xref: /openbmc/linux/fs/nfs/inode.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /*
2  *  linux/fs/nfs/inode.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs inode and superblock handling functions
7  *
8  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
10  *
11  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12  *  J.S.Peatfield@damtp.cam.ac.uk
13  *
14  */
15 
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/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/nfs_idmap.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
43 
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"
50 #include "dns_resolve.h"
51 #include "pnfs.h"
52 
53 #define NFSDBG_FACILITY		NFSDBG_VFS
54 
55 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
56 
57 /* Default is to see 64-bit inode numbers */
58 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
59 
60 static void nfs_invalidate_inode(struct inode *);
61 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
62 
63 static struct kmem_cache * nfs_inode_cachep;
64 
65 static inline unsigned long
66 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
67 {
68 	return nfs_fileid_to_ino_t(fattr->fileid);
69 }
70 
71 /**
72  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
73  * @word: long word containing the bit lock
74  */
75 int nfs_wait_bit_killable(void *word)
76 {
77 	if (fatal_signal_pending(current))
78 		return -ERESTARTSYS;
79 	schedule();
80 	return 0;
81 }
82 
83 /**
84  * nfs_compat_user_ino64 - returns the user-visible inode number
85  * @fileid: 64-bit fileid
86  *
87  * This function returns a 32-bit inode number if the boot parameter
88  * nfs.enable_ino64 is zero.
89  */
90 u64 nfs_compat_user_ino64(u64 fileid)
91 {
92 	int ino;
93 
94 	if (enable_ino64)
95 		return fileid;
96 	ino = fileid;
97 	if (sizeof(ino) < sizeof(fileid))
98 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
99 	return ino;
100 }
101 
102 static void nfs_clear_inode(struct inode *inode)
103 {
104 	/*
105 	 * The following should never happen...
106 	 */
107 	BUG_ON(nfs_have_writebacks(inode));
108 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
109 	nfs_zap_acl_cache(inode);
110 	nfs_access_zap_cache(inode);
111 	nfs_fscache_release_inode_cookie(inode);
112 }
113 
114 void nfs_evict_inode(struct inode *inode)
115 {
116 	truncate_inode_pages(&inode->i_data, 0);
117 	end_writeback(inode);
118 	nfs_clear_inode(inode);
119 }
120 
121 /**
122  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
123  */
124 int nfs_sync_mapping(struct address_space *mapping)
125 {
126 	int ret = 0;
127 
128 	if (mapping->nrpages != 0) {
129 		unmap_mapping_range(mapping, 0, 0, 0);
130 		ret = nfs_wb_all(mapping->host);
131 	}
132 	return ret;
133 }
134 
135 /*
136  * Invalidate the local caches
137  */
138 static void nfs_zap_caches_locked(struct inode *inode)
139 {
140 	struct nfs_inode *nfsi = NFS_I(inode);
141 	int mode = inode->i_mode;
142 
143 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
144 
145 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
146 	nfsi->attrtimeo_timestamp = jiffies;
147 
148 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
149 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
150 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
151 	else
152 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
153 }
154 
155 void nfs_zap_caches(struct inode *inode)
156 {
157 	spin_lock(&inode->i_lock);
158 	nfs_zap_caches_locked(inode);
159 	spin_unlock(&inode->i_lock);
160 }
161 
162 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
163 {
164 	if (mapping->nrpages != 0) {
165 		spin_lock(&inode->i_lock);
166 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
167 		spin_unlock(&inode->i_lock);
168 	}
169 }
170 
171 void nfs_zap_acl_cache(struct inode *inode)
172 {
173 	void (*clear_acl_cache)(struct inode *);
174 
175 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
176 	if (clear_acl_cache != NULL)
177 		clear_acl_cache(inode);
178 	spin_lock(&inode->i_lock);
179 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
180 	spin_unlock(&inode->i_lock);
181 }
182 
183 void nfs_invalidate_atime(struct inode *inode)
184 {
185 	spin_lock(&inode->i_lock);
186 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
187 	spin_unlock(&inode->i_lock);
188 }
189 
190 /*
191  * Invalidate, but do not unhash, the inode.
192  * NB: must be called with inode->i_lock held!
193  */
194 static void nfs_invalidate_inode(struct inode *inode)
195 {
196 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
197 	nfs_zap_caches_locked(inode);
198 }
199 
200 struct nfs_find_desc {
201 	struct nfs_fh		*fh;
202 	struct nfs_fattr	*fattr;
203 };
204 
205 /*
206  * In NFSv3 we can have 64bit inode numbers. In order to support
207  * this, and re-exported directories (also seen in NFSv2)
208  * we are forced to allow 2 different inodes to have the same
209  * i_ino.
210  */
211 static int
212 nfs_find_actor(struct inode *inode, void *opaque)
213 {
214 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
215 	struct nfs_fh		*fh = desc->fh;
216 	struct nfs_fattr	*fattr = desc->fattr;
217 
218 	if (NFS_FILEID(inode) != fattr->fileid)
219 		return 0;
220 	if (nfs_compare_fh(NFS_FH(inode), fh))
221 		return 0;
222 	if (is_bad_inode(inode) || NFS_STALE(inode))
223 		return 0;
224 	return 1;
225 }
226 
227 static int
228 nfs_init_locked(struct inode *inode, void *opaque)
229 {
230 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
231 	struct nfs_fattr	*fattr = desc->fattr;
232 
233 	set_nfs_fileid(inode, fattr->fileid);
234 	nfs_copy_fh(NFS_FH(inode), desc->fh);
235 	return 0;
236 }
237 
238 /*
239  * This is our front-end to iget that looks up inodes by file handle
240  * instead of inode number.
241  */
242 struct inode *
243 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
244 {
245 	struct nfs_find_desc desc = {
246 		.fh	= fh,
247 		.fattr	= fattr
248 	};
249 	struct inode *inode = ERR_PTR(-ENOENT);
250 	unsigned long hash;
251 
252 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
253 		goto out_no_inode;
254 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
255 		goto out_no_inode;
256 
257 	hash = nfs_fattr_to_ino_t(fattr);
258 
259 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
260 	if (inode == NULL) {
261 		inode = ERR_PTR(-ENOMEM);
262 		goto out_no_inode;
263 	}
264 
265 	if (inode->i_state & I_NEW) {
266 		struct nfs_inode *nfsi = NFS_I(inode);
267 		unsigned long now = jiffies;
268 
269 		/* We set i_ino for the few things that still rely on it,
270 		 * such as stat(2) */
271 		inode->i_ino = hash;
272 
273 		/* We can't support update_atime(), since the server will reset it */
274 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
275 		inode->i_mode = fattr->mode;
276 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
277 				&& nfs_server_capable(inode, NFS_CAP_MODE))
278 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
279 				| NFS_INO_INVALID_ACCESS
280 				| NFS_INO_INVALID_ACL;
281 		/* Why so? Because we want revalidate for devices/FIFOs, and
282 		 * that's precisely what we have in nfs_file_inode_operations.
283 		 */
284 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
285 		if (S_ISREG(inode->i_mode)) {
286 			inode->i_fop = &nfs_file_operations;
287 			inode->i_data.a_ops = &nfs_file_aops;
288 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
289 		} else if (S_ISDIR(inode->i_mode)) {
290 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
291 			inode->i_fop = &nfs_dir_operations;
292 			inode->i_data.a_ops = &nfs_dir_aops;
293 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS))
294 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
295 			/* Deal with crossing mountpoints */
296 			if ((fattr->valid & NFS_ATTR_FATTR_FSID)
297 					&& !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
298 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
299 					inode->i_op = &nfs_referral_inode_operations;
300 				else
301 					inode->i_op = &nfs_mountpoint_inode_operations;
302 				inode->i_fop = NULL;
303 				inode->i_flags |= S_AUTOMOUNT;
304 			}
305 		} else if (S_ISLNK(inode->i_mode))
306 			inode->i_op = &nfs_symlink_inode_operations;
307 		else
308 			init_special_inode(inode, inode->i_mode, fattr->rdev);
309 
310 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
311 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
312 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
313 		nfsi->change_attr = 0;
314 		inode->i_size = 0;
315 		inode->i_nlink = 0;
316 		inode->i_uid = -2;
317 		inode->i_gid = -2;
318 		inode->i_blocks = 0;
319 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
320 
321 		nfsi->read_cache_jiffies = fattr->time_start;
322 		nfsi->attr_gencount = fattr->gencount;
323 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
324 			inode->i_atime = fattr->atime;
325 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
326 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
327 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
328 			inode->i_mtime = fattr->mtime;
329 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
330 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
331 				| NFS_INO_INVALID_DATA;
332 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
333 			inode->i_ctime = fattr->ctime;
334 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
335 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
336 				| NFS_INO_INVALID_ACCESS
337 				| NFS_INO_INVALID_ACL;
338 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
339 			nfsi->change_attr = fattr->change_attr;
340 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
341 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
342 				| NFS_INO_INVALID_DATA;
343 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
344 			inode->i_size = nfs_size_to_loff_t(fattr->size);
345 		else
346 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
347 				| NFS_INO_INVALID_DATA
348 				| NFS_INO_REVAL_PAGECACHE;
349 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
350 			inode->i_nlink = fattr->nlink;
351 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
352 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
353 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
354 			inode->i_uid = fattr->uid;
355 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
356 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
357 				| NFS_INO_INVALID_ACCESS
358 				| NFS_INO_INVALID_ACL;
359 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
360 			inode->i_gid = fattr->gid;
361 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
362 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
363 				| NFS_INO_INVALID_ACCESS
364 				| NFS_INO_INVALID_ACL;
365 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
366 			inode->i_blocks = fattr->du.nfs2.blocks;
367 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
368 			/*
369 			 * report the blocks in 512byte units
370 			 */
371 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
372 		}
373 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
374 		nfsi->attrtimeo_timestamp = now;
375 		nfsi->access_cache = RB_ROOT;
376 
377 		nfs_fscache_init_inode_cookie(inode);
378 
379 		unlock_new_inode(inode);
380 	} else
381 		nfs_refresh_inode(inode, fattr);
382 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
383 		inode->i_sb->s_id,
384 		(long long)NFS_FILEID(inode),
385 		atomic_read(&inode->i_count));
386 
387 out:
388 	return inode;
389 
390 out_no_inode:
391 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
392 	goto out;
393 }
394 
395 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
396 
397 int
398 nfs_setattr(struct dentry *dentry, struct iattr *attr)
399 {
400 	struct inode *inode = dentry->d_inode;
401 	struct nfs_fattr *fattr;
402 	int error = -ENOMEM;
403 
404 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
405 
406 	/* skip mode change if it's just for clearing setuid/setgid */
407 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
408 		attr->ia_valid &= ~ATTR_MODE;
409 
410 	if (attr->ia_valid & ATTR_SIZE) {
411 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
412 			attr->ia_valid &= ~ATTR_SIZE;
413 	}
414 
415 	/* Optimization: if the end result is no change, don't RPC */
416 	attr->ia_valid &= NFS_VALID_ATTRS;
417 	if ((attr->ia_valid & ~ATTR_FILE) == 0)
418 		return 0;
419 
420 	/* Write all dirty data */
421 	if (S_ISREG(inode->i_mode))
422 		nfs_wb_all(inode);
423 
424 	fattr = nfs_alloc_fattr();
425 	if (fattr == NULL)
426 		goto out;
427 	/*
428 	 * Return any delegations if we're going to change ACLs
429 	 */
430 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
431 		nfs_inode_return_delegation(inode);
432 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
433 	if (error == 0)
434 		nfs_refresh_inode(inode, fattr);
435 	nfs_free_fattr(fattr);
436 out:
437 	return error;
438 }
439 
440 /**
441  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
442  * @inode: inode of the file used
443  * @offset: file offset to start truncating
444  *
445  * This is a copy of the common vmtruncate, but with the locking
446  * corrected to take into account the fact that NFS requires
447  * inode->i_size to be updated under the inode->i_lock.
448  */
449 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
450 {
451 	loff_t oldsize;
452 	int err;
453 
454 	err = inode_newsize_ok(inode, offset);
455 	if (err)
456 		goto out;
457 
458 	spin_lock(&inode->i_lock);
459 	oldsize = inode->i_size;
460 	i_size_write(inode, offset);
461 	spin_unlock(&inode->i_lock);
462 
463 	truncate_pagecache(inode, oldsize, offset);
464 out:
465 	return err;
466 }
467 
468 /**
469  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
470  * @inode: pointer to struct inode
471  * @attr: pointer to struct iattr
472  *
473  * Note: we do this in the *proc.c in order to ensure that
474  *       it works for things like exclusive creates too.
475  */
476 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
477 {
478 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
479 		spin_lock(&inode->i_lock);
480 		if ((attr->ia_valid & ATTR_MODE) != 0) {
481 			int mode = attr->ia_mode & S_IALLUGO;
482 			mode |= inode->i_mode & ~S_IALLUGO;
483 			inode->i_mode = mode;
484 		}
485 		if ((attr->ia_valid & ATTR_UID) != 0)
486 			inode->i_uid = attr->ia_uid;
487 		if ((attr->ia_valid & ATTR_GID) != 0)
488 			inode->i_gid = attr->ia_gid;
489 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
490 		spin_unlock(&inode->i_lock);
491 	}
492 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
493 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
494 		nfs_vmtruncate(inode, attr->ia_size);
495 	}
496 }
497 
498 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
499 {
500 	struct inode *inode = dentry->d_inode;
501 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
502 	int err;
503 
504 	/* Flush out writes to the server in order to update c/mtime.  */
505 	if (S_ISREG(inode->i_mode)) {
506 		err = filemap_write_and_wait(inode->i_mapping);
507 		if (err)
508 			goto out;
509 	}
510 
511 	/*
512 	 * We may force a getattr if the user cares about atime.
513 	 *
514 	 * Note that we only have to check the vfsmount flags here:
515 	 *  - NFS always sets S_NOATIME by so checking it would give a
516 	 *    bogus result
517 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
518 	 *    no point in checking those.
519 	 */
520  	if ((mnt->mnt_flags & MNT_NOATIME) ||
521  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
522 		need_atime = 0;
523 
524 	if (need_atime)
525 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
526 	else
527 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
528 	if (!err) {
529 		generic_fillattr(inode, stat);
530 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
531 	}
532 out:
533 	return err;
534 }
535 
536 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
537 {
538 	atomic_set(&l_ctx->count, 1);
539 	l_ctx->lockowner = current->files;
540 	l_ctx->pid = current->tgid;
541 	INIT_LIST_HEAD(&l_ctx->list);
542 }
543 
544 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
545 {
546 	struct nfs_lock_context *pos;
547 
548 	list_for_each_entry(pos, &ctx->lock_context.list, list) {
549 		if (pos->lockowner != current->files)
550 			continue;
551 		if (pos->pid != current->tgid)
552 			continue;
553 		atomic_inc(&pos->count);
554 		return pos;
555 	}
556 	return NULL;
557 }
558 
559 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
560 {
561 	struct nfs_lock_context *res, *new = NULL;
562 	struct inode *inode = ctx->path.dentry->d_inode;
563 
564 	spin_lock(&inode->i_lock);
565 	res = __nfs_find_lock_context(ctx);
566 	if (res == NULL) {
567 		spin_unlock(&inode->i_lock);
568 		new = kmalloc(sizeof(*new), GFP_KERNEL);
569 		if (new == NULL)
570 			return NULL;
571 		nfs_init_lock_context(new);
572 		spin_lock(&inode->i_lock);
573 		res = __nfs_find_lock_context(ctx);
574 		if (res == NULL) {
575 			list_add_tail(&new->list, &ctx->lock_context.list);
576 			new->open_context = ctx;
577 			res = new;
578 			new = NULL;
579 		}
580 	}
581 	spin_unlock(&inode->i_lock);
582 	kfree(new);
583 	return res;
584 }
585 
586 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
587 {
588 	struct nfs_open_context *ctx = l_ctx->open_context;
589 	struct inode *inode = ctx->path.dentry->d_inode;
590 
591 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
592 		return;
593 	list_del(&l_ctx->list);
594 	spin_unlock(&inode->i_lock);
595 	kfree(l_ctx);
596 }
597 
598 /**
599  * nfs_close_context - Common close_context() routine NFSv2/v3
600  * @ctx: pointer to context
601  * @is_sync: is this a synchronous close
602  *
603  * always ensure that the attributes are up to date if we're mounted
604  * with close-to-open semantics
605  */
606 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
607 {
608 	struct inode *inode;
609 	struct nfs_server *server;
610 
611 	if (!(ctx->mode & FMODE_WRITE))
612 		return;
613 	if (!is_sync)
614 		return;
615 	inode = ctx->path.dentry->d_inode;
616 	if (!list_empty(&NFS_I(inode)->open_files))
617 		return;
618 	server = NFS_SERVER(inode);
619 	if (server->flags & NFS_MOUNT_NOCTO)
620 		return;
621 	nfs_revalidate_inode(server, inode);
622 }
623 
624 struct nfs_open_context *alloc_nfs_open_context(struct path *path, struct rpc_cred *cred, fmode_t f_mode)
625 {
626 	struct nfs_open_context *ctx;
627 
628 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
629 	if (ctx != NULL) {
630 		ctx->path = *path;
631 		path_get(&ctx->path);
632 		ctx->cred = get_rpccred(cred);
633 		ctx->state = NULL;
634 		ctx->mode = f_mode;
635 		ctx->flags = 0;
636 		ctx->error = 0;
637 		ctx->dir_cookie = 0;
638 		nfs_init_lock_context(&ctx->lock_context);
639 		ctx->lock_context.open_context = ctx;
640 		INIT_LIST_HEAD(&ctx->list);
641 	}
642 	return ctx;
643 }
644 
645 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
646 {
647 	if (ctx != NULL)
648 		atomic_inc(&ctx->lock_context.count);
649 	return ctx;
650 }
651 
652 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
653 {
654 	struct inode *inode = ctx->path.dentry->d_inode;
655 
656 	if (!list_empty(&ctx->list)) {
657 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
658 			return;
659 		list_del(&ctx->list);
660 		spin_unlock(&inode->i_lock);
661 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
662 		return;
663 	if (inode != NULL)
664 		NFS_PROTO(inode)->close_context(ctx, is_sync);
665 	if (ctx->cred != NULL)
666 		put_rpccred(ctx->cred);
667 	path_put(&ctx->path);
668 	kfree(ctx);
669 }
670 
671 void put_nfs_open_context(struct nfs_open_context *ctx)
672 {
673 	__put_nfs_open_context(ctx, 0);
674 }
675 
676 /*
677  * Ensure that mmap has a recent RPC credential for use when writing out
678  * shared pages
679  */
680 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
681 {
682 	struct inode *inode = filp->f_path.dentry->d_inode;
683 	struct nfs_inode *nfsi = NFS_I(inode);
684 
685 	filp->private_data = get_nfs_open_context(ctx);
686 	spin_lock(&inode->i_lock);
687 	list_add(&ctx->list, &nfsi->open_files);
688 	spin_unlock(&inode->i_lock);
689 }
690 
691 /*
692  * Given an inode, search for an open context with the desired characteristics
693  */
694 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
695 {
696 	struct nfs_inode *nfsi = NFS_I(inode);
697 	struct nfs_open_context *pos, *ctx = NULL;
698 
699 	spin_lock(&inode->i_lock);
700 	list_for_each_entry(pos, &nfsi->open_files, list) {
701 		if (cred != NULL && pos->cred != cred)
702 			continue;
703 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
704 			continue;
705 		ctx = get_nfs_open_context(pos);
706 		break;
707 	}
708 	spin_unlock(&inode->i_lock);
709 	return ctx;
710 }
711 
712 static void nfs_file_clear_open_context(struct file *filp)
713 {
714 	struct inode *inode = filp->f_path.dentry->d_inode;
715 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
716 
717 	if (ctx) {
718 		filp->private_data = NULL;
719 		spin_lock(&inode->i_lock);
720 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
721 		spin_unlock(&inode->i_lock);
722 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
723 	}
724 }
725 
726 /*
727  * These allocate and release file read/write context information.
728  */
729 int nfs_open(struct inode *inode, struct file *filp)
730 {
731 	struct nfs_open_context *ctx;
732 	struct rpc_cred *cred;
733 
734 	cred = rpc_lookup_cred();
735 	if (IS_ERR(cred))
736 		return PTR_ERR(cred);
737 	ctx = alloc_nfs_open_context(&filp->f_path, cred, filp->f_mode);
738 	put_rpccred(cred);
739 	if (ctx == NULL)
740 		return -ENOMEM;
741 	nfs_file_set_open_context(filp, ctx);
742 	put_nfs_open_context(ctx);
743 	nfs_fscache_set_inode_cookie(inode, filp);
744 	return 0;
745 }
746 
747 int nfs_release(struct inode *inode, struct file *filp)
748 {
749 	nfs_file_clear_open_context(filp);
750 	return 0;
751 }
752 
753 /*
754  * This function is called whenever some part of NFS notices that
755  * the cached attributes have to be refreshed.
756  */
757 int
758 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
759 {
760 	int		 status = -ESTALE;
761 	struct nfs_fattr *fattr = NULL;
762 	struct nfs_inode *nfsi = NFS_I(inode);
763 
764 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
765 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
766 
767 	if (is_bad_inode(inode))
768 		goto out;
769 	if (NFS_STALE(inode))
770 		goto out;
771 
772 	status = -ENOMEM;
773 	fattr = nfs_alloc_fattr();
774 	if (fattr == NULL)
775 		goto out;
776 
777 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
778 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
779 	if (status != 0) {
780 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
781 			 inode->i_sb->s_id,
782 			 (long long)NFS_FILEID(inode), status);
783 		if (status == -ESTALE) {
784 			nfs_zap_caches(inode);
785 			if (!S_ISDIR(inode->i_mode))
786 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
787 		}
788 		goto out;
789 	}
790 
791 	status = nfs_refresh_inode(inode, fattr);
792 	if (status) {
793 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
794 			 inode->i_sb->s_id,
795 			 (long long)NFS_FILEID(inode), status);
796 		goto out;
797 	}
798 
799 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
800 		nfs_zap_acl_cache(inode);
801 
802 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
803 		inode->i_sb->s_id,
804 		(long long)NFS_FILEID(inode));
805 
806  out:
807 	nfs_free_fattr(fattr);
808 	return status;
809 }
810 
811 int nfs_attribute_timeout(struct inode *inode)
812 {
813 	struct nfs_inode *nfsi = NFS_I(inode);
814 
815 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
816 }
817 
818 static int nfs_attribute_cache_expired(struct inode *inode)
819 {
820 	if (nfs_have_delegated_attributes(inode))
821 		return 0;
822 	return nfs_attribute_timeout(inode);
823 }
824 
825 /**
826  * nfs_revalidate_inode - Revalidate the inode attributes
827  * @server - pointer to nfs_server struct
828  * @inode - pointer to inode struct
829  *
830  * Updates inode attribute information by retrieving the data from the server.
831  */
832 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
833 {
834 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
835 			&& !nfs_attribute_cache_expired(inode))
836 		return NFS_STALE(inode) ? -ESTALE : 0;
837 	return __nfs_revalidate_inode(server, inode);
838 }
839 
840 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
841 {
842 	struct nfs_inode *nfsi = NFS_I(inode);
843 
844 	if (mapping->nrpages != 0) {
845 		int ret = invalidate_inode_pages2(mapping);
846 		if (ret < 0)
847 			return ret;
848 	}
849 	spin_lock(&inode->i_lock);
850 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
851 	if (S_ISDIR(inode->i_mode))
852 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
853 	spin_unlock(&inode->i_lock);
854 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
855 	nfs_fscache_reset_inode_cookie(inode);
856 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
857 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
858 	return 0;
859 }
860 
861 /**
862  * nfs_revalidate_mapping - Revalidate the pagecache
863  * @inode - pointer to host inode
864  * @mapping - pointer to mapping
865  */
866 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
867 {
868 	struct nfs_inode *nfsi = NFS_I(inode);
869 	int ret = 0;
870 
871 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
872 			|| nfs_attribute_cache_expired(inode)
873 			|| NFS_STALE(inode)) {
874 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
875 		if (ret < 0)
876 			goto out;
877 	}
878 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
879 		ret = nfs_invalidate_mapping(inode, mapping);
880 out:
881 	return ret;
882 }
883 
884 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
885 {
886 	struct nfs_inode *nfsi = NFS_I(inode);
887 	unsigned long ret = 0;
888 
889 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
890 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
891 			&& nfsi->change_attr == fattr->pre_change_attr) {
892 		nfsi->change_attr = fattr->change_attr;
893 		if (S_ISDIR(inode->i_mode))
894 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
895 		ret |= NFS_INO_INVALID_ATTR;
896 	}
897 	/* If we have atomic WCC data, we may update some attributes */
898 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
899 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
900 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
901 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
902 		ret |= NFS_INO_INVALID_ATTR;
903 	}
904 
905 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
906 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
907 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
908 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
909 		if (S_ISDIR(inode->i_mode))
910 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
911 		ret |= NFS_INO_INVALID_ATTR;
912 	}
913 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
914 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
915 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
916 			&& nfsi->npages == 0) {
917 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
918 		ret |= NFS_INO_INVALID_ATTR;
919 	}
920 	return ret;
921 }
922 
923 /**
924  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
925  * @inode - pointer to inode
926  * @fattr - updated attributes
927  *
928  * Verifies the attribute cache. If we have just changed the attributes,
929  * so that fattr carries weak cache consistency data, then it may
930  * also update the ctime/mtime/change_attribute.
931  */
932 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
933 {
934 	struct nfs_inode *nfsi = NFS_I(inode);
935 	loff_t cur_size, new_isize;
936 	unsigned long invalid = 0;
937 
938 
939 	/* Has the inode gone and changed behind our back? */
940 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
941 		return -EIO;
942 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
943 		return -EIO;
944 
945 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
946 			nfsi->change_attr != fattr->change_attr)
947 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
948 
949 	/* Verify a few of the more important attributes */
950 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
951 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
952 
953 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
954 		cur_size = i_size_read(inode);
955 		new_isize = nfs_size_to_loff_t(fattr->size);
956 		if (cur_size != new_isize && nfsi->npages == 0)
957 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
958 	}
959 
960 	/* Have any file permissions changed? */
961 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
962 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
963 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
964 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
965 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
966 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
967 
968 	/* Has the link count changed? */
969 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
970 		invalid |= NFS_INO_INVALID_ATTR;
971 
972 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
973 		invalid |= NFS_INO_INVALID_ATIME;
974 
975 	if (invalid != 0)
976 		nfsi->cache_validity |= invalid;
977 
978 	nfsi->read_cache_jiffies = fattr->time_start;
979 	return 0;
980 }
981 
982 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
983 {
984 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
985 		return 0;
986 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
987 }
988 
989 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
990 {
991 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
992 		return 0;
993 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
994 }
995 
996 static atomic_long_t nfs_attr_generation_counter;
997 
998 static unsigned long nfs_read_attr_generation_counter(void)
999 {
1000 	return atomic_long_read(&nfs_attr_generation_counter);
1001 }
1002 
1003 unsigned long nfs_inc_attr_generation_counter(void)
1004 {
1005 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1006 }
1007 
1008 void nfs_fattr_init(struct nfs_fattr *fattr)
1009 {
1010 	fattr->valid = 0;
1011 	fattr->time_start = jiffies;
1012 	fattr->gencount = nfs_inc_attr_generation_counter();
1013 }
1014 
1015 struct nfs_fattr *nfs_alloc_fattr(void)
1016 {
1017 	struct nfs_fattr *fattr;
1018 
1019 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1020 	if (fattr != NULL)
1021 		nfs_fattr_init(fattr);
1022 	return fattr;
1023 }
1024 
1025 struct nfs_fh *nfs_alloc_fhandle(void)
1026 {
1027 	struct nfs_fh *fh;
1028 
1029 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1030 	if (fh != NULL)
1031 		fh->size = 0;
1032 	return fh;
1033 }
1034 
1035 /**
1036  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1037  * @inode - pointer to inode
1038  * @fattr - attributes
1039  *
1040  * Attempt to divine whether or not an RPC call reply carrying stale
1041  * attributes got scheduled after another call carrying updated ones.
1042  *
1043  * To do so, the function first assumes that a more recent ctime means
1044  * that the attributes in fattr are newer, however it also attempt to
1045  * catch the case where ctime either didn't change, or went backwards
1046  * (if someone reset the clock on the server) by looking at whether
1047  * or not this RPC call was started after the inode was last updated.
1048  * Note also the check for wraparound of 'attr_gencount'
1049  *
1050  * The function returns 'true' if it thinks the attributes in 'fattr' are
1051  * more recent than the ones cached in the inode.
1052  *
1053  */
1054 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1055 {
1056 	const struct nfs_inode *nfsi = NFS_I(inode);
1057 
1058 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1059 		nfs_ctime_need_update(inode, fattr) ||
1060 		nfs_size_need_update(inode, fattr) ||
1061 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1062 }
1063 
1064 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1065 {
1066 	if (nfs_inode_attrs_need_update(inode, fattr))
1067 		return nfs_update_inode(inode, fattr);
1068 	return nfs_check_inode_attributes(inode, fattr);
1069 }
1070 
1071 /**
1072  * nfs_refresh_inode - try to update the inode attribute cache
1073  * @inode - pointer to inode
1074  * @fattr - updated attributes
1075  *
1076  * Check that an RPC call that returned attributes has not overlapped with
1077  * other recent updates of the inode metadata, then decide whether it is
1078  * safe to do a full update of the inode attributes, or whether just to
1079  * call nfs_check_inode_attributes.
1080  */
1081 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1082 {
1083 	int status;
1084 
1085 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1086 		return 0;
1087 	spin_lock(&inode->i_lock);
1088 	status = nfs_refresh_inode_locked(inode, fattr);
1089 	spin_unlock(&inode->i_lock);
1090 
1091 	return status;
1092 }
1093 
1094 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1095 {
1096 	struct nfs_inode *nfsi = NFS_I(inode);
1097 
1098 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1099 	if (S_ISDIR(inode->i_mode))
1100 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1101 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1102 		return 0;
1103 	return nfs_refresh_inode_locked(inode, fattr);
1104 }
1105 
1106 /**
1107  * nfs_post_op_update_inode - try to update the inode attribute cache
1108  * @inode - pointer to inode
1109  * @fattr - updated attributes
1110  *
1111  * After an operation that has changed the inode metadata, mark the
1112  * attribute cache as being invalid, then try to update it.
1113  *
1114  * NB: if the server didn't return any post op attributes, this
1115  * function will force the retrieval of attributes before the next
1116  * NFS request.  Thus it should be used only for operations that
1117  * are expected to change one or more attributes, to avoid
1118  * unnecessary NFS requests and trips through nfs_update_inode().
1119  */
1120 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1121 {
1122 	int status;
1123 
1124 	spin_lock(&inode->i_lock);
1125 	status = nfs_post_op_update_inode_locked(inode, fattr);
1126 	spin_unlock(&inode->i_lock);
1127 	return status;
1128 }
1129 
1130 /**
1131  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1132  * @inode - pointer to inode
1133  * @fattr - updated attributes
1134  *
1135  * After an operation that has changed the inode metadata, mark the
1136  * attribute cache as being invalid, then try to update it. Fake up
1137  * weak cache consistency data, if none exist.
1138  *
1139  * This function is mainly designed to be used by the ->write_done() functions.
1140  */
1141 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1142 {
1143 	int status;
1144 
1145 	spin_lock(&inode->i_lock);
1146 	/* Don't do a WCC update if these attributes are already stale */
1147 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1148 			!nfs_inode_attrs_need_update(inode, fattr)) {
1149 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1150 				| NFS_ATTR_FATTR_PRESIZE
1151 				| NFS_ATTR_FATTR_PREMTIME
1152 				| NFS_ATTR_FATTR_PRECTIME);
1153 		goto out_noforce;
1154 	}
1155 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1156 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1157 		fattr->pre_change_attr = NFS_I(inode)->change_attr;
1158 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1159 	}
1160 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1161 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1162 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1163 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1164 	}
1165 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1166 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1167 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1168 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1169 	}
1170 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1171 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1172 		fattr->pre_size = i_size_read(inode);
1173 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1174 	}
1175 out_noforce:
1176 	status = nfs_post_op_update_inode_locked(inode, fattr);
1177 	spin_unlock(&inode->i_lock);
1178 	return status;
1179 }
1180 
1181 /*
1182  * Many nfs protocol calls return the new file attributes after
1183  * an operation.  Here we update the inode to reflect the state
1184  * of the server's inode.
1185  *
1186  * This is a bit tricky because we have to make sure all dirty pages
1187  * have been sent off to the server before calling invalidate_inode_pages.
1188  * To make sure no other process adds more write requests while we try
1189  * our best to flush them, we make them sleep during the attribute refresh.
1190  *
1191  * A very similar scenario holds for the dir cache.
1192  */
1193 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1194 {
1195 	struct nfs_server *server;
1196 	struct nfs_inode *nfsi = NFS_I(inode);
1197 	loff_t cur_isize, new_isize;
1198 	unsigned long invalid = 0;
1199 	unsigned long now = jiffies;
1200 	unsigned long save_cache_validity;
1201 
1202 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1203 			__func__, inode->i_sb->s_id, inode->i_ino,
1204 			atomic_read(&inode->i_count), fattr->valid);
1205 
1206 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1207 		goto out_fileid;
1208 
1209 	/*
1210 	 * Make sure the inode's type hasn't changed.
1211 	 */
1212 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1213 		goto out_changed;
1214 
1215 	server = NFS_SERVER(inode);
1216 	/* Update the fsid? */
1217 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1218 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1219 			!IS_AUTOMOUNT(inode))
1220 		server->fsid = fattr->fsid;
1221 
1222 	/*
1223 	 * Update the read time so we don't revalidate too often.
1224 	 */
1225 	nfsi->read_cache_jiffies = fattr->time_start;
1226 
1227 	save_cache_validity = nfsi->cache_validity;
1228 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1229 			| NFS_INO_INVALID_ATIME
1230 			| NFS_INO_REVAL_FORCED
1231 			| NFS_INO_REVAL_PAGECACHE);
1232 
1233 	/* Do atomic weak cache consistency updates */
1234 	invalid |= nfs_wcc_update_inode(inode, fattr);
1235 
1236 	/* More cache consistency checks */
1237 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1238 		if (nfsi->change_attr != fattr->change_attr) {
1239 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1240 					inode->i_sb->s_id, inode->i_ino);
1241 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1242 			if (S_ISDIR(inode->i_mode))
1243 				nfs_force_lookup_revalidate(inode);
1244 			nfsi->change_attr = fattr->change_attr;
1245 		}
1246 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1247 		invalid |= save_cache_validity;
1248 
1249 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1250 		/* NFSv2/v3: Check if the mtime agrees */
1251 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1252 			dprintk("NFS: mtime change on server for file %s/%ld\n",
1253 					inode->i_sb->s_id, inode->i_ino);
1254 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1255 			if (S_ISDIR(inode->i_mode))
1256 				nfs_force_lookup_revalidate(inode);
1257 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1258 		}
1259 	} else if (server->caps & NFS_CAP_MTIME)
1260 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1261 				| NFS_INO_INVALID_DATA
1262 				| NFS_INO_REVAL_PAGECACHE
1263 				| NFS_INO_REVAL_FORCED);
1264 
1265 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1266 		/* If ctime has changed we should definitely clear access+acl caches */
1267 		if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1268 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1269 			/* and probably clear data for a directory too as utimes can cause
1270 			 * havoc with our cache.
1271 			 */
1272 			if (S_ISDIR(inode->i_mode)) {
1273 				invalid |= NFS_INO_INVALID_DATA;
1274 				nfs_force_lookup_revalidate(inode);
1275 			}
1276 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1277 		}
1278 	} else if (server->caps & NFS_CAP_CTIME)
1279 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1280 				| NFS_INO_INVALID_ACCESS
1281 				| NFS_INO_INVALID_ACL
1282 				| NFS_INO_REVAL_FORCED);
1283 
1284 	/* Check if our cached file size is stale */
1285 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1286 		new_isize = nfs_size_to_loff_t(fattr->size);
1287 		cur_isize = i_size_read(inode);
1288 		if (new_isize != cur_isize) {
1289 			/* Do we perhaps have any outstanding writes, or has
1290 			 * the file grown beyond our last write? */
1291 			if (nfsi->npages == 0 || new_isize > cur_isize) {
1292 				i_size_write(inode, new_isize);
1293 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1294 			}
1295 			dprintk("NFS: isize change on server for file %s/%ld\n",
1296 					inode->i_sb->s_id, inode->i_ino);
1297 		}
1298 	} else
1299 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1300 				| NFS_INO_REVAL_PAGECACHE
1301 				| NFS_INO_REVAL_FORCED);
1302 
1303 
1304 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1305 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1306 	else if (server->caps & NFS_CAP_ATIME)
1307 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1308 				| NFS_INO_REVAL_FORCED);
1309 
1310 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1311 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1312 			umode_t newmode = inode->i_mode & S_IFMT;
1313 			newmode |= fattr->mode & S_IALLUGO;
1314 			inode->i_mode = newmode;
1315 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1316 		}
1317 	} else if (server->caps & NFS_CAP_MODE)
1318 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1319 				| NFS_INO_INVALID_ACCESS
1320 				| NFS_INO_INVALID_ACL
1321 				| NFS_INO_REVAL_FORCED);
1322 
1323 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1324 		if (inode->i_uid != fattr->uid) {
1325 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1326 			inode->i_uid = fattr->uid;
1327 		}
1328 	} else if (server->caps & NFS_CAP_OWNER)
1329 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1330 				| NFS_INO_INVALID_ACCESS
1331 				| NFS_INO_INVALID_ACL
1332 				| NFS_INO_REVAL_FORCED);
1333 
1334 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1335 		if (inode->i_gid != fattr->gid) {
1336 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1337 			inode->i_gid = fattr->gid;
1338 		}
1339 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1340 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1341 				| NFS_INO_INVALID_ACCESS
1342 				| NFS_INO_INVALID_ACL
1343 				| NFS_INO_REVAL_FORCED);
1344 
1345 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1346 		if (inode->i_nlink != fattr->nlink) {
1347 			invalid |= NFS_INO_INVALID_ATTR;
1348 			if (S_ISDIR(inode->i_mode))
1349 				invalid |= NFS_INO_INVALID_DATA;
1350 			inode->i_nlink = fattr->nlink;
1351 		}
1352 	} else if (server->caps & NFS_CAP_NLINK)
1353 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1354 				| NFS_INO_REVAL_FORCED);
1355 
1356 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1357 		/*
1358 		 * report the blocks in 512byte units
1359 		 */
1360 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1361  	}
1362 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1363 		inode->i_blocks = fattr->du.nfs2.blocks;
1364 
1365 	/* Update attrtimeo value if we're out of the unstable period */
1366 	if (invalid & NFS_INO_INVALID_ATTR) {
1367 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1368 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1369 		nfsi->attrtimeo_timestamp = now;
1370 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1371 	} else {
1372 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1373 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1374 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1375 			nfsi->attrtimeo_timestamp = now;
1376 		}
1377 	}
1378 	invalid &= ~NFS_INO_INVALID_ATTR;
1379 	/* Don't invalidate the data if we were to blame */
1380 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1381 				|| S_ISLNK(inode->i_mode)))
1382 		invalid &= ~NFS_INO_INVALID_DATA;
1383 	if (!nfs_have_delegation(inode, FMODE_READ) ||
1384 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1385 		nfsi->cache_validity |= invalid;
1386 
1387 	return 0;
1388  out_changed:
1389 	/*
1390 	 * Big trouble! The inode has become a different object.
1391 	 */
1392 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1393 			__func__, inode->i_ino, inode->i_mode, fattr->mode);
1394  out_err:
1395 	/*
1396 	 * No need to worry about unhashing the dentry, as the
1397 	 * lookup validation will know that the inode is bad.
1398 	 * (But we fall through to invalidate the caches.)
1399 	 */
1400 	nfs_invalidate_inode(inode);
1401 	return -ESTALE;
1402 
1403  out_fileid:
1404 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1405 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1406 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1407 		(long long)nfsi->fileid, (long long)fattr->fileid);
1408 	goto out_err;
1409 }
1410 
1411 
1412 #ifdef CONFIG_NFS_V4
1413 
1414 /*
1415  * Clean out any remaining NFSv4 state that might be left over due
1416  * to open() calls that passed nfs_atomic_lookup, but failed to call
1417  * nfs_open().
1418  */
1419 void nfs4_evict_inode(struct inode *inode)
1420 {
1421 	pnfs_destroy_layout(NFS_I(inode));
1422 	truncate_inode_pages(&inode->i_data, 0);
1423 	end_writeback(inode);
1424 	/* If we are holding a delegation, return it! */
1425 	nfs_inode_return_delegation_noreclaim(inode);
1426 	/* First call standard NFS clear_inode() code */
1427 	nfs_clear_inode(inode);
1428 }
1429 #endif
1430 
1431 struct inode *nfs_alloc_inode(struct super_block *sb)
1432 {
1433 	struct nfs_inode *nfsi;
1434 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1435 	if (!nfsi)
1436 		return NULL;
1437 	nfsi->flags = 0UL;
1438 	nfsi->cache_validity = 0UL;
1439 #ifdef CONFIG_NFS_V3_ACL
1440 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1441 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1442 #endif
1443 #ifdef CONFIG_NFS_V4
1444 	nfsi->nfs4_acl = NULL;
1445 #endif /* CONFIG_NFS_V4 */
1446 	return &nfsi->vfs_inode;
1447 }
1448 
1449 static void nfs_i_callback(struct rcu_head *head)
1450 {
1451 	struct inode *inode = container_of(head, struct inode, i_rcu);
1452 	INIT_LIST_HEAD(&inode->i_dentry);
1453 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1454 }
1455 
1456 void nfs_destroy_inode(struct inode *inode)
1457 {
1458 	call_rcu(&inode->i_rcu, nfs_i_callback);
1459 }
1460 
1461 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1462 {
1463 #ifdef CONFIG_NFS_V4
1464 	INIT_LIST_HEAD(&nfsi->open_states);
1465 	nfsi->delegation = NULL;
1466 	nfsi->delegation_state = 0;
1467 	init_rwsem(&nfsi->rwsem);
1468 	nfsi->layout = NULL;
1469 #endif
1470 }
1471 
1472 static void init_once(void *foo)
1473 {
1474 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1475 
1476 	inode_init_once(&nfsi->vfs_inode);
1477 	INIT_LIST_HEAD(&nfsi->open_files);
1478 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1479 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1480 	INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1481 	nfsi->npages = 0;
1482 	nfsi->ncommit = 0;
1483 	atomic_set(&nfsi->silly_count, 1);
1484 	INIT_HLIST_HEAD(&nfsi->silly_list);
1485 	init_waitqueue_head(&nfsi->waitqueue);
1486 	nfs4_init_once(nfsi);
1487 }
1488 
1489 static int __init nfs_init_inodecache(void)
1490 {
1491 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1492 					     sizeof(struct nfs_inode),
1493 					     0, (SLAB_RECLAIM_ACCOUNT|
1494 						SLAB_MEM_SPREAD),
1495 					     init_once);
1496 	if (nfs_inode_cachep == NULL)
1497 		return -ENOMEM;
1498 
1499 	return 0;
1500 }
1501 
1502 static void nfs_destroy_inodecache(void)
1503 {
1504 	kmem_cache_destroy(nfs_inode_cachep);
1505 }
1506 
1507 struct workqueue_struct *nfsiod_workqueue;
1508 
1509 /*
1510  * start up the nfsiod workqueue
1511  */
1512 static int nfsiod_start(void)
1513 {
1514 	struct workqueue_struct *wq;
1515 	dprintk("RPC:       creating workqueue nfsiod\n");
1516 	wq = alloc_workqueue("nfsiod", WQ_RESCUER, 0);
1517 	if (wq == NULL)
1518 		return -ENOMEM;
1519 	nfsiod_workqueue = wq;
1520 	return 0;
1521 }
1522 
1523 /*
1524  * Destroy the nfsiod workqueue
1525  */
1526 static void nfsiod_stop(void)
1527 {
1528 	struct workqueue_struct *wq;
1529 
1530 	wq = nfsiod_workqueue;
1531 	if (wq == NULL)
1532 		return;
1533 	nfsiod_workqueue = NULL;
1534 	destroy_workqueue(wq);
1535 }
1536 
1537 /*
1538  * Initialize NFS
1539  */
1540 static int __init init_nfs_fs(void)
1541 {
1542 	int err;
1543 
1544 	err = nfs_idmap_init();
1545 	if (err < 0)
1546 		goto out9;
1547 
1548 	err = nfs_dns_resolver_init();
1549 	if (err < 0)
1550 		goto out8;
1551 
1552 	err = nfs_fscache_register();
1553 	if (err < 0)
1554 		goto out7;
1555 
1556 	err = nfsiod_start();
1557 	if (err)
1558 		goto out6;
1559 
1560 	err = nfs_fs_proc_init();
1561 	if (err)
1562 		goto out5;
1563 
1564 	err = nfs_init_nfspagecache();
1565 	if (err)
1566 		goto out4;
1567 
1568 	err = nfs_init_inodecache();
1569 	if (err)
1570 		goto out3;
1571 
1572 	err = nfs_init_readpagecache();
1573 	if (err)
1574 		goto out2;
1575 
1576 	err = nfs_init_writepagecache();
1577 	if (err)
1578 		goto out1;
1579 
1580 	err = nfs_init_directcache();
1581 	if (err)
1582 		goto out0;
1583 
1584 #ifdef CONFIG_PROC_FS
1585 	rpc_proc_register(&nfs_rpcstat);
1586 #endif
1587 	if ((err = register_nfs_fs()) != 0)
1588 		goto out;
1589 	return 0;
1590 out:
1591 #ifdef CONFIG_PROC_FS
1592 	rpc_proc_unregister("nfs");
1593 #endif
1594 	nfs_destroy_directcache();
1595 out0:
1596 	nfs_destroy_writepagecache();
1597 out1:
1598 	nfs_destroy_readpagecache();
1599 out2:
1600 	nfs_destroy_inodecache();
1601 out3:
1602 	nfs_destroy_nfspagecache();
1603 out4:
1604 	nfs_fs_proc_exit();
1605 out5:
1606 	nfsiod_stop();
1607 out6:
1608 	nfs_fscache_unregister();
1609 out7:
1610 	nfs_dns_resolver_destroy();
1611 out8:
1612 	nfs_idmap_quit();
1613 out9:
1614 	return err;
1615 }
1616 
1617 static void __exit exit_nfs_fs(void)
1618 {
1619 	nfs_destroy_directcache();
1620 	nfs_destroy_writepagecache();
1621 	nfs_destroy_readpagecache();
1622 	nfs_destroy_inodecache();
1623 	nfs_destroy_nfspagecache();
1624 	nfs_fscache_unregister();
1625 	nfs_dns_resolver_destroy();
1626 	nfs_idmap_quit();
1627 #ifdef CONFIG_PROC_FS
1628 	rpc_proc_unregister("nfs");
1629 #endif
1630 	nfs_cleanup_cb_ident_idr();
1631 	unregister_nfs_fs();
1632 	nfs_fs_proc_exit();
1633 	nfsiod_stop();
1634 }
1635 
1636 /* Not quite true; I just maintain it */
1637 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1638 MODULE_LICENSE("GPL");
1639 module_param(enable_ino64, bool, 0644);
1640 
1641 module_init(init_nfs_fs)
1642 module_exit(exit_nfs_fs)
1643