• 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_killable(int mode)75 static int nfs_wait_killable(int mode)
76 {
77 	freezable_schedule_unsafe();
78 	if (signal_pending_state(mode, current))
79 		return -ERESTARTSYS;
80 	return 0;
81 }
82 
nfs_wait_bit_killable(struct wait_bit_key * key,int mode)83 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
84 {
85 	return nfs_wait_killable(mode);
86 }
87 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88 
89 /**
90  * nfs_compat_user_ino64 - returns the user-visible inode number
91  * @fileid: 64-bit fileid
92  *
93  * This function returns a 32-bit inode number if the boot parameter
94  * nfs.enable_ino64 is zero.
95  */
nfs_compat_user_ino64(u64 fileid)96 u64 nfs_compat_user_ino64(u64 fileid)
97 {
98 #ifdef CONFIG_COMPAT
99 	compat_ulong_t ino;
100 #else
101 	unsigned long ino;
102 #endif
103 
104 	if (enable_ino64)
105 		return fileid;
106 	ino = fileid;
107 	if (sizeof(ino) < sizeof(fileid))
108 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
109 	return ino;
110 }
111 
nfs_drop_inode(struct inode * inode)112 int nfs_drop_inode(struct inode *inode)
113 {
114 	return NFS_STALE(inode) || generic_drop_inode(inode);
115 }
116 EXPORT_SYMBOL_GPL(nfs_drop_inode);
117 
nfs_clear_inode(struct inode * inode)118 void nfs_clear_inode(struct inode *inode)
119 {
120 	/*
121 	 * The following should never happen...
122 	 */
123 	WARN_ON_ONCE(nfs_have_writebacks(inode));
124 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
125 	nfs_zap_acl_cache(inode);
126 	nfs_access_zap_cache(inode);
127 	nfs_fscache_clear_inode(inode);
128 }
129 EXPORT_SYMBOL_GPL(nfs_clear_inode);
130 
nfs_evict_inode(struct inode * inode)131 void nfs_evict_inode(struct inode *inode)
132 {
133 	truncate_inode_pages_final(&inode->i_data);
134 	clear_inode(inode);
135 	nfs_clear_inode(inode);
136 }
137 
nfs_sync_inode(struct inode * inode)138 int nfs_sync_inode(struct inode *inode)
139 {
140 	inode_dio_wait(inode);
141 	return nfs_wb_all(inode);
142 }
143 EXPORT_SYMBOL_GPL(nfs_sync_inode);
144 
145 /**
146  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147  * @mapping: pointer to struct address_space
148  */
nfs_sync_mapping(struct address_space * mapping)149 int nfs_sync_mapping(struct address_space *mapping)
150 {
151 	int ret = 0;
152 
153 	if (mapping->nrpages != 0) {
154 		unmap_mapping_range(mapping, 0, 0, 0);
155 		ret = nfs_wb_all(mapping->host);
156 	}
157 	return ret;
158 }
159 
nfs_attribute_timeout(struct inode * inode)160 static int nfs_attribute_timeout(struct inode *inode)
161 {
162 	struct nfs_inode *nfsi = NFS_I(inode);
163 
164 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
165 }
166 
nfs_check_cache_flags_invalid(struct inode * inode,unsigned long flags)167 static bool nfs_check_cache_flags_invalid(struct inode *inode,
168 					  unsigned long flags)
169 {
170 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
171 
172 	return (cache_validity & flags) != 0;
173 }
174 
nfs_check_cache_invalid(struct inode * inode,unsigned long flags)175 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
176 {
177 	if (nfs_check_cache_flags_invalid(inode, flags))
178 		return true;
179 	return nfs_attribute_cache_expired(inode);
180 }
181 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
182 
183 #ifdef CONFIG_NFS_V4_2
nfs_has_xattr_cache(const struct nfs_inode * nfsi)184 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
185 {
186 	return nfsi->xattr_cache != NULL;
187 }
188 #else
nfs_has_xattr_cache(const struct nfs_inode * nfsi)189 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
190 {
191 	return false;
192 }
193 #endif
194 
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)195 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
196 {
197 	struct nfs_inode *nfsi = NFS_I(inode);
198 	bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
199 
200 	if (have_delegation) {
201 		if (!(flags & NFS_INO_REVAL_FORCED))
202 			flags &= ~(NFS_INO_INVALID_MODE |
203 				   NFS_INO_INVALID_OTHER |
204 				   NFS_INO_INVALID_XATTR);
205 		flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
206 	} else if (flags & NFS_INO_REVAL_PAGECACHE)
207 		flags |= NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE;
208 
209 	if (!nfs_has_xattr_cache(nfsi))
210 		flags &= ~NFS_INO_INVALID_XATTR;
211 	if (flags & NFS_INO_INVALID_DATA)
212 		nfs_fscache_invalidate(inode);
213 	flags &= ~(NFS_INO_REVAL_PAGECACHE | NFS_INO_REVAL_FORCED);
214 
215 	nfsi->cache_validity |= flags;
216 
217 	if (inode->i_mapping->nrpages == 0)
218 		nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA |
219 					  NFS_INO_DATA_INVAL_DEFER);
220 	else if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
221 		nfsi->cache_validity &= ~NFS_INO_DATA_INVAL_DEFER;
222 }
223 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
224 
225 /*
226  * Invalidate the local caches
227  */
nfs_zap_caches_locked(struct inode * inode)228 static void nfs_zap_caches_locked(struct inode *inode)
229 {
230 	struct nfs_inode *nfsi = NFS_I(inode);
231 	int mode = inode->i_mode;
232 
233 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
234 
235 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
236 	nfsi->attrtimeo_timestamp = jiffies;
237 
238 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
239 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
240 					| NFS_INO_INVALID_DATA
241 					| NFS_INO_INVALID_ACCESS
242 					| NFS_INO_INVALID_ACL
243 					| NFS_INO_INVALID_XATTR
244 					| NFS_INO_REVAL_PAGECACHE);
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_INO_REVAL_PAGECACHE);
251 	nfs_zap_label_cache_locked(nfsi);
252 }
253 
nfs_zap_caches(struct inode * inode)254 void nfs_zap_caches(struct inode *inode)
255 {
256 	spin_lock(&inode->i_lock);
257 	nfs_zap_caches_locked(inode);
258 	spin_unlock(&inode->i_lock);
259 }
260 
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)261 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
262 {
263 	if (mapping->nrpages != 0) {
264 		spin_lock(&inode->i_lock);
265 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
266 		spin_unlock(&inode->i_lock);
267 	}
268 }
269 
nfs_zap_acl_cache(struct inode * inode)270 void nfs_zap_acl_cache(struct inode *inode)
271 {
272 	void (*clear_acl_cache)(struct inode *);
273 
274 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
275 	if (clear_acl_cache != NULL)
276 		clear_acl_cache(inode);
277 	spin_lock(&inode->i_lock);
278 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
279 	spin_unlock(&inode->i_lock);
280 }
281 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
282 
nfs_invalidate_atime(struct inode * inode)283 void nfs_invalidate_atime(struct inode *inode)
284 {
285 	spin_lock(&inode->i_lock);
286 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
287 	spin_unlock(&inode->i_lock);
288 }
289 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
290 
291 /*
292  * Invalidate, but do not unhash, the inode.
293  * NB: must be called with inode->i_lock held!
294  */
nfs_set_inode_stale_locked(struct inode * inode)295 static void nfs_set_inode_stale_locked(struct inode *inode)
296 {
297 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
298 	nfs_zap_caches_locked(inode);
299 	trace_nfs_set_inode_stale(inode);
300 }
301 
nfs_set_inode_stale(struct inode * inode)302 void nfs_set_inode_stale(struct inode *inode)
303 {
304 	spin_lock(&inode->i_lock);
305 	nfs_set_inode_stale_locked(inode);
306 	spin_unlock(&inode->i_lock);
307 }
308 
309 struct nfs_find_desc {
310 	struct nfs_fh		*fh;
311 	struct nfs_fattr	*fattr;
312 };
313 
314 /*
315  * In NFSv3 we can have 64bit inode numbers. In order to support
316  * this, and re-exported directories (also seen in NFSv2)
317  * we are forced to allow 2 different inodes to have the same
318  * i_ino.
319  */
320 static int
nfs_find_actor(struct inode * inode,void * opaque)321 nfs_find_actor(struct inode *inode, void *opaque)
322 {
323 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
324 	struct nfs_fh		*fh = desc->fh;
325 	struct nfs_fattr	*fattr = desc->fattr;
326 
327 	if (NFS_FILEID(inode) != fattr->fileid)
328 		return 0;
329 	if (inode_wrong_type(inode, fattr->mode))
330 		return 0;
331 	if (nfs_compare_fh(NFS_FH(inode), fh))
332 		return 0;
333 	if (is_bad_inode(inode) || NFS_STALE(inode))
334 		return 0;
335 	return 1;
336 }
337 
338 static int
nfs_init_locked(struct inode * inode,void * opaque)339 nfs_init_locked(struct inode *inode, void *opaque)
340 {
341 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
342 	struct nfs_fattr	*fattr = desc->fattr;
343 
344 	set_nfs_fileid(inode, fattr->fileid);
345 	inode->i_mode = fattr->mode;
346 	nfs_copy_fh(NFS_FH(inode), desc->fh);
347 	return 0;
348 }
349 
350 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)351 static void nfs_clear_label_invalid(struct inode *inode)
352 {
353 	spin_lock(&inode->i_lock);
354 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
355 	spin_unlock(&inode->i_lock);
356 }
357 
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)358 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
359 					struct nfs4_label *label)
360 {
361 	int error;
362 
363 	if (label == NULL)
364 		return;
365 
366 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
367 		error = security_inode_notifysecctx(inode, label->label,
368 				label->len);
369 		if (error)
370 			printk(KERN_ERR "%s() %s %d "
371 					"security_inode_notifysecctx() %d\n",
372 					__func__,
373 					(char *)label->label,
374 					label->len, error);
375 		nfs_clear_label_invalid(inode);
376 	}
377 }
378 
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)379 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
380 {
381 	struct nfs4_label *label = NULL;
382 	int minor_version = server->nfs_client->cl_minorversion;
383 
384 	if (minor_version < 2)
385 		return label;
386 
387 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
388 		return label;
389 
390 	label = kzalloc(sizeof(struct nfs4_label), flags);
391 	if (label == NULL)
392 		return ERR_PTR(-ENOMEM);
393 
394 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
395 	if (label->label == NULL) {
396 		kfree(label);
397 		return ERR_PTR(-ENOMEM);
398 	}
399 	label->len = NFS4_MAXLABELLEN;
400 
401 	return label;
402 }
403 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
404 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)405 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
406 					struct nfs4_label *label)
407 {
408 }
409 #endif
410 EXPORT_SYMBOL_GPL(nfs_setsecurity);
411 
412 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
413 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)414 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
415 {
416 	struct nfs_find_desc desc = {
417 		.fh	= fh,
418 		.fattr	= fattr,
419 	};
420 	struct inode *inode;
421 	unsigned long hash;
422 
423 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
424 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
425 		return NULL;
426 
427 	hash = nfs_fattr_to_ino_t(fattr);
428 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
429 
430 	dprintk("%s: returning %p\n", __func__, inode);
431 	return inode;
432 }
433 
nfs_inode_init_regular(struct nfs_inode * nfsi)434 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
435 {
436 	atomic_long_set(&nfsi->nrequests, 0);
437 	atomic_long_set(&nfsi->redirtied_pages, 0);
438 	INIT_LIST_HEAD(&nfsi->commit_info.list);
439 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
440 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
441 	mutex_init(&nfsi->commit_mutex);
442 }
443 
nfs_inode_init_dir(struct nfs_inode * nfsi)444 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
445 {
446 	nfsi->cache_change_attribute = 0;
447 	memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
448 	init_rwsem(&nfsi->rmdir_sem);
449 }
450 
451 /*
452  * This is our front-end to iget that looks up inodes by file handle
453  * instead of inode number.
454  */
455 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr,struct nfs4_label * label)456 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
457 {
458 	struct nfs_find_desc desc = {
459 		.fh	= fh,
460 		.fattr	= fattr
461 	};
462 	struct inode *inode = ERR_PTR(-ENOENT);
463 	u64 fattr_supported = NFS_SB(sb)->fattr_valid;
464 	unsigned long hash;
465 
466 	nfs_attr_check_mountpoint(sb, fattr);
467 
468 	if (nfs_attr_use_mounted_on_fileid(fattr))
469 		fattr->fileid = fattr->mounted_on_fileid;
470 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
471 		goto out_no_inode;
472 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
473 		goto out_no_inode;
474 
475 	hash = nfs_fattr_to_ino_t(fattr);
476 
477 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
478 	if (inode == NULL) {
479 		inode = ERR_PTR(-ENOMEM);
480 		goto out_no_inode;
481 	}
482 
483 	if (inode->i_state & I_NEW) {
484 		struct nfs_inode *nfsi = NFS_I(inode);
485 		unsigned long now = jiffies;
486 
487 		/* We set i_ino for the few things that still rely on it,
488 		 * such as stat(2) */
489 		inode->i_ino = hash;
490 
491 		/* We can't support update_atime(), since the server will reset it */
492 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
493 		inode->i_mode = fattr->mode;
494 		nfsi->cache_validity = 0;
495 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
496 				&& (fattr_supported & NFS_ATTR_FATTR_MODE))
497 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
498 		/* Why so? Because we want revalidate for devices/FIFOs, and
499 		 * that's precisely what we have in nfs_file_inode_operations.
500 		 */
501 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
502 		if (S_ISREG(inode->i_mode)) {
503 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
504 			inode->i_data.a_ops = &nfs_file_aops;
505 			nfs_inode_init_regular(nfsi);
506 		} else if (S_ISDIR(inode->i_mode)) {
507 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
508 			inode->i_fop = &nfs_dir_operations;
509 			inode->i_data.a_ops = &nfs_dir_aops;
510 			nfs_inode_init_dir(nfsi);
511 			/* Deal with crossing mountpoints */
512 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
513 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
514 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
515 					inode->i_op = &nfs_referral_inode_operations;
516 				else
517 					inode->i_op = &nfs_mountpoint_inode_operations;
518 				inode->i_fop = NULL;
519 				inode->i_flags |= S_AUTOMOUNT;
520 			}
521 		} else if (S_ISLNK(inode->i_mode)) {
522 			inode->i_op = &nfs_symlink_inode_operations;
523 			inode_nohighmem(inode);
524 		} else
525 			init_special_inode(inode, inode->i_mode, fattr->rdev);
526 
527 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
528 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
529 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
530 		inode_set_iversion_raw(inode, 0);
531 		inode->i_size = 0;
532 		clear_nlink(inode);
533 		inode->i_uid = make_kuid(&init_user_ns, -2);
534 		inode->i_gid = make_kgid(&init_user_ns, -2);
535 		inode->i_blocks = 0;
536 		nfsi->write_io = 0;
537 		nfsi->read_io = 0;
538 
539 		nfsi->read_cache_jiffies = fattr->time_start;
540 		nfsi->attr_gencount = fattr->gencount;
541 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
542 			inode->i_atime = fattr->atime;
543 		else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
544 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
545 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
546 			inode->i_mtime = fattr->mtime;
547 		else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
548 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
549 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
550 			inode->i_ctime = fattr->ctime;
551 		else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
552 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
553 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
554 			inode_set_iversion_raw(inode, fattr->change_attr);
555 		else
556 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
557 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
558 			inode->i_size = nfs_size_to_loff_t(fattr->size);
559 		else
560 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
561 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
562 			set_nlink(inode, fattr->nlink);
563 		else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
564 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
565 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
566 			inode->i_uid = fattr->uid;
567 		else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
568 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
569 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
570 			inode->i_gid = fattr->gid;
571 		else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
572 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
573 		if (nfs_server_capable(inode, NFS_CAP_XATTR))
574 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
575 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
576 			inode->i_blocks = fattr->du.nfs2.blocks;
577 		else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
578 			 fattr->size != 0)
579 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
580 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
581 			/*
582 			 * report the blocks in 512byte units
583 			 */
584 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
585 		} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
586 			   fattr->size != 0)
587 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
588 
589 		nfs_setsecurity(inode, fattr, label);
590 
591 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
592 		nfsi->attrtimeo_timestamp = now;
593 		nfsi->access_cache = RB_ROOT;
594 
595 		nfs_fscache_init_inode(inode);
596 
597 		unlock_new_inode(inode);
598 	} else {
599 		int err = nfs_refresh_inode(inode, fattr);
600 		if (err < 0) {
601 			iput(inode);
602 			inode = ERR_PTR(err);
603 			goto out_no_inode;
604 		}
605 	}
606 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
607 		inode->i_sb->s_id,
608 		(unsigned long long)NFS_FILEID(inode),
609 		nfs_display_fhandle_hash(fh),
610 		atomic_read(&inode->i_count));
611 
612 out:
613 	return inode;
614 
615 out_no_inode:
616 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
617 	goto out;
618 }
619 EXPORT_SYMBOL_GPL(nfs_fhget);
620 
621 #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)
622 
623 int
nfs_setattr(struct user_namespace * mnt_userns,struct dentry * dentry,struct iattr * attr)624 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
625 	    struct iattr *attr)
626 {
627 	struct inode *inode = d_inode(dentry);
628 	struct nfs_fattr *fattr;
629 	int error = 0;
630 
631 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
632 
633 	/* skip mode change if it's just for clearing setuid/setgid */
634 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
635 		attr->ia_valid &= ~ATTR_MODE;
636 
637 	if (attr->ia_valid & ATTR_SIZE) {
638 		BUG_ON(!S_ISREG(inode->i_mode));
639 
640 		error = inode_newsize_ok(inode, attr->ia_size);
641 		if (error)
642 			return error;
643 
644 		if (attr->ia_size == i_size_read(inode))
645 			attr->ia_valid &= ~ATTR_SIZE;
646 	}
647 
648 	/* Optimization: if the end result is no change, don't RPC */
649 	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
650 		return 0;
651 
652 	trace_nfs_setattr_enter(inode);
653 
654 	/* Write all dirty data */
655 	if (S_ISREG(inode->i_mode))
656 		nfs_sync_inode(inode);
657 
658 	fattr = nfs_alloc_fattr();
659 	if (fattr == NULL) {
660 		error = -ENOMEM;
661 		goto out;
662 	}
663 
664 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
665 	if (error == 0)
666 		error = nfs_refresh_inode(inode, fattr);
667 	nfs_free_fattr(fattr);
668 out:
669 	trace_nfs_setattr_exit(inode, error);
670 	return error;
671 }
672 EXPORT_SYMBOL_GPL(nfs_setattr);
673 
674 /**
675  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
676  * @inode: inode of the file used
677  * @offset: file offset to start truncating
678  *
679  * This is a copy of the common vmtruncate, but with the locking
680  * corrected to take into account the fact that NFS requires
681  * inode->i_size to be updated under the inode->i_lock.
682  * Note: must be called with inode->i_lock held!
683  */
nfs_vmtruncate(struct inode * inode,loff_t offset)684 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
685 {
686 	int err;
687 
688 	err = inode_newsize_ok(inode, offset);
689 	if (err)
690 		goto out;
691 
692 	i_size_write(inode, offset);
693 	/* Optimisation */
694 	if (offset == 0)
695 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
696 				NFS_INO_DATA_INVAL_DEFER);
697 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
698 
699 	spin_unlock(&inode->i_lock);
700 	truncate_pagecache(inode, offset);
701 	spin_lock(&inode->i_lock);
702 out:
703 	return err;
704 }
705 
706 /**
707  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
708  * @inode: pointer to struct inode
709  * @attr: pointer to struct iattr
710  * @fattr: pointer to struct nfs_fattr
711  *
712  * Note: we do this in the *proc.c in order to ensure that
713  *       it works for things like exclusive creates too.
714  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)715 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
716 		struct nfs_fattr *fattr)
717 {
718 	/* Barrier: bump the attribute generation count. */
719 	nfs_fattr_set_barrier(fattr);
720 
721 	spin_lock(&inode->i_lock);
722 	NFS_I(inode)->attr_gencount = fattr->gencount;
723 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
724 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
725 						     NFS_INO_INVALID_BLOCKS);
726 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
727 		nfs_vmtruncate(inode, attr->ia_size);
728 	}
729 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
730 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
731 		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
732 		    inode->i_mode & S_ISUID)
733 			inode->i_mode &= ~S_ISUID;
734 		if (setattr_should_drop_sgid(&init_user_ns, inode))
735 			inode->i_mode &= ~S_ISGID;
736 		if ((attr->ia_valid & ATTR_MODE) != 0) {
737 			int mode = attr->ia_mode & S_IALLUGO;
738 			mode |= inode->i_mode & ~S_IALLUGO;
739 			inode->i_mode = mode;
740 		}
741 		if ((attr->ia_valid & ATTR_UID) != 0)
742 			inode->i_uid = attr->ia_uid;
743 		if ((attr->ia_valid & ATTR_GID) != 0)
744 			inode->i_gid = attr->ia_gid;
745 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
746 			inode->i_ctime = fattr->ctime;
747 		else
748 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
749 					| NFS_INO_INVALID_CTIME);
750 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
751 				| NFS_INO_INVALID_ACL);
752 	}
753 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
754 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
755 				| NFS_INO_INVALID_CTIME);
756 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
757 			inode->i_atime = fattr->atime;
758 		else if (attr->ia_valid & ATTR_ATIME_SET)
759 			inode->i_atime = attr->ia_atime;
760 		else
761 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
762 
763 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
764 			inode->i_ctime = fattr->ctime;
765 		else
766 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
767 					| NFS_INO_INVALID_CTIME);
768 	}
769 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
770 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
771 				| NFS_INO_INVALID_CTIME);
772 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
773 			inode->i_mtime = fattr->mtime;
774 		else if (attr->ia_valid & ATTR_MTIME_SET)
775 			inode->i_mtime = attr->ia_mtime;
776 		else
777 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
778 
779 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
780 			inode->i_ctime = fattr->ctime;
781 		else
782 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
783 					| NFS_INO_INVALID_CTIME);
784 	}
785 	if (fattr->valid)
786 		nfs_update_inode(inode, fattr);
787 	spin_unlock(&inode->i_lock);
788 }
789 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
790 
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)791 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
792 {
793 	struct dentry *parent;
794 
795 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
796 		return;
797 	parent = dget_parent(dentry);
798 	nfs_force_use_readdirplus(d_inode(parent));
799 	dput(parent);
800 }
801 
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)802 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
803 {
804 	struct dentry *parent;
805 
806 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
807 		return;
808 	parent = dget_parent(dentry);
809 	nfs_advise_use_readdirplus(d_inode(parent));
810 	dput(parent);
811 }
812 
nfs_get_valid_attrmask(struct inode * inode)813 static u32 nfs_get_valid_attrmask(struct inode *inode)
814 {
815 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
816 	u32 reply_mask = STATX_INO | STATX_TYPE;
817 
818 	if (!(cache_validity & NFS_INO_INVALID_ATIME))
819 		reply_mask |= STATX_ATIME;
820 	if (!(cache_validity & NFS_INO_INVALID_CTIME))
821 		reply_mask |= STATX_CTIME;
822 	if (!(cache_validity & NFS_INO_INVALID_MTIME))
823 		reply_mask |= STATX_MTIME;
824 	if (!(cache_validity & NFS_INO_INVALID_SIZE))
825 		reply_mask |= STATX_SIZE;
826 	if (!(cache_validity & NFS_INO_INVALID_NLINK))
827 		reply_mask |= STATX_NLINK;
828 	if (!(cache_validity & NFS_INO_INVALID_MODE))
829 		reply_mask |= STATX_MODE;
830 	if (!(cache_validity & NFS_INO_INVALID_OTHER))
831 		reply_mask |= STATX_UID | STATX_GID;
832 	if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
833 		reply_mask |= STATX_BLOCKS;
834 	return reply_mask;
835 }
836 
nfs_getattr(struct user_namespace * mnt_userns,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)837 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
838 		struct kstat *stat, u32 request_mask, unsigned int query_flags)
839 {
840 	struct inode *inode = d_inode(path->dentry);
841 	struct nfs_server *server = NFS_SERVER(inode);
842 	unsigned long cache_validity;
843 	int err = 0;
844 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
845 	bool do_update = false;
846 
847 	trace_nfs_getattr_enter(inode);
848 
849 	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
850 			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
851 			STATX_INO | STATX_SIZE | STATX_BLOCKS;
852 
853 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
854 		nfs_readdirplus_parent_cache_hit(path->dentry);
855 		goto out_no_revalidate;
856 	}
857 
858 	/* Flush out writes to the server in order to update c/mtime.  */
859 	if ((request_mask & (STATX_CTIME | STATX_MTIME)) &&
860 	    S_ISREG(inode->i_mode))
861 		filemap_write_and_wait(inode->i_mapping);
862 
863 	/*
864 	 * We may force a getattr if the user cares about atime.
865 	 *
866 	 * Note that we only have to check the vfsmount flags here:
867 	 *  - NFS always sets S_NOATIME by so checking it would give a
868 	 *    bogus result
869 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
870 	 *    no point in checking those.
871 	 */
872 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
873 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
874 		request_mask &= ~STATX_ATIME;
875 
876 	/* Is the user requesting attributes that might need revalidation? */
877 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
878 					STATX_MTIME|STATX_UID|STATX_GID|
879 					STATX_SIZE|STATX_BLOCKS)))
880 		goto out_no_revalidate;
881 
882 	/* Check whether the cached attributes are stale */
883 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
884 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
885 	do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
886 	if (request_mask & STATX_ATIME)
887 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
888 	if (request_mask & STATX_CTIME)
889 		do_update |= cache_validity & NFS_INO_INVALID_CTIME;
890 	if (request_mask & STATX_MTIME)
891 		do_update |= cache_validity & NFS_INO_INVALID_MTIME;
892 	if (request_mask & STATX_SIZE)
893 		do_update |= cache_validity & NFS_INO_INVALID_SIZE;
894 	if (request_mask & STATX_NLINK)
895 		do_update |= cache_validity & NFS_INO_INVALID_NLINK;
896 	if (request_mask & STATX_MODE)
897 		do_update |= cache_validity & NFS_INO_INVALID_MODE;
898 	if (request_mask & (STATX_UID | STATX_GID))
899 		do_update |= cache_validity & NFS_INO_INVALID_OTHER;
900 	if (request_mask & STATX_BLOCKS)
901 		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
902 
903 	if (do_update) {
904 		/* Update the attribute cache */
905 		if (!(server->flags & NFS_MOUNT_NOAC))
906 			nfs_readdirplus_parent_cache_miss(path->dentry);
907 		else
908 			nfs_readdirplus_parent_cache_hit(path->dentry);
909 		err = __nfs_revalidate_inode(server, inode);
910 		if (err)
911 			goto out;
912 	} else
913 		nfs_readdirplus_parent_cache_hit(path->dentry);
914 out_no_revalidate:
915 	/* Only return attributes that were revalidated. */
916 	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
917 
918 	generic_fillattr(&init_user_ns, inode, stat);
919 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
920 	if (S_ISDIR(inode->i_mode))
921 		stat->blksize = NFS_SERVER(inode)->dtsize;
922 out:
923 	trace_nfs_getattr_exit(inode, err);
924 	return err;
925 }
926 EXPORT_SYMBOL_GPL(nfs_getattr);
927 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)928 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
929 {
930 	refcount_set(&l_ctx->count, 1);
931 	l_ctx->lockowner = current->files;
932 	INIT_LIST_HEAD(&l_ctx->list);
933 	atomic_set(&l_ctx->io_count, 0);
934 }
935 
__nfs_find_lock_context(struct nfs_open_context * ctx)936 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
937 {
938 	struct nfs_lock_context *pos;
939 
940 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
941 		if (pos->lockowner != current->files)
942 			continue;
943 		if (refcount_inc_not_zero(&pos->count))
944 			return pos;
945 	}
946 	return NULL;
947 }
948 
nfs_get_lock_context(struct nfs_open_context * ctx)949 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
950 {
951 	struct nfs_lock_context *res, *new = NULL;
952 	struct inode *inode = d_inode(ctx->dentry);
953 
954 	rcu_read_lock();
955 	res = __nfs_find_lock_context(ctx);
956 	rcu_read_unlock();
957 	if (res == NULL) {
958 		new = kmalloc(sizeof(*new), GFP_KERNEL);
959 		if (new == NULL)
960 			return ERR_PTR(-ENOMEM);
961 		nfs_init_lock_context(new);
962 		spin_lock(&inode->i_lock);
963 		res = __nfs_find_lock_context(ctx);
964 		if (res == NULL) {
965 			new->open_context = get_nfs_open_context(ctx);
966 			if (new->open_context) {
967 				list_add_tail_rcu(&new->list,
968 						&ctx->lock_context.list);
969 				res = new;
970 				new = NULL;
971 			} else
972 				res = ERR_PTR(-EBADF);
973 		}
974 		spin_unlock(&inode->i_lock);
975 		kfree(new);
976 	}
977 	return res;
978 }
979 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
980 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)981 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
982 {
983 	struct nfs_open_context *ctx = l_ctx->open_context;
984 	struct inode *inode = d_inode(ctx->dentry);
985 
986 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
987 		return;
988 	list_del_rcu(&l_ctx->list);
989 	spin_unlock(&inode->i_lock);
990 	put_nfs_open_context(ctx);
991 	kfree_rcu(l_ctx, rcu_head);
992 }
993 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
994 
995 /**
996  * nfs_close_context - Common close_context() routine NFSv2/v3
997  * @ctx: pointer to context
998  * @is_sync: is this a synchronous close
999  *
1000  * Ensure that the attributes are up to date if we're mounted
1001  * with close-to-open semantics and we have cached data that will
1002  * need to be revalidated on open.
1003  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)1004 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1005 {
1006 	struct nfs_inode *nfsi;
1007 	struct inode *inode;
1008 
1009 	if (!(ctx->mode & FMODE_WRITE))
1010 		return;
1011 	if (!is_sync)
1012 		return;
1013 	inode = d_inode(ctx->dentry);
1014 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1015 		return;
1016 	nfsi = NFS_I(inode);
1017 	if (inode->i_mapping->nrpages == 0)
1018 		return;
1019 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1020 		return;
1021 	if (!list_empty(&nfsi->open_files))
1022 		return;
1023 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1024 		return;
1025 	nfs_revalidate_inode(inode,
1026 			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1027 }
1028 EXPORT_SYMBOL_GPL(nfs_close_context);
1029 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1030 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1031 						fmode_t f_mode,
1032 						struct file *filp)
1033 {
1034 	struct nfs_open_context *ctx;
1035 
1036 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
1037 	if (!ctx)
1038 		return ERR_PTR(-ENOMEM);
1039 	nfs_sb_active(dentry->d_sb);
1040 	ctx->dentry = dget(dentry);
1041 	if (filp)
1042 		ctx->cred = get_cred(filp->f_cred);
1043 	else
1044 		ctx->cred = get_current_cred();
1045 	ctx->ll_cred = NULL;
1046 	ctx->state = NULL;
1047 	ctx->mode = f_mode;
1048 	ctx->flags = 0;
1049 	ctx->error = 0;
1050 	ctx->flock_owner = (fl_owner_t)filp;
1051 	nfs_init_lock_context(&ctx->lock_context);
1052 	ctx->lock_context.open_context = ctx;
1053 	INIT_LIST_HEAD(&ctx->list);
1054 	ctx->mdsthreshold = NULL;
1055 	return ctx;
1056 }
1057 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1058 
get_nfs_open_context(struct nfs_open_context * ctx)1059 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1060 {
1061 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1062 		return ctx;
1063 	return NULL;
1064 }
1065 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1066 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1067 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1068 {
1069 	struct inode *inode = d_inode(ctx->dentry);
1070 	struct super_block *sb = ctx->dentry->d_sb;
1071 
1072 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1073 		return;
1074 	if (!list_empty(&ctx->list)) {
1075 		spin_lock(&inode->i_lock);
1076 		list_del_rcu(&ctx->list);
1077 		spin_unlock(&inode->i_lock);
1078 	}
1079 	if (inode != NULL)
1080 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1081 	put_cred(ctx->cred);
1082 	dput(ctx->dentry);
1083 	nfs_sb_deactive(sb);
1084 	put_rpccred(ctx->ll_cred);
1085 	kfree(ctx->mdsthreshold);
1086 	kfree_rcu(ctx, rcu_head);
1087 }
1088 
put_nfs_open_context(struct nfs_open_context * ctx)1089 void put_nfs_open_context(struct nfs_open_context *ctx)
1090 {
1091 	__put_nfs_open_context(ctx, 0);
1092 }
1093 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1094 
put_nfs_open_context_sync(struct nfs_open_context * ctx)1095 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1096 {
1097 	__put_nfs_open_context(ctx, 1);
1098 }
1099 
1100 /*
1101  * Ensure that mmap has a recent RPC credential for use when writing out
1102  * shared pages
1103  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1104 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1105 {
1106 	struct inode *inode = d_inode(ctx->dentry);
1107 	struct nfs_inode *nfsi = NFS_I(inode);
1108 
1109 	spin_lock(&inode->i_lock);
1110 	if (list_empty(&nfsi->open_files) &&
1111 	    (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1112 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1113 						     NFS_INO_REVAL_FORCED);
1114 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1115 	spin_unlock(&inode->i_lock);
1116 }
1117 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1118 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1119 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1120 {
1121 	filp->private_data = get_nfs_open_context(ctx);
1122 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1123 	if (list_empty(&ctx->list))
1124 		nfs_inode_attach_open_context(ctx);
1125 }
1126 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1127 
1128 /*
1129  * Given an inode, search for an open context with the desired characteristics
1130  */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1131 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1132 {
1133 	struct nfs_inode *nfsi = NFS_I(inode);
1134 	struct nfs_open_context *pos, *ctx = NULL;
1135 
1136 	rcu_read_lock();
1137 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1138 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1139 			continue;
1140 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1141 			continue;
1142 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1143 			continue;
1144 		ctx = get_nfs_open_context(pos);
1145 		if (ctx)
1146 			break;
1147 	}
1148 	rcu_read_unlock();
1149 	return ctx;
1150 }
1151 
nfs_file_clear_open_context(struct file * filp)1152 void nfs_file_clear_open_context(struct file *filp)
1153 {
1154 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1155 
1156 	if (ctx) {
1157 		struct inode *inode = d_inode(ctx->dentry);
1158 
1159 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1160 		/*
1161 		 * We fatal error on write before. Try to writeback
1162 		 * every page again.
1163 		 */
1164 		if (ctx->error < 0)
1165 			invalidate_inode_pages2(inode->i_mapping);
1166 		filp->private_data = NULL;
1167 		put_nfs_open_context_sync(ctx);
1168 	}
1169 }
1170 
1171 /*
1172  * These allocate and release file read/write context information.
1173  */
nfs_open(struct inode * inode,struct file * filp)1174 int nfs_open(struct inode *inode, struct file *filp)
1175 {
1176 	struct nfs_open_context *ctx;
1177 
1178 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1179 	if (IS_ERR(ctx))
1180 		return PTR_ERR(ctx);
1181 	nfs_file_set_open_context(filp, ctx);
1182 	put_nfs_open_context(ctx);
1183 	nfs_fscache_open_file(inode, filp);
1184 	return 0;
1185 }
1186 
1187 /*
1188  * This function is called whenever some part of NFS notices that
1189  * the cached attributes have to be refreshed.
1190  */
1191 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1192 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1193 {
1194 	int		 status = -ESTALE;
1195 	struct nfs4_label *label = NULL;
1196 	struct nfs_fattr *fattr = NULL;
1197 	struct nfs_inode *nfsi = NFS_I(inode);
1198 
1199 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1200 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1201 
1202 	trace_nfs_revalidate_inode_enter(inode);
1203 
1204 	if (is_bad_inode(inode))
1205 		goto out;
1206 	if (NFS_STALE(inode))
1207 		goto out;
1208 
1209 	/* pNFS: Attributes aren't updated until we layoutcommit */
1210 	if (S_ISREG(inode->i_mode)) {
1211 		status = pnfs_sync_inode(inode, false);
1212 		if (status)
1213 			goto out;
1214 	}
1215 
1216 	status = -ENOMEM;
1217 	fattr = nfs_alloc_fattr();
1218 	if (fattr == NULL)
1219 		goto out;
1220 
1221 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1222 
1223 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1224 	if (IS_ERR(label)) {
1225 		status = PTR_ERR(label);
1226 		goto out;
1227 	}
1228 
1229 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1230 			label, inode);
1231 	if (status != 0) {
1232 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1233 			 inode->i_sb->s_id,
1234 			 (unsigned long long)NFS_FILEID(inode), status);
1235 		switch (status) {
1236 		case -ETIMEDOUT:
1237 			/* A soft timeout occurred. Use cached information? */
1238 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1239 				status = 0;
1240 			break;
1241 		case -ESTALE:
1242 			if (!S_ISDIR(inode->i_mode))
1243 				nfs_set_inode_stale(inode);
1244 			else
1245 				nfs_zap_caches(inode);
1246 		}
1247 		goto err_out;
1248 	}
1249 
1250 	status = nfs_refresh_inode(inode, fattr);
1251 	if (status) {
1252 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1253 			 inode->i_sb->s_id,
1254 			 (unsigned long long)NFS_FILEID(inode), status);
1255 		goto err_out;
1256 	}
1257 
1258 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1259 		nfs_zap_acl_cache(inode);
1260 
1261 	nfs_setsecurity(inode, fattr, label);
1262 
1263 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1264 		inode->i_sb->s_id,
1265 		(unsigned long long)NFS_FILEID(inode));
1266 
1267 err_out:
1268 	nfs4_label_free(label);
1269 out:
1270 	nfs_free_fattr(fattr);
1271 	trace_nfs_revalidate_inode_exit(inode, status);
1272 	return status;
1273 }
1274 
nfs_attribute_cache_expired(struct inode * inode)1275 int nfs_attribute_cache_expired(struct inode *inode)
1276 {
1277 	if (nfs_have_delegated_attributes(inode))
1278 		return 0;
1279 	return nfs_attribute_timeout(inode);
1280 }
1281 
1282 /**
1283  * nfs_revalidate_inode - Revalidate the inode attributes
1284  * @inode: pointer to inode struct
1285  * @flags: cache flags to check
1286  *
1287  * Updates inode attribute information by retrieving the data from the server.
1288  */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1289 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1290 {
1291 	if (!nfs_check_cache_invalid(inode, flags))
1292 		return NFS_STALE(inode) ? -ESTALE : 0;
1293 	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1294 }
1295 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1296 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1297 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1298 {
1299 	int ret;
1300 
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 	nfs_fscache_wait_on_invalidate(inode);
1313 
1314 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1315 			inode->i_sb->s_id,
1316 			(unsigned long long)NFS_FILEID(inode));
1317 	return 0;
1318 }
1319 
1320 /**
1321  * nfs_clear_invalid_mapping - Conditionally clear a mapping
1322  * @mapping: pointer to mapping
1323  *
1324  * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1325  */
nfs_clear_invalid_mapping(struct address_space * mapping)1326 int nfs_clear_invalid_mapping(struct address_space *mapping)
1327 {
1328 	struct inode *inode = mapping->host;
1329 	struct nfs_inode *nfsi = NFS_I(inode);
1330 	unsigned long *bitlock = &nfsi->flags;
1331 	int ret = 0;
1332 
1333 	/*
1334 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1335 	 * invalidations that come in while we're shooting down the mappings
1336 	 * are respected. But, that leaves a race window where one revalidator
1337 	 * can clear the flag, and then another checks it before the mapping
1338 	 * gets invalidated. Fix that by serializing access to this part of
1339 	 * the function.
1340 	 *
1341 	 * At the same time, we need to allow other tasks to see whether we
1342 	 * might be in the middle of invalidating the pages, so we only set
1343 	 * the bit lock here if it looks like we're going to be doing that.
1344 	 */
1345 	for (;;) {
1346 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1347 					 nfs_wait_bit_killable, TASK_KILLABLE);
1348 		if (ret)
1349 			goto out;
1350 		spin_lock(&inode->i_lock);
1351 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1352 			spin_unlock(&inode->i_lock);
1353 			continue;
1354 		}
1355 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1356 			break;
1357 		spin_unlock(&inode->i_lock);
1358 		goto out;
1359 	}
1360 
1361 	set_bit(NFS_INO_INVALIDATING, bitlock);
1362 	smp_wmb();
1363 	nfsi->cache_validity &=
1364 		~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER);
1365 	spin_unlock(&inode->i_lock);
1366 	trace_nfs_invalidate_mapping_enter(inode);
1367 	ret = nfs_invalidate_mapping(inode, mapping);
1368 	trace_nfs_invalidate_mapping_exit(inode, ret);
1369 
1370 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1371 	smp_mb__after_atomic();
1372 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1373 out:
1374 	return ret;
1375 }
1376 
nfs_mapping_need_revalidate_inode(struct inode * inode)1377 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1378 {
1379 	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1380 		NFS_STALE(inode);
1381 }
1382 
nfs_revalidate_mapping_rcu(struct inode * inode)1383 int nfs_revalidate_mapping_rcu(struct inode *inode)
1384 {
1385 	struct nfs_inode *nfsi = NFS_I(inode);
1386 	unsigned long *bitlock = &nfsi->flags;
1387 	int ret = 0;
1388 
1389 	if (IS_SWAPFILE(inode))
1390 		goto out;
1391 	if (nfs_mapping_need_revalidate_inode(inode)) {
1392 		ret = -ECHILD;
1393 		goto out;
1394 	}
1395 	spin_lock(&inode->i_lock);
1396 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1397 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1398 		ret = -ECHILD;
1399 	spin_unlock(&inode->i_lock);
1400 out:
1401 	return ret;
1402 }
1403 
1404 /**
1405  * nfs_revalidate_mapping - Revalidate the pagecache
1406  * @inode: pointer to host inode
1407  * @mapping: pointer to mapping
1408  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1409 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1410 {
1411 	/* swapfiles are not supposed to be shared. */
1412 	if (IS_SWAPFILE(inode))
1413 		return 0;
1414 
1415 	if (nfs_mapping_need_revalidate_inode(inode)) {
1416 		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1417 		if (ret < 0)
1418 			return ret;
1419 	}
1420 
1421 	return nfs_clear_invalid_mapping(mapping);
1422 }
1423 
nfs_file_has_writers(struct nfs_inode * nfsi)1424 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1425 {
1426 	struct inode *inode = &nfsi->vfs_inode;
1427 
1428 	if (!S_ISREG(inode->i_mode))
1429 		return false;
1430 	if (list_empty(&nfsi->open_files))
1431 		return false;
1432 	return inode_is_open_for_write(inode);
1433 }
1434 
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1435 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1436 {
1437 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1438 }
1439 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1440 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1441 {
1442 	struct timespec64 ts;
1443 
1444 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1445 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1446 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1447 		inode_set_iversion_raw(inode, fattr->change_attr);
1448 		if (S_ISDIR(inode->i_mode))
1449 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1450 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1451 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1452 	}
1453 	/* If we have atomic WCC data, we may update some attributes */
1454 	ts = inode->i_ctime;
1455 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1456 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1457 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1458 		inode->i_ctime = fattr->ctime;
1459 	}
1460 
1461 	ts = inode->i_mtime;
1462 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1463 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1464 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1465 		inode->i_mtime = fattr->mtime;
1466 		if (S_ISDIR(inode->i_mode))
1467 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1468 	}
1469 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1470 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1471 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1472 			&& !nfs_have_writebacks(inode)) {
1473 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1474 	}
1475 }
1476 
1477 /**
1478  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1479  * @inode: pointer to inode
1480  * @fattr: updated attributes
1481  *
1482  * Verifies the attribute cache. If we have just changed the attributes,
1483  * so that fattr carries weak cache consistency data, then it may
1484  * also update the ctime/mtime/change_attribute.
1485  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1486 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1487 {
1488 	struct nfs_inode *nfsi = NFS_I(inode);
1489 	loff_t cur_size, new_isize;
1490 	unsigned long invalid = 0;
1491 	struct timespec64 ts;
1492 
1493 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1494 		return 0;
1495 
1496 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1497 		/* Only a mounted-on-fileid? Just exit */
1498 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1499 			return 0;
1500 	/* Has the inode gone and changed behind our back? */
1501 	} else if (nfsi->fileid != fattr->fileid) {
1502 		/* Is this perhaps the mounted-on fileid? */
1503 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1504 		    nfsi->fileid == fattr->mounted_on_fileid)
1505 			return 0;
1506 		return -ESTALE;
1507 	}
1508 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1509 		return -ESTALE;
1510 
1511 
1512 	if (!nfs_file_has_buffered_writers(nfsi)) {
1513 		/* Verify a few of the more important attributes */
1514 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1515 			invalid |= NFS_INO_INVALID_CHANGE;
1516 
1517 		ts = inode->i_mtime;
1518 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1519 			invalid |= NFS_INO_INVALID_MTIME;
1520 
1521 		ts = inode->i_ctime;
1522 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1523 			invalid |= NFS_INO_INVALID_CTIME;
1524 
1525 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1526 			cur_size = i_size_read(inode);
1527 			new_isize = nfs_size_to_loff_t(fattr->size);
1528 			if (cur_size != new_isize)
1529 				invalid |= NFS_INO_INVALID_SIZE;
1530 		}
1531 	}
1532 
1533 	/* Have any file permissions changed? */
1534 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1535 		invalid |= NFS_INO_INVALID_MODE;
1536 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1537 		invalid |= NFS_INO_INVALID_OTHER;
1538 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1539 		invalid |= NFS_INO_INVALID_OTHER;
1540 
1541 	/* Has the link count changed? */
1542 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1543 		invalid |= NFS_INO_INVALID_NLINK;
1544 
1545 	ts = inode->i_atime;
1546 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1547 		invalid |= NFS_INO_INVALID_ATIME;
1548 
1549 	if (invalid != 0)
1550 		nfs_set_cache_invalid(inode, invalid);
1551 
1552 	nfsi->read_cache_jiffies = fattr->time_start;
1553 	return 0;
1554 }
1555 
1556 static atomic_long_t nfs_attr_generation_counter;
1557 
nfs_read_attr_generation_counter(void)1558 static unsigned long nfs_read_attr_generation_counter(void)
1559 {
1560 	return atomic_long_read(&nfs_attr_generation_counter);
1561 }
1562 
nfs_inc_attr_generation_counter(void)1563 unsigned long nfs_inc_attr_generation_counter(void)
1564 {
1565 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1566 }
1567 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1568 
nfs_fattr_init(struct nfs_fattr * fattr)1569 void nfs_fattr_init(struct nfs_fattr *fattr)
1570 {
1571 	fattr->valid = 0;
1572 	fattr->time_start = jiffies;
1573 	fattr->gencount = nfs_inc_attr_generation_counter();
1574 	fattr->owner_name = NULL;
1575 	fattr->group_name = NULL;
1576 }
1577 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1578 
1579 /**
1580  * nfs_fattr_set_barrier
1581  * @fattr: attributes
1582  *
1583  * Used to set a barrier after an attribute was updated. This
1584  * barrier ensures that older attributes from RPC calls that may
1585  * have raced with our update cannot clobber these new values.
1586  * Note that you are still responsible for ensuring that other
1587  * operations which change the attribute on the server do not
1588  * collide.
1589  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1590 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1591 {
1592 	fattr->gencount = nfs_inc_attr_generation_counter();
1593 }
1594 
nfs_alloc_fattr(void)1595 struct nfs_fattr *nfs_alloc_fattr(void)
1596 {
1597 	struct nfs_fattr *fattr;
1598 
1599 	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1600 	if (fattr != NULL) {
1601 		nfs_fattr_init(fattr);
1602 		fattr->label = NULL;
1603 	}
1604 	return fattr;
1605 }
1606 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1607 
nfs_alloc_fattr_with_label(struct nfs_server * server)1608 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1609 {
1610 	struct nfs_fattr *fattr = nfs_alloc_fattr();
1611 
1612 	if (!fattr)
1613 		return NULL;
1614 
1615 	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1616 	if (IS_ERR(fattr->label)) {
1617 		kfree(fattr);
1618 		return NULL;
1619 	}
1620 
1621 	return fattr;
1622 }
1623 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1624 
nfs_alloc_fhandle(void)1625 struct nfs_fh *nfs_alloc_fhandle(void)
1626 {
1627 	struct nfs_fh *fh;
1628 
1629 	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1630 	if (fh != NULL)
1631 		fh->size = 0;
1632 	return fh;
1633 }
1634 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1635 
1636 #ifdef NFS_DEBUG
1637 /*
1638  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1639  *                             in the same way that wireshark does
1640  *
1641  * @fh: file handle
1642  *
1643  * For debugging only.
1644  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1645 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1646 {
1647 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1648 	 * not on the result */
1649 	return nfs_fhandle_hash(fh);
1650 }
1651 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1652 
1653 /*
1654  * _nfs_display_fhandle - display an NFS file handle on the console
1655  *
1656  * @fh: file handle to display
1657  * @caption: display caption
1658  *
1659  * For debugging only.
1660  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1661 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1662 {
1663 	unsigned short i;
1664 
1665 	if (fh == NULL || fh->size == 0) {
1666 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1667 		return;
1668 	}
1669 
1670 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1671 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1672 	for (i = 0; i < fh->size; i += 16) {
1673 		__be32 *pos = (__be32 *)&fh->data[i];
1674 
1675 		switch ((fh->size - i - 1) >> 2) {
1676 		case 0:
1677 			printk(KERN_DEFAULT " %08x\n",
1678 				be32_to_cpup(pos));
1679 			break;
1680 		case 1:
1681 			printk(KERN_DEFAULT " %08x %08x\n",
1682 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1683 			break;
1684 		case 2:
1685 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1686 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1687 				be32_to_cpup(pos + 2));
1688 			break;
1689 		default:
1690 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1691 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1692 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1693 		}
1694 	}
1695 }
1696 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1697 #endif
1698 
1699 /**
1700  * nfs_inode_attrs_cmp_generic - compare attributes
1701  * @fattr: attributes
1702  * @inode: pointer to inode
1703  *
1704  * Attempt to divine whether or not an RPC call reply carrying stale
1705  * attributes got scheduled after another call carrying updated ones.
1706  * Note also the check for wraparound of 'attr_gencount'
1707  *
1708  * The function returns '1' if it thinks the attributes in @fattr are
1709  * more recent than the ones cached in @inode. Otherwise it returns
1710  * the value '0'.
1711  */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1712 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1713 				       const struct inode *inode)
1714 {
1715 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1716 
1717 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1718 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1719 }
1720 
1721 /**
1722  * nfs_inode_attrs_cmp_monotonic - compare attributes
1723  * @fattr: attributes
1724  * @inode: pointer to inode
1725  *
1726  * Attempt to divine whether or not an RPC call reply carrying stale
1727  * attributes got scheduled after another call carrying updated ones.
1728  *
1729  * We assume that the server observes monotonic semantics for
1730  * the change attribute, so a larger value means that the attributes in
1731  * @fattr are more recent, in which case the function returns the
1732  * value '1'.
1733  * A return value of '0' indicates no measurable change
1734  * A return value of '-1' means that the attributes in @inode are
1735  * more recent.
1736  */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1737 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1738 					 const struct inode *inode)
1739 {
1740 	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1741 	if (diff > 0)
1742 		return 1;
1743 	return diff == 0 ? 0 : -1;
1744 }
1745 
1746 /**
1747  * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1748  * @fattr: attributes
1749  * @inode: pointer to inode
1750  *
1751  * Attempt to divine whether or not an RPC call reply carrying stale
1752  * attributes got scheduled after another call carrying updated ones.
1753  *
1754  * We assume that the server observes strictly monotonic semantics for
1755  * the change attribute, so a larger value means that the attributes in
1756  * @fattr are more recent, in which case the function returns the
1757  * value '1'.
1758  * A return value of '-1' means that the attributes in @inode are
1759  * more recent or unchanged.
1760  */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1761 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1762 						const struct inode *inode)
1763 {
1764 	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1765 }
1766 
1767 /**
1768  * nfs_inode_attrs_cmp - compare attributes
1769  * @fattr: attributes
1770  * @inode: pointer to inode
1771  *
1772  * This function returns '1' if it thinks the attributes in @fattr are
1773  * more recent than the ones cached in @inode. It returns '-1' if
1774  * the attributes in @inode are more recent than the ones in @fattr,
1775  * and it returns 0 if not sure.
1776  */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1777 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1778 			       const struct inode *inode)
1779 {
1780 	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1781 		return 1;
1782 	switch (NFS_SERVER(inode)->change_attr_type) {
1783 	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1784 		break;
1785 	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1786 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1787 			break;
1788 		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1789 	default:
1790 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1791 			break;
1792 		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1793 	}
1794 	return 0;
1795 }
1796 
1797 /**
1798  * nfs_inode_finish_partial_attr_update - complete a previous inode update
1799  * @fattr: attributes
1800  * @inode: pointer to inode
1801  *
1802  * Returns '1' if the last attribute update left the inode cached
1803  * attributes in a partially unrevalidated state, and @fattr
1804  * matches the change attribute of that partial update.
1805  * Otherwise returns '0'.
1806  */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1807 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1808 						const struct inode *inode)
1809 {
1810 	const unsigned long check_valid =
1811 		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1812 		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1813 		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1814 		NFS_INO_INVALID_NLINK;
1815 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1816 	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1817 
1818 	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1819 	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1820 	    (cache_validity & check_valid) != 0 &&
1821 	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1822 	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1823 		return 1;
1824 	return 0;
1825 }
1826 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1827 static int nfs_refresh_inode_locked(struct inode *inode,
1828 				    struct nfs_fattr *fattr)
1829 {
1830 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1831 	int ret = 0;
1832 
1833 	trace_nfs_refresh_inode_enter(inode);
1834 
1835 	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1836 		ret = nfs_update_inode(inode, fattr);
1837 	else if (attr_cmp == 0)
1838 		ret = nfs_check_inode_attributes(inode, fattr);
1839 
1840 	trace_nfs_refresh_inode_exit(inode, ret);
1841 	return ret;
1842 }
1843 
1844 /**
1845  * nfs_refresh_inode - try to update the inode attribute cache
1846  * @inode: pointer to inode
1847  * @fattr: updated attributes
1848  *
1849  * Check that an RPC call that returned attributes has not overlapped with
1850  * other recent updates of the inode metadata, then decide whether it is
1851  * safe to do a full update of the inode attributes, or whether just to
1852  * call nfs_check_inode_attributes.
1853  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1854 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1855 {
1856 	int status;
1857 
1858 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1859 		return 0;
1860 	spin_lock(&inode->i_lock);
1861 	status = nfs_refresh_inode_locked(inode, fattr);
1862 	spin_unlock(&inode->i_lock);
1863 
1864 	return status;
1865 }
1866 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1867 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1868 static int nfs_post_op_update_inode_locked(struct inode *inode,
1869 		struct nfs_fattr *fattr, unsigned int invalid)
1870 {
1871 	if (S_ISDIR(inode->i_mode))
1872 		invalid |= NFS_INO_INVALID_DATA;
1873 	nfs_set_cache_invalid(inode, invalid);
1874 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1875 		return 0;
1876 	return nfs_refresh_inode_locked(inode, fattr);
1877 }
1878 
1879 /**
1880  * nfs_post_op_update_inode - try to update the inode attribute cache
1881  * @inode: pointer to inode
1882  * @fattr: updated attributes
1883  *
1884  * After an operation that has changed the inode metadata, mark the
1885  * attribute cache as being invalid, then try to update it.
1886  *
1887  * NB: if the server didn't return any post op attributes, this
1888  * function will force the retrieval of attributes before the next
1889  * NFS request.  Thus it should be used only for operations that
1890  * are expected to change one or more attributes, to avoid
1891  * unnecessary NFS requests and trips through nfs_update_inode().
1892  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1893 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1894 {
1895 	int status;
1896 
1897 	spin_lock(&inode->i_lock);
1898 	nfs_fattr_set_barrier(fattr);
1899 	status = nfs_post_op_update_inode_locked(inode, fattr,
1900 			NFS_INO_INVALID_CHANGE
1901 			| NFS_INO_INVALID_CTIME
1902 			| NFS_INO_REVAL_FORCED);
1903 	spin_unlock(&inode->i_lock);
1904 
1905 	return status;
1906 }
1907 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1908 
1909 /**
1910  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1911  * @inode: pointer to inode
1912  * @fattr: updated attributes
1913  *
1914  * After an operation that has changed the inode metadata, mark the
1915  * attribute cache as being invalid, then try to update it. Fake up
1916  * weak cache consistency data, if none exist.
1917  *
1918  * This function is mainly designed to be used by the ->write_done() functions.
1919  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1920 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1921 {
1922 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1923 	int status;
1924 
1925 	/* Don't do a WCC update if these attributes are already stale */
1926 	if (attr_cmp < 0)
1927 		return 0;
1928 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1929 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1930 				| NFS_ATTR_FATTR_PRESIZE
1931 				| NFS_ATTR_FATTR_PREMTIME
1932 				| NFS_ATTR_FATTR_PRECTIME);
1933 		goto out_noforce;
1934 	}
1935 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1936 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1937 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1938 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1939 	}
1940 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1941 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1942 		fattr->pre_ctime = inode->i_ctime;
1943 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1944 	}
1945 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1946 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1947 		fattr->pre_mtime = inode->i_mtime;
1948 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1949 	}
1950 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1951 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1952 		fattr->pre_size = i_size_read(inode);
1953 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1954 	}
1955 out_noforce:
1956 	status = nfs_post_op_update_inode_locked(inode, fattr,
1957 			NFS_INO_INVALID_CHANGE
1958 			| NFS_INO_INVALID_CTIME
1959 			| NFS_INO_INVALID_MTIME
1960 			| NFS_INO_INVALID_BLOCKS);
1961 	return status;
1962 }
1963 
1964 /**
1965  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1966  * @inode: pointer to inode
1967  * @fattr: updated attributes
1968  *
1969  * After an operation that has changed the inode metadata, mark the
1970  * attribute cache as being invalid, then try to update it. Fake up
1971  * weak cache consistency data, if none exist.
1972  *
1973  * This function is mainly designed to be used by the ->write_done() functions.
1974  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)1975 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1976 {
1977 	int status;
1978 
1979 	spin_lock(&inode->i_lock);
1980 	nfs_fattr_set_barrier(fattr);
1981 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1982 	spin_unlock(&inode->i_lock);
1983 	return status;
1984 }
1985 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1986 
1987 
1988 /*
1989  * Many nfs protocol calls return the new file attributes after
1990  * an operation.  Here we update the inode to reflect the state
1991  * of the server's inode.
1992  *
1993  * This is a bit tricky because we have to make sure all dirty pages
1994  * have been sent off to the server before calling invalidate_inode_pages.
1995  * To make sure no other process adds more write requests while we try
1996  * our best to flush them, we make them sleep during the attribute refresh.
1997  *
1998  * A very similar scenario holds for the dir cache.
1999  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)2000 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2001 {
2002 	struct nfs_server *server = NFS_SERVER(inode);
2003 	struct nfs_inode *nfsi = NFS_I(inode);
2004 	loff_t cur_isize, new_isize;
2005 	u64 fattr_supported = server->fattr_valid;
2006 	unsigned long invalid = 0;
2007 	unsigned long now = jiffies;
2008 	unsigned long save_cache_validity;
2009 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2010 	bool cache_revalidated = true;
2011 	bool attr_changed = false;
2012 	bool have_delegation;
2013 
2014 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2015 			__func__, inode->i_sb->s_id, inode->i_ino,
2016 			nfs_display_fhandle_hash(NFS_FH(inode)),
2017 			atomic_read(&inode->i_count), fattr->valid);
2018 
2019 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2020 		/* Only a mounted-on-fileid? Just exit */
2021 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2022 			return 0;
2023 	/* Has the inode gone and changed behind our back? */
2024 	} else if (nfsi->fileid != fattr->fileid) {
2025 		/* Is this perhaps the mounted-on fileid? */
2026 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2027 		    nfsi->fileid == fattr->mounted_on_fileid)
2028 			return 0;
2029 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2030 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2031 			NFS_SERVER(inode)->nfs_client->cl_hostname,
2032 			inode->i_sb->s_id, (long long)nfsi->fileid,
2033 			(long long)fattr->fileid);
2034 		goto out_err;
2035 	}
2036 
2037 	/*
2038 	 * Make sure the inode's type hasn't changed.
2039 	 */
2040 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2041 		/*
2042 		* Big trouble! The inode has become a different object.
2043 		*/
2044 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2045 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2046 		goto out_err;
2047 	}
2048 
2049 	/* Update the fsid? */
2050 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2051 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2052 			!IS_AUTOMOUNT(inode))
2053 		server->fsid = fattr->fsid;
2054 
2055 	/* Save the delegation state before clearing cache_validity */
2056 	have_delegation = nfs_have_delegated_attributes(inode);
2057 
2058 	/*
2059 	 * Update the read time so we don't revalidate too often.
2060 	 */
2061 	nfsi->read_cache_jiffies = fattr->time_start;
2062 
2063 	save_cache_validity = nfsi->cache_validity;
2064 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2065 			| NFS_INO_INVALID_ATIME
2066 			| NFS_INO_REVAL_FORCED
2067 			| NFS_INO_INVALID_BLOCKS);
2068 
2069 	/* Do atomic weak cache consistency updates */
2070 	nfs_wcc_update_inode(inode, fattr);
2071 
2072 	if (pnfs_layoutcommit_outstanding(inode)) {
2073 		nfsi->cache_validity |=
2074 			save_cache_validity &
2075 			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2076 			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2077 			 NFS_INO_INVALID_BLOCKS);
2078 		cache_revalidated = false;
2079 	}
2080 
2081 	/* More cache consistency checks */
2082 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2083 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2084 			/* Could it be a race with writeback? */
2085 			if (!(have_writers || have_delegation)) {
2086 				invalid |= NFS_INO_INVALID_DATA
2087 					| NFS_INO_INVALID_ACCESS
2088 					| NFS_INO_INVALID_ACL
2089 					| NFS_INO_INVALID_XATTR;
2090 				/* Force revalidate of all attributes */
2091 				save_cache_validity |= NFS_INO_INVALID_CTIME
2092 					| NFS_INO_INVALID_MTIME
2093 					| NFS_INO_INVALID_SIZE
2094 					| NFS_INO_INVALID_BLOCKS
2095 					| NFS_INO_INVALID_NLINK
2096 					| NFS_INO_INVALID_MODE
2097 					| NFS_INO_INVALID_OTHER;
2098 				if (S_ISDIR(inode->i_mode))
2099 					nfs_force_lookup_revalidate(inode);
2100 				attr_changed = true;
2101 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2102 						inode->i_sb->s_id,
2103 						inode->i_ino);
2104 			} else if (!have_delegation)
2105 				nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2106 			inode_set_iversion_raw(inode, fattr->change_attr);
2107 		}
2108 	} else {
2109 		nfsi->cache_validity |=
2110 			save_cache_validity & NFS_INO_INVALID_CHANGE;
2111 		if (!have_delegation ||
2112 		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2113 			cache_revalidated = false;
2114 	}
2115 
2116 	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2117 		inode->i_mtime = fattr->mtime;
2118 	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2119 		nfsi->cache_validity |=
2120 			save_cache_validity & NFS_INO_INVALID_MTIME;
2121 
2122 	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2123 		inode->i_ctime = fattr->ctime;
2124 	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2125 		nfsi->cache_validity |=
2126 			save_cache_validity & NFS_INO_INVALID_CTIME;
2127 
2128 	/* Check if our cached file size is stale */
2129 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2130 		new_isize = nfs_size_to_loff_t(fattr->size);
2131 		cur_isize = i_size_read(inode);
2132 		if (new_isize != cur_isize && !have_delegation) {
2133 			/* Do we perhaps have any outstanding writes, or has
2134 			 * the file grown beyond our last write? */
2135 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2136 				i_size_write(inode, new_isize);
2137 				if (!have_writers)
2138 					invalid |= NFS_INO_INVALID_DATA;
2139 			}
2140 			dprintk("NFS: isize change on server for file %s/%ld "
2141 					"(%Ld to %Ld)\n",
2142 					inode->i_sb->s_id,
2143 					inode->i_ino,
2144 					(long long)cur_isize,
2145 					(long long)new_isize);
2146 		}
2147 		if (new_isize == 0 &&
2148 		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2149 				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2150 			fattr->du.nfs3.used = 0;
2151 			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2152 		}
2153 	} else
2154 		nfsi->cache_validity |=
2155 			save_cache_validity & NFS_INO_INVALID_SIZE;
2156 
2157 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2158 		inode->i_atime = fattr->atime;
2159 	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2160 		nfsi->cache_validity |=
2161 			save_cache_validity & NFS_INO_INVALID_ATIME;
2162 
2163 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2164 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2165 			umode_t newmode = inode->i_mode & S_IFMT;
2166 			newmode |= fattr->mode & S_IALLUGO;
2167 			inode->i_mode = newmode;
2168 			invalid |= NFS_INO_INVALID_ACCESS
2169 				| NFS_INO_INVALID_ACL;
2170 		}
2171 	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2172 		nfsi->cache_validity |=
2173 			save_cache_validity & NFS_INO_INVALID_MODE;
2174 
2175 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2176 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2177 			invalid |= NFS_INO_INVALID_ACCESS
2178 				| NFS_INO_INVALID_ACL;
2179 			inode->i_uid = fattr->uid;
2180 		}
2181 	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2182 		nfsi->cache_validity |=
2183 			save_cache_validity & NFS_INO_INVALID_OTHER;
2184 
2185 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2186 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2187 			invalid |= NFS_INO_INVALID_ACCESS
2188 				| NFS_INO_INVALID_ACL;
2189 			inode->i_gid = fattr->gid;
2190 		}
2191 	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2192 		nfsi->cache_validity |=
2193 			save_cache_validity & NFS_INO_INVALID_OTHER;
2194 
2195 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2196 		if (inode->i_nlink != fattr->nlink) {
2197 			if (S_ISDIR(inode->i_mode))
2198 				invalid |= NFS_INO_INVALID_DATA;
2199 			set_nlink(inode, fattr->nlink);
2200 		}
2201 	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2202 		nfsi->cache_validity |=
2203 			save_cache_validity & NFS_INO_INVALID_NLINK;
2204 
2205 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2206 		/*
2207 		 * report the blocks in 512byte units
2208 		 */
2209 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2210 	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2211 		nfsi->cache_validity |=
2212 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2213 
2214 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2215 		inode->i_blocks = fattr->du.nfs2.blocks;
2216 	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2217 		nfsi->cache_validity |=
2218 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2219 
2220 	/* Update attrtimeo value if we're out of the unstable period */
2221 	if (attr_changed) {
2222 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2223 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2224 		nfsi->attrtimeo_timestamp = now;
2225 		/* Set barrier to be more recent than all outstanding updates */
2226 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2227 	} else {
2228 		if (cache_revalidated) {
2229 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2230 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2231 				nfsi->attrtimeo <<= 1;
2232 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2233 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2234 			}
2235 			nfsi->attrtimeo_timestamp = now;
2236 		}
2237 		/* Set the barrier to be more recent than this fattr */
2238 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2239 			nfsi->attr_gencount = fattr->gencount;
2240 	}
2241 
2242 	/* Don't invalidate the data if we were to blame */
2243 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2244 				|| S_ISLNK(inode->i_mode)))
2245 		invalid &= ~NFS_INO_INVALID_DATA;
2246 	nfs_set_cache_invalid(inode, invalid);
2247 
2248 	return 0;
2249  out_err:
2250 	/*
2251 	 * No need to worry about unhashing the dentry, as the
2252 	 * lookup validation will know that the inode is bad.
2253 	 * (But we fall through to invalidate the caches.)
2254 	 */
2255 	nfs_set_inode_stale_locked(inode);
2256 	return -ESTALE;
2257 }
2258 
nfs_alloc_inode(struct super_block * sb)2259 struct inode *nfs_alloc_inode(struct super_block *sb)
2260 {
2261 	struct nfs_inode *nfsi;
2262 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2263 	if (!nfsi)
2264 		return NULL;
2265 	nfsi->flags = 0UL;
2266 	nfsi->cache_validity = 0UL;
2267 #if IS_ENABLED(CONFIG_NFS_V4)
2268 	nfsi->nfs4_acl = NULL;
2269 #endif /* CONFIG_NFS_V4 */
2270 #ifdef CONFIG_NFS_V4_2
2271 	nfsi->xattr_cache = NULL;
2272 #endif
2273 	return &nfsi->vfs_inode;
2274 }
2275 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2276 
nfs_free_inode(struct inode * inode)2277 void nfs_free_inode(struct inode *inode)
2278 {
2279 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2280 }
2281 EXPORT_SYMBOL_GPL(nfs_free_inode);
2282 
nfs4_init_once(struct nfs_inode * nfsi)2283 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2284 {
2285 #if IS_ENABLED(CONFIG_NFS_V4)
2286 	INIT_LIST_HEAD(&nfsi->open_states);
2287 	nfsi->delegation = NULL;
2288 	init_rwsem(&nfsi->rwsem);
2289 	nfsi->layout = NULL;
2290 #endif
2291 }
2292 
init_once(void * foo)2293 static void init_once(void *foo)
2294 {
2295 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2296 
2297 	inode_init_once(&nfsi->vfs_inode);
2298 	INIT_LIST_HEAD(&nfsi->open_files);
2299 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2300 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2301 	nfs4_init_once(nfsi);
2302 }
2303 
nfs_init_inodecache(void)2304 static int __init nfs_init_inodecache(void)
2305 {
2306 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2307 					     sizeof(struct nfs_inode),
2308 					     0, (SLAB_RECLAIM_ACCOUNT|
2309 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2310 					     init_once);
2311 	if (nfs_inode_cachep == NULL)
2312 		return -ENOMEM;
2313 
2314 	return 0;
2315 }
2316 
nfs_destroy_inodecache(void)2317 static void nfs_destroy_inodecache(void)
2318 {
2319 	/*
2320 	 * Make sure all delayed rcu free inodes are flushed before we
2321 	 * destroy cache.
2322 	 */
2323 	rcu_barrier();
2324 	kmem_cache_destroy(nfs_inode_cachep);
2325 }
2326 
2327 struct workqueue_struct *nfsiod_workqueue;
2328 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2329 
2330 /*
2331  * start up the nfsiod workqueue
2332  */
nfsiod_start(void)2333 static int nfsiod_start(void)
2334 {
2335 	struct workqueue_struct *wq;
2336 	dprintk("RPC:       creating workqueue nfsiod\n");
2337 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2338 	if (wq == NULL)
2339 		return -ENOMEM;
2340 	nfsiod_workqueue = wq;
2341 	return 0;
2342 }
2343 
2344 /*
2345  * Destroy the nfsiod workqueue
2346  */
nfsiod_stop(void)2347 static void nfsiod_stop(void)
2348 {
2349 	struct workqueue_struct *wq;
2350 
2351 	wq = nfsiod_workqueue;
2352 	if (wq == NULL)
2353 		return;
2354 	nfsiod_workqueue = NULL;
2355 	destroy_workqueue(wq);
2356 }
2357 
2358 unsigned int nfs_net_id;
2359 EXPORT_SYMBOL_GPL(nfs_net_id);
2360 
nfs_net_init(struct net * net)2361 static int nfs_net_init(struct net *net)
2362 {
2363 	nfs_clients_init(net);
2364 	return nfs_fs_proc_net_init(net);
2365 }
2366 
nfs_net_exit(struct net * net)2367 static void nfs_net_exit(struct net *net)
2368 {
2369 	nfs_fs_proc_net_exit(net);
2370 	nfs_clients_exit(net);
2371 }
2372 
2373 static struct pernet_operations nfs_net_ops = {
2374 	.init = nfs_net_init,
2375 	.exit = nfs_net_exit,
2376 	.id   = &nfs_net_id,
2377 	.size = sizeof(struct nfs_net),
2378 };
2379 
2380 /*
2381  * Initialize NFS
2382  */
init_nfs_fs(void)2383 static int __init init_nfs_fs(void)
2384 {
2385 	int err;
2386 
2387 	err = nfs_sysfs_init();
2388 	if (err < 0)
2389 		goto out10;
2390 
2391 	err = register_pernet_subsys(&nfs_net_ops);
2392 	if (err < 0)
2393 		goto out9;
2394 
2395 	err = nfs_fscache_register();
2396 	if (err < 0)
2397 		goto out8;
2398 
2399 	err = nfsiod_start();
2400 	if (err)
2401 		goto out7;
2402 
2403 	err = nfs_fs_proc_init();
2404 	if (err)
2405 		goto out6;
2406 
2407 	err = nfs_init_nfspagecache();
2408 	if (err)
2409 		goto out5;
2410 
2411 	err = nfs_init_inodecache();
2412 	if (err)
2413 		goto out4;
2414 
2415 	err = nfs_init_readpagecache();
2416 	if (err)
2417 		goto out3;
2418 
2419 	err = nfs_init_writepagecache();
2420 	if (err)
2421 		goto out2;
2422 
2423 	err = nfs_init_directcache();
2424 	if (err)
2425 		goto out1;
2426 
2427 	rpc_proc_register(&init_net, &nfs_rpcstat);
2428 
2429 	err = register_nfs_fs();
2430 	if (err)
2431 		goto out0;
2432 
2433 	return 0;
2434 out0:
2435 	rpc_proc_unregister(&init_net, "nfs");
2436 	nfs_destroy_directcache();
2437 out1:
2438 	nfs_destroy_writepagecache();
2439 out2:
2440 	nfs_destroy_readpagecache();
2441 out3:
2442 	nfs_destroy_inodecache();
2443 out4:
2444 	nfs_destroy_nfspagecache();
2445 out5:
2446 	nfs_fs_proc_exit();
2447 out6:
2448 	nfsiod_stop();
2449 out7:
2450 	nfs_fscache_unregister();
2451 out8:
2452 	unregister_pernet_subsys(&nfs_net_ops);
2453 out9:
2454 	nfs_sysfs_exit();
2455 out10:
2456 	return err;
2457 }
2458 
exit_nfs_fs(void)2459 static void __exit exit_nfs_fs(void)
2460 {
2461 	nfs_destroy_directcache();
2462 	nfs_destroy_writepagecache();
2463 	nfs_destroy_readpagecache();
2464 	nfs_destroy_inodecache();
2465 	nfs_destroy_nfspagecache();
2466 	nfs_fscache_unregister();
2467 	unregister_pernet_subsys(&nfs_net_ops);
2468 	rpc_proc_unregister(&init_net, "nfs");
2469 	unregister_nfs_fs();
2470 	nfs_fs_proc_exit();
2471 	nfsiod_stop();
2472 	nfs_sysfs_exit();
2473 }
2474 
2475 /* Not quite true; I just maintain it */
2476 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2477 MODULE_LICENSE("GPL");
2478 MODULE_IMPORT_NS(ANDROID_GKI_VFS_EXPORT_ONLY);
2479 module_param(enable_ino64, bool, 0644);
2480 
2481 module_init(init_nfs_fs)
2482 module_exit(exit_nfs_fs)
2483