4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
42 #include <asm/uaccess.h>
46 #include "delegation.h"
56 #define NFSDBG_FACILITY NFSDBG_VFS
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 static struct kmem_cache * nfs_inode_cachep;
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 return nfs_fileid_to_ino_t(fattr->fileid);
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
78 int nfs_wait_bit_killable(struct wait_bit_key *key)
80 if (fatal_signal_pending(current))
82 freezable_schedule_unsafe();
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
94 u64 nfs_compat_user_ino64(u64 fileid)
105 if (sizeof(ino) < sizeof(fileid))
106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
110 int nfs_drop_inode(struct inode *inode)
112 return NFS_STALE(inode) || generic_drop_inode(inode);
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
116 void nfs_clear_inode(struct inode *inode)
119 * The following should never happen...
121 WARN_ON_ONCE(nfs_have_writebacks(inode));
122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 nfs_zap_acl_cache(inode);
124 nfs_access_zap_cache(inode);
125 nfs_fscache_clear_inode(inode);
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
129 void nfs_evict_inode(struct inode *inode)
131 truncate_inode_pages_final(&inode->i_data);
133 nfs_clear_inode(inode);
136 int nfs_sync_inode(struct inode *inode)
138 nfs_inode_dio_wait(inode);
139 return nfs_wb_all(inode);
141 EXPORT_SYMBOL_GPL(nfs_sync_inode);
144 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
146 int nfs_sync_mapping(struct address_space *mapping)
150 if (mapping->nrpages != 0) {
151 unmap_mapping_range(mapping, 0, 0, 0);
152 ret = nfs_wb_all(mapping->host);
157 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
159 struct nfs_inode *nfsi = NFS_I(inode);
161 if (inode->i_mapping->nrpages == 0)
162 flags &= ~NFS_INO_INVALID_DATA;
163 nfsi->cache_validity |= flags;
164 if (flags & NFS_INO_INVALID_DATA)
165 nfs_fscache_invalidate(inode);
169 * Invalidate the local caches
171 static void nfs_zap_caches_locked(struct inode *inode)
173 struct nfs_inode *nfsi = NFS_I(inode);
174 int mode = inode->i_mode;
176 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
178 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
179 nfsi->attrtimeo_timestamp = jiffies;
181 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
182 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
183 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
184 | NFS_INO_INVALID_DATA
185 | NFS_INO_INVALID_ACCESS
186 | NFS_INO_INVALID_ACL
187 | NFS_INO_REVAL_PAGECACHE);
189 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
190 | NFS_INO_INVALID_ACCESS
191 | NFS_INO_INVALID_ACL
192 | NFS_INO_REVAL_PAGECACHE);
193 nfs_zap_label_cache_locked(nfsi);
196 void nfs_zap_caches(struct inode *inode)
198 spin_lock(&inode->i_lock);
199 nfs_zap_caches_locked(inode);
200 spin_unlock(&inode->i_lock);
203 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
205 if (mapping->nrpages != 0) {
206 spin_lock(&inode->i_lock);
207 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
208 spin_unlock(&inode->i_lock);
212 void nfs_zap_acl_cache(struct inode *inode)
214 void (*clear_acl_cache)(struct inode *);
216 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
217 if (clear_acl_cache != NULL)
218 clear_acl_cache(inode);
219 spin_lock(&inode->i_lock);
220 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
221 spin_unlock(&inode->i_lock);
223 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
225 void nfs_invalidate_atime(struct inode *inode)
227 spin_lock(&inode->i_lock);
228 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
229 spin_unlock(&inode->i_lock);
231 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
234 * Invalidate, but do not unhash, the inode.
235 * NB: must be called with inode->i_lock held!
237 static void nfs_invalidate_inode(struct inode *inode)
239 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
240 nfs_zap_caches_locked(inode);
243 struct nfs_find_desc {
245 struct nfs_fattr *fattr;
249 * In NFSv3 we can have 64bit inode numbers. In order to support
250 * this, and re-exported directories (also seen in NFSv2)
251 * we are forced to allow 2 different inodes to have the same
255 nfs_find_actor(struct inode *inode, void *opaque)
257 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
258 struct nfs_fh *fh = desc->fh;
259 struct nfs_fattr *fattr = desc->fattr;
261 if (NFS_FILEID(inode) != fattr->fileid)
263 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
265 if (nfs_compare_fh(NFS_FH(inode), fh))
267 if (is_bad_inode(inode) || NFS_STALE(inode))
273 nfs_init_locked(struct inode *inode, void *opaque)
275 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
276 struct nfs_fattr *fattr = desc->fattr;
278 set_nfs_fileid(inode, fattr->fileid);
279 nfs_copy_fh(NFS_FH(inode), desc->fh);
283 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
284 static void nfs_clear_label_invalid(struct inode *inode)
286 spin_lock(&inode->i_lock);
287 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
288 spin_unlock(&inode->i_lock);
291 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
292 struct nfs4_label *label)
299 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
300 error = security_inode_notifysecctx(inode, label->label,
303 printk(KERN_ERR "%s() %s %d "
304 "security_inode_notifysecctx() %d\n",
306 (char *)label->label,
308 nfs_clear_label_invalid(inode);
312 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
314 struct nfs4_label *label = NULL;
315 int minor_version = server->nfs_client->cl_minorversion;
317 if (minor_version < 2)
320 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
323 label = kzalloc(sizeof(struct nfs4_label), flags);
325 return ERR_PTR(-ENOMEM);
327 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
328 if (label->label == NULL) {
330 return ERR_PTR(-ENOMEM);
332 label->len = NFS4_MAXLABELLEN;
336 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
338 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
339 struct nfs4_label *label)
343 EXPORT_SYMBOL_GPL(nfs_setsecurity);
346 * This is our front-end to iget that looks up inodes by file handle
347 * instead of inode number.
350 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
352 struct nfs_find_desc desc = {
356 struct inode *inode = ERR_PTR(-ENOENT);
359 nfs_attr_check_mountpoint(sb, fattr);
361 if (nfs_attr_use_mounted_on_fileid(fattr))
362 fattr->fileid = fattr->mounted_on_fileid;
363 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
365 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
368 hash = nfs_fattr_to_ino_t(fattr);
370 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
372 inode = ERR_PTR(-ENOMEM);
376 if (inode->i_state & I_NEW) {
377 struct nfs_inode *nfsi = NFS_I(inode);
378 unsigned long now = jiffies;
380 /* We set i_ino for the few things that still rely on it,
384 /* We can't support update_atime(), since the server will reset it */
385 inode->i_flags |= S_NOATIME|S_NOCMTIME;
386 inode->i_mode = fattr->mode;
387 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
388 && nfs_server_capable(inode, NFS_CAP_MODE))
389 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
390 /* Why so? Because we want revalidate for devices/FIFOs, and
391 * that's precisely what we have in nfs_file_inode_operations.
393 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
394 if (S_ISREG(inode->i_mode)) {
395 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
396 inode->i_data.a_ops = &nfs_file_aops;
397 } else if (S_ISDIR(inode->i_mode)) {
398 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
399 inode->i_fop = &nfs_dir_operations;
400 inode->i_data.a_ops = &nfs_dir_aops;
401 /* Deal with crossing mountpoints */
402 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
403 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
404 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
405 inode->i_op = &nfs_referral_inode_operations;
407 inode->i_op = &nfs_mountpoint_inode_operations;
409 inode->i_flags |= S_AUTOMOUNT;
411 } else if (S_ISLNK(inode->i_mode))
412 inode->i_op = &nfs_symlink_inode_operations;
414 init_special_inode(inode, inode->i_mode, fattr->rdev);
416 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
417 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
418 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
419 inode->i_version = 0;
422 inode->i_uid = make_kuid(&init_user_ns, -2);
423 inode->i_gid = make_kgid(&init_user_ns, -2);
425 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
429 nfsi->read_cache_jiffies = fattr->time_start;
430 nfsi->attr_gencount = fattr->gencount;
431 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
432 inode->i_atime = fattr->atime;
433 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
434 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
435 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
436 inode->i_mtime = fattr->mtime;
437 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
438 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
439 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
440 inode->i_ctime = fattr->ctime;
441 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
442 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
443 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
444 inode->i_version = fattr->change_attr;
446 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
447 | NFS_INO_REVAL_PAGECACHE);
448 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
449 inode->i_size = nfs_size_to_loff_t(fattr->size);
451 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
452 | NFS_INO_REVAL_PAGECACHE);
453 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
454 set_nlink(inode, fattr->nlink);
455 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
456 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
457 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
458 inode->i_uid = fattr->uid;
459 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
460 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
461 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
462 inode->i_gid = fattr->gid;
463 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
464 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
465 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
466 inode->i_blocks = fattr->du.nfs2.blocks;
467 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
469 * report the blocks in 512byte units
471 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
474 nfs_setsecurity(inode, fattr, label);
476 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
477 nfsi->attrtimeo_timestamp = now;
478 nfsi->access_cache = RB_ROOT;
480 nfs_fscache_init_inode(inode);
482 unlock_new_inode(inode);
484 nfs_refresh_inode(inode, fattr);
485 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
487 (unsigned long long)NFS_FILEID(inode),
488 nfs_display_fhandle_hash(fh),
489 atomic_read(&inode->i_count));
495 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
498 EXPORT_SYMBOL_GPL(nfs_fhget);
500 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
503 nfs_setattr(struct dentry *dentry, struct iattr *attr)
505 struct inode *inode = d_inode(dentry);
506 struct nfs_fattr *fattr;
509 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
511 /* skip mode change if it's just for clearing setuid/setgid */
512 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
513 attr->ia_valid &= ~ATTR_MODE;
515 if (attr->ia_valid & ATTR_SIZE) {
516 BUG_ON(!S_ISREG(inode->i_mode));
518 error = inode_newsize_ok(inode, attr->ia_size);
522 if (attr->ia_size == i_size_read(inode))
523 attr->ia_valid &= ~ATTR_SIZE;
526 /* Optimization: if the end result is no change, don't RPC */
527 attr->ia_valid &= NFS_VALID_ATTRS;
528 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
531 trace_nfs_setattr_enter(inode);
533 /* Write all dirty data */
534 if (S_ISREG(inode->i_mode))
535 nfs_sync_inode(inode);
537 fattr = nfs_alloc_fattr();
544 * Return any delegations if we're going to change ACLs
546 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
547 NFS_PROTO(inode)->return_delegation(inode);
548 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
550 error = nfs_refresh_inode(inode, fattr);
551 nfs_free_fattr(fattr);
553 trace_nfs_setattr_exit(inode, error);
556 EXPORT_SYMBOL_GPL(nfs_setattr);
559 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
560 * @inode: inode of the file used
561 * @offset: file offset to start truncating
563 * This is a copy of the common vmtruncate, but with the locking
564 * corrected to take into account the fact that NFS requires
565 * inode->i_size to be updated under the inode->i_lock.
566 * Note: must be called with inode->i_lock held!
568 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
572 err = inode_newsize_ok(inode, offset);
576 i_size_write(inode, offset);
579 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
581 spin_unlock(&inode->i_lock);
582 truncate_pagecache(inode, offset);
583 spin_lock(&inode->i_lock);
589 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
590 * @inode: pointer to struct inode
591 * @attr: pointer to struct iattr
593 * Note: we do this in the *proc.c in order to ensure that
594 * it works for things like exclusive creates too.
596 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
597 struct nfs_fattr *fattr)
599 /* Barrier: bump the attribute generation count. */
600 nfs_fattr_set_barrier(fattr);
602 spin_lock(&inode->i_lock);
603 NFS_I(inode)->attr_gencount = fattr->gencount;
604 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
605 if ((attr->ia_valid & ATTR_MODE) != 0) {
606 int mode = attr->ia_mode & S_IALLUGO;
607 mode |= inode->i_mode & ~S_IALLUGO;
608 inode->i_mode = mode;
610 if ((attr->ia_valid & ATTR_UID) != 0)
611 inode->i_uid = attr->ia_uid;
612 if ((attr->ia_valid & ATTR_GID) != 0)
613 inode->i_gid = attr->ia_gid;
614 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
615 | NFS_INO_INVALID_ACL);
617 if ((attr->ia_valid & ATTR_SIZE) != 0) {
618 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
619 nfs_vmtruncate(inode, attr->ia_size);
622 nfs_update_inode(inode, fattr);
624 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
625 spin_unlock(&inode->i_lock);
627 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
629 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
631 struct dentry *parent;
633 parent = dget_parent(dentry);
634 nfs_force_use_readdirplus(d_inode(parent));
638 static bool nfs_need_revalidate_inode(struct inode *inode)
640 if (NFS_I(inode)->cache_validity &
641 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
643 if (nfs_attribute_cache_expired(inode))
648 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
650 struct inode *inode = d_inode(dentry);
651 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
654 trace_nfs_getattr_enter(inode);
655 /* Flush out writes to the server in order to update c/mtime. */
656 if (S_ISREG(inode->i_mode)) {
657 mutex_lock(&inode->i_mutex);
658 err = nfs_sync_inode(inode);
659 mutex_unlock(&inode->i_mutex);
665 * We may force a getattr if the user cares about atime.
667 * Note that we only have to check the vfsmount flags here:
668 * - NFS always sets S_NOATIME by so checking it would give a
670 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
671 * no point in checking those.
673 if ((mnt->mnt_flags & MNT_NOATIME) ||
674 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
677 if (need_atime || nfs_need_revalidate_inode(inode)) {
678 struct nfs_server *server = NFS_SERVER(inode);
680 if (server->caps & NFS_CAP_READDIRPLUS)
681 nfs_request_parent_use_readdirplus(dentry);
682 err = __nfs_revalidate_inode(server, inode);
685 generic_fillattr(inode, stat);
686 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
687 if (S_ISDIR(inode->i_mode))
688 stat->blksize = NFS_SERVER(inode)->dtsize;
691 trace_nfs_getattr_exit(inode, err);
694 EXPORT_SYMBOL_GPL(nfs_getattr);
696 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
698 atomic_set(&l_ctx->count, 1);
699 l_ctx->lockowner.l_owner = current->files;
700 l_ctx->lockowner.l_pid = current->tgid;
701 INIT_LIST_HEAD(&l_ctx->list);
702 nfs_iocounter_init(&l_ctx->io_count);
705 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
707 struct nfs_lock_context *head = &ctx->lock_context;
708 struct nfs_lock_context *pos = head;
711 if (pos->lockowner.l_owner != current->files)
713 if (pos->lockowner.l_pid != current->tgid)
715 atomic_inc(&pos->count);
717 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
721 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
723 struct nfs_lock_context *res, *new = NULL;
724 struct inode *inode = d_inode(ctx->dentry);
726 spin_lock(&inode->i_lock);
727 res = __nfs_find_lock_context(ctx);
729 spin_unlock(&inode->i_lock);
730 new = kmalloc(sizeof(*new), GFP_KERNEL);
732 return ERR_PTR(-ENOMEM);
733 nfs_init_lock_context(new);
734 spin_lock(&inode->i_lock);
735 res = __nfs_find_lock_context(ctx);
737 list_add_tail(&new->list, &ctx->lock_context.list);
738 new->open_context = ctx;
743 spin_unlock(&inode->i_lock);
747 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
749 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
751 struct nfs_open_context *ctx = l_ctx->open_context;
752 struct inode *inode = d_inode(ctx->dentry);
754 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
756 list_del(&l_ctx->list);
757 spin_unlock(&inode->i_lock);
760 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
763 * nfs_close_context - Common close_context() routine NFSv2/v3
764 * @ctx: pointer to context
765 * @is_sync: is this a synchronous close
767 * Ensure that the attributes are up to date if we're mounted
768 * with close-to-open semantics and we have cached data that will
769 * need to be revalidated on open.
771 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
773 struct nfs_inode *nfsi;
775 struct nfs_server *server;
777 if (!(ctx->mode & FMODE_WRITE))
781 inode = d_inode(ctx->dentry);
783 if (inode->i_mapping->nrpages == 0)
785 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
787 if (!list_empty(&nfsi->open_files))
789 server = NFS_SERVER(inode);
790 if (server->flags & NFS_MOUNT_NOCTO)
792 nfs_revalidate_inode(server, inode);
794 EXPORT_SYMBOL_GPL(nfs_close_context);
796 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
798 struct nfs_open_context *ctx;
799 struct rpc_cred *cred = rpc_lookup_cred();
801 return ERR_CAST(cred);
803 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
806 return ERR_PTR(-ENOMEM);
808 nfs_sb_active(dentry->d_sb);
809 ctx->dentry = dget(dentry);
815 nfs_init_lock_context(&ctx->lock_context);
816 ctx->lock_context.open_context = ctx;
817 INIT_LIST_HEAD(&ctx->list);
818 ctx->mdsthreshold = NULL;
821 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
823 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
826 atomic_inc(&ctx->lock_context.count);
829 EXPORT_SYMBOL_GPL(get_nfs_open_context);
831 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
833 struct inode *inode = d_inode(ctx->dentry);
834 struct super_block *sb = ctx->dentry->d_sb;
836 if (!list_empty(&ctx->list)) {
837 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
839 list_del(&ctx->list);
840 spin_unlock(&inode->i_lock);
841 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
844 NFS_PROTO(inode)->close_context(ctx, is_sync);
845 if (ctx->cred != NULL)
846 put_rpccred(ctx->cred);
849 kfree(ctx->mdsthreshold);
853 void put_nfs_open_context(struct nfs_open_context *ctx)
855 __put_nfs_open_context(ctx, 0);
857 EXPORT_SYMBOL_GPL(put_nfs_open_context);
859 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
861 __put_nfs_open_context(ctx, 1);
865 * Ensure that mmap has a recent RPC credential for use when writing out
868 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
870 struct inode *inode = d_inode(ctx->dentry);
871 struct nfs_inode *nfsi = NFS_I(inode);
873 spin_lock(&inode->i_lock);
874 list_add(&ctx->list, &nfsi->open_files);
875 spin_unlock(&inode->i_lock);
877 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
879 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
881 filp->private_data = get_nfs_open_context(ctx);
882 if (list_empty(&ctx->list))
883 nfs_inode_attach_open_context(ctx);
885 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
888 * Given an inode, search for an open context with the desired characteristics
890 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
892 struct nfs_inode *nfsi = NFS_I(inode);
893 struct nfs_open_context *pos, *ctx = NULL;
895 spin_lock(&inode->i_lock);
896 list_for_each_entry(pos, &nfsi->open_files, list) {
897 if (cred != NULL && pos->cred != cred)
899 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
901 ctx = get_nfs_open_context(pos);
904 spin_unlock(&inode->i_lock);
908 void nfs_file_clear_open_context(struct file *filp)
910 struct nfs_open_context *ctx = nfs_file_open_context(filp);
913 struct inode *inode = d_inode(ctx->dentry);
915 filp->private_data = NULL;
916 spin_lock(&inode->i_lock);
917 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
918 spin_unlock(&inode->i_lock);
919 put_nfs_open_context_sync(ctx);
924 * These allocate and release file read/write context information.
926 int nfs_open(struct inode *inode, struct file *filp)
928 struct nfs_open_context *ctx;
930 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
933 nfs_file_set_open_context(filp, ctx);
934 put_nfs_open_context(ctx);
935 nfs_fscache_open_file(inode, filp);
940 * This function is called whenever some part of NFS notices that
941 * the cached attributes have to be refreshed.
944 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
946 int status = -ESTALE;
947 struct nfs4_label *label = NULL;
948 struct nfs_fattr *fattr = NULL;
949 struct nfs_inode *nfsi = NFS_I(inode);
951 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
952 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
954 trace_nfs_revalidate_inode_enter(inode);
956 if (is_bad_inode(inode))
958 if (NFS_STALE(inode))
962 fattr = nfs_alloc_fattr();
966 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
968 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
970 status = PTR_ERR(label);
974 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
976 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
978 (unsigned long long)NFS_FILEID(inode), status);
979 if (status == -ESTALE) {
980 nfs_zap_caches(inode);
981 if (!S_ISDIR(inode->i_mode))
982 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
987 status = nfs_refresh_inode(inode, fattr);
989 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
991 (unsigned long long)NFS_FILEID(inode), status);
995 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
996 nfs_zap_acl_cache(inode);
998 nfs_setsecurity(inode, fattr, label);
1000 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1002 (unsigned long long)NFS_FILEID(inode));
1005 nfs4_label_free(label);
1007 nfs_free_fattr(fattr);
1008 trace_nfs_revalidate_inode_exit(inode, status);
1012 int nfs_attribute_timeout(struct inode *inode)
1014 struct nfs_inode *nfsi = NFS_I(inode);
1016 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1019 int nfs_attribute_cache_expired(struct inode *inode)
1021 if (nfs_have_delegated_attributes(inode))
1023 return nfs_attribute_timeout(inode);
1027 * nfs_revalidate_inode - Revalidate the inode attributes
1028 * @server - pointer to nfs_server struct
1029 * @inode - pointer to inode struct
1031 * Updates inode attribute information by retrieving the data from the server.
1033 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1035 if (!nfs_need_revalidate_inode(inode))
1036 return NFS_STALE(inode) ? -ESTALE : 0;
1037 return __nfs_revalidate_inode(server, inode);
1039 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1041 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1043 if (!(NFS_I(inode)->cache_validity &
1044 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1045 && !nfs_attribute_cache_expired(inode))
1046 return NFS_STALE(inode) ? -ESTALE : 0;
1050 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1052 struct nfs_inode *nfsi = NFS_I(inode);
1055 if (mapping->nrpages != 0) {
1056 if (S_ISREG(inode->i_mode)) {
1057 unmap_mapping_range(mapping, 0, 0, 0);
1058 ret = nfs_sync_mapping(mapping);
1062 ret = invalidate_inode_pages2(mapping);
1066 if (S_ISDIR(inode->i_mode)) {
1067 spin_lock(&inode->i_lock);
1068 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1069 spin_unlock(&inode->i_lock);
1071 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1072 nfs_fscache_wait_on_invalidate(inode);
1074 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1076 (unsigned long long)NFS_FILEID(inode));
1080 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1082 if (nfs_have_delegated_attributes(inode))
1084 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1085 || nfs_attribute_timeout(inode)
1086 || NFS_STALE(inode);
1090 * __nfs_revalidate_mapping - Revalidate the pagecache
1091 * @inode - pointer to host inode
1092 * @mapping - pointer to mapping
1093 * @may_lock - take inode->i_mutex?
1095 static int __nfs_revalidate_mapping(struct inode *inode,
1096 struct address_space *mapping,
1099 struct nfs_inode *nfsi = NFS_I(inode);
1100 unsigned long *bitlock = &nfsi->flags;
1103 /* swapfiles are not supposed to be shared. */
1104 if (IS_SWAPFILE(inode))
1107 if (nfs_mapping_need_revalidate_inode(inode)) {
1108 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1114 * We must clear NFS_INO_INVALID_DATA first to ensure that
1115 * invalidations that come in while we're shooting down the mappings
1116 * are respected. But, that leaves a race window where one revalidator
1117 * can clear the flag, and then another checks it before the mapping
1118 * gets invalidated. Fix that by serializing access to this part of
1121 * At the same time, we need to allow other tasks to see whether we
1122 * might be in the middle of invalidating the pages, so we only set
1123 * the bit lock here if it looks like we're going to be doing that.
1126 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1127 nfs_wait_bit_killable, TASK_KILLABLE);
1130 spin_lock(&inode->i_lock);
1131 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1132 spin_unlock(&inode->i_lock);
1135 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1137 spin_unlock(&inode->i_lock);
1141 set_bit(NFS_INO_INVALIDATING, bitlock);
1143 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1144 spin_unlock(&inode->i_lock);
1145 trace_nfs_invalidate_mapping_enter(inode);
1147 mutex_lock(&inode->i_mutex);
1148 ret = nfs_invalidate_mapping(inode, mapping);
1149 mutex_unlock(&inode->i_mutex);
1151 ret = nfs_invalidate_mapping(inode, mapping);
1152 trace_nfs_invalidate_mapping_exit(inode, ret);
1154 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1155 smp_mb__after_atomic();
1156 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1162 * nfs_revalidate_mapping - Revalidate the pagecache
1163 * @inode - pointer to host inode
1164 * @mapping - pointer to mapping
1166 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1168 return __nfs_revalidate_mapping(inode, mapping, false);
1172 * nfs_revalidate_mapping_protected - Revalidate the pagecache
1173 * @inode - pointer to host inode
1174 * @mapping - pointer to mapping
1176 * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1177 * while invalidating the mapping.
1179 int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1181 return __nfs_revalidate_mapping(inode, mapping, true);
1184 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1186 struct nfs_inode *nfsi = NFS_I(inode);
1187 unsigned long ret = 0;
1189 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1190 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1191 && inode->i_version == fattr->pre_change_attr) {
1192 inode->i_version = fattr->change_attr;
1193 if (S_ISDIR(inode->i_mode))
1194 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1195 ret |= NFS_INO_INVALID_ATTR;
1197 /* If we have atomic WCC data, we may update some attributes */
1198 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1199 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1200 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1201 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1202 ret |= NFS_INO_INVALID_ATTR;
1205 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1206 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1207 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1208 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1209 if (S_ISDIR(inode->i_mode))
1210 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1211 ret |= NFS_INO_INVALID_ATTR;
1213 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1214 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1215 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1216 && nfsi->nrequests == 0) {
1217 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1218 ret |= NFS_INO_INVALID_ATTR;
1225 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1226 * @inode - pointer to inode
1227 * @fattr - updated attributes
1229 * Verifies the attribute cache. If we have just changed the attributes,
1230 * so that fattr carries weak cache consistency data, then it may
1231 * also update the ctime/mtime/change_attribute.
1233 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1235 struct nfs_inode *nfsi = NFS_I(inode);
1236 loff_t cur_size, new_isize;
1237 unsigned long invalid = 0;
1240 if (nfs_have_delegated_attributes(inode))
1242 /* Has the inode gone and changed behind our back? */
1243 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1245 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1248 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1249 inode->i_version != fattr->change_attr)
1250 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1252 /* Verify a few of the more important attributes */
1253 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1254 invalid |= NFS_INO_INVALID_ATTR;
1256 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1257 cur_size = i_size_read(inode);
1258 new_isize = nfs_size_to_loff_t(fattr->size);
1259 if (cur_size != new_isize)
1260 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1262 if (nfsi->nrequests != 0)
1263 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1265 /* Have any file permissions changed? */
1266 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1267 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1268 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1269 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1270 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1271 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1273 /* Has the link count changed? */
1274 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1275 invalid |= NFS_INO_INVALID_ATTR;
1277 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1278 invalid |= NFS_INO_INVALID_ATIME;
1281 nfs_set_cache_invalid(inode, invalid);
1283 nfsi->read_cache_jiffies = fattr->time_start;
1287 static atomic_long_t nfs_attr_generation_counter;
1289 static unsigned long nfs_read_attr_generation_counter(void)
1291 return atomic_long_read(&nfs_attr_generation_counter);
1294 unsigned long nfs_inc_attr_generation_counter(void)
1296 return atomic_long_inc_return(&nfs_attr_generation_counter);
1298 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1300 void nfs_fattr_init(struct nfs_fattr *fattr)
1303 fattr->time_start = jiffies;
1304 fattr->gencount = nfs_inc_attr_generation_counter();
1305 fattr->owner_name = NULL;
1306 fattr->group_name = NULL;
1308 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1311 * nfs_fattr_set_barrier
1312 * @fattr: attributes
1314 * Used to set a barrier after an attribute was updated. This
1315 * barrier ensures that older attributes from RPC calls that may
1316 * have raced with our update cannot clobber these new values.
1317 * Note that you are still responsible for ensuring that other
1318 * operations which change the attribute on the server do not
1321 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1323 fattr->gencount = nfs_inc_attr_generation_counter();
1326 struct nfs_fattr *nfs_alloc_fattr(void)
1328 struct nfs_fattr *fattr;
1330 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1332 nfs_fattr_init(fattr);
1335 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1337 struct nfs_fh *nfs_alloc_fhandle(void)
1341 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1346 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1350 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1351 * in the same way that wireshark does
1355 * For debugging only.
1357 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1359 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1360 * not on the result */
1361 return nfs_fhandle_hash(fh);
1363 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1366 * _nfs_display_fhandle - display an NFS file handle on the console
1368 * @fh: file handle to display
1369 * @caption: display caption
1371 * For debugging only.
1373 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1377 if (fh == NULL || fh->size == 0) {
1378 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1382 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1383 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1384 for (i = 0; i < fh->size; i += 16) {
1385 __be32 *pos = (__be32 *)&fh->data[i];
1387 switch ((fh->size - i - 1) >> 2) {
1389 printk(KERN_DEFAULT " %08x\n",
1393 printk(KERN_DEFAULT " %08x %08x\n",
1394 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1397 printk(KERN_DEFAULT " %08x %08x %08x\n",
1398 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1399 be32_to_cpup(pos + 2));
1402 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1403 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1404 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1408 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1412 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1413 * @inode - pointer to inode
1414 * @fattr - attributes
1416 * Attempt to divine whether or not an RPC call reply carrying stale
1417 * attributes got scheduled after another call carrying updated ones.
1419 * To do so, the function first assumes that a more recent ctime means
1420 * that the attributes in fattr are newer, however it also attempt to
1421 * catch the case where ctime either didn't change, or went backwards
1422 * (if someone reset the clock on the server) by looking at whether
1423 * or not this RPC call was started after the inode was last updated.
1424 * Note also the check for wraparound of 'attr_gencount'
1426 * The function returns 'true' if it thinks the attributes in 'fattr' are
1427 * more recent than the ones cached in the inode.
1430 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1432 const struct nfs_inode *nfsi = NFS_I(inode);
1434 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1435 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1439 * Don't trust the change_attribute, mtime, ctime or size if
1440 * a pnfs LAYOUTCOMMIT is outstanding
1442 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1443 struct nfs_fattr *fattr)
1445 if (pnfs_layoutcommit_outstanding(inode))
1446 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1447 NFS_ATTR_FATTR_MTIME |
1448 NFS_ATTR_FATTR_CTIME |
1449 NFS_ATTR_FATTR_SIZE);
1452 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1456 trace_nfs_refresh_inode_enter(inode);
1458 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1460 if (nfs_inode_attrs_need_update(inode, fattr))
1461 ret = nfs_update_inode(inode, fattr);
1463 ret = nfs_check_inode_attributes(inode, fattr);
1465 trace_nfs_refresh_inode_exit(inode, ret);
1470 * nfs_refresh_inode - try to update the inode attribute cache
1471 * @inode - pointer to inode
1472 * @fattr - updated attributes
1474 * Check that an RPC call that returned attributes has not overlapped with
1475 * other recent updates of the inode metadata, then decide whether it is
1476 * safe to do a full update of the inode attributes, or whether just to
1477 * call nfs_check_inode_attributes.
1479 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1483 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1485 spin_lock(&inode->i_lock);
1486 status = nfs_refresh_inode_locked(inode, fattr);
1487 spin_unlock(&inode->i_lock);
1491 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1493 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1495 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1498 * Don't revalidate the pagecache if we hold a delegation, but do
1499 * force an attribute update
1501 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1502 invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1504 if (S_ISDIR(inode->i_mode))
1505 invalid |= NFS_INO_INVALID_DATA;
1506 nfs_set_cache_invalid(inode, invalid);
1507 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1509 return nfs_refresh_inode_locked(inode, fattr);
1513 * nfs_post_op_update_inode - try to update the inode attribute cache
1514 * @inode - pointer to inode
1515 * @fattr - updated attributes
1517 * After an operation that has changed the inode metadata, mark the
1518 * attribute cache as being invalid, then try to update it.
1520 * NB: if the server didn't return any post op attributes, this
1521 * function will force the retrieval of attributes before the next
1522 * NFS request. Thus it should be used only for operations that
1523 * are expected to change one or more attributes, to avoid
1524 * unnecessary NFS requests and trips through nfs_update_inode().
1526 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1530 spin_lock(&inode->i_lock);
1531 nfs_fattr_set_barrier(fattr);
1532 status = nfs_post_op_update_inode_locked(inode, fattr);
1533 spin_unlock(&inode->i_lock);
1537 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1540 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1541 * @inode - pointer to inode
1542 * @fattr - updated attributes
1544 * After an operation that has changed the inode metadata, mark the
1545 * attribute cache as being invalid, then try to update it. Fake up
1546 * weak cache consistency data, if none exist.
1548 * This function is mainly designed to be used by the ->write_done() functions.
1550 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1554 /* Don't do a WCC update if these attributes are already stale */
1555 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1556 !nfs_inode_attrs_need_update(inode, fattr)) {
1557 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1558 | NFS_ATTR_FATTR_PRESIZE
1559 | NFS_ATTR_FATTR_PREMTIME
1560 | NFS_ATTR_FATTR_PRECTIME);
1563 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1564 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1565 fattr->pre_change_attr = inode->i_version;
1566 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1568 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1569 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1570 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1571 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1573 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1574 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1575 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1576 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1578 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1579 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1580 fattr->pre_size = i_size_read(inode);
1581 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1584 status = nfs_post_op_update_inode_locked(inode, fattr);
1589 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1590 * @inode - pointer to inode
1591 * @fattr - updated attributes
1593 * After an operation that has changed the inode metadata, mark the
1594 * attribute cache as being invalid, then try to update it. Fake up
1595 * weak cache consistency data, if none exist.
1597 * This function is mainly designed to be used by the ->write_done() functions.
1599 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1603 spin_lock(&inode->i_lock);
1604 nfs_fattr_set_barrier(fattr);
1605 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1606 spin_unlock(&inode->i_lock);
1609 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1612 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1613 struct nfs_fattr *fattr)
1615 bool ret1 = true, ret2 = true;
1617 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1618 ret1 = (nfsi->fileid == fattr->fileid);
1619 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1620 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1621 return ret1 || ret2;
1625 * Many nfs protocol calls return the new file attributes after
1626 * an operation. Here we update the inode to reflect the state
1627 * of the server's inode.
1629 * This is a bit tricky because we have to make sure all dirty pages
1630 * have been sent off to the server before calling invalidate_inode_pages.
1631 * To make sure no other process adds more write requests while we try
1632 * our best to flush them, we make them sleep during the attribute refresh.
1634 * A very similar scenario holds for the dir cache.
1636 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1638 struct nfs_server *server;
1639 struct nfs_inode *nfsi = NFS_I(inode);
1640 loff_t cur_isize, new_isize;
1641 unsigned long invalid = 0;
1642 unsigned long now = jiffies;
1643 unsigned long save_cache_validity;
1645 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1646 __func__, inode->i_sb->s_id, inode->i_ino,
1647 nfs_display_fhandle_hash(NFS_FH(inode)),
1648 atomic_read(&inode->i_count), fattr->valid);
1650 if (!nfs_fileid_valid(nfsi, fattr)) {
1651 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1652 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1653 NFS_SERVER(inode)->nfs_client->cl_hostname,
1654 inode->i_sb->s_id, (long long)nfsi->fileid,
1655 (long long)fattr->fileid);
1660 * Make sure the inode's type hasn't changed.
1662 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1664 * Big trouble! The inode has become a different object.
1666 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1667 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1671 server = NFS_SERVER(inode);
1672 /* Update the fsid? */
1673 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1674 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1675 !IS_AUTOMOUNT(inode))
1676 server->fsid = fattr->fsid;
1679 * Update the read time so we don't revalidate too often.
1681 nfsi->read_cache_jiffies = fattr->time_start;
1683 save_cache_validity = nfsi->cache_validity;
1684 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1685 | NFS_INO_INVALID_ATIME
1686 | NFS_INO_REVAL_FORCED
1687 | NFS_INO_REVAL_PAGECACHE);
1689 /* Do atomic weak cache consistency updates */
1690 invalid |= nfs_wcc_update_inode(inode, fattr);
1692 /* More cache consistency checks */
1693 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1694 if (inode->i_version != fattr->change_attr) {
1695 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1696 inode->i_sb->s_id, inode->i_ino);
1697 invalid |= NFS_INO_INVALID_ATTR
1698 | NFS_INO_INVALID_DATA
1699 | NFS_INO_INVALID_ACCESS
1700 | NFS_INO_INVALID_ACL;
1701 if (S_ISDIR(inode->i_mode))
1702 nfs_force_lookup_revalidate(inode);
1703 inode->i_version = fattr->change_attr;
1706 nfsi->cache_validity |= save_cache_validity;
1708 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1709 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1710 } else if (server->caps & NFS_CAP_MTIME)
1711 nfsi->cache_validity |= save_cache_validity &
1712 (NFS_INO_INVALID_ATTR
1713 | NFS_INO_REVAL_FORCED);
1715 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1716 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1717 } else if (server->caps & NFS_CAP_CTIME)
1718 nfsi->cache_validity |= save_cache_validity &
1719 (NFS_INO_INVALID_ATTR
1720 | NFS_INO_REVAL_FORCED);
1722 /* Check if our cached file size is stale */
1723 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1724 new_isize = nfs_size_to_loff_t(fattr->size);
1725 cur_isize = i_size_read(inode);
1726 if (new_isize != cur_isize) {
1727 /* Do we perhaps have any outstanding writes, or has
1728 * the file grown beyond our last write? */
1729 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1730 i_size_write(inode, new_isize);
1731 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1733 dprintk("NFS: isize change on server for file %s/%ld "
1737 (long long)cur_isize,
1738 (long long)new_isize);
1741 nfsi->cache_validity |= save_cache_validity &
1742 (NFS_INO_INVALID_ATTR
1743 | NFS_INO_REVAL_PAGECACHE
1744 | NFS_INO_REVAL_FORCED);
1747 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1748 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1749 else if (server->caps & NFS_CAP_ATIME)
1750 nfsi->cache_validity |= save_cache_validity &
1751 (NFS_INO_INVALID_ATIME
1752 | NFS_INO_REVAL_FORCED);
1754 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1755 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1756 umode_t newmode = inode->i_mode & S_IFMT;
1757 newmode |= fattr->mode & S_IALLUGO;
1758 inode->i_mode = newmode;
1759 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1761 } else if (server->caps & NFS_CAP_MODE)
1762 nfsi->cache_validity |= save_cache_validity &
1763 (NFS_INO_INVALID_ATTR
1764 | NFS_INO_INVALID_ACCESS
1765 | NFS_INO_INVALID_ACL
1766 | NFS_INO_REVAL_FORCED);
1768 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1769 if (!uid_eq(inode->i_uid, fattr->uid)) {
1770 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1771 inode->i_uid = fattr->uid;
1773 } else if (server->caps & NFS_CAP_OWNER)
1774 nfsi->cache_validity |= save_cache_validity &
1775 (NFS_INO_INVALID_ATTR
1776 | NFS_INO_INVALID_ACCESS
1777 | NFS_INO_INVALID_ACL
1778 | NFS_INO_REVAL_FORCED);
1780 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1781 if (!gid_eq(inode->i_gid, fattr->gid)) {
1782 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1783 inode->i_gid = fattr->gid;
1785 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1786 nfsi->cache_validity |= save_cache_validity &
1787 (NFS_INO_INVALID_ATTR
1788 | NFS_INO_INVALID_ACCESS
1789 | NFS_INO_INVALID_ACL
1790 | NFS_INO_REVAL_FORCED);
1792 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1793 if (inode->i_nlink != fattr->nlink) {
1794 invalid |= NFS_INO_INVALID_ATTR;
1795 if (S_ISDIR(inode->i_mode))
1796 invalid |= NFS_INO_INVALID_DATA;
1797 set_nlink(inode, fattr->nlink);
1799 } else if (server->caps & NFS_CAP_NLINK)
1800 nfsi->cache_validity |= save_cache_validity &
1801 (NFS_INO_INVALID_ATTR
1802 | NFS_INO_REVAL_FORCED);
1804 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1806 * report the blocks in 512byte units
1808 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1810 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1811 inode->i_blocks = fattr->du.nfs2.blocks;
1813 /* Update attrtimeo value if we're out of the unstable period */
1814 if (invalid & NFS_INO_INVALID_ATTR) {
1815 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1816 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1817 nfsi->attrtimeo_timestamp = now;
1818 /* Set barrier to be more recent than all outstanding updates */
1819 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1821 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1822 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1823 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1824 nfsi->attrtimeo_timestamp = now;
1826 /* Set the barrier to be more recent than this fattr */
1827 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1828 nfsi->attr_gencount = fattr->gencount;
1831 /* Don't declare attrcache up to date if there were no attrs! */
1832 if (fattr->valid != 0)
1833 invalid &= ~NFS_INO_INVALID_ATTR;
1835 /* Don't invalidate the data if we were to blame */
1836 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1837 || S_ISLNK(inode->i_mode)))
1838 invalid &= ~NFS_INO_INVALID_DATA;
1839 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1840 (save_cache_validity & NFS_INO_REVAL_FORCED))
1841 nfs_set_cache_invalid(inode, invalid);
1846 * No need to worry about unhashing the dentry, as the
1847 * lookup validation will know that the inode is bad.
1848 * (But we fall through to invalidate the caches.)
1850 nfs_invalidate_inode(inode);
1854 struct inode *nfs_alloc_inode(struct super_block *sb)
1856 struct nfs_inode *nfsi;
1857 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1861 nfsi->cache_validity = 0UL;
1862 #if IS_ENABLED(CONFIG_NFS_V4)
1863 nfsi->nfs4_acl = NULL;
1864 #endif /* CONFIG_NFS_V4 */
1865 return &nfsi->vfs_inode;
1867 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1869 static void nfs_i_callback(struct rcu_head *head)
1871 struct inode *inode = container_of(head, struct inode, i_rcu);
1872 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1875 void nfs_destroy_inode(struct inode *inode)
1877 call_rcu(&inode->i_rcu, nfs_i_callback);
1879 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1881 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1883 #if IS_ENABLED(CONFIG_NFS_V4)
1884 INIT_LIST_HEAD(&nfsi->open_states);
1885 nfsi->delegation = NULL;
1886 init_rwsem(&nfsi->rwsem);
1887 nfsi->layout = NULL;
1891 static void init_once(void *foo)
1893 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1895 inode_init_once(&nfsi->vfs_inode);
1896 INIT_LIST_HEAD(&nfsi->open_files);
1897 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1898 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1899 INIT_LIST_HEAD(&nfsi->commit_info.list);
1900 nfsi->nrequests = 0;
1901 nfsi->commit_info.ncommit = 0;
1902 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1903 atomic_set(&nfsi->silly_count, 1);
1904 INIT_HLIST_HEAD(&nfsi->silly_list);
1905 init_waitqueue_head(&nfsi->waitqueue);
1906 nfs4_init_once(nfsi);
1909 static int __init nfs_init_inodecache(void)
1911 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1912 sizeof(struct nfs_inode),
1913 0, (SLAB_RECLAIM_ACCOUNT|
1916 if (nfs_inode_cachep == NULL)
1922 static void nfs_destroy_inodecache(void)
1925 * Make sure all delayed rcu free inodes are flushed before we
1929 kmem_cache_destroy(nfs_inode_cachep);
1932 struct workqueue_struct *nfsiod_workqueue;
1933 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1936 * start up the nfsiod workqueue
1938 static int nfsiod_start(void)
1940 struct workqueue_struct *wq;
1941 dprintk("RPC: creating workqueue nfsiod\n");
1942 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1945 nfsiod_workqueue = wq;
1950 * Destroy the nfsiod workqueue
1952 static void nfsiod_stop(void)
1954 struct workqueue_struct *wq;
1956 wq = nfsiod_workqueue;
1959 nfsiod_workqueue = NULL;
1960 destroy_workqueue(wq);
1964 EXPORT_SYMBOL_GPL(nfs_net_id);
1966 static int nfs_net_init(struct net *net)
1968 nfs_clients_init(net);
1969 return nfs_fs_proc_net_init(net);
1972 static void nfs_net_exit(struct net *net)
1974 nfs_fs_proc_net_exit(net);
1975 nfs_cleanup_cb_ident_idr(net);
1978 static struct pernet_operations nfs_net_ops = {
1979 .init = nfs_net_init,
1980 .exit = nfs_net_exit,
1982 .size = sizeof(struct nfs_net),
1988 static int __init init_nfs_fs(void)
1992 err = register_pernet_subsys(&nfs_net_ops);
1996 err = nfs_fscache_register();
2000 err = nfsiod_start();
2004 err = nfs_fs_proc_init();
2008 err = nfs_init_nfspagecache();
2012 err = nfs_init_inodecache();
2016 err = nfs_init_readpagecache();
2020 err = nfs_init_writepagecache();
2024 err = nfs_init_directcache();
2028 rpc_proc_register(&init_net, &nfs_rpcstat);
2030 err = register_nfs_fs();
2036 rpc_proc_unregister(&init_net, "nfs");
2037 nfs_destroy_directcache();
2039 nfs_destroy_writepagecache();
2041 nfs_destroy_readpagecache();
2043 nfs_destroy_inodecache();
2045 nfs_destroy_nfspagecache();
2051 nfs_fscache_unregister();
2053 unregister_pernet_subsys(&nfs_net_ops);
2058 static void __exit exit_nfs_fs(void)
2060 nfs_destroy_directcache();
2061 nfs_destroy_writepagecache();
2062 nfs_destroy_readpagecache();
2063 nfs_destroy_inodecache();
2064 nfs_destroy_nfspagecache();
2065 nfs_fscache_unregister();
2066 unregister_pernet_subsys(&nfs_net_ops);
2067 rpc_proc_unregister(&init_net, "nfs");
2068 unregister_nfs_fs();
2073 /* Not quite true; I just maintain it */
2074 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2075 MODULE_LICENSE("GPL");
2076 module_param(enable_ino64, bool, 0644);
2078 module_init(init_nfs_fs)
2079 module_exit(exit_nfs_fs)