xref: /openbmc/linux/fs/nfs/inode.c (revision 01a6e126)
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/signal.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 #include <linux/uaccess.h>
42 #include <linux/iversion.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, unsigned int mode)
89 {
90 	return nfs_wait_killable(mode);
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 int nfs_attribute_timeout(struct inode *inode)
164 {
165 	struct nfs_inode *nfsi = NFS_I(inode);
166 
167 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
168 }
169 
170 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
171 {
172 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
173 
174 	/* Special case for the pagecache or access cache */
175 	if (flags == NFS_INO_REVAL_PAGECACHE &&
176 	    !(cache_validity & NFS_INO_REVAL_FORCED))
177 		return false;
178 	return (cache_validity & flags) != 0;
179 }
180 
181 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
182 {
183 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
184 
185 	if ((cache_validity & flags) != 0)
186 		return true;
187 	if (nfs_attribute_timeout(inode))
188 		return true;
189 	return false;
190 }
191 
192 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
193 {
194 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
195 		return nfs_check_cache_invalid_delegated(inode, flags);
196 
197 	return nfs_check_cache_invalid_not_delegated(inode, flags);
198 }
199 
200 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
201 {
202 	struct nfs_inode *nfsi = NFS_I(inode);
203 
204 	if (inode->i_mapping->nrpages == 0)
205 		flags &= ~NFS_INO_INVALID_DATA;
206 	nfsi->cache_validity |= flags;
207 	if (flags & NFS_INO_INVALID_DATA)
208 		nfs_fscache_invalidate(inode);
209 }
210 
211 /*
212  * Invalidate the local caches
213  */
214 static void nfs_zap_caches_locked(struct inode *inode)
215 {
216 	struct nfs_inode *nfsi = NFS_I(inode);
217 	int mode = inode->i_mode;
218 
219 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
220 
221 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
222 	nfsi->attrtimeo_timestamp = jiffies;
223 
224 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
225 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
226 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
227 					| NFS_INO_INVALID_DATA
228 					| NFS_INO_INVALID_ACCESS
229 					| NFS_INO_INVALID_ACL
230 					| NFS_INO_REVAL_PAGECACHE);
231 	} else
232 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
233 					| NFS_INO_INVALID_ACCESS
234 					| NFS_INO_INVALID_ACL
235 					| NFS_INO_REVAL_PAGECACHE);
236 	nfs_zap_label_cache_locked(nfsi);
237 }
238 
239 void nfs_zap_caches(struct inode *inode)
240 {
241 	spin_lock(&inode->i_lock);
242 	nfs_zap_caches_locked(inode);
243 	spin_unlock(&inode->i_lock);
244 }
245 
246 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
247 {
248 	if (mapping->nrpages != 0) {
249 		spin_lock(&inode->i_lock);
250 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
251 		spin_unlock(&inode->i_lock);
252 	}
253 }
254 
255 void nfs_zap_acl_cache(struct inode *inode)
256 {
257 	void (*clear_acl_cache)(struct inode *);
258 
259 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
260 	if (clear_acl_cache != NULL)
261 		clear_acl_cache(inode);
262 	spin_lock(&inode->i_lock);
263 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
264 	spin_unlock(&inode->i_lock);
265 }
266 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
267 
268 void nfs_invalidate_atime(struct inode *inode)
269 {
270 	spin_lock(&inode->i_lock);
271 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
272 	spin_unlock(&inode->i_lock);
273 }
274 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
275 
276 /*
277  * Invalidate, but do not unhash, the inode.
278  * NB: must be called with inode->i_lock held!
279  */
280 static void nfs_invalidate_inode(struct inode *inode)
281 {
282 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
283 	nfs_zap_caches_locked(inode);
284 }
285 
286 struct nfs_find_desc {
287 	struct nfs_fh		*fh;
288 	struct nfs_fattr	*fattr;
289 };
290 
291 /*
292  * In NFSv3 we can have 64bit inode numbers. In order to support
293  * this, and re-exported directories (also seen in NFSv2)
294  * we are forced to allow 2 different inodes to have the same
295  * i_ino.
296  */
297 static int
298 nfs_find_actor(struct inode *inode, void *opaque)
299 {
300 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
301 	struct nfs_fh		*fh = desc->fh;
302 	struct nfs_fattr	*fattr = desc->fattr;
303 
304 	if (NFS_FILEID(inode) != fattr->fileid)
305 		return 0;
306 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
307 		return 0;
308 	if (nfs_compare_fh(NFS_FH(inode), fh))
309 		return 0;
310 	if (is_bad_inode(inode) || NFS_STALE(inode))
311 		return 0;
312 	return 1;
313 }
314 
315 static int
316 nfs_init_locked(struct inode *inode, void *opaque)
317 {
318 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
319 	struct nfs_fattr	*fattr = desc->fattr;
320 
321 	set_nfs_fileid(inode, fattr->fileid);
322 	inode->i_mode = fattr->mode;
323 	nfs_copy_fh(NFS_FH(inode), desc->fh);
324 	return 0;
325 }
326 
327 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
328 static void nfs_clear_label_invalid(struct inode *inode)
329 {
330 	spin_lock(&inode->i_lock);
331 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
332 	spin_unlock(&inode->i_lock);
333 }
334 
335 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
336 					struct nfs4_label *label)
337 {
338 	int error;
339 
340 	if (label == NULL)
341 		return;
342 
343 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
344 		error = security_inode_notifysecctx(inode, label->label,
345 				label->len);
346 		if (error)
347 			printk(KERN_ERR "%s() %s %d "
348 					"security_inode_notifysecctx() %d\n",
349 					__func__,
350 					(char *)label->label,
351 					label->len, error);
352 		nfs_clear_label_invalid(inode);
353 	}
354 }
355 
356 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
357 {
358 	struct nfs4_label *label = NULL;
359 	int minor_version = server->nfs_client->cl_minorversion;
360 
361 	if (minor_version < 2)
362 		return label;
363 
364 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
365 		return label;
366 
367 	label = kzalloc(sizeof(struct nfs4_label), flags);
368 	if (label == NULL)
369 		return ERR_PTR(-ENOMEM);
370 
371 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
372 	if (label->label == NULL) {
373 		kfree(label);
374 		return ERR_PTR(-ENOMEM);
375 	}
376 	label->len = NFS4_MAXLABELLEN;
377 
378 	return label;
379 }
380 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
381 #else
382 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
383 					struct nfs4_label *label)
384 {
385 }
386 #endif
387 EXPORT_SYMBOL_GPL(nfs_setsecurity);
388 
389 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
390 struct inode *
391 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
392 {
393 	struct nfs_find_desc desc = {
394 		.fh	= fh,
395 		.fattr	= fattr,
396 	};
397 	struct inode *inode;
398 	unsigned long hash;
399 
400 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
401 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
402 		return NULL;
403 
404 	hash = nfs_fattr_to_ino_t(fattr);
405 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
406 
407 	dprintk("%s: returning %p\n", __func__, inode);
408 	return inode;
409 }
410 
411 /*
412  * This is our front-end to iget that looks up inodes by file handle
413  * instead of inode number.
414  */
415 struct inode *
416 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
417 {
418 	struct nfs_find_desc desc = {
419 		.fh	= fh,
420 		.fattr	= fattr
421 	};
422 	struct inode *inode = ERR_PTR(-ENOENT);
423 	unsigned long hash;
424 
425 	nfs_attr_check_mountpoint(sb, fattr);
426 
427 	if (nfs_attr_use_mounted_on_fileid(fattr))
428 		fattr->fileid = fattr->mounted_on_fileid;
429 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
430 		goto out_no_inode;
431 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
432 		goto out_no_inode;
433 
434 	hash = nfs_fattr_to_ino_t(fattr);
435 
436 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
437 	if (inode == NULL) {
438 		inode = ERR_PTR(-ENOMEM);
439 		goto out_no_inode;
440 	}
441 
442 	if (inode->i_state & I_NEW) {
443 		struct nfs_inode *nfsi = NFS_I(inode);
444 		unsigned long now = jiffies;
445 
446 		/* We set i_ino for the few things that still rely on it,
447 		 * such as stat(2) */
448 		inode->i_ino = hash;
449 
450 		/* We can't support update_atime(), since the server will reset it */
451 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
452 		inode->i_mode = fattr->mode;
453 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
454 				&& nfs_server_capable(inode, NFS_CAP_MODE))
455 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
456 		/* Why so? Because we want revalidate for devices/FIFOs, and
457 		 * that's precisely what we have in nfs_file_inode_operations.
458 		 */
459 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
460 		if (S_ISREG(inode->i_mode)) {
461 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
462 			inode->i_data.a_ops = &nfs_file_aops;
463 		} else if (S_ISDIR(inode->i_mode)) {
464 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
465 			inode->i_fop = &nfs_dir_operations;
466 			inode->i_data.a_ops = &nfs_dir_aops;
467 			/* Deal with crossing mountpoints */
468 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
469 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
470 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
471 					inode->i_op = &nfs_referral_inode_operations;
472 				else
473 					inode->i_op = &nfs_mountpoint_inode_operations;
474 				inode->i_fop = NULL;
475 				inode->i_flags |= S_AUTOMOUNT;
476 			}
477 		} else if (S_ISLNK(inode->i_mode)) {
478 			inode->i_op = &nfs_symlink_inode_operations;
479 			inode_nohighmem(inode);
480 		} else
481 			init_special_inode(inode, inode->i_mode, fattr->rdev);
482 
483 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
484 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
485 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
486 		inode_set_iversion_raw(inode, 0);
487 		inode->i_size = 0;
488 		clear_nlink(inode);
489 		inode->i_uid = make_kuid(&init_user_ns, -2);
490 		inode->i_gid = make_kgid(&init_user_ns, -2);
491 		inode->i_blocks = 0;
492 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
493 		nfsi->write_io = 0;
494 		nfsi->read_io = 0;
495 
496 		nfsi->read_cache_jiffies = fattr->time_start;
497 		nfsi->attr_gencount = fattr->gencount;
498 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
499 			inode->i_atime = fattr->atime;
500 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
501 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
502 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
503 			inode->i_mtime = fattr->mtime;
504 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
505 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
506 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
507 			inode->i_ctime = fattr->ctime;
508 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
509 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
510 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
511 			inode_set_iversion_raw(inode, fattr->change_attr);
512 		else
513 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
514 				| NFS_INO_REVAL_PAGECACHE);
515 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
516 			inode->i_size = nfs_size_to_loff_t(fattr->size);
517 		else
518 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
519 				| NFS_INO_REVAL_PAGECACHE);
520 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
521 			set_nlink(inode, fattr->nlink);
522 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
523 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
524 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
525 			inode->i_uid = fattr->uid;
526 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
527 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
528 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
529 			inode->i_gid = fattr->gid;
530 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
531 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
532 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
533 			inode->i_blocks = fattr->du.nfs2.blocks;
534 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
535 			/*
536 			 * report the blocks in 512byte units
537 			 */
538 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
539 		}
540 
541 		nfs_setsecurity(inode, fattr, label);
542 
543 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
544 		nfsi->attrtimeo_timestamp = now;
545 		nfsi->access_cache = RB_ROOT;
546 
547 		nfs_fscache_init_inode(inode);
548 
549 		unlock_new_inode(inode);
550 	} else {
551 		int err = nfs_refresh_inode(inode, fattr);
552 		if (err < 0) {
553 			iput(inode);
554 			inode = ERR_PTR(err);
555 			goto out_no_inode;
556 		}
557 	}
558 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
559 		inode->i_sb->s_id,
560 		(unsigned long long)NFS_FILEID(inode),
561 		nfs_display_fhandle_hash(fh),
562 		atomic_read(&inode->i_count));
563 
564 out:
565 	return inode;
566 
567 out_no_inode:
568 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
569 	goto out;
570 }
571 EXPORT_SYMBOL_GPL(nfs_fhget);
572 
573 #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)
574 
575 int
576 nfs_setattr(struct dentry *dentry, struct iattr *attr)
577 {
578 	struct inode *inode = d_inode(dentry);
579 	struct nfs_fattr *fattr;
580 	int error = 0;
581 
582 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
583 
584 	/* skip mode change if it's just for clearing setuid/setgid */
585 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
586 		attr->ia_valid &= ~ATTR_MODE;
587 
588 	if (attr->ia_valid & ATTR_SIZE) {
589 		BUG_ON(!S_ISREG(inode->i_mode));
590 
591 		error = inode_newsize_ok(inode, attr->ia_size);
592 		if (error)
593 			return error;
594 
595 		if (attr->ia_size == i_size_read(inode))
596 			attr->ia_valid &= ~ATTR_SIZE;
597 	}
598 
599 	/* Optimization: if the end result is no change, don't RPC */
600 	attr->ia_valid &= NFS_VALID_ATTRS;
601 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
602 		return 0;
603 
604 	trace_nfs_setattr_enter(inode);
605 
606 	/* Write all dirty data */
607 	if (S_ISREG(inode->i_mode))
608 		nfs_sync_inode(inode);
609 
610 	fattr = nfs_alloc_fattr();
611 	if (fattr == NULL) {
612 		error = -ENOMEM;
613 		goto out;
614 	}
615 
616 	/*
617 	 * Return any delegations if we're going to change ACLs
618 	 */
619 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
620 		NFS_PROTO(inode)->return_delegation(inode);
621 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
622 	if (error == 0)
623 		error = nfs_refresh_inode(inode, fattr);
624 	nfs_free_fattr(fattr);
625 out:
626 	trace_nfs_setattr_exit(inode, error);
627 	return error;
628 }
629 EXPORT_SYMBOL_GPL(nfs_setattr);
630 
631 /**
632  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
633  * @inode: inode of the file used
634  * @offset: file offset to start truncating
635  *
636  * This is a copy of the common vmtruncate, but with the locking
637  * corrected to take into account the fact that NFS requires
638  * inode->i_size to be updated under the inode->i_lock.
639  * Note: must be called with inode->i_lock held!
640  */
641 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
642 {
643 	int err;
644 
645 	err = inode_newsize_ok(inode, offset);
646 	if (err)
647 		goto out;
648 
649 	i_size_write(inode, offset);
650 	/* Optimisation */
651 	if (offset == 0)
652 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
653 
654 	spin_unlock(&inode->i_lock);
655 	truncate_pagecache(inode, offset);
656 	spin_lock(&inode->i_lock);
657 out:
658 	return err;
659 }
660 
661 /**
662  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
663  * @inode: pointer to struct inode
664  * @attr: pointer to struct iattr
665  *
666  * Note: we do this in the *proc.c in order to ensure that
667  *       it works for things like exclusive creates too.
668  */
669 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
670 		struct nfs_fattr *fattr)
671 {
672 	/* Barrier: bump the attribute generation count. */
673 	nfs_fattr_set_barrier(fattr);
674 
675 	spin_lock(&inode->i_lock);
676 	NFS_I(inode)->attr_gencount = fattr->gencount;
677 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
678 		if ((attr->ia_valid & ATTR_MODE) != 0) {
679 			int mode = attr->ia_mode & S_IALLUGO;
680 			mode |= inode->i_mode & ~S_IALLUGO;
681 			inode->i_mode = mode;
682 		}
683 		if ((attr->ia_valid & ATTR_UID) != 0)
684 			inode->i_uid = attr->ia_uid;
685 		if ((attr->ia_valid & ATTR_GID) != 0)
686 			inode->i_gid = attr->ia_gid;
687 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
688 				| NFS_INO_INVALID_ACL);
689 	}
690 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
691 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
692 		nfs_vmtruncate(inode, attr->ia_size);
693 	}
694 	if (fattr->valid)
695 		nfs_update_inode(inode, fattr);
696 	else
697 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
698 	spin_unlock(&inode->i_lock);
699 }
700 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
701 
702 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
703 {
704 	struct dentry *parent;
705 
706 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
707 		return;
708 	parent = dget_parent(dentry);
709 	nfs_force_use_readdirplus(d_inode(parent));
710 	dput(parent);
711 }
712 
713 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
714 {
715 	struct dentry *parent;
716 
717 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
718 		return;
719 	parent = dget_parent(dentry);
720 	nfs_advise_use_readdirplus(d_inode(parent));
721 	dput(parent);
722 }
723 
724 static bool nfs_need_revalidate_inode(struct inode *inode)
725 {
726 	if (NFS_I(inode)->cache_validity &
727 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
728 		return true;
729 	if (nfs_attribute_cache_expired(inode))
730 		return true;
731 	return false;
732 }
733 
734 int nfs_getattr(const struct path *path, struct kstat *stat,
735 		u32 request_mask, unsigned int query_flags)
736 {
737 	struct inode *inode = d_inode(path->dentry);
738 	struct nfs_server *server = NFS_SERVER(inode);
739 	unsigned long cache_validity;
740 	int err = 0;
741 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
742 	bool do_update = false;
743 
744 	trace_nfs_getattr_enter(inode);
745 
746 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
747 		goto out_no_update;
748 
749 	/* Flush out writes to the server in order to update c/mtime.  */
750 	if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
751 			S_ISREG(inode->i_mode)) {
752 		err = filemap_write_and_wait(inode->i_mapping);
753 		if (err)
754 			goto out;
755 	}
756 
757 	/*
758 	 * We may force a getattr if the user cares about atime.
759 	 *
760 	 * Note that we only have to check the vfsmount flags here:
761 	 *  - NFS always sets S_NOATIME by so checking it would give a
762 	 *    bogus result
763 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
764 	 *    no point in checking those.
765 	 */
766 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
767 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
768 		request_mask &= ~STATX_ATIME;
769 
770 	/* Is the user requesting attributes that might need revalidation? */
771 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
772 					STATX_MTIME|STATX_UID|STATX_GID|
773 					STATX_SIZE|STATX_BLOCKS)))
774 		goto out_no_revalidate;
775 
776 	/* Check whether the cached attributes are stale */
777 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
778 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
779 	do_update |= cache_validity &
780 		(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
781 	if (request_mask & STATX_ATIME)
782 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
783 	if (request_mask & (STATX_CTIME|STATX_MTIME))
784 		do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
785 	if (do_update) {
786 		/* Update the attribute cache */
787 		if (!(server->flags & NFS_MOUNT_NOAC))
788 			nfs_readdirplus_parent_cache_miss(path->dentry);
789 		else
790 			nfs_readdirplus_parent_cache_hit(path->dentry);
791 		err = __nfs_revalidate_inode(server, inode);
792 		if (err)
793 			goto out;
794 	} else
795 		nfs_readdirplus_parent_cache_hit(path->dentry);
796 out_no_revalidate:
797 	/* Only return attributes that were revalidated. */
798 	stat->result_mask &= request_mask;
799 out_no_update:
800 	generic_fillattr(inode, stat);
801 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
802 	if (S_ISDIR(inode->i_mode))
803 		stat->blksize = NFS_SERVER(inode)->dtsize;
804 out:
805 	trace_nfs_getattr_exit(inode, err);
806 	return err;
807 }
808 EXPORT_SYMBOL_GPL(nfs_getattr);
809 
810 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
811 {
812 	refcount_set(&l_ctx->count, 1);
813 	l_ctx->lockowner = current->files;
814 	INIT_LIST_HEAD(&l_ctx->list);
815 	atomic_set(&l_ctx->io_count, 0);
816 }
817 
818 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
819 {
820 	struct nfs_lock_context *head = &ctx->lock_context;
821 	struct nfs_lock_context *pos = head;
822 
823 	do {
824 		if (pos->lockowner != current->files)
825 			continue;
826 		refcount_inc(&pos->count);
827 		return pos;
828 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
829 	return NULL;
830 }
831 
832 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
833 {
834 	struct nfs_lock_context *res, *new = NULL;
835 	struct inode *inode = d_inode(ctx->dentry);
836 
837 	spin_lock(&inode->i_lock);
838 	res = __nfs_find_lock_context(ctx);
839 	if (res == NULL) {
840 		spin_unlock(&inode->i_lock);
841 		new = kmalloc(sizeof(*new), GFP_KERNEL);
842 		if (new == NULL)
843 			return ERR_PTR(-ENOMEM);
844 		nfs_init_lock_context(new);
845 		spin_lock(&inode->i_lock);
846 		res = __nfs_find_lock_context(ctx);
847 		if (res == NULL) {
848 			list_add_tail(&new->list, &ctx->lock_context.list);
849 			new->open_context = ctx;
850 			res = new;
851 			new = NULL;
852 		}
853 	}
854 	spin_unlock(&inode->i_lock);
855 	kfree(new);
856 	return res;
857 }
858 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
859 
860 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
861 {
862 	struct nfs_open_context *ctx = l_ctx->open_context;
863 	struct inode *inode = d_inode(ctx->dentry);
864 
865 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
866 		return;
867 	list_del(&l_ctx->list);
868 	spin_unlock(&inode->i_lock);
869 	kfree(l_ctx);
870 }
871 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
872 
873 /**
874  * nfs_close_context - Common close_context() routine NFSv2/v3
875  * @ctx: pointer to context
876  * @is_sync: is this a synchronous close
877  *
878  * Ensure that the attributes are up to date if we're mounted
879  * with close-to-open semantics and we have cached data that will
880  * need to be revalidated on open.
881  */
882 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
883 {
884 	struct nfs_inode *nfsi;
885 	struct inode *inode;
886 	struct nfs_server *server;
887 
888 	if (!(ctx->mode & FMODE_WRITE))
889 		return;
890 	if (!is_sync)
891 		return;
892 	inode = d_inode(ctx->dentry);
893 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
894 		return;
895 	nfsi = NFS_I(inode);
896 	if (inode->i_mapping->nrpages == 0)
897 		return;
898 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
899 		return;
900 	if (!list_empty(&nfsi->open_files))
901 		return;
902 	server = NFS_SERVER(inode);
903 	if (server->flags & NFS_MOUNT_NOCTO)
904 		return;
905 	nfs_revalidate_inode(server, inode);
906 }
907 EXPORT_SYMBOL_GPL(nfs_close_context);
908 
909 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
910 						fmode_t f_mode,
911 						struct file *filp)
912 {
913 	struct nfs_open_context *ctx;
914 	struct rpc_cred *cred = rpc_lookup_cred();
915 	if (IS_ERR(cred))
916 		return ERR_CAST(cred);
917 
918 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
919 	if (!ctx) {
920 		put_rpccred(cred);
921 		return ERR_PTR(-ENOMEM);
922 	}
923 	nfs_sb_active(dentry->d_sb);
924 	ctx->dentry = dget(dentry);
925 	ctx->cred = cred;
926 	ctx->state = NULL;
927 	ctx->mode = f_mode;
928 	ctx->flags = 0;
929 	ctx->error = 0;
930 	ctx->flock_owner = (fl_owner_t)filp;
931 	nfs_init_lock_context(&ctx->lock_context);
932 	ctx->lock_context.open_context = ctx;
933 	INIT_LIST_HEAD(&ctx->list);
934 	ctx->mdsthreshold = NULL;
935 	return ctx;
936 }
937 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
938 
939 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
940 {
941 	if (ctx != NULL)
942 		refcount_inc(&ctx->lock_context.count);
943 	return ctx;
944 }
945 EXPORT_SYMBOL_GPL(get_nfs_open_context);
946 
947 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
948 {
949 	struct inode *inode = d_inode(ctx->dentry);
950 	struct super_block *sb = ctx->dentry->d_sb;
951 
952 	if (!list_empty(&ctx->list)) {
953 		if (!refcount_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
954 			return;
955 		list_del(&ctx->list);
956 		spin_unlock(&inode->i_lock);
957 	} else if (!refcount_dec_and_test(&ctx->lock_context.count))
958 		return;
959 	if (inode != NULL)
960 		NFS_PROTO(inode)->close_context(ctx, is_sync);
961 	if (ctx->cred != NULL)
962 		put_rpccred(ctx->cred);
963 	dput(ctx->dentry);
964 	nfs_sb_deactive(sb);
965 	kfree(ctx->mdsthreshold);
966 	kfree(ctx);
967 }
968 
969 void put_nfs_open_context(struct nfs_open_context *ctx)
970 {
971 	__put_nfs_open_context(ctx, 0);
972 }
973 EXPORT_SYMBOL_GPL(put_nfs_open_context);
974 
975 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
976 {
977 	__put_nfs_open_context(ctx, 1);
978 }
979 
980 /*
981  * Ensure that mmap has a recent RPC credential for use when writing out
982  * shared pages
983  */
984 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
985 {
986 	struct inode *inode = d_inode(ctx->dentry);
987 	struct nfs_inode *nfsi = NFS_I(inode);
988 
989 	spin_lock(&inode->i_lock);
990 	if (ctx->mode & FMODE_WRITE)
991 		list_add(&ctx->list, &nfsi->open_files);
992 	else
993 		list_add_tail(&ctx->list, &nfsi->open_files);
994 	spin_unlock(&inode->i_lock);
995 }
996 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
997 
998 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
999 {
1000 	filp->private_data = get_nfs_open_context(ctx);
1001 	if (list_empty(&ctx->list))
1002 		nfs_inode_attach_open_context(ctx);
1003 }
1004 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1005 
1006 /*
1007  * Given an inode, search for an open context with the desired characteristics
1008  */
1009 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
1010 {
1011 	struct nfs_inode *nfsi = NFS_I(inode);
1012 	struct nfs_open_context *pos, *ctx = NULL;
1013 
1014 	spin_lock(&inode->i_lock);
1015 	list_for_each_entry(pos, &nfsi->open_files, list) {
1016 		if (cred != NULL && pos->cred != cred)
1017 			continue;
1018 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1019 			continue;
1020 		ctx = get_nfs_open_context(pos);
1021 		break;
1022 	}
1023 	spin_unlock(&inode->i_lock);
1024 	return ctx;
1025 }
1026 
1027 void nfs_file_clear_open_context(struct file *filp)
1028 {
1029 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1030 
1031 	if (ctx) {
1032 		struct inode *inode = d_inode(ctx->dentry);
1033 
1034 		/*
1035 		 * We fatal error on write before. Try to writeback
1036 		 * every page again.
1037 		 */
1038 		if (ctx->error < 0)
1039 			invalidate_inode_pages2(inode->i_mapping);
1040 		filp->private_data = NULL;
1041 		spin_lock(&inode->i_lock);
1042 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1043 		spin_unlock(&inode->i_lock);
1044 		put_nfs_open_context_sync(ctx);
1045 	}
1046 }
1047 
1048 /*
1049  * These allocate and release file read/write context information.
1050  */
1051 int nfs_open(struct inode *inode, struct file *filp)
1052 {
1053 	struct nfs_open_context *ctx;
1054 
1055 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1056 	if (IS_ERR(ctx))
1057 		return PTR_ERR(ctx);
1058 	nfs_file_set_open_context(filp, ctx);
1059 	put_nfs_open_context(ctx);
1060 	nfs_fscache_open_file(inode, filp);
1061 	return 0;
1062 }
1063 
1064 /*
1065  * This function is called whenever some part of NFS notices that
1066  * the cached attributes have to be refreshed.
1067  */
1068 int
1069 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1070 {
1071 	int		 status = -ESTALE;
1072 	struct nfs4_label *label = NULL;
1073 	struct nfs_fattr *fattr = NULL;
1074 	struct nfs_inode *nfsi = NFS_I(inode);
1075 
1076 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1077 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1078 
1079 	trace_nfs_revalidate_inode_enter(inode);
1080 
1081 	if (is_bad_inode(inode))
1082 		goto out;
1083 	if (NFS_STALE(inode))
1084 		goto out;
1085 
1086 	/* pNFS: Attributes aren't updated until we layoutcommit */
1087 	if (S_ISREG(inode->i_mode)) {
1088 		status = pnfs_sync_inode(inode, false);
1089 		if (status)
1090 			goto out;
1091 	}
1092 
1093 	status = -ENOMEM;
1094 	fattr = nfs_alloc_fattr();
1095 	if (fattr == NULL)
1096 		goto out;
1097 
1098 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1099 
1100 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1101 	if (IS_ERR(label)) {
1102 		status = PTR_ERR(label);
1103 		goto out;
1104 	}
1105 
1106 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1107 	if (status != 0) {
1108 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1109 			 inode->i_sb->s_id,
1110 			 (unsigned long long)NFS_FILEID(inode), status);
1111 		if (status == -ESTALE) {
1112 			nfs_zap_caches(inode);
1113 			if (!S_ISDIR(inode->i_mode))
1114 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1115 		}
1116 		goto err_out;
1117 	}
1118 
1119 	status = nfs_refresh_inode(inode, fattr);
1120 	if (status) {
1121 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1122 			 inode->i_sb->s_id,
1123 			 (unsigned long long)NFS_FILEID(inode), status);
1124 		goto err_out;
1125 	}
1126 
1127 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1128 		nfs_zap_acl_cache(inode);
1129 
1130 	nfs_setsecurity(inode, fattr, label);
1131 
1132 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1133 		inode->i_sb->s_id,
1134 		(unsigned long long)NFS_FILEID(inode));
1135 
1136 err_out:
1137 	nfs4_label_free(label);
1138 out:
1139 	nfs_free_fattr(fattr);
1140 	trace_nfs_revalidate_inode_exit(inode, status);
1141 	return status;
1142 }
1143 
1144 int nfs_attribute_cache_expired(struct inode *inode)
1145 {
1146 	if (nfs_have_delegated_attributes(inode))
1147 		return 0;
1148 	return nfs_attribute_timeout(inode);
1149 }
1150 
1151 /**
1152  * nfs_revalidate_inode - Revalidate the inode attributes
1153  * @server - pointer to nfs_server struct
1154  * @inode - pointer to inode struct
1155  *
1156  * Updates inode attribute information by retrieving the data from the server.
1157  */
1158 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1159 {
1160 	if (!nfs_need_revalidate_inode(inode))
1161 		return NFS_STALE(inode) ? -ESTALE : 0;
1162 	return __nfs_revalidate_inode(server, inode);
1163 }
1164 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1165 
1166 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1167 {
1168 	struct nfs_inode *nfsi = NFS_I(inode);
1169 	int ret;
1170 
1171 	if (mapping->nrpages != 0) {
1172 		if (S_ISREG(inode->i_mode)) {
1173 			ret = nfs_sync_mapping(mapping);
1174 			if (ret < 0)
1175 				return ret;
1176 		}
1177 		ret = invalidate_inode_pages2(mapping);
1178 		if (ret < 0)
1179 			return ret;
1180 	}
1181 	if (S_ISDIR(inode->i_mode)) {
1182 		spin_lock(&inode->i_lock);
1183 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1184 		spin_unlock(&inode->i_lock);
1185 	}
1186 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1187 	nfs_fscache_wait_on_invalidate(inode);
1188 
1189 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1190 			inode->i_sb->s_id,
1191 			(unsigned long long)NFS_FILEID(inode));
1192 	return 0;
1193 }
1194 
1195 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1196 {
1197 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1198 		NFS_STALE(inode);
1199 }
1200 
1201 int nfs_revalidate_mapping_rcu(struct inode *inode)
1202 {
1203 	struct nfs_inode *nfsi = NFS_I(inode);
1204 	unsigned long *bitlock = &nfsi->flags;
1205 	int ret = 0;
1206 
1207 	if (IS_SWAPFILE(inode))
1208 		goto out;
1209 	if (nfs_mapping_need_revalidate_inode(inode)) {
1210 		ret = -ECHILD;
1211 		goto out;
1212 	}
1213 	spin_lock(&inode->i_lock);
1214 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1215 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1216 		ret = -ECHILD;
1217 	spin_unlock(&inode->i_lock);
1218 out:
1219 	return ret;
1220 }
1221 
1222 /**
1223  * nfs_revalidate_mapping - Revalidate the pagecache
1224  * @inode - pointer to host inode
1225  * @mapping - pointer to mapping
1226  */
1227 int nfs_revalidate_mapping(struct inode *inode,
1228 		struct address_space *mapping)
1229 {
1230 	struct nfs_inode *nfsi = NFS_I(inode);
1231 	unsigned long *bitlock = &nfsi->flags;
1232 	int ret = 0;
1233 
1234 	/* swapfiles are not supposed to be shared. */
1235 	if (IS_SWAPFILE(inode))
1236 		goto out;
1237 
1238 	if (nfs_mapping_need_revalidate_inode(inode)) {
1239 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1240 		if (ret < 0)
1241 			goto out;
1242 	}
1243 
1244 	/*
1245 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1246 	 * invalidations that come in while we're shooting down the mappings
1247 	 * are respected. But, that leaves a race window where one revalidator
1248 	 * can clear the flag, and then another checks it before the mapping
1249 	 * gets invalidated. Fix that by serializing access to this part of
1250 	 * the function.
1251 	 *
1252 	 * At the same time, we need to allow other tasks to see whether we
1253 	 * might be in the middle of invalidating the pages, so we only set
1254 	 * the bit lock here if it looks like we're going to be doing that.
1255 	 */
1256 	for (;;) {
1257 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1258 					 nfs_wait_bit_killable, TASK_KILLABLE);
1259 		if (ret)
1260 			goto out;
1261 		spin_lock(&inode->i_lock);
1262 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1263 			spin_unlock(&inode->i_lock);
1264 			continue;
1265 		}
1266 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1267 			break;
1268 		spin_unlock(&inode->i_lock);
1269 		goto out;
1270 	}
1271 
1272 	set_bit(NFS_INO_INVALIDATING, bitlock);
1273 	smp_wmb();
1274 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1275 	spin_unlock(&inode->i_lock);
1276 	trace_nfs_invalidate_mapping_enter(inode);
1277 	ret = nfs_invalidate_mapping(inode, mapping);
1278 	trace_nfs_invalidate_mapping_exit(inode, ret);
1279 
1280 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1281 	smp_mb__after_atomic();
1282 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1283 out:
1284 	return ret;
1285 }
1286 
1287 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1288 {
1289 	struct inode *inode = &nfsi->vfs_inode;
1290 
1291 	assert_spin_locked(&inode->i_lock);
1292 
1293 	if (!S_ISREG(inode->i_mode))
1294 		return false;
1295 	if (list_empty(&nfsi->open_files))
1296 		return false;
1297 	/* Note: This relies on nfsi->open_files being ordered with writers
1298 	 *       being placed at the head of the list.
1299 	 *       See nfs_inode_attach_open_context()
1300 	 */
1301 	return (list_first_entry(&nfsi->open_files,
1302 			struct nfs_open_context,
1303 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1304 }
1305 
1306 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1307 {
1308 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1309 }
1310 
1311 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1312 {
1313 	unsigned long ret = 0;
1314 
1315 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1316 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1317 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1318 		inode_set_iversion_raw(inode, fattr->change_attr);
1319 		if (S_ISDIR(inode->i_mode))
1320 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1321 		ret |= NFS_INO_INVALID_ATTR;
1322 	}
1323 	/* If we have atomic WCC data, we may update some attributes */
1324 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1325 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1326 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1327 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1328 		ret |= NFS_INO_INVALID_ATTR;
1329 	}
1330 
1331 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1332 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1333 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1334 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1335 		if (S_ISDIR(inode->i_mode))
1336 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1337 		ret |= NFS_INO_INVALID_ATTR;
1338 	}
1339 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1340 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1341 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1342 			&& !nfs_have_writebacks(inode)) {
1343 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1344 		ret |= NFS_INO_INVALID_ATTR;
1345 	}
1346 
1347 	return ret;
1348 }
1349 
1350 /**
1351  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1352  * @inode - pointer to inode
1353  * @fattr - updated attributes
1354  *
1355  * Verifies the attribute cache. If we have just changed the attributes,
1356  * so that fattr carries weak cache consistency data, then it may
1357  * also update the ctime/mtime/change_attribute.
1358  */
1359 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1360 {
1361 	struct nfs_inode *nfsi = NFS_I(inode);
1362 	loff_t cur_size, new_isize;
1363 	unsigned long invalid = 0;
1364 
1365 
1366 	if (nfs_have_delegated_attributes(inode))
1367 		return 0;
1368 	/* Has the inode gone and changed behind our back? */
1369 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1370 		return -ESTALE;
1371 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1372 		return -ESTALE;
1373 
1374 	if (!nfs_file_has_buffered_writers(nfsi)) {
1375 		/* Verify a few of the more important attributes */
1376 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1377 			invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1378 
1379 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1380 			invalid |= NFS_INO_INVALID_ATTR;
1381 
1382 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1383 			invalid |= NFS_INO_INVALID_ATTR;
1384 
1385 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1386 			cur_size = i_size_read(inode);
1387 			new_isize = nfs_size_to_loff_t(fattr->size);
1388 			if (cur_size != new_isize)
1389 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1390 		}
1391 	}
1392 
1393 	/* Have any file permissions changed? */
1394 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1395 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1396 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1397 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1398 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1399 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1400 
1401 	/* Has the link count changed? */
1402 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1403 		invalid |= NFS_INO_INVALID_ATTR;
1404 
1405 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1406 		invalid |= NFS_INO_INVALID_ATIME;
1407 
1408 	if (invalid != 0)
1409 		nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1410 
1411 	nfsi->read_cache_jiffies = fattr->time_start;
1412 	return 0;
1413 }
1414 
1415 static atomic_long_t nfs_attr_generation_counter;
1416 
1417 static unsigned long nfs_read_attr_generation_counter(void)
1418 {
1419 	return atomic_long_read(&nfs_attr_generation_counter);
1420 }
1421 
1422 unsigned long nfs_inc_attr_generation_counter(void)
1423 {
1424 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1425 }
1426 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1427 
1428 void nfs_fattr_init(struct nfs_fattr *fattr)
1429 {
1430 	fattr->valid = 0;
1431 	fattr->time_start = jiffies;
1432 	fattr->gencount = nfs_inc_attr_generation_counter();
1433 	fattr->owner_name = NULL;
1434 	fattr->group_name = NULL;
1435 }
1436 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1437 
1438 /**
1439  * nfs_fattr_set_barrier
1440  * @fattr: attributes
1441  *
1442  * Used to set a barrier after an attribute was updated. This
1443  * barrier ensures that older attributes from RPC calls that may
1444  * have raced with our update cannot clobber these new values.
1445  * Note that you are still responsible for ensuring that other
1446  * operations which change the attribute on the server do not
1447  * collide.
1448  */
1449 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1450 {
1451 	fattr->gencount = nfs_inc_attr_generation_counter();
1452 }
1453 
1454 struct nfs_fattr *nfs_alloc_fattr(void)
1455 {
1456 	struct nfs_fattr *fattr;
1457 
1458 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1459 	if (fattr != NULL)
1460 		nfs_fattr_init(fattr);
1461 	return fattr;
1462 }
1463 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1464 
1465 struct nfs_fh *nfs_alloc_fhandle(void)
1466 {
1467 	struct nfs_fh *fh;
1468 
1469 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1470 	if (fh != NULL)
1471 		fh->size = 0;
1472 	return fh;
1473 }
1474 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1475 
1476 #ifdef NFS_DEBUG
1477 /*
1478  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1479  *                             in the same way that wireshark does
1480  *
1481  * @fh: file handle
1482  *
1483  * For debugging only.
1484  */
1485 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1486 {
1487 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1488 	 * not on the result */
1489 	return nfs_fhandle_hash(fh);
1490 }
1491 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1492 
1493 /*
1494  * _nfs_display_fhandle - display an NFS file handle on the console
1495  *
1496  * @fh: file handle to display
1497  * @caption: display caption
1498  *
1499  * For debugging only.
1500  */
1501 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1502 {
1503 	unsigned short i;
1504 
1505 	if (fh == NULL || fh->size == 0) {
1506 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1507 		return;
1508 	}
1509 
1510 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1511 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1512 	for (i = 0; i < fh->size; i += 16) {
1513 		__be32 *pos = (__be32 *)&fh->data[i];
1514 
1515 		switch ((fh->size - i - 1) >> 2) {
1516 		case 0:
1517 			printk(KERN_DEFAULT " %08x\n",
1518 				be32_to_cpup(pos));
1519 			break;
1520 		case 1:
1521 			printk(KERN_DEFAULT " %08x %08x\n",
1522 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1523 			break;
1524 		case 2:
1525 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1526 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1527 				be32_to_cpup(pos + 2));
1528 			break;
1529 		default:
1530 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1531 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1532 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1533 		}
1534 	}
1535 }
1536 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1537 #endif
1538 
1539 /**
1540  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1541  * @inode - pointer to inode
1542  * @fattr - attributes
1543  *
1544  * Attempt to divine whether or not an RPC call reply carrying stale
1545  * attributes got scheduled after another call carrying updated ones.
1546  *
1547  * To do so, the function first assumes that a more recent ctime means
1548  * that the attributes in fattr are newer, however it also attempt to
1549  * catch the case where ctime either didn't change, or went backwards
1550  * (if someone reset the clock on the server) by looking at whether
1551  * or not this RPC call was started after the inode was last updated.
1552  * Note also the check for wraparound of 'attr_gencount'
1553  *
1554  * The function returns 'true' if it thinks the attributes in 'fattr' are
1555  * more recent than the ones cached in the inode.
1556  *
1557  */
1558 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1559 {
1560 	const struct nfs_inode *nfsi = NFS_I(inode);
1561 
1562 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1563 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1564 }
1565 
1566 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1567 {
1568 	int ret;
1569 
1570 	trace_nfs_refresh_inode_enter(inode);
1571 
1572 	if (nfs_inode_attrs_need_update(inode, fattr))
1573 		ret = nfs_update_inode(inode, fattr);
1574 	else
1575 		ret = nfs_check_inode_attributes(inode, fattr);
1576 
1577 	trace_nfs_refresh_inode_exit(inode, ret);
1578 	return ret;
1579 }
1580 
1581 /**
1582  * nfs_refresh_inode - try to update the inode attribute cache
1583  * @inode - pointer to inode
1584  * @fattr - updated attributes
1585  *
1586  * Check that an RPC call that returned attributes has not overlapped with
1587  * other recent updates of the inode metadata, then decide whether it is
1588  * safe to do a full update of the inode attributes, or whether just to
1589  * call nfs_check_inode_attributes.
1590  */
1591 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1592 {
1593 	int status;
1594 
1595 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1596 		return 0;
1597 	spin_lock(&inode->i_lock);
1598 	status = nfs_refresh_inode_locked(inode, fattr);
1599 	spin_unlock(&inode->i_lock);
1600 
1601 	return status;
1602 }
1603 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1604 
1605 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1606 {
1607 	unsigned long invalid = NFS_INO_INVALID_ATTR;
1608 
1609 	if (S_ISDIR(inode->i_mode))
1610 		invalid |= NFS_INO_INVALID_DATA;
1611 	nfs_set_cache_invalid(inode, invalid);
1612 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1613 		return 0;
1614 	return nfs_refresh_inode_locked(inode, fattr);
1615 }
1616 
1617 /**
1618  * nfs_post_op_update_inode - try to update the inode attribute cache
1619  * @inode - pointer to inode
1620  * @fattr - updated attributes
1621  *
1622  * After an operation that has changed the inode metadata, mark the
1623  * attribute cache as being invalid, then try to update it.
1624  *
1625  * NB: if the server didn't return any post op attributes, this
1626  * function will force the retrieval of attributes before the next
1627  * NFS request.  Thus it should be used only for operations that
1628  * are expected to change one or more attributes, to avoid
1629  * unnecessary NFS requests and trips through nfs_update_inode().
1630  */
1631 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1632 {
1633 	int status;
1634 
1635 	spin_lock(&inode->i_lock);
1636 	nfs_fattr_set_barrier(fattr);
1637 	status = nfs_post_op_update_inode_locked(inode, fattr);
1638 	spin_unlock(&inode->i_lock);
1639 
1640 	return status;
1641 }
1642 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1643 
1644 /**
1645  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1646  * @inode - pointer to inode
1647  * @fattr - updated attributes
1648  *
1649  * After an operation that has changed the inode metadata, mark the
1650  * attribute cache as being invalid, then try to update it. Fake up
1651  * weak cache consistency data, if none exist.
1652  *
1653  * This function is mainly designed to be used by the ->write_done() functions.
1654  */
1655 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1656 {
1657 	int status;
1658 
1659 	/* Don't do a WCC update if these attributes are already stale */
1660 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1661 			!nfs_inode_attrs_need_update(inode, fattr)) {
1662 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1663 				| NFS_ATTR_FATTR_PRESIZE
1664 				| NFS_ATTR_FATTR_PREMTIME
1665 				| NFS_ATTR_FATTR_PRECTIME);
1666 		goto out_noforce;
1667 	}
1668 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1669 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1670 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1671 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1672 	}
1673 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1674 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1675 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1676 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1677 	}
1678 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1679 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1680 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1681 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1682 	}
1683 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1684 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1685 		fattr->pre_size = i_size_read(inode);
1686 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1687 	}
1688 out_noforce:
1689 	status = nfs_post_op_update_inode_locked(inode, fattr);
1690 	return status;
1691 }
1692 
1693 /**
1694  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1695  * @inode - pointer to inode
1696  * @fattr - updated attributes
1697  *
1698  * After an operation that has changed the inode metadata, mark the
1699  * attribute cache as being invalid, then try to update it. Fake up
1700  * weak cache consistency data, if none exist.
1701  *
1702  * This function is mainly designed to be used by the ->write_done() functions.
1703  */
1704 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1705 {
1706 	int status;
1707 
1708 	spin_lock(&inode->i_lock);
1709 	nfs_fattr_set_barrier(fattr);
1710 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1711 	spin_unlock(&inode->i_lock);
1712 	return status;
1713 }
1714 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1715 
1716 
1717 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1718 				    struct nfs_fattr *fattr)
1719 {
1720 	bool ret1 = true, ret2 = true;
1721 
1722 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1723 		ret1 = (nfsi->fileid == fattr->fileid);
1724 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1725 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1726 	return ret1 || ret2;
1727 }
1728 
1729 /*
1730  * Many nfs protocol calls return the new file attributes after
1731  * an operation.  Here we update the inode to reflect the state
1732  * of the server's inode.
1733  *
1734  * This is a bit tricky because we have to make sure all dirty pages
1735  * have been sent off to the server before calling invalidate_inode_pages.
1736  * To make sure no other process adds more write requests while we try
1737  * our best to flush them, we make them sleep during the attribute refresh.
1738  *
1739  * A very similar scenario holds for the dir cache.
1740  */
1741 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1742 {
1743 	struct nfs_server *server;
1744 	struct nfs_inode *nfsi = NFS_I(inode);
1745 	loff_t cur_isize, new_isize;
1746 	unsigned long invalid = 0;
1747 	unsigned long now = jiffies;
1748 	unsigned long save_cache_validity;
1749 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1750 	bool cache_revalidated = true;
1751 
1752 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1753 			__func__, inode->i_sb->s_id, inode->i_ino,
1754 			nfs_display_fhandle_hash(NFS_FH(inode)),
1755 			atomic_read(&inode->i_count), fattr->valid);
1756 
1757 	if (!nfs_fileid_valid(nfsi, fattr)) {
1758 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1759 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1760 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1761 			inode->i_sb->s_id, (long long)nfsi->fileid,
1762 			(long long)fattr->fileid);
1763 		goto out_err;
1764 	}
1765 
1766 	/*
1767 	 * Make sure the inode's type hasn't changed.
1768 	 */
1769 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1770 		/*
1771 		* Big trouble! The inode has become a different object.
1772 		*/
1773 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1774 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1775 		goto out_err;
1776 	}
1777 
1778 	server = NFS_SERVER(inode);
1779 	/* Update the fsid? */
1780 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1781 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1782 			!IS_AUTOMOUNT(inode))
1783 		server->fsid = fattr->fsid;
1784 
1785 	/*
1786 	 * Update the read time so we don't revalidate too often.
1787 	 */
1788 	nfsi->read_cache_jiffies = fattr->time_start;
1789 
1790 	save_cache_validity = nfsi->cache_validity;
1791 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1792 			| NFS_INO_INVALID_ATIME
1793 			| NFS_INO_REVAL_FORCED
1794 			| NFS_INO_REVAL_PAGECACHE);
1795 
1796 	/* Do atomic weak cache consistency updates */
1797 	invalid |= nfs_wcc_update_inode(inode, fattr);
1798 
1799 	if (pnfs_layoutcommit_outstanding(inode)) {
1800 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1801 		cache_revalidated = false;
1802 	}
1803 
1804 	/* More cache consistency checks */
1805 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1806 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1807 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1808 					inode->i_sb->s_id, inode->i_ino);
1809 			/* Could it be a race with writeback? */
1810 			if (!have_writers) {
1811 				invalid |= NFS_INO_INVALID_ATTR
1812 					| NFS_INO_INVALID_DATA
1813 					| NFS_INO_INVALID_ACCESS
1814 					| NFS_INO_INVALID_ACL;
1815 				if (S_ISDIR(inode->i_mode))
1816 					nfs_force_lookup_revalidate(inode);
1817 			}
1818 			inode_set_iversion_raw(inode, fattr->change_attr);
1819 		}
1820 	} else {
1821 		nfsi->cache_validity |= save_cache_validity;
1822 		cache_revalidated = false;
1823 	}
1824 
1825 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1826 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1827 	} else if (server->caps & NFS_CAP_MTIME) {
1828 		nfsi->cache_validity |= save_cache_validity &
1829 				(NFS_INO_INVALID_ATTR
1830 				| NFS_INO_REVAL_FORCED);
1831 		cache_revalidated = false;
1832 	}
1833 
1834 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1835 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1836 	} else if (server->caps & NFS_CAP_CTIME) {
1837 		nfsi->cache_validity |= save_cache_validity &
1838 				(NFS_INO_INVALID_ATTR
1839 				| NFS_INO_REVAL_FORCED);
1840 		cache_revalidated = false;
1841 	}
1842 
1843 	/* Check if our cached file size is stale */
1844 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1845 		new_isize = nfs_size_to_loff_t(fattr->size);
1846 		cur_isize = i_size_read(inode);
1847 		if (new_isize != cur_isize) {
1848 			/* Do we perhaps have any outstanding writes, or has
1849 			 * the file grown beyond our last write? */
1850 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1851 				i_size_write(inode, new_isize);
1852 				if (!have_writers)
1853 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1854 			}
1855 			dprintk("NFS: isize change on server for file %s/%ld "
1856 					"(%Ld to %Ld)\n",
1857 					inode->i_sb->s_id,
1858 					inode->i_ino,
1859 					(long long)cur_isize,
1860 					(long long)new_isize);
1861 		}
1862 	} else {
1863 		nfsi->cache_validity |= save_cache_validity &
1864 				(NFS_INO_INVALID_ATTR
1865 				| NFS_INO_REVAL_PAGECACHE
1866 				| NFS_INO_REVAL_FORCED);
1867 		cache_revalidated = false;
1868 	}
1869 
1870 
1871 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1872 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1873 	else if (server->caps & NFS_CAP_ATIME) {
1874 		nfsi->cache_validity |= save_cache_validity &
1875 				(NFS_INO_INVALID_ATIME
1876 				| NFS_INO_REVAL_FORCED);
1877 		cache_revalidated = false;
1878 	}
1879 
1880 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1881 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1882 			umode_t newmode = inode->i_mode & S_IFMT;
1883 			newmode |= fattr->mode & S_IALLUGO;
1884 			inode->i_mode = newmode;
1885 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1886 		}
1887 	} else if (server->caps & NFS_CAP_MODE) {
1888 		nfsi->cache_validity |= save_cache_validity &
1889 				(NFS_INO_INVALID_ATTR
1890 				| NFS_INO_INVALID_ACCESS
1891 				| NFS_INO_INVALID_ACL
1892 				| NFS_INO_REVAL_FORCED);
1893 		cache_revalidated = false;
1894 	}
1895 
1896 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1897 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1898 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1899 			inode->i_uid = fattr->uid;
1900 		}
1901 	} else if (server->caps & NFS_CAP_OWNER) {
1902 		nfsi->cache_validity |= save_cache_validity &
1903 				(NFS_INO_INVALID_ATTR
1904 				| NFS_INO_INVALID_ACCESS
1905 				| NFS_INO_INVALID_ACL
1906 				| NFS_INO_REVAL_FORCED);
1907 		cache_revalidated = false;
1908 	}
1909 
1910 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1911 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1912 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1913 			inode->i_gid = fattr->gid;
1914 		}
1915 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1916 		nfsi->cache_validity |= save_cache_validity &
1917 				(NFS_INO_INVALID_ATTR
1918 				| NFS_INO_INVALID_ACCESS
1919 				| NFS_INO_INVALID_ACL
1920 				| NFS_INO_REVAL_FORCED);
1921 		cache_revalidated = false;
1922 	}
1923 
1924 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1925 		if (inode->i_nlink != fattr->nlink) {
1926 			invalid |= NFS_INO_INVALID_ATTR;
1927 			if (S_ISDIR(inode->i_mode))
1928 				invalid |= NFS_INO_INVALID_DATA;
1929 			set_nlink(inode, fattr->nlink);
1930 		}
1931 	} else if (server->caps & NFS_CAP_NLINK) {
1932 		nfsi->cache_validity |= save_cache_validity &
1933 				(NFS_INO_INVALID_ATTR
1934 				| NFS_INO_REVAL_FORCED);
1935 		cache_revalidated = false;
1936 	}
1937 
1938 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1939 		/*
1940 		 * report the blocks in 512byte units
1941 		 */
1942 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1943 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1944 		inode->i_blocks = fattr->du.nfs2.blocks;
1945 	else
1946 		cache_revalidated = false;
1947 
1948 	/* Update attrtimeo value if we're out of the unstable period */
1949 	if (invalid & NFS_INO_INVALID_ATTR) {
1950 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1951 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1952 		nfsi->attrtimeo_timestamp = now;
1953 		/* Set barrier to be more recent than all outstanding updates */
1954 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1955 	} else {
1956 		if (cache_revalidated) {
1957 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1958 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1959 				nfsi->attrtimeo <<= 1;
1960 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1961 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1962 			}
1963 			nfsi->attrtimeo_timestamp = now;
1964 		}
1965 		/* Set the barrier to be more recent than this fattr */
1966 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1967 			nfsi->attr_gencount = fattr->gencount;
1968 	}
1969 
1970 	/* Don't declare attrcache up to date if there were no attrs! */
1971 	if (cache_revalidated)
1972 		invalid &= ~NFS_INO_INVALID_ATTR;
1973 
1974 	/* Don't invalidate the data if we were to blame */
1975 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1976 				|| S_ISLNK(inode->i_mode)))
1977 		invalid &= ~NFS_INO_INVALID_DATA;
1978 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1979 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1980 		nfs_set_cache_invalid(inode, invalid);
1981 
1982 	return 0;
1983  out_err:
1984 	/*
1985 	 * No need to worry about unhashing the dentry, as the
1986 	 * lookup validation will know that the inode is bad.
1987 	 * (But we fall through to invalidate the caches.)
1988 	 */
1989 	nfs_invalidate_inode(inode);
1990 	return -ESTALE;
1991 }
1992 
1993 struct inode *nfs_alloc_inode(struct super_block *sb)
1994 {
1995 	struct nfs_inode *nfsi;
1996 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1997 	if (!nfsi)
1998 		return NULL;
1999 	nfsi->flags = 0UL;
2000 	nfsi->cache_validity = 0UL;
2001 #if IS_ENABLED(CONFIG_NFS_V4)
2002 	nfsi->nfs4_acl = NULL;
2003 #endif /* CONFIG_NFS_V4 */
2004 	return &nfsi->vfs_inode;
2005 }
2006 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2007 
2008 static void nfs_i_callback(struct rcu_head *head)
2009 {
2010 	struct inode *inode = container_of(head, struct inode, i_rcu);
2011 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2012 }
2013 
2014 void nfs_destroy_inode(struct inode *inode)
2015 {
2016 	call_rcu(&inode->i_rcu, nfs_i_callback);
2017 }
2018 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
2019 
2020 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2021 {
2022 #if IS_ENABLED(CONFIG_NFS_V4)
2023 	INIT_LIST_HEAD(&nfsi->open_states);
2024 	nfsi->delegation = NULL;
2025 	init_rwsem(&nfsi->rwsem);
2026 	nfsi->layout = NULL;
2027 #endif
2028 }
2029 
2030 static void init_once(void *foo)
2031 {
2032 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2033 
2034 	inode_init_once(&nfsi->vfs_inode);
2035 	INIT_LIST_HEAD(&nfsi->open_files);
2036 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2037 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2038 	INIT_LIST_HEAD(&nfsi->commit_info.list);
2039 	atomic_long_set(&nfsi->nrequests, 0);
2040 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
2041 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
2042 	init_rwsem(&nfsi->rmdir_sem);
2043 	mutex_init(&nfsi->commit_mutex);
2044 	nfs4_init_once(nfsi);
2045 }
2046 
2047 static int __init nfs_init_inodecache(void)
2048 {
2049 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2050 					     sizeof(struct nfs_inode),
2051 					     0, (SLAB_RECLAIM_ACCOUNT|
2052 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2053 					     init_once);
2054 	if (nfs_inode_cachep == NULL)
2055 		return -ENOMEM;
2056 
2057 	return 0;
2058 }
2059 
2060 static void nfs_destroy_inodecache(void)
2061 {
2062 	/*
2063 	 * Make sure all delayed rcu free inodes are flushed before we
2064 	 * destroy cache.
2065 	 */
2066 	rcu_barrier();
2067 	kmem_cache_destroy(nfs_inode_cachep);
2068 }
2069 
2070 struct workqueue_struct *nfsiod_workqueue;
2071 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2072 
2073 /*
2074  * start up the nfsiod workqueue
2075  */
2076 static int nfsiod_start(void)
2077 {
2078 	struct workqueue_struct *wq;
2079 	dprintk("RPC:       creating workqueue nfsiod\n");
2080 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2081 	if (wq == NULL)
2082 		return -ENOMEM;
2083 	nfsiod_workqueue = wq;
2084 	return 0;
2085 }
2086 
2087 /*
2088  * Destroy the nfsiod workqueue
2089  */
2090 static void nfsiod_stop(void)
2091 {
2092 	struct workqueue_struct *wq;
2093 
2094 	wq = nfsiod_workqueue;
2095 	if (wq == NULL)
2096 		return;
2097 	nfsiod_workqueue = NULL;
2098 	destroy_workqueue(wq);
2099 }
2100 
2101 unsigned int nfs_net_id;
2102 EXPORT_SYMBOL_GPL(nfs_net_id);
2103 
2104 static int nfs_net_init(struct net *net)
2105 {
2106 	nfs_clients_init(net);
2107 	return nfs_fs_proc_net_init(net);
2108 }
2109 
2110 static void nfs_net_exit(struct net *net)
2111 {
2112 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2113 
2114 	nfs_fs_proc_net_exit(net);
2115 	nfs_cleanup_cb_ident_idr(net);
2116 	WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2117 	WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2118 }
2119 
2120 static struct pernet_operations nfs_net_ops = {
2121 	.init = nfs_net_init,
2122 	.exit = nfs_net_exit,
2123 	.id   = &nfs_net_id,
2124 	.size = sizeof(struct nfs_net),
2125 };
2126 
2127 /*
2128  * Initialize NFS
2129  */
2130 static int __init init_nfs_fs(void)
2131 {
2132 	int err;
2133 
2134 	err = register_pernet_subsys(&nfs_net_ops);
2135 	if (err < 0)
2136 		goto out9;
2137 
2138 	err = nfs_fscache_register();
2139 	if (err < 0)
2140 		goto out8;
2141 
2142 	err = nfsiod_start();
2143 	if (err)
2144 		goto out7;
2145 
2146 	err = nfs_fs_proc_init();
2147 	if (err)
2148 		goto out6;
2149 
2150 	err = nfs_init_nfspagecache();
2151 	if (err)
2152 		goto out5;
2153 
2154 	err = nfs_init_inodecache();
2155 	if (err)
2156 		goto out4;
2157 
2158 	err = nfs_init_readpagecache();
2159 	if (err)
2160 		goto out3;
2161 
2162 	err = nfs_init_writepagecache();
2163 	if (err)
2164 		goto out2;
2165 
2166 	err = nfs_init_directcache();
2167 	if (err)
2168 		goto out1;
2169 
2170 	rpc_proc_register(&init_net, &nfs_rpcstat);
2171 
2172 	err = register_nfs_fs();
2173 	if (err)
2174 		goto out0;
2175 
2176 	return 0;
2177 out0:
2178 	rpc_proc_unregister(&init_net, "nfs");
2179 	nfs_destroy_directcache();
2180 out1:
2181 	nfs_destroy_writepagecache();
2182 out2:
2183 	nfs_destroy_readpagecache();
2184 out3:
2185 	nfs_destroy_inodecache();
2186 out4:
2187 	nfs_destroy_nfspagecache();
2188 out5:
2189 	nfs_fs_proc_exit();
2190 out6:
2191 	nfsiod_stop();
2192 out7:
2193 	nfs_fscache_unregister();
2194 out8:
2195 	unregister_pernet_subsys(&nfs_net_ops);
2196 out9:
2197 	return err;
2198 }
2199 
2200 static void __exit exit_nfs_fs(void)
2201 {
2202 	nfs_destroy_directcache();
2203 	nfs_destroy_writepagecache();
2204 	nfs_destroy_readpagecache();
2205 	nfs_destroy_inodecache();
2206 	nfs_destroy_nfspagecache();
2207 	nfs_fscache_unregister();
2208 	unregister_pernet_subsys(&nfs_net_ops);
2209 	rpc_proc_unregister(&init_net, "nfs");
2210 	unregister_nfs_fs();
2211 	nfs_fs_proc_exit();
2212 	nfsiod_stop();
2213 }
2214 
2215 /* Not quite true; I just maintain it */
2216 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2217 MODULE_LICENSE("GPL");
2218 module_param(enable_ino64, bool, 0644);
2219 
2220 module_init(init_nfs_fs)
2221 module_exit(exit_nfs_fs)
2222