xref: /linux/fs/nfs/inode.c (revision ccc1ead23c3311be76e87d1b06620f6cb697b42a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/nfs/inode.c
4  *
5  *  Copyright (C) 1992  Rick Sladkey
6  *
7  *  nfs inode and superblock handling functions
8  *
9  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
11  *
12  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13  *  J.S.Peatfield@damtp.cam.ac.uk
14  *
15  */
16 
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
44 
45 #include "nfs4_fs.h"
46 #include "callback.h"
47 #include "delegation.h"
48 #include "iostat.h"
49 #include "internal.h"
50 #include "fscache.h"
51 #include "pnfs.h"
52 #include "nfs.h"
53 #include "netns.h"
54 #include "sysfs.h"
55 
56 #include "nfstrace.h"
57 
58 #define NFSDBG_FACILITY		NFSDBG_VFS
59 
60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
61 
62 /* Default is to see 64-bit inode numbers */
63 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64 
65 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 
67 static struct kmem_cache * nfs_inode_cachep;
68 
69 static inline unsigned long
nfs_fattr_to_ino_t(struct nfs_fattr * fattr)70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 {
72 	return nfs_fileid_to_ino_t(fattr->fileid);
73 }
74 
nfs_wait_bit_killable(struct wait_bit_key * key,int mode)75 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
76 {
77 	if (unlikely(nfs_current_task_exiting()))
78 		return -EINTR;
79 	schedule();
80 	if (signal_pending_state(mode, current))
81 		return -ERESTARTSYS;
82 	return 0;
83 }
84 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
85 
86 /**
87  * nfs_compat_user_ino64 - returns the user-visible inode number
88  * @fileid: 64-bit fileid
89  *
90  * This function returns a 32-bit inode number if the boot parameter
91  * nfs.enable_ino64 is zero.
92  */
nfs_compat_user_ino64(u64 fileid)93 u64 nfs_compat_user_ino64(u64 fileid)
94 {
95 #ifdef CONFIG_COMPAT
96 	compat_ulong_t ino;
97 #else
98 	unsigned long ino;
99 #endif
100 
101 	if (enable_ino64)
102 		return fileid;
103 	ino = fileid;
104 	if (sizeof(ino) < sizeof(fileid))
105 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
106 	return ino;
107 }
108 
nfs_drop_inode(struct inode * inode)109 int nfs_drop_inode(struct inode *inode)
110 {
111 	return NFS_STALE(inode) || generic_drop_inode(inode);
112 }
113 EXPORT_SYMBOL_GPL(nfs_drop_inode);
114 
nfs_clear_inode(struct inode * inode)115 void nfs_clear_inode(struct inode *inode)
116 {
117 	/*
118 	 * The following should never happen...
119 	 */
120 	WARN_ON_ONCE(nfs_have_writebacks(inode));
121 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
122 	nfs_zap_acl_cache(inode);
123 	nfs_access_zap_cache(inode);
124 	nfs_fscache_clear_inode(inode);
125 }
126 EXPORT_SYMBOL_GPL(nfs_clear_inode);
127 
nfs_evict_inode(struct inode * inode)128 void nfs_evict_inode(struct inode *inode)
129 {
130 	truncate_inode_pages_final(&inode->i_data);
131 	clear_inode(inode);
132 	nfs_clear_inode(inode);
133 }
134 
nfs_sync_inode(struct inode * inode)135 int nfs_sync_inode(struct inode *inode)
136 {
137 	inode_dio_wait(inode);
138 	return nfs_wb_all(inode);
139 }
140 EXPORT_SYMBOL_GPL(nfs_sync_inode);
141 
142 /**
143  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
144  * @mapping: pointer to struct address_space
145  */
nfs_sync_mapping(struct address_space * mapping)146 int nfs_sync_mapping(struct address_space *mapping)
147 {
148 	int ret = 0;
149 
150 	if (mapping->nrpages != 0) {
151 		unmap_mapping_range(mapping, 0, 0, 0);
152 		ret = nfs_wb_all(mapping->host);
153 	}
154 	return ret;
155 }
156 
nfs_attribute_timeout(struct inode * inode)157 static int nfs_attribute_timeout(struct inode *inode)
158 {
159 	struct nfs_inode *nfsi = NFS_I(inode);
160 
161 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
162 }
163 
nfs_check_cache_flags_invalid(struct inode * inode,unsigned long flags)164 static bool nfs_check_cache_flags_invalid(struct inode *inode,
165 					  unsigned long flags)
166 {
167 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
168 
169 	return (cache_validity & flags) != 0;
170 }
171 
nfs_check_cache_invalid(struct inode * inode,unsigned long flags)172 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
173 {
174 	if (nfs_check_cache_flags_invalid(inode, flags))
175 		return true;
176 	return nfs_attribute_cache_expired(inode);
177 }
178 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
179 
180 #ifdef CONFIG_NFS_V4_2
nfs_has_xattr_cache(const struct nfs_inode * nfsi)181 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
182 {
183 	return nfsi->xattr_cache != NULL;
184 }
185 #else
nfs_has_xattr_cache(const struct nfs_inode * nfsi)186 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
187 {
188 	return false;
189 }
190 #endif
191 
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)192 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
193 {
194 	struct nfs_inode *nfsi = NFS_I(inode);
195 
196 	if (nfs_have_delegated_attributes(inode)) {
197 		if (!(flags & NFS_INO_REVAL_FORCED))
198 			flags &= ~(NFS_INO_INVALID_MODE |
199 				   NFS_INO_INVALID_OTHER |
200 				   NFS_INO_INVALID_BTIME |
201 				   NFS_INO_INVALID_XATTR);
202 		flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
203 	}
204 
205 	if (!nfs_has_xattr_cache(nfsi))
206 		flags &= ~NFS_INO_INVALID_XATTR;
207 	if (flags & NFS_INO_INVALID_DATA)
208 		nfs_fscache_invalidate(inode, 0);
209 	flags &= ~NFS_INO_REVAL_FORCED;
210 
211 	flags |= nfsi->cache_validity;
212 	if (inode->i_mapping->nrpages == 0)
213 		flags &= ~NFS_INO_INVALID_DATA;
214 
215 	/* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */
216 	smp_store_release(&nfsi->cache_validity, flags);
217 
218 	if (inode->i_mapping->nrpages == 0 ||
219 	    nfsi->cache_validity & NFS_INO_INVALID_DATA) {
220 		nfs_ooo_clear(nfsi);
221 	}
222 	trace_nfs_set_cache_invalid(inode, 0);
223 }
224 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
225 
226 /*
227  * Invalidate the local caches
228  */
nfs_zap_caches_locked(struct inode * inode)229 static void nfs_zap_caches_locked(struct inode *inode)
230 {
231 	struct nfs_inode *nfsi = NFS_I(inode);
232 	int mode = inode->i_mode;
233 
234 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
235 
236 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
237 	nfsi->attrtimeo_timestamp = jiffies;
238 
239 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
240 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
241 						     NFS_INO_INVALID_DATA |
242 						     NFS_INO_INVALID_ACCESS |
243 						     NFS_INO_INVALID_ACL |
244 						     NFS_INO_INVALID_XATTR);
245 	else
246 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
247 						     NFS_INO_INVALID_ACCESS |
248 						     NFS_INO_INVALID_ACL |
249 						     NFS_INO_INVALID_XATTR);
250 	nfs_zap_label_cache_locked(nfsi);
251 }
252 
nfs_zap_caches(struct inode * inode)253 void nfs_zap_caches(struct inode *inode)
254 {
255 	spin_lock(&inode->i_lock);
256 	nfs_zap_caches_locked(inode);
257 	spin_unlock(&inode->i_lock);
258 }
259 
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)260 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
261 {
262 	if (mapping->nrpages != 0) {
263 		spin_lock(&inode->i_lock);
264 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
265 		spin_unlock(&inode->i_lock);
266 	}
267 }
268 
nfs_zap_acl_cache(struct inode * inode)269 void nfs_zap_acl_cache(struct inode *inode)
270 {
271 	void (*clear_acl_cache)(struct inode *);
272 
273 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
274 	if (clear_acl_cache != NULL)
275 		clear_acl_cache(inode);
276 	spin_lock(&inode->i_lock);
277 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
278 	spin_unlock(&inode->i_lock);
279 }
280 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
281 
nfs_invalidate_atime(struct inode * inode)282 void nfs_invalidate_atime(struct inode *inode)
283 {
284 	if (nfs_have_delegated_atime(inode))
285 		return;
286 	spin_lock(&inode->i_lock);
287 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
288 	spin_unlock(&inode->i_lock);
289 }
290 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
291 
292 /*
293  * Invalidate, but do not unhash, the inode.
294  * NB: must be called with inode->i_lock held!
295  */
nfs_set_inode_stale_locked(struct inode * inode)296 static void nfs_set_inode_stale_locked(struct inode *inode)
297 {
298 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
299 	nfs_zap_caches_locked(inode);
300 	trace_nfs_set_inode_stale(inode);
301 }
302 
nfs_set_inode_stale(struct inode * inode)303 void nfs_set_inode_stale(struct inode *inode)
304 {
305 	spin_lock(&inode->i_lock);
306 	nfs_set_inode_stale_locked(inode);
307 	spin_unlock(&inode->i_lock);
308 }
309 
310 struct nfs_find_desc {
311 	struct nfs_fh		*fh;
312 	struct nfs_fattr	*fattr;
313 };
314 
315 /*
316  * In NFSv3 we can have 64bit inode numbers. In order to support
317  * this, and re-exported directories (also seen in NFSv2)
318  * we are forced to allow 2 different inodes to have the same
319  * i_ino.
320  */
321 static int
nfs_find_actor(struct inode * inode,void * opaque)322 nfs_find_actor(struct inode *inode, void *opaque)
323 {
324 	struct nfs_find_desc	*desc = opaque;
325 	struct nfs_fh		*fh = desc->fh;
326 	struct nfs_fattr	*fattr = desc->fattr;
327 
328 	if (NFS_FILEID(inode) != fattr->fileid)
329 		return 0;
330 	if (inode_wrong_type(inode, fattr->mode))
331 		return 0;
332 	if (nfs_compare_fh(NFS_FH(inode), fh))
333 		return 0;
334 	if (is_bad_inode(inode) || NFS_STALE(inode))
335 		return 0;
336 	return 1;
337 }
338 
339 static int
nfs_init_locked(struct inode * inode,void * opaque)340 nfs_init_locked(struct inode *inode, void *opaque)
341 {
342 	struct nfs_find_desc	*desc = opaque;
343 	struct nfs_fattr	*fattr = desc->fattr;
344 
345 	set_nfs_fileid(inode, fattr->fileid);
346 	inode->i_mode = fattr->mode;
347 	nfs_copy_fh(NFS_FH(inode), desc->fh);
348 	return 0;
349 }
350 
351 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)352 static void nfs_clear_label_invalid(struct inode *inode)
353 {
354 	spin_lock(&inode->i_lock);
355 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
356 	spin_unlock(&inode->i_lock);
357 }
358 
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)359 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
360 {
361 	int error;
362 
363 	if (fattr->label == NULL)
364 		return;
365 
366 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
367 		error = security_inode_notifysecctx(inode, fattr->label->label,
368 				fattr->label->len);
369 		if (error)
370 			printk(KERN_ERR "%s() %s %d "
371 					"security_inode_notifysecctx() %d\n",
372 					__func__,
373 					(char *)fattr->label->label,
374 					fattr->label->len, error);
375 		nfs_clear_label_invalid(inode);
376 	}
377 }
378 
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)379 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
380 {
381 	struct nfs4_label *label;
382 
383 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
384 		return NULL;
385 
386 	label = kzalloc(sizeof(struct nfs4_label), flags);
387 	if (label == NULL)
388 		return ERR_PTR(-ENOMEM);
389 
390 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
391 	if (label->label == NULL) {
392 		kfree(label);
393 		return ERR_PTR(-ENOMEM);
394 	}
395 	label->len = NFS4_MAXLABELLEN;
396 
397 	return label;
398 }
399 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
400 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)401 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
402 {
403 }
404 #endif
405 EXPORT_SYMBOL_GPL(nfs_setsecurity);
406 
407 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
408 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)409 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
410 {
411 	struct nfs_find_desc desc = {
412 		.fh	= fh,
413 		.fattr	= fattr,
414 	};
415 	struct inode *inode;
416 	unsigned long hash;
417 
418 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
419 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
420 		return NULL;
421 
422 	hash = nfs_fattr_to_ino_t(fattr);
423 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
424 
425 	dprintk("%s: returning %p\n", __func__, inode);
426 	return inode;
427 }
428 
nfs_inode_init_regular(struct nfs_inode * nfsi)429 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
430 {
431 	atomic_long_set(&nfsi->nrequests, 0);
432 	atomic_long_set(&nfsi->redirtied_pages, 0);
433 	INIT_LIST_HEAD(&nfsi->commit_info.list);
434 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
435 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
436 	mutex_init(&nfsi->commit_mutex);
437 }
438 
nfs_inode_init_dir(struct nfs_inode * nfsi)439 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
440 {
441 	nfsi->cache_change_attribute = 0;
442 	memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
443 	init_rwsem(&nfsi->rmdir_sem);
444 }
445 
446 /*
447  * This is our front-end to iget that looks up inodes by file handle
448  * instead of inode number.
449  */
450 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr)451 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
452 {
453 	struct nfs_find_desc desc = {
454 		.fh	= fh,
455 		.fattr	= fattr
456 	};
457 	struct inode *inode = ERR_PTR(-ENOENT);
458 	u64 fattr_supported = NFS_SB(sb)->fattr_valid;
459 	unsigned long hash;
460 
461 	nfs_attr_check_mountpoint(sb, fattr);
462 
463 	if (nfs_attr_use_mounted_on_fileid(fattr))
464 		fattr->fileid = fattr->mounted_on_fileid;
465 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
466 		goto out_no_inode;
467 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
468 		goto out_no_inode;
469 
470 	hash = nfs_fattr_to_ino_t(fattr);
471 
472 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
473 	if (inode == NULL) {
474 		inode = ERR_PTR(-ENOMEM);
475 		goto out_no_inode;
476 	}
477 
478 	if (inode->i_state & I_NEW) {
479 		struct nfs_inode *nfsi = NFS_I(inode);
480 		unsigned long now = jiffies;
481 
482 		/* We set i_ino for the few things that still rely on it,
483 		 * such as stat(2) */
484 		inode->i_ino = hash;
485 
486 		/* We can't support update_atime(), since the server will reset it */
487 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
488 		inode->i_mode = fattr->mode;
489 		nfsi->cache_validity = 0;
490 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
491 				&& (fattr_supported & NFS_ATTR_FATTR_MODE))
492 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
493 		/* Why so? Because we want revalidate for devices/FIFOs, and
494 		 * that's precisely what we have in nfs_file_inode_operations.
495 		 */
496 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
497 		if (S_ISREG(inode->i_mode)) {
498 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
499 			inode->i_data.a_ops = &nfs_file_aops;
500 			nfs_inode_init_regular(nfsi);
501 			mapping_set_large_folios(inode->i_mapping);
502 		} else if (S_ISDIR(inode->i_mode)) {
503 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
504 			inode->i_fop = &nfs_dir_operations;
505 			inode->i_data.a_ops = &nfs_dir_aops;
506 			nfs_inode_init_dir(nfsi);
507 			/* Deal with crossing mountpoints */
508 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
509 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
510 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
511 					inode->i_op = &nfs_referral_inode_operations;
512 				else
513 					inode->i_op = &nfs_mountpoint_inode_operations;
514 				inode->i_fop = NULL;
515 				inode->i_flags |= S_AUTOMOUNT;
516 			}
517 		} else if (S_ISLNK(inode->i_mode)) {
518 			inode->i_op = &nfs_symlink_inode_operations;
519 			inode_nohighmem(inode);
520 		} else
521 			init_special_inode(inode, inode->i_mode, fattr->rdev);
522 
523 		inode_set_atime(inode, 0, 0);
524 		inode_set_mtime(inode, 0, 0);
525 		inode_set_ctime(inode, 0, 0);
526 		memset(&nfsi->btime, 0, sizeof(nfsi->btime));
527 		inode_set_iversion_raw(inode, 0);
528 		inode->i_size = 0;
529 		clear_nlink(inode);
530 		inode->i_uid = make_kuid(&init_user_ns, -2);
531 		inode->i_gid = make_kgid(&init_user_ns, -2);
532 		inode->i_blocks = 0;
533 		nfsi->write_io = 0;
534 		nfsi->read_io = 0;
535 
536 		nfsi->read_cache_jiffies = fattr->time_start;
537 		nfsi->attr_gencount = fattr->gencount;
538 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
539 			inode_set_atime_to_ts(inode, fattr->atime);
540 		else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
541 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
542 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
543 			inode_set_mtime_to_ts(inode, fattr->mtime);
544 		else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
545 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
546 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
547 			inode_set_ctime_to_ts(inode, fattr->ctime);
548 		else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
549 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
550 		if (fattr->valid & NFS_ATTR_FATTR_BTIME)
551 			nfsi->btime = fattr->btime;
552 		else if (fattr_supported & NFS_ATTR_FATTR_BTIME)
553 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BTIME);
554 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
555 			inode_set_iversion_raw(inode, fattr->change_attr);
556 		else
557 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
558 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
559 			inode->i_size = nfs_size_to_loff_t(fattr->size);
560 		else
561 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
562 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
563 			set_nlink(inode, fattr->nlink);
564 		else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
565 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
566 		else
567 			set_nlink(inode, 1);
568 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
569 			inode->i_uid = fattr->uid;
570 		else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
571 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
572 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
573 			inode->i_gid = fattr->gid;
574 		else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
575 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
576 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
577 			inode->i_blocks = fattr->du.nfs2.blocks;
578 		else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
579 			 fattr->size != 0)
580 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
581 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
582 			/*
583 			 * report the blocks in 512byte units
584 			 */
585 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
586 		} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
587 			   fattr->size != 0)
588 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
589 
590 		nfs_setsecurity(inode, fattr);
591 
592 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
593 		nfsi->attrtimeo_timestamp = now;
594 		nfsi->access_cache = RB_ROOT;
595 
596 		nfs_fscache_init_inode(inode);
597 
598 		unlock_new_inode(inode);
599 	} else {
600 		int err = nfs_refresh_inode(inode, fattr);
601 		if (err < 0) {
602 			iput(inode);
603 			inode = ERR_PTR(err);
604 			goto out_no_inode;
605 		}
606 	}
607 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
608 		inode->i_sb->s_id,
609 		(unsigned long long)NFS_FILEID(inode),
610 		nfs_display_fhandle_hash(fh),
611 		atomic_read(&inode->i_count));
612 
613 out:
614 	return inode;
615 
616 out_no_inode:
617 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
618 	goto out;
619 }
620 EXPORT_SYMBOL_GPL(nfs_fhget);
621 
622 static void
nfs_fattr_fixup_delegated(struct inode * inode,struct nfs_fattr * fattr)623 nfs_fattr_fixup_delegated(struct inode *inode, struct nfs_fattr *fattr)
624 {
625 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
626 
627 	if (nfs_have_delegated_mtime(inode)) {
628 		if (!(cache_validity & NFS_INO_INVALID_CTIME))
629 			fattr->valid &= ~(NFS_ATTR_FATTR_PRECTIME |
630 					  NFS_ATTR_FATTR_CTIME);
631 
632 		if (!(cache_validity & NFS_INO_INVALID_MTIME))
633 			fattr->valid &= ~(NFS_ATTR_FATTR_PREMTIME |
634 					  NFS_ATTR_FATTR_MTIME);
635 
636 		if (!(cache_validity & NFS_INO_INVALID_ATIME))
637 			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
638 	} else if (nfs_have_delegated_atime(inode)) {
639 		if (!(cache_validity & NFS_INO_INVALID_ATIME))
640 			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
641 	}
642 }
643 
nfs_set_timestamps_to_ts(struct inode * inode,struct iattr * attr)644 static void nfs_set_timestamps_to_ts(struct inode *inode, struct iattr *attr)
645 {
646 	unsigned int cache_flags = 0;
647 
648 	if (attr->ia_valid & ATTR_MTIME_SET) {
649 		struct timespec64 ctime = inode_get_ctime(inode);
650 		struct timespec64 mtime = inode_get_mtime(inode);
651 		struct timespec64 now;
652 		int updated = 0;
653 
654 		now = inode_set_ctime_current(inode);
655 		if (!timespec64_equal(&now, &ctime))
656 			updated |= S_CTIME;
657 
658 		inode_set_mtime_to_ts(inode, attr->ia_mtime);
659 		if (!timespec64_equal(&now, &mtime))
660 			updated |= S_MTIME;
661 
662 		inode_maybe_inc_iversion(inode, updated);
663 		cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME;
664 	}
665 	if (attr->ia_valid & ATTR_ATIME_SET) {
666 		inode_set_atime_to_ts(inode, attr->ia_atime);
667 		cache_flags |= NFS_INO_INVALID_ATIME;
668 	}
669 	NFS_I(inode)->cache_validity &= ~cache_flags;
670 }
671 
nfs_update_timestamps(struct inode * inode,unsigned int ia_valid)672 static void nfs_update_timestamps(struct inode *inode, unsigned int ia_valid)
673 {
674 	enum file_time_flags time_flags = 0;
675 	unsigned int cache_flags = 0;
676 
677 	if (ia_valid & ATTR_MTIME) {
678 		time_flags |= S_MTIME | S_CTIME;
679 		cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME;
680 	}
681 	if (ia_valid & ATTR_ATIME) {
682 		time_flags |= S_ATIME;
683 		cache_flags |= NFS_INO_INVALID_ATIME;
684 	}
685 	inode_update_timestamps(inode, time_flags);
686 	NFS_I(inode)->cache_validity &= ~cache_flags;
687 }
688 
nfs_update_delegated_atime(struct inode * inode)689 void nfs_update_delegated_atime(struct inode *inode)
690 {
691 	spin_lock(&inode->i_lock);
692 	if (nfs_have_delegated_atime(inode))
693 		nfs_update_timestamps(inode, ATTR_ATIME);
694 	spin_unlock(&inode->i_lock);
695 }
696 
nfs_update_delegated_mtime_locked(struct inode * inode)697 void nfs_update_delegated_mtime_locked(struct inode *inode)
698 {
699 	if (nfs_have_delegated_mtime(inode))
700 		nfs_update_timestamps(inode, ATTR_MTIME);
701 }
702 
nfs_update_delegated_mtime(struct inode * inode)703 void nfs_update_delegated_mtime(struct inode *inode)
704 {
705 	spin_lock(&inode->i_lock);
706 	nfs_update_delegated_mtime_locked(inode);
707 	spin_unlock(&inode->i_lock);
708 }
709 EXPORT_SYMBOL_GPL(nfs_update_delegated_mtime);
710 
711 #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)
712 
713 int
nfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)714 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
715 	    struct iattr *attr)
716 {
717 	struct inode *inode = d_inode(dentry);
718 	struct nfs_fattr *fattr;
719 	int error = 0;
720 
721 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
722 
723 	/* skip mode change if it's just for clearing setuid/setgid */
724 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
725 		attr->ia_valid &= ~ATTR_MODE;
726 
727 	if (attr->ia_valid & ATTR_SIZE) {
728 		BUG_ON(!S_ISREG(inode->i_mode));
729 
730 		error = inode_newsize_ok(inode, attr->ia_size);
731 		if (error)
732 			return error;
733 
734 		if (attr->ia_size == i_size_read(inode))
735 			attr->ia_valid &= ~ATTR_SIZE;
736 	}
737 
738 	if (nfs_have_delegated_mtime(inode) && attr->ia_valid & ATTR_MTIME) {
739 		spin_lock(&inode->i_lock);
740 		if (attr->ia_valid & ATTR_MTIME_SET) {
741 			nfs_set_timestamps_to_ts(inode, attr);
742 			attr->ia_valid &= ~(ATTR_MTIME|ATTR_MTIME_SET|
743 						ATTR_ATIME|ATTR_ATIME_SET);
744 		} else {
745 			nfs_update_timestamps(inode, attr->ia_valid);
746 			attr->ia_valid &= ~(ATTR_MTIME|ATTR_ATIME);
747 		}
748 		spin_unlock(&inode->i_lock);
749 	} else if (nfs_have_delegated_atime(inode) &&
750 		   attr->ia_valid & ATTR_ATIME &&
751 		   !(attr->ia_valid & ATTR_MTIME)) {
752 		if (attr->ia_valid & ATTR_ATIME_SET) {
753 			spin_lock(&inode->i_lock);
754 			nfs_set_timestamps_to_ts(inode, attr);
755 			spin_unlock(&inode->i_lock);
756 			attr->ia_valid &= ~(ATTR_ATIME|ATTR_ATIME_SET);
757 		} else {
758 			nfs_update_delegated_atime(inode);
759 			attr->ia_valid &= ~ATTR_ATIME;
760 		}
761 	}
762 
763 	/* Optimization: if the end result is no change, don't RPC */
764 	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
765 		return 0;
766 
767 	trace_nfs_setattr_enter(inode);
768 
769 	/* Write all dirty data */
770 	if (S_ISREG(inode->i_mode))
771 		nfs_sync_inode(inode);
772 
773 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
774 	if (fattr == NULL) {
775 		error = -ENOMEM;
776 		goto out;
777 	}
778 
779 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
780 	if (error == 0)
781 		error = nfs_refresh_inode(inode, fattr);
782 	nfs_free_fattr(fattr);
783 out:
784 	trace_nfs_setattr_exit(inode, error);
785 	return error;
786 }
787 EXPORT_SYMBOL_GPL(nfs_setattr);
788 
789 /**
790  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
791  * @inode: inode of the file used
792  * @offset: file offset to start truncating
793  *
794  * This is a copy of the common vmtruncate, but with the locking
795  * corrected to take into account the fact that NFS requires
796  * inode->i_size to be updated under the inode->i_lock.
797  * Note: must be called with inode->i_lock held!
798  */
nfs_vmtruncate(struct inode * inode,loff_t offset)799 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
800 {
801 	int err;
802 
803 	err = inode_newsize_ok(inode, offset);
804 	if (err)
805 		goto out;
806 
807 	trace_nfs_size_truncate(inode, offset);
808 	i_size_write(inode, offset);
809 	/* Optimisation */
810 	if (offset == 0) {
811 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
812 		nfs_ooo_clear(NFS_I(inode));
813 	}
814 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
815 
816 	spin_unlock(&inode->i_lock);
817 	truncate_pagecache(inode, offset);
818 	nfs_update_delegated_mtime_locked(inode);
819 	spin_lock(&inode->i_lock);
820 out:
821 	return err;
822 }
823 
824 /**
825  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
826  * @inode: pointer to struct inode
827  * @attr: pointer to struct iattr
828  * @fattr: pointer to struct nfs_fattr
829  *
830  * Note: we do this in the *proc.c in order to ensure that
831  *       it works for things like exclusive creates too.
832  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)833 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
834 		struct nfs_fattr *fattr)
835 {
836 	/* Barrier: bump the attribute generation count. */
837 	nfs_fattr_set_barrier(fattr);
838 
839 	spin_lock(&inode->i_lock);
840 	NFS_I(inode)->attr_gencount = fattr->gencount;
841 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
842 		if (!nfs_have_delegated_mtime(inode))
843 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
844 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
845 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
846 		nfs_vmtruncate(inode, attr->ia_size);
847 	}
848 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
849 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
850 		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
851 		    inode->i_mode & S_ISUID)
852 			inode->i_mode &= ~S_ISUID;
853 		if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
854 			inode->i_mode &= ~S_ISGID;
855 		if ((attr->ia_valid & ATTR_MODE) != 0) {
856 			int mode = attr->ia_mode & S_IALLUGO;
857 			mode |= inode->i_mode & ~S_IALLUGO;
858 			inode->i_mode = mode;
859 		}
860 		if ((attr->ia_valid & ATTR_UID) != 0)
861 			inode->i_uid = attr->ia_uid;
862 		if ((attr->ia_valid & ATTR_GID) != 0)
863 			inode->i_gid = attr->ia_gid;
864 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
865 			inode_set_ctime_to_ts(inode, fattr->ctime);
866 		else
867 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
868 					| NFS_INO_INVALID_CTIME);
869 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
870 				| NFS_INO_INVALID_ACL);
871 	}
872 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
873 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
874 				| NFS_INO_INVALID_CTIME);
875 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
876 			inode_set_atime_to_ts(inode, fattr->atime);
877 		else if (attr->ia_valid & ATTR_ATIME_SET)
878 			inode_set_atime_to_ts(inode, attr->ia_atime);
879 		else
880 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
881 
882 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
883 			inode_set_ctime_to_ts(inode, fattr->ctime);
884 		else
885 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
886 					| NFS_INO_INVALID_CTIME);
887 	}
888 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
889 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
890 				| NFS_INO_INVALID_CTIME);
891 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
892 			inode_set_mtime_to_ts(inode, fattr->mtime);
893 		else if (attr->ia_valid & ATTR_MTIME_SET)
894 			inode_set_mtime_to_ts(inode, attr->ia_mtime);
895 		else
896 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
897 
898 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
899 			inode_set_ctime_to_ts(inode, fattr->ctime);
900 		else
901 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
902 					| NFS_INO_INVALID_CTIME);
903 	}
904 	if (fattr->valid)
905 		nfs_update_inode(inode, fattr);
906 	spin_unlock(&inode->i_lock);
907 }
908 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
909 
910 /*
911  * Don't request help from readdirplus if the file is being written to,
912  * or if attribute caching is turned off
913  */
nfs_getattr_readdirplus_enable(const struct inode * inode)914 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
915 {
916 	return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
917 	       !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
918 }
919 
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)920 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
921 {
922 	if (!IS_ROOT(dentry)) {
923 		struct dentry *parent = dget_parent(dentry);
924 		nfs_readdir_record_entry_cache_miss(d_inode(parent));
925 		dput(parent);
926 	}
927 }
928 
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)929 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
930 {
931 	if (!IS_ROOT(dentry)) {
932 		struct dentry *parent = dget_parent(dentry);
933 		nfs_readdir_record_entry_cache_hit(d_inode(parent));
934 		dput(parent);
935 	}
936 }
937 
nfs_get_valid_attrmask(struct inode * inode)938 static u32 nfs_get_valid_attrmask(struct inode *inode)
939 {
940 	u64 fattr_valid = NFS_SERVER(inode)->fattr_valid;
941 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
942 	u32 reply_mask = STATX_INO | STATX_TYPE;
943 
944 	if (!(cache_validity & NFS_INO_INVALID_ATIME))
945 		reply_mask |= STATX_ATIME;
946 	if (!(cache_validity & NFS_INO_INVALID_CTIME))
947 		reply_mask |= STATX_CTIME;
948 	if (!(cache_validity & NFS_INO_INVALID_MTIME))
949 		reply_mask |= STATX_MTIME;
950 	if (!(cache_validity & NFS_INO_INVALID_SIZE))
951 		reply_mask |= STATX_SIZE;
952 	if (!(cache_validity & NFS_INO_INVALID_NLINK))
953 		reply_mask |= STATX_NLINK;
954 	if (!(cache_validity & NFS_INO_INVALID_MODE))
955 		reply_mask |= STATX_MODE;
956 	if (!(cache_validity & NFS_INO_INVALID_OTHER))
957 		reply_mask |= STATX_UID | STATX_GID;
958 	if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
959 		reply_mask |= STATX_BLOCKS;
960 	if (!(cache_validity & NFS_INO_INVALID_BTIME) &&
961 	    (fattr_valid & NFS_ATTR_FATTR_BTIME))
962 		reply_mask |= STATX_BTIME;
963 	if (!(cache_validity & NFS_INO_INVALID_CHANGE))
964 		reply_mask |= STATX_CHANGE_COOKIE;
965 	return reply_mask;
966 }
967 
nfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)968 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
969 		struct kstat *stat, u32 request_mask, unsigned int query_flags)
970 {
971 	struct inode *inode = d_inode(path->dentry);
972 	struct nfs_server *server = NFS_SERVER(inode);
973 	u64 fattr_valid = server->fattr_valid;
974 	unsigned long cache_validity;
975 	int err = 0;
976 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
977 	bool do_update = false;
978 	bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
979 
980 	trace_nfs_getattr_enter(inode);
981 
982 	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
983 			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
984 			STATX_INO | STATX_SIZE | STATX_BLOCKS | STATX_BTIME |
985 			STATX_CHANGE_COOKIE;
986 
987 	if (!(fattr_valid & NFS_ATTR_FATTR_BTIME))
988 		request_mask &= ~STATX_BTIME;
989 
990 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
991 		if (readdirplus_enabled)
992 			nfs_readdirplus_parent_cache_hit(path->dentry);
993 		goto out_no_revalidate;
994 	}
995 
996 	/* Flush out writes to the server in order to update c/mtime/version.  */
997 	if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
998 	    S_ISREG(inode->i_mode)) {
999 		if (nfs_have_delegated_mtime(inode))
1000 			filemap_fdatawrite(inode->i_mapping);
1001 		else
1002 			filemap_write_and_wait(inode->i_mapping);
1003 	}
1004 
1005 	/*
1006 	 * We may force a getattr if the user cares about atime.
1007 	 *
1008 	 * Note that we only have to check the vfsmount flags here:
1009 	 *  - NFS always sets S_NOATIME by so checking it would give a
1010 	 *    bogus result
1011 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
1012 	 *    no point in checking those.
1013 	 */
1014 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
1015 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1016 		request_mask &= ~STATX_ATIME;
1017 
1018 	/* Is the user requesting attributes that might need revalidation? */
1019 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
1020 					STATX_MTIME|STATX_UID|STATX_GID|
1021 					STATX_SIZE|STATX_BLOCKS|STATX_BTIME|
1022 					STATX_CHANGE_COOKIE)))
1023 		goto out_no_revalidate;
1024 
1025 	/* Check whether the cached attributes are stale */
1026 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
1027 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
1028 	do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
1029 	if (request_mask & STATX_ATIME)
1030 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
1031 	if (request_mask & STATX_CTIME)
1032 		do_update |= cache_validity & NFS_INO_INVALID_CTIME;
1033 	if (request_mask & STATX_MTIME)
1034 		do_update |= cache_validity & NFS_INO_INVALID_MTIME;
1035 	if (request_mask & STATX_SIZE)
1036 		do_update |= cache_validity & NFS_INO_INVALID_SIZE;
1037 	if (request_mask & STATX_NLINK)
1038 		do_update |= cache_validity & NFS_INO_INVALID_NLINK;
1039 	if (request_mask & STATX_MODE)
1040 		do_update |= cache_validity & NFS_INO_INVALID_MODE;
1041 	if (request_mask & (STATX_UID | STATX_GID))
1042 		do_update |= cache_validity & NFS_INO_INVALID_OTHER;
1043 	if (request_mask & STATX_BLOCKS)
1044 		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
1045 	if (request_mask & STATX_BTIME)
1046 		do_update |= cache_validity & NFS_INO_INVALID_BTIME;
1047 
1048 	if (do_update) {
1049 		if (readdirplus_enabled)
1050 			nfs_readdirplus_parent_cache_miss(path->dentry);
1051 		err = __nfs_revalidate_inode(server, inode);
1052 		if (err)
1053 			goto out;
1054 	} else if (readdirplus_enabled)
1055 		nfs_readdirplus_parent_cache_hit(path->dentry);
1056 out_no_revalidate:
1057 	/* Only return attributes that were revalidated. */
1058 	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
1059 
1060 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1061 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
1062 	stat->change_cookie = inode_peek_iversion_raw(inode);
1063 	stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
1064 	if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
1065 		stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
1066 	if (S_ISDIR(inode->i_mode))
1067 		stat->blksize = NFS_SERVER(inode)->dtsize;
1068 	stat->btime = NFS_I(inode)->btime;
1069 out:
1070 	trace_nfs_getattr_exit(inode, err);
1071 	return err;
1072 }
1073 EXPORT_SYMBOL_GPL(nfs_getattr);
1074 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)1075 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
1076 {
1077 	refcount_set(&l_ctx->count, 1);
1078 	l_ctx->lockowner = current->files;
1079 	INIT_LIST_HEAD(&l_ctx->list);
1080 	atomic_set(&l_ctx->io_count, 0);
1081 }
1082 
__nfs_find_lock_context(struct nfs_open_context * ctx)1083 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
1084 {
1085 	struct nfs_lock_context *pos;
1086 
1087 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
1088 		if (pos->lockowner != current->files)
1089 			continue;
1090 		if (refcount_inc_not_zero(&pos->count))
1091 			return pos;
1092 	}
1093 	return NULL;
1094 }
1095 
nfs_get_lock_context(struct nfs_open_context * ctx)1096 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
1097 {
1098 	struct nfs_lock_context *res, *new = NULL;
1099 	struct inode *inode = d_inode(ctx->dentry);
1100 
1101 	rcu_read_lock();
1102 	res = __nfs_find_lock_context(ctx);
1103 	rcu_read_unlock();
1104 	if (res == NULL) {
1105 		new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1106 		if (new == NULL)
1107 			return ERR_PTR(-ENOMEM);
1108 		nfs_init_lock_context(new);
1109 		spin_lock(&inode->i_lock);
1110 		res = __nfs_find_lock_context(ctx);
1111 		if (res == NULL) {
1112 			new->open_context = get_nfs_open_context(ctx);
1113 			if (new->open_context) {
1114 				list_add_tail_rcu(&new->list,
1115 						&ctx->lock_context.list);
1116 				res = new;
1117 				new = NULL;
1118 			} else
1119 				res = ERR_PTR(-EBADF);
1120 		}
1121 		spin_unlock(&inode->i_lock);
1122 		kfree(new);
1123 	}
1124 	return res;
1125 }
1126 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
1127 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)1128 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
1129 {
1130 	struct nfs_open_context *ctx = l_ctx->open_context;
1131 	struct inode *inode = d_inode(ctx->dentry);
1132 
1133 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
1134 		return;
1135 	list_del_rcu(&l_ctx->list);
1136 	spin_unlock(&inode->i_lock);
1137 	put_nfs_open_context(ctx);
1138 	kfree_rcu(l_ctx, rcu_head);
1139 }
1140 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
1141 
1142 /**
1143  * nfs_close_context - Common close_context() routine NFSv2/v3
1144  * @ctx: pointer to context
1145  * @is_sync: is this a synchronous close
1146  *
1147  * Ensure that the attributes are up to date if we're mounted
1148  * with close-to-open semantics and we have cached data that will
1149  * need to be revalidated on open.
1150  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)1151 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1152 {
1153 	struct nfs_inode *nfsi;
1154 	struct inode *inode;
1155 
1156 	if (!(ctx->mode & FMODE_WRITE))
1157 		return;
1158 	if (!is_sync)
1159 		return;
1160 	inode = d_inode(ctx->dentry);
1161 	if (nfs_have_read_or_write_delegation(inode))
1162 		return;
1163 	nfsi = NFS_I(inode);
1164 	if (inode->i_mapping->nrpages == 0)
1165 		return;
1166 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1167 		return;
1168 	if (!list_empty(&nfsi->open_files))
1169 		return;
1170 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1171 		return;
1172 	nfs_revalidate_inode(inode,
1173 			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1174 }
1175 EXPORT_SYMBOL_GPL(nfs_close_context);
1176 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1177 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1178 						fmode_t f_mode,
1179 						struct file *filp)
1180 {
1181 	struct nfs_open_context *ctx;
1182 
1183 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1184 	if (!ctx)
1185 		return ERR_PTR(-ENOMEM);
1186 	nfs_sb_active(dentry->d_sb);
1187 	ctx->dentry = dget(dentry);
1188 	if (filp)
1189 		ctx->cred = get_cred(filp->f_cred);
1190 	else
1191 		ctx->cred = get_current_cred();
1192 	rcu_assign_pointer(ctx->ll_cred, NULL);
1193 	ctx->state = NULL;
1194 	ctx->mode = f_mode;
1195 	ctx->flags = 0;
1196 	ctx->error = 0;
1197 	ctx->flock_owner = (fl_owner_t)filp;
1198 	nfs_init_lock_context(&ctx->lock_context);
1199 	ctx->lock_context.open_context = ctx;
1200 	INIT_LIST_HEAD(&ctx->list);
1201 	ctx->mdsthreshold = NULL;
1202 	nfs_localio_file_init(&ctx->nfl);
1203 
1204 	return ctx;
1205 }
1206 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1207 
get_nfs_open_context(struct nfs_open_context * ctx)1208 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1209 {
1210 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1211 		return ctx;
1212 	return NULL;
1213 }
1214 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1215 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1216 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1217 {
1218 	struct inode *inode = d_inode(ctx->dentry);
1219 	struct super_block *sb = ctx->dentry->d_sb;
1220 
1221 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1222 		return;
1223 	if (!list_empty(&ctx->list)) {
1224 		spin_lock(&inode->i_lock);
1225 		list_del_rcu(&ctx->list);
1226 		spin_unlock(&inode->i_lock);
1227 	}
1228 	if (inode != NULL)
1229 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1230 	put_cred(ctx->cred);
1231 	dput(ctx->dentry);
1232 	nfs_sb_deactive(sb);
1233 	put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1234 	kfree(ctx->mdsthreshold);
1235 	nfs_close_local_fh(&ctx->nfl);
1236 	kfree_rcu(ctx, rcu_head);
1237 }
1238 
put_nfs_open_context(struct nfs_open_context * ctx)1239 void put_nfs_open_context(struct nfs_open_context *ctx)
1240 {
1241 	__put_nfs_open_context(ctx, 0);
1242 }
1243 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1244 
put_nfs_open_context_sync(struct nfs_open_context * ctx)1245 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1246 {
1247 	__put_nfs_open_context(ctx, 1);
1248 }
1249 
1250 /*
1251  * Ensure that mmap has a recent RPC credential for use when writing out
1252  * shared pages
1253  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1254 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1255 {
1256 	struct inode *inode = d_inode(ctx->dentry);
1257 	struct nfs_inode *nfsi = NFS_I(inode);
1258 
1259 	spin_lock(&inode->i_lock);
1260 	if (list_empty(&nfsi->open_files) &&
1261 	    nfs_ooo_test(nfsi))
1262 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1263 						     NFS_INO_REVAL_FORCED);
1264 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1265 	spin_unlock(&inode->i_lock);
1266 }
1267 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1268 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1269 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1270 {
1271 	filp->private_data = get_nfs_open_context(ctx);
1272 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1273 	if (list_empty(&ctx->list))
1274 		nfs_inode_attach_open_context(ctx);
1275 }
1276 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1277 
1278 /*
1279  * Given an inode, search for an open context with the desired characteristics
1280  */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1281 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1282 {
1283 	struct nfs_inode *nfsi = NFS_I(inode);
1284 	struct nfs_open_context *pos, *ctx = NULL;
1285 
1286 	rcu_read_lock();
1287 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1288 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1289 			continue;
1290 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1291 			continue;
1292 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1293 			continue;
1294 		ctx = get_nfs_open_context(pos);
1295 		if (ctx)
1296 			break;
1297 	}
1298 	rcu_read_unlock();
1299 	return ctx;
1300 }
1301 
nfs_file_clear_open_context(struct file * filp)1302 void nfs_file_clear_open_context(struct file *filp)
1303 {
1304 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1305 
1306 	if (ctx) {
1307 		struct inode *inode = d_inode(ctx->dentry);
1308 
1309 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1310 		/*
1311 		 * We fatal error on write before. Try to writeback
1312 		 * every page again.
1313 		 */
1314 		if (ctx->error < 0)
1315 			invalidate_inode_pages2(inode->i_mapping);
1316 		filp->private_data = NULL;
1317 		put_nfs_open_context_sync(ctx);
1318 	}
1319 }
1320 
1321 /*
1322  * These allocate and release file read/write context information.
1323  */
nfs_open(struct inode * inode,struct file * filp)1324 int nfs_open(struct inode *inode, struct file *filp)
1325 {
1326 	struct nfs_open_context *ctx;
1327 
1328 	ctx = alloc_nfs_open_context(file_dentry(filp),
1329 				     flags_to_mode(filp->f_flags), filp);
1330 	if (IS_ERR(ctx))
1331 		return PTR_ERR(ctx);
1332 	nfs_file_set_open_context(filp, ctx);
1333 	put_nfs_open_context(ctx);
1334 	nfs_fscache_open_file(inode, filp);
1335 	return 0;
1336 }
1337 
1338 /*
1339  * This function is called whenever some part of NFS notices that
1340  * the cached attributes have to be refreshed.
1341  */
1342 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1343 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1344 {
1345 	int		 status = -ESTALE;
1346 	struct nfs_fattr *fattr = NULL;
1347 	struct nfs_inode *nfsi = NFS_I(inode);
1348 
1349 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1350 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1351 
1352 	trace_nfs_revalidate_inode_enter(inode);
1353 
1354 	if (is_bad_inode(inode))
1355 		goto out;
1356 	if (NFS_STALE(inode))
1357 		goto out;
1358 
1359 	/* pNFS: Attributes aren't updated until we layoutcommit */
1360 	if (S_ISREG(inode->i_mode)) {
1361 		status = pnfs_sync_inode(inode, false);
1362 		if (status)
1363 			goto out;
1364 	}
1365 
1366 	status = -ENOMEM;
1367 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1368 	if (fattr == NULL)
1369 		goto out;
1370 
1371 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1372 
1373 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1374 	if (status != 0) {
1375 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1376 			 inode->i_sb->s_id,
1377 			 (unsigned long long)NFS_FILEID(inode), status);
1378 		switch (status) {
1379 		case -ETIMEDOUT:
1380 			/* A soft timeout occurred. Use cached information? */
1381 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1382 				status = 0;
1383 			break;
1384 		case -ESTALE:
1385 			if (!S_ISDIR(inode->i_mode))
1386 				nfs_set_inode_stale(inode);
1387 			else
1388 				nfs_zap_caches(inode);
1389 		}
1390 		goto out;
1391 	}
1392 
1393 	status = nfs_refresh_inode(inode, fattr);
1394 	if (status) {
1395 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1396 			 inode->i_sb->s_id,
1397 			 (unsigned long long)NFS_FILEID(inode), status);
1398 		goto out;
1399 	}
1400 
1401 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1402 		nfs_zap_acl_cache(inode);
1403 
1404 	nfs_setsecurity(inode, fattr);
1405 
1406 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1407 		inode->i_sb->s_id,
1408 		(unsigned long long)NFS_FILEID(inode));
1409 
1410 out:
1411 	nfs_free_fattr(fattr);
1412 	trace_nfs_revalidate_inode_exit(inode, status);
1413 	return status;
1414 }
1415 
nfs_attribute_cache_expired(struct inode * inode)1416 int nfs_attribute_cache_expired(struct inode *inode)
1417 {
1418 	if (nfs_have_delegated_attributes(inode))
1419 		return 0;
1420 	return nfs_attribute_timeout(inode);
1421 }
1422 
1423 /**
1424  * nfs_revalidate_inode - Revalidate the inode attributes
1425  * @inode: pointer to inode struct
1426  * @flags: cache flags to check
1427  *
1428  * Updates inode attribute information by retrieving the data from the server.
1429  */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1430 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1431 {
1432 	if (!nfs_check_cache_invalid(inode, flags))
1433 		return NFS_STALE(inode) ? -ESTALE : 0;
1434 	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1435 }
1436 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1437 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1438 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1439 {
1440 	int ret;
1441 
1442 	nfs_fscache_invalidate(inode, 0);
1443 	if (mapping->nrpages != 0) {
1444 		if (S_ISREG(inode->i_mode)) {
1445 			ret = nfs_sync_mapping(mapping);
1446 			if (ret < 0)
1447 				return ret;
1448 		}
1449 		ret = invalidate_inode_pages2(mapping);
1450 		if (ret < 0)
1451 			return ret;
1452 	}
1453 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1454 
1455 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1456 			inode->i_sb->s_id,
1457 			(unsigned long long)NFS_FILEID(inode));
1458 	return 0;
1459 }
1460 
1461 /**
1462  * nfs_clear_invalid_mapping - Conditionally clear a mapping
1463  * @mapping: pointer to mapping
1464  *
1465  * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1466  */
nfs_clear_invalid_mapping(struct address_space * mapping)1467 int nfs_clear_invalid_mapping(struct address_space *mapping)
1468 {
1469 	struct inode *inode = mapping->host;
1470 	struct nfs_inode *nfsi = NFS_I(inode);
1471 	unsigned long *bitlock = &nfsi->flags;
1472 	int ret = 0;
1473 
1474 	/*
1475 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1476 	 * invalidations that come in while we're shooting down the mappings
1477 	 * are respected. But, that leaves a race window where one revalidator
1478 	 * can clear the flag, and then another checks it before the mapping
1479 	 * gets invalidated. Fix that by serializing access to this part of
1480 	 * the function.
1481 	 *
1482 	 * At the same time, we need to allow other tasks to see whether we
1483 	 * might be in the middle of invalidating the pages, so we only set
1484 	 * the bit lock here if it looks like we're going to be doing that.
1485 	 */
1486 	for (;;) {
1487 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1488 					 nfs_wait_bit_killable,
1489 					 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1490 		if (ret)
1491 			goto out;
1492 		smp_rmb(); /* pairs with smp_wmb() below */
1493 		if (test_bit(NFS_INO_INVALIDATING, bitlock))
1494 			continue;
1495 		/* pairs with nfs_set_cache_invalid()'s smp_store_release() */
1496 		if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA))
1497 			goto out;
1498 		/* Slow-path that double-checks with spinlock held */
1499 		spin_lock(&inode->i_lock);
1500 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1501 			spin_unlock(&inode->i_lock);
1502 			continue;
1503 		}
1504 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1505 			break;
1506 		spin_unlock(&inode->i_lock);
1507 		goto out;
1508 	}
1509 
1510 	set_bit(NFS_INO_INVALIDATING, bitlock);
1511 	smp_wmb();
1512 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1513 	nfs_ooo_clear(nfsi);
1514 	spin_unlock(&inode->i_lock);
1515 	trace_nfs_invalidate_mapping_enter(inode);
1516 	ret = nfs_invalidate_mapping(inode, mapping);
1517 	trace_nfs_invalidate_mapping_exit(inode, ret);
1518 
1519 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1520 	smp_mb__after_atomic();
1521 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1522 out:
1523 	return ret;
1524 }
1525 
nfs_mapping_need_revalidate_inode(struct inode * inode)1526 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1527 {
1528 	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1529 		NFS_STALE(inode);
1530 }
1531 
nfs_revalidate_mapping_rcu(struct inode * inode)1532 int nfs_revalidate_mapping_rcu(struct inode *inode)
1533 {
1534 	struct nfs_inode *nfsi = NFS_I(inode);
1535 	unsigned long *bitlock = &nfsi->flags;
1536 	int ret = 0;
1537 
1538 	if (IS_SWAPFILE(inode))
1539 		goto out;
1540 	if (nfs_mapping_need_revalidate_inode(inode)) {
1541 		ret = -ECHILD;
1542 		goto out;
1543 	}
1544 	spin_lock(&inode->i_lock);
1545 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1546 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1547 		ret = -ECHILD;
1548 	spin_unlock(&inode->i_lock);
1549 out:
1550 	return ret;
1551 }
1552 
1553 /**
1554  * nfs_revalidate_mapping - Revalidate the pagecache
1555  * @inode: pointer to host inode
1556  * @mapping: pointer to mapping
1557  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1558 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1559 {
1560 	/* swapfiles are not supposed to be shared. */
1561 	if (IS_SWAPFILE(inode))
1562 		return 0;
1563 
1564 	if (nfs_mapping_need_revalidate_inode(inode)) {
1565 		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1566 		if (ret < 0)
1567 			return ret;
1568 	}
1569 
1570 	return nfs_clear_invalid_mapping(mapping);
1571 }
1572 
nfs_file_has_writers(struct nfs_inode * nfsi)1573 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1574 {
1575 	struct inode *inode = &nfsi->vfs_inode;
1576 
1577 	if (!S_ISREG(inode->i_mode))
1578 		return false;
1579 	if (list_empty(&nfsi->open_files))
1580 		return false;
1581 	return inode_is_open_for_write(inode);
1582 }
1583 
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1584 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1585 {
1586 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1587 }
1588 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1589 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1590 {
1591 	struct timespec64 ts;
1592 
1593 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1594 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1595 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1596 		inode_set_iversion_raw(inode, fattr->change_attr);
1597 		if (S_ISDIR(inode->i_mode))
1598 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1599 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1600 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1601 	}
1602 	/* If we have atomic WCC data, we may update some attributes */
1603 	ts = inode_get_ctime(inode);
1604 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1605 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1606 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1607 		inode_set_ctime_to_ts(inode, fattr->ctime);
1608 	}
1609 
1610 	ts = inode_get_mtime(inode);
1611 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1612 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1613 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1614 		inode_set_mtime_to_ts(inode, fattr->mtime);
1615 	}
1616 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1617 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1618 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1619 			&& !nfs_have_writebacks(inode)) {
1620 		trace_nfs_size_wcc(inode, fattr->size);
1621 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1622 	}
1623 }
1624 
1625 /**
1626  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1627  * @inode: pointer to inode
1628  * @fattr: updated attributes
1629  *
1630  * Verifies the attribute cache. If we have just changed the attributes,
1631  * so that fattr carries weak cache consistency data, then it may
1632  * also update the ctime/mtime/change_attribute.
1633  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1634 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1635 {
1636 	struct nfs_inode *nfsi = NFS_I(inode);
1637 	loff_t cur_size, new_isize;
1638 	unsigned long invalid = 0;
1639 	struct timespec64 ts;
1640 
1641 	if (nfs_have_delegated_attributes(inode))
1642 		return 0;
1643 
1644 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1645 		/* Only a mounted-on-fileid? Just exit */
1646 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1647 			return 0;
1648 	/* Has the inode gone and changed behind our back? */
1649 	} else if (nfsi->fileid != fattr->fileid) {
1650 		/* Is this perhaps the mounted-on fileid? */
1651 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1652 		    nfsi->fileid == fattr->mounted_on_fileid)
1653 			return 0;
1654 		return -ESTALE;
1655 	}
1656 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1657 		return -ESTALE;
1658 
1659 
1660 	if (!nfs_file_has_buffered_writers(nfsi)) {
1661 		/* Verify a few of the more important attributes */
1662 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1663 			invalid |= NFS_INO_INVALID_CHANGE;
1664 
1665 		ts = inode_get_mtime(inode);
1666 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1667 			invalid |= NFS_INO_INVALID_MTIME;
1668 
1669 		ts = inode_get_ctime(inode);
1670 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1671 			invalid |= NFS_INO_INVALID_CTIME;
1672 
1673 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1674 			cur_size = i_size_read(inode);
1675 			new_isize = nfs_size_to_loff_t(fattr->size);
1676 			if (cur_size != new_isize)
1677 				invalid |= NFS_INO_INVALID_SIZE;
1678 		}
1679 	}
1680 
1681 	/* Have any file permissions changed? */
1682 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1683 		invalid |= NFS_INO_INVALID_MODE;
1684 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1685 		invalid |= NFS_INO_INVALID_OTHER;
1686 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1687 		invalid |= NFS_INO_INVALID_OTHER;
1688 
1689 	/* Has the link count changed? */
1690 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1691 		invalid |= NFS_INO_INVALID_NLINK;
1692 
1693 	ts = inode_get_atime(inode);
1694 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1695 		invalid |= NFS_INO_INVALID_ATIME;
1696 
1697 	if (invalid != 0)
1698 		nfs_set_cache_invalid(inode, invalid);
1699 
1700 	nfsi->read_cache_jiffies = fattr->time_start;
1701 	return 0;
1702 }
1703 
1704 static atomic_long_t nfs_attr_generation_counter;
1705 
nfs_read_attr_generation_counter(void)1706 static unsigned long nfs_read_attr_generation_counter(void)
1707 {
1708 	return atomic_long_read(&nfs_attr_generation_counter);
1709 }
1710 
nfs_inc_attr_generation_counter(void)1711 unsigned long nfs_inc_attr_generation_counter(void)
1712 {
1713 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1714 }
1715 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1716 
nfs_fattr_init(struct nfs_fattr * fattr)1717 void nfs_fattr_init(struct nfs_fattr *fattr)
1718 {
1719 	fattr->valid = 0;
1720 	fattr->time_start = jiffies;
1721 	fattr->gencount = nfs_inc_attr_generation_counter();
1722 	fattr->owner_name = NULL;
1723 	fattr->group_name = NULL;
1724 	fattr->mdsthreshold = NULL;
1725 }
1726 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1727 
1728 /**
1729  * nfs_fattr_set_barrier
1730  * @fattr: attributes
1731  *
1732  * Used to set a barrier after an attribute was updated. This
1733  * barrier ensures that older attributes from RPC calls that may
1734  * have raced with our update cannot clobber these new values.
1735  * Note that you are still responsible for ensuring that other
1736  * operations which change the attribute on the server do not
1737  * collide.
1738  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1739 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1740 {
1741 	fattr->gencount = nfs_inc_attr_generation_counter();
1742 }
1743 
nfs_alloc_fattr(void)1744 struct nfs_fattr *nfs_alloc_fattr(void)
1745 {
1746 	struct nfs_fattr *fattr;
1747 
1748 	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1749 	if (fattr != NULL) {
1750 		nfs_fattr_init(fattr);
1751 		fattr->label = NULL;
1752 	}
1753 	return fattr;
1754 }
1755 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1756 
nfs_alloc_fattr_with_label(struct nfs_server * server)1757 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1758 {
1759 	struct nfs_fattr *fattr = nfs_alloc_fattr();
1760 
1761 	if (!fattr)
1762 		return NULL;
1763 
1764 	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1765 	if (IS_ERR(fattr->label)) {
1766 		kfree(fattr);
1767 		return NULL;
1768 	}
1769 
1770 	return fattr;
1771 }
1772 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1773 
nfs_alloc_fhandle(void)1774 struct nfs_fh *nfs_alloc_fhandle(void)
1775 {
1776 	struct nfs_fh *fh;
1777 
1778 	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1779 	if (fh != NULL)
1780 		fh->size = 0;
1781 	return fh;
1782 }
1783 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1784 
1785 #ifdef NFS_DEBUG
1786 /*
1787  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1788  *                             in the same way that wireshark does
1789  *
1790  * @fh: file handle
1791  *
1792  * For debugging only.
1793  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1794 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1795 {
1796 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1797 	 * not on the result */
1798 	return nfs_fhandle_hash(fh);
1799 }
1800 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1801 
1802 /*
1803  * _nfs_display_fhandle - display an NFS file handle on the console
1804  *
1805  * @fh: file handle to display
1806  * @caption: display caption
1807  *
1808  * For debugging only.
1809  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1810 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1811 {
1812 	unsigned short i;
1813 
1814 	if (fh == NULL || fh->size == 0) {
1815 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1816 		return;
1817 	}
1818 
1819 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1820 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1821 	for (i = 0; i < fh->size; i += 16) {
1822 		__be32 *pos = (__be32 *)&fh->data[i];
1823 
1824 		switch ((fh->size - i - 1) >> 2) {
1825 		case 0:
1826 			printk(KERN_DEFAULT " %08x\n",
1827 				be32_to_cpup(pos));
1828 			break;
1829 		case 1:
1830 			printk(KERN_DEFAULT " %08x %08x\n",
1831 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1832 			break;
1833 		case 2:
1834 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1835 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1836 				be32_to_cpup(pos + 2));
1837 			break;
1838 		default:
1839 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1840 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1841 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1842 		}
1843 	}
1844 }
1845 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1846 #endif
1847 
1848 /**
1849  * nfs_inode_attrs_cmp_generic - compare attributes
1850  * @fattr: attributes
1851  * @inode: pointer to inode
1852  *
1853  * Attempt to divine whether or not an RPC call reply carrying stale
1854  * attributes got scheduled after another call carrying updated ones.
1855  * Note also the check for wraparound of 'attr_gencount'
1856  *
1857  * The function returns '1' if it thinks the attributes in @fattr are
1858  * more recent than the ones cached in @inode. Otherwise it returns
1859  * the value '0'.
1860  */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1861 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1862 				       const struct inode *inode)
1863 {
1864 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1865 
1866 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1867 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1868 }
1869 
1870 /**
1871  * nfs_inode_attrs_cmp_monotonic - compare attributes
1872  * @fattr: attributes
1873  * @inode: pointer to inode
1874  *
1875  * Attempt to divine whether or not an RPC call reply carrying stale
1876  * attributes got scheduled after another call carrying updated ones.
1877  *
1878  * We assume that the server observes monotonic semantics for
1879  * the change attribute, so a larger value means that the attributes in
1880  * @fattr are more recent, in which case the function returns the
1881  * value '1'.
1882  * A return value of '0' indicates no measurable change
1883  * A return value of '-1' means that the attributes in @inode are
1884  * more recent.
1885  */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1886 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1887 					 const struct inode *inode)
1888 {
1889 	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1890 	if (diff > 0)
1891 		return 1;
1892 	return diff == 0 ? 0 : -1;
1893 }
1894 
1895 /**
1896  * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1897  * @fattr: attributes
1898  * @inode: pointer to inode
1899  *
1900  * Attempt to divine whether or not an RPC call reply carrying stale
1901  * attributes got scheduled after another call carrying updated ones.
1902  *
1903  * We assume that the server observes strictly monotonic semantics for
1904  * the change attribute, so a larger value means that the attributes in
1905  * @fattr are more recent, in which case the function returns the
1906  * value '1'.
1907  * A return value of '-1' means that the attributes in @inode are
1908  * more recent or unchanged.
1909  */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1910 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1911 						const struct inode *inode)
1912 {
1913 	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1914 }
1915 
1916 /**
1917  * nfs_inode_attrs_cmp - compare attributes
1918  * @fattr: attributes
1919  * @inode: pointer to inode
1920  *
1921  * This function returns '1' if it thinks the attributes in @fattr are
1922  * more recent than the ones cached in @inode. It returns '-1' if
1923  * the attributes in @inode are more recent than the ones in @fattr,
1924  * and it returns 0 if not sure.
1925  */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1926 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1927 			       const struct inode *inode)
1928 {
1929 	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1930 		return 1;
1931 	switch (NFS_SERVER(inode)->change_attr_type) {
1932 	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1933 		break;
1934 	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1935 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1936 			break;
1937 		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1938 	default:
1939 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1940 			break;
1941 		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1942 	}
1943 	return 0;
1944 }
1945 
1946 /**
1947  * nfs_inode_finish_partial_attr_update - complete a previous inode update
1948  * @fattr: attributes
1949  * @inode: pointer to inode
1950  *
1951  * Returns '1' if the last attribute update left the inode cached
1952  * attributes in a partially unrevalidated state, and @fattr
1953  * matches the change attribute of that partial update.
1954  * Otherwise returns '0'.
1955  */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1956 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1957 						const struct inode *inode)
1958 {
1959 	const unsigned long check_valid =
1960 		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1961 		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1962 		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1963 		NFS_INO_INVALID_NLINK | NFS_INO_INVALID_BTIME;
1964 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1965 	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1966 
1967 	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1968 	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1969 	    (cache_validity & check_valid) != 0 &&
1970 	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1971 	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1972 		return 1;
1973 	return 0;
1974 }
1975 
nfs_ooo_merge(struct nfs_inode * nfsi,u64 start,u64 end)1976 static void nfs_ooo_merge(struct nfs_inode *nfsi,
1977 			  u64 start, u64 end)
1978 {
1979 	int i, cnt;
1980 
1981 	if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1982 		/* No point merging anything */
1983 		return;
1984 
1985 	if (!nfsi->ooo) {
1986 		nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1987 		if (!nfsi->ooo) {
1988 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1989 			return;
1990 		}
1991 		nfsi->ooo->cnt = 0;
1992 	}
1993 
1994 	/* add this range, merging if possible */
1995 	cnt = nfsi->ooo->cnt;
1996 	for (i = 0; i < cnt; i++) {
1997 		if (end == nfsi->ooo->gap[i].start)
1998 			end = nfsi->ooo->gap[i].end;
1999 		else if (start == nfsi->ooo->gap[i].end)
2000 			start = nfsi->ooo->gap[i].start;
2001 		else
2002 			continue;
2003 		/* Remove 'i' from table and loop to insert the new range */
2004 		cnt -= 1;
2005 		nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
2006 		i = -1;
2007 	}
2008 	if (start != end) {
2009 		if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
2010 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2011 			kfree(nfsi->ooo);
2012 			nfsi->ooo = NULL;
2013 			return;
2014 		}
2015 		nfsi->ooo->gap[cnt].start = start;
2016 		nfsi->ooo->gap[cnt].end = end;
2017 		cnt += 1;
2018 	}
2019 	nfsi->ooo->cnt = cnt;
2020 }
2021 
nfs_ooo_record(struct nfs_inode * nfsi,struct nfs_fattr * fattr)2022 static void nfs_ooo_record(struct nfs_inode *nfsi,
2023 			   struct nfs_fattr *fattr)
2024 {
2025 	/* This reply was out-of-order, so record in the
2026 	 * pre/post change id, possibly cancelling
2027 	 * gaps created when iversion was jumpped forward.
2028 	 */
2029 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
2030 	    (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
2031 		nfs_ooo_merge(nfsi,
2032 			      fattr->change_attr,
2033 			      fattr->pre_change_attr);
2034 }
2035 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)2036 static int nfs_refresh_inode_locked(struct inode *inode,
2037 				    struct nfs_fattr *fattr)
2038 {
2039 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2040 	int ret = 0;
2041 
2042 	trace_nfs_refresh_inode_enter(inode);
2043 
2044 	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
2045 		ret = nfs_update_inode(inode, fattr);
2046 	else {
2047 		nfs_ooo_record(NFS_I(inode), fattr);
2048 
2049 		if (attr_cmp == 0)
2050 			ret = nfs_check_inode_attributes(inode, fattr);
2051 	}
2052 
2053 	trace_nfs_refresh_inode_exit(inode, ret);
2054 	return ret;
2055 }
2056 
2057 /**
2058  * nfs_refresh_inode - try to update the inode attribute cache
2059  * @inode: pointer to inode
2060  * @fattr: updated attributes
2061  *
2062  * Check that an RPC call that returned attributes has not overlapped with
2063  * other recent updates of the inode metadata, then decide whether it is
2064  * safe to do a full update of the inode attributes, or whether just to
2065  * call nfs_check_inode_attributes.
2066  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)2067 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
2068 {
2069 	int status;
2070 
2071 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2072 		return 0;
2073 	spin_lock(&inode->i_lock);
2074 	status = nfs_refresh_inode_locked(inode, fattr);
2075 	spin_unlock(&inode->i_lock);
2076 
2077 	return status;
2078 }
2079 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
2080 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)2081 static int nfs_post_op_update_inode_locked(struct inode *inode,
2082 		struct nfs_fattr *fattr, unsigned int invalid)
2083 {
2084 	if (S_ISDIR(inode->i_mode))
2085 		invalid |= NFS_INO_INVALID_DATA;
2086 	nfs_set_cache_invalid(inode, invalid);
2087 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2088 		return 0;
2089 	return nfs_refresh_inode_locked(inode, fattr);
2090 }
2091 
2092 /**
2093  * nfs_post_op_update_inode - try to update the inode attribute cache
2094  * @inode: pointer to inode
2095  * @fattr: updated attributes
2096  *
2097  * After an operation that has changed the inode metadata, mark the
2098  * attribute cache as being invalid, then try to update it.
2099  *
2100  * NB: if the server didn't return any post op attributes, this
2101  * function will force the retrieval of attributes before the next
2102  * NFS request.  Thus it should be used only for operations that
2103  * are expected to change one or more attributes, to avoid
2104  * unnecessary NFS requests and trips through nfs_update_inode().
2105  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)2106 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2107 {
2108 	int status;
2109 
2110 	spin_lock(&inode->i_lock);
2111 	nfs_fattr_set_barrier(fattr);
2112 	status = nfs_post_op_update_inode_locked(inode, fattr,
2113 			NFS_INO_INVALID_CHANGE
2114 			| NFS_INO_INVALID_CTIME
2115 			| NFS_INO_REVAL_FORCED);
2116 	spin_unlock(&inode->i_lock);
2117 
2118 	return status;
2119 }
2120 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
2121 
2122 /**
2123  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
2124  * @inode: pointer to inode
2125  * @fattr: updated attributes
2126  *
2127  * After an operation that has changed the inode metadata, mark the
2128  * attribute cache as being invalid, then try to update it. Fake up
2129  * weak cache consistency data, if none exist.
2130  *
2131  * This function is mainly designed to be used by the ->write_done() functions.
2132  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)2133 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
2134 {
2135 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2136 	int status;
2137 
2138 	/* Don't do a WCC update if these attributes are already stale */
2139 	if (attr_cmp < 0)
2140 		return 0;
2141 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
2142 		/* Record the pre/post change info before clearing PRECHANGE */
2143 		nfs_ooo_record(NFS_I(inode), fattr);
2144 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
2145 				| NFS_ATTR_FATTR_PRESIZE
2146 				| NFS_ATTR_FATTR_PREMTIME
2147 				| NFS_ATTR_FATTR_PRECTIME);
2148 		goto out_noforce;
2149 	}
2150 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2151 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2152 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2153 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2154 	}
2155 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2156 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2157 		fattr->pre_ctime = inode_get_ctime(inode);
2158 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2159 	}
2160 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2161 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2162 		fattr->pre_mtime = inode_get_mtime(inode);
2163 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2164 	}
2165 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2166 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2167 		fattr->pre_size = i_size_read(inode);
2168 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2169 	}
2170 out_noforce:
2171 	status = nfs_post_op_update_inode_locked(inode, fattr,
2172 			NFS_INO_INVALID_CHANGE
2173 			| NFS_INO_INVALID_CTIME
2174 			| NFS_INO_INVALID_MTIME
2175 			| NFS_INO_INVALID_BLOCKS);
2176 	return status;
2177 }
2178 
2179 /**
2180  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2181  * @inode: pointer to inode
2182  * @fattr: updated attributes
2183  *
2184  * After an operation that has changed the inode metadata, mark the
2185  * attribute cache as being invalid, then try to update it. Fake up
2186  * weak cache consistency data, if none exist.
2187  *
2188  * This function is mainly designed to be used by the ->write_done() functions.
2189  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)2190 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2191 {
2192 	int status;
2193 
2194 	spin_lock(&inode->i_lock);
2195 	nfs_fattr_set_barrier(fattr);
2196 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2197 	spin_unlock(&inode->i_lock);
2198 	return status;
2199 }
2200 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2201 
2202 
2203 /*
2204  * Many nfs protocol calls return the new file attributes after
2205  * an operation.  Here we update the inode to reflect the state
2206  * of the server's inode.
2207  *
2208  * This is a bit tricky because we have to make sure all dirty pages
2209  * have been sent off to the server before calling invalidate_inode_pages.
2210  * To make sure no other process adds more write requests while we try
2211  * our best to flush them, we make them sleep during the attribute refresh.
2212  *
2213  * A very similar scenario holds for the dir cache.
2214  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)2215 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2216 {
2217 	struct nfs_server *server = NFS_SERVER(inode);
2218 	struct nfs_inode *nfsi = NFS_I(inode);
2219 	loff_t cur_isize, new_isize;
2220 	u64 fattr_supported = server->fattr_valid;
2221 	unsigned long invalid = 0;
2222 	unsigned long now = jiffies;
2223 	unsigned long save_cache_validity;
2224 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2225 	bool cache_revalidated = true;
2226 	bool attr_changed = false;
2227 	bool have_delegation;
2228 
2229 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%llx)\n",
2230 			__func__, inode->i_sb->s_id, inode->i_ino,
2231 			nfs_display_fhandle_hash(NFS_FH(inode)),
2232 			atomic_read(&inode->i_count), fattr->valid);
2233 
2234 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2235 		/* Only a mounted-on-fileid? Just exit */
2236 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2237 			return 0;
2238 	/* Has the inode gone and changed behind our back? */
2239 	} else if (nfsi->fileid != fattr->fileid) {
2240 		/* Is this perhaps the mounted-on fileid? */
2241 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2242 		    nfsi->fileid == fattr->mounted_on_fileid)
2243 			return 0;
2244 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2245 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2246 			NFS_SERVER(inode)->nfs_client->cl_hostname,
2247 			inode->i_sb->s_id, (long long)nfsi->fileid,
2248 			(long long)fattr->fileid);
2249 		goto out_err;
2250 	}
2251 
2252 	/*
2253 	 * Make sure the inode's type hasn't changed.
2254 	 */
2255 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2256 		/*
2257 		* Big trouble! The inode has become a different object.
2258 		*/
2259 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2260 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2261 		goto out_err;
2262 	}
2263 
2264 	/* Update the fsid? */
2265 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2266 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2267 			!IS_AUTOMOUNT(inode))
2268 		server->fsid = fattr->fsid;
2269 
2270 	/* Save the delegation state before clearing cache_validity */
2271 	have_delegation = nfs_have_delegated_attributes(inode);
2272 
2273 	/*
2274 	 * Update the read time so we don't revalidate too often.
2275 	 */
2276 	nfsi->read_cache_jiffies = fattr->time_start;
2277 
2278 	/* Fix up any delegated attributes in the struct nfs_fattr */
2279 	nfs_fattr_fixup_delegated(inode, fattr);
2280 
2281 	save_cache_validity = nfsi->cache_validity;
2282 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2283 			| NFS_INO_INVALID_ATIME
2284 			| NFS_INO_REVAL_FORCED
2285 			| NFS_INO_INVALID_BLOCKS);
2286 
2287 	/* Do atomic weak cache consistency updates */
2288 	nfs_wcc_update_inode(inode, fattr);
2289 
2290 	if (pnfs_layoutcommit_outstanding(inode)) {
2291 		nfsi->cache_validity |=
2292 			save_cache_validity &
2293 			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2294 			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2295 			 NFS_INO_INVALID_BLOCKS);
2296 		cache_revalidated = false;
2297 	}
2298 
2299 	/* More cache consistency checks */
2300 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2301 		if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2302 		    nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2303 			/* There is one remaining gap that hasn't been
2304 			 * merged into iversion - do that now.
2305 			 */
2306 			inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2307 			kfree(nfsi->ooo);
2308 			nfsi->ooo = NULL;
2309 		}
2310 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2311 			/* Could it be a race with writeback? */
2312 			if (!(have_writers || have_delegation)) {
2313 				invalid |= NFS_INO_INVALID_DATA
2314 					| NFS_INO_INVALID_ACCESS
2315 					| NFS_INO_INVALID_ACL
2316 					| NFS_INO_INVALID_XATTR;
2317 				/* Force revalidate of all attributes */
2318 				save_cache_validity |= NFS_INO_INVALID_CTIME
2319 					| NFS_INO_INVALID_MTIME
2320 					| NFS_INO_INVALID_SIZE
2321 					| NFS_INO_INVALID_BLOCKS
2322 					| NFS_INO_INVALID_NLINK
2323 					| NFS_INO_INVALID_MODE
2324 					| NFS_INO_INVALID_OTHER
2325 					| NFS_INO_INVALID_BTIME;
2326 				if (S_ISDIR(inode->i_mode))
2327 					nfs_force_lookup_revalidate(inode);
2328 				attr_changed = true;
2329 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2330 						inode->i_sb->s_id,
2331 						inode->i_ino);
2332 			} else if (!have_delegation) {
2333 				nfs_ooo_record(nfsi, fattr);
2334 				nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2335 					      fattr->change_attr);
2336 			}
2337 			inode_set_iversion_raw(inode, fattr->change_attr);
2338 		}
2339 	} else {
2340 		nfsi->cache_validity |=
2341 			save_cache_validity & NFS_INO_INVALID_CHANGE;
2342 		if (!have_delegation ||
2343 		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2344 			cache_revalidated = false;
2345 	}
2346 
2347 	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2348 		inode_set_mtime_to_ts(inode, fattr->mtime);
2349 	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2350 		nfsi->cache_validity |=
2351 			save_cache_validity & NFS_INO_INVALID_MTIME;
2352 
2353 	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2354 		inode_set_ctime_to_ts(inode, fattr->ctime);
2355 	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2356 		nfsi->cache_validity |=
2357 			save_cache_validity & NFS_INO_INVALID_CTIME;
2358 
2359 	if (fattr->valid & NFS_ATTR_FATTR_BTIME)
2360 		nfsi->btime = fattr->btime;
2361 	else if (fattr_supported & NFS_ATTR_FATTR_BTIME)
2362 		nfsi->cache_validity |=
2363 			save_cache_validity & NFS_INO_INVALID_BTIME;
2364 
2365 	/* Check if our cached file size is stale */
2366 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2367 		new_isize = nfs_size_to_loff_t(fattr->size);
2368 		cur_isize = i_size_read(inode);
2369 		if (new_isize != cur_isize && !have_delegation) {
2370 			/* Do we perhaps have any outstanding writes, or has
2371 			 * the file grown beyond our last write? */
2372 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2373 				trace_nfs_size_update(inode, new_isize);
2374 				i_size_write(inode, new_isize);
2375 				if (!have_writers)
2376 					invalid |= NFS_INO_INVALID_DATA;
2377 			}
2378 		}
2379 		if (new_isize == 0 &&
2380 		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2381 				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2382 			fattr->du.nfs3.used = 0;
2383 			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2384 		}
2385 	} else
2386 		nfsi->cache_validity |=
2387 			save_cache_validity & NFS_INO_INVALID_SIZE;
2388 
2389 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2390 		inode_set_atime_to_ts(inode, fattr->atime);
2391 	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2392 		nfsi->cache_validity |=
2393 			save_cache_validity & NFS_INO_INVALID_ATIME;
2394 
2395 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2396 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2397 			umode_t newmode = inode->i_mode & S_IFMT;
2398 			newmode |= fattr->mode & S_IALLUGO;
2399 			inode->i_mode = newmode;
2400 			invalid |= NFS_INO_INVALID_ACCESS
2401 				| NFS_INO_INVALID_ACL;
2402 		}
2403 	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2404 		nfsi->cache_validity |=
2405 			save_cache_validity & NFS_INO_INVALID_MODE;
2406 
2407 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2408 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2409 			invalid |= NFS_INO_INVALID_ACCESS
2410 				| NFS_INO_INVALID_ACL;
2411 			inode->i_uid = fattr->uid;
2412 		}
2413 	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2414 		nfsi->cache_validity |=
2415 			save_cache_validity & NFS_INO_INVALID_OTHER;
2416 
2417 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2418 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2419 			invalid |= NFS_INO_INVALID_ACCESS
2420 				| NFS_INO_INVALID_ACL;
2421 			inode->i_gid = fattr->gid;
2422 		}
2423 	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2424 		nfsi->cache_validity |=
2425 			save_cache_validity & NFS_INO_INVALID_OTHER;
2426 
2427 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2428 		if (inode->i_nlink != fattr->nlink)
2429 			set_nlink(inode, fattr->nlink);
2430 	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2431 		nfsi->cache_validity |=
2432 			save_cache_validity & NFS_INO_INVALID_NLINK;
2433 
2434 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2435 		/*
2436 		 * report the blocks in 512byte units
2437 		 */
2438 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2439 	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2440 		nfsi->cache_validity |=
2441 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2442 
2443 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2444 		inode->i_blocks = fattr->du.nfs2.blocks;
2445 	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2446 		nfsi->cache_validity |=
2447 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2448 
2449 	/* Update attrtimeo value if we're out of the unstable period */
2450 	if (attr_changed) {
2451 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2452 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2453 		nfsi->attrtimeo_timestamp = now;
2454 		/* Set barrier to be more recent than all outstanding updates */
2455 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2456 	} else {
2457 		if (cache_revalidated) {
2458 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2459 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2460 				nfsi->attrtimeo <<= 1;
2461 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2462 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2463 			}
2464 			nfsi->attrtimeo_timestamp = now;
2465 		}
2466 		/* Set the barrier to be more recent than this fattr */
2467 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2468 			nfsi->attr_gencount = fattr->gencount;
2469 	}
2470 
2471 	/* Don't invalidate the data if we were to blame */
2472 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2473 				|| S_ISLNK(inode->i_mode)))
2474 		invalid &= ~NFS_INO_INVALID_DATA;
2475 	nfs_set_cache_invalid(inode, invalid);
2476 
2477 	return 0;
2478  out_err:
2479 	/*
2480 	 * No need to worry about unhashing the dentry, as the
2481 	 * lookup validation will know that the inode is bad.
2482 	 * (But we fall through to invalidate the caches.)
2483 	 */
2484 	nfs_set_inode_stale_locked(inode);
2485 	return -ESTALE;
2486 }
2487 
nfs_alloc_inode(struct super_block * sb)2488 struct inode *nfs_alloc_inode(struct super_block *sb)
2489 {
2490 	struct nfs_inode *nfsi;
2491 	nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2492 	if (!nfsi)
2493 		return NULL;
2494 	nfsi->flags = 0UL;
2495 	nfsi->cache_validity = 0UL;
2496 	nfsi->ooo = NULL;
2497 #if IS_ENABLED(CONFIG_NFS_V4)
2498 	nfsi->nfs4_acl = NULL;
2499 #endif /* CONFIG_NFS_V4 */
2500 #ifdef CONFIG_NFS_V4_2
2501 	nfsi->xattr_cache = NULL;
2502 #endif
2503 	nfs_netfs_inode_init(nfsi);
2504 
2505 	return &nfsi->vfs_inode;
2506 }
2507 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2508 
nfs_free_inode(struct inode * inode)2509 void nfs_free_inode(struct inode *inode)
2510 {
2511 	kfree(NFS_I(inode)->ooo);
2512 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2513 }
2514 EXPORT_SYMBOL_GPL(nfs_free_inode);
2515 
nfs4_init_once(struct nfs_inode * nfsi)2516 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2517 {
2518 #if IS_ENABLED(CONFIG_NFS_V4)
2519 	INIT_LIST_HEAD(&nfsi->open_states);
2520 	nfsi->delegation = NULL;
2521 	init_rwsem(&nfsi->rwsem);
2522 	nfsi->layout = NULL;
2523 #endif
2524 }
2525 
init_once(void * foo)2526 static void init_once(void *foo)
2527 {
2528 	struct nfs_inode *nfsi = foo;
2529 
2530 	inode_init_once(&nfsi->vfs_inode);
2531 	INIT_LIST_HEAD(&nfsi->open_files);
2532 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2533 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2534 	nfs4_init_once(nfsi);
2535 }
2536 
nfs_init_inodecache(void)2537 static int __init nfs_init_inodecache(void)
2538 {
2539 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2540 					     sizeof(struct nfs_inode),
2541 					     0, (SLAB_RECLAIM_ACCOUNT|
2542 						SLAB_ACCOUNT),
2543 					     init_once);
2544 	if (nfs_inode_cachep == NULL)
2545 		return -ENOMEM;
2546 
2547 	return 0;
2548 }
2549 
nfs_destroy_inodecache(void)2550 static void nfs_destroy_inodecache(void)
2551 {
2552 	/*
2553 	 * Make sure all delayed rcu free inodes are flushed before we
2554 	 * destroy cache.
2555 	 */
2556 	rcu_barrier();
2557 	kmem_cache_destroy(nfs_inode_cachep);
2558 }
2559 
2560 struct workqueue_struct *nfslocaliod_workqueue;
2561 struct workqueue_struct *nfsiod_workqueue;
2562 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2563 
2564 /*
2565  * Destroy the nfsiod workqueues
2566  */
nfsiod_stop(void)2567 static void nfsiod_stop(void)
2568 {
2569 	struct workqueue_struct *wq;
2570 
2571 	wq = nfsiod_workqueue;
2572 	if (wq != NULL) {
2573 		nfsiod_workqueue = NULL;
2574 		destroy_workqueue(wq);
2575 	}
2576 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2577 	wq = nfslocaliod_workqueue;
2578 	if (wq != NULL) {
2579 		nfslocaliod_workqueue = NULL;
2580 		destroy_workqueue(wq);
2581 	}
2582 #endif /* CONFIG_NFS_LOCALIO */
2583 }
2584 
2585 /*
2586  * Start the nfsiod workqueues
2587  */
nfsiod_start(void)2588 static int nfsiod_start(void)
2589 {
2590 	dprintk("RPC:       creating workqueue nfsiod\n");
2591 	nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2592 	if (nfsiod_workqueue == NULL)
2593 		return -ENOMEM;
2594 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2595 	/*
2596 	 * localio writes need to use a normal (non-memreclaim) workqueue.
2597 	 * When we start getting low on space, XFS goes and calls flush_work() on
2598 	 * a non-memreclaim work queue, which causes a priority inversion problem.
2599 	 */
2600 	dprintk("RPC:       creating workqueue nfslocaliod\n");
2601 	nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0);
2602 	if (unlikely(nfslocaliod_workqueue == NULL)) {
2603 		nfsiod_stop();
2604 		return -ENOMEM;
2605 	}
2606 #endif /* CONFIG_NFS_LOCALIO */
2607 	return 0;
2608 }
2609 
2610 unsigned int nfs_net_id;
2611 EXPORT_SYMBOL_GPL(nfs_net_id);
2612 
nfs_net_init(struct net * net)2613 static int nfs_net_init(struct net *net)
2614 {
2615 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2616 	int err;
2617 
2618 	nfs_clients_init(net);
2619 
2620 	if (!rpc_proc_register(net, &nn->rpcstats)) {
2621 		err = -ENOMEM;
2622 		goto err_proc_rpc;
2623 	}
2624 
2625 	err = nfs_fs_proc_net_init(net);
2626 	if (err)
2627 		goto err_proc_nfs;
2628 
2629 	return 0;
2630 
2631 err_proc_nfs:
2632 	rpc_proc_unregister(net, "nfs");
2633 err_proc_rpc:
2634 	nfs_clients_exit(net);
2635 	return err;
2636 }
2637 
nfs_net_exit(struct net * net)2638 static void nfs_net_exit(struct net *net)
2639 {
2640 	rpc_proc_unregister(net, "nfs");
2641 	nfs_fs_proc_net_exit(net);
2642 	nfs_clients_exit(net);
2643 }
2644 
2645 static struct pernet_operations nfs_net_ops = {
2646 	.init = nfs_net_init,
2647 	.exit = nfs_net_exit,
2648 	.id   = &nfs_net_id,
2649 	.size = sizeof(struct nfs_net),
2650 };
2651 
2652 #ifdef CONFIG_KEYS
2653 static struct key *nfs_keyring;
2654 
nfs_init_keyring(void)2655 static int __init nfs_init_keyring(void)
2656 {
2657 	nfs_keyring = keyring_alloc(".nfs",
2658 			     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
2659 			     current_cred(),
2660 			     (KEY_POS_ALL & ~KEY_POS_SETATTR) |
2661 			     (KEY_USR_ALL & ~KEY_USR_SETATTR),
2662 			     KEY_ALLOC_NOT_IN_QUOTA, NULL, NULL);
2663 	return PTR_ERR_OR_ZERO(nfs_keyring);
2664 }
2665 
nfs_exit_keyring(void)2666 static void nfs_exit_keyring(void)
2667 {
2668 	key_put(nfs_keyring);
2669 }
2670 #else
nfs_init_keyring(void)2671 static inline int nfs_init_keyring(void)
2672 {
2673 	return 0;
2674 }
2675 
nfs_exit_keyring(void)2676 static inline void nfs_exit_keyring(void)
2677 {
2678 }
2679 #endif /* CONFIG_KEYS */
2680 
2681 /*
2682  * Initialize NFS
2683  */
init_nfs_fs(void)2684 static int __init init_nfs_fs(void)
2685 {
2686 	int err;
2687 
2688 	err = nfs_init_keyring();
2689 	if (err)
2690 		return err;
2691 
2692 	err = nfs_sysfs_init();
2693 	if (err < 0)
2694 		goto out10;
2695 
2696 	err = register_pernet_subsys(&nfs_net_ops);
2697 	if (err < 0)
2698 		goto out9;
2699 
2700 	err = nfsiod_start();
2701 	if (err)
2702 		goto out7;
2703 
2704 	err = nfs_fs_proc_init();
2705 	if (err)
2706 		goto out6;
2707 
2708 	err = nfs_init_nfspagecache();
2709 	if (err)
2710 		goto out5;
2711 
2712 	err = nfs_init_inodecache();
2713 	if (err)
2714 		goto out4;
2715 
2716 	err = nfs_init_readpagecache();
2717 	if (err)
2718 		goto out3;
2719 
2720 	err = nfs_init_writepagecache();
2721 	if (err)
2722 		goto out2;
2723 
2724 	err = nfs_init_directcache();
2725 	if (err)
2726 		goto out1;
2727 
2728 	err = register_nfs_fs();
2729 	if (err)
2730 		goto out0;
2731 
2732 	return 0;
2733 out0:
2734 	nfs_destroy_directcache();
2735 out1:
2736 	nfs_destroy_writepagecache();
2737 out2:
2738 	nfs_destroy_readpagecache();
2739 out3:
2740 	nfs_destroy_inodecache();
2741 out4:
2742 	nfs_destroy_nfspagecache();
2743 out5:
2744 	nfs_fs_proc_exit();
2745 out6:
2746 	nfsiod_stop();
2747 out7:
2748 	unregister_pernet_subsys(&nfs_net_ops);
2749 out9:
2750 	nfs_sysfs_exit();
2751 out10:
2752 	nfs_exit_keyring();
2753 	return err;
2754 }
2755 
exit_nfs_fs(void)2756 static void __exit exit_nfs_fs(void)
2757 {
2758 	nfs_destroy_directcache();
2759 	nfs_destroy_writepagecache();
2760 	nfs_destroy_readpagecache();
2761 	nfs_destroy_inodecache();
2762 	nfs_destroy_nfspagecache();
2763 	unregister_pernet_subsys(&nfs_net_ops);
2764 	unregister_nfs_fs();
2765 	nfs_fs_proc_exit();
2766 	nfsiod_stop();
2767 	nfs_sysfs_exit();
2768 	nfs_exit_keyring();
2769 }
2770 
2771 /* Not quite true; I just maintain it */
2772 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2773 MODULE_DESCRIPTION("NFS client support");
2774 MODULE_LICENSE("GPL");
2775 module_param(enable_ino64, bool, 0644);
2776 
2777 module_init(init_nfs_fs)
2778 module_exit(exit_nfs_fs)
2779