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);
136 int nfs_sync_inode(struct inode *inode)
138 nfs_inode_dio_wait(inode);
139 return nfs_wb_all(inode);
141 EXPORT_SYMBOL_GPL(nfs_sync_inode);
144 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
146 int nfs_sync_mapping(struct address_space *mapping)
150 if (mapping->nrpages != 0) {
151 unmap_mapping_range(mapping, 0, 0, 0);
152 ret = nfs_wb_all(mapping->host);
157 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
159 struct nfs_inode *nfsi = NFS_I(inode);
161 if (inode->i_mapping->nrpages == 0)
162 flags &= ~NFS_INO_INVALID_DATA;
163 nfsi->cache_validity |= flags;
164 if (flags & NFS_INO_INVALID_DATA)
165 nfs_fscache_invalidate(inode);
169 * Invalidate the local caches
171 static void nfs_zap_caches_locked(struct inode *inode)
173 struct nfs_inode *nfsi = NFS_I(inode);
174 int mode = inode->i_mode;
176 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
178 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
179 nfsi->attrtimeo_timestamp = jiffies;
181 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
182 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
183 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
184 | NFS_INO_INVALID_DATA
185 | NFS_INO_INVALID_ACCESS
186 | NFS_INO_INVALID_ACL
187 | NFS_INO_REVAL_PAGECACHE);
189 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
190 | NFS_INO_INVALID_ACCESS
191 | NFS_INO_INVALID_ACL
192 | NFS_INO_REVAL_PAGECACHE);
193 nfs_zap_label_cache_locked(nfsi);
196 void nfs_zap_caches(struct inode *inode)
198 spin_lock(&inode->i_lock);
199 nfs_zap_caches_locked(inode);
200 spin_unlock(&inode->i_lock);
202 EXPORT_SYMBOL_GPL(nfs_zap_caches);
204 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
206 if (mapping->nrpages != 0) {
207 spin_lock(&inode->i_lock);
208 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
209 spin_unlock(&inode->i_lock);
213 void nfs_zap_acl_cache(struct inode *inode)
215 void (*clear_acl_cache)(struct inode *);
217 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
218 if (clear_acl_cache != NULL)
219 clear_acl_cache(inode);
220 spin_lock(&inode->i_lock);
221 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
222 spin_unlock(&inode->i_lock);
224 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
226 void nfs_invalidate_atime(struct inode *inode)
228 spin_lock(&inode->i_lock);
229 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
230 spin_unlock(&inode->i_lock);
232 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
235 * Invalidate, but do not unhash, the inode.
236 * NB: must be called with inode->i_lock held!
238 static void nfs_invalidate_inode(struct inode *inode)
240 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
241 nfs_zap_caches_locked(inode);
244 struct nfs_find_desc {
246 struct nfs_fattr *fattr;
250 * In NFSv3 we can have 64bit inode numbers. In order to support
251 * this, and re-exported directories (also seen in NFSv2)
252 * we are forced to allow 2 different inodes to have the same
256 nfs_find_actor(struct inode *inode, void *opaque)
258 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
259 struct nfs_fh *fh = desc->fh;
260 struct nfs_fattr *fattr = desc->fattr;
262 if (NFS_FILEID(inode) != fattr->fileid)
264 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
266 if (nfs_compare_fh(NFS_FH(inode), fh))
268 if (is_bad_inode(inode) || NFS_STALE(inode))
274 nfs_init_locked(struct inode *inode, void *opaque)
276 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
277 struct nfs_fattr *fattr = desc->fattr;
279 set_nfs_fileid(inode, fattr->fileid);
280 nfs_copy_fh(NFS_FH(inode), desc->fh);
284 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
285 static void nfs_clear_label_invalid(struct inode *inode)
287 spin_lock(&inode->i_lock);
288 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
289 spin_unlock(&inode->i_lock);
292 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
293 struct nfs4_label *label)
300 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
301 error = security_inode_notifysecctx(inode, label->label,
304 printk(KERN_ERR "%s() %s %d "
305 "security_inode_notifysecctx() %d\n",
307 (char *)label->label,
309 nfs_clear_label_invalid(inode);
313 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
315 struct nfs4_label *label = NULL;
316 int minor_version = server->nfs_client->cl_minorversion;
318 if (minor_version < 2)
321 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
324 label = kzalloc(sizeof(struct nfs4_label), flags);
326 return ERR_PTR(-ENOMEM);
328 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
329 if (label->label == NULL) {
331 return ERR_PTR(-ENOMEM);
333 label->len = NFS4_MAXLABELLEN;
337 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
339 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
340 struct nfs4_label *label)
344 EXPORT_SYMBOL_GPL(nfs_setsecurity);
347 * This is our front-end to iget that looks up inodes by file handle
348 * instead of inode number.
351 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
353 struct nfs_find_desc desc = {
357 struct inode *inode = ERR_PTR(-ENOENT);
360 nfs_attr_check_mountpoint(sb, fattr);
362 if (nfs_attr_use_mounted_on_fileid(fattr))
363 fattr->fileid = fattr->mounted_on_fileid;
364 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
366 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
369 hash = nfs_fattr_to_ino_t(fattr);
371 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
373 inode = ERR_PTR(-ENOMEM);
377 if (inode->i_state & I_NEW) {
378 struct nfs_inode *nfsi = NFS_I(inode);
379 unsigned long now = jiffies;
381 /* We set i_ino for the few things that still rely on it,
385 /* We can't support update_atime(), since the server will reset it */
386 inode->i_flags |= S_NOATIME|S_NOCMTIME;
387 inode->i_mode = fattr->mode;
388 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
389 && nfs_server_capable(inode, NFS_CAP_MODE))
390 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
391 /* Why so? Because we want revalidate for devices/FIFOs, and
392 * that's precisely what we have in nfs_file_inode_operations.
394 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
395 if (S_ISREG(inode->i_mode)) {
396 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
397 inode->i_data.a_ops = &nfs_file_aops;
398 } else if (S_ISDIR(inode->i_mode)) {
399 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
400 inode->i_fop = &nfs_dir_operations;
401 inode->i_data.a_ops = &nfs_dir_aops;
402 /* Deal with crossing mountpoints */
403 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
404 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
405 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
406 inode->i_op = &nfs_referral_inode_operations;
408 inode->i_op = &nfs_mountpoint_inode_operations;
410 inode->i_flags |= S_AUTOMOUNT;
412 } else if (S_ISLNK(inode->i_mode))
413 inode->i_op = &nfs_symlink_inode_operations;
415 init_special_inode(inode, inode->i_mode, fattr->rdev);
417 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
418 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
419 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
420 inode->i_version = 0;
423 inode->i_uid = make_kuid(&init_user_ns, -2);
424 inode->i_gid = make_kgid(&init_user_ns, -2);
426 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
430 nfsi->read_cache_jiffies = fattr->time_start;
431 nfsi->attr_gencount = fattr->gencount;
432 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
433 inode->i_atime = fattr->atime;
434 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
435 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
436 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
437 inode->i_mtime = fattr->mtime;
438 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
439 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
440 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
441 inode->i_ctime = fattr->ctime;
442 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
443 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
444 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
445 inode->i_version = fattr->change_attr;
446 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
447 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
448 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
449 inode->i_size = nfs_size_to_loff_t(fattr->size);
451 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
452 | NFS_INO_REVAL_PAGECACHE);
453 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
454 set_nlink(inode, fattr->nlink);
455 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
456 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
457 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
458 inode->i_uid = fattr->uid;
459 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
460 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
461 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
462 inode->i_gid = fattr->gid;
463 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
464 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
465 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
466 inode->i_blocks = fattr->du.nfs2.blocks;
467 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
469 * report the blocks in 512byte units
471 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
474 nfs_setsecurity(inode, fattr, label);
476 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
477 nfsi->attrtimeo_timestamp = now;
478 nfsi->access_cache = RB_ROOT;
480 nfs_fscache_init_inode(inode);
482 unlock_new_inode(inode);
484 nfs_refresh_inode(inode, fattr);
485 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
487 (unsigned long long)NFS_FILEID(inode),
488 nfs_display_fhandle_hash(fh),
489 atomic_read(&inode->i_count));
495 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
498 EXPORT_SYMBOL_GPL(nfs_fhget);
500 #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)
503 nfs_setattr(struct dentry *dentry, struct iattr *attr)
505 struct inode *inode = dentry->d_inode;
506 struct nfs_fattr *fattr;
509 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
511 /* skip mode change if it's just for clearing setuid/setgid */
512 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
513 attr->ia_valid &= ~ATTR_MODE;
515 if (attr->ia_valid & ATTR_SIZE) {
518 BUG_ON(!S_ISREG(inode->i_mode));
520 i_size = i_size_read(inode);
521 if (attr->ia_size == i_size)
522 attr->ia_valid &= ~ATTR_SIZE;
523 else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
527 /* Optimization: if the end result is no change, don't RPC */
528 attr->ia_valid &= NFS_VALID_ATTRS;
529 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
532 trace_nfs_setattr_enter(inode);
534 /* Write all dirty data */
535 if (S_ISREG(inode->i_mode))
536 nfs_sync_inode(inode);
538 fattr = nfs_alloc_fattr();
542 * Return any delegations if we're going to change ACLs
544 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
545 NFS_PROTO(inode)->return_delegation(inode);
546 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
548 error = nfs_refresh_inode(inode, fattr);
549 nfs_free_fattr(fattr);
551 trace_nfs_setattr_exit(inode, error);
554 EXPORT_SYMBOL_GPL(nfs_setattr);
557 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
558 * @inode: inode of the file used
559 * @offset: file offset to start truncating
561 * This is a copy of the common vmtruncate, but with the locking
562 * corrected to take into account the fact that NFS requires
563 * inode->i_size to be updated under the inode->i_lock.
564 * Note: must be called with inode->i_lock held!
566 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
570 err = inode_newsize_ok(inode, offset);
574 i_size_write(inode, offset);
577 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
579 spin_unlock(&inode->i_lock);
580 truncate_pagecache(inode, offset);
581 spin_lock(&inode->i_lock);
587 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
588 * @inode: pointer to struct inode
589 * @attr: pointer to struct iattr
591 * Note: we do this in the *proc.c in order to ensure that
592 * it works for things like exclusive creates too.
594 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
595 struct nfs_fattr *fattr)
597 /* Barrier: bump the attribute generation count. */
598 nfs_fattr_set_barrier(fattr);
600 spin_lock(&inode->i_lock);
601 NFS_I(inode)->attr_gencount = fattr->gencount;
602 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
603 if ((attr->ia_valid & ATTR_MODE) != 0) {
604 int mode = attr->ia_mode & S_IALLUGO;
605 mode |= inode->i_mode & ~S_IALLUGO;
606 inode->i_mode = mode;
608 if ((attr->ia_valid & ATTR_UID) != 0)
609 inode->i_uid = attr->ia_uid;
610 if ((attr->ia_valid & ATTR_GID) != 0)
611 inode->i_gid = attr->ia_gid;
612 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
613 | NFS_INO_INVALID_ACL);
615 if ((attr->ia_valid & ATTR_SIZE) != 0) {
616 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
617 nfs_vmtruncate(inode, attr->ia_size);
619 nfs_update_inode(inode, fattr);
620 spin_unlock(&inode->i_lock);
622 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
624 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
626 struct dentry *parent;
628 parent = dget_parent(dentry);
629 nfs_force_use_readdirplus(parent->d_inode);
633 static bool nfs_need_revalidate_inode(struct inode *inode)
635 if (NFS_I(inode)->cache_validity &
636 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
638 if (nfs_attribute_cache_expired(inode))
643 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
645 struct inode *inode = dentry->d_inode;
646 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
649 trace_nfs_getattr_enter(inode);
650 /* Flush out writes to the server in order to update c/mtime. */
651 if (S_ISREG(inode->i_mode)) {
652 mutex_lock(&inode->i_mutex);
653 err = nfs_sync_inode(inode);
654 mutex_unlock(&inode->i_mutex);
660 * We may force a getattr if the user cares about atime.
662 * Note that we only have to check the vfsmount flags here:
663 * - NFS always sets S_NOATIME by so checking it would give a
665 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
666 * no point in checking those.
668 if ((mnt->mnt_flags & MNT_NOATIME) ||
669 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
672 if (need_atime || nfs_need_revalidate_inode(inode)) {
673 struct nfs_server *server = NFS_SERVER(inode);
675 if (server->caps & NFS_CAP_READDIRPLUS)
676 nfs_request_parent_use_readdirplus(dentry);
677 err = __nfs_revalidate_inode(server, inode);
680 generic_fillattr(inode, stat);
681 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
684 trace_nfs_getattr_exit(inode, err);
687 EXPORT_SYMBOL_GPL(nfs_getattr);
689 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
691 atomic_set(&l_ctx->count, 1);
692 l_ctx->lockowner.l_owner = current->files;
693 l_ctx->lockowner.l_pid = current->tgid;
694 INIT_LIST_HEAD(&l_ctx->list);
695 nfs_iocounter_init(&l_ctx->io_count);
698 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
700 struct nfs_lock_context *head = &ctx->lock_context;
701 struct nfs_lock_context *pos = head;
704 if (pos->lockowner.l_owner != current->files)
706 if (pos->lockowner.l_pid != current->tgid)
708 atomic_inc(&pos->count);
710 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
714 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
716 struct nfs_lock_context *res, *new = NULL;
717 struct inode *inode = ctx->dentry->d_inode;
719 spin_lock(&inode->i_lock);
720 res = __nfs_find_lock_context(ctx);
722 spin_unlock(&inode->i_lock);
723 new = kmalloc(sizeof(*new), GFP_KERNEL);
725 return ERR_PTR(-ENOMEM);
726 nfs_init_lock_context(new);
727 spin_lock(&inode->i_lock);
728 res = __nfs_find_lock_context(ctx);
730 list_add_tail(&new->list, &ctx->lock_context.list);
731 new->open_context = ctx;
736 spin_unlock(&inode->i_lock);
740 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
742 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
744 struct nfs_open_context *ctx = l_ctx->open_context;
745 struct inode *inode = ctx->dentry->d_inode;
747 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
749 list_del(&l_ctx->list);
750 spin_unlock(&inode->i_lock);
753 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
756 * nfs_close_context - Common close_context() routine NFSv2/v3
757 * @ctx: pointer to context
758 * @is_sync: is this a synchronous close
760 * always ensure that the attributes are up to date if we're mounted
761 * with close-to-open semantics
763 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
766 struct nfs_server *server;
768 if (!(ctx->mode & FMODE_WRITE))
772 inode = ctx->dentry->d_inode;
773 if (!list_empty(&NFS_I(inode)->open_files))
775 server = NFS_SERVER(inode);
776 if (server->flags & NFS_MOUNT_NOCTO)
778 nfs_revalidate_inode(server, inode);
780 EXPORT_SYMBOL_GPL(nfs_close_context);
782 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
784 struct nfs_open_context *ctx;
785 struct rpc_cred *cred = rpc_lookup_cred();
787 return ERR_CAST(cred);
789 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
792 return ERR_PTR(-ENOMEM);
794 nfs_sb_active(dentry->d_sb);
795 ctx->dentry = dget(dentry);
801 nfs_init_lock_context(&ctx->lock_context);
802 ctx->lock_context.open_context = ctx;
803 INIT_LIST_HEAD(&ctx->list);
804 ctx->mdsthreshold = NULL;
807 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
809 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
812 atomic_inc(&ctx->lock_context.count);
815 EXPORT_SYMBOL_GPL(get_nfs_open_context);
817 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
819 struct inode *inode = ctx->dentry->d_inode;
820 struct super_block *sb = ctx->dentry->d_sb;
822 if (!list_empty(&ctx->list)) {
823 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
825 list_del(&ctx->list);
826 spin_unlock(&inode->i_lock);
827 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
830 NFS_PROTO(inode)->close_context(ctx, is_sync);
831 if (ctx->cred != NULL)
832 put_rpccred(ctx->cred);
835 kfree(ctx->mdsthreshold);
839 void put_nfs_open_context(struct nfs_open_context *ctx)
841 __put_nfs_open_context(ctx, 0);
843 EXPORT_SYMBOL_GPL(put_nfs_open_context);
846 * Ensure that mmap has a recent RPC credential for use when writing out
849 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
851 struct inode *inode = ctx->dentry->d_inode;
852 struct nfs_inode *nfsi = NFS_I(inode);
854 spin_lock(&inode->i_lock);
855 list_add(&ctx->list, &nfsi->open_files);
856 spin_unlock(&inode->i_lock);
858 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
860 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
862 filp->private_data = get_nfs_open_context(ctx);
863 if (list_empty(&ctx->list))
864 nfs_inode_attach_open_context(ctx);
866 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
869 * Given an inode, search for an open context with the desired characteristics
871 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
873 struct nfs_inode *nfsi = NFS_I(inode);
874 struct nfs_open_context *pos, *ctx = NULL;
876 spin_lock(&inode->i_lock);
877 list_for_each_entry(pos, &nfsi->open_files, list) {
878 if (cred != NULL && pos->cred != cred)
880 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
882 ctx = get_nfs_open_context(pos);
885 spin_unlock(&inode->i_lock);
889 static void nfs_file_clear_open_context(struct file *filp)
891 struct nfs_open_context *ctx = nfs_file_open_context(filp);
894 struct inode *inode = ctx->dentry->d_inode;
896 filp->private_data = NULL;
897 spin_lock(&inode->i_lock);
898 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
899 spin_unlock(&inode->i_lock);
900 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
905 * These allocate and release file read/write context information.
907 int nfs_open(struct inode *inode, struct file *filp)
909 struct nfs_open_context *ctx;
911 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
914 nfs_file_set_open_context(filp, ctx);
915 put_nfs_open_context(ctx);
916 nfs_fscache_open_file(inode, filp);
920 int nfs_release(struct inode *inode, struct file *filp)
922 nfs_file_clear_open_context(filp);
927 * This function is called whenever some part of NFS notices that
928 * the cached attributes have to be refreshed.
931 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
933 int status = -ESTALE;
934 struct nfs4_label *label = NULL;
935 struct nfs_fattr *fattr = NULL;
936 struct nfs_inode *nfsi = NFS_I(inode);
938 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
939 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
941 trace_nfs_revalidate_inode_enter(inode);
943 if (is_bad_inode(inode))
945 if (NFS_STALE(inode))
949 fattr = nfs_alloc_fattr();
953 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
955 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
957 status = PTR_ERR(label);
961 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
963 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
965 (unsigned long long)NFS_FILEID(inode), status);
966 if (status == -ESTALE) {
967 nfs_zap_caches(inode);
968 if (!S_ISDIR(inode->i_mode))
969 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
974 status = nfs_refresh_inode(inode, fattr);
976 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
978 (unsigned long long)NFS_FILEID(inode), status);
982 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
983 nfs_zap_acl_cache(inode);
985 nfs_setsecurity(inode, fattr, label);
987 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
989 (unsigned long long)NFS_FILEID(inode));
992 nfs4_label_free(label);
994 nfs_free_fattr(fattr);
995 trace_nfs_revalidate_inode_exit(inode, status);
999 int nfs_attribute_timeout(struct inode *inode)
1001 struct nfs_inode *nfsi = NFS_I(inode);
1003 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1006 int nfs_attribute_cache_expired(struct inode *inode)
1008 if (nfs_have_delegated_attributes(inode))
1010 return nfs_attribute_timeout(inode);
1014 * nfs_revalidate_inode - Revalidate the inode attributes
1015 * @server - pointer to nfs_server struct
1016 * @inode - pointer to inode struct
1018 * Updates inode attribute information by retrieving the data from the server.
1020 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1022 if (!nfs_need_revalidate_inode(inode))
1023 return NFS_STALE(inode) ? -ESTALE : 0;
1024 return __nfs_revalidate_inode(server, inode);
1026 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1028 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1030 if (!(NFS_I(inode)->cache_validity &
1031 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1032 && !nfs_attribute_cache_expired(inode))
1033 return NFS_STALE(inode) ? -ESTALE : 0;
1037 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1039 struct nfs_inode *nfsi = NFS_I(inode);
1042 if (mapping->nrpages != 0) {
1043 if (S_ISREG(inode->i_mode)) {
1044 unmap_mapping_range(mapping, 0, 0, 0);
1045 ret = nfs_sync_mapping(mapping);
1049 ret = invalidate_inode_pages2(mapping);
1053 if (S_ISDIR(inode->i_mode)) {
1054 spin_lock(&inode->i_lock);
1055 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1056 spin_unlock(&inode->i_lock);
1058 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1059 nfs_fscache_wait_on_invalidate(inode);
1061 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1063 (unsigned long long)NFS_FILEID(inode));
1067 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1069 if (nfs_have_delegated_attributes(inode))
1071 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1072 || nfs_attribute_timeout(inode)
1073 || NFS_STALE(inode);
1077 * __nfs_revalidate_mapping - Revalidate the pagecache
1078 * @inode - pointer to host inode
1079 * @mapping - pointer to mapping
1080 * @may_lock - take inode->i_mutex?
1082 static int __nfs_revalidate_mapping(struct inode *inode,
1083 struct address_space *mapping,
1086 struct nfs_inode *nfsi = NFS_I(inode);
1087 unsigned long *bitlock = &nfsi->flags;
1090 /* swapfiles are not supposed to be shared. */
1091 if (IS_SWAPFILE(inode))
1094 if (nfs_mapping_need_revalidate_inode(inode)) {
1095 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1101 * We must clear NFS_INO_INVALID_DATA first to ensure that
1102 * invalidations that come in while we're shooting down the mappings
1103 * are respected. But, that leaves a race window where one revalidator
1104 * can clear the flag, and then another checks it before the mapping
1105 * gets invalidated. Fix that by serializing access to this part of
1108 * At the same time, we need to allow other tasks to see whether we
1109 * might be in the middle of invalidating the pages, so we only set
1110 * the bit lock here if it looks like we're going to be doing that.
1113 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1114 nfs_wait_bit_killable, TASK_KILLABLE);
1117 spin_lock(&inode->i_lock);
1118 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1119 spin_unlock(&inode->i_lock);
1122 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1124 spin_unlock(&inode->i_lock);
1128 set_bit(NFS_INO_INVALIDATING, bitlock);
1130 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1131 spin_unlock(&inode->i_lock);
1132 trace_nfs_invalidate_mapping_enter(inode);
1134 mutex_lock(&inode->i_mutex);
1135 ret = nfs_invalidate_mapping(inode, mapping);
1136 mutex_unlock(&inode->i_mutex);
1138 ret = nfs_invalidate_mapping(inode, mapping);
1139 trace_nfs_invalidate_mapping_exit(inode, ret);
1141 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1142 smp_mb__after_atomic();
1143 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1149 * nfs_revalidate_mapping - Revalidate the pagecache
1150 * @inode - pointer to host inode
1151 * @mapping - pointer to mapping
1153 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1155 return __nfs_revalidate_mapping(inode, mapping, false);
1159 * nfs_revalidate_mapping_protected - Revalidate the pagecache
1160 * @inode - pointer to host inode
1161 * @mapping - pointer to mapping
1163 * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1164 * while invalidating the mapping.
1166 int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1168 return __nfs_revalidate_mapping(inode, mapping, true);
1171 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1173 struct nfs_inode *nfsi = NFS_I(inode);
1174 unsigned long ret = 0;
1176 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1177 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1178 && inode->i_version == fattr->pre_change_attr) {
1179 inode->i_version = fattr->change_attr;
1180 if (S_ISDIR(inode->i_mode))
1181 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1182 ret |= NFS_INO_INVALID_ATTR;
1184 /* If we have atomic WCC data, we may update some attributes */
1185 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1186 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1187 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1188 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1189 ret |= NFS_INO_INVALID_ATTR;
1192 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1193 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1194 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1195 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1196 if (S_ISDIR(inode->i_mode))
1197 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1198 ret |= NFS_INO_INVALID_ATTR;
1200 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1201 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1202 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1203 && nfsi->nrequests == 0) {
1204 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1205 ret |= NFS_INO_INVALID_ATTR;
1212 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1213 * @inode - pointer to inode
1214 * @fattr - updated attributes
1216 * Verifies the attribute cache. If we have just changed the attributes,
1217 * so that fattr carries weak cache consistency data, then it may
1218 * also update the ctime/mtime/change_attribute.
1220 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1222 struct nfs_inode *nfsi = NFS_I(inode);
1223 loff_t cur_size, new_isize;
1224 unsigned long invalid = 0;
1227 if (nfs_have_delegated_attributes(inode))
1229 /* Has the inode gone and changed behind our back? */
1230 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1232 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1235 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1236 inode->i_version != fattr->change_attr)
1237 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1239 /* Verify a few of the more important attributes */
1240 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1241 invalid |= NFS_INO_INVALID_ATTR;
1243 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1244 cur_size = i_size_read(inode);
1245 new_isize = nfs_size_to_loff_t(fattr->size);
1246 if (cur_size != new_isize && nfsi->nrequests == 0)
1247 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1250 /* Have any file permissions changed? */
1251 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1252 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1253 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1254 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1255 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1256 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1258 /* Has the link count changed? */
1259 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1260 invalid |= NFS_INO_INVALID_ATTR;
1262 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1263 invalid |= NFS_INO_INVALID_ATIME;
1266 nfs_set_cache_invalid(inode, invalid);
1268 nfsi->read_cache_jiffies = fattr->time_start;
1272 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1274 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1276 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1279 static atomic_long_t nfs_attr_generation_counter;
1281 static unsigned long nfs_read_attr_generation_counter(void)
1283 return atomic_long_read(&nfs_attr_generation_counter);
1286 unsigned long nfs_inc_attr_generation_counter(void)
1288 return atomic_long_inc_return(&nfs_attr_generation_counter);
1290 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1292 void nfs_fattr_init(struct nfs_fattr *fattr)
1295 fattr->time_start = jiffies;
1296 fattr->gencount = nfs_inc_attr_generation_counter();
1297 fattr->owner_name = NULL;
1298 fattr->group_name = NULL;
1300 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1303 * nfs_fattr_set_barrier
1304 * @fattr: attributes
1306 * Used to set a barrier after an attribute was updated. This
1307 * barrier ensures that older attributes from RPC calls that may
1308 * have raced with our update cannot clobber these new values.
1309 * Note that you are still responsible for ensuring that other
1310 * operations which change the attribute on the server do not
1313 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1315 fattr->gencount = nfs_inc_attr_generation_counter();
1318 struct nfs_fattr *nfs_alloc_fattr(void)
1320 struct nfs_fattr *fattr;
1322 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1324 nfs_fattr_init(fattr);
1327 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1329 struct nfs_fh *nfs_alloc_fhandle(void)
1333 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1338 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1342 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1343 * in the same way that wireshark does
1347 * For debugging only.
1349 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1351 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1352 * not on the result */
1353 return nfs_fhandle_hash(fh);
1355 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1358 * _nfs_display_fhandle - display an NFS file handle on the console
1360 * @fh: file handle to display
1361 * @caption: display caption
1363 * For debugging only.
1365 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1369 if (fh == NULL || fh->size == 0) {
1370 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1374 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1375 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1376 for (i = 0; i < fh->size; i += 16) {
1377 __be32 *pos = (__be32 *)&fh->data[i];
1379 switch ((fh->size - i - 1) >> 2) {
1381 printk(KERN_DEFAULT " %08x\n",
1385 printk(KERN_DEFAULT " %08x %08x\n",
1386 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1389 printk(KERN_DEFAULT " %08x %08x %08x\n",
1390 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1391 be32_to_cpup(pos + 2));
1394 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1395 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1396 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1400 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1404 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1405 * @inode - pointer to inode
1406 * @fattr - attributes
1408 * Attempt to divine whether or not an RPC call reply carrying stale
1409 * attributes got scheduled after another call carrying updated ones.
1411 * To do so, the function first assumes that a more recent ctime means
1412 * that the attributes in fattr are newer, however it also attempt to
1413 * catch the case where ctime either didn't change, or went backwards
1414 * (if someone reset the clock on the server) by looking at whether
1415 * or not this RPC call was started after the inode was last updated.
1416 * Note also the check for wraparound of 'attr_gencount'
1418 * The function returns 'true' if it thinks the attributes in 'fattr' are
1419 * more recent than the ones cached in the inode.
1422 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1424 const struct nfs_inode *nfsi = NFS_I(inode);
1426 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1427 nfs_ctime_need_update(inode, fattr) ||
1428 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1432 * Don't trust the change_attribute, mtime, ctime or size if
1433 * a pnfs LAYOUTCOMMIT is outstanding
1435 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1436 struct nfs_fattr *fattr)
1438 if (pnfs_layoutcommit_outstanding(inode))
1439 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1440 NFS_ATTR_FATTR_MTIME |
1441 NFS_ATTR_FATTR_CTIME |
1442 NFS_ATTR_FATTR_SIZE);
1445 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1449 trace_nfs_refresh_inode_enter(inode);
1451 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1453 if (nfs_inode_attrs_need_update(inode, fattr))
1454 ret = nfs_update_inode(inode, fattr);
1456 ret = nfs_check_inode_attributes(inode, fattr);
1458 trace_nfs_refresh_inode_exit(inode, ret);
1463 * nfs_refresh_inode - try to update the inode attribute cache
1464 * @inode - pointer to inode
1465 * @fattr - updated attributes
1467 * Check that an RPC call that returned attributes has not overlapped with
1468 * other recent updates of the inode metadata, then decide whether it is
1469 * safe to do a full update of the inode attributes, or whether just to
1470 * call nfs_check_inode_attributes.
1472 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1476 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1478 spin_lock(&inode->i_lock);
1479 status = nfs_refresh_inode_locked(inode, fattr);
1480 spin_unlock(&inode->i_lock);
1484 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1486 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1488 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1490 if (S_ISDIR(inode->i_mode))
1491 invalid |= NFS_INO_INVALID_DATA;
1492 nfs_set_cache_invalid(inode, invalid);
1493 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1495 return nfs_refresh_inode_locked(inode, fattr);
1499 * nfs_post_op_update_inode - try to update the inode attribute cache
1500 * @inode - pointer to inode
1501 * @fattr - updated attributes
1503 * After an operation that has changed the inode metadata, mark the
1504 * attribute cache as being invalid, then try to update it.
1506 * NB: if the server didn't return any post op attributes, this
1507 * function will force the retrieval of attributes before the next
1508 * NFS request. Thus it should be used only for operations that
1509 * are expected to change one or more attributes, to avoid
1510 * unnecessary NFS requests and trips through nfs_update_inode().
1512 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1516 spin_lock(&inode->i_lock);
1517 nfs_fattr_set_barrier(fattr);
1518 status = nfs_post_op_update_inode_locked(inode, fattr);
1519 spin_unlock(&inode->i_lock);
1523 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1526 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1527 * @inode - pointer to inode
1528 * @fattr - updated attributes
1530 * After an operation that has changed the inode metadata, mark the
1531 * attribute cache as being invalid, then try to update it. Fake up
1532 * weak cache consistency data, if none exist.
1534 * This function is mainly designed to be used by the ->write_done() functions.
1536 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1540 /* Don't do a WCC update if these attributes are already stale */
1541 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1542 !nfs_inode_attrs_need_update(inode, fattr)) {
1543 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1544 | NFS_ATTR_FATTR_PRESIZE
1545 | NFS_ATTR_FATTR_PREMTIME
1546 | NFS_ATTR_FATTR_PRECTIME);
1549 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1550 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1551 fattr->pre_change_attr = inode->i_version;
1552 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1554 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1555 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1556 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1557 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1559 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1560 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1561 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1562 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1564 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1565 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1566 fattr->pre_size = i_size_read(inode);
1567 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1570 status = nfs_post_op_update_inode_locked(inode, fattr);
1575 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1576 * @inode - pointer to inode
1577 * @fattr - updated attributes
1579 * After an operation that has changed the inode metadata, mark the
1580 * attribute cache as being invalid, then try to update it. Fake up
1581 * weak cache consistency data, if none exist.
1583 * This function is mainly designed to be used by the ->write_done() functions.
1585 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1589 spin_lock(&inode->i_lock);
1590 nfs_fattr_set_barrier(fattr);
1591 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1592 spin_unlock(&inode->i_lock);
1595 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1598 * Many nfs protocol calls return the new file attributes after
1599 * an operation. Here we update the inode to reflect the state
1600 * of the server's inode.
1602 * This is a bit tricky because we have to make sure all dirty pages
1603 * have been sent off to the server before calling invalidate_inode_pages.
1604 * To make sure no other process adds more write requests while we try
1605 * our best to flush them, we make them sleep during the attribute refresh.
1607 * A very similar scenario holds for the dir cache.
1609 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1611 struct nfs_server *server;
1612 struct nfs_inode *nfsi = NFS_I(inode);
1613 loff_t cur_isize, new_isize;
1614 unsigned long invalid = 0;
1615 unsigned long now = jiffies;
1616 unsigned long save_cache_validity;
1618 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1619 __func__, inode->i_sb->s_id, inode->i_ino,
1620 nfs_display_fhandle_hash(NFS_FH(inode)),
1621 atomic_read(&inode->i_count), fattr->valid);
1623 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1624 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1625 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1626 NFS_SERVER(inode)->nfs_client->cl_hostname,
1627 inode->i_sb->s_id, (long long)nfsi->fileid,
1628 (long long)fattr->fileid);
1633 * Make sure the inode's type hasn't changed.
1635 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1637 * Big trouble! The inode has become a different object.
1639 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1640 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1644 server = NFS_SERVER(inode);
1645 /* Update the fsid? */
1646 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1647 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1648 !IS_AUTOMOUNT(inode))
1649 server->fsid = fattr->fsid;
1652 * Update the read time so we don't revalidate too often.
1654 nfsi->read_cache_jiffies = fattr->time_start;
1656 save_cache_validity = nfsi->cache_validity;
1657 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1658 | NFS_INO_INVALID_ATIME
1659 | NFS_INO_REVAL_FORCED
1660 | NFS_INO_REVAL_PAGECACHE);
1662 /* Do atomic weak cache consistency updates */
1663 invalid |= nfs_wcc_update_inode(inode, fattr);
1665 /* More cache consistency checks */
1666 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1667 if (inode->i_version != fattr->change_attr) {
1668 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1669 inode->i_sb->s_id, inode->i_ino);
1670 invalid |= NFS_INO_INVALID_ATTR
1671 | NFS_INO_INVALID_DATA
1672 | NFS_INO_INVALID_ACCESS
1673 | NFS_INO_INVALID_ACL
1674 | NFS_INO_REVAL_PAGECACHE;
1675 if (S_ISDIR(inode->i_mode))
1676 nfs_force_lookup_revalidate(inode);
1677 inode->i_version = fattr->change_attr;
1679 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1680 nfsi->cache_validity |= save_cache_validity;
1682 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1683 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1684 } else if (server->caps & NFS_CAP_MTIME)
1685 nfsi->cache_validity |= save_cache_validity &
1686 (NFS_INO_INVALID_ATTR
1687 | NFS_INO_REVAL_FORCED);
1689 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1690 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1691 } else if (server->caps & NFS_CAP_CTIME)
1692 nfsi->cache_validity |= save_cache_validity &
1693 (NFS_INO_INVALID_ATTR
1694 | NFS_INO_REVAL_FORCED);
1696 /* Check if our cached file size is stale */
1697 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1698 new_isize = nfs_size_to_loff_t(fattr->size);
1699 cur_isize = i_size_read(inode);
1700 if (new_isize != cur_isize) {
1701 /* Do we perhaps have any outstanding writes, or has
1702 * the file grown beyond our last write? */
1703 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1704 i_size_write(inode, new_isize);
1705 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1706 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1708 dprintk("NFS: isize change on server for file %s/%ld "
1712 (long long)cur_isize,
1713 (long long)new_isize);
1716 nfsi->cache_validity |= save_cache_validity &
1717 (NFS_INO_INVALID_ATTR
1718 | NFS_INO_REVAL_PAGECACHE
1719 | NFS_INO_REVAL_FORCED);
1722 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1723 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1724 else if (server->caps & NFS_CAP_ATIME)
1725 nfsi->cache_validity |= save_cache_validity &
1726 (NFS_INO_INVALID_ATIME
1727 | NFS_INO_REVAL_FORCED);
1729 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1730 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1731 umode_t newmode = inode->i_mode & S_IFMT;
1732 newmode |= fattr->mode & S_IALLUGO;
1733 inode->i_mode = newmode;
1734 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1736 } else if (server->caps & NFS_CAP_MODE)
1737 nfsi->cache_validity |= save_cache_validity &
1738 (NFS_INO_INVALID_ATTR
1739 | NFS_INO_INVALID_ACCESS
1740 | NFS_INO_INVALID_ACL
1741 | NFS_INO_REVAL_FORCED);
1743 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1744 if (!uid_eq(inode->i_uid, fattr->uid)) {
1745 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1746 inode->i_uid = fattr->uid;
1748 } else if (server->caps & NFS_CAP_OWNER)
1749 nfsi->cache_validity |= save_cache_validity &
1750 (NFS_INO_INVALID_ATTR
1751 | NFS_INO_INVALID_ACCESS
1752 | NFS_INO_INVALID_ACL
1753 | NFS_INO_REVAL_FORCED);
1755 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1756 if (!gid_eq(inode->i_gid, fattr->gid)) {
1757 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1758 inode->i_gid = fattr->gid;
1760 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1761 nfsi->cache_validity |= save_cache_validity &
1762 (NFS_INO_INVALID_ATTR
1763 | NFS_INO_INVALID_ACCESS
1764 | NFS_INO_INVALID_ACL
1765 | NFS_INO_REVAL_FORCED);
1767 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1768 if (inode->i_nlink != fattr->nlink) {
1769 invalid |= NFS_INO_INVALID_ATTR;
1770 if (S_ISDIR(inode->i_mode))
1771 invalid |= NFS_INO_INVALID_DATA;
1772 set_nlink(inode, fattr->nlink);
1774 } else if (server->caps & NFS_CAP_NLINK)
1775 nfsi->cache_validity |= save_cache_validity &
1776 (NFS_INO_INVALID_ATTR
1777 | NFS_INO_REVAL_FORCED);
1779 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1781 * report the blocks in 512byte units
1783 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1785 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1786 inode->i_blocks = fattr->du.nfs2.blocks;
1788 /* Update attrtimeo value if we're out of the unstable period */
1789 if (invalid & NFS_INO_INVALID_ATTR) {
1790 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1791 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1792 nfsi->attrtimeo_timestamp = now;
1793 /* Set barrier to be more recent than all outstanding updates */
1794 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1796 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1797 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1798 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1799 nfsi->attrtimeo_timestamp = now;
1801 /* Set the barrier to be more recent than this fattr */
1802 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1803 nfsi->attr_gencount = fattr->gencount;
1805 invalid &= ~NFS_INO_INVALID_ATTR;
1806 /* Don't invalidate the data if we were to blame */
1807 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1808 || S_ISLNK(inode->i_mode)))
1809 invalid &= ~NFS_INO_INVALID_DATA;
1810 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1811 (save_cache_validity & NFS_INO_REVAL_FORCED))
1812 nfs_set_cache_invalid(inode, invalid);
1817 * No need to worry about unhashing the dentry, as the
1818 * lookup validation will know that the inode is bad.
1819 * (But we fall through to invalidate the caches.)
1821 nfs_invalidate_inode(inode);
1825 struct inode *nfs_alloc_inode(struct super_block *sb)
1827 struct nfs_inode *nfsi;
1828 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1832 nfsi->cache_validity = 0UL;
1833 #if IS_ENABLED(CONFIG_NFS_V4)
1834 nfsi->nfs4_acl = NULL;
1835 #endif /* CONFIG_NFS_V4 */
1836 return &nfsi->vfs_inode;
1838 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1840 static void nfs_i_callback(struct rcu_head *head)
1842 struct inode *inode = container_of(head, struct inode, i_rcu);
1843 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1846 void nfs_destroy_inode(struct inode *inode)
1848 call_rcu(&inode->i_rcu, nfs_i_callback);
1850 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1852 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1854 #if IS_ENABLED(CONFIG_NFS_V4)
1855 INIT_LIST_HEAD(&nfsi->open_states);
1856 nfsi->delegation = NULL;
1857 init_rwsem(&nfsi->rwsem);
1858 nfsi->layout = NULL;
1862 static void init_once(void *foo)
1864 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1866 inode_init_once(&nfsi->vfs_inode);
1867 INIT_LIST_HEAD(&nfsi->open_files);
1868 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1869 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1870 INIT_LIST_HEAD(&nfsi->commit_info.list);
1871 nfsi->nrequests = 0;
1872 nfsi->commit_info.ncommit = 0;
1873 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1874 atomic_set(&nfsi->silly_count, 1);
1875 INIT_HLIST_HEAD(&nfsi->silly_list);
1876 init_waitqueue_head(&nfsi->waitqueue);
1877 nfs4_init_once(nfsi);
1880 static int __init nfs_init_inodecache(void)
1882 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1883 sizeof(struct nfs_inode),
1884 0, (SLAB_RECLAIM_ACCOUNT|
1887 if (nfs_inode_cachep == NULL)
1893 static void nfs_destroy_inodecache(void)
1896 * Make sure all delayed rcu free inodes are flushed before we
1900 kmem_cache_destroy(nfs_inode_cachep);
1903 struct workqueue_struct *nfsiod_workqueue;
1904 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1907 * start up the nfsiod workqueue
1909 static int nfsiod_start(void)
1911 struct workqueue_struct *wq;
1912 dprintk("RPC: creating workqueue nfsiod\n");
1913 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1916 nfsiod_workqueue = wq;
1921 * Destroy the nfsiod workqueue
1923 static void nfsiod_stop(void)
1925 struct workqueue_struct *wq;
1927 wq = nfsiod_workqueue;
1930 nfsiod_workqueue = NULL;
1931 destroy_workqueue(wq);
1935 EXPORT_SYMBOL_GPL(nfs_net_id);
1937 static int nfs_net_init(struct net *net)
1939 nfs_clients_init(net);
1940 return nfs_fs_proc_net_init(net);
1943 static void nfs_net_exit(struct net *net)
1945 nfs_fs_proc_net_exit(net);
1946 nfs_cleanup_cb_ident_idr(net);
1949 static struct pernet_operations nfs_net_ops = {
1950 .init = nfs_net_init,
1951 .exit = nfs_net_exit,
1953 .size = sizeof(struct nfs_net),
1959 static int __init init_nfs_fs(void)
1963 err = register_pernet_subsys(&nfs_net_ops);
1967 err = nfs_fscache_register();
1971 err = nfsiod_start();
1975 err = nfs_fs_proc_init();
1979 err = nfs_init_nfspagecache();
1983 err = nfs_init_inodecache();
1987 err = nfs_init_readpagecache();
1991 err = nfs_init_writepagecache();
1995 err = nfs_init_directcache();
1999 #ifdef CONFIG_PROC_FS
2000 rpc_proc_register(&init_net, &nfs_rpcstat);
2002 if ((err = register_nfs_fs()) != 0)
2007 #ifdef CONFIG_PROC_FS
2008 rpc_proc_unregister(&init_net, "nfs");
2010 nfs_destroy_directcache();
2012 nfs_destroy_writepagecache();
2014 nfs_destroy_readpagecache();
2016 nfs_destroy_inodecache();
2018 nfs_destroy_nfspagecache();
2024 nfs_fscache_unregister();
2026 unregister_pernet_subsys(&nfs_net_ops);
2031 static void __exit exit_nfs_fs(void)
2033 nfs_destroy_directcache();
2034 nfs_destroy_writepagecache();
2035 nfs_destroy_readpagecache();
2036 nfs_destroy_inodecache();
2037 nfs_destroy_nfspagecache();
2038 nfs_fscache_unregister();
2039 unregister_pernet_subsys(&nfs_net_ops);
2040 #ifdef CONFIG_PROC_FS
2041 rpc_proc_unregister(&init_net, "nfs");
2043 unregister_nfs_fs();
2048 /* Not quite true; I just maintain it */
2049 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2050 MODULE_LICENSE("GPL");
2051 module_param(enable_ino64, bool, 0644);
2053 module_init(init_nfs_fs)
2054 module_exit(exit_nfs_fs)