• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/nfs/inode.c
4  *
5  *  Copyright (C) 1992  Rick Sladkey
6  *
7  *  nfs inode and superblock handling functions
8  *
9  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
11  *
12  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13  *  J.S.Peatfield@damtp.cam.ac.uk
14  *
15  */
16 
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
44 
45 #include "nfs4_fs.h"
46 #include "callback.h"
47 #include "delegation.h"
48 #include "iostat.h"
49 #include "internal.h"
50 #include "fscache.h"
51 #include "pnfs.h"
52 #include "nfs.h"
53 #include "netns.h"
54 #include "sysfs.h"
55 
56 #include "nfstrace.h"
57 
58 #define NFSDBG_FACILITY		NFSDBG_VFS
59 
60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
61 
62 /* Default is to see 64-bit inode numbers */
63 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64 
65 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 
67 static struct kmem_cache * nfs_inode_cachep;
68 
69 static inline unsigned long
nfs_fattr_to_ino_t(struct nfs_fattr * fattr)70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 {
72 	return nfs_fileid_to_ino_t(fattr->fileid);
73 }
74 
nfs_wait_bit_killable(struct wait_bit_key * key,int mode)75 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
76 {
77 	if (unlikely(nfs_current_task_exiting()))
78 		return -EINTR;
79 	schedule();
80 	if (signal_pending_state(mode, current))
81 		return -ERESTARTSYS;
82 	return 0;
83 }
84 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
85 
86 /**
87  * nfs_compat_user_ino64 - returns the user-visible inode number
88  * @fileid: 64-bit fileid
89  *
90  * This function returns a 32-bit inode number if the boot parameter
91  * nfs.enable_ino64 is zero.
92  */
nfs_compat_user_ino64(u64 fileid)93 u64 nfs_compat_user_ino64(u64 fileid)
94 {
95 #ifdef CONFIG_COMPAT
96 	compat_ulong_t ino;
97 #else
98 	unsigned long ino;
99 #endif
100 
101 	if (enable_ino64)
102 		return fileid;
103 	ino = fileid;
104 	if (sizeof(ino) < sizeof(fileid))
105 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
106 	return ino;
107 }
108 
nfs_drop_inode(struct inode * inode)109 int nfs_drop_inode(struct inode *inode)
110 {
111 	return NFS_STALE(inode) || generic_drop_inode(inode);
112 }
113 EXPORT_SYMBOL_GPL(nfs_drop_inode);
114 
nfs_clear_inode(struct inode * inode)115 void nfs_clear_inode(struct inode *inode)
116 {
117 	/*
118 	 * The following should never happen...
119 	 */
120 	WARN_ON_ONCE(nfs_have_writebacks(inode));
121 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
122 	nfs_zap_acl_cache(inode);
123 	nfs_access_zap_cache(inode);
124 	nfs_fscache_clear_inode(inode);
125 }
126 EXPORT_SYMBOL_GPL(nfs_clear_inode);
127 
nfs_evict_inode(struct inode * inode)128 void nfs_evict_inode(struct inode *inode)
129 {
130 	truncate_inode_pages_final(&inode->i_data);
131 	clear_inode(inode);
132 	nfs_clear_inode(inode);
133 }
134 
nfs_sync_inode(struct inode * inode)135 int nfs_sync_inode(struct inode *inode)
136 {
137 	inode_dio_wait(inode);
138 	return nfs_wb_all(inode);
139 }
140 EXPORT_SYMBOL_GPL(nfs_sync_inode);
141 
142 /**
143  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
144  * @mapping: pointer to struct address_space
145  */
nfs_sync_mapping(struct address_space * mapping)146 int nfs_sync_mapping(struct address_space *mapping)
147 {
148 	int ret = 0;
149 
150 	if (mapping->nrpages != 0) {
151 		unmap_mapping_range(mapping, 0, 0, 0);
152 		ret = nfs_wb_all(mapping->host);
153 	}
154 	return ret;
155 }
156 
nfs_attribute_timeout(struct inode * inode)157 static int nfs_attribute_timeout(struct inode *inode)
158 {
159 	struct nfs_inode *nfsi = NFS_I(inode);
160 
161 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
162 }
163 
nfs_check_cache_flags_invalid(struct inode * inode,unsigned long flags)164 static bool nfs_check_cache_flags_invalid(struct inode *inode,
165 					  unsigned long flags)
166 {
167 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
168 
169 	return (cache_validity & flags) != 0;
170 }
171 
nfs_check_cache_invalid(struct inode * inode,unsigned long flags)172 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
173 {
174 	if (nfs_check_cache_flags_invalid(inode, flags))
175 		return true;
176 	return nfs_attribute_cache_expired(inode);
177 }
178 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
179 
180 #ifdef CONFIG_NFS_V4_2
nfs_has_xattr_cache(const struct nfs_inode * nfsi)181 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
182 {
183 	return nfsi->xattr_cache != NULL;
184 }
185 #else
nfs_has_xattr_cache(const struct nfs_inode * nfsi)186 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
187 {
188 	return false;
189 }
190 #endif
191 
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)192 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
193 {
194 	struct nfs_inode *nfsi = NFS_I(inode);
195 	bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
196 
197 	if (have_delegation) {
198 		if (!(flags & NFS_INO_REVAL_FORCED))
199 			flags &= ~(NFS_INO_INVALID_MODE |
200 				   NFS_INO_INVALID_OTHER |
201 				   NFS_INO_INVALID_XATTR);
202 		flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
203 	}
204 
205 	if (!nfs_has_xattr_cache(nfsi))
206 		flags &= ~NFS_INO_INVALID_XATTR;
207 	if (flags & NFS_INO_INVALID_DATA)
208 		nfs_fscache_invalidate(inode, 0);
209 	flags &= ~NFS_INO_REVAL_FORCED;
210 
211 	flags |= nfsi->cache_validity;
212 	if (inode->i_mapping->nrpages == 0)
213 		flags &= ~NFS_INO_INVALID_DATA;
214 
215 	/* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */
216 	smp_store_release(&nfsi->cache_validity, flags);
217 
218 	if (inode->i_mapping->nrpages == 0 ||
219 	    nfsi->cache_validity & NFS_INO_INVALID_DATA) {
220 		nfs_ooo_clear(nfsi);
221 	}
222 	trace_nfs_set_cache_invalid(inode, 0);
223 }
224 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
225 
226 /*
227  * Invalidate the local caches
228  */
nfs_zap_caches_locked(struct inode * inode)229 static void nfs_zap_caches_locked(struct inode *inode)
230 {
231 	struct nfs_inode *nfsi = NFS_I(inode);
232 	int mode = inode->i_mode;
233 
234 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
235 
236 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
237 	nfsi->attrtimeo_timestamp = jiffies;
238 
239 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
240 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
241 						     NFS_INO_INVALID_DATA |
242 						     NFS_INO_INVALID_ACCESS |
243 						     NFS_INO_INVALID_ACL |
244 						     NFS_INO_INVALID_XATTR);
245 	else
246 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
247 						     NFS_INO_INVALID_ACCESS |
248 						     NFS_INO_INVALID_ACL |
249 						     NFS_INO_INVALID_XATTR);
250 	nfs_zap_label_cache_locked(nfsi);
251 }
252 
nfs_zap_caches(struct inode * inode)253 void nfs_zap_caches(struct inode *inode)
254 {
255 	spin_lock(&inode->i_lock);
256 	nfs_zap_caches_locked(inode);
257 	spin_unlock(&inode->i_lock);
258 }
259 
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)260 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
261 {
262 	if (mapping->nrpages != 0) {
263 		spin_lock(&inode->i_lock);
264 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
265 		spin_unlock(&inode->i_lock);
266 	}
267 }
268 
nfs_zap_acl_cache(struct inode * inode)269 void nfs_zap_acl_cache(struct inode *inode)
270 {
271 	void (*clear_acl_cache)(struct inode *);
272 
273 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
274 	if (clear_acl_cache != NULL)
275 		clear_acl_cache(inode);
276 	spin_lock(&inode->i_lock);
277 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
278 	spin_unlock(&inode->i_lock);
279 }
280 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
281 
nfs_invalidate_atime(struct inode * inode)282 void nfs_invalidate_atime(struct inode *inode)
283 {
284 	spin_lock(&inode->i_lock);
285 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
286 	spin_unlock(&inode->i_lock);
287 }
288 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
289 
290 /*
291  * Invalidate, but do not unhash, the inode.
292  * NB: must be called with inode->i_lock held!
293  */
nfs_set_inode_stale_locked(struct inode * inode)294 static void nfs_set_inode_stale_locked(struct inode *inode)
295 {
296 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
297 	nfs_zap_caches_locked(inode);
298 	trace_nfs_set_inode_stale(inode);
299 }
300 
nfs_set_inode_stale(struct inode * inode)301 void nfs_set_inode_stale(struct inode *inode)
302 {
303 	spin_lock(&inode->i_lock);
304 	nfs_set_inode_stale_locked(inode);
305 	spin_unlock(&inode->i_lock);
306 }
307 
308 struct nfs_find_desc {
309 	struct nfs_fh		*fh;
310 	struct nfs_fattr	*fattr;
311 };
312 
313 /*
314  * In NFSv3 we can have 64bit inode numbers. In order to support
315  * this, and re-exported directories (also seen in NFSv2)
316  * we are forced to allow 2 different inodes to have the same
317  * i_ino.
318  */
319 static int
nfs_find_actor(struct inode * inode,void * opaque)320 nfs_find_actor(struct inode *inode, void *opaque)
321 {
322 	struct nfs_find_desc	*desc = opaque;
323 	struct nfs_fh		*fh = desc->fh;
324 	struct nfs_fattr	*fattr = desc->fattr;
325 
326 	if (NFS_FILEID(inode) != fattr->fileid)
327 		return 0;
328 	if (inode_wrong_type(inode, fattr->mode))
329 		return 0;
330 	if (nfs_compare_fh(NFS_FH(inode), fh))
331 		return 0;
332 	if (is_bad_inode(inode) || NFS_STALE(inode))
333 		return 0;
334 	return 1;
335 }
336 
337 static int
nfs_init_locked(struct inode * inode,void * opaque)338 nfs_init_locked(struct inode *inode, void *opaque)
339 {
340 	struct nfs_find_desc	*desc = opaque;
341 	struct nfs_fattr	*fattr = desc->fattr;
342 
343 	set_nfs_fileid(inode, fattr->fileid);
344 	inode->i_mode = fattr->mode;
345 	nfs_copy_fh(NFS_FH(inode), desc->fh);
346 	return 0;
347 }
348 
349 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)350 static void nfs_clear_label_invalid(struct inode *inode)
351 {
352 	spin_lock(&inode->i_lock);
353 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
354 	spin_unlock(&inode->i_lock);
355 }
356 
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)357 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
358 {
359 	int error;
360 
361 	if (fattr->label == NULL)
362 		return;
363 
364 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
365 		error = security_inode_notifysecctx(inode, fattr->label->label,
366 				fattr->label->len);
367 		if (error)
368 			printk(KERN_ERR "%s() %s %d "
369 					"security_inode_notifysecctx() %d\n",
370 					__func__,
371 					(char *)fattr->label->label,
372 					fattr->label->len, error);
373 		nfs_clear_label_invalid(inode);
374 	}
375 }
376 
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)377 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
378 {
379 	struct nfs4_label *label;
380 
381 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
382 		return NULL;
383 
384 	label = kzalloc(sizeof(struct nfs4_label), flags);
385 	if (label == NULL)
386 		return ERR_PTR(-ENOMEM);
387 
388 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
389 	if (label->label == NULL) {
390 		kfree(label);
391 		return ERR_PTR(-ENOMEM);
392 	}
393 	label->len = NFS4_MAXLABELLEN;
394 
395 	return label;
396 }
397 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
398 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)399 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
400 {
401 }
402 #endif
403 EXPORT_SYMBOL_GPL(nfs_setsecurity);
404 
405 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
406 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)407 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
408 {
409 	struct nfs_find_desc desc = {
410 		.fh	= fh,
411 		.fattr	= fattr,
412 	};
413 	struct inode *inode;
414 	unsigned long hash;
415 
416 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
417 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
418 		return NULL;
419 
420 	hash = nfs_fattr_to_ino_t(fattr);
421 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
422 
423 	dprintk("%s: returning %p\n", __func__, inode);
424 	return inode;
425 }
426 
nfs_inode_init_regular(struct nfs_inode * nfsi)427 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
428 {
429 	atomic_long_set(&nfsi->nrequests, 0);
430 	atomic_long_set(&nfsi->redirtied_pages, 0);
431 	INIT_LIST_HEAD(&nfsi->commit_info.list);
432 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
433 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
434 	mutex_init(&nfsi->commit_mutex);
435 }
436 
nfs_inode_init_dir(struct nfs_inode * nfsi)437 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
438 {
439 	nfsi->cache_change_attribute = 0;
440 	memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
441 	init_rwsem(&nfsi->rmdir_sem);
442 }
443 
444 /*
445  * This is our front-end to iget that looks up inodes by file handle
446  * instead of inode number.
447  */
448 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr)449 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
450 {
451 	struct nfs_find_desc desc = {
452 		.fh	= fh,
453 		.fattr	= fattr
454 	};
455 	struct inode *inode = ERR_PTR(-ENOENT);
456 	u64 fattr_supported = NFS_SB(sb)->fattr_valid;
457 	unsigned long hash;
458 
459 	nfs_attr_check_mountpoint(sb, fattr);
460 
461 	if (nfs_attr_use_mounted_on_fileid(fattr))
462 		fattr->fileid = fattr->mounted_on_fileid;
463 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
464 		goto out_no_inode;
465 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
466 		goto out_no_inode;
467 
468 	hash = nfs_fattr_to_ino_t(fattr);
469 
470 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
471 	if (inode == NULL) {
472 		inode = ERR_PTR(-ENOMEM);
473 		goto out_no_inode;
474 	}
475 
476 	if (inode->i_state & I_NEW) {
477 		struct nfs_inode *nfsi = NFS_I(inode);
478 		unsigned long now = jiffies;
479 
480 		/* We set i_ino for the few things that still rely on it,
481 		 * such as stat(2) */
482 		inode->i_ino = hash;
483 
484 		/* We can't support update_atime(), since the server will reset it */
485 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
486 		inode->i_mode = fattr->mode;
487 		nfsi->cache_validity = 0;
488 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
489 				&& (fattr_supported & NFS_ATTR_FATTR_MODE))
490 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
491 		/* Why so? Because we want revalidate for devices/FIFOs, and
492 		 * that's precisely what we have in nfs_file_inode_operations.
493 		 */
494 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
495 		if (S_ISREG(inode->i_mode)) {
496 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
497 			inode->i_data.a_ops = &nfs_file_aops;
498 			nfs_inode_init_regular(nfsi);
499 		} else if (S_ISDIR(inode->i_mode)) {
500 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
501 			inode->i_fop = &nfs_dir_operations;
502 			inode->i_data.a_ops = &nfs_dir_aops;
503 			nfs_inode_init_dir(nfsi);
504 			/* Deal with crossing mountpoints */
505 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
506 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
507 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
508 					inode->i_op = &nfs_referral_inode_operations;
509 				else
510 					inode->i_op = &nfs_mountpoint_inode_operations;
511 				inode->i_fop = NULL;
512 				inode->i_flags |= S_AUTOMOUNT;
513 			}
514 		} else if (S_ISLNK(inode->i_mode)) {
515 			inode->i_op = &nfs_symlink_inode_operations;
516 			inode_nohighmem(inode);
517 		} else
518 			init_special_inode(inode, inode->i_mode, fattr->rdev);
519 
520 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
521 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
522 		inode_set_ctime(inode, 0, 0);
523 		inode_set_iversion_raw(inode, 0);
524 		inode->i_size = 0;
525 		clear_nlink(inode);
526 		inode->i_uid = make_kuid(&init_user_ns, -2);
527 		inode->i_gid = make_kgid(&init_user_ns, -2);
528 		inode->i_blocks = 0;
529 		nfsi->write_io = 0;
530 		nfsi->read_io = 0;
531 
532 		nfsi->read_cache_jiffies = fattr->time_start;
533 		nfsi->attr_gencount = fattr->gencount;
534 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
535 			inode->i_atime = fattr->atime;
536 		else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
537 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
538 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
539 			inode->i_mtime = fattr->mtime;
540 		else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
541 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
542 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
543 			inode_set_ctime_to_ts(inode, fattr->ctime);
544 		else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
545 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
546 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
547 			inode_set_iversion_raw(inode, fattr->change_attr);
548 		else
549 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
550 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
551 			inode->i_size = nfs_size_to_loff_t(fattr->size);
552 		else
553 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
554 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
555 			set_nlink(inode, fattr->nlink);
556 		else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
557 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
558 		else
559 			set_nlink(inode, 1);
560 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
561 			inode->i_uid = fattr->uid;
562 		else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
563 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
564 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
565 			inode->i_gid = fattr->gid;
566 		else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
567 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
568 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
569 			inode->i_blocks = fattr->du.nfs2.blocks;
570 		else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
571 			 fattr->size != 0)
572 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
573 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
574 			/*
575 			 * report the blocks in 512byte units
576 			 */
577 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
578 		} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
579 			   fattr->size != 0)
580 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
581 
582 		nfs_setsecurity(inode, fattr);
583 
584 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
585 		nfsi->attrtimeo_timestamp = now;
586 		nfsi->access_cache = RB_ROOT;
587 
588 		nfs_fscache_init_inode(inode);
589 
590 		unlock_new_inode(inode);
591 	} else {
592 		int err = nfs_refresh_inode(inode, fattr);
593 		if (err < 0) {
594 			iput(inode);
595 			inode = ERR_PTR(err);
596 			goto out_no_inode;
597 		}
598 	}
599 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
600 		inode->i_sb->s_id,
601 		(unsigned long long)NFS_FILEID(inode),
602 		nfs_display_fhandle_hash(fh),
603 		atomic_read(&inode->i_count));
604 
605 out:
606 	return inode;
607 
608 out_no_inode:
609 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
610 	goto out;
611 }
612 EXPORT_SYMBOL_GPL(nfs_fhget);
613 
614 #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)
615 
616 int
nfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)617 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
618 	    struct iattr *attr)
619 {
620 	struct inode *inode = d_inode(dentry);
621 	struct nfs_fattr *fattr;
622 	int error = 0;
623 
624 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
625 
626 	/* skip mode change if it's just for clearing setuid/setgid */
627 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
628 		attr->ia_valid &= ~ATTR_MODE;
629 
630 	if (attr->ia_valid & ATTR_SIZE) {
631 		BUG_ON(!S_ISREG(inode->i_mode));
632 
633 		error = inode_newsize_ok(inode, attr->ia_size);
634 		if (error)
635 			return error;
636 
637 		if (attr->ia_size == i_size_read(inode))
638 			attr->ia_valid &= ~ATTR_SIZE;
639 	}
640 
641 	/* Optimization: if the end result is no change, don't RPC */
642 	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
643 		return 0;
644 
645 	trace_nfs_setattr_enter(inode);
646 
647 	/* Write all dirty data */
648 	if (S_ISREG(inode->i_mode))
649 		nfs_sync_inode(inode);
650 
651 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
652 	if (fattr == NULL) {
653 		error = -ENOMEM;
654 		goto out;
655 	}
656 
657 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
658 	if (error == 0)
659 		error = nfs_refresh_inode(inode, fattr);
660 	nfs_free_fattr(fattr);
661 out:
662 	trace_nfs_setattr_exit(inode, error);
663 	return error;
664 }
665 EXPORT_SYMBOL_GPL(nfs_setattr);
666 
667 /**
668  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
669  * @inode: inode of the file used
670  * @offset: file offset to start truncating
671  *
672  * This is a copy of the common vmtruncate, but with the locking
673  * corrected to take into account the fact that NFS requires
674  * inode->i_size to be updated under the inode->i_lock.
675  * Note: must be called with inode->i_lock held!
676  */
nfs_vmtruncate(struct inode * inode,loff_t offset)677 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
678 {
679 	int err;
680 
681 	err = inode_newsize_ok(inode, offset);
682 	if (err)
683 		goto out;
684 
685 	trace_nfs_size_truncate(inode, offset);
686 	i_size_write(inode, offset);
687 	/* Optimisation */
688 	if (offset == 0) {
689 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
690 		nfs_ooo_clear(NFS_I(inode));
691 	}
692 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
693 
694 	spin_unlock(&inode->i_lock);
695 	truncate_pagecache(inode, offset);
696 	spin_lock(&inode->i_lock);
697 out:
698 	return err;
699 }
700 
701 /**
702  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
703  * @inode: pointer to struct inode
704  * @attr: pointer to struct iattr
705  * @fattr: pointer to struct nfs_fattr
706  *
707  * Note: we do this in the *proc.c in order to ensure that
708  *       it works for things like exclusive creates too.
709  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)710 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
711 		struct nfs_fattr *fattr)
712 {
713 	/* Barrier: bump the attribute generation count. */
714 	nfs_fattr_set_barrier(fattr);
715 
716 	spin_lock(&inode->i_lock);
717 	NFS_I(inode)->attr_gencount = fattr->gencount;
718 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
719 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
720 						     NFS_INO_INVALID_BLOCKS);
721 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
722 		nfs_vmtruncate(inode, attr->ia_size);
723 	}
724 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
725 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
726 		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
727 		    inode->i_mode & S_ISUID)
728 			inode->i_mode &= ~S_ISUID;
729 		if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
730 			inode->i_mode &= ~S_ISGID;
731 		if ((attr->ia_valid & ATTR_MODE) != 0) {
732 			int mode = attr->ia_mode & S_IALLUGO;
733 			mode |= inode->i_mode & ~S_IALLUGO;
734 			inode->i_mode = mode;
735 		}
736 		if ((attr->ia_valid & ATTR_UID) != 0)
737 			inode->i_uid = attr->ia_uid;
738 		if ((attr->ia_valid & ATTR_GID) != 0)
739 			inode->i_gid = attr->ia_gid;
740 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
741 			inode_set_ctime_to_ts(inode, fattr->ctime);
742 		else
743 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
744 					| NFS_INO_INVALID_CTIME);
745 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
746 				| NFS_INO_INVALID_ACL);
747 	}
748 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
749 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
750 				| NFS_INO_INVALID_CTIME);
751 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
752 			inode->i_atime = fattr->atime;
753 		else if (attr->ia_valid & ATTR_ATIME_SET)
754 			inode->i_atime = attr->ia_atime;
755 		else
756 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
757 
758 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
759 			inode_set_ctime_to_ts(inode, fattr->ctime);
760 		else
761 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
762 					| NFS_INO_INVALID_CTIME);
763 	}
764 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
765 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
766 				| NFS_INO_INVALID_CTIME);
767 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
768 			inode->i_mtime = fattr->mtime;
769 		else if (attr->ia_valid & ATTR_MTIME_SET)
770 			inode->i_mtime = attr->ia_mtime;
771 		else
772 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
773 
774 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
775 			inode_set_ctime_to_ts(inode, fattr->ctime);
776 		else
777 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
778 					| NFS_INO_INVALID_CTIME);
779 	}
780 	if (fattr->valid)
781 		nfs_update_inode(inode, fattr);
782 	spin_unlock(&inode->i_lock);
783 }
784 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
785 
786 /*
787  * Don't request help from readdirplus if the file is being written to,
788  * or if attribute caching is turned off
789  */
nfs_getattr_readdirplus_enable(const struct inode * inode)790 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
791 {
792 	return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
793 	       !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
794 }
795 
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)796 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
797 {
798 	if (!IS_ROOT(dentry)) {
799 		struct dentry *parent = dget_parent(dentry);
800 		nfs_readdir_record_entry_cache_miss(d_inode(parent));
801 		dput(parent);
802 	}
803 }
804 
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)805 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
806 {
807 	if (!IS_ROOT(dentry)) {
808 		struct dentry *parent = dget_parent(dentry);
809 		nfs_readdir_record_entry_cache_hit(d_inode(parent));
810 		dput(parent);
811 	}
812 }
813 
nfs_get_valid_attrmask(struct inode * inode)814 static u32 nfs_get_valid_attrmask(struct inode *inode)
815 {
816 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
817 	u32 reply_mask = STATX_INO | STATX_TYPE;
818 
819 	if (!(cache_validity & NFS_INO_INVALID_ATIME))
820 		reply_mask |= STATX_ATIME;
821 	if (!(cache_validity & NFS_INO_INVALID_CTIME))
822 		reply_mask |= STATX_CTIME;
823 	if (!(cache_validity & NFS_INO_INVALID_MTIME))
824 		reply_mask |= STATX_MTIME;
825 	if (!(cache_validity & NFS_INO_INVALID_SIZE))
826 		reply_mask |= STATX_SIZE;
827 	if (!(cache_validity & NFS_INO_INVALID_NLINK))
828 		reply_mask |= STATX_NLINK;
829 	if (!(cache_validity & NFS_INO_INVALID_MODE))
830 		reply_mask |= STATX_MODE;
831 	if (!(cache_validity & NFS_INO_INVALID_OTHER))
832 		reply_mask |= STATX_UID | STATX_GID;
833 	if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
834 		reply_mask |= STATX_BLOCKS;
835 	if (!(cache_validity & NFS_INO_INVALID_CHANGE))
836 		reply_mask |= STATX_CHANGE_COOKIE;
837 	return reply_mask;
838 }
839 
nfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)840 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
841 		struct kstat *stat, u32 request_mask, unsigned int query_flags)
842 {
843 	struct inode *inode = d_inode(path->dentry);
844 	struct nfs_server *server = NFS_SERVER(inode);
845 	unsigned long cache_validity;
846 	int err = 0;
847 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
848 	bool do_update = false;
849 	bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
850 
851 	trace_nfs_getattr_enter(inode);
852 
853 	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
854 			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
855 			STATX_INO | STATX_SIZE | STATX_BLOCKS |
856 			STATX_CHANGE_COOKIE;
857 
858 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
859 		if (readdirplus_enabled)
860 			nfs_readdirplus_parent_cache_hit(path->dentry);
861 		goto out_no_revalidate;
862 	}
863 
864 	/* Flush out writes to the server in order to update c/mtime/version.  */
865 	if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
866 	    S_ISREG(inode->i_mode))
867 		filemap_write_and_wait(inode->i_mapping);
868 
869 	/*
870 	 * We may force a getattr if the user cares about atime.
871 	 *
872 	 * Note that we only have to check the vfsmount flags here:
873 	 *  - NFS always sets S_NOATIME by so checking it would give a
874 	 *    bogus result
875 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
876 	 *    no point in checking those.
877 	 */
878 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
879 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
880 		request_mask &= ~STATX_ATIME;
881 
882 	/* Is the user requesting attributes that might need revalidation? */
883 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
884 					STATX_MTIME|STATX_UID|STATX_GID|
885 					STATX_SIZE|STATX_BLOCKS|
886 					STATX_CHANGE_COOKIE)))
887 		goto out_no_revalidate;
888 
889 	/* Check whether the cached attributes are stale */
890 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
891 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
892 	do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
893 	if (request_mask & STATX_ATIME)
894 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
895 	if (request_mask & STATX_CTIME)
896 		do_update |= cache_validity & NFS_INO_INVALID_CTIME;
897 	if (request_mask & STATX_MTIME)
898 		do_update |= cache_validity & NFS_INO_INVALID_MTIME;
899 	if (request_mask & STATX_SIZE)
900 		do_update |= cache_validity & NFS_INO_INVALID_SIZE;
901 	if (request_mask & STATX_NLINK)
902 		do_update |= cache_validity & NFS_INO_INVALID_NLINK;
903 	if (request_mask & STATX_MODE)
904 		do_update |= cache_validity & NFS_INO_INVALID_MODE;
905 	if (request_mask & (STATX_UID | STATX_GID))
906 		do_update |= cache_validity & NFS_INO_INVALID_OTHER;
907 	if (request_mask & STATX_BLOCKS)
908 		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
909 
910 	if (do_update) {
911 		if (readdirplus_enabled)
912 			nfs_readdirplus_parent_cache_miss(path->dentry);
913 		err = __nfs_revalidate_inode(server, inode);
914 		if (err)
915 			goto out;
916 	} else if (readdirplus_enabled)
917 		nfs_readdirplus_parent_cache_hit(path->dentry);
918 out_no_revalidate:
919 	/* Only return attributes that were revalidated. */
920 	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
921 
922 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
923 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
924 	stat->change_cookie = inode_peek_iversion_raw(inode);
925 	stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
926 	if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
927 		stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
928 	if (S_ISDIR(inode->i_mode))
929 		stat->blksize = NFS_SERVER(inode)->dtsize;
930 out:
931 	trace_nfs_getattr_exit(inode, err);
932 	return err;
933 }
934 EXPORT_SYMBOL_GPL(nfs_getattr);
935 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)936 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
937 {
938 	refcount_set(&l_ctx->count, 1);
939 	l_ctx->lockowner = current->files;
940 	INIT_LIST_HEAD(&l_ctx->list);
941 	atomic_set(&l_ctx->io_count, 0);
942 }
943 
__nfs_find_lock_context(struct nfs_open_context * ctx)944 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
945 {
946 	struct nfs_lock_context *pos;
947 
948 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
949 		if (pos->lockowner != current->files)
950 			continue;
951 		if (refcount_inc_not_zero(&pos->count))
952 			return pos;
953 	}
954 	return NULL;
955 }
956 
nfs_get_lock_context(struct nfs_open_context * ctx)957 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
958 {
959 	struct nfs_lock_context *res, *new = NULL;
960 	struct inode *inode = d_inode(ctx->dentry);
961 
962 	rcu_read_lock();
963 	res = __nfs_find_lock_context(ctx);
964 	rcu_read_unlock();
965 	if (res == NULL) {
966 		new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
967 		if (new == NULL)
968 			return ERR_PTR(-ENOMEM);
969 		nfs_init_lock_context(new);
970 		spin_lock(&inode->i_lock);
971 		res = __nfs_find_lock_context(ctx);
972 		if (res == NULL) {
973 			new->open_context = get_nfs_open_context(ctx);
974 			if (new->open_context) {
975 				list_add_tail_rcu(&new->list,
976 						&ctx->lock_context.list);
977 				res = new;
978 				new = NULL;
979 			} else
980 				res = ERR_PTR(-EBADF);
981 		}
982 		spin_unlock(&inode->i_lock);
983 		kfree(new);
984 	}
985 	return res;
986 }
987 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
988 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)989 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
990 {
991 	struct nfs_open_context *ctx = l_ctx->open_context;
992 	struct inode *inode = d_inode(ctx->dentry);
993 
994 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
995 		return;
996 	list_del_rcu(&l_ctx->list);
997 	spin_unlock(&inode->i_lock);
998 	put_nfs_open_context(ctx);
999 	kfree_rcu(l_ctx, rcu_head);
1000 }
1001 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
1002 
1003 /**
1004  * nfs_close_context - Common close_context() routine NFSv2/v3
1005  * @ctx: pointer to context
1006  * @is_sync: is this a synchronous close
1007  *
1008  * Ensure that the attributes are up to date if we're mounted
1009  * with close-to-open semantics and we have cached data that will
1010  * need to be revalidated on open.
1011  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)1012 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1013 {
1014 	struct nfs_inode *nfsi;
1015 	struct inode *inode;
1016 
1017 	if (!(ctx->mode & FMODE_WRITE))
1018 		return;
1019 	if (!is_sync)
1020 		return;
1021 	inode = d_inode(ctx->dentry);
1022 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1023 		return;
1024 	nfsi = NFS_I(inode);
1025 	if (inode->i_mapping->nrpages == 0)
1026 		return;
1027 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1028 		return;
1029 	if (!list_empty(&nfsi->open_files))
1030 		return;
1031 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1032 		return;
1033 	nfs_revalidate_inode(inode,
1034 			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1035 }
1036 EXPORT_SYMBOL_GPL(nfs_close_context);
1037 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1038 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1039 						fmode_t f_mode,
1040 						struct file *filp)
1041 {
1042 	struct nfs_open_context *ctx;
1043 
1044 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1045 	if (!ctx)
1046 		return ERR_PTR(-ENOMEM);
1047 	nfs_sb_active(dentry->d_sb);
1048 	ctx->dentry = dget(dentry);
1049 	if (filp)
1050 		ctx->cred = get_cred(filp->f_cred);
1051 	else
1052 		ctx->cred = get_current_cred();
1053 	rcu_assign_pointer(ctx->ll_cred, NULL);
1054 	ctx->state = NULL;
1055 	ctx->mode = f_mode;
1056 	ctx->flags = 0;
1057 	ctx->error = 0;
1058 	ctx->flock_owner = (fl_owner_t)filp;
1059 	nfs_init_lock_context(&ctx->lock_context);
1060 	ctx->lock_context.open_context = ctx;
1061 	INIT_LIST_HEAD(&ctx->list);
1062 	ctx->mdsthreshold = NULL;
1063 	return ctx;
1064 }
1065 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1066 
get_nfs_open_context(struct nfs_open_context * ctx)1067 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1068 {
1069 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1070 		return ctx;
1071 	return NULL;
1072 }
1073 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1074 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1075 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1076 {
1077 	struct inode *inode = d_inode(ctx->dentry);
1078 	struct super_block *sb = ctx->dentry->d_sb;
1079 
1080 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1081 		return;
1082 	if (!list_empty(&ctx->list)) {
1083 		spin_lock(&inode->i_lock);
1084 		list_del_rcu(&ctx->list);
1085 		spin_unlock(&inode->i_lock);
1086 	}
1087 	if (inode != NULL)
1088 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1089 	put_cred(ctx->cred);
1090 	dput(ctx->dentry);
1091 	nfs_sb_deactive(sb);
1092 	put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1093 	kfree(ctx->mdsthreshold);
1094 	kfree_rcu(ctx, rcu_head);
1095 }
1096 
put_nfs_open_context(struct nfs_open_context * ctx)1097 void put_nfs_open_context(struct nfs_open_context *ctx)
1098 {
1099 	__put_nfs_open_context(ctx, 0);
1100 }
1101 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1102 
put_nfs_open_context_sync(struct nfs_open_context * ctx)1103 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1104 {
1105 	__put_nfs_open_context(ctx, 1);
1106 }
1107 
1108 /*
1109  * Ensure that mmap has a recent RPC credential for use when writing out
1110  * shared pages
1111  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1112 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1113 {
1114 	struct inode *inode = d_inode(ctx->dentry);
1115 	struct nfs_inode *nfsi = NFS_I(inode);
1116 
1117 	spin_lock(&inode->i_lock);
1118 	if (list_empty(&nfsi->open_files) &&
1119 	    nfs_ooo_test(nfsi))
1120 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1121 						     NFS_INO_REVAL_FORCED);
1122 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1123 	spin_unlock(&inode->i_lock);
1124 }
1125 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1126 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1127 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1128 {
1129 	filp->private_data = get_nfs_open_context(ctx);
1130 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1131 	if (list_empty(&ctx->list))
1132 		nfs_inode_attach_open_context(ctx);
1133 }
1134 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1135 
1136 /*
1137  * Given an inode, search for an open context with the desired characteristics
1138  */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1139 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1140 {
1141 	struct nfs_inode *nfsi = NFS_I(inode);
1142 	struct nfs_open_context *pos, *ctx = NULL;
1143 
1144 	rcu_read_lock();
1145 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1146 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1147 			continue;
1148 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1149 			continue;
1150 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1151 			continue;
1152 		ctx = get_nfs_open_context(pos);
1153 		if (ctx)
1154 			break;
1155 	}
1156 	rcu_read_unlock();
1157 	return ctx;
1158 }
1159 
nfs_file_clear_open_context(struct file * filp)1160 void nfs_file_clear_open_context(struct file *filp)
1161 {
1162 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1163 
1164 	if (ctx) {
1165 		struct inode *inode = d_inode(ctx->dentry);
1166 
1167 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1168 		/*
1169 		 * We fatal error on write before. Try to writeback
1170 		 * every page again.
1171 		 */
1172 		if (ctx->error < 0)
1173 			invalidate_inode_pages2(inode->i_mapping);
1174 		filp->private_data = NULL;
1175 		put_nfs_open_context_sync(ctx);
1176 	}
1177 }
1178 
1179 /*
1180  * These allocate and release file read/write context information.
1181  */
nfs_open(struct inode * inode,struct file * filp)1182 int nfs_open(struct inode *inode, struct file *filp)
1183 {
1184 	struct nfs_open_context *ctx;
1185 
1186 	ctx = alloc_nfs_open_context(file_dentry(filp),
1187 				     flags_to_mode(filp->f_flags), filp);
1188 	if (IS_ERR(ctx))
1189 		return PTR_ERR(ctx);
1190 	nfs_file_set_open_context(filp, ctx);
1191 	put_nfs_open_context(ctx);
1192 	nfs_fscache_open_file(inode, filp);
1193 	return 0;
1194 }
1195 
1196 /*
1197  * This function is called whenever some part of NFS notices that
1198  * the cached attributes have to be refreshed.
1199  */
1200 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1201 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1202 {
1203 	int		 status = -ESTALE;
1204 	struct nfs_fattr *fattr = NULL;
1205 	struct nfs_inode *nfsi = NFS_I(inode);
1206 
1207 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1208 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1209 
1210 	trace_nfs_revalidate_inode_enter(inode);
1211 
1212 	if (is_bad_inode(inode))
1213 		goto out;
1214 	if (NFS_STALE(inode))
1215 		goto out;
1216 
1217 	/* pNFS: Attributes aren't updated until we layoutcommit */
1218 	if (S_ISREG(inode->i_mode)) {
1219 		status = pnfs_sync_inode(inode, false);
1220 		if (status)
1221 			goto out;
1222 	}
1223 
1224 	status = -ENOMEM;
1225 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1226 	if (fattr == NULL)
1227 		goto out;
1228 
1229 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1230 
1231 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1232 	if (status != 0) {
1233 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1234 			 inode->i_sb->s_id,
1235 			 (unsigned long long)NFS_FILEID(inode), status);
1236 		switch (status) {
1237 		case -ETIMEDOUT:
1238 			/* A soft timeout occurred. Use cached information? */
1239 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1240 				status = 0;
1241 			break;
1242 		case -ESTALE:
1243 			if (!S_ISDIR(inode->i_mode))
1244 				nfs_set_inode_stale(inode);
1245 			else
1246 				nfs_zap_caches(inode);
1247 		}
1248 		goto out;
1249 	}
1250 
1251 	status = nfs_refresh_inode(inode, fattr);
1252 	if (status) {
1253 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1254 			 inode->i_sb->s_id,
1255 			 (unsigned long long)NFS_FILEID(inode), status);
1256 		goto out;
1257 	}
1258 
1259 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1260 		nfs_zap_acl_cache(inode);
1261 
1262 	nfs_setsecurity(inode, fattr);
1263 
1264 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1265 		inode->i_sb->s_id,
1266 		(unsigned long long)NFS_FILEID(inode));
1267 
1268 out:
1269 	nfs_free_fattr(fattr);
1270 	trace_nfs_revalidate_inode_exit(inode, status);
1271 	return status;
1272 }
1273 
nfs_attribute_cache_expired(struct inode * inode)1274 int nfs_attribute_cache_expired(struct inode *inode)
1275 {
1276 	if (nfs_have_delegated_attributes(inode))
1277 		return 0;
1278 	return nfs_attribute_timeout(inode);
1279 }
1280 
1281 /**
1282  * nfs_revalidate_inode - Revalidate the inode attributes
1283  * @inode: pointer to inode struct
1284  * @flags: cache flags to check
1285  *
1286  * Updates inode attribute information by retrieving the data from the server.
1287  */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1288 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1289 {
1290 	if (!nfs_check_cache_invalid(inode, flags))
1291 		return NFS_STALE(inode) ? -ESTALE : 0;
1292 	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1293 }
1294 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1295 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1296 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1297 {
1298 	int ret;
1299 
1300 	nfs_fscache_invalidate(inode, 0);
1301 	if (mapping->nrpages != 0) {
1302 		if (S_ISREG(inode->i_mode)) {
1303 			ret = nfs_sync_mapping(mapping);
1304 			if (ret < 0)
1305 				return ret;
1306 		}
1307 		ret = invalidate_inode_pages2(mapping);
1308 		if (ret < 0)
1309 			return ret;
1310 	}
1311 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1312 
1313 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1314 			inode->i_sb->s_id,
1315 			(unsigned long long)NFS_FILEID(inode));
1316 	return 0;
1317 }
1318 
1319 /**
1320  * nfs_clear_invalid_mapping - Conditionally clear a mapping
1321  * @mapping: pointer to mapping
1322  *
1323  * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1324  */
nfs_clear_invalid_mapping(struct address_space * mapping)1325 int nfs_clear_invalid_mapping(struct address_space *mapping)
1326 {
1327 	struct inode *inode = mapping->host;
1328 	struct nfs_inode *nfsi = NFS_I(inode);
1329 	unsigned long *bitlock = &nfsi->flags;
1330 	int ret = 0;
1331 
1332 	/*
1333 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1334 	 * invalidations that come in while we're shooting down the mappings
1335 	 * are respected. But, that leaves a race window where one revalidator
1336 	 * can clear the flag, and then another checks it before the mapping
1337 	 * gets invalidated. Fix that by serializing access to this part of
1338 	 * the function.
1339 	 *
1340 	 * At the same time, we need to allow other tasks to see whether we
1341 	 * might be in the middle of invalidating the pages, so we only set
1342 	 * the bit lock here if it looks like we're going to be doing that.
1343 	 */
1344 	for (;;) {
1345 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1346 					 nfs_wait_bit_killable,
1347 					 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1348 		if (ret)
1349 			goto out;
1350 		smp_rmb(); /* pairs with smp_wmb() below */
1351 		if (test_bit(NFS_INO_INVALIDATING, bitlock))
1352 			continue;
1353 		/* pairs with nfs_set_cache_invalid()'s smp_store_release() */
1354 		if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA))
1355 			goto out;
1356 		/* Slow-path that double-checks with spinlock held */
1357 		spin_lock(&inode->i_lock);
1358 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1359 			spin_unlock(&inode->i_lock);
1360 			continue;
1361 		}
1362 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1363 			break;
1364 		spin_unlock(&inode->i_lock);
1365 		goto out;
1366 	}
1367 
1368 	set_bit(NFS_INO_INVALIDATING, bitlock);
1369 	smp_wmb();
1370 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1371 	nfs_ooo_clear(nfsi);
1372 	spin_unlock(&inode->i_lock);
1373 	trace_nfs_invalidate_mapping_enter(inode);
1374 	ret = nfs_invalidate_mapping(inode, mapping);
1375 	trace_nfs_invalidate_mapping_exit(inode, ret);
1376 
1377 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1378 	smp_mb__after_atomic();
1379 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1380 out:
1381 	return ret;
1382 }
1383 
nfs_mapping_need_revalidate_inode(struct inode * inode)1384 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1385 {
1386 	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1387 		NFS_STALE(inode);
1388 }
1389 
nfs_revalidate_mapping_rcu(struct inode * inode)1390 int nfs_revalidate_mapping_rcu(struct inode *inode)
1391 {
1392 	struct nfs_inode *nfsi = NFS_I(inode);
1393 	unsigned long *bitlock = &nfsi->flags;
1394 	int ret = 0;
1395 
1396 	if (IS_SWAPFILE(inode))
1397 		goto out;
1398 	if (nfs_mapping_need_revalidate_inode(inode)) {
1399 		ret = -ECHILD;
1400 		goto out;
1401 	}
1402 	spin_lock(&inode->i_lock);
1403 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1404 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1405 		ret = -ECHILD;
1406 	spin_unlock(&inode->i_lock);
1407 out:
1408 	return ret;
1409 }
1410 
1411 /**
1412  * nfs_revalidate_mapping - Revalidate the pagecache
1413  * @inode: pointer to host inode
1414  * @mapping: pointer to mapping
1415  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1416 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1417 {
1418 	/* swapfiles are not supposed to be shared. */
1419 	if (IS_SWAPFILE(inode))
1420 		return 0;
1421 
1422 	if (nfs_mapping_need_revalidate_inode(inode)) {
1423 		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1424 		if (ret < 0)
1425 			return ret;
1426 	}
1427 
1428 	return nfs_clear_invalid_mapping(mapping);
1429 }
1430 
nfs_file_has_writers(struct nfs_inode * nfsi)1431 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1432 {
1433 	struct inode *inode = &nfsi->vfs_inode;
1434 
1435 	if (!S_ISREG(inode->i_mode))
1436 		return false;
1437 	if (list_empty(&nfsi->open_files))
1438 		return false;
1439 	return inode_is_open_for_write(inode);
1440 }
1441 
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1442 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1443 {
1444 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1445 }
1446 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1447 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1448 {
1449 	struct timespec64 ts;
1450 
1451 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1452 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1453 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1454 		inode_set_iversion_raw(inode, fattr->change_attr);
1455 		if (S_ISDIR(inode->i_mode))
1456 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1457 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1458 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1459 	}
1460 	/* If we have atomic WCC data, we may update some attributes */
1461 	ts = inode_get_ctime(inode);
1462 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1463 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1464 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1465 		inode_set_ctime_to_ts(inode, fattr->ctime);
1466 	}
1467 
1468 	ts = inode->i_mtime;
1469 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1470 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1471 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1472 		inode->i_mtime = fattr->mtime;
1473 	}
1474 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1475 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1476 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1477 			&& !nfs_have_writebacks(inode)) {
1478 		trace_nfs_size_wcc(inode, fattr->size);
1479 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1480 	}
1481 }
1482 
1483 /**
1484  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1485  * @inode: pointer to inode
1486  * @fattr: updated attributes
1487  *
1488  * Verifies the attribute cache. If we have just changed the attributes,
1489  * so that fattr carries weak cache consistency data, then it may
1490  * also update the ctime/mtime/change_attribute.
1491  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1492 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1493 {
1494 	struct nfs_inode *nfsi = NFS_I(inode);
1495 	loff_t cur_size, new_isize;
1496 	unsigned long invalid = 0;
1497 	struct timespec64 ts;
1498 
1499 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1500 		return 0;
1501 
1502 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1503 		/* Only a mounted-on-fileid? Just exit */
1504 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1505 			return 0;
1506 	/* Has the inode gone and changed behind our back? */
1507 	} else if (nfsi->fileid != fattr->fileid) {
1508 		/* Is this perhaps the mounted-on fileid? */
1509 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1510 		    nfsi->fileid == fattr->mounted_on_fileid)
1511 			return 0;
1512 		return -ESTALE;
1513 	}
1514 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1515 		return -ESTALE;
1516 
1517 
1518 	if (!nfs_file_has_buffered_writers(nfsi)) {
1519 		/* Verify a few of the more important attributes */
1520 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1521 			invalid |= NFS_INO_INVALID_CHANGE;
1522 
1523 		ts = inode->i_mtime;
1524 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1525 			invalid |= NFS_INO_INVALID_MTIME;
1526 
1527 		ts = inode_get_ctime(inode);
1528 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1529 			invalid |= NFS_INO_INVALID_CTIME;
1530 
1531 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1532 			cur_size = i_size_read(inode);
1533 			new_isize = nfs_size_to_loff_t(fattr->size);
1534 			if (cur_size != new_isize)
1535 				invalid |= NFS_INO_INVALID_SIZE;
1536 		}
1537 	}
1538 
1539 	/* Have any file permissions changed? */
1540 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1541 		invalid |= NFS_INO_INVALID_MODE;
1542 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1543 		invalid |= NFS_INO_INVALID_OTHER;
1544 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1545 		invalid |= NFS_INO_INVALID_OTHER;
1546 
1547 	/* Has the link count changed? */
1548 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1549 		invalid |= NFS_INO_INVALID_NLINK;
1550 
1551 	ts = inode->i_atime;
1552 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1553 		invalid |= NFS_INO_INVALID_ATIME;
1554 
1555 	if (invalid != 0)
1556 		nfs_set_cache_invalid(inode, invalid);
1557 
1558 	nfsi->read_cache_jiffies = fattr->time_start;
1559 	return 0;
1560 }
1561 
1562 static atomic_long_t nfs_attr_generation_counter;
1563 
nfs_read_attr_generation_counter(void)1564 static unsigned long nfs_read_attr_generation_counter(void)
1565 {
1566 	return atomic_long_read(&nfs_attr_generation_counter);
1567 }
1568 
nfs_inc_attr_generation_counter(void)1569 unsigned long nfs_inc_attr_generation_counter(void)
1570 {
1571 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1572 }
1573 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1574 
nfs_fattr_init(struct nfs_fattr * fattr)1575 void nfs_fattr_init(struct nfs_fattr *fattr)
1576 {
1577 	fattr->valid = 0;
1578 	fattr->time_start = jiffies;
1579 	fattr->gencount = nfs_inc_attr_generation_counter();
1580 	fattr->owner_name = NULL;
1581 	fattr->group_name = NULL;
1582 	fattr->mdsthreshold = NULL;
1583 }
1584 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1585 
1586 /**
1587  * nfs_fattr_set_barrier
1588  * @fattr: attributes
1589  *
1590  * Used to set a barrier after an attribute was updated. This
1591  * barrier ensures that older attributes from RPC calls that may
1592  * have raced with our update cannot clobber these new values.
1593  * Note that you are still responsible for ensuring that other
1594  * operations which change the attribute on the server do not
1595  * collide.
1596  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1597 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1598 {
1599 	fattr->gencount = nfs_inc_attr_generation_counter();
1600 }
1601 
nfs_alloc_fattr(void)1602 struct nfs_fattr *nfs_alloc_fattr(void)
1603 {
1604 	struct nfs_fattr *fattr;
1605 
1606 	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1607 	if (fattr != NULL) {
1608 		nfs_fattr_init(fattr);
1609 		fattr->label = NULL;
1610 	}
1611 	return fattr;
1612 }
1613 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1614 
nfs_alloc_fattr_with_label(struct nfs_server * server)1615 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1616 {
1617 	struct nfs_fattr *fattr = nfs_alloc_fattr();
1618 
1619 	if (!fattr)
1620 		return NULL;
1621 
1622 	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1623 	if (IS_ERR(fattr->label)) {
1624 		kfree(fattr);
1625 		return NULL;
1626 	}
1627 
1628 	return fattr;
1629 }
1630 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1631 
nfs_alloc_fhandle(void)1632 struct nfs_fh *nfs_alloc_fhandle(void)
1633 {
1634 	struct nfs_fh *fh;
1635 
1636 	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1637 	if (fh != NULL)
1638 		fh->size = 0;
1639 	return fh;
1640 }
1641 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1642 
1643 #ifdef NFS_DEBUG
1644 /*
1645  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1646  *                             in the same way that wireshark does
1647  *
1648  * @fh: file handle
1649  *
1650  * For debugging only.
1651  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1652 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1653 {
1654 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1655 	 * not on the result */
1656 	return nfs_fhandle_hash(fh);
1657 }
1658 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1659 
1660 /*
1661  * _nfs_display_fhandle - display an NFS file handle on the console
1662  *
1663  * @fh: file handle to display
1664  * @caption: display caption
1665  *
1666  * For debugging only.
1667  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1668 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1669 {
1670 	unsigned short i;
1671 
1672 	if (fh == NULL || fh->size == 0) {
1673 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1674 		return;
1675 	}
1676 
1677 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1678 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1679 	for (i = 0; i < fh->size; i += 16) {
1680 		__be32 *pos = (__be32 *)&fh->data[i];
1681 
1682 		switch ((fh->size - i - 1) >> 2) {
1683 		case 0:
1684 			printk(KERN_DEFAULT " %08x\n",
1685 				be32_to_cpup(pos));
1686 			break;
1687 		case 1:
1688 			printk(KERN_DEFAULT " %08x %08x\n",
1689 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1690 			break;
1691 		case 2:
1692 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1693 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1694 				be32_to_cpup(pos + 2));
1695 			break;
1696 		default:
1697 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1698 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1699 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1700 		}
1701 	}
1702 }
1703 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1704 #endif
1705 
1706 /**
1707  * nfs_inode_attrs_cmp_generic - compare attributes
1708  * @fattr: attributes
1709  * @inode: pointer to inode
1710  *
1711  * Attempt to divine whether or not an RPC call reply carrying stale
1712  * attributes got scheduled after another call carrying updated ones.
1713  * Note also the check for wraparound of 'attr_gencount'
1714  *
1715  * The function returns '1' if it thinks the attributes in @fattr are
1716  * more recent than the ones cached in @inode. Otherwise it returns
1717  * the value '0'.
1718  */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1719 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1720 				       const struct inode *inode)
1721 {
1722 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1723 
1724 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1725 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1726 }
1727 
1728 /**
1729  * nfs_inode_attrs_cmp_monotonic - compare attributes
1730  * @fattr: attributes
1731  * @inode: pointer to inode
1732  *
1733  * Attempt to divine whether or not an RPC call reply carrying stale
1734  * attributes got scheduled after another call carrying updated ones.
1735  *
1736  * We assume that the server observes monotonic semantics for
1737  * the change attribute, so a larger value means that the attributes in
1738  * @fattr are more recent, in which case the function returns the
1739  * value '1'.
1740  * A return value of '0' indicates no measurable change
1741  * A return value of '-1' means that the attributes in @inode are
1742  * more recent.
1743  */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1744 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1745 					 const struct inode *inode)
1746 {
1747 	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1748 	if (diff > 0)
1749 		return 1;
1750 	return diff == 0 ? 0 : -1;
1751 }
1752 
1753 /**
1754  * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1755  * @fattr: attributes
1756  * @inode: pointer to inode
1757  *
1758  * Attempt to divine whether or not an RPC call reply carrying stale
1759  * attributes got scheduled after another call carrying updated ones.
1760  *
1761  * We assume that the server observes strictly monotonic semantics for
1762  * the change attribute, so a larger value means that the attributes in
1763  * @fattr are more recent, in which case the function returns the
1764  * value '1'.
1765  * A return value of '-1' means that the attributes in @inode are
1766  * more recent or unchanged.
1767  */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1768 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1769 						const struct inode *inode)
1770 {
1771 	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1772 }
1773 
1774 /**
1775  * nfs_inode_attrs_cmp - compare attributes
1776  * @fattr: attributes
1777  * @inode: pointer to inode
1778  *
1779  * This function returns '1' if it thinks the attributes in @fattr are
1780  * more recent than the ones cached in @inode. It returns '-1' if
1781  * the attributes in @inode are more recent than the ones in @fattr,
1782  * and it returns 0 if not sure.
1783  */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1784 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1785 			       const struct inode *inode)
1786 {
1787 	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1788 		return 1;
1789 	switch (NFS_SERVER(inode)->change_attr_type) {
1790 	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1791 		break;
1792 	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1793 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1794 			break;
1795 		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1796 	default:
1797 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1798 			break;
1799 		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1800 	}
1801 	return 0;
1802 }
1803 
1804 /**
1805  * nfs_inode_finish_partial_attr_update - complete a previous inode update
1806  * @fattr: attributes
1807  * @inode: pointer to inode
1808  *
1809  * Returns '1' if the last attribute update left the inode cached
1810  * attributes in a partially unrevalidated state, and @fattr
1811  * matches the change attribute of that partial update.
1812  * Otherwise returns '0'.
1813  */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1814 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1815 						const struct inode *inode)
1816 {
1817 	const unsigned long check_valid =
1818 		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1819 		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1820 		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1821 		NFS_INO_INVALID_NLINK;
1822 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1823 	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1824 
1825 	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1826 	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1827 	    (cache_validity & check_valid) != 0 &&
1828 	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1829 	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1830 		return 1;
1831 	return 0;
1832 }
1833 
nfs_ooo_merge(struct nfs_inode * nfsi,u64 start,u64 end)1834 static void nfs_ooo_merge(struct nfs_inode *nfsi,
1835 			  u64 start, u64 end)
1836 {
1837 	int i, cnt;
1838 
1839 	if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1840 		/* No point merging anything */
1841 		return;
1842 
1843 	if (!nfsi->ooo) {
1844 		nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1845 		if (!nfsi->ooo) {
1846 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1847 			return;
1848 		}
1849 		nfsi->ooo->cnt = 0;
1850 	}
1851 
1852 	/* add this range, merging if possible */
1853 	cnt = nfsi->ooo->cnt;
1854 	for (i = 0; i < cnt; i++) {
1855 		if (end == nfsi->ooo->gap[i].start)
1856 			end = nfsi->ooo->gap[i].end;
1857 		else if (start == nfsi->ooo->gap[i].end)
1858 			start = nfsi->ooo->gap[i].start;
1859 		else
1860 			continue;
1861 		/* Remove 'i' from table and loop to insert the new range */
1862 		cnt -= 1;
1863 		nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
1864 		i = -1;
1865 	}
1866 	if (start != end) {
1867 		if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
1868 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1869 			kfree(nfsi->ooo);
1870 			nfsi->ooo = NULL;
1871 			return;
1872 		}
1873 		nfsi->ooo->gap[cnt].start = start;
1874 		nfsi->ooo->gap[cnt].end = end;
1875 		cnt += 1;
1876 	}
1877 	nfsi->ooo->cnt = cnt;
1878 }
1879 
nfs_ooo_record(struct nfs_inode * nfsi,struct nfs_fattr * fattr)1880 static void nfs_ooo_record(struct nfs_inode *nfsi,
1881 			   struct nfs_fattr *fattr)
1882 {
1883 	/* This reply was out-of-order, so record in the
1884 	 * pre/post change id, possibly cancelling
1885 	 * gaps created when iversion was jumpped forward.
1886 	 */
1887 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
1888 	    (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
1889 		nfs_ooo_merge(nfsi,
1890 			      fattr->change_attr,
1891 			      fattr->pre_change_attr);
1892 }
1893 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1894 static int nfs_refresh_inode_locked(struct inode *inode,
1895 				    struct nfs_fattr *fattr)
1896 {
1897 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1898 	int ret = 0;
1899 
1900 	trace_nfs_refresh_inode_enter(inode);
1901 
1902 	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1903 		ret = nfs_update_inode(inode, fattr);
1904 	else {
1905 		nfs_ooo_record(NFS_I(inode), fattr);
1906 
1907 		if (attr_cmp == 0)
1908 			ret = nfs_check_inode_attributes(inode, fattr);
1909 	}
1910 
1911 	trace_nfs_refresh_inode_exit(inode, ret);
1912 	return ret;
1913 }
1914 
1915 /**
1916  * nfs_refresh_inode - try to update the inode attribute cache
1917  * @inode: pointer to inode
1918  * @fattr: updated attributes
1919  *
1920  * Check that an RPC call that returned attributes has not overlapped with
1921  * other recent updates of the inode metadata, then decide whether it is
1922  * safe to do a full update of the inode attributes, or whether just to
1923  * call nfs_check_inode_attributes.
1924  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1925 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1926 {
1927 	int status;
1928 
1929 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1930 		return 0;
1931 	spin_lock(&inode->i_lock);
1932 	status = nfs_refresh_inode_locked(inode, fattr);
1933 	spin_unlock(&inode->i_lock);
1934 
1935 	return status;
1936 }
1937 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1938 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1939 static int nfs_post_op_update_inode_locked(struct inode *inode,
1940 		struct nfs_fattr *fattr, unsigned int invalid)
1941 {
1942 	if (S_ISDIR(inode->i_mode))
1943 		invalid |= NFS_INO_INVALID_DATA;
1944 	nfs_set_cache_invalid(inode, invalid);
1945 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1946 		return 0;
1947 	return nfs_refresh_inode_locked(inode, fattr);
1948 }
1949 
1950 /**
1951  * nfs_post_op_update_inode - try to update the inode attribute cache
1952  * @inode: pointer to inode
1953  * @fattr: updated attributes
1954  *
1955  * After an operation that has changed the inode metadata, mark the
1956  * attribute cache as being invalid, then try to update it.
1957  *
1958  * NB: if the server didn't return any post op attributes, this
1959  * function will force the retrieval of attributes before the next
1960  * NFS request.  Thus it should be used only for operations that
1961  * are expected to change one or more attributes, to avoid
1962  * unnecessary NFS requests and trips through nfs_update_inode().
1963  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1964 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1965 {
1966 	int status;
1967 
1968 	spin_lock(&inode->i_lock);
1969 	nfs_fattr_set_barrier(fattr);
1970 	status = nfs_post_op_update_inode_locked(inode, fattr,
1971 			NFS_INO_INVALID_CHANGE
1972 			| NFS_INO_INVALID_CTIME
1973 			| NFS_INO_REVAL_FORCED);
1974 	spin_unlock(&inode->i_lock);
1975 
1976 	return status;
1977 }
1978 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1979 
1980 /**
1981  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1982  * @inode: pointer to inode
1983  * @fattr: updated attributes
1984  *
1985  * After an operation that has changed the inode metadata, mark the
1986  * attribute cache as being invalid, then try to update it. Fake up
1987  * weak cache consistency data, if none exist.
1988  *
1989  * This function is mainly designed to be used by the ->write_done() functions.
1990  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1991 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1992 {
1993 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1994 	int status;
1995 
1996 	/* Don't do a WCC update if these attributes are already stale */
1997 	if (attr_cmp < 0)
1998 		return 0;
1999 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
2000 		/* Record the pre/post change info before clearing PRECHANGE */
2001 		nfs_ooo_record(NFS_I(inode), fattr);
2002 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
2003 				| NFS_ATTR_FATTR_PRESIZE
2004 				| NFS_ATTR_FATTR_PREMTIME
2005 				| NFS_ATTR_FATTR_PRECTIME);
2006 		goto out_noforce;
2007 	}
2008 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2009 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2010 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2011 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2012 	}
2013 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2014 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2015 		fattr->pre_ctime = inode_get_ctime(inode);
2016 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2017 	}
2018 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2019 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2020 		fattr->pre_mtime = inode->i_mtime;
2021 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2022 	}
2023 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2024 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2025 		fattr->pre_size = i_size_read(inode);
2026 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2027 	}
2028 out_noforce:
2029 	status = nfs_post_op_update_inode_locked(inode, fattr,
2030 			NFS_INO_INVALID_CHANGE
2031 			| NFS_INO_INVALID_CTIME
2032 			| NFS_INO_INVALID_MTIME
2033 			| NFS_INO_INVALID_BLOCKS);
2034 	return status;
2035 }
2036 
2037 /**
2038  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2039  * @inode: pointer to inode
2040  * @fattr: updated attributes
2041  *
2042  * After an operation that has changed the inode metadata, mark the
2043  * attribute cache as being invalid, then try to update it. Fake up
2044  * weak cache consistency data, if none exist.
2045  *
2046  * This function is mainly designed to be used by the ->write_done() functions.
2047  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)2048 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2049 {
2050 	int status;
2051 
2052 	spin_lock(&inode->i_lock);
2053 	nfs_fattr_set_barrier(fattr);
2054 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2055 	spin_unlock(&inode->i_lock);
2056 	return status;
2057 }
2058 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2059 
2060 
2061 /*
2062  * Many nfs protocol calls return the new file attributes after
2063  * an operation.  Here we update the inode to reflect the state
2064  * of the server's inode.
2065  *
2066  * This is a bit tricky because we have to make sure all dirty pages
2067  * have been sent off to the server before calling invalidate_inode_pages.
2068  * To make sure no other process adds more write requests while we try
2069  * our best to flush them, we make them sleep during the attribute refresh.
2070  *
2071  * A very similar scenario holds for the dir cache.
2072  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)2073 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2074 {
2075 	struct nfs_server *server = NFS_SERVER(inode);
2076 	struct nfs_inode *nfsi = NFS_I(inode);
2077 	loff_t cur_isize, new_isize;
2078 	u64 fattr_supported = server->fattr_valid;
2079 	unsigned long invalid = 0;
2080 	unsigned long now = jiffies;
2081 	unsigned long save_cache_validity;
2082 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2083 	bool cache_revalidated = true;
2084 	bool attr_changed = false;
2085 	bool have_delegation;
2086 
2087 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2088 			__func__, inode->i_sb->s_id, inode->i_ino,
2089 			nfs_display_fhandle_hash(NFS_FH(inode)),
2090 			atomic_read(&inode->i_count), fattr->valid);
2091 
2092 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2093 		/* Only a mounted-on-fileid? Just exit */
2094 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2095 			return 0;
2096 	/* Has the inode gone and changed behind our back? */
2097 	} else if (nfsi->fileid != fattr->fileid) {
2098 		/* Is this perhaps the mounted-on fileid? */
2099 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2100 		    nfsi->fileid == fattr->mounted_on_fileid)
2101 			return 0;
2102 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2103 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2104 			NFS_SERVER(inode)->nfs_client->cl_hostname,
2105 			inode->i_sb->s_id, (long long)nfsi->fileid,
2106 			(long long)fattr->fileid);
2107 		goto out_err;
2108 	}
2109 
2110 	/*
2111 	 * Make sure the inode's type hasn't changed.
2112 	 */
2113 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2114 		/*
2115 		* Big trouble! The inode has become a different object.
2116 		*/
2117 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2118 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2119 		goto out_err;
2120 	}
2121 
2122 	/* Update the fsid? */
2123 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2124 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2125 			!IS_AUTOMOUNT(inode))
2126 		server->fsid = fattr->fsid;
2127 
2128 	/* Save the delegation state before clearing cache_validity */
2129 	have_delegation = nfs_have_delegated_attributes(inode);
2130 
2131 	/*
2132 	 * Update the read time so we don't revalidate too often.
2133 	 */
2134 	nfsi->read_cache_jiffies = fattr->time_start;
2135 
2136 	save_cache_validity = nfsi->cache_validity;
2137 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2138 			| NFS_INO_INVALID_ATIME
2139 			| NFS_INO_REVAL_FORCED
2140 			| NFS_INO_INVALID_BLOCKS);
2141 
2142 	/* Do atomic weak cache consistency updates */
2143 	nfs_wcc_update_inode(inode, fattr);
2144 
2145 	if (pnfs_layoutcommit_outstanding(inode)) {
2146 		nfsi->cache_validity |=
2147 			save_cache_validity &
2148 			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2149 			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2150 			 NFS_INO_INVALID_BLOCKS);
2151 		cache_revalidated = false;
2152 	}
2153 
2154 	/* More cache consistency checks */
2155 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2156 		if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2157 		    nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2158 			/* There is one remaining gap that hasn't been
2159 			 * merged into iversion - do that now.
2160 			 */
2161 			inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2162 			kfree(nfsi->ooo);
2163 			nfsi->ooo = NULL;
2164 		}
2165 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2166 			/* Could it be a race with writeback? */
2167 			if (!(have_writers || have_delegation)) {
2168 				invalid |= NFS_INO_INVALID_DATA
2169 					| NFS_INO_INVALID_ACCESS
2170 					| NFS_INO_INVALID_ACL
2171 					| NFS_INO_INVALID_XATTR;
2172 				/* Force revalidate of all attributes */
2173 				save_cache_validity |= NFS_INO_INVALID_CTIME
2174 					| NFS_INO_INVALID_MTIME
2175 					| NFS_INO_INVALID_SIZE
2176 					| NFS_INO_INVALID_BLOCKS
2177 					| NFS_INO_INVALID_NLINK
2178 					| NFS_INO_INVALID_MODE
2179 					| NFS_INO_INVALID_OTHER;
2180 				if (S_ISDIR(inode->i_mode))
2181 					nfs_force_lookup_revalidate(inode);
2182 				attr_changed = true;
2183 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2184 						inode->i_sb->s_id,
2185 						inode->i_ino);
2186 			} else if (!have_delegation) {
2187 				nfs_ooo_record(nfsi, fattr);
2188 				nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2189 					      fattr->change_attr);
2190 			}
2191 			inode_set_iversion_raw(inode, fattr->change_attr);
2192 		}
2193 	} else {
2194 		nfsi->cache_validity |=
2195 			save_cache_validity & NFS_INO_INVALID_CHANGE;
2196 		if (!have_delegation ||
2197 		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2198 			cache_revalidated = false;
2199 	}
2200 
2201 	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2202 		inode->i_mtime = fattr->mtime;
2203 	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2204 		nfsi->cache_validity |=
2205 			save_cache_validity & NFS_INO_INVALID_MTIME;
2206 
2207 	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2208 		inode_set_ctime_to_ts(inode, fattr->ctime);
2209 	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2210 		nfsi->cache_validity |=
2211 			save_cache_validity & NFS_INO_INVALID_CTIME;
2212 
2213 	/* Check if our cached file size is stale */
2214 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2215 		new_isize = nfs_size_to_loff_t(fattr->size);
2216 		cur_isize = i_size_read(inode);
2217 		if (new_isize != cur_isize && !have_delegation) {
2218 			/* Do we perhaps have any outstanding writes, or has
2219 			 * the file grown beyond our last write? */
2220 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2221 				trace_nfs_size_update(inode, new_isize);
2222 				i_size_write(inode, new_isize);
2223 				if (!have_writers)
2224 					invalid |= NFS_INO_INVALID_DATA;
2225 			}
2226 		}
2227 		if (new_isize == 0 &&
2228 		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2229 				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2230 			fattr->du.nfs3.used = 0;
2231 			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2232 		}
2233 	} else
2234 		nfsi->cache_validity |=
2235 			save_cache_validity & NFS_INO_INVALID_SIZE;
2236 
2237 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2238 		inode->i_atime = fattr->atime;
2239 	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2240 		nfsi->cache_validity |=
2241 			save_cache_validity & NFS_INO_INVALID_ATIME;
2242 
2243 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2244 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2245 			umode_t newmode = inode->i_mode & S_IFMT;
2246 			newmode |= fattr->mode & S_IALLUGO;
2247 			inode->i_mode = newmode;
2248 			invalid |= NFS_INO_INVALID_ACCESS
2249 				| NFS_INO_INVALID_ACL;
2250 		}
2251 	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2252 		nfsi->cache_validity |=
2253 			save_cache_validity & NFS_INO_INVALID_MODE;
2254 
2255 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2256 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2257 			invalid |= NFS_INO_INVALID_ACCESS
2258 				| NFS_INO_INVALID_ACL;
2259 			inode->i_uid = fattr->uid;
2260 		}
2261 	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2262 		nfsi->cache_validity |=
2263 			save_cache_validity & NFS_INO_INVALID_OTHER;
2264 
2265 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2266 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2267 			invalid |= NFS_INO_INVALID_ACCESS
2268 				| NFS_INO_INVALID_ACL;
2269 			inode->i_gid = fattr->gid;
2270 		}
2271 	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2272 		nfsi->cache_validity |=
2273 			save_cache_validity & NFS_INO_INVALID_OTHER;
2274 
2275 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2276 		if (inode->i_nlink != fattr->nlink)
2277 			set_nlink(inode, fattr->nlink);
2278 	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2279 		nfsi->cache_validity |=
2280 			save_cache_validity & NFS_INO_INVALID_NLINK;
2281 
2282 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2283 		/*
2284 		 * report the blocks in 512byte units
2285 		 */
2286 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2287 	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2288 		nfsi->cache_validity |=
2289 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2290 
2291 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2292 		inode->i_blocks = fattr->du.nfs2.blocks;
2293 	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2294 		nfsi->cache_validity |=
2295 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2296 
2297 	/* Update attrtimeo value if we're out of the unstable period */
2298 	if (attr_changed) {
2299 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2300 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2301 		nfsi->attrtimeo_timestamp = now;
2302 		/* Set barrier to be more recent than all outstanding updates */
2303 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2304 	} else {
2305 		if (cache_revalidated) {
2306 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2307 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2308 				nfsi->attrtimeo <<= 1;
2309 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2310 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2311 			}
2312 			nfsi->attrtimeo_timestamp = now;
2313 		}
2314 		/* Set the barrier to be more recent than this fattr */
2315 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2316 			nfsi->attr_gencount = fattr->gencount;
2317 	}
2318 
2319 	/* Don't invalidate the data if we were to blame */
2320 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2321 				|| S_ISLNK(inode->i_mode)))
2322 		invalid &= ~NFS_INO_INVALID_DATA;
2323 	nfs_set_cache_invalid(inode, invalid);
2324 
2325 	return 0;
2326  out_err:
2327 	/*
2328 	 * No need to worry about unhashing the dentry, as the
2329 	 * lookup validation will know that the inode is bad.
2330 	 * (But we fall through to invalidate the caches.)
2331 	 */
2332 	nfs_set_inode_stale_locked(inode);
2333 	return -ESTALE;
2334 }
2335 
nfs_alloc_inode(struct super_block * sb)2336 struct inode *nfs_alloc_inode(struct super_block *sb)
2337 {
2338 	struct nfs_inode *nfsi;
2339 	nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2340 	if (!nfsi)
2341 		return NULL;
2342 	nfsi->flags = 0UL;
2343 	nfsi->cache_validity = 0UL;
2344 	nfsi->ooo = NULL;
2345 #if IS_ENABLED(CONFIG_NFS_V4)
2346 	nfsi->nfs4_acl = NULL;
2347 #endif /* CONFIG_NFS_V4 */
2348 #ifdef CONFIG_NFS_V4_2
2349 	nfsi->xattr_cache = NULL;
2350 #endif
2351 	nfs_netfs_inode_init(nfsi);
2352 
2353 	return &nfsi->vfs_inode;
2354 }
2355 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2356 
nfs_free_inode(struct inode * inode)2357 void nfs_free_inode(struct inode *inode)
2358 {
2359 	kfree(NFS_I(inode)->ooo);
2360 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2361 }
2362 EXPORT_SYMBOL_GPL(nfs_free_inode);
2363 
nfs4_init_once(struct nfs_inode * nfsi)2364 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2365 {
2366 #if IS_ENABLED(CONFIG_NFS_V4)
2367 	INIT_LIST_HEAD(&nfsi->open_states);
2368 	nfsi->delegation = NULL;
2369 	init_rwsem(&nfsi->rwsem);
2370 	nfsi->layout = NULL;
2371 #endif
2372 }
2373 
init_once(void * foo)2374 static void init_once(void *foo)
2375 {
2376 	struct nfs_inode *nfsi = foo;
2377 
2378 	inode_init_once(&nfsi->vfs_inode);
2379 	INIT_LIST_HEAD(&nfsi->open_files);
2380 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2381 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2382 	nfs4_init_once(nfsi);
2383 }
2384 
nfs_init_inodecache(void)2385 static int __init nfs_init_inodecache(void)
2386 {
2387 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2388 					     sizeof(struct nfs_inode),
2389 					     0, (SLAB_RECLAIM_ACCOUNT|
2390 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2391 					     init_once);
2392 	if (nfs_inode_cachep == NULL)
2393 		return -ENOMEM;
2394 
2395 	return 0;
2396 }
2397 
nfs_destroy_inodecache(void)2398 static void nfs_destroy_inodecache(void)
2399 {
2400 	/*
2401 	 * Make sure all delayed rcu free inodes are flushed before we
2402 	 * destroy cache.
2403 	 */
2404 	rcu_barrier();
2405 	kmem_cache_destroy(nfs_inode_cachep);
2406 }
2407 
2408 struct workqueue_struct *nfsiod_workqueue;
2409 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2410 
2411 /*
2412  * start up the nfsiod workqueue
2413  */
nfsiod_start(void)2414 static int nfsiod_start(void)
2415 {
2416 	struct workqueue_struct *wq;
2417 	dprintk("RPC:       creating workqueue nfsiod\n");
2418 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2419 	if (wq == NULL)
2420 		return -ENOMEM;
2421 	nfsiod_workqueue = wq;
2422 	return 0;
2423 }
2424 
2425 /*
2426  * Destroy the nfsiod workqueue
2427  */
nfsiod_stop(void)2428 static void nfsiod_stop(void)
2429 {
2430 	struct workqueue_struct *wq;
2431 
2432 	wq = nfsiod_workqueue;
2433 	if (wq == NULL)
2434 		return;
2435 	nfsiod_workqueue = NULL;
2436 	destroy_workqueue(wq);
2437 }
2438 
2439 unsigned int nfs_net_id;
2440 EXPORT_SYMBOL_GPL(nfs_net_id);
2441 
nfs_net_init(struct net * net)2442 static int nfs_net_init(struct net *net)
2443 {
2444 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2445 	int err;
2446 
2447 	nfs_clients_init(net);
2448 
2449 	if (!rpc_proc_register(net, &nn->rpcstats)) {
2450 		err = -ENOMEM;
2451 		goto err_proc_rpc;
2452 	}
2453 
2454 	err = nfs_fs_proc_net_init(net);
2455 	if (err)
2456 		goto err_proc_nfs;
2457 
2458 	return 0;
2459 
2460 err_proc_nfs:
2461 	rpc_proc_unregister(net, "nfs");
2462 err_proc_rpc:
2463 	nfs_clients_exit(net);
2464 	return err;
2465 }
2466 
nfs_net_exit(struct net * net)2467 static void nfs_net_exit(struct net *net)
2468 {
2469 	rpc_proc_unregister(net, "nfs");
2470 	nfs_fs_proc_net_exit(net);
2471 	nfs_clients_exit(net);
2472 }
2473 
2474 static struct pernet_operations nfs_net_ops = {
2475 	.init = nfs_net_init,
2476 	.exit = nfs_net_exit,
2477 	.id   = &nfs_net_id,
2478 	.size = sizeof(struct nfs_net),
2479 };
2480 
2481 /*
2482  * Initialize NFS
2483  */
init_nfs_fs(void)2484 static int __init init_nfs_fs(void)
2485 {
2486 	int err;
2487 
2488 	err = nfs_sysfs_init();
2489 	if (err < 0)
2490 		goto out10;
2491 
2492 	err = register_pernet_subsys(&nfs_net_ops);
2493 	if (err < 0)
2494 		goto out9;
2495 
2496 	err = nfsiod_start();
2497 	if (err)
2498 		goto out7;
2499 
2500 	err = nfs_fs_proc_init();
2501 	if (err)
2502 		goto out6;
2503 
2504 	err = nfs_init_nfspagecache();
2505 	if (err)
2506 		goto out5;
2507 
2508 	err = nfs_init_inodecache();
2509 	if (err)
2510 		goto out4;
2511 
2512 	err = nfs_init_readpagecache();
2513 	if (err)
2514 		goto out3;
2515 
2516 	err = nfs_init_writepagecache();
2517 	if (err)
2518 		goto out2;
2519 
2520 	err = nfs_init_directcache();
2521 	if (err)
2522 		goto out1;
2523 
2524 	err = register_nfs_fs();
2525 	if (err)
2526 		goto out0;
2527 
2528 	return 0;
2529 out0:
2530 	nfs_destroy_directcache();
2531 out1:
2532 	nfs_destroy_writepagecache();
2533 out2:
2534 	nfs_destroy_readpagecache();
2535 out3:
2536 	nfs_destroy_inodecache();
2537 out4:
2538 	nfs_destroy_nfspagecache();
2539 out5:
2540 	nfs_fs_proc_exit();
2541 out6:
2542 	nfsiod_stop();
2543 out7:
2544 	unregister_pernet_subsys(&nfs_net_ops);
2545 out9:
2546 	nfs_sysfs_exit();
2547 out10:
2548 	return err;
2549 }
2550 
exit_nfs_fs(void)2551 static void __exit exit_nfs_fs(void)
2552 {
2553 	nfs_destroy_directcache();
2554 	nfs_destroy_writepagecache();
2555 	nfs_destroy_readpagecache();
2556 	nfs_destroy_inodecache();
2557 	nfs_destroy_nfspagecache();
2558 	unregister_pernet_subsys(&nfs_net_ops);
2559 	unregister_nfs_fs();
2560 	nfs_fs_proc_exit();
2561 	nfsiod_stop();
2562 	nfs_sysfs_exit();
2563 }
2564 
2565 /* Not quite true; I just maintain it */
2566 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2567 MODULE_LICENSE("GPL");
2568 module_param(enable_ino64, bool, 0644);
2569 
2570 module_init(init_nfs_fs)
2571 module_exit(exit_nfs_fs)
2572