xref: /openbmc/linux/fs/nfs/inode.c (revision 8a1f89c9)
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_ooo_clear(nfsi);
214 	} else if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
215 		nfs_ooo_clear(nfsi);
216 	}
217 	trace_nfs_set_cache_invalid(inode, 0);
218 }
219 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
220 
221 /*
222  * Invalidate the local caches
223  */
nfs_zap_caches_locked(struct inode * inode)224 static void nfs_zap_caches_locked(struct inode *inode)
225 {
226 	struct nfs_inode *nfsi = NFS_I(inode);
227 	int mode = inode->i_mode;
228 
229 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
230 
231 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
232 	nfsi->attrtimeo_timestamp = jiffies;
233 
234 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
235 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
236 						     NFS_INO_INVALID_DATA |
237 						     NFS_INO_INVALID_ACCESS |
238 						     NFS_INO_INVALID_ACL |
239 						     NFS_INO_INVALID_XATTR);
240 	else
241 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
242 						     NFS_INO_INVALID_ACCESS |
243 						     NFS_INO_INVALID_ACL |
244 						     NFS_INO_INVALID_XATTR);
245 	nfs_zap_label_cache_locked(nfsi);
246 }
247 
nfs_zap_caches(struct inode * inode)248 void nfs_zap_caches(struct inode *inode)
249 {
250 	spin_lock(&inode->i_lock);
251 	nfs_zap_caches_locked(inode);
252 	spin_unlock(&inode->i_lock);
253 }
254 
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)255 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
256 {
257 	if (mapping->nrpages != 0) {
258 		spin_lock(&inode->i_lock);
259 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
260 		spin_unlock(&inode->i_lock);
261 	}
262 }
263 
nfs_zap_acl_cache(struct inode * inode)264 void nfs_zap_acl_cache(struct inode *inode)
265 {
266 	void (*clear_acl_cache)(struct inode *);
267 
268 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
269 	if (clear_acl_cache != NULL)
270 		clear_acl_cache(inode);
271 	spin_lock(&inode->i_lock);
272 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
273 	spin_unlock(&inode->i_lock);
274 }
275 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
276 
nfs_invalidate_atime(struct inode * inode)277 void nfs_invalidate_atime(struct inode *inode)
278 {
279 	spin_lock(&inode->i_lock);
280 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
281 	spin_unlock(&inode->i_lock);
282 }
283 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
284 
285 /*
286  * Invalidate, but do not unhash, the inode.
287  * NB: must be called with inode->i_lock held!
288  */
nfs_set_inode_stale_locked(struct inode * inode)289 static void nfs_set_inode_stale_locked(struct inode *inode)
290 {
291 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
292 	nfs_zap_caches_locked(inode);
293 	trace_nfs_set_inode_stale(inode);
294 }
295 
nfs_set_inode_stale(struct inode * inode)296 void nfs_set_inode_stale(struct inode *inode)
297 {
298 	spin_lock(&inode->i_lock);
299 	nfs_set_inode_stale_locked(inode);
300 	spin_unlock(&inode->i_lock);
301 }
302 
303 struct nfs_find_desc {
304 	struct nfs_fh		*fh;
305 	struct nfs_fattr	*fattr;
306 };
307 
308 /*
309  * In NFSv3 we can have 64bit inode numbers. In order to support
310  * this, and re-exported directories (also seen in NFSv2)
311  * we are forced to allow 2 different inodes to have the same
312  * i_ino.
313  */
314 static int
nfs_find_actor(struct inode * inode,void * opaque)315 nfs_find_actor(struct inode *inode, void *opaque)
316 {
317 	struct nfs_find_desc	*desc = opaque;
318 	struct nfs_fh		*fh = desc->fh;
319 	struct nfs_fattr	*fattr = desc->fattr;
320 
321 	if (NFS_FILEID(inode) != fattr->fileid)
322 		return 0;
323 	if (inode_wrong_type(inode, fattr->mode))
324 		return 0;
325 	if (nfs_compare_fh(NFS_FH(inode), fh))
326 		return 0;
327 	if (is_bad_inode(inode) || NFS_STALE(inode))
328 		return 0;
329 	return 1;
330 }
331 
332 static int
nfs_init_locked(struct inode * inode,void * opaque)333 nfs_init_locked(struct inode *inode, void *opaque)
334 {
335 	struct nfs_find_desc	*desc = opaque;
336 	struct nfs_fattr	*fattr = desc->fattr;
337 
338 	set_nfs_fileid(inode, fattr->fileid);
339 	inode->i_mode = fattr->mode;
340 	nfs_copy_fh(NFS_FH(inode), desc->fh);
341 	return 0;
342 }
343 
344 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)345 static void nfs_clear_label_invalid(struct inode *inode)
346 {
347 	spin_lock(&inode->i_lock);
348 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
349 	spin_unlock(&inode->i_lock);
350 }
351 
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)352 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
353 {
354 	int error;
355 
356 	if (fattr->label == NULL)
357 		return;
358 
359 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
360 		error = security_inode_notifysecctx(inode, fattr->label->label,
361 				fattr->label->len);
362 		if (error)
363 			printk(KERN_ERR "%s() %s %d "
364 					"security_inode_notifysecctx() %d\n",
365 					__func__,
366 					(char *)fattr->label->label,
367 					fattr->label->len, error);
368 		nfs_clear_label_invalid(inode);
369 	}
370 }
371 
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)372 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
373 {
374 	struct nfs4_label *label;
375 
376 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
377 		return NULL;
378 
379 	label = kzalloc(sizeof(struct nfs4_label), flags);
380 	if (label == NULL)
381 		return ERR_PTR(-ENOMEM);
382 
383 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
384 	if (label->label == NULL) {
385 		kfree(label);
386 		return ERR_PTR(-ENOMEM);
387 	}
388 	label->len = NFS4_MAXLABELLEN;
389 
390 	return label;
391 }
392 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
393 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)394 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
395 {
396 }
397 #endif
398 EXPORT_SYMBOL_GPL(nfs_setsecurity);
399 
400 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
401 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)402 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
403 {
404 	struct nfs_find_desc desc = {
405 		.fh	= fh,
406 		.fattr	= fattr,
407 	};
408 	struct inode *inode;
409 	unsigned long hash;
410 
411 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
412 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
413 		return NULL;
414 
415 	hash = nfs_fattr_to_ino_t(fattr);
416 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
417 
418 	dprintk("%s: returning %p\n", __func__, inode);
419 	return inode;
420 }
421 
nfs_inode_init_regular(struct nfs_inode * nfsi)422 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
423 {
424 	atomic_long_set(&nfsi->nrequests, 0);
425 	atomic_long_set(&nfsi->redirtied_pages, 0);
426 	INIT_LIST_HEAD(&nfsi->commit_info.list);
427 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
428 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
429 	mutex_init(&nfsi->commit_mutex);
430 }
431 
nfs_inode_init_dir(struct nfs_inode * nfsi)432 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
433 {
434 	nfsi->cache_change_attribute = 0;
435 	memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
436 	init_rwsem(&nfsi->rmdir_sem);
437 }
438 
439 /*
440  * This is our front-end to iget that looks up inodes by file handle
441  * instead of inode number.
442  */
443 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr)444 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
445 {
446 	struct nfs_find_desc desc = {
447 		.fh	= fh,
448 		.fattr	= fattr
449 	};
450 	struct inode *inode = ERR_PTR(-ENOENT);
451 	u64 fattr_supported = NFS_SB(sb)->fattr_valid;
452 	unsigned long hash;
453 
454 	nfs_attr_check_mountpoint(sb, fattr);
455 
456 	if (nfs_attr_use_mounted_on_fileid(fattr))
457 		fattr->fileid = fattr->mounted_on_fileid;
458 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
459 		goto out_no_inode;
460 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
461 		goto out_no_inode;
462 
463 	hash = nfs_fattr_to_ino_t(fattr);
464 
465 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
466 	if (inode == NULL) {
467 		inode = ERR_PTR(-ENOMEM);
468 		goto out_no_inode;
469 	}
470 
471 	if (inode->i_state & I_NEW) {
472 		struct nfs_inode *nfsi = NFS_I(inode);
473 		unsigned long now = jiffies;
474 
475 		/* We set i_ino for the few things that still rely on it,
476 		 * such as stat(2) */
477 		inode->i_ino = hash;
478 
479 		/* We can't support update_atime(), since the server will reset it */
480 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
481 		inode->i_mode = fattr->mode;
482 		nfsi->cache_validity = 0;
483 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
484 				&& (fattr_supported & NFS_ATTR_FATTR_MODE))
485 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
486 		/* Why so? Because we want revalidate for devices/FIFOs, and
487 		 * that's precisely what we have in nfs_file_inode_operations.
488 		 */
489 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
490 		if (S_ISREG(inode->i_mode)) {
491 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
492 			inode->i_data.a_ops = &nfs_file_aops;
493 			nfs_inode_init_regular(nfsi);
494 		} else if (S_ISDIR(inode->i_mode)) {
495 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
496 			inode->i_fop = &nfs_dir_operations;
497 			inode->i_data.a_ops = &nfs_dir_aops;
498 			nfs_inode_init_dir(nfsi);
499 			/* Deal with crossing mountpoints */
500 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
501 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
502 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
503 					inode->i_op = &nfs_referral_inode_operations;
504 				else
505 					inode->i_op = &nfs_mountpoint_inode_operations;
506 				inode->i_fop = NULL;
507 				inode->i_flags |= S_AUTOMOUNT;
508 			}
509 		} else if (S_ISLNK(inode->i_mode)) {
510 			inode->i_op = &nfs_symlink_inode_operations;
511 			inode_nohighmem(inode);
512 		} else
513 			init_special_inode(inode, inode->i_mode, fattr->rdev);
514 
515 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
516 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
517 		inode_set_ctime(inode, 0, 0);
518 		inode_set_iversion_raw(inode, 0);
519 		inode->i_size = 0;
520 		clear_nlink(inode);
521 		inode->i_uid = make_kuid(&init_user_ns, -2);
522 		inode->i_gid = make_kgid(&init_user_ns, -2);
523 		inode->i_blocks = 0;
524 		nfsi->write_io = 0;
525 		nfsi->read_io = 0;
526 
527 		nfsi->read_cache_jiffies = fattr->time_start;
528 		nfsi->attr_gencount = fattr->gencount;
529 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
530 			inode->i_atime = fattr->atime;
531 		else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
532 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
533 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
534 			inode->i_mtime = fattr->mtime;
535 		else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
536 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
537 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
538 			inode_set_ctime_to_ts(inode, fattr->ctime);
539 		else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
540 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
541 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
542 			inode_set_iversion_raw(inode, fattr->change_attr);
543 		else
544 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
545 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
546 			inode->i_size = nfs_size_to_loff_t(fattr->size);
547 		else
548 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
549 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
550 			set_nlink(inode, fattr->nlink);
551 		else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
552 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
553 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
554 			inode->i_uid = fattr->uid;
555 		else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
556 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
557 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
558 			inode->i_gid = fattr->gid;
559 		else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
560 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
561 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
562 			inode->i_blocks = fattr->du.nfs2.blocks;
563 		else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
564 			 fattr->size != 0)
565 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
566 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
567 			/*
568 			 * report the blocks in 512byte units
569 			 */
570 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
571 		} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
572 			   fattr->size != 0)
573 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
574 
575 		nfs_setsecurity(inode, fattr);
576 
577 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
578 		nfsi->attrtimeo_timestamp = now;
579 		nfsi->access_cache = RB_ROOT;
580 
581 		nfs_fscache_init_inode(inode);
582 
583 		unlock_new_inode(inode);
584 	} else {
585 		int err = nfs_refresh_inode(inode, fattr);
586 		if (err < 0) {
587 			iput(inode);
588 			inode = ERR_PTR(err);
589 			goto out_no_inode;
590 		}
591 	}
592 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
593 		inode->i_sb->s_id,
594 		(unsigned long long)NFS_FILEID(inode),
595 		nfs_display_fhandle_hash(fh),
596 		atomic_read(&inode->i_count));
597 
598 out:
599 	return inode;
600 
601 out_no_inode:
602 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
603 	goto out;
604 }
605 EXPORT_SYMBOL_GPL(nfs_fhget);
606 
607 #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)
608 
609 int
nfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)610 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
611 	    struct iattr *attr)
612 {
613 	struct inode *inode = d_inode(dentry);
614 	struct nfs_fattr *fattr;
615 	int error = 0;
616 
617 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
618 
619 	/* skip mode change if it's just for clearing setuid/setgid */
620 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
621 		attr->ia_valid &= ~ATTR_MODE;
622 
623 	if (attr->ia_valid & ATTR_SIZE) {
624 		BUG_ON(!S_ISREG(inode->i_mode));
625 
626 		error = inode_newsize_ok(inode, attr->ia_size);
627 		if (error)
628 			return error;
629 
630 		if (attr->ia_size == i_size_read(inode))
631 			attr->ia_valid &= ~ATTR_SIZE;
632 	}
633 
634 	/* Optimization: if the end result is no change, don't RPC */
635 	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
636 		return 0;
637 
638 	trace_nfs_setattr_enter(inode);
639 
640 	/* Write all dirty data */
641 	if (S_ISREG(inode->i_mode))
642 		nfs_sync_inode(inode);
643 
644 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
645 	if (fattr == NULL) {
646 		error = -ENOMEM;
647 		goto out;
648 	}
649 
650 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
651 	if (error == 0)
652 		error = nfs_refresh_inode(inode, fattr);
653 	nfs_free_fattr(fattr);
654 out:
655 	trace_nfs_setattr_exit(inode, error);
656 	return error;
657 }
658 EXPORT_SYMBOL_GPL(nfs_setattr);
659 
660 /**
661  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
662  * @inode: inode of the file used
663  * @offset: file offset to start truncating
664  *
665  * This is a copy of the common vmtruncate, but with the locking
666  * corrected to take into account the fact that NFS requires
667  * inode->i_size to be updated under the inode->i_lock.
668  * Note: must be called with inode->i_lock held!
669  */
nfs_vmtruncate(struct inode * inode,loff_t offset)670 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
671 {
672 	int err;
673 
674 	err = inode_newsize_ok(inode, offset);
675 	if (err)
676 		goto out;
677 
678 	trace_nfs_size_truncate(inode, offset);
679 	i_size_write(inode, offset);
680 	/* Optimisation */
681 	if (offset == 0) {
682 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
683 		nfs_ooo_clear(NFS_I(inode));
684 	}
685 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
686 
687 	spin_unlock(&inode->i_lock);
688 	truncate_pagecache(inode, offset);
689 	spin_lock(&inode->i_lock);
690 out:
691 	return err;
692 }
693 
694 /**
695  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
696  * @inode: pointer to struct inode
697  * @attr: pointer to struct iattr
698  * @fattr: pointer to struct nfs_fattr
699  *
700  * Note: we do this in the *proc.c in order to ensure that
701  *       it works for things like exclusive creates too.
702  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)703 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
704 		struct nfs_fattr *fattr)
705 {
706 	/* Barrier: bump the attribute generation count. */
707 	nfs_fattr_set_barrier(fattr);
708 
709 	spin_lock(&inode->i_lock);
710 	NFS_I(inode)->attr_gencount = fattr->gencount;
711 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
712 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
713 						     NFS_INO_INVALID_BLOCKS);
714 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
715 		nfs_vmtruncate(inode, attr->ia_size);
716 	}
717 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
718 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
719 		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
720 		    inode->i_mode & S_ISUID)
721 			inode->i_mode &= ~S_ISUID;
722 		if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
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_set_ctime_to_ts(inode, 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_set_ctime_to_ts(inode, 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_set_ctime_to_ts(inode, 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 	if (!(cache_validity & NFS_INO_INVALID_CHANGE))
829 		reply_mask |= STATX_CHANGE_COOKIE;
830 	return reply_mask;
831 }
832 
nfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)833 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
834 		struct kstat *stat, u32 request_mask, unsigned int query_flags)
835 {
836 	struct inode *inode = d_inode(path->dentry);
837 	struct nfs_server *server = NFS_SERVER(inode);
838 	unsigned long cache_validity;
839 	int err = 0;
840 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
841 	bool do_update = false;
842 	bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
843 
844 	trace_nfs_getattr_enter(inode);
845 
846 	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
847 			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
848 			STATX_INO | STATX_SIZE | STATX_BLOCKS |
849 			STATX_CHANGE_COOKIE;
850 
851 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
852 		if (readdirplus_enabled)
853 			nfs_readdirplus_parent_cache_hit(path->dentry);
854 		goto out_no_revalidate;
855 	}
856 
857 	/* Flush out writes to the server in order to update c/mtime/version.  */
858 	if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
859 	    S_ISREG(inode->i_mode))
860 		filemap_write_and_wait(inode->i_mapping);
861 
862 	/*
863 	 * We may force a getattr if the user cares about atime.
864 	 *
865 	 * Note that we only have to check the vfsmount flags here:
866 	 *  - NFS always sets S_NOATIME by so checking it would give a
867 	 *    bogus result
868 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
869 	 *    no point in checking those.
870 	 */
871 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
872 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
873 		request_mask &= ~STATX_ATIME;
874 
875 	/* Is the user requesting attributes that might need revalidation? */
876 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
877 					STATX_MTIME|STATX_UID|STATX_GID|
878 					STATX_SIZE|STATX_BLOCKS|
879 					STATX_CHANGE_COOKIE)))
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 		if (readdirplus_enabled)
905 			nfs_readdirplus_parent_cache_miss(path->dentry);
906 		err = __nfs_revalidate_inode(server, inode);
907 		if (err)
908 			goto out;
909 	} else if (readdirplus_enabled)
910 		nfs_readdirplus_parent_cache_hit(path->dentry);
911 out_no_revalidate:
912 	/* Only return attributes that were revalidated. */
913 	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
914 
915 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
916 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
917 	stat->change_cookie = inode_peek_iversion_raw(inode);
918 	stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
919 	if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
920 		stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
921 	if (S_ISDIR(inode->i_mode))
922 		stat->blksize = NFS_SERVER(inode)->dtsize;
923 out:
924 	trace_nfs_getattr_exit(inode, err);
925 	return err;
926 }
927 EXPORT_SYMBOL_GPL(nfs_getattr);
928 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)929 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
930 {
931 	refcount_set(&l_ctx->count, 1);
932 	l_ctx->lockowner = current->files;
933 	INIT_LIST_HEAD(&l_ctx->list);
934 	atomic_set(&l_ctx->io_count, 0);
935 }
936 
__nfs_find_lock_context(struct nfs_open_context * ctx)937 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
938 {
939 	struct nfs_lock_context *pos;
940 
941 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
942 		if (pos->lockowner != current->files)
943 			continue;
944 		if (refcount_inc_not_zero(&pos->count))
945 			return pos;
946 	}
947 	return NULL;
948 }
949 
nfs_get_lock_context(struct nfs_open_context * ctx)950 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
951 {
952 	struct nfs_lock_context *res, *new = NULL;
953 	struct inode *inode = d_inode(ctx->dentry);
954 
955 	rcu_read_lock();
956 	res = __nfs_find_lock_context(ctx);
957 	rcu_read_unlock();
958 	if (res == NULL) {
959 		new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
960 		if (new == NULL)
961 			return ERR_PTR(-ENOMEM);
962 		nfs_init_lock_context(new);
963 		spin_lock(&inode->i_lock);
964 		res = __nfs_find_lock_context(ctx);
965 		if (res == NULL) {
966 			new->open_context = get_nfs_open_context(ctx);
967 			if (new->open_context) {
968 				list_add_tail_rcu(&new->list,
969 						&ctx->lock_context.list);
970 				res = new;
971 				new = NULL;
972 			} else
973 				res = ERR_PTR(-EBADF);
974 		}
975 		spin_unlock(&inode->i_lock);
976 		kfree(new);
977 	}
978 	return res;
979 }
980 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
981 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)982 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
983 {
984 	struct nfs_open_context *ctx = l_ctx->open_context;
985 	struct inode *inode = d_inode(ctx->dentry);
986 
987 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
988 		return;
989 	list_del_rcu(&l_ctx->list);
990 	spin_unlock(&inode->i_lock);
991 	put_nfs_open_context(ctx);
992 	kfree_rcu(l_ctx, rcu_head);
993 }
994 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
995 
996 /**
997  * nfs_close_context - Common close_context() routine NFSv2/v3
998  * @ctx: pointer to context
999  * @is_sync: is this a synchronous close
1000  *
1001  * Ensure that the attributes are up to date if we're mounted
1002  * with close-to-open semantics and we have cached data that will
1003  * need to be revalidated on open.
1004  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)1005 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1006 {
1007 	struct nfs_inode *nfsi;
1008 	struct inode *inode;
1009 
1010 	if (!(ctx->mode & FMODE_WRITE))
1011 		return;
1012 	if (!is_sync)
1013 		return;
1014 	inode = d_inode(ctx->dentry);
1015 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1016 		return;
1017 	nfsi = NFS_I(inode);
1018 	if (inode->i_mapping->nrpages == 0)
1019 		return;
1020 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1021 		return;
1022 	if (!list_empty(&nfsi->open_files))
1023 		return;
1024 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1025 		return;
1026 	nfs_revalidate_inode(inode,
1027 			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1028 }
1029 EXPORT_SYMBOL_GPL(nfs_close_context);
1030 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1031 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1032 						fmode_t f_mode,
1033 						struct file *filp)
1034 {
1035 	struct nfs_open_context *ctx;
1036 
1037 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1038 	if (!ctx)
1039 		return ERR_PTR(-ENOMEM);
1040 	nfs_sb_active(dentry->d_sb);
1041 	ctx->dentry = dget(dentry);
1042 	if (filp)
1043 		ctx->cred = get_cred(filp->f_cred);
1044 	else
1045 		ctx->cred = get_current_cred();
1046 	rcu_assign_pointer(ctx->ll_cred, NULL);
1047 	ctx->state = NULL;
1048 	ctx->mode = f_mode;
1049 	ctx->flags = 0;
1050 	ctx->error = 0;
1051 	ctx->flock_owner = (fl_owner_t)filp;
1052 	nfs_init_lock_context(&ctx->lock_context);
1053 	ctx->lock_context.open_context = ctx;
1054 	INIT_LIST_HEAD(&ctx->list);
1055 	ctx->mdsthreshold = NULL;
1056 	return ctx;
1057 }
1058 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1059 
get_nfs_open_context(struct nfs_open_context * ctx)1060 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1061 {
1062 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1063 		return ctx;
1064 	return NULL;
1065 }
1066 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1067 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1068 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1069 {
1070 	struct inode *inode = d_inode(ctx->dentry);
1071 	struct super_block *sb = ctx->dentry->d_sb;
1072 
1073 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1074 		return;
1075 	if (!list_empty(&ctx->list)) {
1076 		spin_lock(&inode->i_lock);
1077 		list_del_rcu(&ctx->list);
1078 		spin_unlock(&inode->i_lock);
1079 	}
1080 	if (inode != NULL)
1081 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1082 	put_cred(ctx->cred);
1083 	dput(ctx->dentry);
1084 	nfs_sb_deactive(sb);
1085 	put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1086 	kfree(ctx->mdsthreshold);
1087 	kfree_rcu(ctx, rcu_head);
1088 }
1089 
put_nfs_open_context(struct nfs_open_context * ctx)1090 void put_nfs_open_context(struct nfs_open_context *ctx)
1091 {
1092 	__put_nfs_open_context(ctx, 0);
1093 }
1094 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1095 
put_nfs_open_context_sync(struct nfs_open_context * ctx)1096 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1097 {
1098 	__put_nfs_open_context(ctx, 1);
1099 }
1100 
1101 /*
1102  * Ensure that mmap has a recent RPC credential for use when writing out
1103  * shared pages
1104  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1105 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1106 {
1107 	struct inode *inode = d_inode(ctx->dentry);
1108 	struct nfs_inode *nfsi = NFS_I(inode);
1109 
1110 	spin_lock(&inode->i_lock);
1111 	if (list_empty(&nfsi->open_files) &&
1112 	    nfs_ooo_test(nfsi))
1113 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1114 						     NFS_INO_REVAL_FORCED);
1115 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1116 	spin_unlock(&inode->i_lock);
1117 }
1118 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1119 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1120 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1121 {
1122 	filp->private_data = get_nfs_open_context(ctx);
1123 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1124 	if (list_empty(&ctx->list))
1125 		nfs_inode_attach_open_context(ctx);
1126 }
1127 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1128 
1129 /*
1130  * Given an inode, search for an open context with the desired characteristics
1131  */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1132 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1133 {
1134 	struct nfs_inode *nfsi = NFS_I(inode);
1135 	struct nfs_open_context *pos, *ctx = NULL;
1136 
1137 	rcu_read_lock();
1138 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1139 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1140 			continue;
1141 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1142 			continue;
1143 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1144 			continue;
1145 		ctx = get_nfs_open_context(pos);
1146 		if (ctx)
1147 			break;
1148 	}
1149 	rcu_read_unlock();
1150 	return ctx;
1151 }
1152 
nfs_file_clear_open_context(struct file * filp)1153 void nfs_file_clear_open_context(struct file *filp)
1154 {
1155 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1156 
1157 	if (ctx) {
1158 		struct inode *inode = d_inode(ctx->dentry);
1159 
1160 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1161 		/*
1162 		 * We fatal error on write before. Try to writeback
1163 		 * every page again.
1164 		 */
1165 		if (ctx->error < 0)
1166 			invalidate_inode_pages2(inode->i_mapping);
1167 		filp->private_data = NULL;
1168 		put_nfs_open_context_sync(ctx);
1169 	}
1170 }
1171 
1172 /*
1173  * These allocate and release file read/write context information.
1174  */
nfs_open(struct inode * inode,struct file * filp)1175 int nfs_open(struct inode *inode, struct file *filp)
1176 {
1177 	struct nfs_open_context *ctx;
1178 
1179 	ctx = alloc_nfs_open_context(file_dentry(filp),
1180 				     flags_to_mode(filp->f_flags), filp);
1181 	if (IS_ERR(ctx))
1182 		return PTR_ERR(ctx);
1183 	nfs_file_set_open_context(filp, ctx);
1184 	put_nfs_open_context(ctx);
1185 	nfs_fscache_open_file(inode, filp);
1186 	return 0;
1187 }
1188 
1189 /*
1190  * This function is called whenever some part of NFS notices that
1191  * the cached attributes have to be refreshed.
1192  */
1193 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1194 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1195 {
1196 	int		 status = -ESTALE;
1197 	struct nfs_fattr *fattr = NULL;
1198 	struct nfs_inode *nfsi = NFS_I(inode);
1199 
1200 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1201 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1202 
1203 	trace_nfs_revalidate_inode_enter(inode);
1204 
1205 	if (is_bad_inode(inode))
1206 		goto out;
1207 	if (NFS_STALE(inode))
1208 		goto out;
1209 
1210 	/* pNFS: Attributes aren't updated until we layoutcommit */
1211 	if (S_ISREG(inode->i_mode)) {
1212 		status = pnfs_sync_inode(inode, false);
1213 		if (status)
1214 			goto out;
1215 	}
1216 
1217 	status = -ENOMEM;
1218 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1219 	if (fattr == NULL)
1220 		goto out;
1221 
1222 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1223 
1224 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1225 	if (status != 0) {
1226 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1227 			 inode->i_sb->s_id,
1228 			 (unsigned long long)NFS_FILEID(inode), status);
1229 		switch (status) {
1230 		case -ETIMEDOUT:
1231 			/* A soft timeout occurred. Use cached information? */
1232 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1233 				status = 0;
1234 			break;
1235 		case -ESTALE:
1236 			if (!S_ISDIR(inode->i_mode))
1237 				nfs_set_inode_stale(inode);
1238 			else
1239 				nfs_zap_caches(inode);
1240 		}
1241 		goto out;
1242 	}
1243 
1244 	status = nfs_refresh_inode(inode, fattr);
1245 	if (status) {
1246 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1247 			 inode->i_sb->s_id,
1248 			 (unsigned long long)NFS_FILEID(inode), status);
1249 		goto out;
1250 	}
1251 
1252 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1253 		nfs_zap_acl_cache(inode);
1254 
1255 	nfs_setsecurity(inode, fattr);
1256 
1257 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1258 		inode->i_sb->s_id,
1259 		(unsigned long long)NFS_FILEID(inode));
1260 
1261 out:
1262 	nfs_free_fattr(fattr);
1263 	trace_nfs_revalidate_inode_exit(inode, status);
1264 	return status;
1265 }
1266 
nfs_attribute_cache_expired(struct inode * inode)1267 int nfs_attribute_cache_expired(struct inode *inode)
1268 {
1269 	if (nfs_have_delegated_attributes(inode))
1270 		return 0;
1271 	return nfs_attribute_timeout(inode);
1272 }
1273 
1274 /**
1275  * nfs_revalidate_inode - Revalidate the inode attributes
1276  * @inode: pointer to inode struct
1277  * @flags: cache flags to check
1278  *
1279  * Updates inode attribute information by retrieving the data from the server.
1280  */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1281 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1282 {
1283 	if (!nfs_check_cache_invalid(inode, flags))
1284 		return NFS_STALE(inode) ? -ESTALE : 0;
1285 	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1286 }
1287 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1288 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1289 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1290 {
1291 	int ret;
1292 
1293 	nfs_fscache_invalidate(inode, 0);
1294 	if (mapping->nrpages != 0) {
1295 		if (S_ISREG(inode->i_mode)) {
1296 			ret = nfs_sync_mapping(mapping);
1297 			if (ret < 0)
1298 				return ret;
1299 		}
1300 		ret = invalidate_inode_pages2(mapping);
1301 		if (ret < 0)
1302 			return ret;
1303 	}
1304 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1305 
1306 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1307 			inode->i_sb->s_id,
1308 			(unsigned long long)NFS_FILEID(inode));
1309 	return 0;
1310 }
1311 
1312 /**
1313  * nfs_clear_invalid_mapping - Conditionally clear a mapping
1314  * @mapping: pointer to mapping
1315  *
1316  * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1317  */
nfs_clear_invalid_mapping(struct address_space * mapping)1318 int nfs_clear_invalid_mapping(struct address_space *mapping)
1319 {
1320 	struct inode *inode = mapping->host;
1321 	struct nfs_inode *nfsi = NFS_I(inode);
1322 	unsigned long *bitlock = &nfsi->flags;
1323 	int ret = 0;
1324 
1325 	/*
1326 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1327 	 * invalidations that come in while we're shooting down the mappings
1328 	 * are respected. But, that leaves a race window where one revalidator
1329 	 * can clear the flag, and then another checks it before the mapping
1330 	 * gets invalidated. Fix that by serializing access to this part of
1331 	 * the function.
1332 	 *
1333 	 * At the same time, we need to allow other tasks to see whether we
1334 	 * might be in the middle of invalidating the pages, so we only set
1335 	 * the bit lock here if it looks like we're going to be doing that.
1336 	 */
1337 	for (;;) {
1338 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1339 					 nfs_wait_bit_killable,
1340 					 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1341 		if (ret)
1342 			goto out;
1343 		spin_lock(&inode->i_lock);
1344 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1345 			spin_unlock(&inode->i_lock);
1346 			continue;
1347 		}
1348 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1349 			break;
1350 		spin_unlock(&inode->i_lock);
1351 		goto out;
1352 	}
1353 
1354 	set_bit(NFS_INO_INVALIDATING, bitlock);
1355 	smp_wmb();
1356 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1357 	nfs_ooo_clear(nfsi);
1358 	spin_unlock(&inode->i_lock);
1359 	trace_nfs_invalidate_mapping_enter(inode);
1360 	ret = nfs_invalidate_mapping(inode, mapping);
1361 	trace_nfs_invalidate_mapping_exit(inode, ret);
1362 
1363 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1364 	smp_mb__after_atomic();
1365 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1366 out:
1367 	return ret;
1368 }
1369 
nfs_mapping_need_revalidate_inode(struct inode * inode)1370 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1371 {
1372 	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1373 		NFS_STALE(inode);
1374 }
1375 
nfs_revalidate_mapping_rcu(struct inode * inode)1376 int nfs_revalidate_mapping_rcu(struct inode *inode)
1377 {
1378 	struct nfs_inode *nfsi = NFS_I(inode);
1379 	unsigned long *bitlock = &nfsi->flags;
1380 	int ret = 0;
1381 
1382 	if (IS_SWAPFILE(inode))
1383 		goto out;
1384 	if (nfs_mapping_need_revalidate_inode(inode)) {
1385 		ret = -ECHILD;
1386 		goto out;
1387 	}
1388 	spin_lock(&inode->i_lock);
1389 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1390 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1391 		ret = -ECHILD;
1392 	spin_unlock(&inode->i_lock);
1393 out:
1394 	return ret;
1395 }
1396 
1397 /**
1398  * nfs_revalidate_mapping - Revalidate the pagecache
1399  * @inode: pointer to host inode
1400  * @mapping: pointer to mapping
1401  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1402 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1403 {
1404 	/* swapfiles are not supposed to be shared. */
1405 	if (IS_SWAPFILE(inode))
1406 		return 0;
1407 
1408 	if (nfs_mapping_need_revalidate_inode(inode)) {
1409 		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1410 		if (ret < 0)
1411 			return ret;
1412 	}
1413 
1414 	return nfs_clear_invalid_mapping(mapping);
1415 }
1416 
nfs_file_has_writers(struct nfs_inode * nfsi)1417 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1418 {
1419 	struct inode *inode = &nfsi->vfs_inode;
1420 
1421 	if (!S_ISREG(inode->i_mode))
1422 		return false;
1423 	if (list_empty(&nfsi->open_files))
1424 		return false;
1425 	return inode_is_open_for_write(inode);
1426 }
1427 
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1428 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1429 {
1430 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1431 }
1432 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1433 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1434 {
1435 	struct timespec64 ts;
1436 
1437 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1438 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1439 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1440 		inode_set_iversion_raw(inode, fattr->change_attr);
1441 		if (S_ISDIR(inode->i_mode))
1442 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1443 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1444 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1445 	}
1446 	/* If we have atomic WCC data, we may update some attributes */
1447 	ts = inode_get_ctime(inode);
1448 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1449 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1450 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1451 		inode_set_ctime_to_ts(inode, fattr->ctime);
1452 	}
1453 
1454 	ts = inode->i_mtime;
1455 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1456 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1457 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1458 		inode->i_mtime = fattr->mtime;
1459 	}
1460 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1461 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1462 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1463 			&& !nfs_have_writebacks(inode)) {
1464 		trace_nfs_size_wcc(inode, fattr->size);
1465 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1466 	}
1467 }
1468 
1469 /**
1470  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1471  * @inode: pointer to inode
1472  * @fattr: updated attributes
1473  *
1474  * Verifies the attribute cache. If we have just changed the attributes,
1475  * so that fattr carries weak cache consistency data, then it may
1476  * also update the ctime/mtime/change_attribute.
1477  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1478 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1479 {
1480 	struct nfs_inode *nfsi = NFS_I(inode);
1481 	loff_t cur_size, new_isize;
1482 	unsigned long invalid = 0;
1483 	struct timespec64 ts;
1484 
1485 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1486 		return 0;
1487 
1488 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1489 		/* Only a mounted-on-fileid? Just exit */
1490 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1491 			return 0;
1492 	/* Has the inode gone and changed behind our back? */
1493 	} else if (nfsi->fileid != fattr->fileid) {
1494 		/* Is this perhaps the mounted-on fileid? */
1495 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1496 		    nfsi->fileid == fattr->mounted_on_fileid)
1497 			return 0;
1498 		return -ESTALE;
1499 	}
1500 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1501 		return -ESTALE;
1502 
1503 
1504 	if (!nfs_file_has_buffered_writers(nfsi)) {
1505 		/* Verify a few of the more important attributes */
1506 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1507 			invalid |= NFS_INO_INVALID_CHANGE;
1508 
1509 		ts = inode->i_mtime;
1510 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1511 			invalid |= NFS_INO_INVALID_MTIME;
1512 
1513 		ts = inode_get_ctime(inode);
1514 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1515 			invalid |= NFS_INO_INVALID_CTIME;
1516 
1517 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1518 			cur_size = i_size_read(inode);
1519 			new_isize = nfs_size_to_loff_t(fattr->size);
1520 			if (cur_size != new_isize)
1521 				invalid |= NFS_INO_INVALID_SIZE;
1522 		}
1523 	}
1524 
1525 	/* Have any file permissions changed? */
1526 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1527 		invalid |= NFS_INO_INVALID_MODE;
1528 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1529 		invalid |= NFS_INO_INVALID_OTHER;
1530 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1531 		invalid |= NFS_INO_INVALID_OTHER;
1532 
1533 	/* Has the link count changed? */
1534 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1535 		invalid |= NFS_INO_INVALID_NLINK;
1536 
1537 	ts = inode->i_atime;
1538 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1539 		invalid |= NFS_INO_INVALID_ATIME;
1540 
1541 	if (invalid != 0)
1542 		nfs_set_cache_invalid(inode, invalid);
1543 
1544 	nfsi->read_cache_jiffies = fattr->time_start;
1545 	return 0;
1546 }
1547 
1548 static atomic_long_t nfs_attr_generation_counter;
1549 
nfs_read_attr_generation_counter(void)1550 static unsigned long nfs_read_attr_generation_counter(void)
1551 {
1552 	return atomic_long_read(&nfs_attr_generation_counter);
1553 }
1554 
nfs_inc_attr_generation_counter(void)1555 unsigned long nfs_inc_attr_generation_counter(void)
1556 {
1557 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1558 }
1559 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1560 
nfs_fattr_init(struct nfs_fattr * fattr)1561 void nfs_fattr_init(struct nfs_fattr *fattr)
1562 {
1563 	fattr->valid = 0;
1564 	fattr->time_start = jiffies;
1565 	fattr->gencount = nfs_inc_attr_generation_counter();
1566 	fattr->owner_name = NULL;
1567 	fattr->group_name = NULL;
1568 }
1569 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1570 
1571 /**
1572  * nfs_fattr_set_barrier
1573  * @fattr: attributes
1574  *
1575  * Used to set a barrier after an attribute was updated. This
1576  * barrier ensures that older attributes from RPC calls that may
1577  * have raced with our update cannot clobber these new values.
1578  * Note that you are still responsible for ensuring that other
1579  * operations which change the attribute on the server do not
1580  * collide.
1581  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1582 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1583 {
1584 	fattr->gencount = nfs_inc_attr_generation_counter();
1585 }
1586 
nfs_alloc_fattr(void)1587 struct nfs_fattr *nfs_alloc_fattr(void)
1588 {
1589 	struct nfs_fattr *fattr;
1590 
1591 	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1592 	if (fattr != NULL) {
1593 		nfs_fattr_init(fattr);
1594 		fattr->label = NULL;
1595 	}
1596 	return fattr;
1597 }
1598 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1599 
nfs_alloc_fattr_with_label(struct nfs_server * server)1600 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1601 {
1602 	struct nfs_fattr *fattr = nfs_alloc_fattr();
1603 
1604 	if (!fattr)
1605 		return NULL;
1606 
1607 	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1608 	if (IS_ERR(fattr->label)) {
1609 		kfree(fattr);
1610 		return NULL;
1611 	}
1612 
1613 	return fattr;
1614 }
1615 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1616 
nfs_alloc_fhandle(void)1617 struct nfs_fh *nfs_alloc_fhandle(void)
1618 {
1619 	struct nfs_fh *fh;
1620 
1621 	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1622 	if (fh != NULL)
1623 		fh->size = 0;
1624 	return fh;
1625 }
1626 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1627 
1628 #ifdef NFS_DEBUG
1629 /*
1630  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1631  *                             in the same way that wireshark does
1632  *
1633  * @fh: file handle
1634  *
1635  * For debugging only.
1636  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1637 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1638 {
1639 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1640 	 * not on the result */
1641 	return nfs_fhandle_hash(fh);
1642 }
1643 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1644 
1645 /*
1646  * _nfs_display_fhandle - display an NFS file handle on the console
1647  *
1648  * @fh: file handle to display
1649  * @caption: display caption
1650  *
1651  * For debugging only.
1652  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1653 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1654 {
1655 	unsigned short i;
1656 
1657 	if (fh == NULL || fh->size == 0) {
1658 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1659 		return;
1660 	}
1661 
1662 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1663 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1664 	for (i = 0; i < fh->size; i += 16) {
1665 		__be32 *pos = (__be32 *)&fh->data[i];
1666 
1667 		switch ((fh->size - i - 1) >> 2) {
1668 		case 0:
1669 			printk(KERN_DEFAULT " %08x\n",
1670 				be32_to_cpup(pos));
1671 			break;
1672 		case 1:
1673 			printk(KERN_DEFAULT " %08x %08x\n",
1674 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1675 			break;
1676 		case 2:
1677 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1678 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1679 				be32_to_cpup(pos + 2));
1680 			break;
1681 		default:
1682 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1683 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1684 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1685 		}
1686 	}
1687 }
1688 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1689 #endif
1690 
1691 /**
1692  * nfs_inode_attrs_cmp_generic - compare attributes
1693  * @fattr: attributes
1694  * @inode: pointer to inode
1695  *
1696  * Attempt to divine whether or not an RPC call reply carrying stale
1697  * attributes got scheduled after another call carrying updated ones.
1698  * Note also the check for wraparound of 'attr_gencount'
1699  *
1700  * The function returns '1' if it thinks the attributes in @fattr are
1701  * more recent than the ones cached in @inode. Otherwise it returns
1702  * the value '0'.
1703  */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1704 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1705 				       const struct inode *inode)
1706 {
1707 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1708 
1709 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1710 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1711 }
1712 
1713 /**
1714  * nfs_inode_attrs_cmp_monotonic - compare attributes
1715  * @fattr: attributes
1716  * @inode: pointer to inode
1717  *
1718  * Attempt to divine whether or not an RPC call reply carrying stale
1719  * attributes got scheduled after another call carrying updated ones.
1720  *
1721  * We assume that the server observes monotonic semantics for
1722  * the change attribute, so a larger value means that the attributes in
1723  * @fattr are more recent, in which case the function returns the
1724  * value '1'.
1725  * A return value of '0' indicates no measurable change
1726  * A return value of '-1' means that the attributes in @inode are
1727  * more recent.
1728  */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1729 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1730 					 const struct inode *inode)
1731 {
1732 	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1733 	if (diff > 0)
1734 		return 1;
1735 	return diff == 0 ? 0 : -1;
1736 }
1737 
1738 /**
1739  * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1740  * @fattr: attributes
1741  * @inode: pointer to inode
1742  *
1743  * Attempt to divine whether or not an RPC call reply carrying stale
1744  * attributes got scheduled after another call carrying updated ones.
1745  *
1746  * We assume that the server observes strictly monotonic semantics for
1747  * the change attribute, so a larger value means that the attributes in
1748  * @fattr are more recent, in which case the function returns the
1749  * value '1'.
1750  * A return value of '-1' means that the attributes in @inode are
1751  * more recent or unchanged.
1752  */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1753 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1754 						const struct inode *inode)
1755 {
1756 	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1757 }
1758 
1759 /**
1760  * nfs_inode_attrs_cmp - compare attributes
1761  * @fattr: attributes
1762  * @inode: pointer to inode
1763  *
1764  * This function returns '1' if it thinks the attributes in @fattr are
1765  * more recent than the ones cached in @inode. It returns '-1' if
1766  * the attributes in @inode are more recent than the ones in @fattr,
1767  * and it returns 0 if not sure.
1768  */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1769 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1770 			       const struct inode *inode)
1771 {
1772 	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1773 		return 1;
1774 	switch (NFS_SERVER(inode)->change_attr_type) {
1775 	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1776 		break;
1777 	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1778 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1779 			break;
1780 		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1781 	default:
1782 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1783 			break;
1784 		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1785 	}
1786 	return 0;
1787 }
1788 
1789 /**
1790  * nfs_inode_finish_partial_attr_update - complete a previous inode update
1791  * @fattr: attributes
1792  * @inode: pointer to inode
1793  *
1794  * Returns '1' if the last attribute update left the inode cached
1795  * attributes in a partially unrevalidated state, and @fattr
1796  * matches the change attribute of that partial update.
1797  * Otherwise returns '0'.
1798  */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1799 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1800 						const struct inode *inode)
1801 {
1802 	const unsigned long check_valid =
1803 		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1804 		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1805 		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1806 		NFS_INO_INVALID_NLINK;
1807 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1808 	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1809 
1810 	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1811 	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1812 	    (cache_validity & check_valid) != 0 &&
1813 	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1814 	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1815 		return 1;
1816 	return 0;
1817 }
1818 
nfs_ooo_merge(struct nfs_inode * nfsi,u64 start,u64 end)1819 static void nfs_ooo_merge(struct nfs_inode *nfsi,
1820 			  u64 start, u64 end)
1821 {
1822 	int i, cnt;
1823 
1824 	if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1825 		/* No point merging anything */
1826 		return;
1827 
1828 	if (!nfsi->ooo) {
1829 		nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1830 		if (!nfsi->ooo) {
1831 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1832 			return;
1833 		}
1834 		nfsi->ooo->cnt = 0;
1835 	}
1836 
1837 	/* add this range, merging if possible */
1838 	cnt = nfsi->ooo->cnt;
1839 	for (i = 0; i < cnt; i++) {
1840 		if (end == nfsi->ooo->gap[i].start)
1841 			end = nfsi->ooo->gap[i].end;
1842 		else if (start == nfsi->ooo->gap[i].end)
1843 			start = nfsi->ooo->gap[i].start;
1844 		else
1845 			continue;
1846 		/* Remove 'i' from table and loop to insert the new range */
1847 		cnt -= 1;
1848 		nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
1849 		i = -1;
1850 	}
1851 	if (start != end) {
1852 		if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
1853 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1854 			kfree(nfsi->ooo);
1855 			nfsi->ooo = NULL;
1856 			return;
1857 		}
1858 		nfsi->ooo->gap[cnt].start = start;
1859 		nfsi->ooo->gap[cnt].end = end;
1860 		cnt += 1;
1861 	}
1862 	nfsi->ooo->cnt = cnt;
1863 }
1864 
nfs_ooo_record(struct nfs_inode * nfsi,struct nfs_fattr * fattr)1865 static void nfs_ooo_record(struct nfs_inode *nfsi,
1866 			   struct nfs_fattr *fattr)
1867 {
1868 	/* This reply was out-of-order, so record in the
1869 	 * pre/post change id, possibly cancelling
1870 	 * gaps created when iversion was jumpped forward.
1871 	 */
1872 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
1873 	    (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
1874 		nfs_ooo_merge(nfsi,
1875 			      fattr->change_attr,
1876 			      fattr->pre_change_attr);
1877 }
1878 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1879 static int nfs_refresh_inode_locked(struct inode *inode,
1880 				    struct nfs_fattr *fattr)
1881 {
1882 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1883 	int ret = 0;
1884 
1885 	trace_nfs_refresh_inode_enter(inode);
1886 
1887 	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1888 		ret = nfs_update_inode(inode, fattr);
1889 	else {
1890 		nfs_ooo_record(NFS_I(inode), fattr);
1891 
1892 		if (attr_cmp == 0)
1893 			ret = nfs_check_inode_attributes(inode, fattr);
1894 	}
1895 
1896 	trace_nfs_refresh_inode_exit(inode, ret);
1897 	return ret;
1898 }
1899 
1900 /**
1901  * nfs_refresh_inode - try to update the inode attribute cache
1902  * @inode: pointer to inode
1903  * @fattr: updated attributes
1904  *
1905  * Check that an RPC call that returned attributes has not overlapped with
1906  * other recent updates of the inode metadata, then decide whether it is
1907  * safe to do a full update of the inode attributes, or whether just to
1908  * call nfs_check_inode_attributes.
1909  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1910 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1911 {
1912 	int status;
1913 
1914 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1915 		return 0;
1916 	spin_lock(&inode->i_lock);
1917 	status = nfs_refresh_inode_locked(inode, fattr);
1918 	spin_unlock(&inode->i_lock);
1919 
1920 	return status;
1921 }
1922 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1923 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1924 static int nfs_post_op_update_inode_locked(struct inode *inode,
1925 		struct nfs_fattr *fattr, unsigned int invalid)
1926 {
1927 	if (S_ISDIR(inode->i_mode))
1928 		invalid |= NFS_INO_INVALID_DATA;
1929 	nfs_set_cache_invalid(inode, invalid);
1930 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1931 		return 0;
1932 	return nfs_refresh_inode_locked(inode, fattr);
1933 }
1934 
1935 /**
1936  * nfs_post_op_update_inode - try to update the inode attribute cache
1937  * @inode: pointer to inode
1938  * @fattr: updated attributes
1939  *
1940  * After an operation that has changed the inode metadata, mark the
1941  * attribute cache as being invalid, then try to update it.
1942  *
1943  * NB: if the server didn't return any post op attributes, this
1944  * function will force the retrieval of attributes before the next
1945  * NFS request.  Thus it should be used only for operations that
1946  * are expected to change one or more attributes, to avoid
1947  * unnecessary NFS requests and trips through nfs_update_inode().
1948  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1949 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1950 {
1951 	int status;
1952 
1953 	spin_lock(&inode->i_lock);
1954 	nfs_fattr_set_barrier(fattr);
1955 	status = nfs_post_op_update_inode_locked(inode, fattr,
1956 			NFS_INO_INVALID_CHANGE
1957 			| NFS_INO_INVALID_CTIME
1958 			| NFS_INO_REVAL_FORCED);
1959 	spin_unlock(&inode->i_lock);
1960 
1961 	return status;
1962 }
1963 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1964 
1965 /**
1966  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1967  * @inode: pointer to inode
1968  * @fattr: updated attributes
1969  *
1970  * After an operation that has changed the inode metadata, mark the
1971  * attribute cache as being invalid, then try to update it. Fake up
1972  * weak cache consistency data, if none exist.
1973  *
1974  * This function is mainly designed to be used by the ->write_done() functions.
1975  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1976 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1977 {
1978 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1979 	int status;
1980 
1981 	/* Don't do a WCC update if these attributes are already stale */
1982 	if (attr_cmp < 0)
1983 		return 0;
1984 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1985 		/* Record the pre/post change info before clearing PRECHANGE */
1986 		nfs_ooo_record(NFS_I(inode), fattr);
1987 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1988 				| NFS_ATTR_FATTR_PRESIZE
1989 				| NFS_ATTR_FATTR_PREMTIME
1990 				| NFS_ATTR_FATTR_PRECTIME);
1991 		goto out_noforce;
1992 	}
1993 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1994 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1995 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1996 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1997 	}
1998 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1999 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2000 		fattr->pre_ctime = inode_get_ctime(inode);
2001 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2002 	}
2003 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2004 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2005 		fattr->pre_mtime = inode->i_mtime;
2006 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2007 	}
2008 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2009 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2010 		fattr->pre_size = i_size_read(inode);
2011 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2012 	}
2013 out_noforce:
2014 	status = nfs_post_op_update_inode_locked(inode, fattr,
2015 			NFS_INO_INVALID_CHANGE
2016 			| NFS_INO_INVALID_CTIME
2017 			| NFS_INO_INVALID_MTIME
2018 			| NFS_INO_INVALID_BLOCKS);
2019 	return status;
2020 }
2021 
2022 /**
2023  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2024  * @inode: pointer to inode
2025  * @fattr: updated attributes
2026  *
2027  * After an operation that has changed the inode metadata, mark the
2028  * attribute cache as being invalid, then try to update it. Fake up
2029  * weak cache consistency data, if none exist.
2030  *
2031  * This function is mainly designed to be used by the ->write_done() functions.
2032  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)2033 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2034 {
2035 	int status;
2036 
2037 	spin_lock(&inode->i_lock);
2038 	nfs_fattr_set_barrier(fattr);
2039 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2040 	spin_unlock(&inode->i_lock);
2041 	return status;
2042 }
2043 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2044 
2045 
2046 /*
2047  * Many nfs protocol calls return the new file attributes after
2048  * an operation.  Here we update the inode to reflect the state
2049  * of the server's inode.
2050  *
2051  * This is a bit tricky because we have to make sure all dirty pages
2052  * have been sent off to the server before calling invalidate_inode_pages.
2053  * To make sure no other process adds more write requests while we try
2054  * our best to flush them, we make them sleep during the attribute refresh.
2055  *
2056  * A very similar scenario holds for the dir cache.
2057  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)2058 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2059 {
2060 	struct nfs_server *server = NFS_SERVER(inode);
2061 	struct nfs_inode *nfsi = NFS_I(inode);
2062 	loff_t cur_isize, new_isize;
2063 	u64 fattr_supported = server->fattr_valid;
2064 	unsigned long invalid = 0;
2065 	unsigned long now = jiffies;
2066 	unsigned long save_cache_validity;
2067 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2068 	bool cache_revalidated = true;
2069 	bool attr_changed = false;
2070 	bool have_delegation;
2071 
2072 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2073 			__func__, inode->i_sb->s_id, inode->i_ino,
2074 			nfs_display_fhandle_hash(NFS_FH(inode)),
2075 			atomic_read(&inode->i_count), fattr->valid);
2076 
2077 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2078 		/* Only a mounted-on-fileid? Just exit */
2079 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2080 			return 0;
2081 	/* Has the inode gone and changed behind our back? */
2082 	} else if (nfsi->fileid != fattr->fileid) {
2083 		/* Is this perhaps the mounted-on fileid? */
2084 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2085 		    nfsi->fileid == fattr->mounted_on_fileid)
2086 			return 0;
2087 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2088 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2089 			NFS_SERVER(inode)->nfs_client->cl_hostname,
2090 			inode->i_sb->s_id, (long long)nfsi->fileid,
2091 			(long long)fattr->fileid);
2092 		goto out_err;
2093 	}
2094 
2095 	/*
2096 	 * Make sure the inode's type hasn't changed.
2097 	 */
2098 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2099 		/*
2100 		* Big trouble! The inode has become a different object.
2101 		*/
2102 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2103 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2104 		goto out_err;
2105 	}
2106 
2107 	/* Update the fsid? */
2108 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2109 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2110 			!IS_AUTOMOUNT(inode))
2111 		server->fsid = fattr->fsid;
2112 
2113 	/* Save the delegation state before clearing cache_validity */
2114 	have_delegation = nfs_have_delegated_attributes(inode);
2115 
2116 	/*
2117 	 * Update the read time so we don't revalidate too often.
2118 	 */
2119 	nfsi->read_cache_jiffies = fattr->time_start;
2120 
2121 	save_cache_validity = nfsi->cache_validity;
2122 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2123 			| NFS_INO_INVALID_ATIME
2124 			| NFS_INO_REVAL_FORCED
2125 			| NFS_INO_INVALID_BLOCKS);
2126 
2127 	/* Do atomic weak cache consistency updates */
2128 	nfs_wcc_update_inode(inode, fattr);
2129 
2130 	if (pnfs_layoutcommit_outstanding(inode)) {
2131 		nfsi->cache_validity |=
2132 			save_cache_validity &
2133 			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2134 			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2135 			 NFS_INO_INVALID_BLOCKS);
2136 		cache_revalidated = false;
2137 	}
2138 
2139 	/* More cache consistency checks */
2140 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2141 		if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2142 		    nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2143 			/* There is one remaining gap that hasn't been
2144 			 * merged into iversion - do that now.
2145 			 */
2146 			inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2147 			kfree(nfsi->ooo);
2148 			nfsi->ooo = NULL;
2149 		}
2150 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2151 			/* Could it be a race with writeback? */
2152 			if (!(have_writers || have_delegation)) {
2153 				invalid |= NFS_INO_INVALID_DATA
2154 					| NFS_INO_INVALID_ACCESS
2155 					| NFS_INO_INVALID_ACL
2156 					| NFS_INO_INVALID_XATTR;
2157 				/* Force revalidate of all attributes */
2158 				save_cache_validity |= NFS_INO_INVALID_CTIME
2159 					| NFS_INO_INVALID_MTIME
2160 					| NFS_INO_INVALID_SIZE
2161 					| NFS_INO_INVALID_BLOCKS
2162 					| NFS_INO_INVALID_NLINK
2163 					| NFS_INO_INVALID_MODE
2164 					| NFS_INO_INVALID_OTHER;
2165 				if (S_ISDIR(inode->i_mode))
2166 					nfs_force_lookup_revalidate(inode);
2167 				attr_changed = true;
2168 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2169 						inode->i_sb->s_id,
2170 						inode->i_ino);
2171 			} else if (!have_delegation) {
2172 				nfs_ooo_record(nfsi, fattr);
2173 				nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2174 					      fattr->change_attr);
2175 			}
2176 			inode_set_iversion_raw(inode, fattr->change_attr);
2177 		}
2178 	} else {
2179 		nfsi->cache_validity |=
2180 			save_cache_validity & NFS_INO_INVALID_CHANGE;
2181 		if (!have_delegation ||
2182 		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2183 			cache_revalidated = false;
2184 	}
2185 
2186 	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2187 		inode->i_mtime = fattr->mtime;
2188 	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2189 		nfsi->cache_validity |=
2190 			save_cache_validity & NFS_INO_INVALID_MTIME;
2191 
2192 	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2193 		inode_set_ctime_to_ts(inode, fattr->ctime);
2194 	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2195 		nfsi->cache_validity |=
2196 			save_cache_validity & NFS_INO_INVALID_CTIME;
2197 
2198 	/* Check if our cached file size is stale */
2199 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2200 		new_isize = nfs_size_to_loff_t(fattr->size);
2201 		cur_isize = i_size_read(inode);
2202 		if (new_isize != cur_isize && !have_delegation) {
2203 			/* Do we perhaps have any outstanding writes, or has
2204 			 * the file grown beyond our last write? */
2205 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2206 				trace_nfs_size_update(inode, new_isize);
2207 				i_size_write(inode, new_isize);
2208 				if (!have_writers)
2209 					invalid |= NFS_INO_INVALID_DATA;
2210 			}
2211 		}
2212 		if (new_isize == 0 &&
2213 		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2214 				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2215 			fattr->du.nfs3.used = 0;
2216 			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2217 		}
2218 	} else
2219 		nfsi->cache_validity |=
2220 			save_cache_validity & NFS_INO_INVALID_SIZE;
2221 
2222 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2223 		inode->i_atime = fattr->atime;
2224 	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2225 		nfsi->cache_validity |=
2226 			save_cache_validity & NFS_INO_INVALID_ATIME;
2227 
2228 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2229 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2230 			umode_t newmode = inode->i_mode & S_IFMT;
2231 			newmode |= fattr->mode & S_IALLUGO;
2232 			inode->i_mode = newmode;
2233 			invalid |= NFS_INO_INVALID_ACCESS
2234 				| NFS_INO_INVALID_ACL;
2235 		}
2236 	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2237 		nfsi->cache_validity |=
2238 			save_cache_validity & NFS_INO_INVALID_MODE;
2239 
2240 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2241 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2242 			invalid |= NFS_INO_INVALID_ACCESS
2243 				| NFS_INO_INVALID_ACL;
2244 			inode->i_uid = fattr->uid;
2245 		}
2246 	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2247 		nfsi->cache_validity |=
2248 			save_cache_validity & NFS_INO_INVALID_OTHER;
2249 
2250 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2251 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2252 			invalid |= NFS_INO_INVALID_ACCESS
2253 				| NFS_INO_INVALID_ACL;
2254 			inode->i_gid = fattr->gid;
2255 		}
2256 	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2257 		nfsi->cache_validity |=
2258 			save_cache_validity & NFS_INO_INVALID_OTHER;
2259 
2260 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2261 		if (inode->i_nlink != fattr->nlink)
2262 			set_nlink(inode, fattr->nlink);
2263 	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2264 		nfsi->cache_validity |=
2265 			save_cache_validity & NFS_INO_INVALID_NLINK;
2266 
2267 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2268 		/*
2269 		 * report the blocks in 512byte units
2270 		 */
2271 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2272 	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2273 		nfsi->cache_validity |=
2274 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2275 
2276 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2277 		inode->i_blocks = fattr->du.nfs2.blocks;
2278 	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2279 		nfsi->cache_validity |=
2280 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2281 
2282 	/* Update attrtimeo value if we're out of the unstable period */
2283 	if (attr_changed) {
2284 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2285 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2286 		nfsi->attrtimeo_timestamp = now;
2287 		/* Set barrier to be more recent than all outstanding updates */
2288 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2289 	} else {
2290 		if (cache_revalidated) {
2291 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2292 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2293 				nfsi->attrtimeo <<= 1;
2294 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2295 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2296 			}
2297 			nfsi->attrtimeo_timestamp = now;
2298 		}
2299 		/* Set the barrier to be more recent than this fattr */
2300 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2301 			nfsi->attr_gencount = fattr->gencount;
2302 	}
2303 
2304 	/* Don't invalidate the data if we were to blame */
2305 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2306 				|| S_ISLNK(inode->i_mode)))
2307 		invalid &= ~NFS_INO_INVALID_DATA;
2308 	nfs_set_cache_invalid(inode, invalid);
2309 
2310 	return 0;
2311  out_err:
2312 	/*
2313 	 * No need to worry about unhashing the dentry, as the
2314 	 * lookup validation will know that the inode is bad.
2315 	 * (But we fall through to invalidate the caches.)
2316 	 */
2317 	nfs_set_inode_stale_locked(inode);
2318 	return -ESTALE;
2319 }
2320 
nfs_alloc_inode(struct super_block * sb)2321 struct inode *nfs_alloc_inode(struct super_block *sb)
2322 {
2323 	struct nfs_inode *nfsi;
2324 	nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2325 	if (!nfsi)
2326 		return NULL;
2327 	nfsi->flags = 0UL;
2328 	nfsi->cache_validity = 0UL;
2329 	nfsi->ooo = NULL;
2330 #if IS_ENABLED(CONFIG_NFS_V4)
2331 	nfsi->nfs4_acl = NULL;
2332 #endif /* CONFIG_NFS_V4 */
2333 #ifdef CONFIG_NFS_V4_2
2334 	nfsi->xattr_cache = NULL;
2335 #endif
2336 	nfs_netfs_inode_init(nfsi);
2337 
2338 	return &nfsi->vfs_inode;
2339 }
2340 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2341 
nfs_free_inode(struct inode * inode)2342 void nfs_free_inode(struct inode *inode)
2343 {
2344 	kfree(NFS_I(inode)->ooo);
2345 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2346 }
2347 EXPORT_SYMBOL_GPL(nfs_free_inode);
2348 
nfs4_init_once(struct nfs_inode * nfsi)2349 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2350 {
2351 #if IS_ENABLED(CONFIG_NFS_V4)
2352 	INIT_LIST_HEAD(&nfsi->open_states);
2353 	nfsi->delegation = NULL;
2354 	init_rwsem(&nfsi->rwsem);
2355 	nfsi->layout = NULL;
2356 #endif
2357 }
2358 
init_once(void * foo)2359 static void init_once(void *foo)
2360 {
2361 	struct nfs_inode *nfsi = foo;
2362 
2363 	inode_init_once(&nfsi->vfs_inode);
2364 	INIT_LIST_HEAD(&nfsi->open_files);
2365 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2366 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2367 	nfs4_init_once(nfsi);
2368 }
2369 
nfs_init_inodecache(void)2370 static int __init nfs_init_inodecache(void)
2371 {
2372 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2373 					     sizeof(struct nfs_inode),
2374 					     0, (SLAB_RECLAIM_ACCOUNT|
2375 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2376 					     init_once);
2377 	if (nfs_inode_cachep == NULL)
2378 		return -ENOMEM;
2379 
2380 	return 0;
2381 }
2382 
nfs_destroy_inodecache(void)2383 static void nfs_destroy_inodecache(void)
2384 {
2385 	/*
2386 	 * Make sure all delayed rcu free inodes are flushed before we
2387 	 * destroy cache.
2388 	 */
2389 	rcu_barrier();
2390 	kmem_cache_destroy(nfs_inode_cachep);
2391 }
2392 
2393 struct workqueue_struct *nfsiod_workqueue;
2394 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2395 
2396 /*
2397  * start up the nfsiod workqueue
2398  */
nfsiod_start(void)2399 static int nfsiod_start(void)
2400 {
2401 	struct workqueue_struct *wq;
2402 	dprintk("RPC:       creating workqueue nfsiod\n");
2403 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2404 	if (wq == NULL)
2405 		return -ENOMEM;
2406 	nfsiod_workqueue = wq;
2407 	return 0;
2408 }
2409 
2410 /*
2411  * Destroy the nfsiod workqueue
2412  */
nfsiod_stop(void)2413 static void nfsiod_stop(void)
2414 {
2415 	struct workqueue_struct *wq;
2416 
2417 	wq = nfsiod_workqueue;
2418 	if (wq == NULL)
2419 		return;
2420 	nfsiod_workqueue = NULL;
2421 	destroy_workqueue(wq);
2422 }
2423 
2424 unsigned int nfs_net_id;
2425 EXPORT_SYMBOL_GPL(nfs_net_id);
2426 
nfs_net_init(struct net * net)2427 static int nfs_net_init(struct net *net)
2428 {
2429 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2430 
2431 	nfs_clients_init(net);
2432 
2433 	if (!rpc_proc_register(net, &nn->rpcstats)) {
2434 		nfs_clients_exit(net);
2435 		return -ENOMEM;
2436 	}
2437 
2438 	return nfs_fs_proc_net_init(net);
2439 }
2440 
nfs_net_exit(struct net * net)2441 static void nfs_net_exit(struct net *net)
2442 {
2443 	rpc_proc_unregister(net, "nfs");
2444 	nfs_fs_proc_net_exit(net);
2445 	nfs_clients_exit(net);
2446 }
2447 
2448 static struct pernet_operations nfs_net_ops = {
2449 	.init = nfs_net_init,
2450 	.exit = nfs_net_exit,
2451 	.id   = &nfs_net_id,
2452 	.size = sizeof(struct nfs_net),
2453 };
2454 
2455 /*
2456  * Initialize NFS
2457  */
init_nfs_fs(void)2458 static int __init init_nfs_fs(void)
2459 {
2460 	int err;
2461 
2462 	err = nfs_sysfs_init();
2463 	if (err < 0)
2464 		goto out10;
2465 
2466 	err = register_pernet_subsys(&nfs_net_ops);
2467 	if (err < 0)
2468 		goto out9;
2469 
2470 	err = nfsiod_start();
2471 	if (err)
2472 		goto out7;
2473 
2474 	err = nfs_fs_proc_init();
2475 	if (err)
2476 		goto out6;
2477 
2478 	err = nfs_init_nfspagecache();
2479 	if (err)
2480 		goto out5;
2481 
2482 	err = nfs_init_inodecache();
2483 	if (err)
2484 		goto out4;
2485 
2486 	err = nfs_init_readpagecache();
2487 	if (err)
2488 		goto out3;
2489 
2490 	err = nfs_init_writepagecache();
2491 	if (err)
2492 		goto out2;
2493 
2494 	err = nfs_init_directcache();
2495 	if (err)
2496 		goto out1;
2497 
2498 	err = register_nfs_fs();
2499 	if (err)
2500 		goto out0;
2501 
2502 	return 0;
2503 out0:
2504 	nfs_destroy_directcache();
2505 out1:
2506 	nfs_destroy_writepagecache();
2507 out2:
2508 	nfs_destroy_readpagecache();
2509 out3:
2510 	nfs_destroy_inodecache();
2511 out4:
2512 	nfs_destroy_nfspagecache();
2513 out5:
2514 	nfs_fs_proc_exit();
2515 out6:
2516 	nfsiod_stop();
2517 out7:
2518 	unregister_pernet_subsys(&nfs_net_ops);
2519 out9:
2520 	nfs_sysfs_exit();
2521 out10:
2522 	return err;
2523 }
2524 
exit_nfs_fs(void)2525 static void __exit exit_nfs_fs(void)
2526 {
2527 	nfs_destroy_directcache();
2528 	nfs_destroy_writepagecache();
2529 	nfs_destroy_readpagecache();
2530 	nfs_destroy_inodecache();
2531 	nfs_destroy_nfspagecache();
2532 	unregister_pernet_subsys(&nfs_net_ops);
2533 	unregister_nfs_fs();
2534 	nfs_fs_proc_exit();
2535 	nfsiod_stop();
2536 	nfs_sysfs_exit();
2537 }
2538 
2539 /* Not quite true; I just maintain it */
2540 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2541 MODULE_LICENSE("GPL");
2542 module_param(enable_ino64, bool, 0644);
2543 
2544 module_init(init_nfs_fs)
2545 module_exit(exit_nfs_fs)
2546