xref: /openbmc/linux/fs/nfs/inode.c (revision 23c2b932)
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/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
41 
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 "pnfs.h"
51 #include "nfs.h"
52 #include "netns.h"
53 
54 #include "nfstrace.h"
55 
56 #define NFSDBG_FACILITY		NFSDBG_VFS
57 
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
59 
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62 
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65 
66 static struct kmem_cache * nfs_inode_cachep;
67 
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 	return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73 
74 static int nfs_wait_killable(int mode)
75 {
76 	freezable_schedule_unsafe();
77 	if (signal_pending_state(mode, current))
78 		return -ERESTARTSYS;
79 	return 0;
80 }
81 
82 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
83 {
84 	return nfs_wait_killable(mode);
85 }
86 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
87 
88 int nfs_wait_atomic_killable(atomic_t *p)
89 {
90 	return nfs_wait_killable(TASK_KILLABLE);
91 }
92 
93 /**
94  * nfs_compat_user_ino64 - returns the user-visible inode number
95  * @fileid: 64-bit fileid
96  *
97  * This function returns a 32-bit inode number if the boot parameter
98  * nfs.enable_ino64 is zero.
99  */
100 u64 nfs_compat_user_ino64(u64 fileid)
101 {
102 #ifdef CONFIG_COMPAT
103 	compat_ulong_t ino;
104 #else
105 	unsigned long ino;
106 #endif
107 
108 	if (enable_ino64)
109 		return fileid;
110 	ino = fileid;
111 	if (sizeof(ino) < sizeof(fileid))
112 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
113 	return ino;
114 }
115 
116 int nfs_drop_inode(struct inode *inode)
117 {
118 	return NFS_STALE(inode) || generic_drop_inode(inode);
119 }
120 EXPORT_SYMBOL_GPL(nfs_drop_inode);
121 
122 void nfs_clear_inode(struct inode *inode)
123 {
124 	/*
125 	 * The following should never happen...
126 	 */
127 	WARN_ON_ONCE(nfs_have_writebacks(inode));
128 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
129 	nfs_zap_acl_cache(inode);
130 	nfs_access_zap_cache(inode);
131 	nfs_fscache_clear_inode(inode);
132 }
133 EXPORT_SYMBOL_GPL(nfs_clear_inode);
134 
135 void nfs_evict_inode(struct inode *inode)
136 {
137 	truncate_inode_pages_final(&inode->i_data);
138 	clear_inode(inode);
139 	nfs_clear_inode(inode);
140 }
141 
142 int nfs_sync_inode(struct inode *inode)
143 {
144 	inode_dio_wait(inode);
145 	return nfs_wb_all(inode);
146 }
147 EXPORT_SYMBOL_GPL(nfs_sync_inode);
148 
149 /**
150  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
151  */
152 int nfs_sync_mapping(struct address_space *mapping)
153 {
154 	int ret = 0;
155 
156 	if (mapping->nrpages != 0) {
157 		unmap_mapping_range(mapping, 0, 0, 0);
158 		ret = nfs_wb_all(mapping->host);
159 	}
160 	return ret;
161 }
162 
163 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
164 {
165 	struct nfs_inode *nfsi = NFS_I(inode);
166 
167 	if (inode->i_mapping->nrpages == 0)
168 		flags &= ~NFS_INO_INVALID_DATA;
169 	nfsi->cache_validity |= flags;
170 	if (flags & NFS_INO_INVALID_DATA)
171 		nfs_fscache_invalidate(inode);
172 }
173 
174 /*
175  * Invalidate the local caches
176  */
177 static void nfs_zap_caches_locked(struct inode *inode)
178 {
179 	struct nfs_inode *nfsi = NFS_I(inode);
180 	int mode = inode->i_mode;
181 
182 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
183 
184 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
185 	nfsi->attrtimeo_timestamp = jiffies;
186 
187 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
188 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
189 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
190 					| NFS_INO_INVALID_DATA
191 					| NFS_INO_INVALID_ACCESS
192 					| NFS_INO_INVALID_ACL
193 					| NFS_INO_REVAL_PAGECACHE);
194 	} else
195 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
196 					| NFS_INO_INVALID_ACCESS
197 					| NFS_INO_INVALID_ACL
198 					| NFS_INO_REVAL_PAGECACHE);
199 	nfs_zap_label_cache_locked(nfsi);
200 }
201 
202 void nfs_zap_caches(struct inode *inode)
203 {
204 	spin_lock(&inode->i_lock);
205 	nfs_zap_caches_locked(inode);
206 	spin_unlock(&inode->i_lock);
207 }
208 
209 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
210 {
211 	if (mapping->nrpages != 0) {
212 		spin_lock(&inode->i_lock);
213 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
214 		spin_unlock(&inode->i_lock);
215 	}
216 }
217 
218 void nfs_zap_acl_cache(struct inode *inode)
219 {
220 	void (*clear_acl_cache)(struct inode *);
221 
222 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
223 	if (clear_acl_cache != NULL)
224 		clear_acl_cache(inode);
225 	spin_lock(&inode->i_lock);
226 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
227 	spin_unlock(&inode->i_lock);
228 }
229 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
230 
231 void nfs_invalidate_atime(struct inode *inode)
232 {
233 	spin_lock(&inode->i_lock);
234 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
235 	spin_unlock(&inode->i_lock);
236 }
237 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
238 
239 /*
240  * Invalidate, but do not unhash, the inode.
241  * NB: must be called with inode->i_lock held!
242  */
243 static void nfs_invalidate_inode(struct inode *inode)
244 {
245 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
246 	nfs_zap_caches_locked(inode);
247 }
248 
249 struct nfs_find_desc {
250 	struct nfs_fh		*fh;
251 	struct nfs_fattr	*fattr;
252 };
253 
254 /*
255  * In NFSv3 we can have 64bit inode numbers. In order to support
256  * this, and re-exported directories (also seen in NFSv2)
257  * we are forced to allow 2 different inodes to have the same
258  * i_ino.
259  */
260 static int
261 nfs_find_actor(struct inode *inode, void *opaque)
262 {
263 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
264 	struct nfs_fh		*fh = desc->fh;
265 	struct nfs_fattr	*fattr = desc->fattr;
266 
267 	if (NFS_FILEID(inode) != fattr->fileid)
268 		return 0;
269 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
270 		return 0;
271 	if (nfs_compare_fh(NFS_FH(inode), fh))
272 		return 0;
273 	if (is_bad_inode(inode) || NFS_STALE(inode))
274 		return 0;
275 	return 1;
276 }
277 
278 static int
279 nfs_init_locked(struct inode *inode, void *opaque)
280 {
281 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
282 	struct nfs_fattr	*fattr = desc->fattr;
283 
284 	set_nfs_fileid(inode, fattr->fileid);
285 	inode->i_mode = fattr->mode;
286 	nfs_copy_fh(NFS_FH(inode), desc->fh);
287 	return 0;
288 }
289 
290 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
291 static void nfs_clear_label_invalid(struct inode *inode)
292 {
293 	spin_lock(&inode->i_lock);
294 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
295 	spin_unlock(&inode->i_lock);
296 }
297 
298 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
299 					struct nfs4_label *label)
300 {
301 	int error;
302 
303 	if (label == NULL)
304 		return;
305 
306 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
307 		error = security_inode_notifysecctx(inode, label->label,
308 				label->len);
309 		if (error)
310 			printk(KERN_ERR "%s() %s %d "
311 					"security_inode_notifysecctx() %d\n",
312 					__func__,
313 					(char *)label->label,
314 					label->len, error);
315 		nfs_clear_label_invalid(inode);
316 	}
317 }
318 
319 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
320 {
321 	struct nfs4_label *label = NULL;
322 	int minor_version = server->nfs_client->cl_minorversion;
323 
324 	if (minor_version < 2)
325 		return label;
326 
327 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
328 		return label;
329 
330 	label = kzalloc(sizeof(struct nfs4_label), flags);
331 	if (label == NULL)
332 		return ERR_PTR(-ENOMEM);
333 
334 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
335 	if (label->label == NULL) {
336 		kfree(label);
337 		return ERR_PTR(-ENOMEM);
338 	}
339 	label->len = NFS4_MAXLABELLEN;
340 
341 	return label;
342 }
343 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
344 #else
345 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
346 					struct nfs4_label *label)
347 {
348 }
349 #endif
350 EXPORT_SYMBOL_GPL(nfs_setsecurity);
351 
352 /*
353  * This is our front-end to iget that looks up inodes by file handle
354  * instead of inode number.
355  */
356 struct inode *
357 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
358 {
359 	struct nfs_find_desc desc = {
360 		.fh	= fh,
361 		.fattr	= fattr
362 	};
363 	struct inode *inode = ERR_PTR(-ENOENT);
364 	unsigned long hash;
365 
366 	nfs_attr_check_mountpoint(sb, fattr);
367 
368 	if (nfs_attr_use_mounted_on_fileid(fattr))
369 		fattr->fileid = fattr->mounted_on_fileid;
370 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
371 		goto out_no_inode;
372 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
373 		goto out_no_inode;
374 
375 	hash = nfs_fattr_to_ino_t(fattr);
376 
377 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
378 	if (inode == NULL) {
379 		inode = ERR_PTR(-ENOMEM);
380 		goto out_no_inode;
381 	}
382 
383 	if (inode->i_state & I_NEW) {
384 		struct nfs_inode *nfsi = NFS_I(inode);
385 		unsigned long now = jiffies;
386 
387 		/* We set i_ino for the few things that still rely on it,
388 		 * such as stat(2) */
389 		inode->i_ino = hash;
390 
391 		/* We can't support update_atime(), since the server will reset it */
392 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
393 		inode->i_mode = fattr->mode;
394 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
395 				&& nfs_server_capable(inode, NFS_CAP_MODE))
396 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
397 		/* Why so? Because we want revalidate for devices/FIFOs, and
398 		 * that's precisely what we have in nfs_file_inode_operations.
399 		 */
400 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
401 		if (S_ISREG(inode->i_mode)) {
402 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
403 			inode->i_data.a_ops = &nfs_file_aops;
404 		} else if (S_ISDIR(inode->i_mode)) {
405 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
406 			inode->i_fop = &nfs_dir_operations;
407 			inode->i_data.a_ops = &nfs_dir_aops;
408 			/* Deal with crossing mountpoints */
409 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
410 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
411 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
412 					inode->i_op = &nfs_referral_inode_operations;
413 				else
414 					inode->i_op = &nfs_mountpoint_inode_operations;
415 				inode->i_fop = NULL;
416 				inode->i_flags |= S_AUTOMOUNT;
417 			}
418 		} else if (S_ISLNK(inode->i_mode)) {
419 			inode->i_op = &nfs_symlink_inode_operations;
420 			inode_nohighmem(inode);
421 		} else
422 			init_special_inode(inode, inode->i_mode, fattr->rdev);
423 
424 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
425 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
426 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
427 		inode->i_version = 0;
428 		inode->i_size = 0;
429 		clear_nlink(inode);
430 		inode->i_uid = make_kuid(&init_user_ns, -2);
431 		inode->i_gid = make_kgid(&init_user_ns, -2);
432 		inode->i_blocks = 0;
433 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
434 		nfsi->write_io = 0;
435 		nfsi->read_io = 0;
436 
437 		nfsi->read_cache_jiffies = fattr->time_start;
438 		nfsi->attr_gencount = fattr->gencount;
439 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
440 			inode->i_atime = fattr->atime;
441 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
442 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
443 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
444 			inode->i_mtime = fattr->mtime;
445 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
446 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
447 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
448 			inode->i_ctime = fattr->ctime;
449 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
450 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
451 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
452 			inode->i_version = fattr->change_attr;
453 		else
454 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
455 				| NFS_INO_REVAL_PAGECACHE);
456 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
457 			inode->i_size = nfs_size_to_loff_t(fattr->size);
458 		else
459 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
460 				| NFS_INO_REVAL_PAGECACHE);
461 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
462 			set_nlink(inode, fattr->nlink);
463 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
464 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
465 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
466 			inode->i_uid = fattr->uid;
467 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
468 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
469 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
470 			inode->i_gid = fattr->gid;
471 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
472 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
473 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
474 			inode->i_blocks = fattr->du.nfs2.blocks;
475 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
476 			/*
477 			 * report the blocks in 512byte units
478 			 */
479 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
480 		}
481 
482 		nfs_setsecurity(inode, fattr, label);
483 
484 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
485 		nfsi->attrtimeo_timestamp = now;
486 		nfsi->access_cache = RB_ROOT;
487 
488 		nfs_fscache_init_inode(inode);
489 
490 		unlock_new_inode(inode);
491 	} else
492 		nfs_refresh_inode(inode, fattr);
493 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
494 		inode->i_sb->s_id,
495 		(unsigned long long)NFS_FILEID(inode),
496 		nfs_display_fhandle_hash(fh),
497 		atomic_read(&inode->i_count));
498 
499 out:
500 	return inode;
501 
502 out_no_inode:
503 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
504 	goto out;
505 }
506 EXPORT_SYMBOL_GPL(nfs_fhget);
507 
508 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
509 
510 int
511 nfs_setattr(struct dentry *dentry, struct iattr *attr)
512 {
513 	struct inode *inode = d_inode(dentry);
514 	struct nfs_fattr *fattr;
515 	int error = 0;
516 
517 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
518 
519 	/* skip mode change if it's just for clearing setuid/setgid */
520 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
521 		attr->ia_valid &= ~ATTR_MODE;
522 
523 	if (attr->ia_valid & ATTR_SIZE) {
524 		BUG_ON(!S_ISREG(inode->i_mode));
525 
526 		error = inode_newsize_ok(inode, attr->ia_size);
527 		if (error)
528 			return error;
529 
530 		if (attr->ia_size == i_size_read(inode))
531 			attr->ia_valid &= ~ATTR_SIZE;
532 	}
533 
534 	/* Optimization: if the end result is no change, don't RPC */
535 	attr->ia_valid &= NFS_VALID_ATTRS;
536 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
537 		return 0;
538 
539 	trace_nfs_setattr_enter(inode);
540 
541 	/* Write all dirty data */
542 	if (S_ISREG(inode->i_mode))
543 		nfs_sync_inode(inode);
544 
545 	fattr = nfs_alloc_fattr();
546 	if (fattr == NULL) {
547 		error = -ENOMEM;
548 		goto out;
549 	}
550 
551 	/*
552 	 * Return any delegations if we're going to change ACLs
553 	 */
554 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
555 		NFS_PROTO(inode)->return_delegation(inode);
556 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
557 	if (error == 0)
558 		error = nfs_refresh_inode(inode, fattr);
559 	nfs_free_fattr(fattr);
560 out:
561 	trace_nfs_setattr_exit(inode, error);
562 	return error;
563 }
564 EXPORT_SYMBOL_GPL(nfs_setattr);
565 
566 /**
567  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
568  * @inode: inode of the file used
569  * @offset: file offset to start truncating
570  *
571  * This is a copy of the common vmtruncate, but with the locking
572  * corrected to take into account the fact that NFS requires
573  * inode->i_size to be updated under the inode->i_lock.
574  * Note: must be called with inode->i_lock held!
575  */
576 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
577 {
578 	int err;
579 
580 	err = inode_newsize_ok(inode, offset);
581 	if (err)
582 		goto out;
583 
584 	i_size_write(inode, offset);
585 	/* Optimisation */
586 	if (offset == 0)
587 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
588 
589 	spin_unlock(&inode->i_lock);
590 	truncate_pagecache(inode, offset);
591 	spin_lock(&inode->i_lock);
592 out:
593 	return err;
594 }
595 
596 /**
597  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
598  * @inode: pointer to struct inode
599  * @attr: pointer to struct iattr
600  *
601  * Note: we do this in the *proc.c in order to ensure that
602  *       it works for things like exclusive creates too.
603  */
604 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
605 		struct nfs_fattr *fattr)
606 {
607 	/* Barrier: bump the attribute generation count. */
608 	nfs_fattr_set_barrier(fattr);
609 
610 	spin_lock(&inode->i_lock);
611 	NFS_I(inode)->attr_gencount = fattr->gencount;
612 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
613 		if ((attr->ia_valid & ATTR_MODE) != 0) {
614 			int mode = attr->ia_mode & S_IALLUGO;
615 			mode |= inode->i_mode & ~S_IALLUGO;
616 			inode->i_mode = mode;
617 		}
618 		if ((attr->ia_valid & ATTR_UID) != 0)
619 			inode->i_uid = attr->ia_uid;
620 		if ((attr->ia_valid & ATTR_GID) != 0)
621 			inode->i_gid = attr->ia_gid;
622 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
623 				| NFS_INO_INVALID_ACL);
624 	}
625 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
626 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
627 		nfs_vmtruncate(inode, attr->ia_size);
628 	}
629 	if (fattr->valid)
630 		nfs_update_inode(inode, fattr);
631 	else
632 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
633 	spin_unlock(&inode->i_lock);
634 }
635 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
636 
637 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
638 {
639 	struct dentry *parent;
640 
641 	parent = dget_parent(dentry);
642 	nfs_force_use_readdirplus(d_inode(parent));
643 	dput(parent);
644 }
645 
646 static bool nfs_need_revalidate_inode(struct inode *inode)
647 {
648 	if (NFS_I(inode)->cache_validity &
649 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
650 		return true;
651 	if (nfs_attribute_cache_expired(inode))
652 		return true;
653 	return false;
654 }
655 
656 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
657 {
658 	struct inode *inode = d_inode(dentry);
659 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
660 	int err = 0;
661 
662 	trace_nfs_getattr_enter(inode);
663 	/* Flush out writes to the server in order to update c/mtime.  */
664 	if (S_ISREG(inode->i_mode)) {
665 		inode_lock(inode);
666 		err = nfs_sync_inode(inode);
667 		inode_unlock(inode);
668 		if (err)
669 			goto out;
670 	}
671 
672 	/*
673 	 * We may force a getattr if the user cares about atime.
674 	 *
675 	 * Note that we only have to check the vfsmount flags here:
676 	 *  - NFS always sets S_NOATIME by so checking it would give a
677 	 *    bogus result
678 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
679 	 *    no point in checking those.
680 	 */
681  	if ((mnt->mnt_flags & MNT_NOATIME) ||
682  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
683 		need_atime = 0;
684 
685 	if (need_atime || nfs_need_revalidate_inode(inode)) {
686 		struct nfs_server *server = NFS_SERVER(inode);
687 
688 		if (server->caps & NFS_CAP_READDIRPLUS)
689 			nfs_request_parent_use_readdirplus(dentry);
690 		err = __nfs_revalidate_inode(server, inode);
691 	}
692 	if (!err) {
693 		generic_fillattr(inode, stat);
694 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
695 		if (S_ISDIR(inode->i_mode))
696 			stat->blksize = NFS_SERVER(inode)->dtsize;
697 	}
698 out:
699 	trace_nfs_getattr_exit(inode, err);
700 	return err;
701 }
702 EXPORT_SYMBOL_GPL(nfs_getattr);
703 
704 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
705 {
706 	atomic_set(&l_ctx->count, 1);
707 	l_ctx->lockowner.l_owner = current->files;
708 	l_ctx->lockowner.l_pid = current->tgid;
709 	INIT_LIST_HEAD(&l_ctx->list);
710 	atomic_set(&l_ctx->io_count, 0);
711 }
712 
713 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
714 {
715 	struct nfs_lock_context *head = &ctx->lock_context;
716 	struct nfs_lock_context *pos = head;
717 
718 	do {
719 		if (pos->lockowner.l_owner != current->files)
720 			continue;
721 		if (pos->lockowner.l_pid != current->tgid)
722 			continue;
723 		atomic_inc(&pos->count);
724 		return pos;
725 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
726 	return NULL;
727 }
728 
729 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
730 {
731 	struct nfs_lock_context *res, *new = NULL;
732 	struct inode *inode = d_inode(ctx->dentry);
733 
734 	spin_lock(&inode->i_lock);
735 	res = __nfs_find_lock_context(ctx);
736 	if (res == NULL) {
737 		spin_unlock(&inode->i_lock);
738 		new = kmalloc(sizeof(*new), GFP_KERNEL);
739 		if (new == NULL)
740 			return ERR_PTR(-ENOMEM);
741 		nfs_init_lock_context(new);
742 		spin_lock(&inode->i_lock);
743 		res = __nfs_find_lock_context(ctx);
744 		if (res == NULL) {
745 			list_add_tail(&new->list, &ctx->lock_context.list);
746 			new->open_context = ctx;
747 			res = new;
748 			new = NULL;
749 		}
750 	}
751 	spin_unlock(&inode->i_lock);
752 	kfree(new);
753 	return res;
754 }
755 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
756 
757 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
758 {
759 	struct nfs_open_context *ctx = l_ctx->open_context;
760 	struct inode *inode = d_inode(ctx->dentry);
761 
762 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
763 		return;
764 	list_del(&l_ctx->list);
765 	spin_unlock(&inode->i_lock);
766 	kfree(l_ctx);
767 }
768 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
769 
770 /**
771  * nfs_close_context - Common close_context() routine NFSv2/v3
772  * @ctx: pointer to context
773  * @is_sync: is this a synchronous close
774  *
775  * Ensure that the attributes are up to date if we're mounted
776  * with close-to-open semantics and we have cached data that will
777  * need to be revalidated on open.
778  */
779 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
780 {
781 	struct nfs_inode *nfsi;
782 	struct inode *inode;
783 	struct nfs_server *server;
784 
785 	if (!(ctx->mode & FMODE_WRITE))
786 		return;
787 	if (!is_sync)
788 		return;
789 	inode = d_inode(ctx->dentry);
790 	nfsi = NFS_I(inode);
791 	if (inode->i_mapping->nrpages == 0)
792 		return;
793 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
794 		return;
795 	if (!list_empty(&nfsi->open_files))
796 		return;
797 	server = NFS_SERVER(inode);
798 	if (server->flags & NFS_MOUNT_NOCTO)
799 		return;
800 	nfs_revalidate_inode(server, inode);
801 }
802 EXPORT_SYMBOL_GPL(nfs_close_context);
803 
804 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
805 {
806 	struct nfs_open_context *ctx;
807 	struct rpc_cred *cred = rpc_lookup_cred();
808 	if (IS_ERR(cred))
809 		return ERR_CAST(cred);
810 
811 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
812 	if (!ctx) {
813 		put_rpccred(cred);
814 		return ERR_PTR(-ENOMEM);
815 	}
816 	nfs_sb_active(dentry->d_sb);
817 	ctx->dentry = dget(dentry);
818 	ctx->cred = cred;
819 	ctx->state = NULL;
820 	ctx->mode = f_mode;
821 	ctx->flags = 0;
822 	ctx->error = 0;
823 	nfs_init_lock_context(&ctx->lock_context);
824 	ctx->lock_context.open_context = ctx;
825 	INIT_LIST_HEAD(&ctx->list);
826 	ctx->mdsthreshold = NULL;
827 	return ctx;
828 }
829 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
830 
831 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
832 {
833 	if (ctx != NULL)
834 		atomic_inc(&ctx->lock_context.count);
835 	return ctx;
836 }
837 EXPORT_SYMBOL_GPL(get_nfs_open_context);
838 
839 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
840 {
841 	struct inode *inode = d_inode(ctx->dentry);
842 	struct super_block *sb = ctx->dentry->d_sb;
843 
844 	if (!list_empty(&ctx->list)) {
845 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
846 			return;
847 		list_del(&ctx->list);
848 		spin_unlock(&inode->i_lock);
849 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
850 		return;
851 	if (inode != NULL)
852 		NFS_PROTO(inode)->close_context(ctx, is_sync);
853 	if (ctx->cred != NULL)
854 		put_rpccred(ctx->cred);
855 	dput(ctx->dentry);
856 	nfs_sb_deactive(sb);
857 	kfree(ctx->mdsthreshold);
858 	kfree(ctx);
859 }
860 
861 void put_nfs_open_context(struct nfs_open_context *ctx)
862 {
863 	__put_nfs_open_context(ctx, 0);
864 }
865 EXPORT_SYMBOL_GPL(put_nfs_open_context);
866 
867 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
868 {
869 	__put_nfs_open_context(ctx, 1);
870 }
871 
872 /*
873  * Ensure that mmap has a recent RPC credential for use when writing out
874  * shared pages
875  */
876 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
877 {
878 	struct inode *inode = d_inode(ctx->dentry);
879 	struct nfs_inode *nfsi = NFS_I(inode);
880 
881 	spin_lock(&inode->i_lock);
882 	list_add(&ctx->list, &nfsi->open_files);
883 	spin_unlock(&inode->i_lock);
884 }
885 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
886 
887 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
888 {
889 	filp->private_data = get_nfs_open_context(ctx);
890 	if (list_empty(&ctx->list))
891 		nfs_inode_attach_open_context(ctx);
892 }
893 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
894 
895 /*
896  * Given an inode, search for an open context with the desired characteristics
897  */
898 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
899 {
900 	struct nfs_inode *nfsi = NFS_I(inode);
901 	struct nfs_open_context *pos, *ctx = NULL;
902 
903 	spin_lock(&inode->i_lock);
904 	list_for_each_entry(pos, &nfsi->open_files, list) {
905 		if (cred != NULL && pos->cred != cred)
906 			continue;
907 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
908 			continue;
909 		ctx = get_nfs_open_context(pos);
910 		break;
911 	}
912 	spin_unlock(&inode->i_lock);
913 	return ctx;
914 }
915 
916 void nfs_file_clear_open_context(struct file *filp)
917 {
918 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
919 
920 	if (ctx) {
921 		struct inode *inode = d_inode(ctx->dentry);
922 
923 		/*
924 		 * We fatal error on write before. Try to writeback
925 		 * every page again.
926 		 */
927 		if (ctx->error < 0)
928 			invalidate_inode_pages2(inode->i_mapping);
929 		filp->private_data = NULL;
930 		spin_lock(&inode->i_lock);
931 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
932 		spin_unlock(&inode->i_lock);
933 		put_nfs_open_context_sync(ctx);
934 	}
935 }
936 
937 /*
938  * These allocate and release file read/write context information.
939  */
940 int nfs_open(struct inode *inode, struct file *filp)
941 {
942 	struct nfs_open_context *ctx;
943 
944 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode);
945 	if (IS_ERR(ctx))
946 		return PTR_ERR(ctx);
947 	nfs_file_set_open_context(filp, ctx);
948 	put_nfs_open_context(ctx);
949 	nfs_fscache_open_file(inode, filp);
950 	return 0;
951 }
952 
953 /*
954  * This function is called whenever some part of NFS notices that
955  * the cached attributes have to be refreshed.
956  */
957 int
958 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
959 {
960 	int		 status = -ESTALE;
961 	struct nfs4_label *label = NULL;
962 	struct nfs_fattr *fattr = NULL;
963 	struct nfs_inode *nfsi = NFS_I(inode);
964 
965 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
966 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
967 
968 	trace_nfs_revalidate_inode_enter(inode);
969 
970 	if (is_bad_inode(inode))
971 		goto out;
972 	if (NFS_STALE(inode))
973 		goto out;
974 
975 	status = -ENOMEM;
976 	fattr = nfs_alloc_fattr();
977 	if (fattr == NULL)
978 		goto out;
979 
980 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
981 
982 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
983 	if (IS_ERR(label)) {
984 		status = PTR_ERR(label);
985 		goto out;
986 	}
987 
988 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
989 	if (status != 0) {
990 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
991 			 inode->i_sb->s_id,
992 			 (unsigned long long)NFS_FILEID(inode), status);
993 		if (status == -ESTALE) {
994 			nfs_zap_caches(inode);
995 			if (!S_ISDIR(inode->i_mode))
996 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
997 		}
998 		goto err_out;
999 	}
1000 
1001 	status = nfs_refresh_inode(inode, fattr);
1002 	if (status) {
1003 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1004 			 inode->i_sb->s_id,
1005 			 (unsigned long long)NFS_FILEID(inode), status);
1006 		goto err_out;
1007 	}
1008 
1009 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1010 		nfs_zap_acl_cache(inode);
1011 
1012 	nfs_setsecurity(inode, fattr, label);
1013 
1014 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1015 		inode->i_sb->s_id,
1016 		(unsigned long long)NFS_FILEID(inode));
1017 
1018 err_out:
1019 	nfs4_label_free(label);
1020 out:
1021 	nfs_free_fattr(fattr);
1022 	trace_nfs_revalidate_inode_exit(inode, status);
1023 	return status;
1024 }
1025 
1026 int nfs_attribute_timeout(struct inode *inode)
1027 {
1028 	struct nfs_inode *nfsi = NFS_I(inode);
1029 
1030 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1031 }
1032 
1033 int nfs_attribute_cache_expired(struct inode *inode)
1034 {
1035 	if (nfs_have_delegated_attributes(inode))
1036 		return 0;
1037 	return nfs_attribute_timeout(inode);
1038 }
1039 
1040 /**
1041  * nfs_revalidate_inode - Revalidate the inode attributes
1042  * @server - pointer to nfs_server struct
1043  * @inode - pointer to inode struct
1044  *
1045  * Updates inode attribute information by retrieving the data from the server.
1046  */
1047 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1048 {
1049 	if (!nfs_need_revalidate_inode(inode))
1050 		return NFS_STALE(inode) ? -ESTALE : 0;
1051 	return __nfs_revalidate_inode(server, inode);
1052 }
1053 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1054 
1055 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1056 {
1057 	if (!(NFS_I(inode)->cache_validity &
1058 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1059 			&& !nfs_attribute_cache_expired(inode))
1060 		return NFS_STALE(inode) ? -ESTALE : 0;
1061 	return -ECHILD;
1062 }
1063 
1064 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1065 {
1066 	struct nfs_inode *nfsi = NFS_I(inode);
1067 	int ret;
1068 
1069 	if (mapping->nrpages != 0) {
1070 		if (S_ISREG(inode->i_mode)) {
1071 			unmap_mapping_range(mapping, 0, 0, 0);
1072 			ret = nfs_sync_mapping(mapping);
1073 			if (ret < 0)
1074 				return ret;
1075 		}
1076 		ret = invalidate_inode_pages2(mapping);
1077 		if (ret < 0)
1078 			return ret;
1079 	}
1080 	if (S_ISDIR(inode->i_mode)) {
1081 		spin_lock(&inode->i_lock);
1082 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1083 		spin_unlock(&inode->i_lock);
1084 	}
1085 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1086 	nfs_fscache_wait_on_invalidate(inode);
1087 
1088 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1089 			inode->i_sb->s_id,
1090 			(unsigned long long)NFS_FILEID(inode));
1091 	return 0;
1092 }
1093 
1094 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1095 {
1096 	if (nfs_have_delegated_attributes(inode))
1097 		return false;
1098 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1099 		|| nfs_attribute_timeout(inode)
1100 		|| NFS_STALE(inode);
1101 }
1102 
1103 int nfs_revalidate_mapping_rcu(struct inode *inode)
1104 {
1105 	struct nfs_inode *nfsi = NFS_I(inode);
1106 	unsigned long *bitlock = &nfsi->flags;
1107 	int ret = 0;
1108 
1109 	if (IS_SWAPFILE(inode))
1110 		goto out;
1111 	if (nfs_mapping_need_revalidate_inode(inode)) {
1112 		ret = -ECHILD;
1113 		goto out;
1114 	}
1115 	spin_lock(&inode->i_lock);
1116 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1117 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1118 		ret = -ECHILD;
1119 	spin_unlock(&inode->i_lock);
1120 out:
1121 	return ret;
1122 }
1123 
1124 /**
1125  * __nfs_revalidate_mapping - Revalidate the pagecache
1126  * @inode - pointer to host inode
1127  * @mapping - pointer to mapping
1128  * @may_lock - take inode->i_mutex?
1129  */
1130 static int __nfs_revalidate_mapping(struct inode *inode,
1131 		struct address_space *mapping,
1132 		bool may_lock)
1133 {
1134 	struct nfs_inode *nfsi = NFS_I(inode);
1135 	unsigned long *bitlock = &nfsi->flags;
1136 	int ret = 0;
1137 
1138 	/* swapfiles are not supposed to be shared. */
1139 	if (IS_SWAPFILE(inode))
1140 		goto out;
1141 
1142 	if (nfs_mapping_need_revalidate_inode(inode)) {
1143 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1144 		if (ret < 0)
1145 			goto out;
1146 	}
1147 
1148 	/*
1149 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1150 	 * invalidations that come in while we're shooting down the mappings
1151 	 * are respected. But, that leaves a race window where one revalidator
1152 	 * can clear the flag, and then another checks it before the mapping
1153 	 * gets invalidated. Fix that by serializing access to this part of
1154 	 * the function.
1155 	 *
1156 	 * At the same time, we need to allow other tasks to see whether we
1157 	 * might be in the middle of invalidating the pages, so we only set
1158 	 * the bit lock here if it looks like we're going to be doing that.
1159 	 */
1160 	for (;;) {
1161 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1162 					 nfs_wait_bit_killable, TASK_KILLABLE);
1163 		if (ret)
1164 			goto out;
1165 		spin_lock(&inode->i_lock);
1166 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1167 			spin_unlock(&inode->i_lock);
1168 			continue;
1169 		}
1170 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1171 			break;
1172 		spin_unlock(&inode->i_lock);
1173 		goto out;
1174 	}
1175 
1176 	set_bit(NFS_INO_INVALIDATING, bitlock);
1177 	smp_wmb();
1178 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1179 	spin_unlock(&inode->i_lock);
1180 	trace_nfs_invalidate_mapping_enter(inode);
1181 	if (may_lock) {
1182 		inode_lock(inode);
1183 		ret = nfs_invalidate_mapping(inode, mapping);
1184 		inode_unlock(inode);
1185 	} else
1186 		ret = nfs_invalidate_mapping(inode, mapping);
1187 	trace_nfs_invalidate_mapping_exit(inode, ret);
1188 
1189 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1190 	smp_mb__after_atomic();
1191 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1192 out:
1193 	return ret;
1194 }
1195 
1196 /**
1197  * nfs_revalidate_mapping - Revalidate the pagecache
1198  * @inode - pointer to host inode
1199  * @mapping - pointer to mapping
1200  */
1201 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1202 {
1203 	return __nfs_revalidate_mapping(inode, mapping, false);
1204 }
1205 
1206 /**
1207  * nfs_revalidate_mapping_protected - Revalidate the pagecache
1208  * @inode - pointer to host inode
1209  * @mapping - pointer to mapping
1210  *
1211  * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1212  * while invalidating the mapping.
1213  */
1214 int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1215 {
1216 	return __nfs_revalidate_mapping(inode, mapping, true);
1217 }
1218 
1219 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1220 {
1221 	struct nfs_inode *nfsi = NFS_I(inode);
1222 	unsigned long ret = 0;
1223 
1224 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1225 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1226 			&& inode->i_version == fattr->pre_change_attr) {
1227 		inode->i_version = fattr->change_attr;
1228 		if (S_ISDIR(inode->i_mode))
1229 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1230 		ret |= NFS_INO_INVALID_ATTR;
1231 	}
1232 	/* If we have atomic WCC data, we may update some attributes */
1233 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1234 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1235 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1236 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1237 		ret |= NFS_INO_INVALID_ATTR;
1238 	}
1239 
1240 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1241 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1242 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1243 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1244 		if (S_ISDIR(inode->i_mode))
1245 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1246 		ret |= NFS_INO_INVALID_ATTR;
1247 	}
1248 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1249 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1250 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1251 			&& nfsi->nrequests == 0) {
1252 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1253 		ret |= NFS_INO_INVALID_ATTR;
1254 	}
1255 
1256 	return ret;
1257 }
1258 
1259 /**
1260  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1261  * @inode - pointer to inode
1262  * @fattr - updated attributes
1263  *
1264  * Verifies the attribute cache. If we have just changed the attributes,
1265  * so that fattr carries weak cache consistency data, then it may
1266  * also update the ctime/mtime/change_attribute.
1267  */
1268 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1269 {
1270 	struct nfs_inode *nfsi = NFS_I(inode);
1271 	loff_t cur_size, new_isize;
1272 	unsigned long invalid = 0;
1273 
1274 
1275 	if (nfs_have_delegated_attributes(inode))
1276 		return 0;
1277 	/* Has the inode gone and changed behind our back? */
1278 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1279 		return -EIO;
1280 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1281 		return -EIO;
1282 
1283 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1284 			inode->i_version != fattr->change_attr)
1285 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1286 
1287 	/* Verify a few of the more important attributes */
1288 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1289 		invalid |= NFS_INO_INVALID_ATTR;
1290 
1291 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1292 		cur_size = i_size_read(inode);
1293 		new_isize = nfs_size_to_loff_t(fattr->size);
1294 		if (cur_size != new_isize)
1295 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1296 	}
1297 	if (nfsi->nrequests != 0)
1298 		invalid &= ~NFS_INO_REVAL_PAGECACHE;
1299 
1300 	/* Have any file permissions changed? */
1301 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1302 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1303 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1304 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1305 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1306 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1307 
1308 	/* Has the link count changed? */
1309 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1310 		invalid |= NFS_INO_INVALID_ATTR;
1311 
1312 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1313 		invalid |= NFS_INO_INVALID_ATIME;
1314 
1315 	if (invalid != 0)
1316 		nfs_set_cache_invalid(inode, invalid);
1317 
1318 	nfsi->read_cache_jiffies = fattr->time_start;
1319 	return 0;
1320 }
1321 
1322 static atomic_long_t nfs_attr_generation_counter;
1323 
1324 static unsigned long nfs_read_attr_generation_counter(void)
1325 {
1326 	return atomic_long_read(&nfs_attr_generation_counter);
1327 }
1328 
1329 unsigned long nfs_inc_attr_generation_counter(void)
1330 {
1331 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1332 }
1333 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1334 
1335 void nfs_fattr_init(struct nfs_fattr *fattr)
1336 {
1337 	fattr->valid = 0;
1338 	fattr->time_start = jiffies;
1339 	fattr->gencount = nfs_inc_attr_generation_counter();
1340 	fattr->owner_name = NULL;
1341 	fattr->group_name = NULL;
1342 }
1343 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1344 
1345 /**
1346  * nfs_fattr_set_barrier
1347  * @fattr: attributes
1348  *
1349  * Used to set a barrier after an attribute was updated. This
1350  * barrier ensures that older attributes from RPC calls that may
1351  * have raced with our update cannot clobber these new values.
1352  * Note that you are still responsible for ensuring that other
1353  * operations which change the attribute on the server do not
1354  * collide.
1355  */
1356 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1357 {
1358 	fattr->gencount = nfs_inc_attr_generation_counter();
1359 }
1360 
1361 struct nfs_fattr *nfs_alloc_fattr(void)
1362 {
1363 	struct nfs_fattr *fattr;
1364 
1365 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1366 	if (fattr != NULL)
1367 		nfs_fattr_init(fattr);
1368 	return fattr;
1369 }
1370 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1371 
1372 struct nfs_fh *nfs_alloc_fhandle(void)
1373 {
1374 	struct nfs_fh *fh;
1375 
1376 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1377 	if (fh != NULL)
1378 		fh->size = 0;
1379 	return fh;
1380 }
1381 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1382 
1383 #ifdef NFS_DEBUG
1384 /*
1385  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1386  *                             in the same way that wireshark does
1387  *
1388  * @fh: file handle
1389  *
1390  * For debugging only.
1391  */
1392 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1393 {
1394 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1395 	 * not on the result */
1396 	return nfs_fhandle_hash(fh);
1397 }
1398 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1399 
1400 /*
1401  * _nfs_display_fhandle - display an NFS file handle on the console
1402  *
1403  * @fh: file handle to display
1404  * @caption: display caption
1405  *
1406  * For debugging only.
1407  */
1408 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1409 {
1410 	unsigned short i;
1411 
1412 	if (fh == NULL || fh->size == 0) {
1413 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1414 		return;
1415 	}
1416 
1417 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1418 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1419 	for (i = 0; i < fh->size; i += 16) {
1420 		__be32 *pos = (__be32 *)&fh->data[i];
1421 
1422 		switch ((fh->size - i - 1) >> 2) {
1423 		case 0:
1424 			printk(KERN_DEFAULT " %08x\n",
1425 				be32_to_cpup(pos));
1426 			break;
1427 		case 1:
1428 			printk(KERN_DEFAULT " %08x %08x\n",
1429 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1430 			break;
1431 		case 2:
1432 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1433 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1434 				be32_to_cpup(pos + 2));
1435 			break;
1436 		default:
1437 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1438 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1439 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1440 		}
1441 	}
1442 }
1443 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1444 #endif
1445 
1446 /**
1447  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1448  * @inode - pointer to inode
1449  * @fattr - attributes
1450  *
1451  * Attempt to divine whether or not an RPC call reply carrying stale
1452  * attributes got scheduled after another call carrying updated ones.
1453  *
1454  * To do so, the function first assumes that a more recent ctime means
1455  * that the attributes in fattr are newer, however it also attempt to
1456  * catch the case where ctime either didn't change, or went backwards
1457  * (if someone reset the clock on the server) by looking at whether
1458  * or not this RPC call was started after the inode was last updated.
1459  * Note also the check for wraparound of 'attr_gencount'
1460  *
1461  * The function returns 'true' if it thinks the attributes in 'fattr' are
1462  * more recent than the ones cached in the inode.
1463  *
1464  */
1465 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1466 {
1467 	const struct nfs_inode *nfsi = NFS_I(inode);
1468 
1469 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1470 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1471 }
1472 
1473 /*
1474  * Don't trust the change_attribute, mtime, ctime or size if
1475  * a pnfs LAYOUTCOMMIT is outstanding
1476  */
1477 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1478 		struct nfs_fattr *fattr)
1479 {
1480 	if (pnfs_layoutcommit_outstanding(inode))
1481 		fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1482 				NFS_ATTR_FATTR_MTIME |
1483 				NFS_ATTR_FATTR_CTIME |
1484 				NFS_ATTR_FATTR_SIZE);
1485 }
1486 
1487 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1488 {
1489 	int ret;
1490 
1491 	trace_nfs_refresh_inode_enter(inode);
1492 
1493 	nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1494 
1495 	if (nfs_inode_attrs_need_update(inode, fattr))
1496 		ret = nfs_update_inode(inode, fattr);
1497 	else
1498 		ret = nfs_check_inode_attributes(inode, fattr);
1499 
1500 	trace_nfs_refresh_inode_exit(inode, ret);
1501 	return ret;
1502 }
1503 
1504 /**
1505  * nfs_refresh_inode - try to update the inode attribute cache
1506  * @inode - pointer to inode
1507  * @fattr - updated attributes
1508  *
1509  * Check that an RPC call that returned attributes has not overlapped with
1510  * other recent updates of the inode metadata, then decide whether it is
1511  * safe to do a full update of the inode attributes, or whether just to
1512  * call nfs_check_inode_attributes.
1513  */
1514 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1515 {
1516 	int status;
1517 
1518 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1519 		return 0;
1520 	spin_lock(&inode->i_lock);
1521 	status = nfs_refresh_inode_locked(inode, fattr);
1522 	spin_unlock(&inode->i_lock);
1523 
1524 	return status;
1525 }
1526 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1527 
1528 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1529 {
1530 	unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1531 
1532 	/*
1533 	 * Don't revalidate the pagecache if we hold a delegation, but do
1534 	 * force an attribute update
1535 	 */
1536 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1537 		invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1538 
1539 	if (S_ISDIR(inode->i_mode))
1540 		invalid |= NFS_INO_INVALID_DATA;
1541 	nfs_set_cache_invalid(inode, invalid);
1542 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1543 		return 0;
1544 	return nfs_refresh_inode_locked(inode, fattr);
1545 }
1546 
1547 /**
1548  * nfs_post_op_update_inode - try to update the inode attribute cache
1549  * @inode - pointer to inode
1550  * @fattr - updated attributes
1551  *
1552  * After an operation that has changed the inode metadata, mark the
1553  * attribute cache as being invalid, then try to update it.
1554  *
1555  * NB: if the server didn't return any post op attributes, this
1556  * function will force the retrieval of attributes before the next
1557  * NFS request.  Thus it should be used only for operations that
1558  * are expected to change one or more attributes, to avoid
1559  * unnecessary NFS requests and trips through nfs_update_inode().
1560  */
1561 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1562 {
1563 	int status;
1564 
1565 	spin_lock(&inode->i_lock);
1566 	nfs_fattr_set_barrier(fattr);
1567 	status = nfs_post_op_update_inode_locked(inode, fattr);
1568 	spin_unlock(&inode->i_lock);
1569 
1570 	return status;
1571 }
1572 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1573 
1574 /**
1575  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1576  * @inode - pointer to inode
1577  * @fattr - updated attributes
1578  *
1579  * After an operation that has changed the inode metadata, mark the
1580  * attribute cache as being invalid, then try to update it. Fake up
1581  * weak cache consistency data, if none exist.
1582  *
1583  * This function is mainly designed to be used by the ->write_done() functions.
1584  */
1585 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1586 {
1587 	int status;
1588 
1589 	/* Don't do a WCC update if these attributes are already stale */
1590 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1591 			!nfs_inode_attrs_need_update(inode, fattr)) {
1592 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1593 				| NFS_ATTR_FATTR_PRESIZE
1594 				| NFS_ATTR_FATTR_PREMTIME
1595 				| NFS_ATTR_FATTR_PRECTIME);
1596 		goto out_noforce;
1597 	}
1598 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1599 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1600 		fattr->pre_change_attr = inode->i_version;
1601 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1602 	}
1603 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1604 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1605 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1606 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1607 	}
1608 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1609 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1610 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1611 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1612 	}
1613 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1614 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1615 		fattr->pre_size = i_size_read(inode);
1616 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1617 	}
1618 out_noforce:
1619 	status = nfs_post_op_update_inode_locked(inode, fattr);
1620 	return status;
1621 }
1622 
1623 /**
1624  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1625  * @inode - pointer to inode
1626  * @fattr - updated attributes
1627  *
1628  * After an operation that has changed the inode metadata, mark the
1629  * attribute cache as being invalid, then try to update it. Fake up
1630  * weak cache consistency data, if none exist.
1631  *
1632  * This function is mainly designed to be used by the ->write_done() functions.
1633  */
1634 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1635 {
1636 	int status;
1637 
1638 	spin_lock(&inode->i_lock);
1639 	nfs_fattr_set_barrier(fattr);
1640 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1641 	spin_unlock(&inode->i_lock);
1642 	return status;
1643 }
1644 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1645 
1646 
1647 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1648 				    struct nfs_fattr *fattr)
1649 {
1650 	bool ret1 = true, ret2 = true;
1651 
1652 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1653 		ret1 = (nfsi->fileid == fattr->fileid);
1654 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1655 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1656 	return ret1 || ret2;
1657 }
1658 
1659 /*
1660  * Many nfs protocol calls return the new file attributes after
1661  * an operation.  Here we update the inode to reflect the state
1662  * of the server's inode.
1663  *
1664  * This is a bit tricky because we have to make sure all dirty pages
1665  * have been sent off to the server before calling invalidate_inode_pages.
1666  * To make sure no other process adds more write requests while we try
1667  * our best to flush them, we make them sleep during the attribute refresh.
1668  *
1669  * A very similar scenario holds for the dir cache.
1670  */
1671 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1672 {
1673 	struct nfs_server *server;
1674 	struct nfs_inode *nfsi = NFS_I(inode);
1675 	loff_t cur_isize, new_isize;
1676 	unsigned long invalid = 0;
1677 	unsigned long now = jiffies;
1678 	unsigned long save_cache_validity;
1679 	bool cache_revalidated = true;
1680 
1681 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1682 			__func__, inode->i_sb->s_id, inode->i_ino,
1683 			nfs_display_fhandle_hash(NFS_FH(inode)),
1684 			atomic_read(&inode->i_count), fattr->valid);
1685 
1686 	if (!nfs_fileid_valid(nfsi, fattr)) {
1687 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1688 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1689 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1690 			inode->i_sb->s_id, (long long)nfsi->fileid,
1691 			(long long)fattr->fileid);
1692 		goto out_err;
1693 	}
1694 
1695 	/*
1696 	 * Make sure the inode's type hasn't changed.
1697 	 */
1698 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1699 		/*
1700 		* Big trouble! The inode has become a different object.
1701 		*/
1702 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1703 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1704 		goto out_err;
1705 	}
1706 
1707 	server = NFS_SERVER(inode);
1708 	/* Update the fsid? */
1709 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1710 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1711 			!IS_AUTOMOUNT(inode))
1712 		server->fsid = fattr->fsid;
1713 
1714 	/*
1715 	 * Update the read time so we don't revalidate too often.
1716 	 */
1717 	nfsi->read_cache_jiffies = fattr->time_start;
1718 
1719 	save_cache_validity = nfsi->cache_validity;
1720 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1721 			| NFS_INO_INVALID_ATIME
1722 			| NFS_INO_REVAL_FORCED
1723 			| NFS_INO_REVAL_PAGECACHE);
1724 
1725 	/* Do atomic weak cache consistency updates */
1726 	invalid |= nfs_wcc_update_inode(inode, fattr);
1727 
1728 	/* More cache consistency checks */
1729 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1730 		if (inode->i_version != fattr->change_attr) {
1731 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1732 					inode->i_sb->s_id, inode->i_ino);
1733 			invalid |= NFS_INO_INVALID_ATTR
1734 				| NFS_INO_INVALID_DATA
1735 				| NFS_INO_INVALID_ACCESS
1736 				| NFS_INO_INVALID_ACL;
1737 			if (S_ISDIR(inode->i_mode))
1738 				nfs_force_lookup_revalidate(inode);
1739 			inode->i_version = fattr->change_attr;
1740 		}
1741 	} else {
1742 		nfsi->cache_validity |= save_cache_validity;
1743 		cache_revalidated = false;
1744 	}
1745 
1746 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1747 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1748 	} else if (server->caps & NFS_CAP_MTIME) {
1749 		nfsi->cache_validity |= save_cache_validity &
1750 				(NFS_INO_INVALID_ATTR
1751 				| NFS_INO_REVAL_FORCED);
1752 		cache_revalidated = false;
1753 	}
1754 
1755 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1756 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1757 	} else if (server->caps & NFS_CAP_CTIME) {
1758 		nfsi->cache_validity |= save_cache_validity &
1759 				(NFS_INO_INVALID_ATTR
1760 				| NFS_INO_REVAL_FORCED);
1761 		cache_revalidated = false;
1762 	}
1763 
1764 	/* Check if our cached file size is stale */
1765 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1766 		new_isize = nfs_size_to_loff_t(fattr->size);
1767 		cur_isize = i_size_read(inode);
1768 		if (new_isize != cur_isize) {
1769 			/* Do we perhaps have any outstanding writes, or has
1770 			 * the file grown beyond our last write? */
1771 			if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1772 				i_size_write(inode, new_isize);
1773 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1774 			}
1775 			dprintk("NFS: isize change on server for file %s/%ld "
1776 					"(%Ld to %Ld)\n",
1777 					inode->i_sb->s_id,
1778 					inode->i_ino,
1779 					(long long)cur_isize,
1780 					(long long)new_isize);
1781 		}
1782 	} else {
1783 		nfsi->cache_validity |= save_cache_validity &
1784 				(NFS_INO_INVALID_ATTR
1785 				| NFS_INO_REVAL_PAGECACHE
1786 				| NFS_INO_REVAL_FORCED);
1787 		cache_revalidated = false;
1788 	}
1789 
1790 
1791 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1792 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1793 	else if (server->caps & NFS_CAP_ATIME) {
1794 		nfsi->cache_validity |= save_cache_validity &
1795 				(NFS_INO_INVALID_ATIME
1796 				| NFS_INO_REVAL_FORCED);
1797 		cache_revalidated = false;
1798 	}
1799 
1800 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1801 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1802 			umode_t newmode = inode->i_mode & S_IFMT;
1803 			newmode |= fattr->mode & S_IALLUGO;
1804 			inode->i_mode = newmode;
1805 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1806 		}
1807 	} else if (server->caps & NFS_CAP_MODE) {
1808 		nfsi->cache_validity |= save_cache_validity &
1809 				(NFS_INO_INVALID_ATTR
1810 				| NFS_INO_INVALID_ACCESS
1811 				| NFS_INO_INVALID_ACL
1812 				| NFS_INO_REVAL_FORCED);
1813 		cache_revalidated = false;
1814 	}
1815 
1816 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1817 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1818 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1819 			inode->i_uid = fattr->uid;
1820 		}
1821 	} else if (server->caps & NFS_CAP_OWNER) {
1822 		nfsi->cache_validity |= save_cache_validity &
1823 				(NFS_INO_INVALID_ATTR
1824 				| NFS_INO_INVALID_ACCESS
1825 				| NFS_INO_INVALID_ACL
1826 				| NFS_INO_REVAL_FORCED);
1827 		cache_revalidated = false;
1828 	}
1829 
1830 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1831 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1832 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1833 			inode->i_gid = fattr->gid;
1834 		}
1835 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1836 		nfsi->cache_validity |= save_cache_validity &
1837 				(NFS_INO_INVALID_ATTR
1838 				| NFS_INO_INVALID_ACCESS
1839 				| NFS_INO_INVALID_ACL
1840 				| NFS_INO_REVAL_FORCED);
1841 		cache_revalidated = false;
1842 	}
1843 
1844 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1845 		if (inode->i_nlink != fattr->nlink) {
1846 			invalid |= NFS_INO_INVALID_ATTR;
1847 			if (S_ISDIR(inode->i_mode))
1848 				invalid |= NFS_INO_INVALID_DATA;
1849 			set_nlink(inode, fattr->nlink);
1850 		}
1851 	} else if (server->caps & NFS_CAP_NLINK) {
1852 		nfsi->cache_validity |= save_cache_validity &
1853 				(NFS_INO_INVALID_ATTR
1854 				| NFS_INO_REVAL_FORCED);
1855 		cache_revalidated = false;
1856 	}
1857 
1858 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1859 		/*
1860 		 * report the blocks in 512byte units
1861 		 */
1862 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1863 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1864 		inode->i_blocks = fattr->du.nfs2.blocks;
1865 	else
1866 		cache_revalidated = false;
1867 
1868 	/* Update attrtimeo value if we're out of the unstable period */
1869 	if (invalid & NFS_INO_INVALID_ATTR) {
1870 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1871 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1872 		nfsi->attrtimeo_timestamp = now;
1873 		/* Set barrier to be more recent than all outstanding updates */
1874 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1875 	} else {
1876 		if (cache_revalidated) {
1877 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1878 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1879 				nfsi->attrtimeo <<= 1;
1880 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1881 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1882 			}
1883 			nfsi->attrtimeo_timestamp = now;
1884 		}
1885 		/* Set the barrier to be more recent than this fattr */
1886 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1887 			nfsi->attr_gencount = fattr->gencount;
1888 	}
1889 
1890 	/* Don't declare attrcache up to date if there were no attrs! */
1891 	if (cache_revalidated)
1892 		invalid &= ~NFS_INO_INVALID_ATTR;
1893 
1894 	/* Don't invalidate the data if we were to blame */
1895 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1896 				|| S_ISLNK(inode->i_mode)))
1897 		invalid &= ~NFS_INO_INVALID_DATA;
1898 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1899 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1900 		nfs_set_cache_invalid(inode, invalid);
1901 
1902 	return 0;
1903  out_err:
1904 	/*
1905 	 * No need to worry about unhashing the dentry, as the
1906 	 * lookup validation will know that the inode is bad.
1907 	 * (But we fall through to invalidate the caches.)
1908 	 */
1909 	nfs_invalidate_inode(inode);
1910 	return -ESTALE;
1911 }
1912 
1913 struct inode *nfs_alloc_inode(struct super_block *sb)
1914 {
1915 	struct nfs_inode *nfsi;
1916 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1917 	if (!nfsi)
1918 		return NULL;
1919 	nfsi->flags = 0UL;
1920 	nfsi->cache_validity = 0UL;
1921 #if IS_ENABLED(CONFIG_NFS_V4)
1922 	nfsi->nfs4_acl = NULL;
1923 #endif /* CONFIG_NFS_V4 */
1924 	return &nfsi->vfs_inode;
1925 }
1926 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1927 
1928 static void nfs_i_callback(struct rcu_head *head)
1929 {
1930 	struct inode *inode = container_of(head, struct inode, i_rcu);
1931 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1932 }
1933 
1934 void nfs_destroy_inode(struct inode *inode)
1935 {
1936 	call_rcu(&inode->i_rcu, nfs_i_callback);
1937 }
1938 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1939 
1940 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1941 {
1942 #if IS_ENABLED(CONFIG_NFS_V4)
1943 	INIT_LIST_HEAD(&nfsi->open_states);
1944 	nfsi->delegation = NULL;
1945 	init_rwsem(&nfsi->rwsem);
1946 	nfsi->layout = NULL;
1947 #endif
1948 }
1949 
1950 static void init_once(void *foo)
1951 {
1952 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1953 
1954 	inode_init_once(&nfsi->vfs_inode);
1955 	INIT_LIST_HEAD(&nfsi->open_files);
1956 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1957 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1958 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1959 	nfsi->nrequests = 0;
1960 	nfsi->commit_info.ncommit = 0;
1961 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1962 	init_rwsem(&nfsi->rmdir_sem);
1963 	nfs4_init_once(nfsi);
1964 }
1965 
1966 static int __init nfs_init_inodecache(void)
1967 {
1968 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1969 					     sizeof(struct nfs_inode),
1970 					     0, (SLAB_RECLAIM_ACCOUNT|
1971 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1972 					     init_once);
1973 	if (nfs_inode_cachep == NULL)
1974 		return -ENOMEM;
1975 
1976 	return 0;
1977 }
1978 
1979 static void nfs_destroy_inodecache(void)
1980 {
1981 	/*
1982 	 * Make sure all delayed rcu free inodes are flushed before we
1983 	 * destroy cache.
1984 	 */
1985 	rcu_barrier();
1986 	kmem_cache_destroy(nfs_inode_cachep);
1987 }
1988 
1989 struct workqueue_struct *nfsiod_workqueue;
1990 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1991 
1992 /*
1993  * start up the nfsiod workqueue
1994  */
1995 static int nfsiod_start(void)
1996 {
1997 	struct workqueue_struct *wq;
1998 	dprintk("RPC:       creating workqueue nfsiod\n");
1999 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2000 	if (wq == NULL)
2001 		return -ENOMEM;
2002 	nfsiod_workqueue = wq;
2003 	return 0;
2004 }
2005 
2006 /*
2007  * Destroy the nfsiod workqueue
2008  */
2009 static void nfsiod_stop(void)
2010 {
2011 	struct workqueue_struct *wq;
2012 
2013 	wq = nfsiod_workqueue;
2014 	if (wq == NULL)
2015 		return;
2016 	nfsiod_workqueue = NULL;
2017 	destroy_workqueue(wq);
2018 }
2019 
2020 int nfs_net_id;
2021 EXPORT_SYMBOL_GPL(nfs_net_id);
2022 
2023 static int nfs_net_init(struct net *net)
2024 {
2025 	nfs_clients_init(net);
2026 	return nfs_fs_proc_net_init(net);
2027 }
2028 
2029 static void nfs_net_exit(struct net *net)
2030 {
2031 	nfs_fs_proc_net_exit(net);
2032 	nfs_cleanup_cb_ident_idr(net);
2033 }
2034 
2035 static struct pernet_operations nfs_net_ops = {
2036 	.init = nfs_net_init,
2037 	.exit = nfs_net_exit,
2038 	.id   = &nfs_net_id,
2039 	.size = sizeof(struct nfs_net),
2040 };
2041 
2042 /*
2043  * Initialize NFS
2044  */
2045 static int __init init_nfs_fs(void)
2046 {
2047 	int err;
2048 
2049 	err = register_pernet_subsys(&nfs_net_ops);
2050 	if (err < 0)
2051 		goto out9;
2052 
2053 	err = nfs_fscache_register();
2054 	if (err < 0)
2055 		goto out8;
2056 
2057 	err = nfsiod_start();
2058 	if (err)
2059 		goto out7;
2060 
2061 	err = nfs_fs_proc_init();
2062 	if (err)
2063 		goto out6;
2064 
2065 	err = nfs_init_nfspagecache();
2066 	if (err)
2067 		goto out5;
2068 
2069 	err = nfs_init_inodecache();
2070 	if (err)
2071 		goto out4;
2072 
2073 	err = nfs_init_readpagecache();
2074 	if (err)
2075 		goto out3;
2076 
2077 	err = nfs_init_writepagecache();
2078 	if (err)
2079 		goto out2;
2080 
2081 	err = nfs_init_directcache();
2082 	if (err)
2083 		goto out1;
2084 
2085 	rpc_proc_register(&init_net, &nfs_rpcstat);
2086 
2087 	err = register_nfs_fs();
2088 	if (err)
2089 		goto out0;
2090 
2091 	return 0;
2092 out0:
2093 	rpc_proc_unregister(&init_net, "nfs");
2094 	nfs_destroy_directcache();
2095 out1:
2096 	nfs_destroy_writepagecache();
2097 out2:
2098 	nfs_destroy_readpagecache();
2099 out3:
2100 	nfs_destroy_inodecache();
2101 out4:
2102 	nfs_destroy_nfspagecache();
2103 out5:
2104 	nfs_fs_proc_exit();
2105 out6:
2106 	nfsiod_stop();
2107 out7:
2108 	nfs_fscache_unregister();
2109 out8:
2110 	unregister_pernet_subsys(&nfs_net_ops);
2111 out9:
2112 	return err;
2113 }
2114 
2115 static void __exit exit_nfs_fs(void)
2116 {
2117 	nfs_destroy_directcache();
2118 	nfs_destroy_writepagecache();
2119 	nfs_destroy_readpagecache();
2120 	nfs_destroy_inodecache();
2121 	nfs_destroy_nfspagecache();
2122 	nfs_fscache_unregister();
2123 	unregister_pernet_subsys(&nfs_net_ops);
2124 	rpc_proc_unregister(&init_net, "nfs");
2125 	unregister_nfs_fs();
2126 	nfs_fs_proc_exit();
2127 	nfsiod_stop();
2128 }
2129 
2130 /* Not quite true; I just maintain it */
2131 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2132 MODULE_LICENSE("GPL");
2133 module_param(enable_ino64, bool, 0644);
2134 
2135 module_init(init_nfs_fs)
2136 module_exit(exit_nfs_fs)
2137