xref: /linux/fs/nfs/inode.c (revision 6bb34aff1ebdd4ee8ea1721068f74d476d707f01)
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) || inode_generic_drop(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_state_read_once(inode) & 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 		icount_read(inode));
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 	loff_t oldsize = i_size_read(inode);
720 	int error = 0;
721 	kuid_t task_uid = current_fsuid();
722 	kuid_t owner_uid = inode->i_uid;
723 
724 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
725 
726 	/* skip mode change if it's just for clearing setuid/setgid */
727 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
728 		attr->ia_valid &= ~ATTR_MODE;
729 
730 	if (attr->ia_valid & ATTR_SIZE) {
731 		BUG_ON(!S_ISREG(inode->i_mode));
732 
733 		error = inode_newsize_ok(inode, attr->ia_size);
734 		if (error)
735 			return error;
736 
737 		if (attr->ia_size == oldsize)
738 			attr->ia_valid &= ~ATTR_SIZE;
739 	}
740 
741 	if (nfs_have_delegated_mtime(inode) && attr->ia_valid & ATTR_MTIME) {
742 		spin_lock(&inode->i_lock);
743 		if (attr->ia_valid & ATTR_MTIME_SET) {
744 			if (uid_eq(task_uid, owner_uid)) {
745 				nfs_set_timestamps_to_ts(inode, attr);
746 				attr->ia_valid &= ~(ATTR_MTIME|ATTR_MTIME_SET|
747 						ATTR_ATIME|ATTR_ATIME_SET);
748 			}
749 		} else {
750 			nfs_update_timestamps(inode, attr->ia_valid);
751 			attr->ia_valid &= ~(ATTR_MTIME|ATTR_ATIME);
752 		}
753 		spin_unlock(&inode->i_lock);
754 	} else if (nfs_have_delegated_atime(inode) &&
755 		   attr->ia_valid & ATTR_ATIME &&
756 		   !(attr->ia_valid & ATTR_MTIME)) {
757 		if (attr->ia_valid & ATTR_ATIME_SET) {
758 			if (uid_eq(task_uid, owner_uid)) {
759 				spin_lock(&inode->i_lock);
760 				nfs_set_timestamps_to_ts(inode, attr);
761 				spin_unlock(&inode->i_lock);
762 				attr->ia_valid &= ~(ATTR_ATIME|ATTR_ATIME_SET);
763 			}
764 		} else {
765 			nfs_update_delegated_atime(inode);
766 			attr->ia_valid &= ~ATTR_ATIME;
767 		}
768 	}
769 
770 	/* Optimization: if the end result is no change, don't RPC */
771 	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
772 		return 0;
773 
774 	trace_nfs_setattr_enter(inode);
775 
776 	/* Write all dirty data */
777 	if (S_ISREG(inode->i_mode)) {
778 		nfs_file_block_o_direct(NFS_I(inode));
779 		nfs_sync_inode(inode);
780 	}
781 
782 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
783 	if (fattr == NULL) {
784 		error = -ENOMEM;
785 		goto out;
786 	}
787 
788 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
789 	if (error == 0) {
790 		if (attr->ia_valid & ATTR_SIZE)
791 			nfs_truncate_last_folio(inode->i_mapping, oldsize,
792 						attr->ia_size);
793 		error = nfs_refresh_inode(inode, fattr);
794 	}
795 	nfs_free_fattr(fattr);
796 out:
797 	trace_nfs_setattr_exit(inode, error);
798 	return error;
799 }
800 EXPORT_SYMBOL_GPL(nfs_setattr);
801 
802 /**
803  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
804  * @inode: inode of the file used
805  * @offset: file offset to start truncating
806  *
807  * This is a copy of the common vmtruncate, but with the locking
808  * corrected to take into account the fact that NFS requires
809  * inode->i_size to be updated under the inode->i_lock.
810  * Note: must be called with inode->i_lock held!
811  */
nfs_vmtruncate(struct inode * inode,loff_t offset)812 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
813 {
814 	int err;
815 
816 	err = inode_newsize_ok(inode, offset);
817 	if (err)
818 		goto out;
819 
820 	trace_nfs_size_truncate(inode, offset);
821 	i_size_write(inode, offset);
822 	/* Optimisation */
823 	if (offset == 0) {
824 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
825 		nfs_ooo_clear(NFS_I(inode));
826 	}
827 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
828 
829 	spin_unlock(&inode->i_lock);
830 	truncate_pagecache(inode, offset);
831 	nfs_update_delegated_mtime_locked(inode);
832 	spin_lock(&inode->i_lock);
833 out:
834 	return err;
835 }
836 
837 /**
838  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
839  * @inode: pointer to struct inode
840  * @attr: pointer to struct iattr
841  * @fattr: pointer to struct nfs_fattr
842  *
843  * Note: we do this in the *proc.c in order to ensure that
844  *       it works for things like exclusive creates too.
845  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)846 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
847 		struct nfs_fattr *fattr)
848 {
849 	/* Barrier: bump the attribute generation count. */
850 	nfs_fattr_set_barrier(fattr);
851 
852 	spin_lock(&inode->i_lock);
853 	NFS_I(inode)->attr_gencount = fattr->gencount;
854 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
855 		if (!nfs_have_delegated_mtime(inode))
856 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
857 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
858 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
859 		nfs_vmtruncate(inode, attr->ia_size);
860 	}
861 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
862 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
863 		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
864 		    inode->i_mode & S_ISUID)
865 			inode->i_mode &= ~S_ISUID;
866 		if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
867 			inode->i_mode &= ~S_ISGID;
868 		if ((attr->ia_valid & ATTR_MODE) != 0) {
869 			int mode = attr->ia_mode & S_IALLUGO;
870 			mode |= inode->i_mode & ~S_IALLUGO;
871 			inode->i_mode = mode;
872 		}
873 		if ((attr->ia_valid & ATTR_UID) != 0)
874 			inode->i_uid = attr->ia_uid;
875 		if ((attr->ia_valid & ATTR_GID) != 0)
876 			inode->i_gid = attr->ia_gid;
877 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
878 			inode_set_ctime_to_ts(inode, fattr->ctime);
879 		else
880 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
881 					| NFS_INO_INVALID_CTIME);
882 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
883 				| NFS_INO_INVALID_ACL);
884 	}
885 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
886 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
887 				| NFS_INO_INVALID_CTIME);
888 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
889 			inode_set_atime_to_ts(inode, fattr->atime);
890 		else if (attr->ia_valid & ATTR_ATIME_SET)
891 			inode_set_atime_to_ts(inode, attr->ia_atime);
892 		else
893 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
894 
895 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
896 			inode_set_ctime_to_ts(inode, fattr->ctime);
897 		else
898 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
899 					| NFS_INO_INVALID_CTIME);
900 	}
901 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
902 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
903 				| NFS_INO_INVALID_CTIME);
904 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
905 			inode_set_mtime_to_ts(inode, fattr->mtime);
906 		else if (attr->ia_valid & ATTR_MTIME_SET)
907 			inode_set_mtime_to_ts(inode, attr->ia_mtime);
908 		else
909 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
910 
911 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
912 			inode_set_ctime_to_ts(inode, fattr->ctime);
913 		else
914 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
915 					| NFS_INO_INVALID_CTIME);
916 	}
917 	if (fattr->valid)
918 		nfs_update_inode(inode, fattr);
919 	spin_unlock(&inode->i_lock);
920 }
921 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
922 
923 /*
924  * Don't request help from readdirplus if the file is being written to,
925  * or if attribute caching is turned off
926  */
nfs_getattr_readdirplus_enable(const struct inode * inode)927 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
928 {
929 	return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
930 	       !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
931 }
932 
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)933 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
934 {
935 	if (!IS_ROOT(dentry)) {
936 		struct dentry *parent = dget_parent(dentry);
937 		nfs_readdir_record_entry_cache_miss(d_inode(parent));
938 		dput(parent);
939 	}
940 }
941 
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)942 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
943 {
944 	if (!IS_ROOT(dentry)) {
945 		struct dentry *parent = dget_parent(dentry);
946 		nfs_readdir_record_entry_cache_hit(d_inode(parent));
947 		dput(parent);
948 	}
949 }
950 
nfs_get_valid_attrmask(struct inode * inode)951 static u32 nfs_get_valid_attrmask(struct inode *inode)
952 {
953 	u64 fattr_valid = NFS_SERVER(inode)->fattr_valid;
954 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
955 	u32 reply_mask = STATX_INO | STATX_TYPE;
956 
957 	if (!(cache_validity & NFS_INO_INVALID_ATIME))
958 		reply_mask |= STATX_ATIME;
959 	if (!(cache_validity & NFS_INO_INVALID_CTIME))
960 		reply_mask |= STATX_CTIME;
961 	if (!(cache_validity & NFS_INO_INVALID_MTIME))
962 		reply_mask |= STATX_MTIME;
963 	if (!(cache_validity & NFS_INO_INVALID_SIZE))
964 		reply_mask |= STATX_SIZE;
965 	if (!(cache_validity & NFS_INO_INVALID_NLINK))
966 		reply_mask |= STATX_NLINK;
967 	if (!(cache_validity & NFS_INO_INVALID_MODE))
968 		reply_mask |= STATX_MODE;
969 	if (!(cache_validity & NFS_INO_INVALID_OTHER))
970 		reply_mask |= STATX_UID | STATX_GID;
971 	if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
972 		reply_mask |= STATX_BLOCKS;
973 	if (!(cache_validity & NFS_INO_INVALID_BTIME) &&
974 	    (fattr_valid & NFS_ATTR_FATTR_BTIME))
975 		reply_mask |= STATX_BTIME;
976 	if (!(cache_validity & NFS_INO_INVALID_CHANGE))
977 		reply_mask |= STATX_CHANGE_COOKIE;
978 	return reply_mask;
979 }
980 
nfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)981 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
982 		struct kstat *stat, u32 request_mask, unsigned int query_flags)
983 {
984 	struct inode *inode = d_inode(path->dentry);
985 	struct nfs_server *server = NFS_SERVER(inode);
986 	u64 fattr_valid = server->fattr_valid;
987 	unsigned long cache_validity;
988 	int err = 0;
989 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
990 	bool do_update = false;
991 	bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
992 
993 	trace_nfs_getattr_enter(inode);
994 
995 	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
996 			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
997 			STATX_INO | STATX_SIZE | STATX_BLOCKS | STATX_BTIME |
998 			STATX_CHANGE_COOKIE;
999 
1000 	if (!(fattr_valid & NFS_ATTR_FATTR_BTIME))
1001 		request_mask &= ~STATX_BTIME;
1002 
1003 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
1004 		if (readdirplus_enabled)
1005 			nfs_readdirplus_parent_cache_hit(path->dentry);
1006 		goto out_no_revalidate;
1007 	}
1008 
1009 	/* Flush out writes to the server in order to update c/mtime/version.  */
1010 	if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
1011 	    S_ISREG(inode->i_mode)) {
1012 		if (nfs_have_delegated_mtime(inode))
1013 			filemap_fdatawrite(inode->i_mapping);
1014 		else
1015 			filemap_write_and_wait(inode->i_mapping);
1016 	}
1017 
1018 	/*
1019 	 * We may force a getattr if the user cares about atime.
1020 	 *
1021 	 * Note that we only have to check the vfsmount flags here:
1022 	 *  - NFS always sets S_NOATIME by so checking it would give a
1023 	 *    bogus result
1024 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
1025 	 *    no point in checking those.
1026 	 */
1027 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
1028 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1029 		request_mask &= ~STATX_ATIME;
1030 
1031 	/* Is the user requesting attributes that might need revalidation? */
1032 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
1033 					STATX_MTIME|STATX_UID|STATX_GID|
1034 					STATX_SIZE|STATX_BLOCKS|STATX_BTIME|
1035 					STATX_CHANGE_COOKIE)))
1036 		goto out_no_revalidate;
1037 
1038 	/* Check whether the cached attributes are stale */
1039 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
1040 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
1041 	do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
1042 	if (request_mask & STATX_ATIME)
1043 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
1044 	if (request_mask & STATX_CTIME)
1045 		do_update |= cache_validity & NFS_INO_INVALID_CTIME;
1046 	if (request_mask & STATX_MTIME)
1047 		do_update |= cache_validity & NFS_INO_INVALID_MTIME;
1048 	if (request_mask & STATX_SIZE)
1049 		do_update |= cache_validity & NFS_INO_INVALID_SIZE;
1050 	if (request_mask & STATX_NLINK)
1051 		do_update |= cache_validity & NFS_INO_INVALID_NLINK;
1052 	if (request_mask & STATX_MODE)
1053 		do_update |= cache_validity & NFS_INO_INVALID_MODE;
1054 	if (request_mask & (STATX_UID | STATX_GID))
1055 		do_update |= cache_validity & NFS_INO_INVALID_OTHER;
1056 	if (request_mask & STATX_BLOCKS)
1057 		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
1058 	if (request_mask & STATX_BTIME)
1059 		do_update |= cache_validity & NFS_INO_INVALID_BTIME;
1060 
1061 	if (do_update) {
1062 		if (readdirplus_enabled)
1063 			nfs_readdirplus_parent_cache_miss(path->dentry);
1064 		err = __nfs_revalidate_inode(server, inode);
1065 		if (err)
1066 			goto out;
1067 	} else if (readdirplus_enabled)
1068 		nfs_readdirplus_parent_cache_hit(path->dentry);
1069 out_no_revalidate:
1070 	/* Only return attributes that were revalidated. */
1071 	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
1072 
1073 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1074 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
1075 	stat->change_cookie = inode_peek_iversion_raw(inode);
1076 	stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
1077 	if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
1078 		stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
1079 	if (S_ISDIR(inode->i_mode))
1080 		stat->blksize = NFS_SERVER(inode)->dtsize;
1081 	stat->btime = NFS_I(inode)->btime;
1082 
1083 	/* Special handling for STATX_DIOALIGN and STATX_DIO_READ_ALIGN
1084 	 * - NFS doesn't have DIO alignment constraints, avoid getting
1085 	 *   these DIO attrs from remote and just respond with most
1086 	 *   accommodating limits (so client will issue supported DIO).
1087 	 * - this is unintuitive, but the most coarse-grained
1088 	 *   dio_offset_align is the most accommodating.
1089 	 */
1090 	if ((request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) &&
1091 	    S_ISREG(inode->i_mode)) {
1092 		stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN;
1093 		stat->dio_mem_align = 4; /* 4-byte alignment */
1094 		stat->dio_offset_align = PAGE_SIZE;
1095 		stat->dio_read_offset_align = stat->dio_offset_align;
1096 	}
1097 out:
1098 	trace_nfs_getattr_exit(inode, err);
1099 	return err;
1100 }
1101 EXPORT_SYMBOL_GPL(nfs_getattr);
1102 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)1103 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
1104 {
1105 	refcount_set(&l_ctx->count, 1);
1106 	l_ctx->lockowner = current->files;
1107 	INIT_LIST_HEAD(&l_ctx->list);
1108 	atomic_set(&l_ctx->io_count, 0);
1109 }
1110 
__nfs_find_lock_context(struct nfs_open_context * ctx)1111 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
1112 {
1113 	struct nfs_lock_context *pos;
1114 
1115 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
1116 		if (pos->lockowner != current->files)
1117 			continue;
1118 		if (refcount_inc_not_zero(&pos->count))
1119 			return pos;
1120 	}
1121 	return NULL;
1122 }
1123 
nfs_get_lock_context(struct nfs_open_context * ctx)1124 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
1125 {
1126 	struct nfs_lock_context *res, *new = NULL;
1127 	struct inode *inode = d_inode(ctx->dentry);
1128 
1129 	rcu_read_lock();
1130 	res = __nfs_find_lock_context(ctx);
1131 	rcu_read_unlock();
1132 	if (res == NULL) {
1133 		new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1134 		if (new == NULL)
1135 			return ERR_PTR(-ENOMEM);
1136 		nfs_init_lock_context(new);
1137 		spin_lock(&inode->i_lock);
1138 		res = __nfs_find_lock_context(ctx);
1139 		if (res == NULL) {
1140 			new->open_context = get_nfs_open_context(ctx);
1141 			if (new->open_context) {
1142 				list_add_tail_rcu(&new->list,
1143 						&ctx->lock_context.list);
1144 				res = new;
1145 				new = NULL;
1146 			} else
1147 				res = ERR_PTR(-EBADF);
1148 		}
1149 		spin_unlock(&inode->i_lock);
1150 		kfree(new);
1151 	}
1152 	return res;
1153 }
1154 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
1155 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)1156 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
1157 {
1158 	struct nfs_open_context *ctx = l_ctx->open_context;
1159 	struct inode *inode = d_inode(ctx->dentry);
1160 
1161 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
1162 		return;
1163 	list_del_rcu(&l_ctx->list);
1164 	spin_unlock(&inode->i_lock);
1165 	put_nfs_open_context(ctx);
1166 	kfree_rcu(l_ctx, rcu_head);
1167 }
1168 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
1169 
1170 /**
1171  * nfs_close_context - Common close_context() routine NFSv2/v3
1172  * @ctx: pointer to context
1173  * @is_sync: is this a synchronous close
1174  *
1175  * Ensure that the attributes are up to date if we're mounted
1176  * with close-to-open semantics and we have cached data that will
1177  * need to be revalidated on open.
1178  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)1179 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1180 {
1181 	struct nfs_inode *nfsi;
1182 	struct inode *inode;
1183 
1184 	if (!(ctx->mode & FMODE_WRITE))
1185 		return;
1186 	if (!is_sync)
1187 		return;
1188 	inode = d_inode(ctx->dentry);
1189 	if (nfs_have_read_or_write_delegation(inode))
1190 		return;
1191 	nfsi = NFS_I(inode);
1192 	if (inode->i_mapping->nrpages == 0)
1193 		return;
1194 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1195 		return;
1196 	if (!list_empty(&nfsi->open_files))
1197 		return;
1198 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1199 		return;
1200 	nfs_revalidate_inode(inode,
1201 			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1202 }
1203 EXPORT_SYMBOL_GPL(nfs_close_context);
1204 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1205 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1206 						fmode_t f_mode,
1207 						struct file *filp)
1208 {
1209 	struct nfs_open_context *ctx;
1210 
1211 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1212 	if (!ctx)
1213 		return ERR_PTR(-ENOMEM);
1214 	nfs_sb_active(dentry->d_sb);
1215 	ctx->dentry = dget(dentry);
1216 	if (filp)
1217 		ctx->cred = get_cred(filp->f_cred);
1218 	else
1219 		ctx->cred = get_current_cred();
1220 	rcu_assign_pointer(ctx->ll_cred, NULL);
1221 	ctx->state = NULL;
1222 	ctx->mode = f_mode;
1223 	ctx->flags = 0;
1224 	ctx->error = 0;
1225 	ctx->flock_owner = (fl_owner_t)filp;
1226 	nfs_init_lock_context(&ctx->lock_context);
1227 	ctx->lock_context.open_context = ctx;
1228 	INIT_LIST_HEAD(&ctx->list);
1229 	ctx->mdsthreshold = NULL;
1230 	nfs_localio_file_init(&ctx->nfl);
1231 
1232 	return ctx;
1233 }
1234 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1235 
get_nfs_open_context(struct nfs_open_context * ctx)1236 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1237 {
1238 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1239 		return ctx;
1240 	return NULL;
1241 }
1242 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1243 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1244 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1245 {
1246 	struct inode *inode = d_inode(ctx->dentry);
1247 	struct super_block *sb = ctx->dentry->d_sb;
1248 
1249 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1250 		return;
1251 	if (!list_empty(&ctx->list)) {
1252 		spin_lock(&inode->i_lock);
1253 		list_del_rcu(&ctx->list);
1254 		spin_unlock(&inode->i_lock);
1255 	}
1256 	if (inode != NULL)
1257 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1258 	put_cred(ctx->cred);
1259 	dput(ctx->dentry);
1260 	nfs_sb_deactive(sb);
1261 	put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1262 	kfree(ctx->mdsthreshold);
1263 	nfs_close_local_fh(&ctx->nfl);
1264 	kfree_rcu(ctx, rcu_head);
1265 }
1266 
put_nfs_open_context(struct nfs_open_context * ctx)1267 void put_nfs_open_context(struct nfs_open_context *ctx)
1268 {
1269 	__put_nfs_open_context(ctx, 0);
1270 }
1271 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1272 
put_nfs_open_context_sync(struct nfs_open_context * ctx)1273 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1274 {
1275 	__put_nfs_open_context(ctx, 1);
1276 }
1277 
1278 /*
1279  * Ensure that mmap has a recent RPC credential for use when writing out
1280  * shared pages
1281  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1282 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1283 {
1284 	struct inode *inode = d_inode(ctx->dentry);
1285 	struct nfs_inode *nfsi = NFS_I(inode);
1286 
1287 	spin_lock(&inode->i_lock);
1288 	if (list_empty(&nfsi->open_files) &&
1289 	    nfs_ooo_test(nfsi))
1290 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1291 						     NFS_INO_REVAL_FORCED);
1292 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1293 	spin_unlock(&inode->i_lock);
1294 }
1295 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1296 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1297 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1298 {
1299 	filp->private_data = get_nfs_open_context(ctx);
1300 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1301 	if (list_empty(&ctx->list))
1302 		nfs_inode_attach_open_context(ctx);
1303 }
1304 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1305 
1306 /*
1307  * Given an inode, search for an open context with the desired characteristics
1308  */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1309 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1310 {
1311 	struct nfs_inode *nfsi = NFS_I(inode);
1312 	struct nfs_open_context *pos, *ctx = NULL;
1313 
1314 	rcu_read_lock();
1315 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1316 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1317 			continue;
1318 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1319 			continue;
1320 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1321 			continue;
1322 		ctx = get_nfs_open_context(pos);
1323 		if (ctx)
1324 			break;
1325 	}
1326 	rcu_read_unlock();
1327 	return ctx;
1328 }
1329 
nfs_file_clear_open_context(struct file * filp)1330 void nfs_file_clear_open_context(struct file *filp)
1331 {
1332 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1333 
1334 	if (ctx) {
1335 		struct inode *inode = d_inode(ctx->dentry);
1336 
1337 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1338 		/*
1339 		 * We fatal error on write before. Try to writeback
1340 		 * every page again.
1341 		 */
1342 		if (ctx->error < 0)
1343 			invalidate_inode_pages2(inode->i_mapping);
1344 		filp->private_data = NULL;
1345 		put_nfs_open_context_sync(ctx);
1346 	}
1347 }
1348 
1349 /*
1350  * These allocate and release file read/write context information.
1351  */
nfs_open(struct inode * inode,struct file * filp)1352 int nfs_open(struct inode *inode, struct file *filp)
1353 {
1354 	struct nfs_open_context *ctx;
1355 
1356 	ctx = alloc_nfs_open_context(file_dentry(filp),
1357 				     flags_to_mode(filp->f_flags), filp);
1358 	if (IS_ERR(ctx))
1359 		return PTR_ERR(ctx);
1360 	nfs_file_set_open_context(filp, ctx);
1361 	put_nfs_open_context(ctx);
1362 	nfs_fscache_open_file(inode, filp);
1363 	return 0;
1364 }
1365 
1366 /*
1367  * This function is called whenever some part of NFS notices that
1368  * the cached attributes have to be refreshed.
1369  */
1370 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1371 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1372 {
1373 	int		 status = -ESTALE;
1374 	struct nfs_fattr *fattr = NULL;
1375 	struct nfs_inode *nfsi = NFS_I(inode);
1376 
1377 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1378 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1379 
1380 	trace_nfs_revalidate_inode_enter(inode);
1381 
1382 	if (is_bad_inode(inode))
1383 		goto out;
1384 	if (NFS_STALE(inode))
1385 		goto out;
1386 
1387 	/* pNFS: Attributes aren't updated until we layoutcommit */
1388 	if (S_ISREG(inode->i_mode)) {
1389 		status = pnfs_sync_inode(inode, false);
1390 		if (status)
1391 			goto out;
1392 	} else if (nfs_have_directory_delegation(inode)) {
1393 		status = 0;
1394 		goto out;
1395 	}
1396 
1397 	status = -ENOMEM;
1398 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1399 	if (fattr == NULL)
1400 		goto out;
1401 
1402 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1403 
1404 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1405 	if (status != 0) {
1406 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1407 			 inode->i_sb->s_id,
1408 			 (unsigned long long)NFS_FILEID(inode), status);
1409 		switch (status) {
1410 		case -ETIMEDOUT:
1411 			/* A soft timeout occurred. Use cached information? */
1412 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1413 				status = 0;
1414 			break;
1415 		case -ESTALE:
1416 			if (!S_ISDIR(inode->i_mode))
1417 				nfs_set_inode_stale(inode);
1418 			else
1419 				nfs_zap_caches(inode);
1420 		}
1421 		goto out;
1422 	}
1423 
1424 	status = nfs_refresh_inode(inode, fattr);
1425 	if (status) {
1426 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1427 			 inode->i_sb->s_id,
1428 			 (unsigned long long)NFS_FILEID(inode), status);
1429 		goto out;
1430 	}
1431 
1432 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1433 		nfs_zap_acl_cache(inode);
1434 
1435 	nfs_setsecurity(inode, fattr);
1436 
1437 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1438 		inode->i_sb->s_id,
1439 		(unsigned long long)NFS_FILEID(inode));
1440 
1441 out:
1442 	nfs_free_fattr(fattr);
1443 	trace_nfs_revalidate_inode_exit(inode, status);
1444 	return status;
1445 }
1446 
nfs_attribute_cache_expired(struct inode * inode)1447 int nfs_attribute_cache_expired(struct inode *inode)
1448 {
1449 	if (nfs_have_delegated_attributes(inode))
1450 		return 0;
1451 	return nfs_attribute_timeout(inode);
1452 }
1453 
1454 /**
1455  * nfs_revalidate_inode - Revalidate the inode attributes
1456  * @inode: pointer to inode struct
1457  * @flags: cache flags to check
1458  *
1459  * Updates inode attribute information by retrieving the data from the server.
1460  */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1461 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1462 {
1463 	if (!nfs_check_cache_invalid(inode, flags))
1464 		return NFS_STALE(inode) ? -ESTALE : 0;
1465 	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1466 }
1467 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1468 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1469 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1470 {
1471 	int ret;
1472 
1473 	nfs_fscache_invalidate(inode, 0);
1474 	if (mapping->nrpages != 0) {
1475 		if (S_ISREG(inode->i_mode)) {
1476 			ret = nfs_sync_mapping(mapping);
1477 			if (ret < 0)
1478 				return ret;
1479 		}
1480 		ret = invalidate_inode_pages2(mapping);
1481 		if (ret < 0)
1482 			return ret;
1483 	}
1484 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1485 
1486 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1487 			inode->i_sb->s_id,
1488 			(unsigned long long)NFS_FILEID(inode));
1489 	return 0;
1490 }
1491 
1492 /**
1493  * nfs_clear_invalid_mapping - Conditionally clear a mapping
1494  * @mapping: pointer to mapping
1495  *
1496  * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1497  */
nfs_clear_invalid_mapping(struct address_space * mapping)1498 int nfs_clear_invalid_mapping(struct address_space *mapping)
1499 {
1500 	struct inode *inode = mapping->host;
1501 	struct nfs_inode *nfsi = NFS_I(inode);
1502 	unsigned long *bitlock = &nfsi->flags;
1503 	int ret = 0;
1504 
1505 	/*
1506 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1507 	 * invalidations that come in while we're shooting down the mappings
1508 	 * are respected. But, that leaves a race window where one revalidator
1509 	 * can clear the flag, and then another checks it before the mapping
1510 	 * gets invalidated. Fix that by serializing access to this part of
1511 	 * the function.
1512 	 *
1513 	 * At the same time, we need to allow other tasks to see whether we
1514 	 * might be in the middle of invalidating the pages, so we only set
1515 	 * the bit lock here if it looks like we're going to be doing that.
1516 	 */
1517 	for (;;) {
1518 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1519 					 nfs_wait_bit_killable,
1520 					 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1521 		if (ret)
1522 			goto out;
1523 		smp_rmb(); /* pairs with smp_wmb() below */
1524 		if (test_bit(NFS_INO_INVALIDATING, bitlock))
1525 			continue;
1526 		/* pairs with nfs_set_cache_invalid()'s smp_store_release() */
1527 		if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA))
1528 			goto out;
1529 		/* Slow-path that double-checks with spinlock held */
1530 		spin_lock(&inode->i_lock);
1531 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1532 			spin_unlock(&inode->i_lock);
1533 			continue;
1534 		}
1535 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1536 			break;
1537 		spin_unlock(&inode->i_lock);
1538 		goto out;
1539 	}
1540 
1541 	set_bit(NFS_INO_INVALIDATING, bitlock);
1542 	smp_wmb();
1543 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1544 	nfs_ooo_clear(nfsi);
1545 	spin_unlock(&inode->i_lock);
1546 	trace_nfs_invalidate_mapping_enter(inode);
1547 	ret = nfs_invalidate_mapping(inode, mapping);
1548 	trace_nfs_invalidate_mapping_exit(inode, ret);
1549 
1550 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1551 	smp_mb__after_atomic();
1552 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1553 out:
1554 	return ret;
1555 }
1556 
nfs_mapping_need_revalidate_inode(struct inode * inode)1557 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1558 {
1559 	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1560 		NFS_STALE(inode);
1561 }
1562 
nfs_revalidate_mapping_rcu(struct inode * inode)1563 int nfs_revalidate_mapping_rcu(struct inode *inode)
1564 {
1565 	struct nfs_inode *nfsi = NFS_I(inode);
1566 	unsigned long *bitlock = &nfsi->flags;
1567 	int ret = 0;
1568 
1569 	if (IS_SWAPFILE(inode))
1570 		goto out;
1571 	if (nfs_mapping_need_revalidate_inode(inode)) {
1572 		ret = -ECHILD;
1573 		goto out;
1574 	}
1575 	spin_lock(&inode->i_lock);
1576 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1577 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1578 		ret = -ECHILD;
1579 	spin_unlock(&inode->i_lock);
1580 out:
1581 	return ret;
1582 }
1583 
1584 /**
1585  * nfs_revalidate_mapping - Revalidate the pagecache
1586  * @inode: pointer to host inode
1587  * @mapping: pointer to mapping
1588  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1589 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1590 {
1591 	/* swapfiles are not supposed to be shared. */
1592 	if (IS_SWAPFILE(inode))
1593 		return 0;
1594 
1595 	if (nfs_mapping_need_revalidate_inode(inode)) {
1596 		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1597 		if (ret < 0)
1598 			return ret;
1599 	}
1600 
1601 	return nfs_clear_invalid_mapping(mapping);
1602 }
1603 
nfs_file_has_writers(struct nfs_inode * nfsi)1604 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1605 {
1606 	struct inode *inode = &nfsi->vfs_inode;
1607 
1608 	if (!S_ISREG(inode->i_mode))
1609 		return false;
1610 	if (list_empty(&nfsi->open_files))
1611 		return false;
1612 	return inode_is_open_for_write(inode);
1613 }
1614 
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1615 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1616 {
1617 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1618 }
1619 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1620 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1621 {
1622 	struct timespec64 ts;
1623 
1624 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1625 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1626 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1627 		inode_set_iversion_raw(inode, fattr->change_attr);
1628 		if (S_ISDIR(inode->i_mode))
1629 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1630 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1631 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1632 	}
1633 	/* If we have atomic WCC data, we may update some attributes */
1634 	ts = inode_get_ctime(inode);
1635 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1636 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1637 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1638 		inode_set_ctime_to_ts(inode, fattr->ctime);
1639 	}
1640 
1641 	ts = inode_get_mtime(inode);
1642 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1643 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1644 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1645 		inode_set_mtime_to_ts(inode, fattr->mtime);
1646 	}
1647 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1648 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1649 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1650 			&& !nfs_have_writebacks(inode)) {
1651 		trace_nfs_size_wcc(inode, fattr->size);
1652 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1653 	}
1654 }
1655 
1656 /**
1657  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1658  * @inode: pointer to inode
1659  * @fattr: updated attributes
1660  *
1661  * Verifies the attribute cache. If we have just changed the attributes,
1662  * so that fattr carries weak cache consistency data, then it may
1663  * also update the ctime/mtime/change_attribute.
1664  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1665 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1666 {
1667 	struct nfs_inode *nfsi = NFS_I(inode);
1668 	loff_t cur_size, new_isize;
1669 	unsigned long invalid = 0;
1670 	struct timespec64 ts;
1671 
1672 	if (nfs_have_delegated_attributes(inode))
1673 		return 0;
1674 
1675 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1676 		/* Only a mounted-on-fileid? Just exit */
1677 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1678 			return 0;
1679 	/* Has the inode gone and changed behind our back? */
1680 	} else if (nfsi->fileid != fattr->fileid) {
1681 		/* Is this perhaps the mounted-on fileid? */
1682 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1683 		    nfsi->fileid == fattr->mounted_on_fileid)
1684 			return 0;
1685 		return -ESTALE;
1686 	}
1687 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1688 		return -ESTALE;
1689 
1690 
1691 	if (!nfs_file_has_buffered_writers(nfsi)) {
1692 		/* Verify a few of the more important attributes */
1693 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1694 			invalid |= NFS_INO_INVALID_CHANGE;
1695 
1696 		ts = inode_get_mtime(inode);
1697 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1698 			invalid |= NFS_INO_INVALID_MTIME;
1699 
1700 		ts = inode_get_ctime(inode);
1701 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1702 			invalid |= NFS_INO_INVALID_CTIME;
1703 
1704 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1705 			cur_size = i_size_read(inode);
1706 			new_isize = nfs_size_to_loff_t(fattr->size);
1707 			if (cur_size != new_isize)
1708 				invalid |= NFS_INO_INVALID_SIZE;
1709 		}
1710 	}
1711 
1712 	/* Have any file permissions changed? */
1713 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1714 		invalid |= NFS_INO_INVALID_MODE;
1715 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1716 		invalid |= NFS_INO_INVALID_OTHER;
1717 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1718 		invalid |= NFS_INO_INVALID_OTHER;
1719 
1720 	/* Has the link count changed? */
1721 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1722 		invalid |= NFS_INO_INVALID_NLINK;
1723 
1724 	ts = inode_get_atime(inode);
1725 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1726 		invalid |= NFS_INO_INVALID_ATIME;
1727 
1728 	if (invalid != 0)
1729 		nfs_set_cache_invalid(inode, invalid);
1730 
1731 	nfsi->read_cache_jiffies = fattr->time_start;
1732 	return 0;
1733 }
1734 
1735 static atomic_long_t nfs_attr_generation_counter;
1736 
nfs_read_attr_generation_counter(void)1737 static unsigned long nfs_read_attr_generation_counter(void)
1738 {
1739 	return atomic_long_read(&nfs_attr_generation_counter);
1740 }
1741 
nfs_inc_attr_generation_counter(void)1742 unsigned long nfs_inc_attr_generation_counter(void)
1743 {
1744 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1745 }
1746 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1747 
nfs_fattr_init(struct nfs_fattr * fattr)1748 void nfs_fattr_init(struct nfs_fattr *fattr)
1749 {
1750 	fattr->valid = 0;
1751 	fattr->time_start = jiffies;
1752 	fattr->gencount = nfs_inc_attr_generation_counter();
1753 	fattr->owner_name = NULL;
1754 	fattr->group_name = NULL;
1755 	fattr->mdsthreshold = NULL;
1756 }
1757 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1758 
1759 /**
1760  * nfs_fattr_set_barrier
1761  * @fattr: attributes
1762  *
1763  * Used to set a barrier after an attribute was updated. This
1764  * barrier ensures that older attributes from RPC calls that may
1765  * have raced with our update cannot clobber these new values.
1766  * Note that you are still responsible for ensuring that other
1767  * operations which change the attribute on the server do not
1768  * collide.
1769  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1770 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1771 {
1772 	fattr->gencount = nfs_inc_attr_generation_counter();
1773 }
1774 
nfs_alloc_fattr(void)1775 struct nfs_fattr *nfs_alloc_fattr(void)
1776 {
1777 	struct nfs_fattr *fattr;
1778 
1779 	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1780 	if (fattr != NULL) {
1781 		nfs_fattr_init(fattr);
1782 		fattr->label = NULL;
1783 	}
1784 	return fattr;
1785 }
1786 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1787 
nfs_alloc_fattr_with_label(struct nfs_server * server)1788 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1789 {
1790 	struct nfs_fattr *fattr = nfs_alloc_fattr();
1791 
1792 	if (!fattr)
1793 		return NULL;
1794 
1795 	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1796 	if (IS_ERR(fattr->label)) {
1797 		kfree(fattr);
1798 		return NULL;
1799 	}
1800 
1801 	return fattr;
1802 }
1803 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1804 
nfs_alloc_fhandle(void)1805 struct nfs_fh *nfs_alloc_fhandle(void)
1806 {
1807 	struct nfs_fh *fh;
1808 
1809 	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1810 	if (fh != NULL)
1811 		fh->size = 0;
1812 	return fh;
1813 }
1814 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1815 
1816 #ifdef NFS_DEBUG
1817 /*
1818  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1819  *                             in the same way that wireshark does
1820  *
1821  * @fh: file handle
1822  *
1823  * For debugging only.
1824  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1825 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1826 {
1827 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1828 	 * not on the result */
1829 	return nfs_fhandle_hash(fh);
1830 }
1831 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1832 
1833 /*
1834  * _nfs_display_fhandle - display an NFS file handle on the console
1835  *
1836  * @fh: file handle to display
1837  * @caption: display caption
1838  *
1839  * For debugging only.
1840  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1841 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1842 {
1843 	unsigned short i;
1844 
1845 	if (fh == NULL || fh->size == 0) {
1846 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1847 		return;
1848 	}
1849 
1850 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1851 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1852 	for (i = 0; i < fh->size; i += 16) {
1853 		__be32 *pos = (__be32 *)&fh->data[i];
1854 
1855 		switch ((fh->size - i - 1) >> 2) {
1856 		case 0:
1857 			printk(KERN_DEFAULT " %08x\n",
1858 				be32_to_cpup(pos));
1859 			break;
1860 		case 1:
1861 			printk(KERN_DEFAULT " %08x %08x\n",
1862 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1863 			break;
1864 		case 2:
1865 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1866 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1867 				be32_to_cpup(pos + 2));
1868 			break;
1869 		default:
1870 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1871 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1872 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1873 		}
1874 	}
1875 }
1876 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1877 #endif
1878 
1879 /**
1880  * nfs_inode_attrs_cmp_generic - compare attributes
1881  * @fattr: attributes
1882  * @inode: pointer to inode
1883  *
1884  * Attempt to divine whether or not an RPC call reply carrying stale
1885  * attributes got scheduled after another call carrying updated ones.
1886  * Note also the check for wraparound of 'attr_gencount'
1887  *
1888  * The function returns '1' if it thinks the attributes in @fattr are
1889  * more recent than the ones cached in @inode. Otherwise it returns
1890  * the value '0'.
1891  */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1892 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1893 				       const struct inode *inode)
1894 {
1895 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1896 
1897 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1898 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1899 }
1900 
1901 /**
1902  * nfs_inode_attrs_cmp_monotonic - compare attributes
1903  * @fattr: attributes
1904  * @inode: pointer to inode
1905  *
1906  * Attempt to divine whether or not an RPC call reply carrying stale
1907  * attributes got scheduled after another call carrying updated ones.
1908  *
1909  * We assume that the server observes monotonic semantics for
1910  * the change attribute, so a larger value means that the attributes in
1911  * @fattr are more recent, in which case the function returns the
1912  * value '1'.
1913  * A return value of '0' indicates no measurable change
1914  * A return value of '-1' means that the attributes in @inode are
1915  * more recent.
1916  */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1917 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1918 					 const struct inode *inode)
1919 {
1920 	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1921 	if (diff > 0)
1922 		return 1;
1923 	return diff == 0 ? 0 : -1;
1924 }
1925 
1926 /**
1927  * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1928  * @fattr: attributes
1929  * @inode: pointer to inode
1930  *
1931  * Attempt to divine whether or not an RPC call reply carrying stale
1932  * attributes got scheduled after another call carrying updated ones.
1933  *
1934  * We assume that the server observes strictly monotonic semantics for
1935  * the change attribute, so a larger value means that the attributes in
1936  * @fattr are more recent, in which case the function returns the
1937  * value '1'.
1938  * A return value of '-1' means that the attributes in @inode are
1939  * more recent or unchanged.
1940  */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1941 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1942 						const struct inode *inode)
1943 {
1944 	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1945 }
1946 
1947 /**
1948  * nfs_inode_attrs_cmp - compare attributes
1949  * @fattr: attributes
1950  * @inode: pointer to inode
1951  *
1952  * This function returns '1' if it thinks the attributes in @fattr are
1953  * more recent than the ones cached in @inode. It returns '-1' if
1954  * the attributes in @inode are more recent than the ones in @fattr,
1955  * and it returns 0 if not sure.
1956  */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1957 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1958 			       const struct inode *inode)
1959 {
1960 	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1961 		return 1;
1962 	switch (NFS_SERVER(inode)->change_attr_type) {
1963 	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1964 		break;
1965 	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1966 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1967 			break;
1968 		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1969 	default:
1970 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1971 			break;
1972 		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1973 	}
1974 	return 0;
1975 }
1976 
1977 /**
1978  * nfs_inode_finish_partial_attr_update - complete a previous inode update
1979  * @fattr: attributes
1980  * @inode: pointer to inode
1981  *
1982  * Returns '1' if the last attribute update left the inode cached
1983  * attributes in a partially unrevalidated state, and @fattr
1984  * matches the change attribute of that partial update.
1985  * Otherwise returns '0'.
1986  */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1987 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1988 						const struct inode *inode)
1989 {
1990 	const unsigned long check_valid =
1991 		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1992 		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1993 		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1994 		NFS_INO_INVALID_NLINK | NFS_INO_INVALID_BTIME;
1995 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1996 	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1997 
1998 	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1999 	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
2000 	    (cache_validity & check_valid) != 0 &&
2001 	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2002 	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
2003 		return 1;
2004 	return 0;
2005 }
2006 
nfs_ooo_merge(struct nfs_inode * nfsi,u64 start,u64 end)2007 static void nfs_ooo_merge(struct nfs_inode *nfsi,
2008 			  u64 start, u64 end)
2009 {
2010 	int i, cnt;
2011 
2012 	if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
2013 		/* No point merging anything */
2014 		return;
2015 
2016 	if (!nfsi->ooo) {
2017 		nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
2018 		if (!nfsi->ooo) {
2019 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2020 			return;
2021 		}
2022 		nfsi->ooo->cnt = 0;
2023 	}
2024 
2025 	/* add this range, merging if possible */
2026 	cnt = nfsi->ooo->cnt;
2027 	for (i = 0; i < cnt; i++) {
2028 		if (end == nfsi->ooo->gap[i].start)
2029 			end = nfsi->ooo->gap[i].end;
2030 		else if (start == nfsi->ooo->gap[i].end)
2031 			start = nfsi->ooo->gap[i].start;
2032 		else
2033 			continue;
2034 		/* Remove 'i' from table and loop to insert the new range */
2035 		cnt -= 1;
2036 		nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
2037 		i = -1;
2038 	}
2039 	if (start != end) {
2040 		if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
2041 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2042 			kfree(nfsi->ooo);
2043 			nfsi->ooo = NULL;
2044 			return;
2045 		}
2046 		nfsi->ooo->gap[cnt].start = start;
2047 		nfsi->ooo->gap[cnt].end = end;
2048 		cnt += 1;
2049 	}
2050 	nfsi->ooo->cnt = cnt;
2051 }
2052 
nfs_ooo_record(struct nfs_inode * nfsi,struct nfs_fattr * fattr)2053 static void nfs_ooo_record(struct nfs_inode *nfsi,
2054 			   struct nfs_fattr *fattr)
2055 {
2056 	/* This reply was out-of-order, so record in the
2057 	 * pre/post change id, possibly cancelling
2058 	 * gaps created when iversion was jumpped forward.
2059 	 */
2060 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
2061 	    (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
2062 		nfs_ooo_merge(nfsi,
2063 			      fattr->change_attr,
2064 			      fattr->pre_change_attr);
2065 }
2066 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)2067 static int nfs_refresh_inode_locked(struct inode *inode,
2068 				    struct nfs_fattr *fattr)
2069 {
2070 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2071 	int ret = 0;
2072 
2073 	trace_nfs_refresh_inode_enter(inode);
2074 
2075 	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
2076 		ret = nfs_update_inode(inode, fattr);
2077 	else {
2078 		nfs_ooo_record(NFS_I(inode), fattr);
2079 
2080 		if (attr_cmp == 0)
2081 			ret = nfs_check_inode_attributes(inode, fattr);
2082 	}
2083 
2084 	trace_nfs_refresh_inode_exit(inode, ret);
2085 	return ret;
2086 }
2087 
2088 /**
2089  * nfs_refresh_inode - try to update the inode attribute cache
2090  * @inode: pointer to inode
2091  * @fattr: updated attributes
2092  *
2093  * Check that an RPC call that returned attributes has not overlapped with
2094  * other recent updates of the inode metadata, then decide whether it is
2095  * safe to do a full update of the inode attributes, or whether just to
2096  * call nfs_check_inode_attributes.
2097  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)2098 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
2099 {
2100 	int status;
2101 
2102 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2103 		return 0;
2104 	spin_lock(&inode->i_lock);
2105 	status = nfs_refresh_inode_locked(inode, fattr);
2106 	spin_unlock(&inode->i_lock);
2107 
2108 	return status;
2109 }
2110 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
2111 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)2112 static int nfs_post_op_update_inode_locked(struct inode *inode,
2113 		struct nfs_fattr *fattr, unsigned int invalid)
2114 {
2115 	if (S_ISDIR(inode->i_mode))
2116 		invalid |= NFS_INO_INVALID_DATA;
2117 	nfs_set_cache_invalid(inode, invalid);
2118 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2119 		return 0;
2120 	return nfs_refresh_inode_locked(inode, fattr);
2121 }
2122 
2123 /**
2124  * nfs_post_op_update_inode - try to update the inode attribute cache
2125  * @inode: pointer to inode
2126  * @fattr: updated attributes
2127  *
2128  * After an operation that has changed the inode metadata, mark the
2129  * attribute cache as being invalid, then try to update it.
2130  *
2131  * NB: if the server didn't return any post op attributes, this
2132  * function will force the retrieval of attributes before the next
2133  * NFS request.  Thus it should be used only for operations that
2134  * are expected to change one or more attributes, to avoid
2135  * unnecessary NFS requests and trips through nfs_update_inode().
2136  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)2137 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2138 {
2139 	int status;
2140 
2141 	spin_lock(&inode->i_lock);
2142 	nfs_fattr_set_barrier(fattr);
2143 	status = nfs_post_op_update_inode_locked(inode, fattr,
2144 			NFS_INO_INVALID_CHANGE
2145 			| NFS_INO_INVALID_CTIME
2146 			| NFS_INO_REVAL_FORCED);
2147 	spin_unlock(&inode->i_lock);
2148 
2149 	return status;
2150 }
2151 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
2152 
2153 /**
2154  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
2155  * @inode: pointer to inode
2156  * @fattr: updated attributes
2157  *
2158  * After an operation that has changed the inode metadata, mark the
2159  * attribute cache as being invalid, then try to update it. Fake up
2160  * weak cache consistency data, if none exist.
2161  *
2162  * This function is mainly designed to be used by the ->write_done() functions.
2163  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)2164 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
2165 {
2166 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2167 	int status;
2168 
2169 	/* Don't do a WCC update if these attributes are already stale */
2170 	if (attr_cmp < 0)
2171 		return 0;
2172 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
2173 		/* Record the pre/post change info before clearing PRECHANGE */
2174 		nfs_ooo_record(NFS_I(inode), fattr);
2175 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
2176 				| NFS_ATTR_FATTR_PRESIZE
2177 				| NFS_ATTR_FATTR_PREMTIME
2178 				| NFS_ATTR_FATTR_PRECTIME);
2179 		goto out_noforce;
2180 	}
2181 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2182 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2183 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2184 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2185 	}
2186 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2187 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2188 		fattr->pre_ctime = inode_get_ctime(inode);
2189 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2190 	}
2191 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2192 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2193 		fattr->pre_mtime = inode_get_mtime(inode);
2194 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2195 	}
2196 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2197 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2198 		fattr->pre_size = i_size_read(inode);
2199 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2200 	}
2201 out_noforce:
2202 	status = nfs_post_op_update_inode_locked(inode, fattr,
2203 			NFS_INO_INVALID_CHANGE
2204 			| NFS_INO_INVALID_CTIME
2205 			| NFS_INO_INVALID_MTIME
2206 			| NFS_INO_INVALID_BLOCKS);
2207 	return status;
2208 }
2209 
2210 /**
2211  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2212  * @inode: pointer to inode
2213  * @fattr: updated attributes
2214  *
2215  * After an operation that has changed the inode metadata, mark the
2216  * attribute cache as being invalid, then try to update it. Fake up
2217  * weak cache consistency data, if none exist.
2218  *
2219  * This function is mainly designed to be used by the ->write_done() functions.
2220  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)2221 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2222 {
2223 	int status;
2224 
2225 	spin_lock(&inode->i_lock);
2226 	nfs_fattr_set_barrier(fattr);
2227 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2228 	spin_unlock(&inode->i_lock);
2229 	return status;
2230 }
2231 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2232 
2233 
2234 /*
2235  * Many nfs protocol calls return the new file attributes after
2236  * an operation.  Here we update the inode to reflect the state
2237  * of the server's inode.
2238  *
2239  * This is a bit tricky because we have to make sure all dirty pages
2240  * have been sent off to the server before calling invalidate_inode_pages.
2241  * To make sure no other process adds more write requests while we try
2242  * our best to flush them, we make them sleep during the attribute refresh.
2243  *
2244  * A very similar scenario holds for the dir cache.
2245  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)2246 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2247 {
2248 	struct nfs_server *server = NFS_SERVER(inode);
2249 	struct nfs_inode *nfsi = NFS_I(inode);
2250 	loff_t cur_isize, new_isize;
2251 	u64 fattr_supported = server->fattr_valid;
2252 	unsigned long invalid = 0;
2253 	unsigned long now = jiffies;
2254 	unsigned long save_cache_validity;
2255 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2256 	bool cache_revalidated = true;
2257 	bool attr_changed = false;
2258 	bool have_delegation;
2259 
2260 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%llx)\n",
2261 			__func__, inode->i_sb->s_id, inode->i_ino,
2262 			nfs_display_fhandle_hash(NFS_FH(inode)),
2263 			icount_read(inode), fattr->valid);
2264 
2265 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2266 		/* Only a mounted-on-fileid? Just exit */
2267 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2268 			return 0;
2269 	/* Has the inode gone and changed behind our back? */
2270 	} else if (nfsi->fileid != fattr->fileid) {
2271 		/* Is this perhaps the mounted-on fileid? */
2272 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2273 		    nfsi->fileid == fattr->mounted_on_fileid)
2274 			return 0;
2275 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2276 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2277 			NFS_SERVER(inode)->nfs_client->cl_hostname,
2278 			inode->i_sb->s_id, (long long)nfsi->fileid,
2279 			(long long)fattr->fileid);
2280 		goto out_err;
2281 	}
2282 
2283 	/*
2284 	 * Make sure the inode's type hasn't changed.
2285 	 */
2286 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2287 		/*
2288 		* Big trouble! The inode has become a different object.
2289 		*/
2290 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2291 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2292 		goto out_err;
2293 	}
2294 
2295 	/* Update the fsid? */
2296 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2297 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2298 			!IS_AUTOMOUNT(inode))
2299 		server->fsid = fattr->fsid;
2300 
2301 	/* Save the delegation state before clearing cache_validity */
2302 	have_delegation = nfs_have_delegated_attributes(inode);
2303 
2304 	/*
2305 	 * Update the read time so we don't revalidate too often.
2306 	 */
2307 	nfsi->read_cache_jiffies = fattr->time_start;
2308 
2309 	/* Fix up any delegated attributes in the struct nfs_fattr */
2310 	nfs_fattr_fixup_delegated(inode, fattr);
2311 
2312 	save_cache_validity = nfsi->cache_validity;
2313 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2314 			| NFS_INO_INVALID_ATIME
2315 			| NFS_INO_REVAL_FORCED
2316 			| NFS_INO_INVALID_BLOCKS);
2317 
2318 	/* Do atomic weak cache consistency updates */
2319 	nfs_wcc_update_inode(inode, fattr);
2320 
2321 	if (pnfs_layoutcommit_outstanding(inode)) {
2322 		nfsi->cache_validity |=
2323 			save_cache_validity &
2324 			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2325 			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2326 			 NFS_INO_INVALID_BLOCKS);
2327 		cache_revalidated = false;
2328 	}
2329 
2330 	/* More cache consistency checks */
2331 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2332 		if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2333 		    nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2334 			/* There is one remaining gap that hasn't been
2335 			 * merged into iversion - do that now.
2336 			 */
2337 			inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2338 			kfree(nfsi->ooo);
2339 			nfsi->ooo = NULL;
2340 		}
2341 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2342 			/* Could it be a race with writeback? */
2343 			if (!(have_writers || have_delegation)) {
2344 				invalid |= NFS_INO_INVALID_DATA
2345 					| NFS_INO_INVALID_ACCESS
2346 					| NFS_INO_INVALID_ACL
2347 					| NFS_INO_INVALID_XATTR;
2348 				/* Force revalidate of all attributes */
2349 				save_cache_validity |= NFS_INO_INVALID_CTIME
2350 					| NFS_INO_INVALID_MTIME
2351 					| NFS_INO_INVALID_SIZE
2352 					| NFS_INO_INVALID_BLOCKS
2353 					| NFS_INO_INVALID_NLINK
2354 					| NFS_INO_INVALID_MODE
2355 					| NFS_INO_INVALID_OTHER
2356 					| NFS_INO_INVALID_BTIME;
2357 				if (S_ISDIR(inode->i_mode))
2358 					nfs_force_lookup_revalidate(inode);
2359 				attr_changed = true;
2360 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2361 						inode->i_sb->s_id,
2362 						inode->i_ino);
2363 			} else if (!have_delegation) {
2364 				nfs_ooo_record(nfsi, fattr);
2365 				nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2366 					      fattr->change_attr);
2367 			}
2368 			inode_set_iversion_raw(inode, fattr->change_attr);
2369 		}
2370 	} else {
2371 		nfsi->cache_validity |=
2372 			save_cache_validity & NFS_INO_INVALID_CHANGE;
2373 		if (!have_delegation ||
2374 		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2375 			cache_revalidated = false;
2376 	}
2377 
2378 	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2379 		inode_set_mtime_to_ts(inode, fattr->mtime);
2380 	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2381 		nfsi->cache_validity |=
2382 			save_cache_validity & NFS_INO_INVALID_MTIME;
2383 
2384 	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2385 		inode_set_ctime_to_ts(inode, fattr->ctime);
2386 	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2387 		nfsi->cache_validity |=
2388 			save_cache_validity & NFS_INO_INVALID_CTIME;
2389 
2390 	if (fattr->valid & NFS_ATTR_FATTR_BTIME)
2391 		nfsi->btime = fattr->btime;
2392 	else if (fattr_supported & NFS_ATTR_FATTR_BTIME)
2393 		nfsi->cache_validity |=
2394 			save_cache_validity & NFS_INO_INVALID_BTIME;
2395 
2396 	/* Check if our cached file size is stale */
2397 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2398 		new_isize = nfs_size_to_loff_t(fattr->size);
2399 		cur_isize = i_size_read(inode);
2400 		if (new_isize != cur_isize && !have_delegation) {
2401 			/* Do we perhaps have any outstanding writes, or has
2402 			 * the file grown beyond our last write? */
2403 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2404 				trace_nfs_size_update(inode, new_isize);
2405 				i_size_write(inode, new_isize);
2406 				if (!have_writers)
2407 					invalid |= NFS_INO_INVALID_DATA;
2408 			}
2409 		}
2410 		if (new_isize == 0 &&
2411 		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2412 				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2413 			fattr->du.nfs3.used = 0;
2414 			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2415 		}
2416 	} else
2417 		nfsi->cache_validity |=
2418 			save_cache_validity & NFS_INO_INVALID_SIZE;
2419 
2420 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2421 		inode_set_atime_to_ts(inode, fattr->atime);
2422 	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2423 		nfsi->cache_validity |=
2424 			save_cache_validity & NFS_INO_INVALID_ATIME;
2425 
2426 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2427 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2428 			umode_t newmode = inode->i_mode & S_IFMT;
2429 			newmode |= fattr->mode & S_IALLUGO;
2430 			inode->i_mode = newmode;
2431 			invalid |= NFS_INO_INVALID_ACCESS
2432 				| NFS_INO_INVALID_ACL;
2433 		}
2434 	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2435 		nfsi->cache_validity |=
2436 			save_cache_validity & NFS_INO_INVALID_MODE;
2437 
2438 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2439 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2440 			invalid |= NFS_INO_INVALID_ACCESS
2441 				| NFS_INO_INVALID_ACL;
2442 			inode->i_uid = fattr->uid;
2443 		}
2444 	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2445 		nfsi->cache_validity |=
2446 			save_cache_validity & NFS_INO_INVALID_OTHER;
2447 
2448 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2449 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2450 			invalid |= NFS_INO_INVALID_ACCESS
2451 				| NFS_INO_INVALID_ACL;
2452 			inode->i_gid = fattr->gid;
2453 		}
2454 	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2455 		nfsi->cache_validity |=
2456 			save_cache_validity & NFS_INO_INVALID_OTHER;
2457 
2458 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2459 		if (inode->i_nlink != fattr->nlink)
2460 			set_nlink(inode, fattr->nlink);
2461 	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2462 		nfsi->cache_validity |=
2463 			save_cache_validity & NFS_INO_INVALID_NLINK;
2464 
2465 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2466 		/*
2467 		 * report the blocks in 512byte units
2468 		 */
2469 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2470 	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2471 		nfsi->cache_validity |=
2472 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2473 
2474 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2475 		inode->i_blocks = fattr->du.nfs2.blocks;
2476 	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2477 		nfsi->cache_validity |=
2478 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2479 
2480 	/* Update attrtimeo value if we're out of the unstable period */
2481 	if (attr_changed) {
2482 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2483 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2484 		nfsi->attrtimeo_timestamp = now;
2485 		/* Set barrier to be more recent than all outstanding updates */
2486 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2487 	} else {
2488 		if (cache_revalidated) {
2489 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2490 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2491 				nfsi->attrtimeo <<= 1;
2492 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2493 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2494 			}
2495 			nfsi->attrtimeo_timestamp = now;
2496 		}
2497 		/* Set the barrier to be more recent than this fattr */
2498 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2499 			nfsi->attr_gencount = fattr->gencount;
2500 	}
2501 
2502 	/* Don't invalidate the data if we were to blame */
2503 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2504 				|| S_ISLNK(inode->i_mode)))
2505 		invalid &= ~NFS_INO_INVALID_DATA;
2506 	nfs_set_cache_invalid(inode, invalid);
2507 
2508 	return 0;
2509  out_err:
2510 	/*
2511 	 * No need to worry about unhashing the dentry, as the
2512 	 * lookup validation will know that the inode is bad.
2513 	 * (But we fall through to invalidate the caches.)
2514 	 */
2515 	nfs_set_inode_stale_locked(inode);
2516 	return -ESTALE;
2517 }
2518 
nfs_alloc_inode(struct super_block * sb)2519 struct inode *nfs_alloc_inode(struct super_block *sb)
2520 {
2521 	struct nfs_inode *nfsi;
2522 	nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2523 	if (!nfsi)
2524 		return NULL;
2525 	nfsi->flags = 0UL;
2526 	nfsi->cache_validity = 0UL;
2527 	nfsi->ooo = NULL;
2528 #if IS_ENABLED(CONFIG_NFS_V4)
2529 	nfsi->nfs4_acl = NULL;
2530 #endif /* CONFIG_NFS_V4 */
2531 #ifdef CONFIG_NFS_V4_2
2532 	nfsi->xattr_cache = NULL;
2533 #endif
2534 	nfs_netfs_inode_init(nfsi);
2535 
2536 	return &nfsi->vfs_inode;
2537 }
2538 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2539 
nfs_free_inode(struct inode * inode)2540 void nfs_free_inode(struct inode *inode)
2541 {
2542 	kfree(NFS_I(inode)->ooo);
2543 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2544 }
2545 EXPORT_SYMBOL_GPL(nfs_free_inode);
2546 
nfs4_init_once(struct nfs_inode * nfsi)2547 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2548 {
2549 #if IS_ENABLED(CONFIG_NFS_V4)
2550 	INIT_LIST_HEAD(&nfsi->open_states);
2551 	nfsi->delegation = NULL;
2552 	init_rwsem(&nfsi->rwsem);
2553 	nfsi->layout = NULL;
2554 #endif
2555 }
2556 
init_once(void * foo)2557 static void init_once(void *foo)
2558 {
2559 	struct nfs_inode *nfsi = foo;
2560 
2561 	inode_init_once(&nfsi->vfs_inode);
2562 	INIT_LIST_HEAD(&nfsi->open_files);
2563 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2564 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2565 	nfs4_init_once(nfsi);
2566 }
2567 
nfs_init_inodecache(void)2568 static int __init nfs_init_inodecache(void)
2569 {
2570 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2571 					     sizeof(struct nfs_inode),
2572 					     0, (SLAB_RECLAIM_ACCOUNT|
2573 						SLAB_ACCOUNT),
2574 					     init_once);
2575 	if (nfs_inode_cachep == NULL)
2576 		return -ENOMEM;
2577 
2578 	return 0;
2579 }
2580 
nfs_destroy_inodecache(void)2581 static void nfs_destroy_inodecache(void)
2582 {
2583 	/*
2584 	 * Make sure all delayed rcu free inodes are flushed before we
2585 	 * destroy cache.
2586 	 */
2587 	rcu_barrier();
2588 	kmem_cache_destroy(nfs_inode_cachep);
2589 }
2590 
2591 struct workqueue_struct *nfslocaliod_workqueue;
2592 struct workqueue_struct *nfsiod_workqueue;
2593 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2594 
2595 /*
2596  * Destroy the nfsiod workqueues
2597  */
nfsiod_stop(void)2598 static void nfsiod_stop(void)
2599 {
2600 	struct workqueue_struct *wq;
2601 
2602 	wq = nfsiod_workqueue;
2603 	if (wq != NULL) {
2604 		nfsiod_workqueue = NULL;
2605 		destroy_workqueue(wq);
2606 	}
2607 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2608 	wq = nfslocaliod_workqueue;
2609 	if (wq != NULL) {
2610 		nfslocaliod_workqueue = NULL;
2611 		destroy_workqueue(wq);
2612 	}
2613 #endif /* CONFIG_NFS_LOCALIO */
2614 }
2615 
2616 /*
2617  * Start the nfsiod workqueues
2618  */
nfsiod_start(void)2619 static int nfsiod_start(void)
2620 {
2621 	dprintk("RPC:       creating workqueue nfsiod\n");
2622 	nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2623 	if (nfsiod_workqueue == NULL)
2624 		return -ENOMEM;
2625 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2626 	/*
2627 	 * localio writes need to use a normal (non-memreclaim) workqueue.
2628 	 * When we start getting low on space, XFS goes and calls flush_work() on
2629 	 * a non-memreclaim work queue, which causes a priority inversion problem.
2630 	 */
2631 	dprintk("RPC:       creating workqueue nfslocaliod\n");
2632 	nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0);
2633 	if (unlikely(nfslocaliod_workqueue == NULL)) {
2634 		nfsiod_stop();
2635 		return -ENOMEM;
2636 	}
2637 #endif /* CONFIG_NFS_LOCALIO */
2638 	return 0;
2639 }
2640 
2641 unsigned int nfs_net_id;
2642 EXPORT_SYMBOL_GPL(nfs_net_id);
2643 
nfs_net_init(struct net * net)2644 static int nfs_net_init(struct net *net)
2645 {
2646 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2647 	int err;
2648 
2649 	nfs_clients_init(net);
2650 
2651 	if (!rpc_proc_register(net, &nn->rpcstats)) {
2652 		err = -ENOMEM;
2653 		goto err_proc_rpc;
2654 	}
2655 
2656 	err = nfs_fs_proc_net_init(net);
2657 	if (err)
2658 		goto err_proc_nfs;
2659 
2660 	return 0;
2661 
2662 err_proc_nfs:
2663 	rpc_proc_unregister(net, "nfs");
2664 err_proc_rpc:
2665 	nfs_clients_exit(net);
2666 	return err;
2667 }
2668 
nfs_net_exit(struct net * net)2669 static void nfs_net_exit(struct net *net)
2670 {
2671 	rpc_proc_unregister(net, "nfs");
2672 	nfs_fs_proc_net_exit(net);
2673 	nfs_clients_exit(net);
2674 }
2675 
2676 static struct pernet_operations nfs_net_ops = {
2677 	.init = nfs_net_init,
2678 	.exit = nfs_net_exit,
2679 	.id   = &nfs_net_id,
2680 	.size = sizeof(struct nfs_net),
2681 };
2682 
2683 #ifdef CONFIG_KEYS
2684 static struct key *nfs_keyring;
2685 
nfs_init_keyring(void)2686 static int __init nfs_init_keyring(void)
2687 {
2688 	nfs_keyring = keyring_alloc(".nfs",
2689 			     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
2690 			     current_cred(),
2691 			     (KEY_POS_ALL & ~KEY_POS_SETATTR) |
2692 			     (KEY_USR_ALL & ~KEY_USR_SETATTR),
2693 			     KEY_ALLOC_NOT_IN_QUOTA, NULL, NULL);
2694 	return PTR_ERR_OR_ZERO(nfs_keyring);
2695 }
2696 
nfs_exit_keyring(void)2697 static void nfs_exit_keyring(void)
2698 {
2699 	key_put(nfs_keyring);
2700 }
2701 #else
nfs_init_keyring(void)2702 static inline int nfs_init_keyring(void)
2703 {
2704 	return 0;
2705 }
2706 
nfs_exit_keyring(void)2707 static inline void nfs_exit_keyring(void)
2708 {
2709 }
2710 #endif /* CONFIG_KEYS */
2711 
2712 /*
2713  * Initialize NFS
2714  */
init_nfs_fs(void)2715 static int __init init_nfs_fs(void)
2716 {
2717 	int err;
2718 
2719 	err = nfs_init_keyring();
2720 	if (err)
2721 		return err;
2722 
2723 	err = nfs_sysfs_init();
2724 	if (err < 0)
2725 		goto out10;
2726 
2727 	err = register_pernet_subsys(&nfs_net_ops);
2728 	if (err < 0)
2729 		goto out9;
2730 
2731 	err = nfsiod_start();
2732 	if (err)
2733 		goto out7;
2734 
2735 	err = nfs_fs_proc_init();
2736 	if (err)
2737 		goto out6;
2738 
2739 	err = nfs_init_nfspagecache();
2740 	if (err)
2741 		goto out5;
2742 
2743 	err = nfs_init_inodecache();
2744 	if (err)
2745 		goto out4;
2746 
2747 	err = nfs_init_readpagecache();
2748 	if (err)
2749 		goto out3;
2750 
2751 	err = nfs_init_writepagecache();
2752 	if (err)
2753 		goto out2;
2754 
2755 	err = nfs_init_directcache();
2756 	if (err)
2757 		goto out1;
2758 
2759 	err = register_nfs_fs();
2760 	if (err)
2761 		goto out0;
2762 
2763 	return 0;
2764 out0:
2765 	nfs_destroy_directcache();
2766 out1:
2767 	nfs_destroy_writepagecache();
2768 out2:
2769 	nfs_destroy_readpagecache();
2770 out3:
2771 	nfs_destroy_inodecache();
2772 out4:
2773 	nfs_destroy_nfspagecache();
2774 out5:
2775 	nfs_fs_proc_exit();
2776 out6:
2777 	nfsiod_stop();
2778 out7:
2779 	unregister_pernet_subsys(&nfs_net_ops);
2780 out9:
2781 	nfs_sysfs_exit();
2782 out10:
2783 	nfs_exit_keyring();
2784 	return err;
2785 }
2786 
exit_nfs_fs(void)2787 static void __exit exit_nfs_fs(void)
2788 {
2789 	nfs_destroy_directcache();
2790 	nfs_destroy_writepagecache();
2791 	nfs_destroy_readpagecache();
2792 	nfs_destroy_inodecache();
2793 	nfs_destroy_nfspagecache();
2794 	unregister_pernet_subsys(&nfs_net_ops);
2795 	unregister_nfs_fs();
2796 	nfs_fs_proc_exit();
2797 	nfsiod_stop();
2798 	nfs_sysfs_exit();
2799 	nfs_exit_keyring();
2800 }
2801 
2802 /* Not quite true; I just maintain it */
2803 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2804 MODULE_DESCRIPTION("NFS client support");
2805 MODULE_LICENSE("GPL");
2806 module_param(enable_ino64, bool, 0644);
2807 
2808 module_init(init_nfs_fs)
2809 module_exit(exit_nfs_fs)
2810