gre: Fix MTU calculation for bound GRE tunnels
[linux-2.6/mini2440.git] / fs / nfs / inode.c
blob64f87194d3907620709c5413f23c95324b23afd9
1 /*
2 * linux/fs/nfs/inode.c
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>
21 #include <linux/mm.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/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38 #include <linux/inet.h>
39 #include <linux/nfs_xdr.h>
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
44 #include "nfs4_fs.h"
45 #include "callback.h"
46 #include "delegation.h"
47 #include "iostat.h"
48 #include "internal.h"
49 #include "fscache.h"
51 #define NFSDBG_FACILITY NFSDBG_VFS
53 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
55 /* Default is to see 64-bit inode numbers */
56 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
58 static void nfs_invalidate_inode(struct inode *);
59 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
61 static struct kmem_cache * nfs_inode_cachep;
63 static inline unsigned long
64 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
66 return nfs_fileid_to_ino_t(fattr->fileid);
69 /**
70 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
71 * @word: long word containing the bit lock
73 int nfs_wait_bit_killable(void *word)
75 if (fatal_signal_pending(current))
76 return -ERESTARTSYS;
77 schedule();
78 return 0;
81 /**
82 * nfs_compat_user_ino64 - returns the user-visible inode number
83 * @fileid: 64-bit fileid
85 * This function returns a 32-bit inode number if the boot parameter
86 * nfs.enable_ino64 is zero.
88 u64 nfs_compat_user_ino64(u64 fileid)
90 int ino;
92 if (enable_ino64)
93 return fileid;
94 ino = fileid;
95 if (sizeof(ino) < sizeof(fileid))
96 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
97 return ino;
100 int nfs_write_inode(struct inode *inode, int sync)
102 int ret;
104 if (sync) {
105 ret = filemap_fdatawait(inode->i_mapping);
106 if (ret == 0)
107 ret = nfs_commit_inode(inode, FLUSH_SYNC);
108 } else
109 ret = nfs_commit_inode(inode, 0);
110 if (ret >= 0)
111 return 0;
112 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
113 return ret;
116 void nfs_clear_inode(struct inode *inode)
119 * The following should never happen...
121 BUG_ON(nfs_have_writebacks(inode));
122 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
123 nfs_zap_acl_cache(inode);
124 nfs_access_zap_cache(inode);
125 nfs_fscache_release_inode_cookie(inode);
129 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
131 int nfs_sync_mapping(struct address_space *mapping)
133 int ret;
135 if (mapping->nrpages == 0)
136 return 0;
137 unmap_mapping_range(mapping, 0, 0, 0);
138 ret = filemap_write_and_wait(mapping);
139 if (ret != 0)
140 goto out;
141 ret = nfs_wb_all(mapping->host);
142 out:
143 return ret;
147 * Invalidate the local caches
149 static void nfs_zap_caches_locked(struct inode *inode)
151 struct nfs_inode *nfsi = NFS_I(inode);
152 int mode = inode->i_mode;
154 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
156 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
157 nfsi->attrtimeo_timestamp = jiffies;
159 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
160 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
161 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
162 else
163 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
166 void nfs_zap_caches(struct inode *inode)
168 spin_lock(&inode->i_lock);
169 nfs_zap_caches_locked(inode);
170 spin_unlock(&inode->i_lock);
173 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
175 if (mapping->nrpages != 0) {
176 spin_lock(&inode->i_lock);
177 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
178 spin_unlock(&inode->i_lock);
182 void nfs_zap_acl_cache(struct inode *inode)
184 void (*clear_acl_cache)(struct inode *);
186 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
187 if (clear_acl_cache != NULL)
188 clear_acl_cache(inode);
189 spin_lock(&inode->i_lock);
190 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
191 spin_unlock(&inode->i_lock);
194 void nfs_invalidate_atime(struct inode *inode)
196 spin_lock(&inode->i_lock);
197 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
198 spin_unlock(&inode->i_lock);
202 * Invalidate, but do not unhash, the inode.
203 * NB: must be called with inode->i_lock held!
205 static void nfs_invalidate_inode(struct inode *inode)
207 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
208 nfs_zap_caches_locked(inode);
211 struct nfs_find_desc {
212 struct nfs_fh *fh;
213 struct nfs_fattr *fattr;
217 * In NFSv3 we can have 64bit inode numbers. In order to support
218 * this, and re-exported directories (also seen in NFSv2)
219 * we are forced to allow 2 different inodes to have the same
220 * i_ino.
222 static int
223 nfs_find_actor(struct inode *inode, void *opaque)
225 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
226 struct nfs_fh *fh = desc->fh;
227 struct nfs_fattr *fattr = desc->fattr;
229 if (NFS_FILEID(inode) != fattr->fileid)
230 return 0;
231 if (nfs_compare_fh(NFS_FH(inode), fh))
232 return 0;
233 if (is_bad_inode(inode) || NFS_STALE(inode))
234 return 0;
235 return 1;
238 static int
239 nfs_init_locked(struct inode *inode, void *opaque)
241 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
242 struct nfs_fattr *fattr = desc->fattr;
244 set_nfs_fileid(inode, fattr->fileid);
245 nfs_copy_fh(NFS_FH(inode), desc->fh);
246 return 0;
249 /* Don't use READDIRPLUS on directories that we believe are too large */
250 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
253 * This is our front-end to iget that looks up inodes by file handle
254 * instead of inode number.
256 struct inode *
257 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
259 struct nfs_find_desc desc = {
260 .fh = fh,
261 .fattr = fattr
263 struct inode *inode = ERR_PTR(-ENOENT);
264 unsigned long hash;
266 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
267 goto out_no_inode;
268 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
269 goto out_no_inode;
271 hash = nfs_fattr_to_ino_t(fattr);
273 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
274 if (inode == NULL) {
275 inode = ERR_PTR(-ENOMEM);
276 goto out_no_inode;
279 if (inode->i_state & I_NEW) {
280 struct nfs_inode *nfsi = NFS_I(inode);
281 unsigned long now = jiffies;
283 /* We set i_ino for the few things that still rely on it,
284 * such as stat(2) */
285 inode->i_ino = hash;
287 /* We can't support update_atime(), since the server will reset it */
288 inode->i_flags |= S_NOATIME|S_NOCMTIME;
289 inode->i_mode = fattr->mode;
290 /* Why so? Because we want revalidate for devices/FIFOs, and
291 * that's precisely what we have in nfs_file_inode_operations.
293 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
294 if (S_ISREG(inode->i_mode)) {
295 inode->i_fop = &nfs_file_operations;
296 inode->i_data.a_ops = &nfs_file_aops;
297 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
298 } else if (S_ISDIR(inode->i_mode)) {
299 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
300 inode->i_fop = &nfs_dir_operations;
301 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
302 && fattr->size <= NFS_LIMIT_READDIRPLUS)
303 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
304 /* Deal with crossing mountpoints */
305 if ((fattr->valid & NFS_ATTR_FATTR_FSID)
306 && !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
307 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
308 inode->i_op = &nfs_referral_inode_operations;
309 else
310 inode->i_op = &nfs_mountpoint_inode_operations;
311 inode->i_fop = NULL;
312 set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
314 } else if (S_ISLNK(inode->i_mode))
315 inode->i_op = &nfs_symlink_inode_operations;
316 else
317 init_special_inode(inode, inode->i_mode, fattr->rdev);
319 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
320 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
321 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
322 nfsi->change_attr = 0;
323 inode->i_size = 0;
324 inode->i_nlink = 0;
325 inode->i_uid = -2;
326 inode->i_gid = -2;
327 inode->i_blocks = 0;
328 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
330 nfsi->read_cache_jiffies = fattr->time_start;
331 nfsi->attr_gencount = fattr->gencount;
332 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
333 inode->i_atime = fattr->atime;
334 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
335 inode->i_mtime = fattr->mtime;
336 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
337 inode->i_ctime = fattr->ctime;
338 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
339 nfsi->change_attr = fattr->change_attr;
340 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
341 inode->i_size = nfs_size_to_loff_t(fattr->size);
342 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
343 inode->i_nlink = fattr->nlink;
344 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
345 inode->i_uid = fattr->uid;
346 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
347 inode->i_gid = fattr->gid;
348 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
349 inode->i_blocks = fattr->du.nfs2.blocks;
350 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
352 * report the blocks in 512byte units
354 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
356 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
357 nfsi->attrtimeo_timestamp = now;
358 nfsi->access_cache = RB_ROOT;
360 nfs_fscache_init_inode_cookie(inode);
362 unlock_new_inode(inode);
363 } else
364 nfs_refresh_inode(inode, fattr);
365 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
366 inode->i_sb->s_id,
367 (long long)NFS_FILEID(inode),
368 atomic_read(&inode->i_count));
370 out:
371 return inode;
373 out_no_inode:
374 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
375 goto out;
378 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
381 nfs_setattr(struct dentry *dentry, struct iattr *attr)
383 struct inode *inode = dentry->d_inode;
384 struct nfs_fattr fattr;
385 int error;
387 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
389 /* skip mode change if it's just for clearing setuid/setgid */
390 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
391 attr->ia_valid &= ~ATTR_MODE;
393 if (attr->ia_valid & ATTR_SIZE) {
394 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
395 attr->ia_valid &= ~ATTR_SIZE;
398 /* Optimization: if the end result is no change, don't RPC */
399 attr->ia_valid &= NFS_VALID_ATTRS;
400 if ((attr->ia_valid & ~ATTR_FILE) == 0)
401 return 0;
403 /* Write all dirty data */
404 if (S_ISREG(inode->i_mode)) {
405 filemap_write_and_wait(inode->i_mapping);
406 nfs_wb_all(inode);
409 * Return any delegations if we're going to change ACLs
411 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
412 nfs_inode_return_delegation(inode);
413 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
414 if (error == 0)
415 nfs_refresh_inode(inode, &fattr);
416 return error;
420 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
421 * @inode: inode of the file used
422 * @offset: file offset to start truncating
424 * This is a copy of the common vmtruncate, but with the locking
425 * corrected to take into account the fact that NFS requires
426 * inode->i_size to be updated under the inode->i_lock.
428 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
430 if (i_size_read(inode) < offset) {
431 unsigned long limit;
433 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
434 if (limit != RLIM_INFINITY && offset > limit)
435 goto out_sig;
436 if (offset > inode->i_sb->s_maxbytes)
437 goto out_big;
438 spin_lock(&inode->i_lock);
439 i_size_write(inode, offset);
440 spin_unlock(&inode->i_lock);
441 } else {
442 struct address_space *mapping = inode->i_mapping;
445 * truncation of in-use swapfiles is disallowed - it would
446 * cause subsequent swapout to scribble on the now-freed
447 * blocks.
449 if (IS_SWAPFILE(inode))
450 return -ETXTBSY;
451 spin_lock(&inode->i_lock);
452 i_size_write(inode, offset);
453 spin_unlock(&inode->i_lock);
456 * unmap_mapping_range is called twice, first simply for
457 * efficiency so that truncate_inode_pages does fewer
458 * single-page unmaps. However after this first call, and
459 * before truncate_inode_pages finishes, it is possible for
460 * private pages to be COWed, which remain after
461 * truncate_inode_pages finishes, hence the second
462 * unmap_mapping_range call must be made for correctness.
464 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
465 truncate_inode_pages(mapping, offset);
466 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
468 return 0;
469 out_sig:
470 send_sig(SIGXFSZ, current, 0);
471 out_big:
472 return -EFBIG;
476 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
477 * @inode: pointer to struct inode
478 * @attr: pointer to struct iattr
480 * Note: we do this in the *proc.c in order to ensure that
481 * it works for things like exclusive creates too.
483 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
485 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
486 spin_lock(&inode->i_lock);
487 if ((attr->ia_valid & ATTR_MODE) != 0) {
488 int mode = attr->ia_mode & S_IALLUGO;
489 mode |= inode->i_mode & ~S_IALLUGO;
490 inode->i_mode = mode;
492 if ((attr->ia_valid & ATTR_UID) != 0)
493 inode->i_uid = attr->ia_uid;
494 if ((attr->ia_valid & ATTR_GID) != 0)
495 inode->i_gid = attr->ia_gid;
496 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
497 spin_unlock(&inode->i_lock);
499 if ((attr->ia_valid & ATTR_SIZE) != 0) {
500 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
501 nfs_vmtruncate(inode, attr->ia_size);
505 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
507 struct inode *inode = dentry->d_inode;
508 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
509 int err;
512 * Flush out writes to the server in order to update c/mtime.
514 * Hold the i_mutex to suspend application writes temporarily;
515 * this prevents long-running writing applications from blocking
516 * nfs_wb_nocommit.
518 if (S_ISREG(inode->i_mode)) {
519 mutex_lock(&inode->i_mutex);
520 nfs_wb_nocommit(inode);
521 mutex_unlock(&inode->i_mutex);
525 * We may force a getattr if the user cares about atime.
527 * Note that we only have to check the vfsmount flags here:
528 * - NFS always sets S_NOATIME by so checking it would give a
529 * bogus result
530 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
531 * no point in checking those.
533 if ((mnt->mnt_flags & MNT_NOATIME) ||
534 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
535 need_atime = 0;
537 if (need_atime)
538 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
539 else
540 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
541 if (!err) {
542 generic_fillattr(inode, stat);
543 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
545 return err;
549 * nfs_close_context - Common close_context() routine NFSv2/v3
550 * @ctx: pointer to context
551 * @is_sync: is this a synchronous close
553 * always ensure that the attributes are up to date if we're mounted
554 * with close-to-open semantics
556 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
558 struct inode *inode;
559 struct nfs_server *server;
561 if (!(ctx->mode & FMODE_WRITE))
562 return;
563 if (!is_sync)
564 return;
565 inode = ctx->path.dentry->d_inode;
566 if (!list_empty(&NFS_I(inode)->open_files))
567 return;
568 server = NFS_SERVER(inode);
569 if (server->flags & NFS_MOUNT_NOCTO)
570 return;
571 nfs_revalidate_inode(server, inode);
574 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
576 struct nfs_open_context *ctx;
578 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
579 if (ctx != NULL) {
580 ctx->path.dentry = dget(dentry);
581 ctx->path.mnt = mntget(mnt);
582 ctx->cred = get_rpccred(cred);
583 ctx->state = NULL;
584 ctx->lockowner = current->files;
585 ctx->flags = 0;
586 ctx->error = 0;
587 ctx->dir_cookie = 0;
588 atomic_set(&ctx->count, 1);
590 return ctx;
593 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
595 if (ctx != NULL)
596 atomic_inc(&ctx->count);
597 return ctx;
600 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
602 struct inode *inode = ctx->path.dentry->d_inode;
604 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
605 return;
606 list_del(&ctx->list);
607 spin_unlock(&inode->i_lock);
608 NFS_PROTO(inode)->close_context(ctx, is_sync);
609 if (ctx->cred != NULL)
610 put_rpccred(ctx->cred);
611 path_put(&ctx->path);
612 kfree(ctx);
615 void put_nfs_open_context(struct nfs_open_context *ctx)
617 __put_nfs_open_context(ctx, 0);
620 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
622 __put_nfs_open_context(ctx, 1);
626 * Ensure that mmap has a recent RPC credential for use when writing out
627 * shared pages
629 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
631 struct inode *inode = filp->f_path.dentry->d_inode;
632 struct nfs_inode *nfsi = NFS_I(inode);
634 filp->private_data = get_nfs_open_context(ctx);
635 spin_lock(&inode->i_lock);
636 list_add(&ctx->list, &nfsi->open_files);
637 spin_unlock(&inode->i_lock);
641 * Given an inode, search for an open context with the desired characteristics
643 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
645 struct nfs_inode *nfsi = NFS_I(inode);
646 struct nfs_open_context *pos, *ctx = NULL;
648 spin_lock(&inode->i_lock);
649 list_for_each_entry(pos, &nfsi->open_files, list) {
650 if (cred != NULL && pos->cred != cred)
651 continue;
652 if ((pos->mode & mode) == mode) {
653 ctx = get_nfs_open_context(pos);
654 break;
657 spin_unlock(&inode->i_lock);
658 return ctx;
661 static void nfs_file_clear_open_context(struct file *filp)
663 struct inode *inode = filp->f_path.dentry->d_inode;
664 struct nfs_open_context *ctx = nfs_file_open_context(filp);
666 if (ctx) {
667 filp->private_data = NULL;
668 spin_lock(&inode->i_lock);
669 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
670 spin_unlock(&inode->i_lock);
671 put_nfs_open_context_sync(ctx);
676 * These allocate and release file read/write context information.
678 int nfs_open(struct inode *inode, struct file *filp)
680 struct nfs_open_context *ctx;
681 struct rpc_cred *cred;
683 cred = rpc_lookup_cred();
684 if (IS_ERR(cred))
685 return PTR_ERR(cred);
686 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
687 put_rpccred(cred);
688 if (ctx == NULL)
689 return -ENOMEM;
690 ctx->mode = filp->f_mode;
691 nfs_file_set_open_context(filp, ctx);
692 put_nfs_open_context(ctx);
693 nfs_fscache_set_inode_cookie(inode, filp);
694 return 0;
697 int nfs_release(struct inode *inode, struct file *filp)
699 nfs_file_clear_open_context(filp);
700 return 0;
704 * This function is called whenever some part of NFS notices that
705 * the cached attributes have to be refreshed.
708 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
710 int status = -ESTALE;
711 struct nfs_fattr fattr;
712 struct nfs_inode *nfsi = NFS_I(inode);
714 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
715 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
717 if (is_bad_inode(inode))
718 goto out;
719 if (NFS_STALE(inode))
720 goto out;
722 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
723 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
724 if (status != 0) {
725 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
726 inode->i_sb->s_id,
727 (long long)NFS_FILEID(inode), status);
728 if (status == -ESTALE) {
729 nfs_zap_caches(inode);
730 if (!S_ISDIR(inode->i_mode))
731 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
733 goto out;
736 status = nfs_refresh_inode(inode, &fattr);
737 if (status) {
738 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
739 inode->i_sb->s_id,
740 (long long)NFS_FILEID(inode), status);
741 goto out;
744 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
745 nfs_zap_acl_cache(inode);
747 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
748 inode->i_sb->s_id,
749 (long long)NFS_FILEID(inode));
751 out:
752 return status;
755 int nfs_attribute_timeout(struct inode *inode)
757 struct nfs_inode *nfsi = NFS_I(inode);
759 if (nfs_have_delegation(inode, FMODE_READ))
760 return 0;
761 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
765 * nfs_revalidate_inode - Revalidate the inode attributes
766 * @server - pointer to nfs_server struct
767 * @inode - pointer to inode struct
769 * Updates inode attribute information by retrieving the data from the server.
771 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
773 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
774 && !nfs_attribute_timeout(inode))
775 return NFS_STALE(inode) ? -ESTALE : 0;
776 return __nfs_revalidate_inode(server, inode);
779 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
781 struct nfs_inode *nfsi = NFS_I(inode);
783 if (mapping->nrpages != 0) {
784 int ret = invalidate_inode_pages2(mapping);
785 if (ret < 0)
786 return ret;
788 spin_lock(&inode->i_lock);
789 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
790 if (S_ISDIR(inode->i_mode))
791 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
792 spin_unlock(&inode->i_lock);
793 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
794 nfs_fscache_reset_inode_cookie(inode);
795 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
796 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
797 return 0;
800 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
802 int ret = 0;
804 mutex_lock(&inode->i_mutex);
805 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
806 ret = nfs_sync_mapping(mapping);
807 if (ret == 0)
808 ret = nfs_invalidate_mapping_nolock(inode, mapping);
810 mutex_unlock(&inode->i_mutex);
811 return ret;
815 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
816 * @inode - pointer to host inode
817 * @mapping - pointer to mapping
819 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
821 struct nfs_inode *nfsi = NFS_I(inode);
822 int ret = 0;
824 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
825 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
826 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
827 if (ret < 0)
828 goto out;
830 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
831 ret = nfs_invalidate_mapping_nolock(inode, mapping);
832 out:
833 return ret;
837 * nfs_revalidate_mapping - Revalidate the pagecache
838 * @inode - pointer to host inode
839 * @mapping - pointer to mapping
841 * This version of the function will take the inode->i_mutex and attempt to
842 * flush out all dirty data if it needs to invalidate the page cache.
844 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
846 struct nfs_inode *nfsi = NFS_I(inode);
847 int ret = 0;
849 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
850 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
851 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
852 if (ret < 0)
853 goto out;
855 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
856 ret = nfs_invalidate_mapping(inode, mapping);
857 out:
858 return ret;
861 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
863 struct nfs_inode *nfsi = NFS_I(inode);
865 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
866 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
867 && nfsi->change_attr == fattr->pre_change_attr) {
868 nfsi->change_attr = fattr->change_attr;
869 if (S_ISDIR(inode->i_mode))
870 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
872 /* If we have atomic WCC data, we may update some attributes */
873 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
874 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
875 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
876 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
878 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
879 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
880 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
881 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
882 if (S_ISDIR(inode->i_mode))
883 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
885 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
886 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
887 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
888 && nfsi->npages == 0)
889 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
893 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
894 * @inode - pointer to inode
895 * @fattr - updated attributes
897 * Verifies the attribute cache. If we have just changed the attributes,
898 * so that fattr carries weak cache consistency data, then it may
899 * also update the ctime/mtime/change_attribute.
901 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
903 struct nfs_inode *nfsi = NFS_I(inode);
904 loff_t cur_size, new_isize;
905 unsigned long invalid = 0;
908 /* Has the inode gone and changed behind our back? */
909 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
910 return -EIO;
911 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
912 return -EIO;
914 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
915 nfsi->change_attr != fattr->change_attr)
916 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
918 /* Verify a few of the more important attributes */
919 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
920 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
922 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
923 cur_size = i_size_read(inode);
924 new_isize = nfs_size_to_loff_t(fattr->size);
925 if (cur_size != new_isize && nfsi->npages == 0)
926 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
929 /* Have any file permissions changed? */
930 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
931 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
932 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
933 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
934 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
935 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
937 /* Has the link count changed? */
938 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
939 invalid |= NFS_INO_INVALID_ATTR;
941 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
942 invalid |= NFS_INO_INVALID_ATIME;
944 if (invalid != 0)
945 nfsi->cache_validity |= invalid;
947 nfsi->read_cache_jiffies = fattr->time_start;
948 return 0;
951 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
953 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
954 return 0;
955 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
958 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
960 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
961 return 0;
962 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
965 static atomic_long_t nfs_attr_generation_counter;
967 static unsigned long nfs_read_attr_generation_counter(void)
969 return atomic_long_read(&nfs_attr_generation_counter);
972 unsigned long nfs_inc_attr_generation_counter(void)
974 return atomic_long_inc_return(&nfs_attr_generation_counter);
977 void nfs_fattr_init(struct nfs_fattr *fattr)
979 fattr->valid = 0;
980 fattr->time_start = jiffies;
981 fattr->gencount = nfs_inc_attr_generation_counter();
985 * nfs_inode_attrs_need_update - check if the inode attributes need updating
986 * @inode - pointer to inode
987 * @fattr - attributes
989 * Attempt to divine whether or not an RPC call reply carrying stale
990 * attributes got scheduled after another call carrying updated ones.
992 * To do so, the function first assumes that a more recent ctime means
993 * that the attributes in fattr are newer, however it also attempt to
994 * catch the case where ctime either didn't change, or went backwards
995 * (if someone reset the clock on the server) by looking at whether
996 * or not this RPC call was started after the inode was last updated.
997 * Note also the check for wraparound of 'attr_gencount'
999 * The function returns 'true' if it thinks the attributes in 'fattr' are
1000 * more recent than the ones cached in the inode.
1003 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1005 const struct nfs_inode *nfsi = NFS_I(inode);
1007 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1008 nfs_ctime_need_update(inode, fattr) ||
1009 nfs_size_need_update(inode, fattr) ||
1010 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1013 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1015 if (nfs_inode_attrs_need_update(inode, fattr))
1016 return nfs_update_inode(inode, fattr);
1017 return nfs_check_inode_attributes(inode, fattr);
1021 * nfs_refresh_inode - try to update the inode attribute cache
1022 * @inode - pointer to inode
1023 * @fattr - updated attributes
1025 * Check that an RPC call that returned attributes has not overlapped with
1026 * other recent updates of the inode metadata, then decide whether it is
1027 * safe to do a full update of the inode attributes, or whether just to
1028 * call nfs_check_inode_attributes.
1030 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1032 int status;
1034 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1035 return 0;
1036 spin_lock(&inode->i_lock);
1037 status = nfs_refresh_inode_locked(inode, fattr);
1038 spin_unlock(&inode->i_lock);
1040 return status;
1043 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1045 struct nfs_inode *nfsi = NFS_I(inode);
1047 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1048 if (S_ISDIR(inode->i_mode))
1049 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1050 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1051 return 0;
1052 return nfs_refresh_inode_locked(inode, fattr);
1056 * nfs_post_op_update_inode - try to update the inode attribute cache
1057 * @inode - pointer to inode
1058 * @fattr - updated attributes
1060 * After an operation that has changed the inode metadata, mark the
1061 * attribute cache as being invalid, then try to update it.
1063 * NB: if the server didn't return any post op attributes, this
1064 * function will force the retrieval of attributes before the next
1065 * NFS request. Thus it should be used only for operations that
1066 * are expected to change one or more attributes, to avoid
1067 * unnecessary NFS requests and trips through nfs_update_inode().
1069 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1071 int status;
1073 spin_lock(&inode->i_lock);
1074 status = nfs_post_op_update_inode_locked(inode, fattr);
1075 spin_unlock(&inode->i_lock);
1076 return status;
1080 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1081 * @inode - pointer to inode
1082 * @fattr - updated attributes
1084 * After an operation that has changed the inode metadata, mark the
1085 * attribute cache as being invalid, then try to update it. Fake up
1086 * weak cache consistency data, if none exist.
1088 * This function is mainly designed to be used by the ->write_done() functions.
1090 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1092 int status;
1094 spin_lock(&inode->i_lock);
1095 /* Don't do a WCC update if these attributes are already stale */
1096 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1097 !nfs_inode_attrs_need_update(inode, fattr)) {
1098 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1099 | NFS_ATTR_FATTR_PRESIZE
1100 | NFS_ATTR_FATTR_PREMTIME
1101 | NFS_ATTR_FATTR_PRECTIME);
1102 goto out_noforce;
1104 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1105 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1106 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1107 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1109 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1110 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1111 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1112 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1114 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1115 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1116 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1117 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1119 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1120 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1121 fattr->pre_size = i_size_read(inode);
1122 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1124 out_noforce:
1125 status = nfs_post_op_update_inode_locked(inode, fattr);
1126 spin_unlock(&inode->i_lock);
1127 return status;
1131 * Many nfs protocol calls return the new file attributes after
1132 * an operation. Here we update the inode to reflect the state
1133 * of the server's inode.
1135 * This is a bit tricky because we have to make sure all dirty pages
1136 * have been sent off to the server before calling invalidate_inode_pages.
1137 * To make sure no other process adds more write requests while we try
1138 * our best to flush them, we make them sleep during the attribute refresh.
1140 * A very similar scenario holds for the dir cache.
1142 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1144 struct nfs_server *server;
1145 struct nfs_inode *nfsi = NFS_I(inode);
1146 loff_t cur_isize, new_isize;
1147 unsigned long invalid = 0;
1148 unsigned long now = jiffies;
1150 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1151 __func__, inode->i_sb->s_id, inode->i_ino,
1152 atomic_read(&inode->i_count), fattr->valid);
1154 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1155 goto out_fileid;
1158 * Make sure the inode's type hasn't changed.
1160 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1161 goto out_changed;
1163 server = NFS_SERVER(inode);
1164 /* Update the fsid? */
1165 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1166 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1167 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1168 server->fsid = fattr->fsid;
1171 * Update the read time so we don't revalidate too often.
1173 nfsi->read_cache_jiffies = fattr->time_start;
1175 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) || (fattr->valid & (NFS_ATTR_FATTR_MTIME|NFS_ATTR_FATTR_CTIME)))
1176 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1177 | NFS_INO_INVALID_ATIME
1178 | NFS_INO_REVAL_PAGECACHE);
1180 /* Do atomic weak cache consistency updates */
1181 nfs_wcc_update_inode(inode, fattr);
1183 /* More cache consistency checks */
1184 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1185 if (nfsi->change_attr != fattr->change_attr) {
1186 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1187 inode->i_sb->s_id, inode->i_ino);
1188 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1189 if (S_ISDIR(inode->i_mode))
1190 nfs_force_lookup_revalidate(inode);
1191 nfsi->change_attr = fattr->change_attr;
1195 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1196 /* NFSv2/v3: Check if the mtime agrees */
1197 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1198 dprintk("NFS: mtime change on server for file %s/%ld\n",
1199 inode->i_sb->s_id, inode->i_ino);
1200 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1201 if (S_ISDIR(inode->i_mode))
1202 nfs_force_lookup_revalidate(inode);
1203 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1206 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1207 /* If ctime has changed we should definitely clear access+acl caches */
1208 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1209 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1210 /* and probably clear data for a directory too as utimes can cause
1211 * havoc with our cache.
1213 if (S_ISDIR(inode->i_mode)) {
1214 invalid |= NFS_INO_INVALID_DATA;
1215 nfs_force_lookup_revalidate(inode);
1217 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1221 /* Check if our cached file size is stale */
1222 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1223 new_isize = nfs_size_to_loff_t(fattr->size);
1224 cur_isize = i_size_read(inode);
1225 if (new_isize != cur_isize) {
1226 /* Do we perhaps have any outstanding writes, or has
1227 * the file grown beyond our last write? */
1228 if (nfsi->npages == 0 || new_isize > cur_isize) {
1229 i_size_write(inode, new_isize);
1230 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1232 dprintk("NFS: isize change on server for file %s/%ld\n",
1233 inode->i_sb->s_id, inode->i_ino);
1238 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1239 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1241 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1242 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1243 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1244 inode->i_mode = fattr->mode;
1247 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1248 if (inode->i_uid != fattr->uid) {
1249 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1250 inode->i_uid = fattr->uid;
1253 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1254 if (inode->i_gid != fattr->gid) {
1255 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1256 inode->i_gid = fattr->gid;
1260 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1261 if (inode->i_nlink != fattr->nlink) {
1262 invalid |= NFS_INO_INVALID_ATTR;
1263 if (S_ISDIR(inode->i_mode))
1264 invalid |= NFS_INO_INVALID_DATA;
1265 inode->i_nlink = fattr->nlink;
1269 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1271 * report the blocks in 512byte units
1273 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1275 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1276 inode->i_blocks = fattr->du.nfs2.blocks;
1278 /* Update attrtimeo value if we're out of the unstable period */
1279 if (invalid & NFS_INO_INVALID_ATTR) {
1280 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1281 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1282 nfsi->attrtimeo_timestamp = now;
1283 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1284 } else {
1285 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1286 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1287 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1288 nfsi->attrtimeo_timestamp = now;
1291 invalid &= ~NFS_INO_INVALID_ATTR;
1292 /* Don't invalidate the data if we were to blame */
1293 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1294 || S_ISLNK(inode->i_mode)))
1295 invalid &= ~NFS_INO_INVALID_DATA;
1296 if (!nfs_have_delegation(inode, FMODE_READ) ||
1297 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1298 nfsi->cache_validity |= invalid;
1299 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1301 return 0;
1302 out_changed:
1304 * Big trouble! The inode has become a different object.
1306 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1307 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1308 out_err:
1310 * No need to worry about unhashing the dentry, as the
1311 * lookup validation will know that the inode is bad.
1312 * (But we fall through to invalidate the caches.)
1314 nfs_invalidate_inode(inode);
1315 return -ESTALE;
1317 out_fileid:
1318 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1319 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1320 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1321 (long long)nfsi->fileid, (long long)fattr->fileid);
1322 goto out_err;
1326 #ifdef CONFIG_NFS_V4
1329 * Clean out any remaining NFSv4 state that might be left over due
1330 * to open() calls that passed nfs_atomic_lookup, but failed to call
1331 * nfs_open().
1333 void nfs4_clear_inode(struct inode *inode)
1335 /* If we are holding a delegation, return it! */
1336 nfs_inode_return_delegation_noreclaim(inode);
1337 /* First call standard NFS clear_inode() code */
1338 nfs_clear_inode(inode);
1340 #endif
1342 struct inode *nfs_alloc_inode(struct super_block *sb)
1344 struct nfs_inode *nfsi;
1345 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1346 if (!nfsi)
1347 return NULL;
1348 nfsi->flags = 0UL;
1349 nfsi->cache_validity = 0UL;
1350 #ifdef CONFIG_NFS_V3_ACL
1351 nfsi->acl_access = ERR_PTR(-EAGAIN);
1352 nfsi->acl_default = ERR_PTR(-EAGAIN);
1353 #endif
1354 #ifdef CONFIG_NFS_V4
1355 nfsi->nfs4_acl = NULL;
1356 #endif /* CONFIG_NFS_V4 */
1357 return &nfsi->vfs_inode;
1360 void nfs_destroy_inode(struct inode *inode)
1362 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1365 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1367 #ifdef CONFIG_NFS_V4
1368 INIT_LIST_HEAD(&nfsi->open_states);
1369 nfsi->delegation = NULL;
1370 nfsi->delegation_state = 0;
1371 init_rwsem(&nfsi->rwsem);
1372 #endif
1375 static void init_once(void *foo)
1377 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1379 inode_init_once(&nfsi->vfs_inode);
1380 INIT_LIST_HEAD(&nfsi->open_files);
1381 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1382 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1383 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1384 nfsi->npages = 0;
1385 atomic_set(&nfsi->silly_count, 1);
1386 INIT_HLIST_HEAD(&nfsi->silly_list);
1387 init_waitqueue_head(&nfsi->waitqueue);
1388 nfs4_init_once(nfsi);
1391 static int __init nfs_init_inodecache(void)
1393 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1394 sizeof(struct nfs_inode),
1395 0, (SLAB_RECLAIM_ACCOUNT|
1396 SLAB_MEM_SPREAD),
1397 init_once);
1398 if (nfs_inode_cachep == NULL)
1399 return -ENOMEM;
1401 return 0;
1404 static void nfs_destroy_inodecache(void)
1406 kmem_cache_destroy(nfs_inode_cachep);
1409 struct workqueue_struct *nfsiod_workqueue;
1412 * start up the nfsiod workqueue
1414 static int nfsiod_start(void)
1416 struct workqueue_struct *wq;
1417 dprintk("RPC: creating workqueue nfsiod\n");
1418 wq = create_singlethread_workqueue("nfsiod");
1419 if (wq == NULL)
1420 return -ENOMEM;
1421 nfsiod_workqueue = wq;
1422 return 0;
1426 * Destroy the nfsiod workqueue
1428 static void nfsiod_stop(void)
1430 struct workqueue_struct *wq;
1432 wq = nfsiod_workqueue;
1433 if (wq == NULL)
1434 return;
1435 nfsiod_workqueue = NULL;
1436 destroy_workqueue(wq);
1440 * Initialize NFS
1442 static int __init init_nfs_fs(void)
1444 int err;
1446 err = nfs_fscache_register();
1447 if (err < 0)
1448 goto out7;
1450 err = nfsiod_start();
1451 if (err)
1452 goto out6;
1454 err = nfs_fs_proc_init();
1455 if (err)
1456 goto out5;
1458 err = nfs_init_nfspagecache();
1459 if (err)
1460 goto out4;
1462 err = nfs_init_inodecache();
1463 if (err)
1464 goto out3;
1466 err = nfs_init_readpagecache();
1467 if (err)
1468 goto out2;
1470 err = nfs_init_writepagecache();
1471 if (err)
1472 goto out1;
1474 err = nfs_init_directcache();
1475 if (err)
1476 goto out0;
1478 #ifdef CONFIG_PROC_FS
1479 rpc_proc_register(&nfs_rpcstat);
1480 #endif
1481 if ((err = register_nfs_fs()) != 0)
1482 goto out;
1483 return 0;
1484 out:
1485 #ifdef CONFIG_PROC_FS
1486 rpc_proc_unregister("nfs");
1487 #endif
1488 nfs_destroy_directcache();
1489 out0:
1490 nfs_destroy_writepagecache();
1491 out1:
1492 nfs_destroy_readpagecache();
1493 out2:
1494 nfs_destroy_inodecache();
1495 out3:
1496 nfs_destroy_nfspagecache();
1497 out4:
1498 nfs_fs_proc_exit();
1499 out5:
1500 nfsiod_stop();
1501 out6:
1502 nfs_fscache_unregister();
1503 out7:
1504 return err;
1507 static void __exit exit_nfs_fs(void)
1509 nfs_destroy_directcache();
1510 nfs_destroy_writepagecache();
1511 nfs_destroy_readpagecache();
1512 nfs_destroy_inodecache();
1513 nfs_destroy_nfspagecache();
1514 nfs_fscache_unregister();
1515 #ifdef CONFIG_PROC_FS
1516 rpc_proc_unregister("nfs");
1517 #endif
1518 unregister_nfs_fs();
1519 nfs_fs_proc_exit();
1520 nfsiod_stop();
1523 /* Not quite true; I just maintain it */
1524 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1525 MODULE_LICENSE("GPL");
1526 module_param(enable_ino64, bool, 0644);
1528 module_init(init_nfs_fs)
1529 module_exit(exit_nfs_fs)