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_release_inode_cookie(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 (nfs_server_capable(inode, NFS_CAP_SECURITY_LABEL) == 0)
280 if (NFS_SERVER(inode)->nfs_client->cl_minorversion < 2)
283 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
284 error = security_inode_notifysecctx(inode, label->label,
287 printk(KERN_ERR "%s() %s %d "
288 "security_inode_notifysecctx() %d\n",
290 (char *)label->label,
295 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
297 struct nfs4_label *label = NULL;
298 int minor_version = server->nfs_client->cl_minorversion;
300 if (minor_version < 2)
303 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
306 label = kzalloc(sizeof(struct nfs4_label), flags);
308 return ERR_PTR(-ENOMEM);
310 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
311 if (label->label == NULL) {
313 return ERR_PTR(-ENOMEM);
315 label->len = NFS4_MAXLABELLEN;
319 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
321 void inline nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
322 struct nfs4_label *label)
326 EXPORT_SYMBOL_GPL(nfs_setsecurity);
329 * This is our front-end to iget that looks up inodes by file handle
330 * instead of inode number.
333 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
335 struct nfs_find_desc desc = {
339 struct inode *inode = ERR_PTR(-ENOENT);
342 nfs_attr_check_mountpoint(sb, fattr);
344 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
345 !nfs_attr_use_mounted_on_fileid(fattr))
347 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
350 hash = nfs_fattr_to_ino_t(fattr);
352 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
354 inode = ERR_PTR(-ENOMEM);
358 if (inode->i_state & I_NEW) {
359 struct nfs_inode *nfsi = NFS_I(inode);
360 unsigned long now = jiffies;
362 /* We set i_ino for the few things that still rely on it,
366 /* We can't support update_atime(), since the server will reset it */
367 inode->i_flags |= S_NOATIME|S_NOCMTIME;
368 inode->i_mode = fattr->mode;
369 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
370 && nfs_server_capable(inode, NFS_CAP_MODE))
371 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
372 /* Why so? Because we want revalidate for devices/FIFOs, and
373 * that's precisely what we have in nfs_file_inode_operations.
375 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
376 if (S_ISREG(inode->i_mode)) {
377 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
378 inode->i_data.a_ops = &nfs_file_aops;
379 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
380 } else if (S_ISDIR(inode->i_mode)) {
381 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
382 inode->i_fop = &nfs_dir_operations;
383 inode->i_data.a_ops = &nfs_dir_aops;
384 /* Deal with crossing mountpoints */
385 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
386 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
387 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
388 inode->i_op = &nfs_referral_inode_operations;
390 inode->i_op = &nfs_mountpoint_inode_operations;
392 inode->i_flags |= S_AUTOMOUNT;
394 } else if (S_ISLNK(inode->i_mode))
395 inode->i_op = &nfs_symlink_inode_operations;
397 init_special_inode(inode, inode->i_mode, fattr->rdev);
399 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
400 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
401 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
402 inode->i_version = 0;
405 inode->i_uid = make_kuid(&init_user_ns, -2);
406 inode->i_gid = make_kgid(&init_user_ns, -2);
408 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
412 nfsi->read_cache_jiffies = fattr->time_start;
413 nfsi->attr_gencount = fattr->gencount;
414 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
415 inode->i_atime = fattr->atime;
416 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
417 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
418 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
419 inode->i_mtime = fattr->mtime;
420 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
421 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
422 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
423 inode->i_ctime = fattr->ctime;
424 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
425 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
426 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
427 inode->i_version = fattr->change_attr;
428 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
429 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
430 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
431 inode->i_size = nfs_size_to_loff_t(fattr->size);
433 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
434 | NFS_INO_REVAL_PAGECACHE;
435 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
436 set_nlink(inode, fattr->nlink);
437 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
438 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
439 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
440 inode->i_uid = fattr->uid;
441 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
442 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
443 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
444 inode->i_gid = fattr->gid;
445 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
446 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
447 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
448 inode->i_blocks = fattr->du.nfs2.blocks;
449 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
451 * report the blocks in 512byte units
453 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
456 nfs_setsecurity(inode, fattr, label);
458 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
459 nfsi->attrtimeo_timestamp = now;
460 nfsi->access_cache = RB_ROOT;
462 nfs_fscache_init_inode_cookie(inode);
464 unlock_new_inode(inode);
466 nfs_refresh_inode(inode, fattr);
467 dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n",
469 (long long)NFS_FILEID(inode),
470 nfs_display_fhandle_hash(fh),
471 atomic_read(&inode->i_count));
477 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
480 EXPORT_SYMBOL_GPL(nfs_fhget);
482 #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)
485 nfs_setattr(struct dentry *dentry, struct iattr *attr)
487 struct inode *inode = dentry->d_inode;
488 struct nfs_fattr *fattr;
491 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
493 /* skip mode change if it's just for clearing setuid/setgid */
494 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
495 attr->ia_valid &= ~ATTR_MODE;
497 if (attr->ia_valid & ATTR_SIZE) {
498 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
499 attr->ia_valid &= ~ATTR_SIZE;
502 /* Optimization: if the end result is no change, don't RPC */
503 attr->ia_valid &= NFS_VALID_ATTRS;
504 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
507 trace_nfs_setattr_enter(inode);
509 /* Write all dirty data */
510 if (S_ISREG(inode->i_mode)) {
511 nfs_inode_dio_wait(inode);
515 fattr = nfs_alloc_fattr();
519 * Return any delegations if we're going to change ACLs
521 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
522 NFS_PROTO(inode)->return_delegation(inode);
523 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
525 error = nfs_refresh_inode(inode, fattr);
526 nfs_free_fattr(fattr);
528 trace_nfs_setattr_exit(inode, error);
531 EXPORT_SYMBOL_GPL(nfs_setattr);
534 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
535 * @inode: inode of the file used
536 * @offset: file offset to start truncating
538 * This is a copy of the common vmtruncate, but with the locking
539 * corrected to take into account the fact that NFS requires
540 * inode->i_size to be updated under the inode->i_lock.
542 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
547 err = inode_newsize_ok(inode, offset);
551 spin_lock(&inode->i_lock);
552 oldsize = inode->i_size;
553 i_size_write(inode, offset);
554 spin_unlock(&inode->i_lock);
556 truncate_pagecache(inode, oldsize, offset);
562 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
563 * @inode: pointer to struct inode
564 * @attr: pointer to struct iattr
566 * Note: we do this in the *proc.c in order to ensure that
567 * it works for things like exclusive creates too.
569 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
571 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
572 spin_lock(&inode->i_lock);
573 if ((attr->ia_valid & ATTR_MODE) != 0) {
574 int mode = attr->ia_mode & S_IALLUGO;
575 mode |= inode->i_mode & ~S_IALLUGO;
576 inode->i_mode = mode;
578 if ((attr->ia_valid & ATTR_UID) != 0)
579 inode->i_uid = attr->ia_uid;
580 if ((attr->ia_valid & ATTR_GID) != 0)
581 inode->i_gid = attr->ia_gid;
582 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
583 spin_unlock(&inode->i_lock);
585 if ((attr->ia_valid & ATTR_SIZE) != 0) {
586 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
587 nfs_vmtruncate(inode, attr->ia_size);
590 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
592 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
594 struct inode *inode = dentry->d_inode;
595 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
598 trace_nfs_getattr_enter(inode);
599 /* Flush out writes to the server in order to update c/mtime. */
600 if (S_ISREG(inode->i_mode)) {
601 nfs_inode_dio_wait(inode);
602 err = filemap_write_and_wait(inode->i_mapping);
608 * We may force a getattr if the user cares about atime.
610 * Note that we only have to check the vfsmount flags here:
611 * - NFS always sets S_NOATIME by so checking it would give a
613 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
614 * no point in checking those.
616 if ((mnt->mnt_flags & MNT_NOATIME) ||
617 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
621 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
623 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
625 generic_fillattr(inode, stat);
626 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
629 trace_nfs_getattr_exit(inode, err);
632 EXPORT_SYMBOL_GPL(nfs_getattr);
634 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
636 atomic_set(&l_ctx->count, 1);
637 l_ctx->lockowner.l_owner = current->files;
638 l_ctx->lockowner.l_pid = current->tgid;
639 INIT_LIST_HEAD(&l_ctx->list);
640 nfs_iocounter_init(&l_ctx->io_count);
643 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
645 struct nfs_lock_context *head = &ctx->lock_context;
646 struct nfs_lock_context *pos = head;
649 if (pos->lockowner.l_owner != current->files)
651 if (pos->lockowner.l_pid != current->tgid)
653 atomic_inc(&pos->count);
655 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
659 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
661 struct nfs_lock_context *res, *new = NULL;
662 struct inode *inode = ctx->dentry->d_inode;
664 spin_lock(&inode->i_lock);
665 res = __nfs_find_lock_context(ctx);
667 spin_unlock(&inode->i_lock);
668 new = kmalloc(sizeof(*new), GFP_KERNEL);
670 return ERR_PTR(-ENOMEM);
671 nfs_init_lock_context(new);
672 spin_lock(&inode->i_lock);
673 res = __nfs_find_lock_context(ctx);
675 list_add_tail(&new->list, &ctx->lock_context.list);
676 new->open_context = ctx;
681 spin_unlock(&inode->i_lock);
686 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
688 struct nfs_open_context *ctx = l_ctx->open_context;
689 struct inode *inode = ctx->dentry->d_inode;
691 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
693 list_del(&l_ctx->list);
694 spin_unlock(&inode->i_lock);
699 * nfs_close_context - Common close_context() routine NFSv2/v3
700 * @ctx: pointer to context
701 * @is_sync: is this a synchronous close
703 * always ensure that the attributes are up to date if we're mounted
704 * with close-to-open semantics
706 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
709 struct nfs_server *server;
711 if (!(ctx->mode & FMODE_WRITE))
715 inode = ctx->dentry->d_inode;
716 if (!list_empty(&NFS_I(inode)->open_files))
718 server = NFS_SERVER(inode);
719 if (server->flags & NFS_MOUNT_NOCTO)
721 nfs_revalidate_inode(server, inode);
723 EXPORT_SYMBOL_GPL(nfs_close_context);
725 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
727 struct nfs_open_context *ctx;
728 struct rpc_cred *cred = rpc_lookup_cred();
730 return ERR_CAST(cred);
732 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
735 return ERR_PTR(-ENOMEM);
737 nfs_sb_active(dentry->d_sb);
738 ctx->dentry = dget(dentry);
744 nfs_init_lock_context(&ctx->lock_context);
745 ctx->lock_context.open_context = ctx;
746 INIT_LIST_HEAD(&ctx->list);
747 ctx->mdsthreshold = NULL;
750 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
752 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
755 atomic_inc(&ctx->lock_context.count);
758 EXPORT_SYMBOL_GPL(get_nfs_open_context);
760 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
762 struct inode *inode = ctx->dentry->d_inode;
763 struct super_block *sb = ctx->dentry->d_sb;
765 if (!list_empty(&ctx->list)) {
766 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
768 list_del(&ctx->list);
769 spin_unlock(&inode->i_lock);
770 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
773 NFS_PROTO(inode)->close_context(ctx, is_sync);
774 if (ctx->cred != NULL)
775 put_rpccred(ctx->cred);
778 kfree(ctx->mdsthreshold);
782 void put_nfs_open_context(struct nfs_open_context *ctx)
784 __put_nfs_open_context(ctx, 0);
786 EXPORT_SYMBOL_GPL(put_nfs_open_context);
789 * Ensure that mmap has a recent RPC credential for use when writing out
792 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
794 struct inode *inode = ctx->dentry->d_inode;
795 struct nfs_inode *nfsi = NFS_I(inode);
797 spin_lock(&inode->i_lock);
798 list_add(&ctx->list, &nfsi->open_files);
799 spin_unlock(&inode->i_lock);
801 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
803 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
805 filp->private_data = get_nfs_open_context(ctx);
806 if (list_empty(&ctx->list))
807 nfs_inode_attach_open_context(ctx);
809 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
812 * Given an inode, search for an open context with the desired characteristics
814 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
816 struct nfs_inode *nfsi = NFS_I(inode);
817 struct nfs_open_context *pos, *ctx = NULL;
819 spin_lock(&inode->i_lock);
820 list_for_each_entry(pos, &nfsi->open_files, list) {
821 if (cred != NULL && pos->cred != cred)
823 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
825 ctx = get_nfs_open_context(pos);
828 spin_unlock(&inode->i_lock);
832 static void nfs_file_clear_open_context(struct file *filp)
834 struct nfs_open_context *ctx = nfs_file_open_context(filp);
837 struct inode *inode = ctx->dentry->d_inode;
839 filp->private_data = NULL;
840 spin_lock(&inode->i_lock);
841 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
842 spin_unlock(&inode->i_lock);
843 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
848 * These allocate and release file read/write context information.
850 int nfs_open(struct inode *inode, struct file *filp)
852 struct nfs_open_context *ctx;
854 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
857 nfs_file_set_open_context(filp, ctx);
858 put_nfs_open_context(ctx);
859 nfs_fscache_set_inode_cookie(inode, filp);
863 int nfs_release(struct inode *inode, struct file *filp)
865 nfs_file_clear_open_context(filp);
870 * This function is called whenever some part of NFS notices that
871 * the cached attributes have to be refreshed.
874 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
876 int status = -ESTALE;
877 struct nfs4_label *label = NULL;
878 struct nfs_fattr *fattr = NULL;
879 struct nfs_inode *nfsi = NFS_I(inode);
881 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
882 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
884 trace_nfs_revalidate_inode_enter(inode);
886 if (is_bad_inode(inode))
888 if (NFS_STALE(inode))
892 fattr = nfs_alloc_fattr();
896 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
898 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
900 status = PTR_ERR(label);
904 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
906 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
908 (long long)NFS_FILEID(inode), status);
909 if (status == -ESTALE) {
910 nfs_zap_caches(inode);
911 if (!S_ISDIR(inode->i_mode))
912 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
917 status = nfs_refresh_inode(inode, fattr);
919 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
921 (long long)NFS_FILEID(inode), status);
925 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
926 nfs_zap_acl_cache(inode);
928 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
930 (long long)NFS_FILEID(inode));
933 nfs4_label_free(label);
935 nfs_free_fattr(fattr);
936 trace_nfs_revalidate_inode_exit(inode, status);
940 int nfs_attribute_timeout(struct inode *inode)
942 struct nfs_inode *nfsi = NFS_I(inode);
944 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
947 int nfs_attribute_cache_expired(struct inode *inode)
949 if (nfs_have_delegated_attributes(inode))
951 return nfs_attribute_timeout(inode);
955 * nfs_revalidate_inode - Revalidate the inode attributes
956 * @server - pointer to nfs_server struct
957 * @inode - pointer to inode struct
959 * Updates inode attribute information by retrieving the data from the server.
961 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
963 if (!(NFS_I(inode)->cache_validity &
964 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
965 && !nfs_attribute_cache_expired(inode))
966 return NFS_STALE(inode) ? -ESTALE : 0;
967 return __nfs_revalidate_inode(server, inode);
969 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
971 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
973 struct nfs_inode *nfsi = NFS_I(inode);
976 if (mapping->nrpages != 0) {
977 if (S_ISREG(inode->i_mode)) {
978 ret = nfs_sync_mapping(mapping);
982 ret = invalidate_inode_pages2(mapping);
986 spin_lock(&inode->i_lock);
987 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
988 if (S_ISDIR(inode->i_mode))
989 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
990 spin_unlock(&inode->i_lock);
991 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
992 nfs_fscache_wait_on_invalidate(inode);
994 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
995 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
999 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1001 if (nfs_have_delegated_attributes(inode))
1003 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1004 || nfs_attribute_timeout(inode)
1005 || NFS_STALE(inode);
1009 * nfs_revalidate_mapping - Revalidate the pagecache
1010 * @inode - pointer to host inode
1011 * @mapping - pointer to mapping
1013 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1015 struct nfs_inode *nfsi = NFS_I(inode);
1018 /* swapfiles are not supposed to be shared. */
1019 if (IS_SWAPFILE(inode))
1022 if (nfs_mapping_need_revalidate_inode(inode)) {
1023 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1027 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1028 trace_nfs_invalidate_mapping_enter(inode);
1029 ret = nfs_invalidate_mapping(inode, mapping);
1030 trace_nfs_invalidate_mapping_exit(inode, ret);
1037 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1039 struct nfs_inode *nfsi = NFS_I(inode);
1040 unsigned long ret = 0;
1042 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1043 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1044 && inode->i_version == fattr->pre_change_attr) {
1045 inode->i_version = fattr->change_attr;
1046 if (S_ISDIR(inode->i_mode))
1047 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1048 ret |= NFS_INO_INVALID_ATTR;
1050 /* If we have atomic WCC data, we may update some attributes */
1051 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1052 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1053 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1054 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1055 ret |= NFS_INO_INVALID_ATTR;
1058 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1059 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1060 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1061 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1062 if (S_ISDIR(inode->i_mode))
1063 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1064 ret |= NFS_INO_INVALID_ATTR;
1066 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1067 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1068 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1069 && nfsi->npages == 0) {
1070 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1071 ret |= NFS_INO_INVALID_ATTR;
1074 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1075 nfs_fscache_invalidate(inode);
1081 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1082 * @inode - pointer to inode
1083 * @fattr - updated attributes
1085 * Verifies the attribute cache. If we have just changed the attributes,
1086 * so that fattr carries weak cache consistency data, then it may
1087 * also update the ctime/mtime/change_attribute.
1089 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1091 struct nfs_inode *nfsi = NFS_I(inode);
1092 loff_t cur_size, new_isize;
1093 unsigned long invalid = 0;
1096 if (nfs_have_delegated_attributes(inode))
1098 /* Has the inode gone and changed behind our back? */
1099 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1101 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1104 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1105 inode->i_version != fattr->change_attr)
1106 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1108 /* Verify a few of the more important attributes */
1109 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1110 invalid |= NFS_INO_INVALID_ATTR;
1112 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1113 cur_size = i_size_read(inode);
1114 new_isize = nfs_size_to_loff_t(fattr->size);
1115 if (cur_size != new_isize && nfsi->npages == 0)
1116 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1119 /* Have any file permissions changed? */
1120 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1121 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1122 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1123 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1124 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1125 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1127 /* Has the link count changed? */
1128 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1129 invalid |= NFS_INO_INVALID_ATTR;
1131 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1132 invalid |= NFS_INO_INVALID_ATIME;
1135 nfsi->cache_validity |= invalid;
1137 nfsi->read_cache_jiffies = fattr->time_start;
1141 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1143 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1145 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1148 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1150 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1152 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1155 static atomic_long_t nfs_attr_generation_counter;
1157 static unsigned long nfs_read_attr_generation_counter(void)
1159 return atomic_long_read(&nfs_attr_generation_counter);
1162 unsigned long nfs_inc_attr_generation_counter(void)
1164 return atomic_long_inc_return(&nfs_attr_generation_counter);
1167 void nfs_fattr_init(struct nfs_fattr *fattr)
1170 fattr->time_start = jiffies;
1171 fattr->gencount = nfs_inc_attr_generation_counter();
1172 fattr->owner_name = NULL;
1173 fattr->group_name = NULL;
1175 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1177 struct nfs_fattr *nfs_alloc_fattr(void)
1179 struct nfs_fattr *fattr;
1181 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1183 nfs_fattr_init(fattr);
1186 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1188 struct nfs_fh *nfs_alloc_fhandle(void)
1192 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1197 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1201 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1202 * in the same way that wireshark does
1206 * For debugging only.
1208 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1210 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1211 * not on the result */
1212 return nfs_fhandle_hash(fh);
1216 * _nfs_display_fhandle - display an NFS file handle on the console
1218 * @fh: file handle to display
1219 * @caption: display caption
1221 * For debugging only.
1223 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1227 if (fh == NULL || fh->size == 0) {
1228 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1232 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1233 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1234 for (i = 0; i < fh->size; i += 16) {
1235 __be32 *pos = (__be32 *)&fh->data[i];
1237 switch ((fh->size - i - 1) >> 2) {
1239 printk(KERN_DEFAULT " %08x\n",
1243 printk(KERN_DEFAULT " %08x %08x\n",
1244 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1247 printk(KERN_DEFAULT " %08x %08x %08x\n",
1248 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1249 be32_to_cpup(pos + 2));
1252 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1253 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1254 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1261 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1262 * @inode - pointer to inode
1263 * @fattr - attributes
1265 * Attempt to divine whether or not an RPC call reply carrying stale
1266 * attributes got scheduled after another call carrying updated ones.
1268 * To do so, the function first assumes that a more recent ctime means
1269 * that the attributes in fattr are newer, however it also attempt to
1270 * catch the case where ctime either didn't change, or went backwards
1271 * (if someone reset the clock on the server) by looking at whether
1272 * or not this RPC call was started after the inode was last updated.
1273 * Note also the check for wraparound of 'attr_gencount'
1275 * The function returns 'true' if it thinks the attributes in 'fattr' are
1276 * more recent than the ones cached in the inode.
1279 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1281 const struct nfs_inode *nfsi = NFS_I(inode);
1283 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1284 nfs_ctime_need_update(inode, fattr) ||
1285 nfs_size_need_update(inode, fattr) ||
1286 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1289 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1293 trace_nfs_refresh_inode_enter(inode);
1295 if (nfs_inode_attrs_need_update(inode, fattr))
1296 ret = nfs_update_inode(inode, fattr);
1298 ret = nfs_check_inode_attributes(inode, fattr);
1300 trace_nfs_refresh_inode_exit(inode, ret);
1305 * nfs_refresh_inode - try to update the inode attribute cache
1306 * @inode - pointer to inode
1307 * @fattr - updated attributes
1309 * Check that an RPC call that returned attributes has not overlapped with
1310 * other recent updates of the inode metadata, then decide whether it is
1311 * safe to do a full update of the inode attributes, or whether just to
1312 * call nfs_check_inode_attributes.
1314 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1318 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1320 spin_lock(&inode->i_lock);
1321 status = nfs_refresh_inode_locked(inode, fattr);
1322 spin_unlock(&inode->i_lock);
1326 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1328 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1330 struct nfs_inode *nfsi = NFS_I(inode);
1332 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1333 if (S_ISDIR(inode->i_mode)) {
1334 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1335 nfs_fscache_invalidate(inode);
1337 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1339 return nfs_refresh_inode_locked(inode, fattr);
1343 * nfs_post_op_update_inode - try to update the inode attribute cache
1344 * @inode - pointer to inode
1345 * @fattr - updated attributes
1347 * After an operation that has changed the inode metadata, mark the
1348 * attribute cache as being invalid, then try to update it.
1350 * NB: if the server didn't return any post op attributes, this
1351 * function will force the retrieval of attributes before the next
1352 * NFS request. Thus it should be used only for operations that
1353 * are expected to change one or more attributes, to avoid
1354 * unnecessary NFS requests and trips through nfs_update_inode().
1356 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1360 spin_lock(&inode->i_lock);
1361 status = nfs_post_op_update_inode_locked(inode, fattr);
1362 spin_unlock(&inode->i_lock);
1366 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1369 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1370 * @inode - pointer to inode
1371 * @fattr - updated attributes
1373 * After an operation that has changed the inode metadata, mark the
1374 * attribute cache as being invalid, then try to update it. Fake up
1375 * weak cache consistency data, if none exist.
1377 * This function is mainly designed to be used by the ->write_done() functions.
1379 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1383 spin_lock(&inode->i_lock);
1384 /* Don't do a WCC update if these attributes are already stale */
1385 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1386 !nfs_inode_attrs_need_update(inode, fattr)) {
1387 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1388 | NFS_ATTR_FATTR_PRESIZE
1389 | NFS_ATTR_FATTR_PREMTIME
1390 | NFS_ATTR_FATTR_PRECTIME);
1393 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1394 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1395 fattr->pre_change_attr = inode->i_version;
1396 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1398 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1399 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1400 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1401 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1403 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1404 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1405 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1406 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1408 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1409 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1410 fattr->pre_size = i_size_read(inode);
1411 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1414 status = nfs_post_op_update_inode_locked(inode, fattr);
1415 spin_unlock(&inode->i_lock);
1418 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1421 * Many nfs protocol calls return the new file attributes after
1422 * an operation. Here we update the inode to reflect the state
1423 * of the server's inode.
1425 * This is a bit tricky because we have to make sure all dirty pages
1426 * have been sent off to the server before calling invalidate_inode_pages.
1427 * To make sure no other process adds more write requests while we try
1428 * our best to flush them, we make them sleep during the attribute refresh.
1430 * A very similar scenario holds for the dir cache.
1432 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1434 struct nfs_server *server;
1435 struct nfs_inode *nfsi = NFS_I(inode);
1436 loff_t cur_isize, new_isize;
1437 unsigned long invalid = 0;
1438 unsigned long now = jiffies;
1439 unsigned long save_cache_validity;
1441 dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n",
1442 __func__, inode->i_sb->s_id, inode->i_ino,
1443 nfs_display_fhandle_hash(NFS_FH(inode)),
1444 atomic_read(&inode->i_count), fattr->valid);
1446 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1447 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1448 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1449 NFS_SERVER(inode)->nfs_client->cl_hostname,
1450 inode->i_sb->s_id, (long long)nfsi->fileid,
1451 (long long)fattr->fileid);
1456 * Make sure the inode's type hasn't changed.
1458 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1460 * Big trouble! The inode has become a different object.
1462 printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n",
1463 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1467 server = NFS_SERVER(inode);
1468 /* Update the fsid? */
1469 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1470 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1471 !IS_AUTOMOUNT(inode))
1472 server->fsid = fattr->fsid;
1475 * Update the read time so we don't revalidate too often.
1477 nfsi->read_cache_jiffies = fattr->time_start;
1479 save_cache_validity = nfsi->cache_validity;
1480 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1481 | NFS_INO_INVALID_ATIME
1482 | NFS_INO_REVAL_FORCED
1483 | NFS_INO_REVAL_PAGECACHE);
1485 /* Do atomic weak cache consistency updates */
1486 invalid |= nfs_wcc_update_inode(inode, fattr);
1488 /* More cache consistency checks */
1489 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1490 if (inode->i_version != fattr->change_attr) {
1491 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1492 inode->i_sb->s_id, inode->i_ino);
1493 invalid |= NFS_INO_INVALID_ATTR
1494 | NFS_INO_INVALID_DATA
1495 | NFS_INO_INVALID_ACCESS
1496 | NFS_INO_INVALID_ACL
1497 | NFS_INO_REVAL_PAGECACHE;
1498 if (S_ISDIR(inode->i_mode))
1499 nfs_force_lookup_revalidate(inode);
1500 inode->i_version = fattr->change_attr;
1502 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1503 invalid |= save_cache_validity;
1505 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1506 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1507 } else if (server->caps & NFS_CAP_MTIME)
1508 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1509 | NFS_INO_REVAL_FORCED);
1511 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1512 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1513 } else if (server->caps & NFS_CAP_CTIME)
1514 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1515 | NFS_INO_REVAL_FORCED);
1517 /* Check if our cached file size is stale */
1518 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1519 new_isize = nfs_size_to_loff_t(fattr->size);
1520 cur_isize = i_size_read(inode);
1521 if (new_isize != cur_isize) {
1522 /* Do we perhaps have any outstanding writes, or has
1523 * the file grown beyond our last write? */
1524 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1525 new_isize > cur_isize) {
1526 i_size_write(inode, new_isize);
1527 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1529 dprintk("NFS: isize change on server for file %s/%ld "
1533 (long long)cur_isize,
1534 (long long)new_isize);
1537 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1538 | NFS_INO_REVAL_PAGECACHE
1539 | NFS_INO_REVAL_FORCED);
1542 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1543 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1544 else if (server->caps & NFS_CAP_ATIME)
1545 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1546 | NFS_INO_REVAL_FORCED);
1548 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1549 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1550 umode_t newmode = inode->i_mode & S_IFMT;
1551 newmode |= fattr->mode & S_IALLUGO;
1552 inode->i_mode = newmode;
1553 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1555 } else if (server->caps & NFS_CAP_MODE)
1556 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1557 | NFS_INO_INVALID_ACCESS
1558 | NFS_INO_INVALID_ACL
1559 | NFS_INO_REVAL_FORCED);
1561 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1562 if (!uid_eq(inode->i_uid, fattr->uid)) {
1563 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1564 inode->i_uid = fattr->uid;
1566 } else if (server->caps & NFS_CAP_OWNER)
1567 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1568 | NFS_INO_INVALID_ACCESS
1569 | NFS_INO_INVALID_ACL
1570 | NFS_INO_REVAL_FORCED);
1572 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1573 if (!gid_eq(inode->i_gid, fattr->gid)) {
1574 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1575 inode->i_gid = fattr->gid;
1577 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1578 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1579 | NFS_INO_INVALID_ACCESS
1580 | NFS_INO_INVALID_ACL
1581 | NFS_INO_REVAL_FORCED);
1583 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1584 if (inode->i_nlink != fattr->nlink) {
1585 invalid |= NFS_INO_INVALID_ATTR;
1586 if (S_ISDIR(inode->i_mode))
1587 invalid |= NFS_INO_INVALID_DATA;
1588 set_nlink(inode, fattr->nlink);
1590 } else if (server->caps & NFS_CAP_NLINK)
1591 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1592 | NFS_INO_REVAL_FORCED);
1594 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1596 * report the blocks in 512byte units
1598 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1600 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1601 inode->i_blocks = fattr->du.nfs2.blocks;
1603 /* Update attrtimeo value if we're out of the unstable period */
1604 if (invalid & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) {
1605 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1606 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1607 nfsi->attrtimeo_timestamp = now;
1608 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1610 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1611 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1612 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1613 nfsi->attrtimeo_timestamp = now;
1616 invalid &= ~NFS_INO_INVALID_ATTR;
1617 invalid &= ~NFS_INO_INVALID_LABEL;
1618 /* Don't invalidate the data if we were to blame */
1619 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1620 || S_ISLNK(inode->i_mode)))
1621 invalid &= ~NFS_INO_INVALID_DATA;
1622 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1623 (save_cache_validity & NFS_INO_REVAL_FORCED))
1624 nfsi->cache_validity |= invalid;
1626 if (invalid & NFS_INO_INVALID_DATA)
1627 nfs_fscache_invalidate(inode);
1632 * No need to worry about unhashing the dentry, as the
1633 * lookup validation will know that the inode is bad.
1634 * (But we fall through to invalidate the caches.)
1636 nfs_invalidate_inode(inode);
1640 struct inode *nfs_alloc_inode(struct super_block *sb)
1642 struct nfs_inode *nfsi;
1643 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1647 nfsi->cache_validity = 0UL;
1648 #ifdef CONFIG_NFS_V3_ACL
1649 nfsi->acl_access = ERR_PTR(-EAGAIN);
1650 nfsi->acl_default = ERR_PTR(-EAGAIN);
1652 #if IS_ENABLED(CONFIG_NFS_V4)
1653 nfsi->nfs4_acl = NULL;
1654 #endif /* CONFIG_NFS_V4 */
1655 return &nfsi->vfs_inode;
1657 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1659 static void nfs_i_callback(struct rcu_head *head)
1661 struct inode *inode = container_of(head, struct inode, i_rcu);
1662 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1665 void nfs_destroy_inode(struct inode *inode)
1667 call_rcu(&inode->i_rcu, nfs_i_callback);
1669 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1671 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1673 #if IS_ENABLED(CONFIG_NFS_V4)
1674 INIT_LIST_HEAD(&nfsi->open_states);
1675 nfsi->delegation = NULL;
1676 nfsi->delegation_state = 0;
1677 init_rwsem(&nfsi->rwsem);
1678 nfsi->layout = NULL;
1682 static void init_once(void *foo)
1684 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1686 inode_init_once(&nfsi->vfs_inode);
1687 INIT_LIST_HEAD(&nfsi->open_files);
1688 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1689 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1690 INIT_LIST_HEAD(&nfsi->commit_info.list);
1692 nfsi->commit_info.ncommit = 0;
1693 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1694 atomic_set(&nfsi->silly_count, 1);
1695 INIT_HLIST_HEAD(&nfsi->silly_list);
1696 init_waitqueue_head(&nfsi->waitqueue);
1697 nfs4_init_once(nfsi);
1700 static int __init nfs_init_inodecache(void)
1702 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1703 sizeof(struct nfs_inode),
1704 0, (SLAB_RECLAIM_ACCOUNT|
1707 if (nfs_inode_cachep == NULL)
1713 static void nfs_destroy_inodecache(void)
1716 * Make sure all delayed rcu free inodes are flushed before we
1720 kmem_cache_destroy(nfs_inode_cachep);
1723 struct workqueue_struct *nfsiod_workqueue;
1724 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1727 * start up the nfsiod workqueue
1729 static int nfsiod_start(void)
1731 struct workqueue_struct *wq;
1732 dprintk("RPC: creating workqueue nfsiod\n");
1733 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1736 nfsiod_workqueue = wq;
1741 * Destroy the nfsiod workqueue
1743 static void nfsiod_stop(void)
1745 struct workqueue_struct *wq;
1747 wq = nfsiod_workqueue;
1750 nfsiod_workqueue = NULL;
1751 destroy_workqueue(wq);
1755 EXPORT_SYMBOL_GPL(nfs_net_id);
1757 static int nfs_net_init(struct net *net)
1759 nfs_clients_init(net);
1763 static void nfs_net_exit(struct net *net)
1765 nfs_cleanup_cb_ident_idr(net);
1768 static struct pernet_operations nfs_net_ops = {
1769 .init = nfs_net_init,
1770 .exit = nfs_net_exit,
1772 .size = sizeof(struct nfs_net),
1778 static int __init init_nfs_fs(void)
1782 err = register_pernet_subsys(&nfs_net_ops);
1786 err = nfs_fscache_register();
1790 err = nfsiod_start();
1794 err = nfs_fs_proc_init();
1798 err = nfs_init_nfspagecache();
1802 err = nfs_init_inodecache();
1806 err = nfs_init_readpagecache();
1810 err = nfs_init_writepagecache();
1814 err = nfs_init_directcache();
1818 #ifdef CONFIG_PROC_FS
1819 rpc_proc_register(&init_net, &nfs_rpcstat);
1821 if ((err = register_nfs_fs()) != 0)
1826 #ifdef CONFIG_PROC_FS
1827 rpc_proc_unregister(&init_net, "nfs");
1829 nfs_destroy_directcache();
1831 nfs_destroy_writepagecache();
1833 nfs_destroy_readpagecache();
1835 nfs_destroy_inodecache();
1837 nfs_destroy_nfspagecache();
1843 nfs_fscache_unregister();
1845 unregister_pernet_subsys(&nfs_net_ops);
1850 static void __exit exit_nfs_fs(void)
1852 nfs_destroy_directcache();
1853 nfs_destroy_writepagecache();
1854 nfs_destroy_readpagecache();
1855 nfs_destroy_inodecache();
1856 nfs_destroy_nfspagecache();
1857 nfs_fscache_unregister();
1858 unregister_pernet_subsys(&nfs_net_ops);
1859 #ifdef CONFIG_PROC_FS
1860 rpc_proc_unregister(&init_net, "nfs");
1862 unregister_nfs_fs();
1867 /* Not quite true; I just maintain it */
1868 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1869 MODULE_LICENSE("GPL");
1870 module_param(enable_ino64, bool, 0644);
1872 module_init(init_nfs_fs)
1873 module_exit(exit_nfs_fs)