• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/nfs/inode.c
4  *
5  *  Copyright (C) 1992  Rick Sladkey
6  *
7  *  nfs inode and superblock handling functions
8  *
9  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
11  *
12  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13  *  J.S.Peatfield@damtp.cam.ac.uk
14  *
15  */
16 
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
44 
45 #include "nfs4_fs.h"
46 #include "callback.h"
47 #include "delegation.h"
48 #include "iostat.h"
49 #include "internal.h"
50 #include "fscache.h"
51 #include "pnfs.h"
52 #include "nfs.h"
53 #include "netns.h"
54 #include "sysfs.h"
55 
56 #include "nfstrace.h"
57 
58 #define NFSDBG_FACILITY		NFSDBG_VFS
59 
60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
61 
62 /* Default is to see 64-bit inode numbers */
63 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64 
65 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 
67 static struct kmem_cache * nfs_inode_cachep;
68 
69 static inline unsigned long
nfs_fattr_to_ino_t(struct nfs_fattr * fattr)70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 {
72 	return nfs_fileid_to_ino_t(fattr->fileid);
73 }
74 
nfs_wait_killable(int mode)75 static int nfs_wait_killable(int mode)
76 {
77 	freezable_schedule_unsafe();
78 	if (signal_pending_state(mode, current))
79 		return -ERESTARTSYS;
80 	return 0;
81 }
82 
nfs_wait_bit_killable(struct wait_bit_key * key,int mode)83 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
84 {
85 	return nfs_wait_killable(mode);
86 }
87 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88 
89 /**
90  * nfs_compat_user_ino64 - returns the user-visible inode number
91  * @fileid: 64-bit fileid
92  *
93  * This function returns a 32-bit inode number if the boot parameter
94  * nfs.enable_ino64 is zero.
95  */
nfs_compat_user_ino64(u64 fileid)96 u64 nfs_compat_user_ino64(u64 fileid)
97 {
98 #ifdef CONFIG_COMPAT
99 	compat_ulong_t ino;
100 #else
101 	unsigned long ino;
102 #endif
103 
104 	if (enable_ino64)
105 		return fileid;
106 	ino = fileid;
107 	if (sizeof(ino) < sizeof(fileid))
108 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
109 	return ino;
110 }
111 
nfs_drop_inode(struct inode * inode)112 int nfs_drop_inode(struct inode *inode)
113 {
114 	return NFS_STALE(inode) || generic_drop_inode(inode);
115 }
116 EXPORT_SYMBOL_GPL(nfs_drop_inode);
117 
nfs_clear_inode(struct inode * inode)118 void nfs_clear_inode(struct inode *inode)
119 {
120 	/*
121 	 * The following should never happen...
122 	 */
123 	WARN_ON_ONCE(nfs_have_writebacks(inode));
124 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
125 	nfs_zap_acl_cache(inode);
126 	nfs_access_zap_cache(inode);
127 	nfs_fscache_clear_inode(inode);
128 }
129 EXPORT_SYMBOL_GPL(nfs_clear_inode);
130 
nfs_evict_inode(struct inode * inode)131 void nfs_evict_inode(struct inode *inode)
132 {
133 	truncate_inode_pages_final(&inode->i_data);
134 	clear_inode(inode);
135 	nfs_clear_inode(inode);
136 }
137 
nfs_sync_inode(struct inode * inode)138 int nfs_sync_inode(struct inode *inode)
139 {
140 	inode_dio_wait(inode);
141 	return nfs_wb_all(inode);
142 }
143 EXPORT_SYMBOL_GPL(nfs_sync_inode);
144 
145 /**
146  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147  * @mapping: pointer to struct address_space
148  */
nfs_sync_mapping(struct address_space * mapping)149 int nfs_sync_mapping(struct address_space *mapping)
150 {
151 	int ret = 0;
152 
153 	if (mapping->nrpages != 0) {
154 		unmap_mapping_range(mapping, 0, 0, 0);
155 		ret = nfs_wb_all(mapping->host);
156 	}
157 	return ret;
158 }
159 
nfs_attribute_timeout(struct inode * inode)160 static int nfs_attribute_timeout(struct inode *inode)
161 {
162 	struct nfs_inode *nfsi = NFS_I(inode);
163 
164 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
165 }
166 
nfs_check_cache_invalid_delegated(struct inode * inode,unsigned long flags)167 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
168 {
169 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
170 
171 	/* Special case for the pagecache or access cache */
172 	if (flags == NFS_INO_REVAL_PAGECACHE &&
173 	    !(cache_validity & NFS_INO_REVAL_FORCED))
174 		return false;
175 	return (cache_validity & flags) != 0;
176 }
177 
nfs_check_cache_invalid_not_delegated(struct inode * inode,unsigned long flags)178 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
179 {
180 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
181 
182 	if ((cache_validity & flags) != 0)
183 		return true;
184 	if (nfs_attribute_timeout(inode))
185 		return true;
186 	return false;
187 }
188 
nfs_check_cache_invalid(struct inode * inode,unsigned long flags)189 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
190 {
191 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
192 		return nfs_check_cache_invalid_delegated(inode, flags);
193 
194 	return nfs_check_cache_invalid_not_delegated(inode, flags);
195 }
196 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
197 
198 #ifdef CONFIG_NFS_V4_2
nfs_has_xattr_cache(const struct nfs_inode * nfsi)199 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
200 {
201 	return nfsi->xattr_cache != NULL;
202 }
203 #else
nfs_has_xattr_cache(const struct nfs_inode * nfsi)204 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
205 {
206 	return false;
207 }
208 #endif
209 
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)210 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
211 {
212 	struct nfs_inode *nfsi = NFS_I(inode);
213 	bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
214 
215 	if (have_delegation) {
216 		if (!(flags & NFS_INO_REVAL_FORCED))
217 			flags &= ~NFS_INO_INVALID_OTHER;
218 		flags &= ~(NFS_INO_INVALID_CHANGE
219 				| NFS_INO_INVALID_SIZE
220 				| NFS_INO_REVAL_PAGECACHE
221 				| NFS_INO_INVALID_XATTR);
222 	} else if (flags & NFS_INO_REVAL_PAGECACHE)
223 		flags |= NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE;
224 
225 	if (!nfs_has_xattr_cache(nfsi))
226 		flags &= ~NFS_INO_INVALID_XATTR;
227 	if (inode->i_mapping->nrpages == 0)
228 		flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
229 	nfsi->cache_validity |= flags;
230 	if (flags & NFS_INO_INVALID_DATA)
231 		nfs_fscache_invalidate(inode);
232 }
233 
234 /*
235  * Invalidate the local caches
236  */
nfs_zap_caches_locked(struct inode * inode)237 static void nfs_zap_caches_locked(struct inode *inode)
238 {
239 	struct nfs_inode *nfsi = NFS_I(inode);
240 	int mode = inode->i_mode;
241 
242 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
243 
244 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
245 	nfsi->attrtimeo_timestamp = jiffies;
246 
247 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
248 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
249 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
250 					| NFS_INO_INVALID_DATA
251 					| NFS_INO_INVALID_ACCESS
252 					| NFS_INO_INVALID_ACL
253 					| NFS_INO_INVALID_XATTR
254 					| NFS_INO_REVAL_PAGECACHE);
255 	} else
256 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
257 					| NFS_INO_INVALID_ACCESS
258 					| NFS_INO_INVALID_ACL
259 					| NFS_INO_INVALID_XATTR
260 					| NFS_INO_REVAL_PAGECACHE);
261 	nfs_zap_label_cache_locked(nfsi);
262 }
263 
nfs_zap_caches(struct inode * inode)264 void nfs_zap_caches(struct inode *inode)
265 {
266 	spin_lock(&inode->i_lock);
267 	nfs_zap_caches_locked(inode);
268 	spin_unlock(&inode->i_lock);
269 }
270 
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)271 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
272 {
273 	if (mapping->nrpages != 0) {
274 		spin_lock(&inode->i_lock);
275 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
276 		spin_unlock(&inode->i_lock);
277 	}
278 }
279 
nfs_zap_acl_cache(struct inode * inode)280 void nfs_zap_acl_cache(struct inode *inode)
281 {
282 	void (*clear_acl_cache)(struct inode *);
283 
284 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
285 	if (clear_acl_cache != NULL)
286 		clear_acl_cache(inode);
287 	spin_lock(&inode->i_lock);
288 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
289 	spin_unlock(&inode->i_lock);
290 }
291 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
292 
nfs_invalidate_atime(struct inode * inode)293 void nfs_invalidate_atime(struct inode *inode)
294 {
295 	spin_lock(&inode->i_lock);
296 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
297 	spin_unlock(&inode->i_lock);
298 }
299 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
300 
301 /*
302  * Invalidate, but do not unhash, the inode.
303  * NB: must be called with inode->i_lock held!
304  */
nfs_set_inode_stale_locked(struct inode * inode)305 static void nfs_set_inode_stale_locked(struct inode *inode)
306 {
307 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
308 	nfs_zap_caches_locked(inode);
309 	trace_nfs_set_inode_stale(inode);
310 }
311 
nfs_set_inode_stale(struct inode * inode)312 void nfs_set_inode_stale(struct inode *inode)
313 {
314 	spin_lock(&inode->i_lock);
315 	nfs_set_inode_stale_locked(inode);
316 	spin_unlock(&inode->i_lock);
317 }
318 
319 struct nfs_find_desc {
320 	struct nfs_fh		*fh;
321 	struct nfs_fattr	*fattr;
322 };
323 
324 /*
325  * In NFSv3 we can have 64bit inode numbers. In order to support
326  * this, and re-exported directories (also seen in NFSv2)
327  * we are forced to allow 2 different inodes to have the same
328  * i_ino.
329  */
330 static int
nfs_find_actor(struct inode * inode,void * opaque)331 nfs_find_actor(struct inode *inode, void *opaque)
332 {
333 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
334 	struct nfs_fh		*fh = desc->fh;
335 	struct nfs_fattr	*fattr = desc->fattr;
336 
337 	if (NFS_FILEID(inode) != fattr->fileid)
338 		return 0;
339 	if (inode_wrong_type(inode, fattr->mode))
340 		return 0;
341 	if (nfs_compare_fh(NFS_FH(inode), fh))
342 		return 0;
343 	if (is_bad_inode(inode) || NFS_STALE(inode))
344 		return 0;
345 	return 1;
346 }
347 
348 static int
nfs_init_locked(struct inode * inode,void * opaque)349 nfs_init_locked(struct inode *inode, void *opaque)
350 {
351 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
352 	struct nfs_fattr	*fattr = desc->fattr;
353 
354 	set_nfs_fileid(inode, fattr->fileid);
355 	inode->i_mode = fattr->mode;
356 	nfs_copy_fh(NFS_FH(inode), desc->fh);
357 	return 0;
358 }
359 
360 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)361 static void nfs_clear_label_invalid(struct inode *inode)
362 {
363 	spin_lock(&inode->i_lock);
364 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
365 	spin_unlock(&inode->i_lock);
366 }
367 
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)368 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
369 					struct nfs4_label *label)
370 {
371 	int error;
372 
373 	if (label == NULL)
374 		return;
375 
376 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
377 		error = security_inode_notifysecctx(inode, label->label,
378 				label->len);
379 		if (error)
380 			printk(KERN_ERR "%s() %s %d "
381 					"security_inode_notifysecctx() %d\n",
382 					__func__,
383 					(char *)label->label,
384 					label->len, error);
385 		nfs_clear_label_invalid(inode);
386 	}
387 }
388 
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)389 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
390 {
391 	struct nfs4_label *label = NULL;
392 	int minor_version = server->nfs_client->cl_minorversion;
393 
394 	if (minor_version < 2)
395 		return label;
396 
397 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
398 		return label;
399 
400 	label = kzalloc(sizeof(struct nfs4_label), flags);
401 	if (label == NULL)
402 		return ERR_PTR(-ENOMEM);
403 
404 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
405 	if (label->label == NULL) {
406 		kfree(label);
407 		return ERR_PTR(-ENOMEM);
408 	}
409 	label->len = NFS4_MAXLABELLEN;
410 
411 	return label;
412 }
413 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
414 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr,struct nfs4_label * label)415 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
416 					struct nfs4_label *label)
417 {
418 }
419 #endif
420 EXPORT_SYMBOL_GPL(nfs_setsecurity);
421 
422 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
423 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)424 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
425 {
426 	struct nfs_find_desc desc = {
427 		.fh	= fh,
428 		.fattr	= fattr,
429 	};
430 	struct inode *inode;
431 	unsigned long hash;
432 
433 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
434 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
435 		return NULL;
436 
437 	hash = nfs_fattr_to_ino_t(fattr);
438 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
439 
440 	dprintk("%s: returning %p\n", __func__, inode);
441 	return inode;
442 }
443 
444 /*
445  * This is our front-end to iget that looks up inodes by file handle
446  * instead of inode number.
447  */
448 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr,struct nfs4_label * label)449 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
450 {
451 	struct nfs_find_desc desc = {
452 		.fh	= fh,
453 		.fattr	= fattr
454 	};
455 	struct inode *inode = ERR_PTR(-ENOENT);
456 	unsigned long hash;
457 
458 	nfs_attr_check_mountpoint(sb, fattr);
459 
460 	if (nfs_attr_use_mounted_on_fileid(fattr))
461 		fattr->fileid = fattr->mounted_on_fileid;
462 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
463 		goto out_no_inode;
464 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
465 		goto out_no_inode;
466 
467 	hash = nfs_fattr_to_ino_t(fattr);
468 
469 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
470 	if (inode == NULL) {
471 		inode = ERR_PTR(-ENOMEM);
472 		goto out_no_inode;
473 	}
474 
475 	if (inode->i_state & I_NEW) {
476 		struct nfs_inode *nfsi = NFS_I(inode);
477 		unsigned long now = jiffies;
478 
479 		/* We set i_ino for the few things that still rely on it,
480 		 * such as stat(2) */
481 		inode->i_ino = hash;
482 
483 		/* We can't support update_atime(), since the server will reset it */
484 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
485 		inode->i_mode = fattr->mode;
486 		nfsi->cache_validity = 0;
487 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
488 				&& nfs_server_capable(inode, NFS_CAP_MODE))
489 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
490 		/* Why so? Because we want revalidate for devices/FIFOs, and
491 		 * that's precisely what we have in nfs_file_inode_operations.
492 		 */
493 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
494 		if (S_ISREG(inode->i_mode)) {
495 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
496 			inode->i_data.a_ops = &nfs_file_aops;
497 		} else if (S_ISDIR(inode->i_mode)) {
498 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
499 			inode->i_fop = &nfs_dir_operations;
500 			inode->i_data.a_ops = &nfs_dir_aops;
501 			/* Deal with crossing mountpoints */
502 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
503 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
504 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
505 					inode->i_op = &nfs_referral_inode_operations;
506 				else
507 					inode->i_op = &nfs_mountpoint_inode_operations;
508 				inode->i_fop = NULL;
509 				inode->i_flags |= S_AUTOMOUNT;
510 			}
511 		} else if (S_ISLNK(inode->i_mode)) {
512 			inode->i_op = &nfs_symlink_inode_operations;
513 			inode_nohighmem(inode);
514 		} else
515 			init_special_inode(inode, inode->i_mode, fattr->rdev);
516 
517 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
518 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
519 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
520 		inode_set_iversion_raw(inode, 0);
521 		inode->i_size = 0;
522 		clear_nlink(inode);
523 		inode->i_uid = make_kuid(&init_user_ns, -2);
524 		inode->i_gid = make_kgid(&init_user_ns, -2);
525 		inode->i_blocks = 0;
526 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
527 		nfsi->write_io = 0;
528 		nfsi->read_io = 0;
529 
530 		nfsi->read_cache_jiffies = fattr->time_start;
531 		nfsi->attr_gencount = fattr->gencount;
532 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
533 			inode->i_atime = fattr->atime;
534 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
535 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
536 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
537 			inode->i_mtime = fattr->mtime;
538 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
539 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
540 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
541 			inode->i_ctime = fattr->ctime;
542 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
543 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
544 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
545 			inode_set_iversion_raw(inode, fattr->change_attr);
546 		else
547 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
548 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
549 			inode->i_size = nfs_size_to_loff_t(fattr->size);
550 		else
551 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
552 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
553 			set_nlink(inode, fattr->nlink);
554 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
555 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
556 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
557 			inode->i_uid = fattr->uid;
558 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
559 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
560 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
561 			inode->i_gid = fattr->gid;
562 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
563 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
564 		if (nfs_server_capable(inode, NFS_CAP_XATTR))
565 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
566 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
567 			inode->i_blocks = fattr->du.nfs2.blocks;
568 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
569 			/*
570 			 * report the blocks in 512byte units
571 			 */
572 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
573 		}
574 
575 		if (nfsi->cache_validity != 0)
576 			nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
577 
578 		nfs_setsecurity(inode, fattr, label);
579 
580 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
581 		nfsi->attrtimeo_timestamp = now;
582 		nfsi->access_cache = RB_ROOT;
583 
584 		nfs_fscache_init_inode(inode);
585 
586 		unlock_new_inode(inode);
587 	} else {
588 		int err = nfs_refresh_inode(inode, fattr);
589 		if (err < 0) {
590 			iput(inode);
591 			inode = ERR_PTR(err);
592 			goto out_no_inode;
593 		}
594 	}
595 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
596 		inode->i_sb->s_id,
597 		(unsigned long long)NFS_FILEID(inode),
598 		nfs_display_fhandle_hash(fh),
599 		atomic_read(&inode->i_count));
600 
601 out:
602 	return inode;
603 
604 out_no_inode:
605 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
606 	goto out;
607 }
608 EXPORT_SYMBOL_GPL(nfs_fhget);
609 
610 #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)
611 
612 int
nfs_setattr(struct dentry * dentry,struct iattr * attr)613 nfs_setattr(struct dentry *dentry, struct iattr *attr)
614 {
615 	struct inode *inode = d_inode(dentry);
616 	struct nfs_fattr *fattr;
617 	int error = 0;
618 
619 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
620 
621 	/* skip mode change if it's just for clearing setuid/setgid */
622 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
623 		attr->ia_valid &= ~ATTR_MODE;
624 
625 	if (attr->ia_valid & ATTR_SIZE) {
626 		BUG_ON(!S_ISREG(inode->i_mode));
627 
628 		error = inode_newsize_ok(inode, attr->ia_size);
629 		if (error)
630 			return error;
631 
632 		if (attr->ia_size == i_size_read(inode))
633 			attr->ia_valid &= ~ATTR_SIZE;
634 	}
635 
636 	/* Optimization: if the end result is no change, don't RPC */
637 	attr->ia_valid &= NFS_VALID_ATTRS;
638 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
639 		return 0;
640 
641 	trace_nfs_setattr_enter(inode);
642 
643 	/* Write all dirty data */
644 	if (S_ISREG(inode->i_mode))
645 		nfs_sync_inode(inode);
646 
647 	fattr = nfs_alloc_fattr();
648 	if (fattr == NULL) {
649 		error = -ENOMEM;
650 		goto out;
651 	}
652 
653 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
654 	if (error == 0)
655 		error = nfs_refresh_inode(inode, fattr);
656 	nfs_free_fattr(fattr);
657 out:
658 	trace_nfs_setattr_exit(inode, error);
659 	return error;
660 }
661 EXPORT_SYMBOL_GPL(nfs_setattr);
662 
663 /**
664  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
665  * @inode: inode of the file used
666  * @offset: file offset to start truncating
667  *
668  * This is a copy of the common vmtruncate, but with the locking
669  * corrected to take into account the fact that NFS requires
670  * inode->i_size to be updated under the inode->i_lock.
671  * Note: must be called with inode->i_lock held!
672  */
nfs_vmtruncate(struct inode * inode,loff_t offset)673 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
674 {
675 	int err;
676 
677 	err = inode_newsize_ok(inode, offset);
678 	if (err)
679 		goto out;
680 
681 	i_size_write(inode, offset);
682 	/* Optimisation */
683 	if (offset == 0)
684 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
685 				NFS_INO_DATA_INVAL_DEFER);
686 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
687 
688 	spin_unlock(&inode->i_lock);
689 	truncate_pagecache(inode, offset);
690 	spin_lock(&inode->i_lock);
691 out:
692 	return err;
693 }
694 
695 /**
696  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
697  * @inode: pointer to struct inode
698  * @attr: pointer to struct iattr
699  * @fattr: pointer to struct nfs_fattr
700  *
701  * Note: we do this in the *proc.c in order to ensure that
702  *       it works for things like exclusive creates too.
703  */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)704 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
705 		struct nfs_fattr *fattr)
706 {
707 	/* Barrier: bump the attribute generation count. */
708 	nfs_fattr_set_barrier(fattr);
709 
710 	spin_lock(&inode->i_lock);
711 	NFS_I(inode)->attr_gencount = fattr->gencount;
712 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
713 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
714 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
715 		nfs_vmtruncate(inode, attr->ia_size);
716 	}
717 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
718 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
719 		if ((attr->ia_valid & ATTR_MODE) != 0) {
720 			int mode = attr->ia_mode & S_IALLUGO;
721 			mode |= inode->i_mode & ~S_IALLUGO;
722 			inode->i_mode = mode;
723 		}
724 		if ((attr->ia_valid & ATTR_UID) != 0)
725 			inode->i_uid = attr->ia_uid;
726 		if ((attr->ia_valid & ATTR_GID) != 0)
727 			inode->i_gid = attr->ia_gid;
728 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
729 			inode->i_ctime = fattr->ctime;
730 		else
731 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
732 					| NFS_INO_INVALID_CTIME);
733 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
734 				| NFS_INO_INVALID_ACL);
735 	}
736 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
737 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
738 				| NFS_INO_INVALID_CTIME);
739 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
740 			inode->i_atime = fattr->atime;
741 		else if (attr->ia_valid & ATTR_ATIME_SET)
742 			inode->i_atime = attr->ia_atime;
743 		else
744 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
745 
746 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
747 			inode->i_ctime = fattr->ctime;
748 		else
749 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
750 					| NFS_INO_INVALID_CTIME);
751 	}
752 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
753 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
754 				| NFS_INO_INVALID_CTIME);
755 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
756 			inode->i_mtime = fattr->mtime;
757 		else if (attr->ia_valid & ATTR_MTIME_SET)
758 			inode->i_mtime = attr->ia_mtime;
759 		else
760 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
761 
762 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
763 			inode->i_ctime = fattr->ctime;
764 		else
765 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
766 					| NFS_INO_INVALID_CTIME);
767 	}
768 	if (fattr->valid)
769 		nfs_update_inode(inode, fattr);
770 	spin_unlock(&inode->i_lock);
771 }
772 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
773 
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)774 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
775 {
776 	struct dentry *parent;
777 
778 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
779 		return;
780 	parent = dget_parent(dentry);
781 	nfs_force_use_readdirplus(d_inode(parent));
782 	dput(parent);
783 }
784 
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)785 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
786 {
787 	struct dentry *parent;
788 
789 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
790 		return;
791 	parent = dget_parent(dentry);
792 	nfs_advise_use_readdirplus(d_inode(parent));
793 	dput(parent);
794 }
795 
nfs_need_revalidate_inode(struct inode * inode)796 static bool nfs_need_revalidate_inode(struct inode *inode)
797 {
798 	if (NFS_I(inode)->cache_validity &
799 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
800 		return true;
801 	if (nfs_attribute_cache_expired(inode))
802 		return true;
803 	return false;
804 }
805 
nfs_getattr(const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)806 int nfs_getattr(const struct path *path, struct kstat *stat,
807 		u32 request_mask, unsigned int query_flags)
808 {
809 	struct inode *inode = d_inode(path->dentry);
810 	struct nfs_server *server = NFS_SERVER(inode);
811 	unsigned long cache_validity;
812 	int err = 0;
813 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
814 	bool do_update = false;
815 
816 	trace_nfs_getattr_enter(inode);
817 
818 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
819 		nfs_readdirplus_parent_cache_hit(path->dentry);
820 		goto out_no_update;
821 	}
822 
823 	/* Flush out writes to the server in order to update c/mtime.  */
824 	if ((request_mask & (STATX_CTIME | STATX_MTIME)) &&
825 	    S_ISREG(inode->i_mode))
826 		filemap_write_and_wait(inode->i_mapping);
827 
828 	/*
829 	 * We may force a getattr if the user cares about atime.
830 	 *
831 	 * Note that we only have to check the vfsmount flags here:
832 	 *  - NFS always sets S_NOATIME by so checking it would give a
833 	 *    bogus result
834 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
835 	 *    no point in checking those.
836 	 */
837 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
838 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
839 		request_mask &= ~STATX_ATIME;
840 
841 	/* Is the user requesting attributes that might need revalidation? */
842 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
843 					STATX_MTIME|STATX_UID|STATX_GID|
844 					STATX_SIZE|STATX_BLOCKS)))
845 		goto out_no_revalidate;
846 
847 	/* Check whether the cached attributes are stale */
848 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
849 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
850 	do_update |= cache_validity &
851 		(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
852 	if (request_mask & STATX_ATIME)
853 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
854 	if (request_mask & (STATX_CTIME|STATX_MTIME))
855 		do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
856 	if (request_mask & STATX_BLOCKS)
857 		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
858 	if (do_update) {
859 		/* Update the attribute cache */
860 		if (!(server->flags & NFS_MOUNT_NOAC))
861 			nfs_readdirplus_parent_cache_miss(path->dentry);
862 		else
863 			nfs_readdirplus_parent_cache_hit(path->dentry);
864 		err = __nfs_revalidate_inode(server, inode);
865 		if (err)
866 			goto out;
867 	} else
868 		nfs_readdirplus_parent_cache_hit(path->dentry);
869 out_no_revalidate:
870 	/* Only return attributes that were revalidated. */
871 	stat->result_mask &= request_mask;
872 out_no_update:
873 	generic_fillattr(inode, stat);
874 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
875 	if (S_ISDIR(inode->i_mode))
876 		stat->blksize = NFS_SERVER(inode)->dtsize;
877 out:
878 	trace_nfs_getattr_exit(inode, err);
879 	return err;
880 }
881 EXPORT_SYMBOL_GPL(nfs_getattr);
882 
nfs_init_lock_context(struct nfs_lock_context * l_ctx)883 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
884 {
885 	refcount_set(&l_ctx->count, 1);
886 	l_ctx->lockowner = current->files;
887 	INIT_LIST_HEAD(&l_ctx->list);
888 	atomic_set(&l_ctx->io_count, 0);
889 }
890 
__nfs_find_lock_context(struct nfs_open_context * ctx)891 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
892 {
893 	struct nfs_lock_context *pos;
894 
895 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
896 		if (pos->lockowner != current->files)
897 			continue;
898 		if (refcount_inc_not_zero(&pos->count))
899 			return pos;
900 	}
901 	return NULL;
902 }
903 
nfs_get_lock_context(struct nfs_open_context * ctx)904 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
905 {
906 	struct nfs_lock_context *res, *new = NULL;
907 	struct inode *inode = d_inode(ctx->dentry);
908 
909 	rcu_read_lock();
910 	res = __nfs_find_lock_context(ctx);
911 	rcu_read_unlock();
912 	if (res == NULL) {
913 		new = kmalloc(sizeof(*new), GFP_KERNEL);
914 		if (new == NULL)
915 			return ERR_PTR(-ENOMEM);
916 		nfs_init_lock_context(new);
917 		spin_lock(&inode->i_lock);
918 		res = __nfs_find_lock_context(ctx);
919 		if (res == NULL) {
920 			new->open_context = get_nfs_open_context(ctx);
921 			if (new->open_context) {
922 				list_add_tail_rcu(&new->list,
923 						&ctx->lock_context.list);
924 				res = new;
925 				new = NULL;
926 			} else
927 				res = ERR_PTR(-EBADF);
928 		}
929 		spin_unlock(&inode->i_lock);
930 		kfree(new);
931 	}
932 	return res;
933 }
934 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
935 
nfs_put_lock_context(struct nfs_lock_context * l_ctx)936 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
937 {
938 	struct nfs_open_context *ctx = l_ctx->open_context;
939 	struct inode *inode = d_inode(ctx->dentry);
940 
941 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
942 		return;
943 	list_del_rcu(&l_ctx->list);
944 	spin_unlock(&inode->i_lock);
945 	put_nfs_open_context(ctx);
946 	kfree_rcu(l_ctx, rcu_head);
947 }
948 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
949 
950 /**
951  * nfs_close_context - Common close_context() routine NFSv2/v3
952  * @ctx: pointer to context
953  * @is_sync: is this a synchronous close
954  *
955  * Ensure that the attributes are up to date if we're mounted
956  * with close-to-open semantics and we have cached data that will
957  * need to be revalidated on open.
958  */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)959 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
960 {
961 	struct nfs_inode *nfsi;
962 	struct inode *inode;
963 	struct nfs_server *server;
964 
965 	if (!(ctx->mode & FMODE_WRITE))
966 		return;
967 	if (!is_sync)
968 		return;
969 	inode = d_inode(ctx->dentry);
970 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
971 		return;
972 	nfsi = NFS_I(inode);
973 	if (inode->i_mapping->nrpages == 0)
974 		return;
975 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
976 		return;
977 	if (!list_empty(&nfsi->open_files))
978 		return;
979 	server = NFS_SERVER(inode);
980 	if (server->flags & NFS_MOUNT_NOCTO)
981 		return;
982 	nfs_revalidate_inode(server, inode);
983 }
984 EXPORT_SYMBOL_GPL(nfs_close_context);
985 
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)986 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
987 						fmode_t f_mode,
988 						struct file *filp)
989 {
990 	struct nfs_open_context *ctx;
991 
992 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
993 	if (!ctx)
994 		return ERR_PTR(-ENOMEM);
995 	nfs_sb_active(dentry->d_sb);
996 	ctx->dentry = dget(dentry);
997 	if (filp)
998 		ctx->cred = get_cred(filp->f_cred);
999 	else
1000 		ctx->cred = get_current_cred();
1001 	ctx->ll_cred = NULL;
1002 	ctx->state = NULL;
1003 	ctx->mode = f_mode;
1004 	ctx->flags = 0;
1005 	ctx->error = 0;
1006 	ctx->flock_owner = (fl_owner_t)filp;
1007 	nfs_init_lock_context(&ctx->lock_context);
1008 	ctx->lock_context.open_context = ctx;
1009 	INIT_LIST_HEAD(&ctx->list);
1010 	ctx->mdsthreshold = NULL;
1011 	return ctx;
1012 }
1013 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1014 
get_nfs_open_context(struct nfs_open_context * ctx)1015 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1016 {
1017 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1018 		return ctx;
1019 	return NULL;
1020 }
1021 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1022 
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1023 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1024 {
1025 	struct inode *inode = d_inode(ctx->dentry);
1026 	struct super_block *sb = ctx->dentry->d_sb;
1027 
1028 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1029 		return;
1030 	if (!list_empty(&ctx->list)) {
1031 		spin_lock(&inode->i_lock);
1032 		list_del_rcu(&ctx->list);
1033 		spin_unlock(&inode->i_lock);
1034 	}
1035 	if (inode != NULL)
1036 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1037 	put_cred(ctx->cred);
1038 	dput(ctx->dentry);
1039 	nfs_sb_deactive(sb);
1040 	put_rpccred(ctx->ll_cred);
1041 	kfree(ctx->mdsthreshold);
1042 	kfree_rcu(ctx, rcu_head);
1043 }
1044 
put_nfs_open_context(struct nfs_open_context * ctx)1045 void put_nfs_open_context(struct nfs_open_context *ctx)
1046 {
1047 	__put_nfs_open_context(ctx, 0);
1048 }
1049 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1050 
put_nfs_open_context_sync(struct nfs_open_context * ctx)1051 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1052 {
1053 	__put_nfs_open_context(ctx, 1);
1054 }
1055 
1056 /*
1057  * Ensure that mmap has a recent RPC credential for use when writing out
1058  * shared pages
1059  */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1060 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1061 {
1062 	struct inode *inode = d_inode(ctx->dentry);
1063 	struct nfs_inode *nfsi = NFS_I(inode);
1064 
1065 	spin_lock(&inode->i_lock);
1066 	if (list_empty(&nfsi->open_files) &&
1067 	    (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1068 		nfsi->cache_validity |= NFS_INO_INVALID_DATA |
1069 			NFS_INO_REVAL_FORCED;
1070 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1071 	spin_unlock(&inode->i_lock);
1072 }
1073 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1074 
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1075 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1076 {
1077 	filp->private_data = get_nfs_open_context(ctx);
1078 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1079 	if (list_empty(&ctx->list))
1080 		nfs_inode_attach_open_context(ctx);
1081 }
1082 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1083 
1084 /*
1085  * Given an inode, search for an open context with the desired characteristics
1086  */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1087 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1088 {
1089 	struct nfs_inode *nfsi = NFS_I(inode);
1090 	struct nfs_open_context *pos, *ctx = NULL;
1091 
1092 	rcu_read_lock();
1093 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1094 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1095 			continue;
1096 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1097 			continue;
1098 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1099 			continue;
1100 		ctx = get_nfs_open_context(pos);
1101 		if (ctx)
1102 			break;
1103 	}
1104 	rcu_read_unlock();
1105 	return ctx;
1106 }
1107 
nfs_file_clear_open_context(struct file * filp)1108 void nfs_file_clear_open_context(struct file *filp)
1109 {
1110 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1111 
1112 	if (ctx) {
1113 		struct inode *inode = d_inode(ctx->dentry);
1114 
1115 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1116 		/*
1117 		 * We fatal error on write before. Try to writeback
1118 		 * every page again.
1119 		 */
1120 		if (ctx->error < 0)
1121 			invalidate_inode_pages2(inode->i_mapping);
1122 		filp->private_data = NULL;
1123 		put_nfs_open_context_sync(ctx);
1124 	}
1125 }
1126 
1127 /*
1128  * These allocate and release file read/write context information.
1129  */
nfs_open(struct inode * inode,struct file * filp)1130 int nfs_open(struct inode *inode, struct file *filp)
1131 {
1132 	struct nfs_open_context *ctx;
1133 
1134 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1135 	if (IS_ERR(ctx))
1136 		return PTR_ERR(ctx);
1137 	nfs_file_set_open_context(filp, ctx);
1138 	put_nfs_open_context(ctx);
1139 	nfs_fscache_open_file(inode, filp);
1140 	return 0;
1141 }
1142 
1143 /*
1144  * This function is called whenever some part of NFS notices that
1145  * the cached attributes have to be refreshed.
1146  */
1147 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1148 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1149 {
1150 	int		 status = -ESTALE;
1151 	struct nfs4_label *label = NULL;
1152 	struct nfs_fattr *fattr = NULL;
1153 	struct nfs_inode *nfsi = NFS_I(inode);
1154 
1155 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1156 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1157 
1158 	trace_nfs_revalidate_inode_enter(inode);
1159 
1160 	if (is_bad_inode(inode))
1161 		goto out;
1162 	if (NFS_STALE(inode))
1163 		goto out;
1164 
1165 	/* pNFS: Attributes aren't updated until we layoutcommit */
1166 	if (S_ISREG(inode->i_mode)) {
1167 		status = pnfs_sync_inode(inode, false);
1168 		if (status)
1169 			goto out;
1170 	}
1171 
1172 	status = -ENOMEM;
1173 	fattr = nfs_alloc_fattr();
1174 	if (fattr == NULL)
1175 		goto out;
1176 
1177 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1178 
1179 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1180 	if (IS_ERR(label)) {
1181 		status = PTR_ERR(label);
1182 		goto out;
1183 	}
1184 
1185 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1186 			label, inode);
1187 	if (status != 0) {
1188 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1189 			 inode->i_sb->s_id,
1190 			 (unsigned long long)NFS_FILEID(inode), status);
1191 		switch (status) {
1192 		case -ETIMEDOUT:
1193 			/* A soft timeout occurred. Use cached information? */
1194 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1195 				status = 0;
1196 			break;
1197 		case -ESTALE:
1198 			if (!S_ISDIR(inode->i_mode))
1199 				nfs_set_inode_stale(inode);
1200 			else
1201 				nfs_zap_caches(inode);
1202 		}
1203 		goto err_out;
1204 	}
1205 
1206 	status = nfs_refresh_inode(inode, fattr);
1207 	if (status) {
1208 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1209 			 inode->i_sb->s_id,
1210 			 (unsigned long long)NFS_FILEID(inode), status);
1211 		goto err_out;
1212 	}
1213 
1214 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1215 		nfs_zap_acl_cache(inode);
1216 
1217 	nfs_setsecurity(inode, fattr, label);
1218 
1219 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1220 		inode->i_sb->s_id,
1221 		(unsigned long long)NFS_FILEID(inode));
1222 
1223 err_out:
1224 	nfs4_label_free(label);
1225 out:
1226 	nfs_free_fattr(fattr);
1227 	trace_nfs_revalidate_inode_exit(inode, status);
1228 	return status;
1229 }
1230 
nfs_attribute_cache_expired(struct inode * inode)1231 int nfs_attribute_cache_expired(struct inode *inode)
1232 {
1233 	if (nfs_have_delegated_attributes(inode))
1234 		return 0;
1235 	return nfs_attribute_timeout(inode);
1236 }
1237 
1238 /**
1239  * nfs_revalidate_inode - Revalidate the inode attributes
1240  * @server: pointer to nfs_server struct
1241  * @inode: pointer to inode struct
1242  *
1243  * Updates inode attribute information by retrieving the data from the server.
1244  */
nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1245 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1246 {
1247 	if (!nfs_need_revalidate_inode(inode))
1248 		return NFS_STALE(inode) ? -ESTALE : 0;
1249 	return __nfs_revalidate_inode(server, inode);
1250 }
1251 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1252 
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1253 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1254 {
1255 	struct nfs_inode *nfsi = NFS_I(inode);
1256 	int ret;
1257 
1258 	if (mapping->nrpages != 0) {
1259 		if (S_ISREG(inode->i_mode)) {
1260 			ret = nfs_sync_mapping(mapping);
1261 			if (ret < 0)
1262 				return ret;
1263 		}
1264 		ret = invalidate_inode_pages2(mapping);
1265 		if (ret < 0)
1266 			return ret;
1267 	}
1268 	if (S_ISDIR(inode->i_mode)) {
1269 		spin_lock(&inode->i_lock);
1270 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1271 		spin_unlock(&inode->i_lock);
1272 	}
1273 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1274 	nfs_fscache_wait_on_invalidate(inode);
1275 
1276 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1277 			inode->i_sb->s_id,
1278 			(unsigned long long)NFS_FILEID(inode));
1279 	return 0;
1280 }
1281 
nfs_mapping_need_revalidate_inode(struct inode * inode)1282 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1283 {
1284 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1285 		NFS_STALE(inode);
1286 }
1287 
nfs_revalidate_mapping_rcu(struct inode * inode)1288 int nfs_revalidate_mapping_rcu(struct inode *inode)
1289 {
1290 	struct nfs_inode *nfsi = NFS_I(inode);
1291 	unsigned long *bitlock = &nfsi->flags;
1292 	int ret = 0;
1293 
1294 	if (IS_SWAPFILE(inode))
1295 		goto out;
1296 	if (nfs_mapping_need_revalidate_inode(inode)) {
1297 		ret = -ECHILD;
1298 		goto out;
1299 	}
1300 	spin_lock(&inode->i_lock);
1301 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1302 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1303 		ret = -ECHILD;
1304 	spin_unlock(&inode->i_lock);
1305 out:
1306 	return ret;
1307 }
1308 
1309 /**
1310  * nfs_revalidate_mapping - Revalidate the pagecache
1311  * @inode: pointer to host inode
1312  * @mapping: pointer to mapping
1313  */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1314 int nfs_revalidate_mapping(struct inode *inode,
1315 		struct address_space *mapping)
1316 {
1317 	struct nfs_inode *nfsi = NFS_I(inode);
1318 	unsigned long *bitlock = &nfsi->flags;
1319 	int ret = 0;
1320 
1321 	/* swapfiles are not supposed to be shared. */
1322 	if (IS_SWAPFILE(inode))
1323 		goto out;
1324 
1325 	if (nfs_mapping_need_revalidate_inode(inode)) {
1326 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1327 		if (ret < 0)
1328 			goto out;
1329 	}
1330 
1331 	/*
1332 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1333 	 * invalidations that come in while we're shooting down the mappings
1334 	 * are respected. But, that leaves a race window where one revalidator
1335 	 * can clear the flag, and then another checks it before the mapping
1336 	 * gets invalidated. Fix that by serializing access to this part of
1337 	 * the function.
1338 	 *
1339 	 * At the same time, we need to allow other tasks to see whether we
1340 	 * might be in the middle of invalidating the pages, so we only set
1341 	 * the bit lock here if it looks like we're going to be doing that.
1342 	 */
1343 	for (;;) {
1344 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1345 					 nfs_wait_bit_killable, TASK_KILLABLE);
1346 		if (ret)
1347 			goto out;
1348 		spin_lock(&inode->i_lock);
1349 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1350 			spin_unlock(&inode->i_lock);
1351 			continue;
1352 		}
1353 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1354 			break;
1355 		spin_unlock(&inode->i_lock);
1356 		goto out;
1357 	}
1358 
1359 	set_bit(NFS_INO_INVALIDATING, bitlock);
1360 	smp_wmb();
1361 	nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA|
1362 			NFS_INO_DATA_INVAL_DEFER);
1363 	spin_unlock(&inode->i_lock);
1364 	trace_nfs_invalidate_mapping_enter(inode);
1365 	ret = nfs_invalidate_mapping(inode, mapping);
1366 	trace_nfs_invalidate_mapping_exit(inode, ret);
1367 
1368 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1369 	smp_mb__after_atomic();
1370 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1371 out:
1372 	return ret;
1373 }
1374 
nfs_file_has_writers(struct nfs_inode * nfsi)1375 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1376 {
1377 	struct inode *inode = &nfsi->vfs_inode;
1378 
1379 	if (!S_ISREG(inode->i_mode))
1380 		return false;
1381 	if (list_empty(&nfsi->open_files))
1382 		return false;
1383 	return inode_is_open_for_write(inode);
1384 }
1385 
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1386 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1387 {
1388 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1389 }
1390 
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1391 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1392 {
1393 	struct timespec64 ts;
1394 
1395 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1396 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1397 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1398 		inode_set_iversion_raw(inode, fattr->change_attr);
1399 		if (S_ISDIR(inode->i_mode))
1400 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1401 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1402 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1403 	}
1404 	/* If we have atomic WCC data, we may update some attributes */
1405 	ts = inode->i_ctime;
1406 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1407 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1408 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1409 		inode->i_ctime = fattr->ctime;
1410 	}
1411 
1412 	ts = inode->i_mtime;
1413 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1414 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1415 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1416 		inode->i_mtime = fattr->mtime;
1417 		if (S_ISDIR(inode->i_mode))
1418 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1419 	}
1420 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1421 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1422 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1423 			&& !nfs_have_writebacks(inode)) {
1424 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1425 	}
1426 }
1427 
1428 /**
1429  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1430  * @inode: pointer to inode
1431  * @fattr: updated attributes
1432  *
1433  * Verifies the attribute cache. If we have just changed the attributes,
1434  * so that fattr carries weak cache consistency data, then it may
1435  * also update the ctime/mtime/change_attribute.
1436  */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1437 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1438 {
1439 	struct nfs_inode *nfsi = NFS_I(inode);
1440 	loff_t cur_size, new_isize;
1441 	unsigned long invalid = 0;
1442 	struct timespec64 ts;
1443 
1444 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1445 		return 0;
1446 
1447 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1448 		/* Only a mounted-on-fileid? Just exit */
1449 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1450 			return 0;
1451 	/* Has the inode gone and changed behind our back? */
1452 	} else if (nfsi->fileid != fattr->fileid) {
1453 		/* Is this perhaps the mounted-on fileid? */
1454 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1455 		    nfsi->fileid == fattr->mounted_on_fileid)
1456 			return 0;
1457 		return -ESTALE;
1458 	}
1459 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1460 		return -ESTALE;
1461 
1462 
1463 	if (!nfs_file_has_buffered_writers(nfsi)) {
1464 		/* Verify a few of the more important attributes */
1465 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1466 			invalid |= NFS_INO_INVALID_CHANGE
1467 				| NFS_INO_REVAL_PAGECACHE;
1468 
1469 		ts = inode->i_mtime;
1470 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1471 			invalid |= NFS_INO_INVALID_MTIME;
1472 
1473 		ts = inode->i_ctime;
1474 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1475 			invalid |= NFS_INO_INVALID_CTIME;
1476 
1477 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1478 			cur_size = i_size_read(inode);
1479 			new_isize = nfs_size_to_loff_t(fattr->size);
1480 			if (cur_size != new_isize)
1481 				invalid |= NFS_INO_INVALID_SIZE
1482 					| NFS_INO_REVAL_PAGECACHE;
1483 		}
1484 	}
1485 
1486 	/* Have any file permissions changed? */
1487 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1488 		invalid |= NFS_INO_INVALID_ACCESS
1489 			| NFS_INO_INVALID_ACL
1490 			| NFS_INO_INVALID_OTHER;
1491 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1492 		invalid |= NFS_INO_INVALID_ACCESS
1493 			| NFS_INO_INVALID_ACL
1494 			| NFS_INO_INVALID_OTHER;
1495 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1496 		invalid |= NFS_INO_INVALID_ACCESS
1497 			| NFS_INO_INVALID_ACL
1498 			| NFS_INO_INVALID_OTHER;
1499 
1500 	/* Has the link count changed? */
1501 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1502 		invalid |= NFS_INO_INVALID_OTHER;
1503 
1504 	ts = inode->i_atime;
1505 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1506 		invalid |= NFS_INO_INVALID_ATIME;
1507 
1508 	if (invalid != 0)
1509 		nfs_set_cache_invalid(inode, invalid);
1510 
1511 	nfsi->read_cache_jiffies = fattr->time_start;
1512 	return 0;
1513 }
1514 
1515 static atomic_long_t nfs_attr_generation_counter;
1516 
nfs_read_attr_generation_counter(void)1517 static unsigned long nfs_read_attr_generation_counter(void)
1518 {
1519 	return atomic_long_read(&nfs_attr_generation_counter);
1520 }
1521 
nfs_inc_attr_generation_counter(void)1522 unsigned long nfs_inc_attr_generation_counter(void)
1523 {
1524 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1525 }
1526 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1527 
nfs_fattr_init(struct nfs_fattr * fattr)1528 void nfs_fattr_init(struct nfs_fattr *fattr)
1529 {
1530 	fattr->valid = 0;
1531 	fattr->time_start = jiffies;
1532 	fattr->gencount = nfs_inc_attr_generation_counter();
1533 	fattr->owner_name = NULL;
1534 	fattr->group_name = NULL;
1535 }
1536 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1537 
1538 /**
1539  * nfs_fattr_set_barrier
1540  * @fattr: attributes
1541  *
1542  * Used to set a barrier after an attribute was updated. This
1543  * barrier ensures that older attributes from RPC calls that may
1544  * have raced with our update cannot clobber these new values.
1545  * Note that you are still responsible for ensuring that other
1546  * operations which change the attribute on the server do not
1547  * collide.
1548  */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1549 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1550 {
1551 	fattr->gencount = nfs_inc_attr_generation_counter();
1552 }
1553 
nfs_alloc_fattr(void)1554 struct nfs_fattr *nfs_alloc_fattr(void)
1555 {
1556 	struct nfs_fattr *fattr;
1557 
1558 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1559 	if (fattr != NULL)
1560 		nfs_fattr_init(fattr);
1561 	return fattr;
1562 }
1563 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1564 
nfs_alloc_fhandle(void)1565 struct nfs_fh *nfs_alloc_fhandle(void)
1566 {
1567 	struct nfs_fh *fh;
1568 
1569 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1570 	if (fh != NULL)
1571 		fh->size = 0;
1572 	return fh;
1573 }
1574 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1575 
1576 #ifdef NFS_DEBUG
1577 /*
1578  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1579  *                             in the same way that wireshark does
1580  *
1581  * @fh: file handle
1582  *
1583  * For debugging only.
1584  */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1585 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1586 {
1587 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1588 	 * not on the result */
1589 	return nfs_fhandle_hash(fh);
1590 }
1591 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1592 
1593 /*
1594  * _nfs_display_fhandle - display an NFS file handle on the console
1595  *
1596  * @fh: file handle to display
1597  * @caption: display caption
1598  *
1599  * For debugging only.
1600  */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1601 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1602 {
1603 	unsigned short i;
1604 
1605 	if (fh == NULL || fh->size == 0) {
1606 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1607 		return;
1608 	}
1609 
1610 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1611 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1612 	for (i = 0; i < fh->size; i += 16) {
1613 		__be32 *pos = (__be32 *)&fh->data[i];
1614 
1615 		switch ((fh->size - i - 1) >> 2) {
1616 		case 0:
1617 			printk(KERN_DEFAULT " %08x\n",
1618 				be32_to_cpup(pos));
1619 			break;
1620 		case 1:
1621 			printk(KERN_DEFAULT " %08x %08x\n",
1622 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1623 			break;
1624 		case 2:
1625 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1626 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1627 				be32_to_cpup(pos + 2));
1628 			break;
1629 		default:
1630 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1631 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1632 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1633 		}
1634 	}
1635 }
1636 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1637 #endif
1638 
1639 /**
1640  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1641  * @inode: pointer to inode
1642  * @fattr: attributes
1643  *
1644  * Attempt to divine whether or not an RPC call reply carrying stale
1645  * attributes got scheduled after another call carrying updated ones.
1646  *
1647  * To do so, the function first assumes that a more recent ctime means
1648  * that the attributes in fattr are newer, however it also attempt to
1649  * catch the case where ctime either didn't change, or went backwards
1650  * (if someone reset the clock on the server) by looking at whether
1651  * or not this RPC call was started after the inode was last updated.
1652  * Note also the check for wraparound of 'attr_gencount'
1653  *
1654  * The function returns 'true' if it thinks the attributes in 'fattr' are
1655  * more recent than the ones cached in the inode.
1656  *
1657  */
nfs_inode_attrs_need_update(const struct inode * inode,const struct nfs_fattr * fattr)1658 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1659 {
1660 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1661 
1662 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1663 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1664 }
1665 
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1666 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1667 {
1668 	int ret;
1669 
1670 	trace_nfs_refresh_inode_enter(inode);
1671 
1672 	if (nfs_inode_attrs_need_update(inode, fattr))
1673 		ret = nfs_update_inode(inode, fattr);
1674 	else
1675 		ret = nfs_check_inode_attributes(inode, fattr);
1676 
1677 	trace_nfs_refresh_inode_exit(inode, ret);
1678 	return ret;
1679 }
1680 
1681 /**
1682  * nfs_refresh_inode - try to update the inode attribute cache
1683  * @inode: pointer to inode
1684  * @fattr: updated attributes
1685  *
1686  * Check that an RPC call that returned attributes has not overlapped with
1687  * other recent updates of the inode metadata, then decide whether it is
1688  * safe to do a full update of the inode attributes, or whether just to
1689  * call nfs_check_inode_attributes.
1690  */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1691 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1692 {
1693 	int status;
1694 
1695 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1696 		return 0;
1697 	spin_lock(&inode->i_lock);
1698 	status = nfs_refresh_inode_locked(inode, fattr);
1699 	spin_unlock(&inode->i_lock);
1700 
1701 	return status;
1702 }
1703 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1704 
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1705 static int nfs_post_op_update_inode_locked(struct inode *inode,
1706 		struct nfs_fattr *fattr, unsigned int invalid)
1707 {
1708 	if (S_ISDIR(inode->i_mode))
1709 		invalid |= NFS_INO_INVALID_DATA;
1710 	nfs_set_cache_invalid(inode, invalid);
1711 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1712 		return 0;
1713 	return nfs_refresh_inode_locked(inode, fattr);
1714 }
1715 
1716 /**
1717  * nfs_post_op_update_inode - try to update the inode attribute cache
1718  * @inode: pointer to inode
1719  * @fattr: updated attributes
1720  *
1721  * After an operation that has changed the inode metadata, mark the
1722  * attribute cache as being invalid, then try to update it.
1723  *
1724  * NB: if the server didn't return any post op attributes, this
1725  * function will force the retrieval of attributes before the next
1726  * NFS request.  Thus it should be used only for operations that
1727  * are expected to change one or more attributes, to avoid
1728  * unnecessary NFS requests and trips through nfs_update_inode().
1729  */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1730 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1731 {
1732 	int status;
1733 
1734 	spin_lock(&inode->i_lock);
1735 	nfs_fattr_set_barrier(fattr);
1736 	status = nfs_post_op_update_inode_locked(inode, fattr,
1737 			NFS_INO_INVALID_CHANGE
1738 			| NFS_INO_INVALID_CTIME
1739 			| NFS_INO_REVAL_FORCED);
1740 	spin_unlock(&inode->i_lock);
1741 
1742 	return status;
1743 }
1744 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1745 
1746 /**
1747  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1748  * @inode: pointer to inode
1749  * @fattr: updated attributes
1750  *
1751  * After an operation that has changed the inode metadata, mark the
1752  * attribute cache as being invalid, then try to update it. Fake up
1753  * weak cache consistency data, if none exist.
1754  *
1755  * This function is mainly designed to be used by the ->write_done() functions.
1756  */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1757 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1758 {
1759 	int status;
1760 
1761 	/* Don't do a WCC update if these attributes are already stale */
1762 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1763 			!nfs_inode_attrs_need_update(inode, fattr)) {
1764 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1765 				| NFS_ATTR_FATTR_PRESIZE
1766 				| NFS_ATTR_FATTR_PREMTIME
1767 				| NFS_ATTR_FATTR_PRECTIME);
1768 		goto out_noforce;
1769 	}
1770 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1771 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1772 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1773 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1774 	}
1775 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1776 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1777 		fattr->pre_ctime = inode->i_ctime;
1778 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1779 	}
1780 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1781 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1782 		fattr->pre_mtime = inode->i_mtime;
1783 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1784 	}
1785 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1786 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1787 		fattr->pre_size = i_size_read(inode);
1788 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1789 	}
1790 out_noforce:
1791 	status = nfs_post_op_update_inode_locked(inode, fattr,
1792 			NFS_INO_INVALID_CHANGE
1793 			| NFS_INO_INVALID_CTIME
1794 			| NFS_INO_INVALID_MTIME
1795 			| NFS_INO_INVALID_BLOCKS);
1796 	return status;
1797 }
1798 
1799 /**
1800  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1801  * @inode: pointer to inode
1802  * @fattr: updated attributes
1803  *
1804  * After an operation that has changed the inode metadata, mark the
1805  * attribute cache as being invalid, then try to update it. Fake up
1806  * weak cache consistency data, if none exist.
1807  *
1808  * This function is mainly designed to be used by the ->write_done() functions.
1809  */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)1810 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1811 {
1812 	int status;
1813 
1814 	spin_lock(&inode->i_lock);
1815 	nfs_fattr_set_barrier(fattr);
1816 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1817 	spin_unlock(&inode->i_lock);
1818 	return status;
1819 }
1820 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1821 
1822 
1823 /*
1824  * Many nfs protocol calls return the new file attributes after
1825  * an operation.  Here we update the inode to reflect the state
1826  * of the server's inode.
1827  *
1828  * This is a bit tricky because we have to make sure all dirty pages
1829  * have been sent off to the server before calling invalidate_inode_pages.
1830  * To make sure no other process adds more write requests while we try
1831  * our best to flush them, we make them sleep during the attribute refresh.
1832  *
1833  * A very similar scenario holds for the dir cache.
1834  */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)1835 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1836 {
1837 	struct nfs_server *server;
1838 	struct nfs_inode *nfsi = NFS_I(inode);
1839 	loff_t cur_isize, new_isize;
1840 	unsigned long invalid = 0;
1841 	unsigned long now = jiffies;
1842 	unsigned long save_cache_validity;
1843 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1844 	bool cache_revalidated = true;
1845 	bool attr_changed = false;
1846 	bool have_delegation;
1847 
1848 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1849 			__func__, inode->i_sb->s_id, inode->i_ino,
1850 			nfs_display_fhandle_hash(NFS_FH(inode)),
1851 			atomic_read(&inode->i_count), fattr->valid);
1852 
1853 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1854 		/* Only a mounted-on-fileid? Just exit */
1855 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1856 			return 0;
1857 	/* Has the inode gone and changed behind our back? */
1858 	} else if (nfsi->fileid != fattr->fileid) {
1859 		/* Is this perhaps the mounted-on fileid? */
1860 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1861 		    nfsi->fileid == fattr->mounted_on_fileid)
1862 			return 0;
1863 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1864 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1865 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1866 			inode->i_sb->s_id, (long long)nfsi->fileid,
1867 			(long long)fattr->fileid);
1868 		goto out_err;
1869 	}
1870 
1871 	/*
1872 	 * Make sure the inode's type hasn't changed.
1873 	 */
1874 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
1875 		/*
1876 		* Big trouble! The inode has become a different object.
1877 		*/
1878 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1879 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1880 		goto out_err;
1881 	}
1882 
1883 	server = NFS_SERVER(inode);
1884 	/* Update the fsid? */
1885 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1886 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1887 			!IS_AUTOMOUNT(inode))
1888 		server->fsid = fattr->fsid;
1889 
1890 	/* Save the delegation state before clearing cache_validity */
1891 	have_delegation = nfs_have_delegated_attributes(inode);
1892 
1893 	/*
1894 	 * Update the read time so we don't revalidate too often.
1895 	 */
1896 	nfsi->read_cache_jiffies = fattr->time_start;
1897 
1898 	save_cache_validity = nfsi->cache_validity;
1899 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1900 			| NFS_INO_INVALID_ATIME
1901 			| NFS_INO_REVAL_FORCED
1902 			| NFS_INO_REVAL_PAGECACHE
1903 			| NFS_INO_INVALID_BLOCKS);
1904 
1905 	/* Do atomic weak cache consistency updates */
1906 	nfs_wcc_update_inode(inode, fattr);
1907 
1908 	if (pnfs_layoutcommit_outstanding(inode)) {
1909 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1910 		cache_revalidated = false;
1911 	}
1912 
1913 	/* More cache consistency checks */
1914 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1915 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1916 			/* Could it be a race with writeback? */
1917 			if (!(have_writers || have_delegation)) {
1918 				invalid |= NFS_INO_INVALID_DATA
1919 					| NFS_INO_INVALID_ACCESS
1920 					| NFS_INO_INVALID_ACL
1921 					| NFS_INO_INVALID_XATTR;
1922 				/* Force revalidate of all attributes */
1923 				save_cache_validity |= NFS_INO_INVALID_CTIME
1924 					| NFS_INO_INVALID_MTIME
1925 					| NFS_INO_INVALID_SIZE
1926 					| NFS_INO_INVALID_OTHER;
1927 				if (S_ISDIR(inode->i_mode))
1928 					nfs_force_lookup_revalidate(inode);
1929 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
1930 						inode->i_sb->s_id,
1931 						inode->i_ino);
1932 			} else if (!have_delegation)
1933 				nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1934 			inode_set_iversion_raw(inode, fattr->change_attr);
1935 			attr_changed = true;
1936 		}
1937 	} else {
1938 		nfsi->cache_validity |= save_cache_validity &
1939 				(NFS_INO_INVALID_CHANGE
1940 				| NFS_INO_REVAL_PAGECACHE
1941 				| NFS_INO_REVAL_FORCED);
1942 		cache_revalidated = false;
1943 	}
1944 
1945 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1946 		inode->i_mtime = fattr->mtime;
1947 	} else if (server->caps & NFS_CAP_MTIME) {
1948 		nfsi->cache_validity |= save_cache_validity &
1949 				(NFS_INO_INVALID_MTIME
1950 				| NFS_INO_REVAL_FORCED);
1951 		cache_revalidated = false;
1952 	}
1953 
1954 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1955 		inode->i_ctime = fattr->ctime;
1956 	} else if (server->caps & NFS_CAP_CTIME) {
1957 		nfsi->cache_validity |= save_cache_validity &
1958 				(NFS_INO_INVALID_CTIME
1959 				| NFS_INO_REVAL_FORCED);
1960 		cache_revalidated = false;
1961 	}
1962 
1963 	/* Check if our cached file size is stale */
1964 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1965 		new_isize = nfs_size_to_loff_t(fattr->size);
1966 		cur_isize = i_size_read(inode);
1967 		if (new_isize != cur_isize && !have_delegation) {
1968 			/* Do we perhaps have any outstanding writes, or has
1969 			 * the file grown beyond our last write? */
1970 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1971 				i_size_write(inode, new_isize);
1972 				if (!have_writers)
1973 					invalid |= NFS_INO_INVALID_DATA;
1974 				attr_changed = true;
1975 			}
1976 			dprintk("NFS: isize change on server for file %s/%ld "
1977 					"(%Ld to %Ld)\n",
1978 					inode->i_sb->s_id,
1979 					inode->i_ino,
1980 					(long long)cur_isize,
1981 					(long long)new_isize);
1982 		}
1983 	} else {
1984 		nfsi->cache_validity |= save_cache_validity &
1985 				(NFS_INO_INVALID_SIZE
1986 				| NFS_INO_REVAL_PAGECACHE
1987 				| NFS_INO_REVAL_FORCED);
1988 		cache_revalidated = false;
1989 	}
1990 
1991 
1992 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1993 		inode->i_atime = fattr->atime;
1994 	else if (server->caps & NFS_CAP_ATIME) {
1995 		nfsi->cache_validity |= save_cache_validity &
1996 				(NFS_INO_INVALID_ATIME
1997 				| NFS_INO_REVAL_FORCED);
1998 		cache_revalidated = false;
1999 	}
2000 
2001 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2002 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2003 			umode_t newmode = inode->i_mode & S_IFMT;
2004 			newmode |= fattr->mode & S_IALLUGO;
2005 			inode->i_mode = newmode;
2006 			invalid |= NFS_INO_INVALID_ACCESS
2007 				| NFS_INO_INVALID_ACL;
2008 			attr_changed = true;
2009 		}
2010 	} else if (server->caps & NFS_CAP_MODE) {
2011 		nfsi->cache_validity |= save_cache_validity &
2012 				(NFS_INO_INVALID_OTHER
2013 				| NFS_INO_REVAL_FORCED);
2014 		cache_revalidated = false;
2015 	}
2016 
2017 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2018 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2019 			invalid |= NFS_INO_INVALID_ACCESS
2020 				| NFS_INO_INVALID_ACL;
2021 			inode->i_uid = fattr->uid;
2022 			attr_changed = true;
2023 		}
2024 	} else if (server->caps & NFS_CAP_OWNER) {
2025 		nfsi->cache_validity |= save_cache_validity &
2026 				(NFS_INO_INVALID_OTHER
2027 				| NFS_INO_REVAL_FORCED);
2028 		cache_revalidated = false;
2029 	}
2030 
2031 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2032 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2033 			invalid |= NFS_INO_INVALID_ACCESS
2034 				| NFS_INO_INVALID_ACL;
2035 			inode->i_gid = fattr->gid;
2036 			attr_changed = true;
2037 		}
2038 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
2039 		nfsi->cache_validity |= save_cache_validity &
2040 				(NFS_INO_INVALID_OTHER
2041 				| NFS_INO_REVAL_FORCED);
2042 		cache_revalidated = false;
2043 	}
2044 
2045 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2046 		if (inode->i_nlink != fattr->nlink) {
2047 			if (S_ISDIR(inode->i_mode))
2048 				invalid |= NFS_INO_INVALID_DATA;
2049 			set_nlink(inode, fattr->nlink);
2050 			attr_changed = true;
2051 		}
2052 	} else if (server->caps & NFS_CAP_NLINK) {
2053 		nfsi->cache_validity |= save_cache_validity &
2054 				(NFS_INO_INVALID_OTHER
2055 				| NFS_INO_REVAL_FORCED);
2056 		cache_revalidated = false;
2057 	}
2058 
2059 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2060 		/*
2061 		 * report the blocks in 512byte units
2062 		 */
2063 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2064 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2065 		inode->i_blocks = fattr->du.nfs2.blocks;
2066 	else {
2067 		nfsi->cache_validity |= save_cache_validity &
2068 				(NFS_INO_INVALID_BLOCKS
2069 				| NFS_INO_REVAL_FORCED);
2070 		cache_revalidated = false;
2071 	}
2072 
2073 	/* Update attrtimeo value if we're out of the unstable period */
2074 	if (attr_changed) {
2075 		invalid &= ~NFS_INO_INVALID_ATTR;
2076 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2077 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2078 		nfsi->attrtimeo_timestamp = now;
2079 		/* Set barrier to be more recent than all outstanding updates */
2080 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2081 	} else {
2082 		if (cache_revalidated) {
2083 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2084 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2085 				nfsi->attrtimeo <<= 1;
2086 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2087 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2088 			}
2089 			nfsi->attrtimeo_timestamp = now;
2090 		}
2091 		/* Set the barrier to be more recent than this fattr */
2092 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2093 			nfsi->attr_gencount = fattr->gencount;
2094 	}
2095 
2096 	/* Don't invalidate the data if we were to blame */
2097 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2098 				|| S_ISLNK(inode->i_mode)))
2099 		invalid &= ~NFS_INO_INVALID_DATA;
2100 	nfs_set_cache_invalid(inode, invalid);
2101 
2102 	return 0;
2103  out_err:
2104 	/*
2105 	 * No need to worry about unhashing the dentry, as the
2106 	 * lookup validation will know that the inode is bad.
2107 	 * (But we fall through to invalidate the caches.)
2108 	 */
2109 	nfs_set_inode_stale_locked(inode);
2110 	return -ESTALE;
2111 }
2112 
nfs_alloc_inode(struct super_block * sb)2113 struct inode *nfs_alloc_inode(struct super_block *sb)
2114 {
2115 	struct nfs_inode *nfsi;
2116 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2117 	if (!nfsi)
2118 		return NULL;
2119 	nfsi->flags = 0UL;
2120 	nfsi->cache_validity = 0UL;
2121 #if IS_ENABLED(CONFIG_NFS_V4)
2122 	nfsi->nfs4_acl = NULL;
2123 #endif /* CONFIG_NFS_V4 */
2124 #ifdef CONFIG_NFS_V4_2
2125 	nfsi->xattr_cache = NULL;
2126 #endif
2127 	return &nfsi->vfs_inode;
2128 }
2129 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2130 
nfs_free_inode(struct inode * inode)2131 void nfs_free_inode(struct inode *inode)
2132 {
2133 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2134 }
2135 EXPORT_SYMBOL_GPL(nfs_free_inode);
2136 
nfs4_init_once(struct nfs_inode * nfsi)2137 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2138 {
2139 #if IS_ENABLED(CONFIG_NFS_V4)
2140 	INIT_LIST_HEAD(&nfsi->open_states);
2141 	nfsi->delegation = NULL;
2142 	init_rwsem(&nfsi->rwsem);
2143 	nfsi->layout = NULL;
2144 #endif
2145 }
2146 
init_once(void * foo)2147 static void init_once(void *foo)
2148 {
2149 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2150 
2151 	inode_init_once(&nfsi->vfs_inode);
2152 	INIT_LIST_HEAD(&nfsi->open_files);
2153 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2154 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2155 	INIT_LIST_HEAD(&nfsi->commit_info.list);
2156 	atomic_long_set(&nfsi->nrequests, 0);
2157 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
2158 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
2159 	init_rwsem(&nfsi->rmdir_sem);
2160 	mutex_init(&nfsi->commit_mutex);
2161 	nfs4_init_once(nfsi);
2162 	nfsi->cache_change_attribute = 0;
2163 }
2164 
nfs_init_inodecache(void)2165 static int __init nfs_init_inodecache(void)
2166 {
2167 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2168 					     sizeof(struct nfs_inode),
2169 					     0, (SLAB_RECLAIM_ACCOUNT|
2170 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2171 					     init_once);
2172 	if (nfs_inode_cachep == NULL)
2173 		return -ENOMEM;
2174 
2175 	return 0;
2176 }
2177 
nfs_destroy_inodecache(void)2178 static void nfs_destroy_inodecache(void)
2179 {
2180 	/*
2181 	 * Make sure all delayed rcu free inodes are flushed before we
2182 	 * destroy cache.
2183 	 */
2184 	rcu_barrier();
2185 	kmem_cache_destroy(nfs_inode_cachep);
2186 }
2187 
2188 struct workqueue_struct *nfsiod_workqueue;
2189 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2190 
2191 /*
2192  * start up the nfsiod workqueue
2193  */
nfsiod_start(void)2194 static int nfsiod_start(void)
2195 {
2196 	struct workqueue_struct *wq;
2197 	dprintk("RPC:       creating workqueue nfsiod\n");
2198 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2199 	if (wq == NULL)
2200 		return -ENOMEM;
2201 	nfsiod_workqueue = wq;
2202 	return 0;
2203 }
2204 
2205 /*
2206  * Destroy the nfsiod workqueue
2207  */
nfsiod_stop(void)2208 static void nfsiod_stop(void)
2209 {
2210 	struct workqueue_struct *wq;
2211 
2212 	wq = nfsiod_workqueue;
2213 	if (wq == NULL)
2214 		return;
2215 	nfsiod_workqueue = NULL;
2216 	destroy_workqueue(wq);
2217 }
2218 
2219 unsigned int nfs_net_id;
2220 EXPORT_SYMBOL_GPL(nfs_net_id);
2221 
nfs_net_init(struct net * net)2222 static int nfs_net_init(struct net *net)
2223 {
2224 	nfs_clients_init(net);
2225 	return nfs_fs_proc_net_init(net);
2226 }
2227 
nfs_net_exit(struct net * net)2228 static void nfs_net_exit(struct net *net)
2229 {
2230 	nfs_fs_proc_net_exit(net);
2231 	nfs_clients_exit(net);
2232 }
2233 
2234 static struct pernet_operations nfs_net_ops = {
2235 	.init = nfs_net_init,
2236 	.exit = nfs_net_exit,
2237 	.id   = &nfs_net_id,
2238 	.size = sizeof(struct nfs_net),
2239 };
2240 
2241 /*
2242  * Initialize NFS
2243  */
init_nfs_fs(void)2244 static int __init init_nfs_fs(void)
2245 {
2246 	int err;
2247 
2248 	err = nfs_sysfs_init();
2249 	if (err < 0)
2250 		goto out10;
2251 
2252 	err = register_pernet_subsys(&nfs_net_ops);
2253 	if (err < 0)
2254 		goto out9;
2255 
2256 	err = nfs_fscache_register();
2257 	if (err < 0)
2258 		goto out8;
2259 
2260 	err = nfsiod_start();
2261 	if (err)
2262 		goto out7;
2263 
2264 	err = nfs_fs_proc_init();
2265 	if (err)
2266 		goto out6;
2267 
2268 	err = nfs_init_nfspagecache();
2269 	if (err)
2270 		goto out5;
2271 
2272 	err = nfs_init_inodecache();
2273 	if (err)
2274 		goto out4;
2275 
2276 	err = nfs_init_readpagecache();
2277 	if (err)
2278 		goto out3;
2279 
2280 	err = nfs_init_writepagecache();
2281 	if (err)
2282 		goto out2;
2283 
2284 	err = nfs_init_directcache();
2285 	if (err)
2286 		goto out1;
2287 
2288 	rpc_proc_register(&init_net, &nfs_rpcstat);
2289 
2290 	err = register_nfs_fs();
2291 	if (err)
2292 		goto out0;
2293 
2294 	return 0;
2295 out0:
2296 	rpc_proc_unregister(&init_net, "nfs");
2297 	nfs_destroy_directcache();
2298 out1:
2299 	nfs_destroy_writepagecache();
2300 out2:
2301 	nfs_destroy_readpagecache();
2302 out3:
2303 	nfs_destroy_inodecache();
2304 out4:
2305 	nfs_destroy_nfspagecache();
2306 out5:
2307 	nfs_fs_proc_exit();
2308 out6:
2309 	nfsiod_stop();
2310 out7:
2311 	nfs_fscache_unregister();
2312 out8:
2313 	unregister_pernet_subsys(&nfs_net_ops);
2314 out9:
2315 	nfs_sysfs_exit();
2316 out10:
2317 	return err;
2318 }
2319 
exit_nfs_fs(void)2320 static void __exit exit_nfs_fs(void)
2321 {
2322 	nfs_destroy_directcache();
2323 	nfs_destroy_writepagecache();
2324 	nfs_destroy_readpagecache();
2325 	nfs_destroy_inodecache();
2326 	nfs_destroy_nfspagecache();
2327 	nfs_fscache_unregister();
2328 	unregister_pernet_subsys(&nfs_net_ops);
2329 	rpc_proc_unregister(&init_net, "nfs");
2330 	unregister_nfs_fs();
2331 	nfs_fs_proc_exit();
2332 	nfsiod_stop();
2333 	nfs_sysfs_exit();
2334 }
2335 
2336 /* Not quite true; I just maintain it */
2337 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2338 MODULE_LICENSE("GPL");
2339 module_param(enable_ino64, bool, 0644);
2340 
2341 module_init(init_nfs_fs)
2342 module_exit(exit_nfs_fs)
2343