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;
445 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
446 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
447 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
448 inode->i_size = nfs_size_to_loff_t(fattr->size);
450 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
451 | NFS_INO_REVAL_PAGECACHE);
452 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
453 set_nlink(inode, fattr->nlink);
454 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
455 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
456 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
457 inode->i_uid = fattr->uid;
458 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
459 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
460 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
461 inode->i_gid = fattr->gid;
462 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
463 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
464 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
465 inode->i_blocks = fattr->du.nfs2.blocks;
466 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
468 * report the blocks in 512byte units
470 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
473 nfs_setsecurity(inode, fattr, label);
475 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
476 nfsi->attrtimeo_timestamp = now;
477 nfsi->access_cache = RB_ROOT;
479 nfs_fscache_init_inode(inode);
481 unlock_new_inode(inode);
483 nfs_refresh_inode(inode, fattr);
484 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
486 (unsigned long long)NFS_FILEID(inode),
487 nfs_display_fhandle_hash(fh),
488 atomic_read(&inode->i_count));
494 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
497 EXPORT_SYMBOL_GPL(nfs_fhget);
499 #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)
502 nfs_setattr(struct dentry *dentry, struct iattr *attr)
504 struct inode *inode = d_inode(dentry);
505 struct nfs_fattr *fattr;
508 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
510 /* skip mode change if it's just for clearing setuid/setgid */
511 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
512 attr->ia_valid &= ~ATTR_MODE;
514 if (attr->ia_valid & ATTR_SIZE) {
517 BUG_ON(!S_ISREG(inode->i_mode));
519 i_size = i_size_read(inode);
520 if (attr->ia_size == i_size)
521 attr->ia_valid &= ~ATTR_SIZE;
522 else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
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();
541 * Return any delegations if we're going to change ACLs
543 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
544 NFS_PROTO(inode)->return_delegation(inode);
545 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
547 error = nfs_refresh_inode(inode, fattr);
548 nfs_free_fattr(fattr);
550 trace_nfs_setattr_exit(inode, error);
553 EXPORT_SYMBOL_GPL(nfs_setattr);
556 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
557 * @inode: inode of the file used
558 * @offset: file offset to start truncating
560 * This is a copy of the common vmtruncate, but with the locking
561 * corrected to take into account the fact that NFS requires
562 * inode->i_size to be updated under the inode->i_lock.
563 * Note: must be called with inode->i_lock held!
565 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
569 err = inode_newsize_ok(inode, offset);
573 i_size_write(inode, offset);
576 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
578 spin_unlock(&inode->i_lock);
579 truncate_pagecache(inode, offset);
580 spin_lock(&inode->i_lock);
586 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
587 * @inode: pointer to struct inode
588 * @attr: pointer to struct iattr
590 * Note: we do this in the *proc.c in order to ensure that
591 * it works for things like exclusive creates too.
593 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
594 struct nfs_fattr *fattr)
596 /* Barrier: bump the attribute generation count. */
597 nfs_fattr_set_barrier(fattr);
599 spin_lock(&inode->i_lock);
600 NFS_I(inode)->attr_gencount = fattr->gencount;
601 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
602 if ((attr->ia_valid & ATTR_MODE) != 0) {
603 int mode = attr->ia_mode & S_IALLUGO;
604 mode |= inode->i_mode & ~S_IALLUGO;
605 inode->i_mode = mode;
607 if ((attr->ia_valid & ATTR_UID) != 0)
608 inode->i_uid = attr->ia_uid;
609 if ((attr->ia_valid & ATTR_GID) != 0)
610 inode->i_gid = attr->ia_gid;
611 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
612 | NFS_INO_INVALID_ACL);
614 if ((attr->ia_valid & ATTR_SIZE) != 0) {
615 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
616 nfs_vmtruncate(inode, attr->ia_size);
618 nfs_update_inode(inode, fattr);
619 spin_unlock(&inode->i_lock);
621 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
623 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
625 struct dentry *parent;
627 parent = dget_parent(dentry);
628 nfs_force_use_readdirplus(d_inode(parent));
632 static bool nfs_need_revalidate_inode(struct inode *inode)
634 if (NFS_I(inode)->cache_validity &
635 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
637 if (nfs_attribute_cache_expired(inode))
642 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
644 struct inode *inode = d_inode(dentry);
645 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
648 trace_nfs_getattr_enter(inode);
649 /* Flush out writes to the server in order to update c/mtime. */
650 if (S_ISREG(inode->i_mode)) {
651 mutex_lock(&inode->i_mutex);
652 err = nfs_sync_inode(inode);
653 mutex_unlock(&inode->i_mutex);
659 * We may force a getattr if the user cares about atime.
661 * Note that we only have to check the vfsmount flags here:
662 * - NFS always sets S_NOATIME by so checking it would give a
664 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
665 * no point in checking those.
667 if ((mnt->mnt_flags & MNT_NOATIME) ||
668 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
671 if (need_atime || nfs_need_revalidate_inode(inode)) {
672 struct nfs_server *server = NFS_SERVER(inode);
674 if (server->caps & NFS_CAP_READDIRPLUS)
675 nfs_request_parent_use_readdirplus(dentry);
676 err = __nfs_revalidate_inode(server, inode);
679 generic_fillattr(inode, stat);
680 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
683 trace_nfs_getattr_exit(inode, err);
686 EXPORT_SYMBOL_GPL(nfs_getattr);
688 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
690 atomic_set(&l_ctx->count, 1);
691 l_ctx->lockowner.l_owner = current->files;
692 l_ctx->lockowner.l_pid = current->tgid;
693 INIT_LIST_HEAD(&l_ctx->list);
694 nfs_iocounter_init(&l_ctx->io_count);
697 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
699 struct nfs_lock_context *head = &ctx->lock_context;
700 struct nfs_lock_context *pos = head;
703 if (pos->lockowner.l_owner != current->files)
705 if (pos->lockowner.l_pid != current->tgid)
707 atomic_inc(&pos->count);
709 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
713 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
715 struct nfs_lock_context *res, *new = NULL;
716 struct inode *inode = d_inode(ctx->dentry);
718 spin_lock(&inode->i_lock);
719 res = __nfs_find_lock_context(ctx);
721 spin_unlock(&inode->i_lock);
722 new = kmalloc(sizeof(*new), GFP_KERNEL);
724 return ERR_PTR(-ENOMEM);
725 nfs_init_lock_context(new);
726 spin_lock(&inode->i_lock);
727 res = __nfs_find_lock_context(ctx);
729 list_add_tail(&new->list, &ctx->lock_context.list);
730 new->open_context = ctx;
735 spin_unlock(&inode->i_lock);
739 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
741 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
743 struct nfs_open_context *ctx = l_ctx->open_context;
744 struct inode *inode = d_inode(ctx->dentry);
746 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
748 list_del(&l_ctx->list);
749 spin_unlock(&inode->i_lock);
752 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
755 * nfs_close_context - Common close_context() routine NFSv2/v3
756 * @ctx: pointer to context
757 * @is_sync: is this a synchronous close
759 * always ensure that the attributes are up to date if we're mounted
760 * with close-to-open semantics
762 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
765 struct nfs_server *server;
767 if (!(ctx->mode & FMODE_WRITE))
771 inode = d_inode(ctx->dentry);
772 if (!list_empty(&NFS_I(inode)->open_files))
774 server = NFS_SERVER(inode);
775 if (server->flags & NFS_MOUNT_NOCTO)
777 nfs_revalidate_inode(server, inode);
779 EXPORT_SYMBOL_GPL(nfs_close_context);
781 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
783 struct nfs_open_context *ctx;
784 struct rpc_cred *cred = rpc_lookup_cred();
786 return ERR_CAST(cred);
788 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
791 return ERR_PTR(-ENOMEM);
793 nfs_sb_active(dentry->d_sb);
794 ctx->dentry = dget(dentry);
800 nfs_init_lock_context(&ctx->lock_context);
801 ctx->lock_context.open_context = ctx;
802 INIT_LIST_HEAD(&ctx->list);
803 ctx->mdsthreshold = NULL;
806 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
808 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
811 atomic_inc(&ctx->lock_context.count);
814 EXPORT_SYMBOL_GPL(get_nfs_open_context);
816 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
818 struct inode *inode = d_inode(ctx->dentry);
819 struct super_block *sb = ctx->dentry->d_sb;
821 if (!list_empty(&ctx->list)) {
822 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
824 list_del(&ctx->list);
825 spin_unlock(&inode->i_lock);
826 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
829 NFS_PROTO(inode)->close_context(ctx, is_sync);
830 if (ctx->cred != NULL)
831 put_rpccred(ctx->cred);
834 kfree(ctx->mdsthreshold);
838 void put_nfs_open_context(struct nfs_open_context *ctx)
840 __put_nfs_open_context(ctx, 0);
842 EXPORT_SYMBOL_GPL(put_nfs_open_context);
845 * Ensure that mmap has a recent RPC credential for use when writing out
848 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
850 struct inode *inode = d_inode(ctx->dentry);
851 struct nfs_inode *nfsi = NFS_I(inode);
853 spin_lock(&inode->i_lock);
854 list_add(&ctx->list, &nfsi->open_files);
855 spin_unlock(&inode->i_lock);
857 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
859 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
861 filp->private_data = get_nfs_open_context(ctx);
862 if (list_empty(&ctx->list))
863 nfs_inode_attach_open_context(ctx);
865 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
868 * Given an inode, search for an open context with the desired characteristics
870 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
872 struct nfs_inode *nfsi = NFS_I(inode);
873 struct nfs_open_context *pos, *ctx = NULL;
875 spin_lock(&inode->i_lock);
876 list_for_each_entry(pos, &nfsi->open_files, list) {
877 if (cred != NULL && pos->cred != cred)
879 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
881 ctx = get_nfs_open_context(pos);
884 spin_unlock(&inode->i_lock);
888 static void nfs_file_clear_open_context(struct file *filp)
890 struct nfs_open_context *ctx = nfs_file_open_context(filp);
893 struct inode *inode = d_inode(ctx->dentry);
895 filp->private_data = NULL;
896 spin_lock(&inode->i_lock);
897 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
898 spin_unlock(&inode->i_lock);
899 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
904 * These allocate and release file read/write context information.
906 int nfs_open(struct inode *inode, struct file *filp)
908 struct nfs_open_context *ctx;
910 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
913 nfs_file_set_open_context(filp, ctx);
914 put_nfs_open_context(ctx);
915 nfs_fscache_open_file(inode, filp);
919 int nfs_release(struct inode *inode, struct file *filp)
921 nfs_file_clear_open_context(filp);
926 * This function is called whenever some part of NFS notices that
927 * the cached attributes have to be refreshed.
930 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
932 int status = -ESTALE;
933 struct nfs4_label *label = NULL;
934 struct nfs_fattr *fattr = NULL;
935 struct nfs_inode *nfsi = NFS_I(inode);
937 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
938 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
940 trace_nfs_revalidate_inode_enter(inode);
942 if (is_bad_inode(inode))
944 if (NFS_STALE(inode))
948 fattr = nfs_alloc_fattr();
952 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
954 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
956 status = PTR_ERR(label);
960 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
962 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
964 (unsigned long long)NFS_FILEID(inode), status);
965 if (status == -ESTALE) {
966 nfs_zap_caches(inode);
967 if (!S_ISDIR(inode->i_mode))
968 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
973 status = nfs_refresh_inode(inode, fattr);
975 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
977 (unsigned long long)NFS_FILEID(inode), status);
981 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
982 nfs_zap_acl_cache(inode);
984 nfs_setsecurity(inode, fattr, label);
986 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
988 (unsigned long long)NFS_FILEID(inode));
991 nfs4_label_free(label);
993 nfs_free_fattr(fattr);
994 trace_nfs_revalidate_inode_exit(inode, status);
998 int nfs_attribute_timeout(struct inode *inode)
1000 struct nfs_inode *nfsi = NFS_I(inode);
1002 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1005 int nfs_attribute_cache_expired(struct inode *inode)
1007 if (nfs_have_delegated_attributes(inode))
1009 return nfs_attribute_timeout(inode);
1013 * nfs_revalidate_inode - Revalidate the inode attributes
1014 * @server - pointer to nfs_server struct
1015 * @inode - pointer to inode struct
1017 * Updates inode attribute information by retrieving the data from the server.
1019 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1021 if (!nfs_need_revalidate_inode(inode))
1022 return NFS_STALE(inode) ? -ESTALE : 0;
1023 return __nfs_revalidate_inode(server, inode);
1025 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1027 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1029 if (!(NFS_I(inode)->cache_validity &
1030 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1031 && !nfs_attribute_cache_expired(inode))
1032 return NFS_STALE(inode) ? -ESTALE : 0;
1036 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1038 struct nfs_inode *nfsi = NFS_I(inode);
1041 if (mapping->nrpages != 0) {
1042 if (S_ISREG(inode->i_mode)) {
1043 unmap_mapping_range(mapping, 0, 0, 0);
1044 ret = nfs_sync_mapping(mapping);
1048 ret = invalidate_inode_pages2(mapping);
1052 if (S_ISDIR(inode->i_mode)) {
1053 spin_lock(&inode->i_lock);
1054 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1055 spin_unlock(&inode->i_lock);
1057 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1058 nfs_fscache_wait_on_invalidate(inode);
1060 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1062 (unsigned long long)NFS_FILEID(inode));
1066 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1068 if (nfs_have_delegated_attributes(inode))
1070 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1071 || nfs_attribute_timeout(inode)
1072 || NFS_STALE(inode);
1076 * __nfs_revalidate_mapping - Revalidate the pagecache
1077 * @inode - pointer to host inode
1078 * @mapping - pointer to mapping
1079 * @may_lock - take inode->i_mutex?
1081 static int __nfs_revalidate_mapping(struct inode *inode,
1082 struct address_space *mapping,
1085 struct nfs_inode *nfsi = NFS_I(inode);
1086 unsigned long *bitlock = &nfsi->flags;
1089 /* swapfiles are not supposed to be shared. */
1090 if (IS_SWAPFILE(inode))
1093 if (nfs_mapping_need_revalidate_inode(inode)) {
1094 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1100 * We must clear NFS_INO_INVALID_DATA first to ensure that
1101 * invalidations that come in while we're shooting down the mappings
1102 * are respected. But, that leaves a race window where one revalidator
1103 * can clear the flag, and then another checks it before the mapping
1104 * gets invalidated. Fix that by serializing access to this part of
1107 * At the same time, we need to allow other tasks to see whether we
1108 * might be in the middle of invalidating the pages, so we only set
1109 * the bit lock here if it looks like we're going to be doing that.
1112 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1113 nfs_wait_bit_killable, TASK_KILLABLE);
1116 spin_lock(&inode->i_lock);
1117 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1118 spin_unlock(&inode->i_lock);
1121 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1123 spin_unlock(&inode->i_lock);
1127 set_bit(NFS_INO_INVALIDATING, bitlock);
1129 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1130 spin_unlock(&inode->i_lock);
1131 trace_nfs_invalidate_mapping_enter(inode);
1133 mutex_lock(&inode->i_mutex);
1134 ret = nfs_invalidate_mapping(inode, mapping);
1135 mutex_unlock(&inode->i_mutex);
1137 ret = nfs_invalidate_mapping(inode, mapping);
1138 trace_nfs_invalidate_mapping_exit(inode, ret);
1140 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1141 smp_mb__after_atomic();
1142 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1148 * nfs_revalidate_mapping - Revalidate the pagecache
1149 * @inode - pointer to host inode
1150 * @mapping - pointer to mapping
1152 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1154 return __nfs_revalidate_mapping(inode, mapping, false);
1158 * nfs_revalidate_mapping_protected - Revalidate the pagecache
1159 * @inode - pointer to host inode
1160 * @mapping - pointer to mapping
1162 * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1163 * while invalidating the mapping.
1165 int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1167 return __nfs_revalidate_mapping(inode, mapping, true);
1170 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1172 struct nfs_inode *nfsi = NFS_I(inode);
1173 unsigned long ret = 0;
1175 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1176 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1177 && inode->i_version == fattr->pre_change_attr) {
1178 inode->i_version = fattr->change_attr;
1179 if (S_ISDIR(inode->i_mode))
1180 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1181 ret |= NFS_INO_INVALID_ATTR;
1183 /* If we have atomic WCC data, we may update some attributes */
1184 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1185 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1186 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1187 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1188 ret |= NFS_INO_INVALID_ATTR;
1191 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1192 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1193 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1194 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1195 if (S_ISDIR(inode->i_mode))
1196 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1197 ret |= NFS_INO_INVALID_ATTR;
1199 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1200 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1201 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1202 && nfsi->nrequests == 0) {
1203 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1204 ret |= NFS_INO_INVALID_ATTR;
1211 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1212 * @inode - pointer to inode
1213 * @fattr - updated attributes
1215 * Verifies the attribute cache. If we have just changed the attributes,
1216 * so that fattr carries weak cache consistency data, then it may
1217 * also update the ctime/mtime/change_attribute.
1219 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1221 struct nfs_inode *nfsi = NFS_I(inode);
1222 loff_t cur_size, new_isize;
1223 unsigned long invalid = 0;
1226 if (nfs_have_delegated_attributes(inode))
1228 /* Has the inode gone and changed behind our back? */
1229 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1231 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1234 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1235 inode->i_version != fattr->change_attr)
1236 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1238 /* Verify a few of the more important attributes */
1239 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1240 invalid |= NFS_INO_INVALID_ATTR;
1242 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1243 cur_size = i_size_read(inode);
1244 new_isize = nfs_size_to_loff_t(fattr->size);
1245 if (cur_size != new_isize)
1246 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1248 if (nfsi->nrequests != 0)
1249 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1251 /* Have any file permissions changed? */
1252 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1253 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1254 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1255 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1256 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1257 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1259 /* Has the link count changed? */
1260 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1261 invalid |= NFS_INO_INVALID_ATTR;
1263 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1264 invalid |= NFS_INO_INVALID_ATIME;
1267 nfs_set_cache_invalid(inode, invalid);
1269 nfsi->read_cache_jiffies = fattr->time_start;
1273 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1275 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1277 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1280 static atomic_long_t nfs_attr_generation_counter;
1282 static unsigned long nfs_read_attr_generation_counter(void)
1284 return atomic_long_read(&nfs_attr_generation_counter);
1287 unsigned long nfs_inc_attr_generation_counter(void)
1289 return atomic_long_inc_return(&nfs_attr_generation_counter);
1291 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1293 void nfs_fattr_init(struct nfs_fattr *fattr)
1296 fattr->time_start = jiffies;
1297 fattr->gencount = nfs_inc_attr_generation_counter();
1298 fattr->owner_name = NULL;
1299 fattr->group_name = NULL;
1301 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1304 * nfs_fattr_set_barrier
1305 * @fattr: attributes
1307 * Used to set a barrier after an attribute was updated. This
1308 * barrier ensures that older attributes from RPC calls that may
1309 * have raced with our update cannot clobber these new values.
1310 * Note that you are still responsible for ensuring that other
1311 * operations which change the attribute on the server do not
1314 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1316 fattr->gencount = nfs_inc_attr_generation_counter();
1319 struct nfs_fattr *nfs_alloc_fattr(void)
1321 struct nfs_fattr *fattr;
1323 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1325 nfs_fattr_init(fattr);
1328 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1330 struct nfs_fh *nfs_alloc_fhandle(void)
1334 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1339 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1343 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1344 * in the same way that wireshark does
1348 * For debugging only.
1350 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1352 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1353 * not on the result */
1354 return nfs_fhandle_hash(fh);
1356 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1359 * _nfs_display_fhandle - display an NFS file handle on the console
1361 * @fh: file handle to display
1362 * @caption: display caption
1364 * For debugging only.
1366 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1370 if (fh == NULL || fh->size == 0) {
1371 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1375 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1376 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1377 for (i = 0; i < fh->size; i += 16) {
1378 __be32 *pos = (__be32 *)&fh->data[i];
1380 switch ((fh->size - i - 1) >> 2) {
1382 printk(KERN_DEFAULT " %08x\n",
1386 printk(KERN_DEFAULT " %08x %08x\n",
1387 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1390 printk(KERN_DEFAULT " %08x %08x %08x\n",
1391 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1392 be32_to_cpup(pos + 2));
1395 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1396 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1397 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1401 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1405 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1406 * @inode - pointer to inode
1407 * @fattr - attributes
1409 * Attempt to divine whether or not an RPC call reply carrying stale
1410 * attributes got scheduled after another call carrying updated ones.
1412 * To do so, the function first assumes that a more recent ctime means
1413 * that the attributes in fattr are newer, however it also attempt to
1414 * catch the case where ctime either didn't change, or went backwards
1415 * (if someone reset the clock on the server) by looking at whether
1416 * or not this RPC call was started after the inode was last updated.
1417 * Note also the check for wraparound of 'attr_gencount'
1419 * The function returns 'true' if it thinks the attributes in 'fattr' are
1420 * more recent than the ones cached in the inode.
1423 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1425 const struct nfs_inode *nfsi = NFS_I(inode);
1427 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1428 nfs_ctime_need_update(inode, fattr) ||
1429 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1433 * Don't trust the change_attribute, mtime, ctime or size if
1434 * a pnfs LAYOUTCOMMIT is outstanding
1436 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1437 struct nfs_fattr *fattr)
1439 if (pnfs_layoutcommit_outstanding(inode))
1440 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1441 NFS_ATTR_FATTR_MTIME |
1442 NFS_ATTR_FATTR_CTIME |
1443 NFS_ATTR_FATTR_SIZE);
1446 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1450 trace_nfs_refresh_inode_enter(inode);
1452 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1454 if (nfs_inode_attrs_need_update(inode, fattr))
1455 ret = nfs_update_inode(inode, fattr);
1457 ret = nfs_check_inode_attributes(inode, fattr);
1459 trace_nfs_refresh_inode_exit(inode, ret);
1464 * nfs_refresh_inode - try to update the inode attribute cache
1465 * @inode - pointer to inode
1466 * @fattr - updated attributes
1468 * Check that an RPC call that returned attributes has not overlapped with
1469 * other recent updates of the inode metadata, then decide whether it is
1470 * safe to do a full update of the inode attributes, or whether just to
1471 * call nfs_check_inode_attributes.
1473 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1477 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1479 spin_lock(&inode->i_lock);
1480 status = nfs_refresh_inode_locked(inode, fattr);
1481 spin_unlock(&inode->i_lock);
1485 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1487 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1489 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1491 if (S_ISDIR(inode->i_mode))
1492 invalid |= NFS_INO_INVALID_DATA;
1493 nfs_set_cache_invalid(inode, invalid);
1494 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1496 return nfs_refresh_inode_locked(inode, fattr);
1500 * nfs_post_op_update_inode - try to update the inode attribute cache
1501 * @inode - pointer to inode
1502 * @fattr - updated attributes
1504 * After an operation that has changed the inode metadata, mark the
1505 * attribute cache as being invalid, then try to update it.
1507 * NB: if the server didn't return any post op attributes, this
1508 * function will force the retrieval of attributes before the next
1509 * NFS request. Thus it should be used only for operations that
1510 * are expected to change one or more attributes, to avoid
1511 * unnecessary NFS requests and trips through nfs_update_inode().
1513 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1517 spin_lock(&inode->i_lock);
1518 nfs_fattr_set_barrier(fattr);
1519 status = nfs_post_op_update_inode_locked(inode, fattr);
1520 spin_unlock(&inode->i_lock);
1524 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1527 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1528 * @inode - pointer to inode
1529 * @fattr - updated attributes
1531 * After an operation that has changed the inode metadata, mark the
1532 * attribute cache as being invalid, then try to update it. Fake up
1533 * weak cache consistency data, if none exist.
1535 * This function is mainly designed to be used by the ->write_done() functions.
1537 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1541 /* Don't do a WCC update if these attributes are already stale */
1542 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1543 !nfs_inode_attrs_need_update(inode, fattr)) {
1544 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1545 | NFS_ATTR_FATTR_PRESIZE
1546 | NFS_ATTR_FATTR_PREMTIME
1547 | NFS_ATTR_FATTR_PRECTIME);
1550 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1551 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1552 fattr->pre_change_attr = inode->i_version;
1553 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1555 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1556 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1557 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1558 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1560 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1561 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1562 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1563 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1565 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1566 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1567 fattr->pre_size = i_size_read(inode);
1568 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1571 status = nfs_post_op_update_inode_locked(inode, fattr);
1576 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1577 * @inode - pointer to inode
1578 * @fattr - updated attributes
1580 * After an operation that has changed the inode metadata, mark the
1581 * attribute cache as being invalid, then try to update it. Fake up
1582 * weak cache consistency data, if none exist.
1584 * This function is mainly designed to be used by the ->write_done() functions.
1586 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1590 spin_lock(&inode->i_lock);
1591 nfs_fattr_set_barrier(fattr);
1592 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1593 spin_unlock(&inode->i_lock);
1596 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1599 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1600 struct nfs_fattr *fattr)
1602 bool ret1 = true, ret2 = true;
1604 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1605 ret1 = (nfsi->fileid == fattr->fileid);
1606 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1607 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1608 return ret1 || ret2;
1612 * Many nfs protocol calls return the new file attributes after
1613 * an operation. Here we update the inode to reflect the state
1614 * of the server's inode.
1616 * This is a bit tricky because we have to make sure all dirty pages
1617 * have been sent off to the server before calling invalidate_inode_pages.
1618 * To make sure no other process adds more write requests while we try
1619 * our best to flush them, we make them sleep during the attribute refresh.
1621 * A very similar scenario holds for the dir cache.
1623 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1625 struct nfs_server *server;
1626 struct nfs_inode *nfsi = NFS_I(inode);
1627 loff_t cur_isize, new_isize;
1628 unsigned long invalid = 0;
1629 unsigned long now = jiffies;
1630 unsigned long save_cache_validity;
1632 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1633 __func__, inode->i_sb->s_id, inode->i_ino,
1634 nfs_display_fhandle_hash(NFS_FH(inode)),
1635 atomic_read(&inode->i_count), fattr->valid);
1637 if (!nfs_fileid_valid(nfsi, fattr)) {
1638 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1639 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1640 NFS_SERVER(inode)->nfs_client->cl_hostname,
1641 inode->i_sb->s_id, (long long)nfsi->fileid,
1642 (long long)fattr->fileid);
1647 * Make sure the inode's type hasn't changed.
1649 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1651 * Big trouble! The inode has become a different object.
1653 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1654 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1658 server = NFS_SERVER(inode);
1659 /* Update the fsid? */
1660 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1661 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1662 !IS_AUTOMOUNT(inode))
1663 server->fsid = fattr->fsid;
1666 * Update the read time so we don't revalidate too often.
1668 nfsi->read_cache_jiffies = fattr->time_start;
1670 save_cache_validity = nfsi->cache_validity;
1671 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1672 | NFS_INO_INVALID_ATIME
1673 | NFS_INO_REVAL_FORCED
1674 | NFS_INO_REVAL_PAGECACHE);
1676 /* Do atomic weak cache consistency updates */
1677 invalid |= nfs_wcc_update_inode(inode, fattr);
1679 /* More cache consistency checks */
1680 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1681 if (inode->i_version != fattr->change_attr) {
1682 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1683 inode->i_sb->s_id, inode->i_ino);
1684 invalid |= NFS_INO_INVALID_ATTR
1685 | NFS_INO_INVALID_DATA
1686 | NFS_INO_INVALID_ACCESS
1687 | NFS_INO_INVALID_ACL;
1688 if (S_ISDIR(inode->i_mode))
1689 nfs_force_lookup_revalidate(inode);
1690 inode->i_version = fattr->change_attr;
1692 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1693 nfsi->cache_validity |= save_cache_validity;
1695 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1696 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1697 } else if (server->caps & NFS_CAP_MTIME)
1698 nfsi->cache_validity |= save_cache_validity &
1699 (NFS_INO_INVALID_ATTR
1700 | NFS_INO_REVAL_FORCED);
1702 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1703 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1704 } else if (server->caps & NFS_CAP_CTIME)
1705 nfsi->cache_validity |= save_cache_validity &
1706 (NFS_INO_INVALID_ATTR
1707 | NFS_INO_REVAL_FORCED);
1709 /* Check if our cached file size is stale */
1710 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1711 new_isize = nfs_size_to_loff_t(fattr->size);
1712 cur_isize = i_size_read(inode);
1713 if (new_isize != cur_isize) {
1714 /* Do we perhaps have any outstanding writes, or has
1715 * the file grown beyond our last write? */
1716 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1717 i_size_write(inode, new_isize);
1718 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1720 dprintk("NFS: isize change on server for file %s/%ld "
1724 (long long)cur_isize,
1725 (long long)new_isize);
1728 nfsi->cache_validity |= save_cache_validity &
1729 (NFS_INO_INVALID_ATTR
1730 | NFS_INO_REVAL_PAGECACHE
1731 | NFS_INO_REVAL_FORCED);
1734 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1735 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1736 else if (server->caps & NFS_CAP_ATIME)
1737 nfsi->cache_validity |= save_cache_validity &
1738 (NFS_INO_INVALID_ATIME
1739 | NFS_INO_REVAL_FORCED);
1741 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1742 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1743 umode_t newmode = inode->i_mode & S_IFMT;
1744 newmode |= fattr->mode & S_IALLUGO;
1745 inode->i_mode = newmode;
1746 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1748 } else if (server->caps & NFS_CAP_MODE)
1749 nfsi->cache_validity |= save_cache_validity &
1750 (NFS_INO_INVALID_ATTR
1751 | NFS_INO_INVALID_ACCESS
1752 | NFS_INO_INVALID_ACL
1753 | NFS_INO_REVAL_FORCED);
1755 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1756 if (!uid_eq(inode->i_uid, fattr->uid)) {
1757 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1758 inode->i_uid = fattr->uid;
1760 } else if (server->caps & NFS_CAP_OWNER)
1761 nfsi->cache_validity |= save_cache_validity &
1762 (NFS_INO_INVALID_ATTR
1763 | NFS_INO_INVALID_ACCESS
1764 | NFS_INO_INVALID_ACL
1765 | NFS_INO_REVAL_FORCED);
1767 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1768 if (!gid_eq(inode->i_gid, fattr->gid)) {
1769 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1770 inode->i_gid = fattr->gid;
1772 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1773 nfsi->cache_validity |= save_cache_validity &
1774 (NFS_INO_INVALID_ATTR
1775 | NFS_INO_INVALID_ACCESS
1776 | NFS_INO_INVALID_ACL
1777 | NFS_INO_REVAL_FORCED);
1779 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1780 if (inode->i_nlink != fattr->nlink) {
1781 invalid |= NFS_INO_INVALID_ATTR;
1782 if (S_ISDIR(inode->i_mode))
1783 invalid |= NFS_INO_INVALID_DATA;
1784 set_nlink(inode, fattr->nlink);
1786 } else if (server->caps & NFS_CAP_NLINK)
1787 nfsi->cache_validity |= save_cache_validity &
1788 (NFS_INO_INVALID_ATTR
1789 | NFS_INO_REVAL_FORCED);
1791 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1793 * report the blocks in 512byte units
1795 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1797 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1798 inode->i_blocks = fattr->du.nfs2.blocks;
1800 /* Update attrtimeo value if we're out of the unstable period */
1801 if (invalid & NFS_INO_INVALID_ATTR) {
1802 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1803 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1804 nfsi->attrtimeo_timestamp = now;
1805 /* Set barrier to be more recent than all outstanding updates */
1806 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1808 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1809 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1810 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1811 nfsi->attrtimeo_timestamp = now;
1813 /* Set the barrier to be more recent than this fattr */
1814 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1815 nfsi->attr_gencount = fattr->gencount;
1817 invalid &= ~NFS_INO_INVALID_ATTR;
1818 /* Don't invalidate the data if we were to blame */
1819 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1820 || S_ISLNK(inode->i_mode)))
1821 invalid &= ~NFS_INO_INVALID_DATA;
1822 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1823 (save_cache_validity & NFS_INO_REVAL_FORCED))
1824 nfs_set_cache_invalid(inode, invalid);
1829 * No need to worry about unhashing the dentry, as the
1830 * lookup validation will know that the inode is bad.
1831 * (But we fall through to invalidate the caches.)
1833 nfs_invalidate_inode(inode);
1837 struct inode *nfs_alloc_inode(struct super_block *sb)
1839 struct nfs_inode *nfsi;
1840 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1844 nfsi->cache_validity = 0UL;
1845 #if IS_ENABLED(CONFIG_NFS_V4)
1846 nfsi->nfs4_acl = NULL;
1847 #endif /* CONFIG_NFS_V4 */
1848 return &nfsi->vfs_inode;
1850 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1852 static void nfs_i_callback(struct rcu_head *head)
1854 struct inode *inode = container_of(head, struct inode, i_rcu);
1855 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1858 void nfs_destroy_inode(struct inode *inode)
1860 call_rcu(&inode->i_rcu, nfs_i_callback);
1862 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1864 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1866 #if IS_ENABLED(CONFIG_NFS_V4)
1867 INIT_LIST_HEAD(&nfsi->open_states);
1868 nfsi->delegation = NULL;
1869 init_rwsem(&nfsi->rwsem);
1870 nfsi->layout = NULL;
1874 static void init_once(void *foo)
1876 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1878 inode_init_once(&nfsi->vfs_inode);
1879 INIT_LIST_HEAD(&nfsi->open_files);
1880 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1881 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1882 INIT_LIST_HEAD(&nfsi->commit_info.list);
1883 nfsi->nrequests = 0;
1884 nfsi->commit_info.ncommit = 0;
1885 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1886 atomic_set(&nfsi->silly_count, 1);
1887 INIT_HLIST_HEAD(&nfsi->silly_list);
1888 init_waitqueue_head(&nfsi->waitqueue);
1889 nfs4_init_once(nfsi);
1892 static int __init nfs_init_inodecache(void)
1894 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1895 sizeof(struct nfs_inode),
1896 0, (SLAB_RECLAIM_ACCOUNT|
1899 if (nfs_inode_cachep == NULL)
1905 static void nfs_destroy_inodecache(void)
1908 * Make sure all delayed rcu free inodes are flushed before we
1912 kmem_cache_destroy(nfs_inode_cachep);
1915 struct workqueue_struct *nfsiod_workqueue;
1916 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1919 * start up the nfsiod workqueue
1921 static int nfsiod_start(void)
1923 struct workqueue_struct *wq;
1924 dprintk("RPC: creating workqueue nfsiod\n");
1925 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1928 nfsiod_workqueue = wq;
1933 * Destroy the nfsiod workqueue
1935 static void nfsiod_stop(void)
1937 struct workqueue_struct *wq;
1939 wq = nfsiod_workqueue;
1942 nfsiod_workqueue = NULL;
1943 destroy_workqueue(wq);
1947 EXPORT_SYMBOL_GPL(nfs_net_id);
1949 static int nfs_net_init(struct net *net)
1951 nfs_clients_init(net);
1952 return nfs_fs_proc_net_init(net);
1955 static void nfs_net_exit(struct net *net)
1957 nfs_fs_proc_net_exit(net);
1958 nfs_cleanup_cb_ident_idr(net);
1961 static struct pernet_operations nfs_net_ops = {
1962 .init = nfs_net_init,
1963 .exit = nfs_net_exit,
1965 .size = sizeof(struct nfs_net),
1971 static int __init init_nfs_fs(void)
1975 err = register_pernet_subsys(&nfs_net_ops);
1979 err = nfs_fscache_register();
1983 err = nfsiod_start();
1987 err = nfs_fs_proc_init();
1991 err = nfs_init_nfspagecache();
1995 err = nfs_init_inodecache();
1999 err = nfs_init_readpagecache();
2003 err = nfs_init_writepagecache();
2007 err = nfs_init_directcache();
2011 #ifdef CONFIG_PROC_FS
2012 rpc_proc_register(&init_net, &nfs_rpcstat);
2014 if ((err = register_nfs_fs()) != 0)
2019 #ifdef CONFIG_PROC_FS
2020 rpc_proc_unregister(&init_net, "nfs");
2022 nfs_destroy_directcache();
2024 nfs_destroy_writepagecache();
2026 nfs_destroy_readpagecache();
2028 nfs_destroy_inodecache();
2030 nfs_destroy_nfspagecache();
2036 nfs_fscache_unregister();
2038 unregister_pernet_subsys(&nfs_net_ops);
2043 static void __exit exit_nfs_fs(void)
2045 nfs_destroy_directcache();
2046 nfs_destroy_writepagecache();
2047 nfs_destroy_readpagecache();
2048 nfs_destroy_inodecache();
2049 nfs_destroy_nfspagecache();
2050 nfs_fscache_unregister();
2051 unregister_pernet_subsys(&nfs_net_ops);
2052 #ifdef CONFIG_PROC_FS
2053 rpc_proc_unregister(&init_net, "nfs");
2055 unregister_nfs_fs();
2060 /* Not quite true; I just maintain it */
2061 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2062 MODULE_LICENSE("GPL");
2063 module_param(enable_ino64, bool, 0644);
2065 module_init(init_nfs_fs)
2066 module_exit(exit_nfs_fs)