xref: /linux/fs/nfs/inode.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
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 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 
43 #include <asm/system.h>
44 #include <asm/uaccess.h>
45 
46 #include "nfs4_fs.h"
47 #include "callback.h"
48 #include "delegation.h"
49 #include "iostat.h"
50 #include "internal.h"
51 #include "fscache.h"
52 #include "dns_resolve.h"
53 #include "pnfs.h"
54 
55 #define NFSDBG_FACILITY		NFSDBG_VFS
56 
57 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
58 
59 /* Default is to see 64-bit inode numbers */
60 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
61 
62 static void nfs_invalidate_inode(struct inode *);
63 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
64 
65 static struct kmem_cache * nfs_inode_cachep;
66 
67 static inline unsigned long
68 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
69 {
70 	return nfs_fileid_to_ino_t(fattr->fileid);
71 }
72 
73 /**
74  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
75  * @word: long word containing the bit lock
76  */
77 int nfs_wait_bit_killable(void *word)
78 {
79 	if (fatal_signal_pending(current))
80 		return -ERESTARTSYS;
81 	freezable_schedule();
82 	return 0;
83 }
84 
85 /**
86  * nfs_compat_user_ino64 - returns the user-visible inode number
87  * @fileid: 64-bit fileid
88  *
89  * This function returns a 32-bit inode number if the boot parameter
90  * nfs.enable_ino64 is zero.
91  */
92 u64 nfs_compat_user_ino64(u64 fileid)
93 {
94 #ifdef CONFIG_COMPAT
95 	compat_ulong_t ino;
96 #else
97 	unsigned long ino;
98 #endif
99 
100 	if (enable_ino64)
101 		return fileid;
102 	ino = fileid;
103 	if (sizeof(ino) < sizeof(fileid))
104 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
105 	return ino;
106 }
107 
108 static void nfs_clear_inode(struct inode *inode)
109 {
110 	/*
111 	 * The following should never happen...
112 	 */
113 	BUG_ON(nfs_have_writebacks(inode));
114 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
115 	nfs_zap_acl_cache(inode);
116 	nfs_access_zap_cache(inode);
117 	nfs_fscache_release_inode_cookie(inode);
118 }
119 
120 void nfs_evict_inode(struct inode *inode)
121 {
122 	truncate_inode_pages(&inode->i_data, 0);
123 	end_writeback(inode);
124 	nfs_clear_inode(inode);
125 }
126 
127 /**
128  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
129  */
130 int nfs_sync_mapping(struct address_space *mapping)
131 {
132 	int ret = 0;
133 
134 	if (mapping->nrpages != 0) {
135 		unmap_mapping_range(mapping, 0, 0, 0);
136 		ret = nfs_wb_all(mapping->host);
137 	}
138 	return ret;
139 }
140 
141 /*
142  * Invalidate the local caches
143  */
144 static void nfs_zap_caches_locked(struct inode *inode)
145 {
146 	struct nfs_inode *nfsi = NFS_I(inode);
147 	int mode = inode->i_mode;
148 
149 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
150 
151 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
152 	nfsi->attrtimeo_timestamp = jiffies;
153 
154 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
155 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
156 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
157 	else
158 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
159 }
160 
161 void nfs_zap_caches(struct inode *inode)
162 {
163 	spin_lock(&inode->i_lock);
164 	nfs_zap_caches_locked(inode);
165 	spin_unlock(&inode->i_lock);
166 }
167 
168 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
169 {
170 	if (mapping->nrpages != 0) {
171 		spin_lock(&inode->i_lock);
172 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
173 		spin_unlock(&inode->i_lock);
174 	}
175 }
176 
177 void nfs_zap_acl_cache(struct inode *inode)
178 {
179 	void (*clear_acl_cache)(struct inode *);
180 
181 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
182 	if (clear_acl_cache != NULL)
183 		clear_acl_cache(inode);
184 	spin_lock(&inode->i_lock);
185 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
186 	spin_unlock(&inode->i_lock);
187 }
188 
189 void nfs_invalidate_atime(struct inode *inode)
190 {
191 	spin_lock(&inode->i_lock);
192 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
193 	spin_unlock(&inode->i_lock);
194 }
195 
196 /*
197  * Invalidate, but do not unhash, the inode.
198  * NB: must be called with inode->i_lock held!
199  */
200 static void nfs_invalidate_inode(struct inode *inode)
201 {
202 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
203 	nfs_zap_caches_locked(inode);
204 }
205 
206 struct nfs_find_desc {
207 	struct nfs_fh		*fh;
208 	struct nfs_fattr	*fattr;
209 };
210 
211 /*
212  * In NFSv3 we can have 64bit inode numbers. In order to support
213  * this, and re-exported directories (also seen in NFSv2)
214  * we are forced to allow 2 different inodes to have the same
215  * i_ino.
216  */
217 static int
218 nfs_find_actor(struct inode *inode, void *opaque)
219 {
220 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
221 	struct nfs_fh		*fh = desc->fh;
222 	struct nfs_fattr	*fattr = desc->fattr;
223 
224 	if (NFS_FILEID(inode) != fattr->fileid)
225 		return 0;
226 	if (nfs_compare_fh(NFS_FH(inode), fh))
227 		return 0;
228 	if (is_bad_inode(inode) || NFS_STALE(inode))
229 		return 0;
230 	return 1;
231 }
232 
233 static int
234 nfs_init_locked(struct inode *inode, void *opaque)
235 {
236 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
237 	struct nfs_fattr	*fattr = desc->fattr;
238 
239 	set_nfs_fileid(inode, fattr->fileid);
240 	nfs_copy_fh(NFS_FH(inode), desc->fh);
241 	return 0;
242 }
243 
244 /*
245  * This is our front-end to iget that looks up inodes by file handle
246  * instead of inode number.
247  */
248 struct inode *
249 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
250 {
251 	struct nfs_find_desc desc = {
252 		.fh	= fh,
253 		.fattr	= fattr
254 	};
255 	struct inode *inode = ERR_PTR(-ENOENT);
256 	unsigned long hash;
257 
258 	nfs_attr_check_mountpoint(sb, fattr);
259 
260 	if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
261 	    !nfs_attr_use_mounted_on_fileid(fattr))
262 		goto out_no_inode;
263 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
264 		goto out_no_inode;
265 
266 	hash = nfs_fattr_to_ino_t(fattr);
267 
268 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
269 	if (inode == NULL) {
270 		inode = ERR_PTR(-ENOMEM);
271 		goto out_no_inode;
272 	}
273 
274 	if (inode->i_state & I_NEW) {
275 		struct nfs_inode *nfsi = NFS_I(inode);
276 		unsigned long now = jiffies;
277 
278 		/* We set i_ino for the few things that still rely on it,
279 		 * such as stat(2) */
280 		inode->i_ino = hash;
281 
282 		/* We can't support update_atime(), since the server will reset it */
283 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
284 		inode->i_mode = fattr->mode;
285 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
286 				&& nfs_server_capable(inode, NFS_CAP_MODE))
287 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
288 				| NFS_INO_INVALID_ACCESS
289 				| NFS_INO_INVALID_ACL;
290 		/* Why so? Because we want revalidate for devices/FIFOs, and
291 		 * that's precisely what we have in nfs_file_inode_operations.
292 		 */
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_SB(sb)->nfs_client->rpc_ops->file_ops;
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 			inode->i_data.a_ops = &nfs_dir_aops;
302 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS))
303 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
304 			/* Deal with crossing mountpoints */
305 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
306 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
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 				inode->i_flags |= S_AUTOMOUNT;
313 			}
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);
318 
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 		inode->i_version = 0;
323 		inode->i_size = 0;
324 		clear_nlink(inode);
325 		inode->i_uid = -2;
326 		inode->i_gid = -2;
327 		inode->i_blocks = 0;
328 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
329 
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 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
335 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
336 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
337 			inode->i_mtime = fattr->mtime;
338 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
339 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
340 				| NFS_INO_INVALID_DATA;
341 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
342 			inode->i_ctime = fattr->ctime;
343 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
344 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
345 				| NFS_INO_INVALID_ACCESS
346 				| NFS_INO_INVALID_ACL;
347 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
348 			inode->i_version = fattr->change_attr;
349 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
350 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
351 				| NFS_INO_INVALID_DATA;
352 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
353 			inode->i_size = nfs_size_to_loff_t(fattr->size);
354 		else
355 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
356 				| NFS_INO_INVALID_DATA
357 				| NFS_INO_REVAL_PAGECACHE;
358 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
359 			set_nlink(inode, fattr->nlink);
360 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
361 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
362 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
363 			inode->i_uid = fattr->uid;
364 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
365 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
366 				| NFS_INO_INVALID_ACCESS
367 				| NFS_INO_INVALID_ACL;
368 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
369 			inode->i_gid = fattr->gid;
370 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
371 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
372 				| NFS_INO_INVALID_ACCESS
373 				| NFS_INO_INVALID_ACL;
374 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
375 			inode->i_blocks = fattr->du.nfs2.blocks;
376 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
377 			/*
378 			 * report the blocks in 512byte units
379 			 */
380 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
381 		}
382 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
383 		nfsi->attrtimeo_timestamp = now;
384 		nfsi->access_cache = RB_ROOT;
385 
386 		nfs_fscache_init_inode_cookie(inode);
387 
388 		unlock_new_inode(inode);
389 	} else
390 		nfs_refresh_inode(inode, fattr);
391 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
392 		inode->i_sb->s_id,
393 		(long long)NFS_FILEID(inode),
394 		atomic_read(&inode->i_count));
395 
396 out:
397 	return inode;
398 
399 out_no_inode:
400 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
401 	goto out;
402 }
403 
404 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
405 
406 int
407 nfs_setattr(struct dentry *dentry, struct iattr *attr)
408 {
409 	struct inode *inode = dentry->d_inode;
410 	struct nfs_fattr *fattr;
411 	int error = -ENOMEM;
412 
413 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
414 
415 	/* skip mode change if it's just for clearing setuid/setgid */
416 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
417 		attr->ia_valid &= ~ATTR_MODE;
418 
419 	if (attr->ia_valid & ATTR_SIZE) {
420 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
421 			attr->ia_valid &= ~ATTR_SIZE;
422 	}
423 
424 	/* Optimization: if the end result is no change, don't RPC */
425 	attr->ia_valid &= NFS_VALID_ATTRS;
426 	if ((attr->ia_valid & ~ATTR_FILE) == 0)
427 		return 0;
428 
429 	/* Write all dirty data */
430 	if (S_ISREG(inode->i_mode))
431 		nfs_wb_all(inode);
432 
433 	fattr = nfs_alloc_fattr();
434 	if (fattr == NULL)
435 		goto out;
436 	/*
437 	 * Return any delegations if we're going to change ACLs
438 	 */
439 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
440 		nfs_inode_return_delegation(inode);
441 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
442 	if (error == 0)
443 		nfs_refresh_inode(inode, fattr);
444 	nfs_free_fattr(fattr);
445 out:
446 	return error;
447 }
448 
449 /**
450  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
451  * @inode: inode of the file used
452  * @offset: file offset to start truncating
453  *
454  * This is a copy of the common vmtruncate, but with the locking
455  * corrected to take into account the fact that NFS requires
456  * inode->i_size to be updated under the inode->i_lock.
457  */
458 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
459 {
460 	loff_t oldsize;
461 	int err;
462 
463 	err = inode_newsize_ok(inode, offset);
464 	if (err)
465 		goto out;
466 
467 	spin_lock(&inode->i_lock);
468 	oldsize = inode->i_size;
469 	i_size_write(inode, offset);
470 	spin_unlock(&inode->i_lock);
471 
472 	truncate_pagecache(inode, oldsize, offset);
473 out:
474 	return err;
475 }
476 
477 /**
478  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
479  * @inode: pointer to struct inode
480  * @attr: pointer to struct iattr
481  *
482  * Note: we do this in the *proc.c in order to ensure that
483  *       it works for things like exclusive creates too.
484  */
485 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
486 {
487 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
488 		spin_lock(&inode->i_lock);
489 		if ((attr->ia_valid & ATTR_MODE) != 0) {
490 			int mode = attr->ia_mode & S_IALLUGO;
491 			mode |= inode->i_mode & ~S_IALLUGO;
492 			inode->i_mode = mode;
493 		}
494 		if ((attr->ia_valid & ATTR_UID) != 0)
495 			inode->i_uid = attr->ia_uid;
496 		if ((attr->ia_valid & ATTR_GID) != 0)
497 			inode->i_gid = attr->ia_gid;
498 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
499 		spin_unlock(&inode->i_lock);
500 	}
501 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
502 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
503 		nfs_vmtruncate(inode, attr->ia_size);
504 	}
505 }
506 
507 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
508 {
509 	struct inode *inode = dentry->d_inode;
510 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
511 	int err;
512 
513 	/* Flush out writes to the server in order to update c/mtime.  */
514 	if (S_ISREG(inode->i_mode)) {
515 		err = filemap_write_and_wait(inode->i_mapping);
516 		if (err)
517 			goto out;
518 	}
519 
520 	/*
521 	 * We may force a getattr if the user cares about atime.
522 	 *
523 	 * Note that we only have to check the vfsmount flags here:
524 	 *  - NFS always sets S_NOATIME by so checking it would give a
525 	 *    bogus result
526 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
527 	 *    no point in checking those.
528 	 */
529  	if ((mnt->mnt_flags & MNT_NOATIME) ||
530  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
531 		need_atime = 0;
532 
533 	if (need_atime)
534 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
535 	else
536 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
537 	if (!err) {
538 		generic_fillattr(inode, stat);
539 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
540 	}
541 out:
542 	return err;
543 }
544 
545 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
546 {
547 	atomic_set(&l_ctx->count, 1);
548 	l_ctx->lockowner = current->files;
549 	l_ctx->pid = current->tgid;
550 	INIT_LIST_HEAD(&l_ctx->list);
551 }
552 
553 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
554 {
555 	struct nfs_lock_context *pos;
556 
557 	list_for_each_entry(pos, &ctx->lock_context.list, list) {
558 		if (pos->lockowner != current->files)
559 			continue;
560 		if (pos->pid != current->tgid)
561 			continue;
562 		atomic_inc(&pos->count);
563 		return pos;
564 	}
565 	return NULL;
566 }
567 
568 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
569 {
570 	struct nfs_lock_context *res, *new = NULL;
571 	struct inode *inode = ctx->dentry->d_inode;
572 
573 	spin_lock(&inode->i_lock);
574 	res = __nfs_find_lock_context(ctx);
575 	if (res == NULL) {
576 		spin_unlock(&inode->i_lock);
577 		new = kmalloc(sizeof(*new), GFP_KERNEL);
578 		if (new == NULL)
579 			return NULL;
580 		nfs_init_lock_context(new);
581 		spin_lock(&inode->i_lock);
582 		res = __nfs_find_lock_context(ctx);
583 		if (res == NULL) {
584 			list_add_tail(&new->list, &ctx->lock_context.list);
585 			new->open_context = ctx;
586 			res = new;
587 			new = NULL;
588 		}
589 	}
590 	spin_unlock(&inode->i_lock);
591 	kfree(new);
592 	return res;
593 }
594 
595 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
596 {
597 	struct nfs_open_context *ctx = l_ctx->open_context;
598 	struct inode *inode = ctx->dentry->d_inode;
599 
600 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
601 		return;
602 	list_del(&l_ctx->list);
603 	spin_unlock(&inode->i_lock);
604 	kfree(l_ctx);
605 }
606 
607 /**
608  * nfs_close_context - Common close_context() routine NFSv2/v3
609  * @ctx: pointer to context
610  * @is_sync: is this a synchronous close
611  *
612  * always ensure that the attributes are up to date if we're mounted
613  * with close-to-open semantics
614  */
615 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
616 {
617 	struct inode *inode;
618 	struct nfs_server *server;
619 
620 	if (!(ctx->mode & FMODE_WRITE))
621 		return;
622 	if (!is_sync)
623 		return;
624 	inode = ctx->dentry->d_inode;
625 	if (!list_empty(&NFS_I(inode)->open_files))
626 		return;
627 	server = NFS_SERVER(inode);
628 	if (server->flags & NFS_MOUNT_NOCTO)
629 		return;
630 	nfs_revalidate_inode(server, inode);
631 }
632 
633 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
634 {
635 	struct nfs_open_context *ctx;
636 	struct rpc_cred *cred = rpc_lookup_cred();
637 	if (IS_ERR(cred))
638 		return ERR_CAST(cred);
639 
640 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
641 	if (!ctx) {
642 		put_rpccred(cred);
643 		return ERR_PTR(-ENOMEM);
644 	}
645 	nfs_sb_active(dentry->d_sb);
646 	ctx->dentry = dget(dentry);
647 	ctx->cred = cred;
648 	ctx->state = NULL;
649 	ctx->mode = f_mode;
650 	ctx->flags = 0;
651 	ctx->error = 0;
652 	nfs_init_lock_context(&ctx->lock_context);
653 	ctx->lock_context.open_context = ctx;
654 	INIT_LIST_HEAD(&ctx->list);
655 	return ctx;
656 }
657 
658 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
659 {
660 	if (ctx != NULL)
661 		atomic_inc(&ctx->lock_context.count);
662 	return ctx;
663 }
664 
665 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
666 {
667 	struct inode *inode = ctx->dentry->d_inode;
668 	struct super_block *sb = ctx->dentry->d_sb;
669 
670 	if (!list_empty(&ctx->list)) {
671 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
672 			return;
673 		list_del(&ctx->list);
674 		spin_unlock(&inode->i_lock);
675 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
676 		return;
677 	if (inode != NULL)
678 		NFS_PROTO(inode)->close_context(ctx, is_sync);
679 	if (ctx->cred != NULL)
680 		put_rpccred(ctx->cred);
681 	dput(ctx->dentry);
682 	nfs_sb_deactive(sb);
683 	kfree(ctx);
684 }
685 
686 void put_nfs_open_context(struct nfs_open_context *ctx)
687 {
688 	__put_nfs_open_context(ctx, 0);
689 }
690 
691 /*
692  * Ensure that mmap has a recent RPC credential for use when writing out
693  * shared pages
694  */
695 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
696 {
697 	struct inode *inode = filp->f_path.dentry->d_inode;
698 	struct nfs_inode *nfsi = NFS_I(inode);
699 
700 	filp->private_data = get_nfs_open_context(ctx);
701 	spin_lock(&inode->i_lock);
702 	list_add(&ctx->list, &nfsi->open_files);
703 	spin_unlock(&inode->i_lock);
704 }
705 
706 /*
707  * Given an inode, search for an open context with the desired characteristics
708  */
709 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
710 {
711 	struct nfs_inode *nfsi = NFS_I(inode);
712 	struct nfs_open_context *pos, *ctx = NULL;
713 
714 	spin_lock(&inode->i_lock);
715 	list_for_each_entry(pos, &nfsi->open_files, list) {
716 		if (cred != NULL && pos->cred != cred)
717 			continue;
718 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
719 			continue;
720 		ctx = get_nfs_open_context(pos);
721 		break;
722 	}
723 	spin_unlock(&inode->i_lock);
724 	return ctx;
725 }
726 
727 static void nfs_file_clear_open_context(struct file *filp)
728 {
729 	struct inode *inode = filp->f_path.dentry->d_inode;
730 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
731 
732 	if (ctx) {
733 		filp->private_data = NULL;
734 		spin_lock(&inode->i_lock);
735 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
736 		spin_unlock(&inode->i_lock);
737 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
738 	}
739 }
740 
741 /*
742  * These allocate and release file read/write context information.
743  */
744 int nfs_open(struct inode *inode, struct file *filp)
745 {
746 	struct nfs_open_context *ctx;
747 
748 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
749 	if (IS_ERR(ctx))
750 		return PTR_ERR(ctx);
751 	nfs_file_set_open_context(filp, ctx);
752 	put_nfs_open_context(ctx);
753 	nfs_fscache_set_inode_cookie(inode, filp);
754 	return 0;
755 }
756 
757 int nfs_release(struct inode *inode, struct file *filp)
758 {
759 	nfs_file_clear_open_context(filp);
760 	return 0;
761 }
762 
763 /*
764  * This function is called whenever some part of NFS notices that
765  * the cached attributes have to be refreshed.
766  */
767 int
768 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
769 {
770 	int		 status = -ESTALE;
771 	struct nfs_fattr *fattr = NULL;
772 	struct nfs_inode *nfsi = NFS_I(inode);
773 
774 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
775 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
776 
777 	if (is_bad_inode(inode))
778 		goto out;
779 	if (NFS_STALE(inode))
780 		goto out;
781 
782 	status = -ENOMEM;
783 	fattr = nfs_alloc_fattr();
784 	if (fattr == NULL)
785 		goto out;
786 
787 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
788 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
789 	if (status != 0) {
790 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
791 			 inode->i_sb->s_id,
792 			 (long long)NFS_FILEID(inode), status);
793 		if (status == -ESTALE) {
794 			nfs_zap_caches(inode);
795 			if (!S_ISDIR(inode->i_mode))
796 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
797 		}
798 		goto out;
799 	}
800 
801 	status = nfs_refresh_inode(inode, fattr);
802 	if (status) {
803 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
804 			 inode->i_sb->s_id,
805 			 (long long)NFS_FILEID(inode), status);
806 		goto out;
807 	}
808 
809 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
810 		nfs_zap_acl_cache(inode);
811 
812 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
813 		inode->i_sb->s_id,
814 		(long long)NFS_FILEID(inode));
815 
816  out:
817 	nfs_free_fattr(fattr);
818 	return status;
819 }
820 
821 int nfs_attribute_timeout(struct inode *inode)
822 {
823 	struct nfs_inode *nfsi = NFS_I(inode);
824 
825 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
826 }
827 
828 static int nfs_attribute_cache_expired(struct inode *inode)
829 {
830 	if (nfs_have_delegated_attributes(inode))
831 		return 0;
832 	return nfs_attribute_timeout(inode);
833 }
834 
835 /**
836  * nfs_revalidate_inode - Revalidate the inode attributes
837  * @server - pointer to nfs_server struct
838  * @inode - pointer to inode struct
839  *
840  * Updates inode attribute information by retrieving the data from the server.
841  */
842 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
843 {
844 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
845 			&& !nfs_attribute_cache_expired(inode))
846 		return NFS_STALE(inode) ? -ESTALE : 0;
847 	return __nfs_revalidate_inode(server, inode);
848 }
849 
850 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
851 {
852 	struct nfs_inode *nfsi = NFS_I(inode);
853 
854 	if (mapping->nrpages != 0) {
855 		int ret = invalidate_inode_pages2(mapping);
856 		if (ret < 0)
857 			return ret;
858 	}
859 	spin_lock(&inode->i_lock);
860 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
861 	if (S_ISDIR(inode->i_mode))
862 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
863 	spin_unlock(&inode->i_lock);
864 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
865 	nfs_fscache_reset_inode_cookie(inode);
866 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
867 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
868 	return 0;
869 }
870 
871 /**
872  * nfs_revalidate_mapping - Revalidate the pagecache
873  * @inode - pointer to host inode
874  * @mapping - pointer to mapping
875  */
876 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
877 {
878 	struct nfs_inode *nfsi = NFS_I(inode);
879 	int ret = 0;
880 
881 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
882 			|| nfs_attribute_cache_expired(inode)
883 			|| NFS_STALE(inode)) {
884 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
885 		if (ret < 0)
886 			goto out;
887 	}
888 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
889 		ret = nfs_invalidate_mapping(inode, mapping);
890 out:
891 	return ret;
892 }
893 
894 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
895 {
896 	struct nfs_inode *nfsi = NFS_I(inode);
897 	unsigned long ret = 0;
898 
899 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
900 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
901 			&& inode->i_version == fattr->pre_change_attr) {
902 		inode->i_version = fattr->change_attr;
903 		if (S_ISDIR(inode->i_mode))
904 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
905 		ret |= NFS_INO_INVALID_ATTR;
906 	}
907 	/* If we have atomic WCC data, we may update some attributes */
908 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
909 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
910 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
911 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
912 		ret |= NFS_INO_INVALID_ATTR;
913 	}
914 
915 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
916 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
917 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
918 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
919 		if (S_ISDIR(inode->i_mode))
920 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
921 		ret |= NFS_INO_INVALID_ATTR;
922 	}
923 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
924 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
925 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
926 			&& nfsi->npages == 0) {
927 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
928 		ret |= NFS_INO_INVALID_ATTR;
929 	}
930 	return ret;
931 }
932 
933 /**
934  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
935  * @inode - pointer to inode
936  * @fattr - updated attributes
937  *
938  * Verifies the attribute cache. If we have just changed the attributes,
939  * so that fattr carries weak cache consistency data, then it may
940  * also update the ctime/mtime/change_attribute.
941  */
942 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
943 {
944 	struct nfs_inode *nfsi = NFS_I(inode);
945 	loff_t cur_size, new_isize;
946 	unsigned long invalid = 0;
947 
948 
949 	/* Has the inode gone and changed behind our back? */
950 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
951 		return -EIO;
952 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
953 		return -EIO;
954 
955 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
956 			inode->i_version != fattr->change_attr)
957 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
958 
959 	/* Verify a few of the more important attributes */
960 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
961 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
962 
963 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
964 		cur_size = i_size_read(inode);
965 		new_isize = nfs_size_to_loff_t(fattr->size);
966 		if (cur_size != new_isize && nfsi->npages == 0)
967 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
968 	}
969 
970 	/* Have any file permissions changed? */
971 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
972 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
973 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
974 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
975 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
976 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
977 
978 	/* Has the link count changed? */
979 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
980 		invalid |= NFS_INO_INVALID_ATTR;
981 
982 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
983 		invalid |= NFS_INO_INVALID_ATIME;
984 
985 	if (invalid != 0)
986 		nfsi->cache_validity |= invalid;
987 
988 	nfsi->read_cache_jiffies = fattr->time_start;
989 	return 0;
990 }
991 
992 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
993 {
994 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
995 		return 0;
996 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
997 }
998 
999 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1000 {
1001 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1002 		return 0;
1003 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1004 }
1005 
1006 static atomic_long_t nfs_attr_generation_counter;
1007 
1008 static unsigned long nfs_read_attr_generation_counter(void)
1009 {
1010 	return atomic_long_read(&nfs_attr_generation_counter);
1011 }
1012 
1013 unsigned long nfs_inc_attr_generation_counter(void)
1014 {
1015 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1016 }
1017 
1018 void nfs_fattr_init(struct nfs_fattr *fattr)
1019 {
1020 	fattr->valid = 0;
1021 	fattr->time_start = jiffies;
1022 	fattr->gencount = nfs_inc_attr_generation_counter();
1023 	fattr->owner_name = NULL;
1024 	fattr->group_name = NULL;
1025 }
1026 
1027 struct nfs_fattr *nfs_alloc_fattr(void)
1028 {
1029 	struct nfs_fattr *fattr;
1030 
1031 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1032 	if (fattr != NULL)
1033 		nfs_fattr_init(fattr);
1034 	return fattr;
1035 }
1036 
1037 struct nfs_fh *nfs_alloc_fhandle(void)
1038 {
1039 	struct nfs_fh *fh;
1040 
1041 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1042 	if (fh != NULL)
1043 		fh->size = 0;
1044 	return fh;
1045 }
1046 
1047 /**
1048  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1049  * @inode - pointer to inode
1050  * @fattr - attributes
1051  *
1052  * Attempt to divine whether or not an RPC call reply carrying stale
1053  * attributes got scheduled after another call carrying updated ones.
1054  *
1055  * To do so, the function first assumes that a more recent ctime means
1056  * that the attributes in fattr are newer, however it also attempt to
1057  * catch the case where ctime either didn't change, or went backwards
1058  * (if someone reset the clock on the server) by looking at whether
1059  * or not this RPC call was started after the inode was last updated.
1060  * Note also the check for wraparound of 'attr_gencount'
1061  *
1062  * The function returns 'true' if it thinks the attributes in 'fattr' are
1063  * more recent than the ones cached in the inode.
1064  *
1065  */
1066 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1067 {
1068 	const struct nfs_inode *nfsi = NFS_I(inode);
1069 
1070 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1071 		nfs_ctime_need_update(inode, fattr) ||
1072 		nfs_size_need_update(inode, fattr) ||
1073 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1074 }
1075 
1076 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1077 {
1078 	if (nfs_inode_attrs_need_update(inode, fattr))
1079 		return nfs_update_inode(inode, fattr);
1080 	return nfs_check_inode_attributes(inode, fattr);
1081 }
1082 
1083 /**
1084  * nfs_refresh_inode - try to update the inode attribute cache
1085  * @inode - pointer to inode
1086  * @fattr - updated attributes
1087  *
1088  * Check that an RPC call that returned attributes has not overlapped with
1089  * other recent updates of the inode metadata, then decide whether it is
1090  * safe to do a full update of the inode attributes, or whether just to
1091  * call nfs_check_inode_attributes.
1092  */
1093 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1094 {
1095 	int status;
1096 
1097 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1098 		return 0;
1099 	spin_lock(&inode->i_lock);
1100 	status = nfs_refresh_inode_locked(inode, fattr);
1101 	spin_unlock(&inode->i_lock);
1102 
1103 	return status;
1104 }
1105 
1106 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1107 {
1108 	struct nfs_inode *nfsi = NFS_I(inode);
1109 
1110 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1111 	if (S_ISDIR(inode->i_mode))
1112 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1113 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1114 		return 0;
1115 	return nfs_refresh_inode_locked(inode, fattr);
1116 }
1117 
1118 /**
1119  * nfs_post_op_update_inode - try to update the inode attribute cache
1120  * @inode - pointer to inode
1121  * @fattr - updated attributes
1122  *
1123  * After an operation that has changed the inode metadata, mark the
1124  * attribute cache as being invalid, then try to update it.
1125  *
1126  * NB: if the server didn't return any post op attributes, this
1127  * function will force the retrieval of attributes before the next
1128  * NFS request.  Thus it should be used only for operations that
1129  * are expected to change one or more attributes, to avoid
1130  * unnecessary NFS requests and trips through nfs_update_inode().
1131  */
1132 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1133 {
1134 	int status;
1135 
1136 	spin_lock(&inode->i_lock);
1137 	status = nfs_post_op_update_inode_locked(inode, fattr);
1138 	spin_unlock(&inode->i_lock);
1139 	return status;
1140 }
1141 
1142 /**
1143  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1144  * @inode - pointer to inode
1145  * @fattr - updated attributes
1146  *
1147  * After an operation that has changed the inode metadata, mark the
1148  * attribute cache as being invalid, then try to update it. Fake up
1149  * weak cache consistency data, if none exist.
1150  *
1151  * This function is mainly designed to be used by the ->write_done() functions.
1152  */
1153 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1154 {
1155 	int status;
1156 
1157 	spin_lock(&inode->i_lock);
1158 	/* Don't do a WCC update if these attributes are already stale */
1159 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1160 			!nfs_inode_attrs_need_update(inode, fattr)) {
1161 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1162 				| NFS_ATTR_FATTR_PRESIZE
1163 				| NFS_ATTR_FATTR_PREMTIME
1164 				| NFS_ATTR_FATTR_PRECTIME);
1165 		goto out_noforce;
1166 	}
1167 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1168 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1169 		fattr->pre_change_attr = inode->i_version;
1170 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1171 	}
1172 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1173 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1174 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1175 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1176 	}
1177 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1178 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1179 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1180 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1181 	}
1182 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1183 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1184 		fattr->pre_size = i_size_read(inode);
1185 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1186 	}
1187 out_noforce:
1188 	status = nfs_post_op_update_inode_locked(inode, fattr);
1189 	spin_unlock(&inode->i_lock);
1190 	return status;
1191 }
1192 
1193 /*
1194  * Many nfs protocol calls return the new file attributes after
1195  * an operation.  Here we update the inode to reflect the state
1196  * of the server's inode.
1197  *
1198  * This is a bit tricky because we have to make sure all dirty pages
1199  * have been sent off to the server before calling invalidate_inode_pages.
1200  * To make sure no other process adds more write requests while we try
1201  * our best to flush them, we make them sleep during the attribute refresh.
1202  *
1203  * A very similar scenario holds for the dir cache.
1204  */
1205 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1206 {
1207 	struct nfs_server *server;
1208 	struct nfs_inode *nfsi = NFS_I(inode);
1209 	loff_t cur_isize, new_isize;
1210 	unsigned long invalid = 0;
1211 	unsigned long now = jiffies;
1212 	unsigned long save_cache_validity;
1213 
1214 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1215 			__func__, inode->i_sb->s_id, inode->i_ino,
1216 			atomic_read(&inode->i_count), fattr->valid);
1217 
1218 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1219 		goto out_fileid;
1220 
1221 	/*
1222 	 * Make sure the inode's type hasn't changed.
1223 	 */
1224 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1225 		goto out_changed;
1226 
1227 	server = NFS_SERVER(inode);
1228 	/* Update the fsid? */
1229 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1230 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1231 			!IS_AUTOMOUNT(inode))
1232 		server->fsid = fattr->fsid;
1233 
1234 	/*
1235 	 * Update the read time so we don't revalidate too often.
1236 	 */
1237 	nfsi->read_cache_jiffies = fattr->time_start;
1238 
1239 	save_cache_validity = nfsi->cache_validity;
1240 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1241 			| NFS_INO_INVALID_ATIME
1242 			| NFS_INO_REVAL_FORCED
1243 			| NFS_INO_REVAL_PAGECACHE);
1244 
1245 	/* Do atomic weak cache consistency updates */
1246 	invalid |= nfs_wcc_update_inode(inode, fattr);
1247 
1248 	/* More cache consistency checks */
1249 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1250 		if (inode->i_version != fattr->change_attr) {
1251 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1252 					inode->i_sb->s_id, inode->i_ino);
1253 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1254 			if (S_ISDIR(inode->i_mode))
1255 				nfs_force_lookup_revalidate(inode);
1256 			inode->i_version = fattr->change_attr;
1257 		}
1258 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1259 		invalid |= save_cache_validity;
1260 
1261 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1262 		/* NFSv2/v3: Check if the mtime agrees */
1263 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1264 			dprintk("NFS: mtime change on server for file %s/%ld\n",
1265 					inode->i_sb->s_id, inode->i_ino);
1266 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1267 			if (S_ISDIR(inode->i_mode))
1268 				nfs_force_lookup_revalidate(inode);
1269 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1270 		}
1271 	} else if (server->caps & NFS_CAP_MTIME)
1272 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1273 				| NFS_INO_INVALID_DATA
1274 				| NFS_INO_REVAL_PAGECACHE
1275 				| NFS_INO_REVAL_FORCED);
1276 
1277 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1278 		/* If ctime has changed we should definitely clear access+acl caches */
1279 		if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1280 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1281 			/* and probably clear data for a directory too as utimes can cause
1282 			 * havoc with our cache.
1283 			 */
1284 			if (S_ISDIR(inode->i_mode)) {
1285 				invalid |= NFS_INO_INVALID_DATA;
1286 				nfs_force_lookup_revalidate(inode);
1287 			}
1288 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1289 		}
1290 	} else if (server->caps & NFS_CAP_CTIME)
1291 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1292 				| NFS_INO_INVALID_ACCESS
1293 				| NFS_INO_INVALID_ACL
1294 				| NFS_INO_REVAL_FORCED);
1295 
1296 	/* Check if our cached file size is stale */
1297 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1298 		new_isize = nfs_size_to_loff_t(fattr->size);
1299 		cur_isize = i_size_read(inode);
1300 		if (new_isize != cur_isize) {
1301 			/* Do we perhaps have any outstanding writes, or has
1302 			 * the file grown beyond our last write? */
1303 			if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1304 			     new_isize > cur_isize) {
1305 				i_size_write(inode, new_isize);
1306 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1307 			}
1308 			dprintk("NFS: isize change on server for file %s/%ld "
1309 					"(%Ld to %Ld)\n",
1310 					inode->i_sb->s_id,
1311 					inode->i_ino,
1312 					(long long)cur_isize,
1313 					(long long)new_isize);
1314 		}
1315 	} else
1316 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1317 				| NFS_INO_REVAL_PAGECACHE
1318 				| NFS_INO_REVAL_FORCED);
1319 
1320 
1321 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1322 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1323 	else if (server->caps & NFS_CAP_ATIME)
1324 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1325 				| NFS_INO_REVAL_FORCED);
1326 
1327 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1328 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1329 			umode_t newmode = inode->i_mode & S_IFMT;
1330 			newmode |= fattr->mode & S_IALLUGO;
1331 			inode->i_mode = newmode;
1332 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1333 		}
1334 	} else if (server->caps & NFS_CAP_MODE)
1335 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1336 				| NFS_INO_INVALID_ACCESS
1337 				| NFS_INO_INVALID_ACL
1338 				| NFS_INO_REVAL_FORCED);
1339 
1340 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1341 		if (inode->i_uid != fattr->uid) {
1342 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1343 			inode->i_uid = fattr->uid;
1344 		}
1345 	} else if (server->caps & NFS_CAP_OWNER)
1346 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1347 				| NFS_INO_INVALID_ACCESS
1348 				| NFS_INO_INVALID_ACL
1349 				| NFS_INO_REVAL_FORCED);
1350 
1351 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1352 		if (inode->i_gid != fattr->gid) {
1353 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1354 			inode->i_gid = fattr->gid;
1355 		}
1356 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1357 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1358 				| NFS_INO_INVALID_ACCESS
1359 				| NFS_INO_INVALID_ACL
1360 				| NFS_INO_REVAL_FORCED);
1361 
1362 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1363 		if (inode->i_nlink != fattr->nlink) {
1364 			invalid |= NFS_INO_INVALID_ATTR;
1365 			if (S_ISDIR(inode->i_mode))
1366 				invalid |= NFS_INO_INVALID_DATA;
1367 			set_nlink(inode, fattr->nlink);
1368 		}
1369 	} else if (server->caps & NFS_CAP_NLINK)
1370 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1371 				| NFS_INO_REVAL_FORCED);
1372 
1373 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1374 		/*
1375 		 * report the blocks in 512byte units
1376 		 */
1377 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1378  	}
1379 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1380 		inode->i_blocks = fattr->du.nfs2.blocks;
1381 
1382 	/* Update attrtimeo value if we're out of the unstable period */
1383 	if (invalid & NFS_INO_INVALID_ATTR) {
1384 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1385 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1386 		nfsi->attrtimeo_timestamp = now;
1387 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1388 	} else {
1389 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1390 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1391 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1392 			nfsi->attrtimeo_timestamp = now;
1393 		}
1394 	}
1395 	invalid &= ~NFS_INO_INVALID_ATTR;
1396 	/* Don't invalidate the data if we were to blame */
1397 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1398 				|| S_ISLNK(inode->i_mode)))
1399 		invalid &= ~NFS_INO_INVALID_DATA;
1400 	if (!nfs_have_delegation(inode, FMODE_READ) ||
1401 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1402 		nfsi->cache_validity |= invalid;
1403 
1404 	return 0;
1405  out_changed:
1406 	/*
1407 	 * Big trouble! The inode has become a different object.
1408 	 */
1409 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1410 			__func__, inode->i_ino, inode->i_mode, fattr->mode);
1411  out_err:
1412 	/*
1413 	 * No need to worry about unhashing the dentry, as the
1414 	 * lookup validation will know that the inode is bad.
1415 	 * (But we fall through to invalidate the caches.)
1416 	 */
1417 	nfs_invalidate_inode(inode);
1418 	return -ESTALE;
1419 
1420  out_fileid:
1421 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1422 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1423 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1424 		(long long)nfsi->fileid, (long long)fattr->fileid);
1425 	goto out_err;
1426 }
1427 
1428 
1429 #ifdef CONFIG_NFS_V4
1430 
1431 /*
1432  * Clean out any remaining NFSv4 state that might be left over due
1433  * to open() calls that passed nfs_atomic_lookup, but failed to call
1434  * nfs_open().
1435  */
1436 void nfs4_evict_inode(struct inode *inode)
1437 {
1438 	truncate_inode_pages(&inode->i_data, 0);
1439 	end_writeback(inode);
1440 	pnfs_return_layout(inode);
1441 	pnfs_destroy_layout(NFS_I(inode));
1442 	/* If we are holding a delegation, return it! */
1443 	nfs_inode_return_delegation_noreclaim(inode);
1444 	/* First call standard NFS clear_inode() code */
1445 	nfs_clear_inode(inode);
1446 }
1447 #endif
1448 
1449 struct inode *nfs_alloc_inode(struct super_block *sb)
1450 {
1451 	struct nfs_inode *nfsi;
1452 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1453 	if (!nfsi)
1454 		return NULL;
1455 	nfsi->flags = 0UL;
1456 	nfsi->cache_validity = 0UL;
1457 #ifdef CONFIG_NFS_V3_ACL
1458 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1459 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1460 #endif
1461 #ifdef CONFIG_NFS_V4
1462 	nfsi->nfs4_acl = NULL;
1463 #endif /* CONFIG_NFS_V4 */
1464 	return &nfsi->vfs_inode;
1465 }
1466 
1467 static void nfs_i_callback(struct rcu_head *head)
1468 {
1469 	struct inode *inode = container_of(head, struct inode, i_rcu);
1470 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1471 }
1472 
1473 void nfs_destroy_inode(struct inode *inode)
1474 {
1475 	call_rcu(&inode->i_rcu, nfs_i_callback);
1476 }
1477 
1478 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1479 {
1480 #ifdef CONFIG_NFS_V4
1481 	INIT_LIST_HEAD(&nfsi->open_states);
1482 	nfsi->delegation = NULL;
1483 	nfsi->delegation_state = 0;
1484 	init_rwsem(&nfsi->rwsem);
1485 	nfsi->layout = NULL;
1486 	atomic_set(&nfsi->commits_outstanding, 0);
1487 #endif
1488 }
1489 
1490 static void init_once(void *foo)
1491 {
1492 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1493 
1494 	inode_init_once(&nfsi->vfs_inode);
1495 	INIT_LIST_HEAD(&nfsi->open_files);
1496 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1497 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1498 	INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1499 	nfsi->npages = 0;
1500 	nfsi->ncommit = 0;
1501 	atomic_set(&nfsi->silly_count, 1);
1502 	INIT_HLIST_HEAD(&nfsi->silly_list);
1503 	init_waitqueue_head(&nfsi->waitqueue);
1504 	nfs4_init_once(nfsi);
1505 }
1506 
1507 static int __init nfs_init_inodecache(void)
1508 {
1509 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1510 					     sizeof(struct nfs_inode),
1511 					     0, (SLAB_RECLAIM_ACCOUNT|
1512 						SLAB_MEM_SPREAD),
1513 					     init_once);
1514 	if (nfs_inode_cachep == NULL)
1515 		return -ENOMEM;
1516 
1517 	return 0;
1518 }
1519 
1520 static void nfs_destroy_inodecache(void)
1521 {
1522 	kmem_cache_destroy(nfs_inode_cachep);
1523 }
1524 
1525 struct workqueue_struct *nfsiod_workqueue;
1526 
1527 /*
1528  * start up the nfsiod workqueue
1529  */
1530 static int nfsiod_start(void)
1531 {
1532 	struct workqueue_struct *wq;
1533 	dprintk("RPC:       creating workqueue nfsiod\n");
1534 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1535 	if (wq == NULL)
1536 		return -ENOMEM;
1537 	nfsiod_workqueue = wq;
1538 	return 0;
1539 }
1540 
1541 /*
1542  * Destroy the nfsiod workqueue
1543  */
1544 static void nfsiod_stop(void)
1545 {
1546 	struct workqueue_struct *wq;
1547 
1548 	wq = nfsiod_workqueue;
1549 	if (wq == NULL)
1550 		return;
1551 	nfsiod_workqueue = NULL;
1552 	destroy_workqueue(wq);
1553 }
1554 
1555 /*
1556  * Initialize NFS
1557  */
1558 static int __init init_nfs_fs(void)
1559 {
1560 	int err;
1561 
1562 	err = nfs_idmap_init();
1563 	if (err < 0)
1564 		goto out9;
1565 
1566 	err = nfs_dns_resolver_init();
1567 	if (err < 0)
1568 		goto out8;
1569 
1570 	err = nfs_fscache_register();
1571 	if (err < 0)
1572 		goto out7;
1573 
1574 	err = nfsiod_start();
1575 	if (err)
1576 		goto out6;
1577 
1578 	err = nfs_fs_proc_init();
1579 	if (err)
1580 		goto out5;
1581 
1582 	err = nfs_init_nfspagecache();
1583 	if (err)
1584 		goto out4;
1585 
1586 	err = nfs_init_inodecache();
1587 	if (err)
1588 		goto out3;
1589 
1590 	err = nfs_init_readpagecache();
1591 	if (err)
1592 		goto out2;
1593 
1594 	err = nfs_init_writepagecache();
1595 	if (err)
1596 		goto out1;
1597 
1598 	err = nfs_init_directcache();
1599 	if (err)
1600 		goto out0;
1601 
1602 #ifdef CONFIG_PROC_FS
1603 	rpc_proc_register(&nfs_rpcstat);
1604 #endif
1605 	if ((err = register_nfs_fs()) != 0)
1606 		goto out;
1607 	return 0;
1608 out:
1609 #ifdef CONFIG_PROC_FS
1610 	rpc_proc_unregister("nfs");
1611 #endif
1612 	nfs_destroy_directcache();
1613 out0:
1614 	nfs_destroy_writepagecache();
1615 out1:
1616 	nfs_destroy_readpagecache();
1617 out2:
1618 	nfs_destroy_inodecache();
1619 out3:
1620 	nfs_destroy_nfspagecache();
1621 out4:
1622 	nfs_fs_proc_exit();
1623 out5:
1624 	nfsiod_stop();
1625 out6:
1626 	nfs_fscache_unregister();
1627 out7:
1628 	nfs_dns_resolver_destroy();
1629 out8:
1630 	nfs_idmap_quit();
1631 out9:
1632 	return err;
1633 }
1634 
1635 static void __exit exit_nfs_fs(void)
1636 {
1637 	nfs_destroy_directcache();
1638 	nfs_destroy_writepagecache();
1639 	nfs_destroy_readpagecache();
1640 	nfs_destroy_inodecache();
1641 	nfs_destroy_nfspagecache();
1642 	nfs_fscache_unregister();
1643 	nfs_dns_resolver_destroy();
1644 	nfs_idmap_quit();
1645 #ifdef CONFIG_PROC_FS
1646 	rpc_proc_unregister("nfs");
1647 #endif
1648 	nfs_cleanup_cb_ident_idr();
1649 	unregister_nfs_fs();
1650 	nfs_fs_proc_exit();
1651 	nfsiod_stop();
1652 }
1653 
1654 /* Not quite true; I just maintain it */
1655 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1656 MODULE_LICENSE("GPL");
1657 module_param(enable_ino64, bool, 0644);
1658 
1659 module_init(init_nfs_fs)
1660 module_exit(exit_nfs_fs)
1661