1865 nfs_fscache_unregister();
1866 unregister_pernet_subsys(&nfs_net_ops);
1867 #ifdef CONFIG_PROC_FS
1868 rpc_proc_unregister(&init_net, "nfs");
1870 unregister_nfs_fs();
1875 /* Not quite true; I just maintain it */
1876 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1877 MODULE_LICENSE("GPL");
1878 module_param(enable_ino64, bool, 0644);
1880 module_init(init_nfs_fs)
1881 module_exit(exit_nfs_fs)