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