4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
42 #include <asm/uaccess.h>
46 #include "delegation.h"
56 #define NFSDBG_FACILITY NFSDBG_VFS
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 static struct kmem_cache * nfs_inode_cachep;
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 return nfs_fileid_to_ino_t(fattr->fileid);
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
78 int nfs_wait_bit_killable(struct wait_bit_key *key)
80 if (fatal_signal_pending(current))
82 freezable_schedule_unsafe();
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
94 u64 nfs_compat_user_ino64(u64 fileid)
105 if (sizeof(ino) < sizeof(fileid))
106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
110 int nfs_drop_inode(struct inode *inode)
112 return NFS_STALE(inode) || generic_drop_inode(inode);
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
116 void nfs_clear_inode(struct inode *inode)
119 * The following should never happen...
121 WARN_ON_ONCE(nfs_have_writebacks(inode));
122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 nfs_zap_acl_cache(inode);
124 nfs_access_zap_cache(inode);
125 nfs_fscache_clear_inode(inode);
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
129 void nfs_evict_inode(struct inode *inode)
131 truncate_inode_pages_final(&inode->i_data);
133 nfs_clear_inode(inode);
137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
139 int nfs_sync_mapping(struct address_space *mapping)
143 if (mapping->nrpages != 0) {
144 unmap_mapping_range(mapping, 0, 0, 0);
145 ret = nfs_wb_all(mapping->host);
150 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
152 struct nfs_inode *nfsi = NFS_I(inode);
154 if (inode->i_mapping->nrpages == 0)
155 flags &= ~NFS_INO_INVALID_DATA;
156 nfsi->cache_validity |= flags;
157 if (flags & NFS_INO_INVALID_DATA)
158 nfs_fscache_invalidate(inode);
162 * Invalidate the local caches
164 static void nfs_zap_caches_locked(struct inode *inode)
166 struct nfs_inode *nfsi = NFS_I(inode);
167 int mode = inode->i_mode;
169 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
171 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
172 nfsi->attrtimeo_timestamp = jiffies;
174 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
175 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
176 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
177 | NFS_INO_INVALID_DATA
178 | NFS_INO_INVALID_ACCESS
179 | NFS_INO_INVALID_ACL
180 | NFS_INO_REVAL_PAGECACHE);
182 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
183 | NFS_INO_INVALID_ACCESS
184 | NFS_INO_INVALID_ACL
185 | NFS_INO_REVAL_PAGECACHE);
186 nfs_zap_label_cache_locked(nfsi);
189 void nfs_zap_caches(struct inode *inode)
191 spin_lock(&inode->i_lock);
192 nfs_zap_caches_locked(inode);
193 spin_unlock(&inode->i_lock);
195 EXPORT_SYMBOL_GPL(nfs_zap_caches);
197 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
199 if (mapping->nrpages != 0) {
200 spin_lock(&inode->i_lock);
201 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
202 spin_unlock(&inode->i_lock);
206 void nfs_zap_acl_cache(struct inode *inode)
208 void (*clear_acl_cache)(struct inode *);
210 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
211 if (clear_acl_cache != NULL)
212 clear_acl_cache(inode);
213 spin_lock(&inode->i_lock);
214 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
215 spin_unlock(&inode->i_lock);
217 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
219 void nfs_invalidate_atime(struct inode *inode)
221 spin_lock(&inode->i_lock);
222 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
223 spin_unlock(&inode->i_lock);
225 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
228 * Invalidate, but do not unhash, the inode.
229 * NB: must be called with inode->i_lock held!
231 static void nfs_invalidate_inode(struct inode *inode)
233 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
234 nfs_zap_caches_locked(inode);
237 struct nfs_find_desc {
239 struct nfs_fattr *fattr;
243 * In NFSv3 we can have 64bit inode numbers. In order to support
244 * this, and re-exported directories (also seen in NFSv2)
245 * we are forced to allow 2 different inodes to have the same
249 nfs_find_actor(struct inode *inode, void *opaque)
251 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
252 struct nfs_fh *fh = desc->fh;
253 struct nfs_fattr *fattr = desc->fattr;
255 if (NFS_FILEID(inode) != fattr->fileid)
257 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
259 if (nfs_compare_fh(NFS_FH(inode), fh))
261 if (is_bad_inode(inode) || NFS_STALE(inode))
267 nfs_init_locked(struct inode *inode, void *opaque)
269 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
270 struct nfs_fattr *fattr = desc->fattr;
272 set_nfs_fileid(inode, fattr->fileid);
273 nfs_copy_fh(NFS_FH(inode), desc->fh);
277 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
278 static void nfs_clear_label_invalid(struct inode *inode)
280 spin_lock(&inode->i_lock);
281 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
282 spin_unlock(&inode->i_lock);
285 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
286 struct nfs4_label *label)
293 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
294 error = security_inode_notifysecctx(inode, label->label,
297 printk(KERN_ERR "%s() %s %d "
298 "security_inode_notifysecctx() %d\n",
300 (char *)label->label,
302 nfs_clear_label_invalid(inode);
306 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
308 struct nfs4_label *label = NULL;
309 int minor_version = server->nfs_client->cl_minorversion;
311 if (minor_version < 2)
314 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
317 label = kzalloc(sizeof(struct nfs4_label), flags);
319 return ERR_PTR(-ENOMEM);
321 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
322 if (label->label == NULL) {
324 return ERR_PTR(-ENOMEM);
326 label->len = NFS4_MAXLABELLEN;
330 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
332 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
333 struct nfs4_label *label)
337 EXPORT_SYMBOL_GPL(nfs_setsecurity);
340 * This is our front-end to iget that looks up inodes by file handle
341 * instead of inode number.
344 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
346 struct nfs_find_desc desc = {
350 struct inode *inode = ERR_PTR(-ENOENT);
353 nfs_attr_check_mountpoint(sb, fattr);
355 if (nfs_attr_use_mounted_on_fileid(fattr))
356 fattr->fileid = fattr->mounted_on_fileid;
357 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
359 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
362 hash = nfs_fattr_to_ino_t(fattr);
364 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
366 inode = ERR_PTR(-ENOMEM);
370 if (inode->i_state & I_NEW) {
371 struct nfs_inode *nfsi = NFS_I(inode);
372 unsigned long now = jiffies;
374 /* We set i_ino for the few things that still rely on it,
378 /* We can't support update_atime(), since the server will reset it */
379 inode->i_flags |= S_NOATIME|S_NOCMTIME;
380 inode->i_mode = fattr->mode;
381 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
382 && nfs_server_capable(inode, NFS_CAP_MODE))
383 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
384 /* Why so? Because we want revalidate for devices/FIFOs, and
385 * that's precisely what we have in nfs_file_inode_operations.
387 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
388 if (S_ISREG(inode->i_mode)) {
389 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
390 inode->i_data.a_ops = &nfs_file_aops;
391 } else if (S_ISDIR(inode->i_mode)) {
392 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
393 inode->i_fop = &nfs_dir_operations;
394 inode->i_data.a_ops = &nfs_dir_aops;
395 /* Deal with crossing mountpoints */
396 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
397 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
398 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
399 inode->i_op = &nfs_referral_inode_operations;
401 inode->i_op = &nfs_mountpoint_inode_operations;
403 inode->i_flags |= S_AUTOMOUNT;
405 } else if (S_ISLNK(inode->i_mode))
406 inode->i_op = &nfs_symlink_inode_operations;
408 init_special_inode(inode, inode->i_mode, fattr->rdev);
410 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
411 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
412 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
413 inode->i_version = 0;
416 inode->i_uid = make_kuid(&init_user_ns, -2);
417 inode->i_gid = make_kgid(&init_user_ns, -2);
419 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
423 nfsi->read_cache_jiffies = fattr->time_start;
424 nfsi->attr_gencount = fattr->gencount;
425 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
426 inode->i_atime = fattr->atime;
427 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
428 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
429 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
430 inode->i_mtime = fattr->mtime;
431 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
432 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
433 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
434 inode->i_ctime = fattr->ctime;
435 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
436 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
437 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
438 inode->i_version = fattr->change_attr;
439 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
440 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
441 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
442 inode->i_size = nfs_size_to_loff_t(fattr->size);
444 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
445 | NFS_INO_REVAL_PAGECACHE);
446 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
447 set_nlink(inode, fattr->nlink);
448 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
449 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
450 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
451 inode->i_uid = fattr->uid;
452 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
453 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
454 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
455 inode->i_gid = fattr->gid;
456 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
457 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
458 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
459 inode->i_blocks = fattr->du.nfs2.blocks;
460 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
462 * report the blocks in 512byte units
464 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
467 nfs_setsecurity(inode, fattr, label);
469 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
470 nfsi->attrtimeo_timestamp = now;
471 nfsi->access_cache = RB_ROOT;
473 nfs_fscache_init_inode(inode);
475 unlock_new_inode(inode);
477 nfs_refresh_inode(inode, fattr);
478 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
480 (unsigned long long)NFS_FILEID(inode),
481 nfs_display_fhandle_hash(fh),
482 atomic_read(&inode->i_count));
488 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
491 EXPORT_SYMBOL_GPL(nfs_fhget);
493 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
496 nfs_setattr(struct dentry *dentry, struct iattr *attr)
498 struct inode *inode = dentry->d_inode;
499 struct nfs_fattr *fattr;
502 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
504 /* skip mode change if it's just for clearing setuid/setgid */
505 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
506 attr->ia_valid &= ~ATTR_MODE;
508 if (attr->ia_valid & ATTR_SIZE) {
511 BUG_ON(!S_ISREG(inode->i_mode));
513 i_size = i_size_read(inode);
514 if (attr->ia_size == i_size)
515 attr->ia_valid &= ~ATTR_SIZE;
516 else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
520 /* Optimization: if the end result is no change, don't RPC */
521 attr->ia_valid &= NFS_VALID_ATTRS;
522 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
525 trace_nfs_setattr_enter(inode);
527 /* Write all dirty data */
528 if (S_ISREG(inode->i_mode)) {
529 nfs_inode_dio_wait(inode);
533 fattr = nfs_alloc_fattr();
537 * Return any delegations if we're going to change ACLs
539 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
540 NFS_PROTO(inode)->return_delegation(inode);
541 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
543 error = nfs_refresh_inode(inode, fattr);
544 nfs_free_fattr(fattr);
546 trace_nfs_setattr_exit(inode, error);
549 EXPORT_SYMBOL_GPL(nfs_setattr);
552 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
553 * @inode: inode of the file used
554 * @offset: file offset to start truncating
556 * This is a copy of the common vmtruncate, but with the locking
557 * corrected to take into account the fact that NFS requires
558 * inode->i_size to be updated under the inode->i_lock.
560 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
564 err = inode_newsize_ok(inode, offset);
568 spin_lock(&inode->i_lock);
569 i_size_write(inode, offset);
572 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
573 spin_unlock(&inode->i_lock);
575 truncate_pagecache(inode, offset);
581 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
582 * @inode: pointer to struct inode
583 * @attr: pointer to struct iattr
585 * Note: we do this in the *proc.c in order to ensure that
586 * it works for things like exclusive creates too.
588 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
590 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
591 spin_lock(&inode->i_lock);
592 if ((attr->ia_valid & ATTR_MODE) != 0) {
593 int mode = attr->ia_mode & S_IALLUGO;
594 mode |= inode->i_mode & ~S_IALLUGO;
595 inode->i_mode = mode;
597 if ((attr->ia_valid & ATTR_UID) != 0)
598 inode->i_uid = attr->ia_uid;
599 if ((attr->ia_valid & ATTR_GID) != 0)
600 inode->i_gid = attr->ia_gid;
601 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
602 | NFS_INO_INVALID_ACL);
603 spin_unlock(&inode->i_lock);
605 if ((attr->ia_valid & ATTR_SIZE) != 0) {
606 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
607 nfs_vmtruncate(inode, attr->ia_size);
610 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
612 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
614 struct dentry *parent;
616 parent = dget_parent(dentry);
617 nfs_force_use_readdirplus(parent->d_inode);
621 static bool nfs_need_revalidate_inode(struct inode *inode)
623 if (NFS_I(inode)->cache_validity &
624 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
626 if (nfs_attribute_cache_expired(inode))
631 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
633 struct inode *inode = dentry->d_inode;
634 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
637 trace_nfs_getattr_enter(inode);
638 /* Flush out writes to the server in order to update c/mtime. */
639 if (S_ISREG(inode->i_mode)) {
640 nfs_inode_dio_wait(inode);
641 err = filemap_write_and_wait(inode->i_mapping);
647 * We may force a getattr if the user cares about atime.
649 * Note that we only have to check the vfsmount flags here:
650 * - NFS always sets S_NOATIME by so checking it would give a
652 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
653 * no point in checking those.
655 if ((mnt->mnt_flags & MNT_NOATIME) ||
656 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
659 if (need_atime || nfs_need_revalidate_inode(inode)) {
660 struct nfs_server *server = NFS_SERVER(inode);
662 if (server->caps & NFS_CAP_READDIRPLUS)
663 nfs_request_parent_use_readdirplus(dentry);
664 err = __nfs_revalidate_inode(server, inode);
667 generic_fillattr(inode, stat);
668 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
671 trace_nfs_getattr_exit(inode, err);
674 EXPORT_SYMBOL_GPL(nfs_getattr);
676 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
678 atomic_set(&l_ctx->count, 1);
679 l_ctx->lockowner.l_owner = current->files;
680 l_ctx->lockowner.l_pid = current->tgid;
681 INIT_LIST_HEAD(&l_ctx->list);
682 nfs_iocounter_init(&l_ctx->io_count);
685 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
687 struct nfs_lock_context *head = &ctx->lock_context;
688 struct nfs_lock_context *pos = head;
691 if (pos->lockowner.l_owner != current->files)
693 if (pos->lockowner.l_pid != current->tgid)
695 atomic_inc(&pos->count);
697 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
701 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
703 struct nfs_lock_context *res, *new = NULL;
704 struct inode *inode = ctx->dentry->d_inode;
706 spin_lock(&inode->i_lock);
707 res = __nfs_find_lock_context(ctx);
709 spin_unlock(&inode->i_lock);
710 new = kmalloc(sizeof(*new), GFP_KERNEL);
712 return ERR_PTR(-ENOMEM);
713 nfs_init_lock_context(new);
714 spin_lock(&inode->i_lock);
715 res = __nfs_find_lock_context(ctx);
717 list_add_tail(&new->list, &ctx->lock_context.list);
718 new->open_context = ctx;
723 spin_unlock(&inode->i_lock);
727 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
729 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
731 struct nfs_open_context *ctx = l_ctx->open_context;
732 struct inode *inode = ctx->dentry->d_inode;
734 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
736 list_del(&l_ctx->list);
737 spin_unlock(&inode->i_lock);
740 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
743 * nfs_close_context - Common close_context() routine NFSv2/v3
744 * @ctx: pointer to context
745 * @is_sync: is this a synchronous close
747 * always ensure that the attributes are up to date if we're mounted
748 * with close-to-open semantics
750 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
753 struct nfs_server *server;
755 if (!(ctx->mode & FMODE_WRITE))
759 inode = ctx->dentry->d_inode;
760 if (!list_empty(&NFS_I(inode)->open_files))
762 server = NFS_SERVER(inode);
763 if (server->flags & NFS_MOUNT_NOCTO)
765 nfs_revalidate_inode(server, inode);
767 EXPORT_SYMBOL_GPL(nfs_close_context);
769 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
771 struct nfs_open_context *ctx;
772 struct rpc_cred *cred = rpc_lookup_cred();
774 return ERR_CAST(cred);
776 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
779 return ERR_PTR(-ENOMEM);
781 nfs_sb_active(dentry->d_sb);
782 ctx->dentry = dget(dentry);
788 nfs_init_lock_context(&ctx->lock_context);
789 ctx->lock_context.open_context = ctx;
790 INIT_LIST_HEAD(&ctx->list);
791 ctx->mdsthreshold = NULL;
794 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
796 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
799 atomic_inc(&ctx->lock_context.count);
802 EXPORT_SYMBOL_GPL(get_nfs_open_context);
804 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
806 struct inode *inode = ctx->dentry->d_inode;
807 struct super_block *sb = ctx->dentry->d_sb;
809 if (!list_empty(&ctx->list)) {
810 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
812 list_del(&ctx->list);
813 spin_unlock(&inode->i_lock);
814 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
817 NFS_PROTO(inode)->close_context(ctx, is_sync);
818 if (ctx->cred != NULL)
819 put_rpccred(ctx->cred);
822 kfree(ctx->mdsthreshold);
826 void put_nfs_open_context(struct nfs_open_context *ctx)
828 __put_nfs_open_context(ctx, 0);
830 EXPORT_SYMBOL_GPL(put_nfs_open_context);
833 * Ensure that mmap has a recent RPC credential for use when writing out
836 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
838 struct inode *inode = ctx->dentry->d_inode;
839 struct nfs_inode *nfsi = NFS_I(inode);
841 spin_lock(&inode->i_lock);
842 list_add(&ctx->list, &nfsi->open_files);
843 spin_unlock(&inode->i_lock);
845 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
847 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
849 filp->private_data = get_nfs_open_context(ctx);
850 if (list_empty(&ctx->list))
851 nfs_inode_attach_open_context(ctx);
853 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
856 * Given an inode, search for an open context with the desired characteristics
858 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
860 struct nfs_inode *nfsi = NFS_I(inode);
861 struct nfs_open_context *pos, *ctx = NULL;
863 spin_lock(&inode->i_lock);
864 list_for_each_entry(pos, &nfsi->open_files, list) {
865 if (cred != NULL && pos->cred != cred)
867 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
869 ctx = get_nfs_open_context(pos);
872 spin_unlock(&inode->i_lock);
876 static void nfs_file_clear_open_context(struct file *filp)
878 struct nfs_open_context *ctx = nfs_file_open_context(filp);
881 struct inode *inode = ctx->dentry->d_inode;
883 filp->private_data = NULL;
884 spin_lock(&inode->i_lock);
885 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
886 spin_unlock(&inode->i_lock);
887 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
892 * These allocate and release file read/write context information.
894 int nfs_open(struct inode *inode, struct file *filp)
896 struct nfs_open_context *ctx;
898 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
901 nfs_file_set_open_context(filp, ctx);
902 put_nfs_open_context(ctx);
903 nfs_fscache_open_file(inode, filp);
907 int nfs_release(struct inode *inode, struct file *filp)
909 nfs_file_clear_open_context(filp);
914 * This function is called whenever some part of NFS notices that
915 * the cached attributes have to be refreshed.
918 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
920 int status = -ESTALE;
921 struct nfs4_label *label = NULL;
922 struct nfs_fattr *fattr = NULL;
923 struct nfs_inode *nfsi = NFS_I(inode);
925 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
926 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
928 trace_nfs_revalidate_inode_enter(inode);
930 if (is_bad_inode(inode))
932 if (NFS_STALE(inode))
936 fattr = nfs_alloc_fattr();
940 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
942 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
944 status = PTR_ERR(label);
948 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
950 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
952 (unsigned long long)NFS_FILEID(inode), status);
953 if (status == -ESTALE) {
954 nfs_zap_caches(inode);
955 if (!S_ISDIR(inode->i_mode))
956 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
961 status = nfs_refresh_inode(inode, fattr);
963 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
965 (unsigned long long)NFS_FILEID(inode), status);
969 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
970 nfs_zap_acl_cache(inode);
972 nfs_setsecurity(inode, fattr, label);
974 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
976 (unsigned long long)NFS_FILEID(inode));
979 nfs4_label_free(label);
981 nfs_free_fattr(fattr);
982 trace_nfs_revalidate_inode_exit(inode, status);
986 int nfs_attribute_timeout(struct inode *inode)
988 struct nfs_inode *nfsi = NFS_I(inode);
990 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
993 int nfs_attribute_cache_expired(struct inode *inode)
995 if (nfs_have_delegated_attributes(inode))
997 return nfs_attribute_timeout(inode);
1001 * nfs_revalidate_inode - Revalidate the inode attributes
1002 * @server - pointer to nfs_server struct
1003 * @inode - pointer to inode struct
1005 * Updates inode attribute information by retrieving the data from the server.
1007 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1009 if (!nfs_need_revalidate_inode(inode))
1010 return NFS_STALE(inode) ? -ESTALE : 0;
1011 return __nfs_revalidate_inode(server, inode);
1013 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1015 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1017 if (!(NFS_I(inode)->cache_validity &
1018 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1019 && !nfs_attribute_cache_expired(inode))
1020 return NFS_STALE(inode) ? -ESTALE : 0;
1024 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1026 struct nfs_inode *nfsi = NFS_I(inode);
1029 if (mapping->nrpages != 0) {
1030 if (S_ISREG(inode->i_mode)) {
1031 ret = nfs_sync_mapping(mapping);
1035 ret = invalidate_inode_pages2(mapping);
1039 if (S_ISDIR(inode->i_mode)) {
1040 spin_lock(&inode->i_lock);
1041 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1042 spin_unlock(&inode->i_lock);
1044 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1045 nfs_fscache_wait_on_invalidate(inode);
1047 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1049 (unsigned long long)NFS_FILEID(inode));
1053 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1055 if (nfs_have_delegated_attributes(inode))
1057 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1058 || nfs_attribute_timeout(inode)
1059 || NFS_STALE(inode);
1063 * nfs_revalidate_mapping - Revalidate the pagecache
1064 * @inode - pointer to host inode
1065 * @mapping - pointer to mapping
1067 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1069 struct nfs_inode *nfsi = NFS_I(inode);
1070 unsigned long *bitlock = &nfsi->flags;
1073 /* swapfiles are not supposed to be shared. */
1074 if (IS_SWAPFILE(inode))
1077 if (nfs_mapping_need_revalidate_inode(inode)) {
1078 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1084 * We must clear NFS_INO_INVALID_DATA first to ensure that
1085 * invalidations that come in while we're shooting down the mappings
1086 * are respected. But, that leaves a race window where one revalidator
1087 * can clear the flag, and then another checks it before the mapping
1088 * gets invalidated. Fix that by serializing access to this part of
1091 * At the same time, we need to allow other tasks to see whether we
1092 * might be in the middle of invalidating the pages, so we only set
1093 * the bit lock here if it looks like we're going to be doing that.
1096 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1097 nfs_wait_bit_killable, TASK_KILLABLE);
1100 spin_lock(&inode->i_lock);
1101 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1102 spin_unlock(&inode->i_lock);
1105 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1107 spin_unlock(&inode->i_lock);
1111 set_bit(NFS_INO_INVALIDATING, bitlock);
1113 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1114 spin_unlock(&inode->i_lock);
1115 trace_nfs_invalidate_mapping_enter(inode);
1116 ret = nfs_invalidate_mapping(inode, mapping);
1117 trace_nfs_invalidate_mapping_exit(inode, ret);
1119 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1120 smp_mb__after_atomic();
1121 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1126 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1128 struct nfs_inode *nfsi = NFS_I(inode);
1129 unsigned long ret = 0;
1131 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1132 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1133 && inode->i_version == fattr->pre_change_attr) {
1134 inode->i_version = fattr->change_attr;
1135 if (S_ISDIR(inode->i_mode))
1136 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1137 ret |= NFS_INO_INVALID_ATTR;
1139 /* If we have atomic WCC data, we may update some attributes */
1140 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1141 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1142 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1143 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1144 ret |= NFS_INO_INVALID_ATTR;
1147 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1148 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1149 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1150 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1151 if (S_ISDIR(inode->i_mode))
1152 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1153 ret |= NFS_INO_INVALID_ATTR;
1155 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1156 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1157 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1158 && nfsi->nrequests == 0) {
1159 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1160 ret |= NFS_INO_INVALID_ATTR;
1167 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1168 * @inode - pointer to inode
1169 * @fattr - updated attributes
1171 * Verifies the attribute cache. If we have just changed the attributes,
1172 * so that fattr carries weak cache consistency data, then it may
1173 * also update the ctime/mtime/change_attribute.
1175 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1177 struct nfs_inode *nfsi = NFS_I(inode);
1178 loff_t cur_size, new_isize;
1179 unsigned long invalid = 0;
1182 if (nfs_have_delegated_attributes(inode))
1184 /* Has the inode gone and changed behind our back? */
1185 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1187 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1190 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1191 inode->i_version != fattr->change_attr)
1192 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1194 /* Verify a few of the more important attributes */
1195 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1196 invalid |= NFS_INO_INVALID_ATTR;
1198 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1199 cur_size = i_size_read(inode);
1200 new_isize = nfs_size_to_loff_t(fattr->size);
1201 if (cur_size != new_isize && nfsi->nrequests == 0)
1202 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1205 /* Have any file permissions changed? */
1206 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1207 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1208 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1209 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1210 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1211 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1213 /* Has the link count changed? */
1214 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1215 invalid |= NFS_INO_INVALID_ATTR;
1217 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1218 invalid |= NFS_INO_INVALID_ATIME;
1221 nfs_set_cache_invalid(inode, invalid);
1223 nfsi->read_cache_jiffies = fattr->time_start;
1227 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1229 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1231 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1234 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1236 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1238 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1241 static atomic_long_t nfs_attr_generation_counter;
1243 static unsigned long nfs_read_attr_generation_counter(void)
1245 return atomic_long_read(&nfs_attr_generation_counter);
1248 unsigned long nfs_inc_attr_generation_counter(void)
1250 return atomic_long_inc_return(&nfs_attr_generation_counter);
1253 void nfs_fattr_init(struct nfs_fattr *fattr)
1256 fattr->time_start = jiffies;
1257 fattr->gencount = nfs_inc_attr_generation_counter();
1258 fattr->owner_name = NULL;
1259 fattr->group_name = NULL;
1261 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1263 struct nfs_fattr *nfs_alloc_fattr(void)
1265 struct nfs_fattr *fattr;
1267 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1269 nfs_fattr_init(fattr);
1272 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1274 struct nfs_fh *nfs_alloc_fhandle(void)
1278 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1283 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1287 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1288 * in the same way that wireshark does
1292 * For debugging only.
1294 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1296 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1297 * not on the result */
1298 return nfs_fhandle_hash(fh);
1300 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1303 * _nfs_display_fhandle - display an NFS file handle on the console
1305 * @fh: file handle to display
1306 * @caption: display caption
1308 * For debugging only.
1310 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1314 if (fh == NULL || fh->size == 0) {
1315 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1319 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1320 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1321 for (i = 0; i < fh->size; i += 16) {
1322 __be32 *pos = (__be32 *)&fh->data[i];
1324 switch ((fh->size - i - 1) >> 2) {
1326 printk(KERN_DEFAULT " %08x\n",
1330 printk(KERN_DEFAULT " %08x %08x\n",
1331 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1334 printk(KERN_DEFAULT " %08x %08x %08x\n",
1335 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1336 be32_to_cpup(pos + 2));
1339 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1340 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1341 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1345 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1349 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1350 * @inode - pointer to inode
1351 * @fattr - attributes
1353 * Attempt to divine whether or not an RPC call reply carrying stale
1354 * attributes got scheduled after another call carrying updated ones.
1356 * To do so, the function first assumes that a more recent ctime means
1357 * that the attributes in fattr are newer, however it also attempt to
1358 * catch the case where ctime either didn't change, or went backwards
1359 * (if someone reset the clock on the server) by looking at whether
1360 * or not this RPC call was started after the inode was last updated.
1361 * Note also the check for wraparound of 'attr_gencount'
1363 * The function returns 'true' if it thinks the attributes in 'fattr' are
1364 * more recent than the ones cached in the inode.
1367 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1369 const struct nfs_inode *nfsi = NFS_I(inode);
1371 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1372 nfs_ctime_need_update(inode, fattr) ||
1373 nfs_size_need_update(inode, fattr) ||
1374 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1378 * Don't trust the change_attribute, mtime, ctime or size if
1379 * a pnfs LAYOUTCOMMIT is outstanding
1381 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1382 struct nfs_fattr *fattr)
1384 if (pnfs_layoutcommit_outstanding(inode))
1385 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1386 NFS_ATTR_FATTR_MTIME |
1387 NFS_ATTR_FATTR_CTIME |
1388 NFS_ATTR_FATTR_SIZE);
1391 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1395 trace_nfs_refresh_inode_enter(inode);
1397 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1399 if (nfs_inode_attrs_need_update(inode, fattr))
1400 ret = nfs_update_inode(inode, fattr);
1402 ret = nfs_check_inode_attributes(inode, fattr);
1404 trace_nfs_refresh_inode_exit(inode, ret);
1409 * nfs_refresh_inode - try to update the inode attribute cache
1410 * @inode - pointer to inode
1411 * @fattr - updated attributes
1413 * Check that an RPC call that returned attributes has not overlapped with
1414 * other recent updates of the inode metadata, then decide whether it is
1415 * safe to do a full update of the inode attributes, or whether just to
1416 * call nfs_check_inode_attributes.
1418 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1422 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1424 spin_lock(&inode->i_lock);
1425 status = nfs_refresh_inode_locked(inode, fattr);
1426 spin_unlock(&inode->i_lock);
1430 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1432 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1434 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1436 if (S_ISDIR(inode->i_mode))
1437 invalid |= NFS_INO_INVALID_DATA;
1438 nfs_set_cache_invalid(inode, invalid);
1439 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1441 return nfs_refresh_inode_locked(inode, fattr);
1445 * nfs_post_op_update_inode - try to update the inode attribute cache
1446 * @inode - pointer to inode
1447 * @fattr - updated attributes
1449 * After an operation that has changed the inode metadata, mark the
1450 * attribute cache as being invalid, then try to update it.
1452 * NB: if the server didn't return any post op attributes, this
1453 * function will force the retrieval of attributes before the next
1454 * NFS request. Thus it should be used only for operations that
1455 * are expected to change one or more attributes, to avoid
1456 * unnecessary NFS requests and trips through nfs_update_inode().
1458 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1462 spin_lock(&inode->i_lock);
1463 status = nfs_post_op_update_inode_locked(inode, fattr);
1464 spin_unlock(&inode->i_lock);
1468 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1471 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1472 * @inode - pointer to inode
1473 * @fattr - updated attributes
1475 * After an operation that has changed the inode metadata, mark the
1476 * attribute cache as being invalid, then try to update it. Fake up
1477 * weak cache consistency data, if none exist.
1479 * This function is mainly designed to be used by the ->write_done() functions.
1481 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1485 spin_lock(&inode->i_lock);
1486 /* Don't do a WCC update if these attributes are already stale */
1487 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1488 !nfs_inode_attrs_need_update(inode, fattr)) {
1489 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1490 | NFS_ATTR_FATTR_PRESIZE
1491 | NFS_ATTR_FATTR_PREMTIME
1492 | NFS_ATTR_FATTR_PRECTIME);
1495 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1496 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1497 fattr->pre_change_attr = inode->i_version;
1498 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1500 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1501 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1502 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1503 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1505 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1506 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1507 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1508 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1510 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1511 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1512 fattr->pre_size = i_size_read(inode);
1513 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1516 status = nfs_post_op_update_inode_locked(inode, fattr);
1517 spin_unlock(&inode->i_lock);
1520 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1523 * Many nfs protocol calls return the new file attributes after
1524 * an operation. Here we update the inode to reflect the state
1525 * of the server's inode.
1527 * This is a bit tricky because we have to make sure all dirty pages
1528 * have been sent off to the server before calling invalidate_inode_pages.
1529 * To make sure no other process adds more write requests while we try
1530 * our best to flush them, we make them sleep during the attribute refresh.
1532 * A very similar scenario holds for the dir cache.
1534 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1536 struct nfs_server *server;
1537 struct nfs_inode *nfsi = NFS_I(inode);
1538 loff_t cur_isize, new_isize;
1539 unsigned long invalid = 0;
1540 unsigned long now = jiffies;
1541 unsigned long save_cache_validity;
1543 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1544 __func__, inode->i_sb->s_id, inode->i_ino,
1545 nfs_display_fhandle_hash(NFS_FH(inode)),
1546 atomic_read(&inode->i_count), fattr->valid);
1548 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1549 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1550 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1551 NFS_SERVER(inode)->nfs_client->cl_hostname,
1552 inode->i_sb->s_id, (long long)nfsi->fileid,
1553 (long long)fattr->fileid);
1558 * Make sure the inode's type hasn't changed.
1560 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1562 * Big trouble! The inode has become a different object.
1564 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1565 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1569 server = NFS_SERVER(inode);
1570 /* Update the fsid? */
1571 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1572 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1573 !IS_AUTOMOUNT(inode))
1574 server->fsid = fattr->fsid;
1577 * Update the read time so we don't revalidate too often.
1579 nfsi->read_cache_jiffies = fattr->time_start;
1581 save_cache_validity = nfsi->cache_validity;
1582 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1583 | NFS_INO_INVALID_ATIME
1584 | NFS_INO_REVAL_FORCED
1585 | NFS_INO_REVAL_PAGECACHE);
1587 /* Do atomic weak cache consistency updates */
1588 invalid |= nfs_wcc_update_inode(inode, fattr);
1590 /* More cache consistency checks */
1591 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1592 if (inode->i_version != fattr->change_attr) {
1593 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1594 inode->i_sb->s_id, inode->i_ino);
1595 invalid |= NFS_INO_INVALID_ATTR
1596 | NFS_INO_INVALID_DATA
1597 | NFS_INO_INVALID_ACCESS
1598 | NFS_INO_INVALID_ACL
1599 | NFS_INO_REVAL_PAGECACHE;
1600 if (S_ISDIR(inode->i_mode))
1601 nfs_force_lookup_revalidate(inode);
1602 inode->i_version = fattr->change_attr;
1604 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1605 nfsi->cache_validity |= save_cache_validity;
1607 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1608 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1609 } else if (server->caps & NFS_CAP_MTIME)
1610 nfsi->cache_validity |= save_cache_validity &
1611 (NFS_INO_INVALID_ATTR
1612 | NFS_INO_REVAL_FORCED);
1614 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1615 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1616 } else if (server->caps & NFS_CAP_CTIME)
1617 nfsi->cache_validity |= save_cache_validity &
1618 (NFS_INO_INVALID_ATTR
1619 | NFS_INO_REVAL_FORCED);
1621 /* Check if our cached file size is stale */
1622 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1623 new_isize = nfs_size_to_loff_t(fattr->size);
1624 cur_isize = i_size_read(inode);
1625 if (new_isize != cur_isize) {
1626 /* Do we perhaps have any outstanding writes, or has
1627 * the file grown beyond our last write? */
1628 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1629 i_size_write(inode, new_isize);
1630 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1631 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1633 dprintk("NFS: isize change on server for file %s/%ld "
1637 (long long)cur_isize,
1638 (long long)new_isize);
1641 nfsi->cache_validity |= save_cache_validity &
1642 (NFS_INO_INVALID_ATTR
1643 | NFS_INO_REVAL_PAGECACHE
1644 | NFS_INO_REVAL_FORCED);
1647 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1648 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1649 else if (server->caps & NFS_CAP_ATIME)
1650 nfsi->cache_validity |= save_cache_validity &
1651 (NFS_INO_INVALID_ATIME
1652 | NFS_INO_REVAL_FORCED);
1654 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1655 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1656 umode_t newmode = inode->i_mode & S_IFMT;
1657 newmode |= fattr->mode & S_IALLUGO;
1658 inode->i_mode = newmode;
1659 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1661 } else if (server->caps & NFS_CAP_MODE)
1662 nfsi->cache_validity |= save_cache_validity &
1663 (NFS_INO_INVALID_ATTR
1664 | NFS_INO_INVALID_ACCESS
1665 | NFS_INO_INVALID_ACL
1666 | NFS_INO_REVAL_FORCED);
1668 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1669 if (!uid_eq(inode->i_uid, fattr->uid)) {
1670 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1671 inode->i_uid = fattr->uid;
1673 } else if (server->caps & NFS_CAP_OWNER)
1674 nfsi->cache_validity |= save_cache_validity &
1675 (NFS_INO_INVALID_ATTR
1676 | NFS_INO_INVALID_ACCESS
1677 | NFS_INO_INVALID_ACL
1678 | NFS_INO_REVAL_FORCED);
1680 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1681 if (!gid_eq(inode->i_gid, fattr->gid)) {
1682 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1683 inode->i_gid = fattr->gid;
1685 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1686 nfsi->cache_validity |= save_cache_validity &
1687 (NFS_INO_INVALID_ATTR
1688 | NFS_INO_INVALID_ACCESS
1689 | NFS_INO_INVALID_ACL
1690 | NFS_INO_REVAL_FORCED);
1692 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1693 if (inode->i_nlink != fattr->nlink) {
1694 invalid |= NFS_INO_INVALID_ATTR;
1695 if (S_ISDIR(inode->i_mode))
1696 invalid |= NFS_INO_INVALID_DATA;
1697 set_nlink(inode, fattr->nlink);
1699 } else if (server->caps & NFS_CAP_NLINK)
1700 nfsi->cache_validity |= save_cache_validity &
1701 (NFS_INO_INVALID_ATTR
1702 | NFS_INO_REVAL_FORCED);
1704 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1706 * report the blocks in 512byte units
1708 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1710 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1711 inode->i_blocks = fattr->du.nfs2.blocks;
1713 /* Update attrtimeo value if we're out of the unstable period */
1714 if (invalid & NFS_INO_INVALID_ATTR) {
1715 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1716 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1717 nfsi->attrtimeo_timestamp = now;
1718 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1720 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1721 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1722 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1723 nfsi->attrtimeo_timestamp = now;
1726 invalid &= ~NFS_INO_INVALID_ATTR;
1727 /* Don't invalidate the data if we were to blame */
1728 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1729 || S_ISLNK(inode->i_mode)))
1730 invalid &= ~NFS_INO_INVALID_DATA;
1731 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1732 (save_cache_validity & NFS_INO_REVAL_FORCED))
1733 nfs_set_cache_invalid(inode, invalid);
1738 * No need to worry about unhashing the dentry, as the
1739 * lookup validation will know that the inode is bad.
1740 * (But we fall through to invalidate the caches.)
1742 nfs_invalidate_inode(inode);
1746 struct inode *nfs_alloc_inode(struct super_block *sb)
1748 struct nfs_inode *nfsi;
1749 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1753 nfsi->cache_validity = 0UL;
1754 #if IS_ENABLED(CONFIG_NFS_V4)
1755 nfsi->nfs4_acl = NULL;
1756 #endif /* CONFIG_NFS_V4 */
1757 return &nfsi->vfs_inode;
1759 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1761 static void nfs_i_callback(struct rcu_head *head)
1763 struct inode *inode = container_of(head, struct inode, i_rcu);
1764 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1767 void nfs_destroy_inode(struct inode *inode)
1769 call_rcu(&inode->i_rcu, nfs_i_callback);
1771 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1773 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1775 #if IS_ENABLED(CONFIG_NFS_V4)
1776 INIT_LIST_HEAD(&nfsi->open_states);
1777 nfsi->delegation = NULL;
1778 init_rwsem(&nfsi->rwsem);
1779 nfsi->layout = NULL;
1783 static void init_once(void *foo)
1785 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1787 inode_init_once(&nfsi->vfs_inode);
1788 INIT_LIST_HEAD(&nfsi->open_files);
1789 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1790 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1791 INIT_LIST_HEAD(&nfsi->commit_info.list);
1792 nfsi->nrequests = 0;
1793 nfsi->commit_info.ncommit = 0;
1794 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1795 atomic_set(&nfsi->silly_count, 1);
1796 INIT_HLIST_HEAD(&nfsi->silly_list);
1797 init_waitqueue_head(&nfsi->waitqueue);
1798 nfs4_init_once(nfsi);
1801 static int __init nfs_init_inodecache(void)
1803 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1804 sizeof(struct nfs_inode),
1805 0, (SLAB_RECLAIM_ACCOUNT|
1808 if (nfs_inode_cachep == NULL)
1814 static void nfs_destroy_inodecache(void)
1817 * Make sure all delayed rcu free inodes are flushed before we
1821 kmem_cache_destroy(nfs_inode_cachep);
1824 struct workqueue_struct *nfsiod_workqueue;
1825 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1828 * start up the nfsiod workqueue
1830 static int nfsiod_start(void)
1832 struct workqueue_struct *wq;
1833 dprintk("RPC: creating workqueue nfsiod\n");
1834 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1837 nfsiod_workqueue = wq;
1842 * Destroy the nfsiod workqueue
1844 static void nfsiod_stop(void)
1846 struct workqueue_struct *wq;
1848 wq = nfsiod_workqueue;
1851 nfsiod_workqueue = NULL;
1852 destroy_workqueue(wq);
1856 EXPORT_SYMBOL_GPL(nfs_net_id);
1858 static int nfs_net_init(struct net *net)
1860 nfs_clients_init(net);
1861 return nfs_fs_proc_net_init(net);
1864 static void nfs_net_exit(struct net *net)
1866 nfs_fs_proc_net_exit(net);
1867 nfs_cleanup_cb_ident_idr(net);
1870 static struct pernet_operations nfs_net_ops = {
1871 .init = nfs_net_init,
1872 .exit = nfs_net_exit,
1874 .size = sizeof(struct nfs_net),
1880 static int __init init_nfs_fs(void)
1884 err = register_pernet_subsys(&nfs_net_ops);
1888 err = nfs_fscache_register();
1892 err = nfsiod_start();
1896 err = nfs_fs_proc_init();
1900 err = nfs_init_nfspagecache();
1904 err = nfs_init_inodecache();
1908 err = nfs_init_readpagecache();
1912 err = nfs_init_writepagecache();
1916 err = nfs_init_directcache();
1920 #ifdef CONFIG_PROC_FS
1921 rpc_proc_register(&init_net, &nfs_rpcstat);
1923 if ((err = register_nfs_fs()) != 0)
1928 #ifdef CONFIG_PROC_FS
1929 rpc_proc_unregister(&init_net, "nfs");
1931 nfs_destroy_directcache();
1933 nfs_destroy_writepagecache();
1935 nfs_destroy_readpagecache();
1937 nfs_destroy_inodecache();
1939 nfs_destroy_nfspagecache();
1945 nfs_fscache_unregister();
1947 unregister_pernet_subsys(&nfs_net_ops);
1952 static void __exit exit_nfs_fs(void)
1954 nfs_destroy_directcache();
1955 nfs_destroy_writepagecache();
1956 nfs_destroy_readpagecache();
1957 nfs_destroy_inodecache();
1958 nfs_destroy_nfspagecache();
1959 nfs_fscache_unregister();
1960 unregister_pernet_subsys(&nfs_net_ops);
1961 #ifdef CONFIG_PROC_FS
1962 rpc_proc_unregister(&init_net, "nfs");
1964 unregister_nfs_fs();
1969 /* Not quite true; I just maintain it */
1970 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1971 MODULE_LICENSE("GPL");
1972 module_param(enable_ino64, bool, 0644);
1974 module_init(init_nfs_fs)
1975 module_exit(exit_nfs_fs)