?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mm: numa: Do not mark PTEs pte_numa when splitting huge pages
[linux/fpc-iii.git] / fs / nfs / inode.c
blob9927913c97c2eb80909d7e4a06f033046f1bcdc5
1 /*
2 * linux/fs/nfs/inode.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs inode and superblock handling functions
7 *
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
10 *
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
13 *
14 */
15
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
41
42 #include <asm/uaccess.h>
43
44 #include "nfs4_fs.h"
45 #include "callback.h"
46 #include "delegation.h"
47 #include "iostat.h"
48 #include "internal.h"
49 #include "fscache.h"
50 #include "pnfs.h"
51 #include "nfs.h"
52 #include "netns.h"
53
54 #include "nfstrace.h"
55
56 #define NFSDBG_FACILITY NFSDBG_VFS
57
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
59
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65
66 static struct kmem_cache * nfs_inode_cachep;
67
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73
74 /**
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
77 */
78 int nfs_wait_bit_killable(void *word)
79 {
80 if (fatal_signal_pending(current))
81 return -ERESTARTSYS;
82 freezable_schedule_unsafe();
83 return 0;
84 }
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
86
87 /**
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
90 *
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
93 */
94 u64 nfs_compat_user_ino64(u64 fileid)
95 {
96 #ifdef CONFIG_COMPAT
97 compat_ulong_t ino;
98 #else
99 unsigned long ino;
100 #endif
101
102 if (enable_ino64)
103 return fileid;
104 ino = fileid;
105 if (sizeof(ino) < sizeof(fileid))
106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
107 return ino;
108 }
109
110 int nfs_drop_inode(struct inode *inode)
111 {
112 return NFS_STALE(inode) || generic_drop_inode(inode);
113 }
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
115
116 void nfs_clear_inode(struct inode *inode)
117 {
118 /*
119 * The following should never happen...
120 */
121 WARN_ON_ONCE(nfs_have_writebacks(inode));
122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 nfs_zap_acl_cache(inode);
124 nfs_access_zap_cache(inode);
125 nfs_fscache_clear_inode(inode);
126 }
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
128
129 void nfs_evict_inode(struct inode *inode)
130 {
131 truncate_inode_pages_final(&inode->i_data);
132 clear_inode(inode);
133 nfs_clear_inode(inode);
134 }
135
136 /**
137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
138 */
139 int nfs_sync_mapping(struct address_space *mapping)
140 {
141 int ret = 0;
142
143 if (mapping->nrpages != 0) {
144 unmap_mapping_range(mapping, 0, 0, 0);
145 ret = nfs_wb_all(mapping->host);
146 }
147 return ret;
148 }
149
150 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
151 {
152 struct nfs_inode *nfsi = NFS_I(inode);
153
154 if (inode->i_mapping->nrpages == 0)
155 flags &= ~NFS_INO_INVALID_DATA;
156 nfsi->cache_validity |= flags;
157 if (flags & NFS_INO_INVALID_DATA)
158 nfs_fscache_invalidate(inode);
159 }
160
161 /*
162 * Invalidate the local caches
163 */
164 static void nfs_zap_caches_locked(struct inode *inode)
165 {
166 struct nfs_inode *nfsi = NFS_I(inode);
167 int mode = inode->i_mode;
168
169 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
170
171 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
172 nfsi->attrtimeo_timestamp = jiffies;
173
174 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
175 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
176 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
177 | NFS_INO_INVALID_DATA
178 | NFS_INO_INVALID_ACCESS
179 | NFS_INO_INVALID_ACL
180 | NFS_INO_REVAL_PAGECACHE);
181 } else
182 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
183 | NFS_INO_INVALID_ACCESS
184 | NFS_INO_INVALID_ACL
185 | NFS_INO_REVAL_PAGECACHE);
186 nfs_zap_label_cache_locked(nfsi);
187 }
188
189 void nfs_zap_caches(struct inode *inode)
190 {
191 spin_lock(&inode->i_lock);
192 nfs_zap_caches_locked(inode);
193 spin_unlock(&inode->i_lock);
194 }
195
196 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
197 {
198 if (mapping->nrpages != 0) {
199 spin_lock(&inode->i_lock);
200 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
201 spin_unlock(&inode->i_lock);
202 }
203 }
204
205 void nfs_zap_acl_cache(struct inode *inode)
206 {
207 void (*clear_acl_cache)(struct inode *);
208
209 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
210 if (clear_acl_cache != NULL)
211 clear_acl_cache(inode);
212 spin_lock(&inode->i_lock);
213 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
214 spin_unlock(&inode->i_lock);
215 }
216 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
217
218 void nfs_invalidate_atime(struct inode *inode)
219 {
220 spin_lock(&inode->i_lock);
221 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
222 spin_unlock(&inode->i_lock);
223 }
224 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
225
226 /*
227 * Invalidate, but do not unhash, the inode.
228 * NB: must be called with inode->i_lock held!
229 */
230 static void nfs_invalidate_inode(struct inode *inode)
231 {
232 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
233 nfs_zap_caches_locked(inode);
234 }
235
236 struct nfs_find_desc {
237 struct nfs_fh *fh;
238 struct nfs_fattr *fattr;
239 };
240
241 /*
242 * In NFSv3 we can have 64bit inode numbers. In order to support
243 * this, and re-exported directories (also seen in NFSv2)
244 * we are forced to allow 2 different inodes to have the same
245 * i_ino.
246 */
247 static int
248 nfs_find_actor(struct inode *inode, void *opaque)
249 {
250 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
251 struct nfs_fh *fh = desc->fh;
252 struct nfs_fattr *fattr = desc->fattr;
253
254 if (NFS_FILEID(inode) != fattr->fileid)
255 return 0;
256 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
257 return 0;
258 if (nfs_compare_fh(NFS_FH(inode), fh))
259 return 0;
260 if (is_bad_inode(inode) || NFS_STALE(inode))
261 return 0;
262 return 1;
263 }
264
265 static int
266 nfs_init_locked(struct inode *inode, void *opaque)
267 {
268 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
269 struct nfs_fattr *fattr = desc->fattr;
270
271 set_nfs_fileid(inode, fattr->fileid);
272 nfs_copy_fh(NFS_FH(inode), desc->fh);
273 return 0;
274 }
275
276 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
277 static void nfs_clear_label_invalid(struct inode *inode)
278 {
279 spin_lock(&inode->i_lock);
280 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
281 spin_unlock(&inode->i_lock);
282 }
283
284 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
285 struct nfs4_label *label)
286 {
287 int error;
288
289 if (label == NULL)
290 return;
291
292 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
293 error = security_inode_notifysecctx(inode, label->label,
294 label->len);
295 if (error)
296 printk(KERN_ERR "%s() %s %d "
297 "security_inode_notifysecctx() %d\n",
298 __func__,
299 (char *)label->label,
300 label->len, error);
301 nfs_clear_label_invalid(inode);
302 }
303 }
304
305 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
306 {
307 struct nfs4_label *label = NULL;
308 int minor_version = server->nfs_client->cl_minorversion;
309
310 if (minor_version < 2)
311 return label;
312
313 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
314 return label;
315
316 label = kzalloc(sizeof(struct nfs4_label), flags);
317 if (label == NULL)
318 return ERR_PTR(-ENOMEM);
319
320 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
321 if (label->label == NULL) {
322 kfree(label);
323 return ERR_PTR(-ENOMEM);
324 }
325 label->len = NFS4_MAXLABELLEN;
326
327 return label;
328 }
329 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
330 #else
331 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
332 struct nfs4_label *label)
333 {
334 }
335 #endif
336 EXPORT_SYMBOL_GPL(nfs_setsecurity);
337
338 /*
339 * This is our front-end to iget that looks up inodes by file handle
340 * instead of inode number.
341 */
342 struct inode *
343 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
344 {
345 struct nfs_find_desc desc = {
346 .fh = fh,
347 .fattr = fattr
348 };
349 struct inode *inode = ERR_PTR(-ENOENT);
350 unsigned long hash;
351
352 nfs_attr_check_mountpoint(sb, fattr);
353
354 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
355 !nfs_attr_use_mounted_on_fileid(fattr))
356 goto out_no_inode;
357 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
358 goto out_no_inode;
359
360 hash = nfs_fattr_to_ino_t(fattr);
361
362 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
363 if (inode == NULL) {
364 inode = ERR_PTR(-ENOMEM);
365 goto out_no_inode;
366 }
367
368 if (inode->i_state & I_NEW) {
369 struct nfs_inode *nfsi = NFS_I(inode);
370 unsigned long now = jiffies;
371
372 /* We set i_ino for the few things that still rely on it,
373 * such as stat(2) */
374 inode->i_ino = hash;
375
376 /* We can't support update_atime(), since the server will reset it */
377 inode->i_flags |= S_NOATIME|S_NOCMTIME;
378 inode->i_mode = fattr->mode;
379 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
380 && nfs_server_capable(inode, NFS_CAP_MODE))
381 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
382 /* Why so? Because we want revalidate for devices/FIFOs, and
383 * that's precisely what we have in nfs_file_inode_operations.
384 */
385 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
386 if (S_ISREG(inode->i_mode)) {
387 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
388 inode->i_data.a_ops = &nfs_file_aops;
389 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
390 } else if (S_ISDIR(inode->i_mode)) {
391 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
392 inode->i_fop = &nfs_dir_operations;
393 inode->i_data.a_ops = &nfs_dir_aops;
394 /* Deal with crossing mountpoints */
395 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
396 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
397 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
398 inode->i_op = &nfs_referral_inode_operations;
399 else
400 inode->i_op = &nfs_mountpoint_inode_operations;
401 inode->i_fop = NULL;
402 inode->i_flags |= S_AUTOMOUNT;
403 }
404 } else if (S_ISLNK(inode->i_mode))
405 inode->i_op = &nfs_symlink_inode_operations;
406 else
407 init_special_inode(inode, inode->i_mode, fattr->rdev);
408
409 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
410 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
411 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
412 inode->i_version = 0;
413 inode->i_size = 0;
414 clear_nlink(inode);
415 inode->i_uid = make_kuid(&init_user_ns, -2);
416 inode->i_gid = make_kgid(&init_user_ns, -2);
417 inode->i_blocks = 0;
418 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
419 nfsi->write_io = 0;
420 nfsi->read_io = 0;
421
422 nfsi->read_cache_jiffies = fattr->time_start;
423 nfsi->attr_gencount = fattr->gencount;
424 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
425 inode->i_atime = fattr->atime;
426 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
427 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
428 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
429 inode->i_mtime = fattr->mtime;
430 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
431 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
432 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
433 inode->i_ctime = fattr->ctime;
434 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
435 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
436 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
437 inode->i_version = fattr->change_attr;
438 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
439 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
440 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
441 inode->i_size = nfs_size_to_loff_t(fattr->size);
442 else
443 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
444 | NFS_INO_REVAL_PAGECACHE);
445 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
446 set_nlink(inode, fattr->nlink);
447 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
448 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
449 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
450 inode->i_uid = fattr->uid;
451 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
452 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
453 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
454 inode->i_gid = fattr->gid;
455 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
456 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
457 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
458 inode->i_blocks = fattr->du.nfs2.blocks;
459 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
460 /*
461 * report the blocks in 512byte units
462 */
463 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
464 }
465
466 nfs_setsecurity(inode, fattr, label);
467
468 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
469 nfsi->attrtimeo_timestamp = now;
470 nfsi->access_cache = RB_ROOT;
471
472 nfs_fscache_init_inode(inode);
473
474 unlock_new_inode(inode);
475 } else
476 nfs_refresh_inode(inode, fattr);
477 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
478 inode->i_sb->s_id,
479 (unsigned long long)NFS_FILEID(inode),
480 nfs_display_fhandle_hash(fh),
481 atomic_read(&inode->i_count));
482
483 out:
484 return inode;
485
486 out_no_inode:
487 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
488 goto out;
489 }
490 EXPORT_SYMBOL_GPL(nfs_fhget);
491
492 #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)
493
494 int
495 nfs_setattr(struct dentry *dentry, struct iattr *attr)
496 {
497 struct inode *inode = dentry->d_inode;
498 struct nfs_fattr *fattr;
499 int error = -ENOMEM;
500
501 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
502
503 /* skip mode change if it's just for clearing setuid/setgid */
504 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
505 attr->ia_valid &= ~ATTR_MODE;
506
507 if (attr->ia_valid & ATTR_SIZE) {
508 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
509 attr->ia_valid &= ~ATTR_SIZE;
510 }
511
512 /* Optimization: if the end result is no change, don't RPC */
513 attr->ia_valid &= NFS_VALID_ATTRS;
514 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
515 return 0;
516
517 trace_nfs_setattr_enter(inode);
518
519 /* Write all dirty data */
520 if (S_ISREG(inode->i_mode)) {
521 nfs_inode_dio_wait(inode);
522 nfs_wb_all(inode);
523 }
524
525 fattr = nfs_alloc_fattr();
526 if (fattr == NULL)
527 goto out;
528 /*
529 * Return any delegations if we're going to change ACLs
530 */
531 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
532 NFS_PROTO(inode)->return_delegation(inode);
533 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
534 if (error == 0)
535 error = nfs_refresh_inode(inode, fattr);
536 nfs_free_fattr(fattr);
537 out:
538 trace_nfs_setattr_exit(inode, error);
539 return error;
540 }
541 EXPORT_SYMBOL_GPL(nfs_setattr);
542
543 /**
544 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
545 * @inode: inode of the file used
546 * @offset: file offset to start truncating
547 *
548 * This is a copy of the common vmtruncate, but with the locking
549 * corrected to take into account the fact that NFS requires
550 * inode->i_size to be updated under the inode->i_lock.
551 */
552 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
553 {
554 int err;
555
556 err = inode_newsize_ok(inode, offset);
557 if (err)
558 goto out;
559
560 spin_lock(&inode->i_lock);
561 i_size_write(inode, offset);
562 /* Optimisation */
563 if (offset == 0)
564 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
565 spin_unlock(&inode->i_lock);
566
567 truncate_pagecache(inode, offset);
568 out:
569 return err;
570 }
571
572 /**
573 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
574 * @inode: pointer to struct inode
575 * @attr: pointer to struct iattr
576 *
577 * Note: we do this in the *proc.c in order to ensure that
578 * it works for things like exclusive creates too.
579 */
580 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
581 {
582 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
583 spin_lock(&inode->i_lock);
584 if ((attr->ia_valid & ATTR_MODE) != 0) {
585 int mode = attr->ia_mode & S_IALLUGO;
586 mode |= inode->i_mode & ~S_IALLUGO;
587 inode->i_mode = mode;
588 }
589 if ((attr->ia_valid & ATTR_UID) != 0)
590 inode->i_uid = attr->ia_uid;
591 if ((attr->ia_valid & ATTR_GID) != 0)
592 inode->i_gid = attr->ia_gid;
593 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
594 | NFS_INO_INVALID_ACL);
595 spin_unlock(&inode->i_lock);
596 }
597 if ((attr->ia_valid & ATTR_SIZE) != 0) {
598 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
599 nfs_vmtruncate(inode, attr->ia_size);
600 }
601 }
602 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
603
604 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
605 {
606 struct dentry *parent;
607
608 parent = dget_parent(dentry);
609 nfs_force_use_readdirplus(parent->d_inode);
610 dput(parent);
611 }
612
613 static bool nfs_need_revalidate_inode(struct inode *inode)
614 {
615 if (NFS_I(inode)->cache_validity &
616 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
617 return true;
618 if (nfs_attribute_cache_expired(inode))
619 return true;
620 return false;
621 }
622
623 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
624 {
625 struct inode *inode = dentry->d_inode;
626 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
627 int err;
628
629 trace_nfs_getattr_enter(inode);
630 /* Flush out writes to the server in order to update c/mtime. */
631 if (S_ISREG(inode->i_mode)) {
632 nfs_inode_dio_wait(inode);
633 err = filemap_write_and_wait(inode->i_mapping);
634 if (err)
635 goto out;
636 }
637
638 /*
639 * We may force a getattr if the user cares about atime.
640 *
641 * Note that we only have to check the vfsmount flags here:
642 * - NFS always sets S_NOATIME by so checking it would give a
643 * bogus result
644 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
645 * no point in checking those.
646 */
647 if ((mnt->mnt_flags & MNT_NOATIME) ||
648 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
649 need_atime = 0;
650
651 if (need_atime || nfs_need_revalidate_inode(inode)) {
652 struct nfs_server *server = NFS_SERVER(inode);
653
654 if (server->caps & NFS_CAP_READDIRPLUS)
655 nfs_request_parent_use_readdirplus(dentry);
656 err = __nfs_revalidate_inode(server, inode);
657 }
658 if (!err) {
659 generic_fillattr(inode, stat);
660 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
661 }
662 out:
663 trace_nfs_getattr_exit(inode, err);
664 return err;
665 }
666 EXPORT_SYMBOL_GPL(nfs_getattr);
667
668 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
669 {
670 atomic_set(&l_ctx->count, 1);
671 l_ctx->lockowner.l_owner = current->files;
672 l_ctx->lockowner.l_pid = current->tgid;
673 INIT_LIST_HEAD(&l_ctx->list);
674 nfs_iocounter_init(&l_ctx->io_count);
675 }
676
677 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
678 {
679 struct nfs_lock_context *head = &ctx->lock_context;
680 struct nfs_lock_context *pos = head;
681
682 do {
683 if (pos->lockowner.l_owner != current->files)
684 continue;
685 if (pos->lockowner.l_pid != current->tgid)
686 continue;
687 atomic_inc(&pos->count);
688 return pos;
689 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
690 return NULL;
691 }
692
693 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
694 {
695 struct nfs_lock_context *res, *new = NULL;
696 struct inode *inode = ctx->dentry->d_inode;
697
698 spin_lock(&inode->i_lock);
699 res = __nfs_find_lock_context(ctx);
700 if (res == NULL) {
701 spin_unlock(&inode->i_lock);
702 new = kmalloc(sizeof(*new), GFP_KERNEL);
703 if (new == NULL)
704 return ERR_PTR(-ENOMEM);
705 nfs_init_lock_context(new);
706 spin_lock(&inode->i_lock);
707 res = __nfs_find_lock_context(ctx);
708 if (res == NULL) {
709 list_add_tail(&new->list, &ctx->lock_context.list);
710 new->open_context = ctx;
711 res = new;
712 new = NULL;
713 }
714 }
715 spin_unlock(&inode->i_lock);
716 kfree(new);
717 return res;
718 }
719
720 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
721 {
722 struct nfs_open_context *ctx = l_ctx->open_context;
723 struct inode *inode = ctx->dentry->d_inode;
724
725 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
726 return;
727 list_del(&l_ctx->list);
728 spin_unlock(&inode->i_lock);
729 kfree(l_ctx);
730 }
731
732 /**
733 * nfs_close_context - Common close_context() routine NFSv2/v3
734 * @ctx: pointer to context
735 * @is_sync: is this a synchronous close
736 *
737 * always ensure that the attributes are up to date if we're mounted
738 * with close-to-open semantics
739 */
740 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
741 {
742 struct inode *inode;
743 struct nfs_server *server;
744
745 if (!(ctx->mode & FMODE_WRITE))
746 return;
747 if (!is_sync)
748 return;
749 inode = ctx->dentry->d_inode;
750 if (!list_empty(&NFS_I(inode)->open_files))
751 return;
752 server = NFS_SERVER(inode);
753 if (server->flags & NFS_MOUNT_NOCTO)
754 return;
755 nfs_revalidate_inode(server, inode);
756 }
757 EXPORT_SYMBOL_GPL(nfs_close_context);
758
759 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
760 {
761 struct nfs_open_context *ctx;
762 struct rpc_cred *cred = rpc_lookup_cred();
763 if (IS_ERR(cred))
764 return ERR_CAST(cred);
765
766 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
767 if (!ctx) {
768 put_rpccred(cred);
769 return ERR_PTR(-ENOMEM);
770 }
771 nfs_sb_active(dentry->d_sb);
772 ctx->dentry = dget(dentry);
773 ctx->cred = cred;
774 ctx->state = NULL;
775 ctx->mode = f_mode;
776 ctx->flags = 0;
777 ctx->error = 0;
778 nfs_init_lock_context(&ctx->lock_context);
779 ctx->lock_context.open_context = ctx;
780 INIT_LIST_HEAD(&ctx->list);
781 ctx->mdsthreshold = NULL;
782 return ctx;
783 }
784 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
785
786 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
787 {
788 if (ctx != NULL)
789 atomic_inc(&ctx->lock_context.count);
790 return ctx;
791 }
792 EXPORT_SYMBOL_GPL(get_nfs_open_context);
793
794 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
795 {
796 struct inode *inode = ctx->dentry->d_inode;
797 struct super_block *sb = ctx->dentry->d_sb;
798
799 if (!list_empty(&ctx->list)) {
800 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
801 return;
802 list_del(&ctx->list);
803 spin_unlock(&inode->i_lock);
804 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
805 return;
806 if (inode != NULL)
807 NFS_PROTO(inode)->close_context(ctx, is_sync);
808 if (ctx->cred != NULL)
809 put_rpccred(ctx->cred);
810 dput(ctx->dentry);
811 nfs_sb_deactive(sb);
812 kfree(ctx->mdsthreshold);
813 kfree(ctx);
814 }
815
816 void put_nfs_open_context(struct nfs_open_context *ctx)
817 {
818 __put_nfs_open_context(ctx, 0);
819 }
820 EXPORT_SYMBOL_GPL(put_nfs_open_context);
821
822 /*
823 * Ensure that mmap has a recent RPC credential for use when writing out
824 * shared pages
825 */
826 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
827 {
828 struct inode *inode = ctx->dentry->d_inode;
829 struct nfs_inode *nfsi = NFS_I(inode);
830
831 spin_lock(&inode->i_lock);
832 list_add(&ctx->list, &nfsi->open_files);
833 spin_unlock(&inode->i_lock);
834 }
835 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
836
837 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
838 {
839 filp->private_data = get_nfs_open_context(ctx);
840 if (list_empty(&ctx->list))
841 nfs_inode_attach_open_context(ctx);
842 }
843 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
844
845 /*
846 * Given an inode, search for an open context with the desired characteristics
847 */
848 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
849 {
850 struct nfs_inode *nfsi = NFS_I(inode);
851 struct nfs_open_context *pos, *ctx = NULL;
852
853 spin_lock(&inode->i_lock);
854 list_for_each_entry(pos, &nfsi->open_files, list) {
855 if (cred != NULL && pos->cred != cred)
856 continue;
857 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
858 continue;
859 ctx = get_nfs_open_context(pos);
860 break;
861 }
862 spin_unlock(&inode->i_lock);
863 return ctx;
864 }
865
866 static void nfs_file_clear_open_context(struct file *filp)
867 {
868 struct nfs_open_context *ctx = nfs_file_open_context(filp);
869
870 if (ctx) {
871 struct inode *inode = ctx->dentry->d_inode;
872
873 filp->private_data = NULL;
874 spin_lock(&inode->i_lock);
875 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
876 spin_unlock(&inode->i_lock);
877 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
878 }
879 }
880
881 /*
882 * These allocate and release file read/write context information.
883 */
884 int nfs_open(struct inode *inode, struct file *filp)
885 {
886 struct nfs_open_context *ctx;
887
888 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
889 if (IS_ERR(ctx))
890 return PTR_ERR(ctx);
891 nfs_file_set_open_context(filp, ctx);
892 put_nfs_open_context(ctx);
893 nfs_fscache_open_file(inode, filp);
894 return 0;
895 }
896
897 int nfs_release(struct inode *inode, struct file *filp)
898 {
899 nfs_file_clear_open_context(filp);
900 return 0;
901 }
902
903 /*
904 * This function is called whenever some part of NFS notices that
905 * the cached attributes have to be refreshed.
906 */
907 int
908 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
909 {
910 int status = -ESTALE;
911 struct nfs4_label *label = NULL;
912 struct nfs_fattr *fattr = NULL;
913 struct nfs_inode *nfsi = NFS_I(inode);
914
915 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
916 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
917
918 trace_nfs_revalidate_inode_enter(inode);
919
920 if (is_bad_inode(inode))
921 goto out;
922 if (NFS_STALE(inode))
923 goto out;
924
925 status = -ENOMEM;
926 fattr = nfs_alloc_fattr();
927 if (fattr == NULL)
928 goto out;
929
930 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
931
932 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
933 if (IS_ERR(label)) {
934 status = PTR_ERR(label);
935 goto out;
936 }
937
938 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
939 if (status != 0) {
940 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
941 inode->i_sb->s_id,
942 (unsigned long long)NFS_FILEID(inode), status);
943 if (status == -ESTALE) {
944 nfs_zap_caches(inode);
945 if (!S_ISDIR(inode->i_mode))
946 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
947 }
948 goto err_out;
949 }
950
951 status = nfs_refresh_inode(inode, fattr);
952 if (status) {
953 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
954 inode->i_sb->s_id,
955 (unsigned long long)NFS_FILEID(inode), status);
956 goto err_out;
957 }
958
959 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
960 nfs_zap_acl_cache(inode);
961
962 nfs_setsecurity(inode, fattr, label);
963
964 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
965 inode->i_sb->s_id,
966 (unsigned long long)NFS_FILEID(inode));
967
968 err_out:
969 nfs4_label_free(label);
970 out:
971 nfs_free_fattr(fattr);
972 trace_nfs_revalidate_inode_exit(inode, status);
973 return status;
974 }
975
976 int nfs_attribute_timeout(struct inode *inode)
977 {
978 struct nfs_inode *nfsi = NFS_I(inode);
979
980 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
981 }
982
983 int nfs_attribute_cache_expired(struct inode *inode)
984 {
985 if (nfs_have_delegated_attributes(inode))
986 return 0;
987 return nfs_attribute_timeout(inode);
988 }
989
990 /**
991 * nfs_revalidate_inode - Revalidate the inode attributes
992 * @server - pointer to nfs_server struct
993 * @inode - pointer to inode struct
994 *
995 * Updates inode attribute information by retrieving the data from the server.
996 */
997 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
998 {
999 if (!nfs_need_revalidate_inode(inode))
1000 return NFS_STALE(inode) ? -ESTALE : 0;
1001 return __nfs_revalidate_inode(server, inode);
1002 }
1003 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1004
1005 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1006 {
1007 struct nfs_inode *nfsi = NFS_I(inode);
1008 int ret;
1009
1010 if (mapping->nrpages != 0) {
1011 if (S_ISREG(inode->i_mode)) {
1012 ret = nfs_sync_mapping(mapping);
1013 if (ret < 0)
1014 return ret;
1015 }
1016 ret = invalidate_inode_pages2(mapping);
1017 if (ret < 0)
1018 return ret;
1019 }
1020 if (S_ISDIR(inode->i_mode)) {
1021 spin_lock(&inode->i_lock);
1022 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1023 spin_unlock(&inode->i_lock);
1024 }
1025 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1026 nfs_fscache_wait_on_invalidate(inode);
1027
1028 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1029 inode->i_sb->s_id,
1030 (unsigned long long)NFS_FILEID(inode));
1031 return 0;
1032 }
1033
1034 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1035 {
1036 if (nfs_have_delegated_attributes(inode))
1037 return false;
1038 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1039 || nfs_attribute_timeout(inode)
1040 || NFS_STALE(inode);
1041 }
1042
1043 /**
1044 * nfs_revalidate_mapping - Revalidate the pagecache
1045 * @inode - pointer to host inode
1046 * @mapping - pointer to mapping
1047 */
1048 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1049 {
1050 struct nfs_inode *nfsi = NFS_I(inode);
1051 unsigned long *bitlock = &nfsi->flags;
1052 int ret = 0;
1053
1054 /* swapfiles are not supposed to be shared. */
1055 if (IS_SWAPFILE(inode))
1056 goto out;
1057
1058 if (nfs_mapping_need_revalidate_inode(inode)) {
1059 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1060 if (ret < 0)
1061 goto out;
1062 }
1063
1064 /*
1065 * We must clear NFS_INO_INVALID_DATA first to ensure that
1066 * invalidations that come in while we're shooting down the mappings
1067 * are respected. But, that leaves a race window where one revalidator
1068 * can clear the flag, and then another checks it before the mapping
1069 * gets invalidated. Fix that by serializing access to this part of
1070 * the function.
1071 *
1072 * At the same time, we need to allow other tasks to see whether we
1073 * might be in the middle of invalidating the pages, so we only set
1074 * the bit lock here if it looks like we're going to be doing that.
1075 */
1076 for (;;) {
1077 ret = wait_on_bit(bitlock, NFS_INO_INVALIDATING,
1078 nfs_wait_bit_killable, TASK_KILLABLE);
1079 if (ret)
1080 goto out;
1081 spin_lock(&inode->i_lock);
1082 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1083 spin_unlock(&inode->i_lock);
1084 continue;
1085 }
1086 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1087 break;
1088 spin_unlock(&inode->i_lock);
1089 goto out;
1090 }
1091
1092 set_bit(NFS_INO_INVALIDATING, bitlock);
1093 smp_wmb();
1094 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1095 spin_unlock(&inode->i_lock);
1096 trace_nfs_invalidate_mapping_enter(inode);
1097 ret = nfs_invalidate_mapping(inode, mapping);
1098 trace_nfs_invalidate_mapping_exit(inode, ret);
1099
1100 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1101 smp_mb__after_atomic();
1102 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1103 out:
1104 return ret;
1105 }
1106
1107 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1108 {
1109 struct nfs_inode *nfsi = NFS_I(inode);
1110 unsigned long ret = 0;
1111
1112 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1113 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1114 && inode->i_version == fattr->pre_change_attr) {
1115 inode->i_version = fattr->change_attr;
1116 if (S_ISDIR(inode->i_mode))
1117 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1118 ret |= NFS_INO_INVALID_ATTR;
1119 }
1120 /* If we have atomic WCC data, we may update some attributes */
1121 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1122 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1123 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1124 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1125 ret |= NFS_INO_INVALID_ATTR;
1126 }
1127
1128 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1129 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1130 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1131 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1132 if (S_ISDIR(inode->i_mode))
1133 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1134 ret |= NFS_INO_INVALID_ATTR;
1135 }
1136 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1137 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1138 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1139 && nfsi->npages == 0) {
1140 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1141 ret |= NFS_INO_INVALID_ATTR;
1142 }
1143
1144 return ret;
1145 }
1146
1147 /**
1148 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1149 * @inode - pointer to inode
1150 * @fattr - updated attributes
1151 *
1152 * Verifies the attribute cache. If we have just changed the attributes,
1153 * so that fattr carries weak cache consistency data, then it may
1154 * also update the ctime/mtime/change_attribute.
1155 */
1156 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1157 {
1158 struct nfs_inode *nfsi = NFS_I(inode);
1159 loff_t cur_size, new_isize;
1160 unsigned long invalid = 0;
1161
1162
1163 if (nfs_have_delegated_attributes(inode))
1164 return 0;
1165 /* Has the inode gone and changed behind our back? */
1166 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1167 return -EIO;
1168 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1169 return -EIO;
1170
1171 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1172 inode->i_version != fattr->change_attr)
1173 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1174
1175 /* Verify a few of the more important attributes */
1176 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1177 invalid |= NFS_INO_INVALID_ATTR;
1178
1179 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1180 cur_size = i_size_read(inode);
1181 new_isize = nfs_size_to_loff_t(fattr->size);
1182 if (cur_size != new_isize && nfsi->npages == 0)
1183 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1184 }
1185
1186 /* Have any file permissions changed? */
1187 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1188 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1189 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1190 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1191 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1192 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1193
1194 /* Has the link count changed? */
1195 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1196 invalid |= NFS_INO_INVALID_ATTR;
1197
1198 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1199 invalid |= NFS_INO_INVALID_ATIME;
1200
1201 if (invalid != 0)
1202 nfs_set_cache_invalid(inode, invalid);
1203
1204 nfsi->read_cache_jiffies = fattr->time_start;
1205 return 0;
1206 }
1207
1208 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1209 {
1210 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1211 return 0;
1212 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1213 }
1214
1215 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1216 {
1217 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1218 return 0;
1219 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1220 }
1221
1222 static atomic_long_t nfs_attr_generation_counter;
1223
1224 static unsigned long nfs_read_attr_generation_counter(void)
1225 {
1226 return atomic_long_read(&nfs_attr_generation_counter);
1227 }
1228
1229 unsigned long nfs_inc_attr_generation_counter(void)
1230 {
1231 return atomic_long_inc_return(&nfs_attr_generation_counter);
1232 }
1233
1234 void nfs_fattr_init(struct nfs_fattr *fattr)
1235 {
1236 fattr->valid = 0;
1237 fattr->time_start = jiffies;
1238 fattr->gencount = nfs_inc_attr_generation_counter();
1239 fattr->owner_name = NULL;
1240 fattr->group_name = NULL;
1241 }
1242 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1243
1244 struct nfs_fattr *nfs_alloc_fattr(void)
1245 {
1246 struct nfs_fattr *fattr;
1247
1248 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1249 if (fattr != NULL)
1250 nfs_fattr_init(fattr);
1251 return fattr;
1252 }
1253 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1254
1255 struct nfs_fh *nfs_alloc_fhandle(void)
1256 {
1257 struct nfs_fh *fh;
1258
1259 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1260 if (fh != NULL)
1261 fh->size = 0;
1262 return fh;
1263 }
1264 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1265
1266 #ifdef NFS_DEBUG
1267 /*
1268 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1269 * in the same way that wireshark does
1270 *
1271 * @fh: file handle
1272 *
1273 * For debugging only.
1274 */
1275 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1276 {
1277 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1278 * not on the result */
1279 return nfs_fhandle_hash(fh);
1280 }
1281 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1282
1283 /*
1284 * _nfs_display_fhandle - display an NFS file handle on the console
1285 *
1286 * @fh: file handle to display
1287 * @caption: display caption
1288 *
1289 * For debugging only.
1290 */
1291 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1292 {
1293 unsigned short i;
1294
1295 if (fh == NULL || fh->size == 0) {
1296 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1297 return;
1298 }
1299
1300 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1301 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1302 for (i = 0; i < fh->size; i += 16) {
1303 __be32 *pos = (__be32 *)&fh->data[i];
1304
1305 switch ((fh->size - i - 1) >> 2) {
1306 case 0:
1307 printk(KERN_DEFAULT " %08x\n",
1308 be32_to_cpup(pos));
1309 break;
1310 case 1:
1311 printk(KERN_DEFAULT " %08x %08x\n",
1312 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1313 break;
1314 case 2:
1315 printk(KERN_DEFAULT " %08x %08x %08x\n",
1316 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1317 be32_to_cpup(pos + 2));
1318 break;
1319 default:
1320 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1321 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1322 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1323 }
1324 }
1325 }
1326 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1327 #endif
1328
1329 /**
1330 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1331 * @inode - pointer to inode
1332 * @fattr - attributes
1333 *
1334 * Attempt to divine whether or not an RPC call reply carrying stale
1335 * attributes got scheduled after another call carrying updated ones.
1336 *
1337 * To do so, the function first assumes that a more recent ctime means
1338 * that the attributes in fattr are newer, however it also attempt to
1339 * catch the case where ctime either didn't change, or went backwards
1340 * (if someone reset the clock on the server) by looking at whether
1341 * or not this RPC call was started after the inode was last updated.
1342 * Note also the check for wraparound of 'attr_gencount'
1343 *
1344 * The function returns 'true' if it thinks the attributes in 'fattr' are
1345 * more recent than the ones cached in the inode.
1346 *
1347 */
1348 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1349 {
1350 const struct nfs_inode *nfsi = NFS_I(inode);
1351
1352 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1353 nfs_ctime_need_update(inode, fattr) ||
1354 nfs_size_need_update(inode, fattr) ||
1355 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1356 }
1357
1358 /*
1359 * Don't trust the change_attribute, mtime, ctime or size if
1360 * a pnfs LAYOUTCOMMIT is outstanding
1361 */
1362 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1363 struct nfs_fattr *fattr)
1364 {
1365 if (pnfs_layoutcommit_outstanding(inode))
1366 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1367 NFS_ATTR_FATTR_MTIME |
1368 NFS_ATTR_FATTR_CTIME |
1369 NFS_ATTR_FATTR_SIZE);
1370 }
1371
1372 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1373 {
1374 int ret;
1375
1376 trace_nfs_refresh_inode_enter(inode);
1377
1378 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1379
1380 if (nfs_inode_attrs_need_update(inode, fattr))
1381 ret = nfs_update_inode(inode, fattr);
1382 else
1383 ret = nfs_check_inode_attributes(inode, fattr);
1384
1385 trace_nfs_refresh_inode_exit(inode, ret);
1386 return ret;
1387 }
1388
1389 /**
1390 * nfs_refresh_inode - try to update the inode attribute cache
1391 * @inode - pointer to inode
1392 * @fattr - updated attributes
1393 *
1394 * Check that an RPC call that returned attributes has not overlapped with
1395 * other recent updates of the inode metadata, then decide whether it is
1396 * safe to do a full update of the inode attributes, or whether just to
1397 * call nfs_check_inode_attributes.
1398 */
1399 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1400 {
1401 int status;
1402
1403 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1404 return 0;
1405 spin_lock(&inode->i_lock);
1406 status = nfs_refresh_inode_locked(inode, fattr);
1407 spin_unlock(&inode->i_lock);
1408
1409 return status;
1410 }
1411 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1412
1413 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1414 {
1415 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1416
1417 if (S_ISDIR(inode->i_mode))
1418 invalid |= NFS_INO_INVALID_DATA;
1419 nfs_set_cache_invalid(inode, invalid);
1420 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1421 return 0;
1422 return nfs_refresh_inode_locked(inode, fattr);
1423 }
1424
1425 /**
1426 * nfs_post_op_update_inode - try to update the inode attribute cache
1427 * @inode - pointer to inode
1428 * @fattr - updated attributes
1429 *
1430 * After an operation that has changed the inode metadata, mark the
1431 * attribute cache as being invalid, then try to update it.
1432 *
1433 * NB: if the server didn't return any post op attributes, this
1434 * function will force the retrieval of attributes before the next
1435 * NFS request. Thus it should be used only for operations that
1436 * are expected to change one or more attributes, to avoid
1437 * unnecessary NFS requests and trips through nfs_update_inode().
1438 */
1439 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1440 {
1441 int status;
1442
1443 spin_lock(&inode->i_lock);
1444 status = nfs_post_op_update_inode_locked(inode, fattr);
1445 spin_unlock(&inode->i_lock);
1446
1447 return status;
1448 }
1449 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1450
1451 /**
1452 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1453 * @inode - pointer to inode
1454 * @fattr - updated attributes
1455 *
1456 * After an operation that has changed the inode metadata, mark the
1457 * attribute cache as being invalid, then try to update it. Fake up
1458 * weak cache consistency data, if none exist.
1459 *
1460 * This function is mainly designed to be used by the ->write_done() functions.
1461 */
1462 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1463 {
1464 int status;
1465
1466 spin_lock(&inode->i_lock);
1467 /* Don't do a WCC update if these attributes are already stale */
1468 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1469 !nfs_inode_attrs_need_update(inode, fattr)) {
1470 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1471 | NFS_ATTR_FATTR_PRESIZE
1472 | NFS_ATTR_FATTR_PREMTIME
1473 | NFS_ATTR_FATTR_PRECTIME);
1474 goto out_noforce;
1475 }
1476 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1477 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1478 fattr->pre_change_attr = inode->i_version;
1479 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1480 }
1481 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1482 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1483 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1484 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1485 }
1486 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1487 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1488 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1489 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1490 }
1491 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1492 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1493 fattr->pre_size = i_size_read(inode);
1494 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1495 }
1496 out_noforce:
1497 status = nfs_post_op_update_inode_locked(inode, fattr);
1498 spin_unlock(&inode->i_lock);
1499 return status;
1500 }
1501 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1502
1503 /*
1504 * Many nfs protocol calls return the new file attributes after
1505 * an operation. Here we update the inode to reflect the state
1506 * of the server's inode.
1507 *
1508 * This is a bit tricky because we have to make sure all dirty pages
1509 * have been sent off to the server before calling invalidate_inode_pages.
1510 * To make sure no other process adds more write requests while we try
1511 * our best to flush them, we make them sleep during the attribute refresh.
1512 *
1513 * A very similar scenario holds for the dir cache.
1514 */
1515 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1516 {
1517 struct nfs_server *server;
1518 struct nfs_inode *nfsi = NFS_I(inode);
1519 loff_t cur_isize, new_isize;
1520 unsigned long invalid = 0;
1521 unsigned long now = jiffies;
1522 unsigned long save_cache_validity;
1523
1524 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1525 __func__, inode->i_sb->s_id, inode->i_ino,
1526 nfs_display_fhandle_hash(NFS_FH(inode)),
1527 atomic_read(&inode->i_count), fattr->valid);
1528
1529 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1530 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1531 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1532 NFS_SERVER(inode)->nfs_client->cl_hostname,
1533 inode->i_sb->s_id, (long long)nfsi->fileid,
1534 (long long)fattr->fileid);
1535 goto out_err;
1536 }
1537
1538 /*
1539 * Make sure the inode's type hasn't changed.
1540 */
1541 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1542 /*
1543 * Big trouble! The inode has become a different object.
1544 */
1545 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1546 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1547 goto out_err;
1548 }
1549
1550 server = NFS_SERVER(inode);
1551 /* Update the fsid? */
1552 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1553 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1554 !IS_AUTOMOUNT(inode))
1555 server->fsid = fattr->fsid;
1556
1557 /*
1558 * Update the read time so we don't revalidate too often.
1559 */
1560 nfsi->read_cache_jiffies = fattr->time_start;
1561
1562 save_cache_validity = nfsi->cache_validity;
1563 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1564 | NFS_INO_INVALID_ATIME
1565 | NFS_INO_REVAL_FORCED
1566 | NFS_INO_REVAL_PAGECACHE);
1567
1568 /* Do atomic weak cache consistency updates */
1569 invalid |= nfs_wcc_update_inode(inode, fattr);
1570
1571 /* More cache consistency checks */
1572 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1573 if (inode->i_version != fattr->change_attr) {
1574 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1575 inode->i_sb->s_id, inode->i_ino);
1576 invalid |= NFS_INO_INVALID_ATTR
1577 | NFS_INO_INVALID_DATA
1578 | NFS_INO_INVALID_ACCESS
1579 | NFS_INO_INVALID_ACL
1580 | NFS_INO_REVAL_PAGECACHE;
1581 if (S_ISDIR(inode->i_mode))
1582 nfs_force_lookup_revalidate(inode);
1583 inode->i_version = fattr->change_attr;
1584 }
1585 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1586 nfsi->cache_validity |= save_cache_validity;
1587
1588 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1589 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1590 } else if (server->caps & NFS_CAP_MTIME)
1591 nfsi->cache_validity |= save_cache_validity &
1592 (NFS_INO_INVALID_ATTR
1593 | NFS_INO_REVAL_FORCED);
1594
1595 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1596 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1597 } else if (server->caps & NFS_CAP_CTIME)
1598 nfsi->cache_validity |= save_cache_validity &
1599 (NFS_INO_INVALID_ATTR
1600 | NFS_INO_REVAL_FORCED);
1601
1602 /* Check if our cached file size is stale */
1603 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1604 new_isize = nfs_size_to_loff_t(fattr->size);
1605 cur_isize = i_size_read(inode);
1606 if (new_isize != cur_isize) {
1607 /* Do we perhaps have any outstanding writes, or has
1608 * the file grown beyond our last write? */
1609 if ((nfsi->npages == 0) || new_isize > cur_isize) {
1610 i_size_write(inode, new_isize);
1611 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1612 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1613 }
1614 dprintk("NFS: isize change on server for file %s/%ld "
1615 "(%Ld to %Ld)\n",
1616 inode->i_sb->s_id,
1617 inode->i_ino,
1618 (long long)cur_isize,
1619 (long long)new_isize);
1620 }
1621 } else
1622 nfsi->cache_validity |= save_cache_validity &
1623 (NFS_INO_INVALID_ATTR
1624 | NFS_INO_REVAL_PAGECACHE
1625 | NFS_INO_REVAL_FORCED);
1626
1627
1628 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1629 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1630 else if (server->caps & NFS_CAP_ATIME)
1631 nfsi->cache_validity |= save_cache_validity &
1632 (NFS_INO_INVALID_ATIME
1633 | NFS_INO_REVAL_FORCED);
1634
1635 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1636 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1637 umode_t newmode = inode->i_mode & S_IFMT;
1638 newmode |= fattr->mode & S_IALLUGO;
1639 inode->i_mode = newmode;
1640 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1641 }
1642 } else if (server->caps & NFS_CAP_MODE)
1643 nfsi->cache_validity |= save_cache_validity &
1644 (NFS_INO_INVALID_ATTR
1645 | NFS_INO_INVALID_ACCESS
1646 | NFS_INO_INVALID_ACL
1647 | NFS_INO_REVAL_FORCED);
1648
1649 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1650 if (!uid_eq(inode->i_uid, fattr->uid)) {
1651 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1652 inode->i_uid = fattr->uid;
1653 }
1654 } else if (server->caps & NFS_CAP_OWNER)
1655 nfsi->cache_validity |= save_cache_validity &
1656 (NFS_INO_INVALID_ATTR
1657 | NFS_INO_INVALID_ACCESS
1658 | NFS_INO_INVALID_ACL
1659 | NFS_INO_REVAL_FORCED);
1660
1661 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1662 if (!gid_eq(inode->i_gid, fattr->gid)) {
1663 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1664 inode->i_gid = fattr->gid;
1665 }
1666 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1667 nfsi->cache_validity |= save_cache_validity &
1668 (NFS_INO_INVALID_ATTR
1669 | NFS_INO_INVALID_ACCESS
1670 | NFS_INO_INVALID_ACL
1671 | NFS_INO_REVAL_FORCED);
1672
1673 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1674 if (inode->i_nlink != fattr->nlink) {
1675 invalid |= NFS_INO_INVALID_ATTR;
1676 if (S_ISDIR(inode->i_mode))
1677 invalid |= NFS_INO_INVALID_DATA;
1678 set_nlink(inode, fattr->nlink);
1679 }
1680 } else if (server->caps & NFS_CAP_NLINK)
1681 nfsi->cache_validity |= save_cache_validity &
1682 (NFS_INO_INVALID_ATTR
1683 | NFS_INO_REVAL_FORCED);
1684
1685 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1686 /*
1687 * report the blocks in 512byte units
1688 */
1689 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1690 }
1691 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1692 inode->i_blocks = fattr->du.nfs2.blocks;
1693
1694 /* Update attrtimeo value if we're out of the unstable period */
1695 if (invalid & NFS_INO_INVALID_ATTR) {
1696 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1697 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1698 nfsi->attrtimeo_timestamp = now;
1699 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1700 } else {
1701 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1702 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1703 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1704 nfsi->attrtimeo_timestamp = now;
1705 }
1706 }
1707 invalid &= ~NFS_INO_INVALID_ATTR;
1708 /* Don't invalidate the data if we were to blame */
1709 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1710 || S_ISLNK(inode->i_mode)))
1711 invalid &= ~NFS_INO_INVALID_DATA;
1712 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1713 (save_cache_validity & NFS_INO_REVAL_FORCED))
1714 nfs_set_cache_invalid(inode, invalid);
1715
1716 return 0;
1717 out_err:
1718 /*
1719 * No need to worry about unhashing the dentry, as the
1720 * lookup validation will know that the inode is bad.
1721 * (But we fall through to invalidate the caches.)
1722 */
1723 nfs_invalidate_inode(inode);
1724 return -ESTALE;
1725 }
1726
1727 struct inode *nfs_alloc_inode(struct super_block *sb)
1728 {
1729 struct nfs_inode *nfsi;
1730 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1731 if (!nfsi)
1732 return NULL;
1733 nfsi->flags = 0UL;
1734 nfsi->cache_validity = 0UL;
1735 #if IS_ENABLED(CONFIG_NFS_V4)
1736 nfsi->nfs4_acl = NULL;
1737 #endif /* CONFIG_NFS_V4 */
1738 return &nfsi->vfs_inode;
1739 }
1740 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1741
1742 static void nfs_i_callback(struct rcu_head *head)
1743 {
1744 struct inode *inode = container_of(head, struct inode, i_rcu);
1745 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1746 }
1747
1748 void nfs_destroy_inode(struct inode *inode)
1749 {
1750 call_rcu(&inode->i_rcu, nfs_i_callback);
1751 }
1752 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1753
1754 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1755 {
1756 #if IS_ENABLED(CONFIG_NFS_V4)
1757 INIT_LIST_HEAD(&nfsi->open_states);
1758 nfsi->delegation = NULL;
1759 nfsi->delegation_state = 0;
1760 init_rwsem(&nfsi->rwsem);
1761 nfsi->layout = NULL;
1762 #endif
1763 }
1764
1765 static void init_once(void *foo)
1766 {
1767 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1768
1769 inode_init_once(&nfsi->vfs_inode);
1770 INIT_LIST_HEAD(&nfsi->open_files);
1771 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1772 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1773 INIT_LIST_HEAD(&nfsi->commit_info.list);
1774 nfsi->npages = 0;
1775 nfsi->commit_info.ncommit = 0;
1776 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1777 atomic_set(&nfsi->silly_count, 1);
1778 INIT_HLIST_HEAD(&nfsi->silly_list);
1779 init_waitqueue_head(&nfsi->waitqueue);
1780 nfs4_init_once(nfsi);
1781 }
1782
1783 static int __init nfs_init_inodecache(void)
1784 {
1785 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1786 sizeof(struct nfs_inode),
1787 0, (SLAB_RECLAIM_ACCOUNT|
1788 SLAB_MEM_SPREAD),
1789 init_once);
1790 if (nfs_inode_cachep == NULL)
1791 return -ENOMEM;
1792
1793 return 0;
1794 }
1795
1796 static void nfs_destroy_inodecache(void)
1797 {
1798 /*
1799 * Make sure all delayed rcu free inodes are flushed before we
1800 * destroy cache.
1801 */
1802 rcu_barrier();
1803 kmem_cache_destroy(nfs_inode_cachep);
1804 }
1805
1806 struct workqueue_struct *nfsiod_workqueue;
1807 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1808
1809 /*
1810 * start up the nfsiod workqueue
1811 */
1812 static int nfsiod_start(void)
1813 {
1814 struct workqueue_struct *wq;
1815 dprintk("RPC: creating workqueue nfsiod\n");
1816 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1817 if (wq == NULL)
1818 return -ENOMEM;
1819 nfsiod_workqueue = wq;
1820 return 0;
1821 }
1822
1823 /*
1824 * Destroy the nfsiod workqueue
1825 */
1826 static void nfsiod_stop(void)
1827 {
1828 struct workqueue_struct *wq;
1829
1830 wq = nfsiod_workqueue;
1831 if (wq == NULL)
1832 return;
1833 nfsiod_workqueue = NULL;
1834 destroy_workqueue(wq);
1835 }
1836
1837 int nfs_net_id;
1838 EXPORT_SYMBOL_GPL(nfs_net_id);
1839
1840 static int nfs_net_init(struct net *net)
1841 {
1842 nfs_clients_init(net);
1843 return 0;
1844 }
1845
1846 static void nfs_net_exit(struct net *net)
1847 {
1848 nfs_cleanup_cb_ident_idr(net);
1849 }
1850
1851 static struct pernet_operations nfs_net_ops = {
1852 .init = nfs_net_init,
1853 .exit = nfs_net_exit,
1854 .id = &nfs_net_id,
1855 .size = sizeof(struct nfs_net),
1856 };
1857
1858 /*
1859 * Initialize NFS
1860 */
1861 static int __init init_nfs_fs(void)
1862 {
1863 int err;
1864
1865 err = register_pernet_subsys(&nfs_net_ops);
1866 if (err < 0)
1867 goto out9;
1868
1869 err = nfs_fscache_register();
1870 if (err < 0)
1871 goto out8;
1872
1873 err = nfsiod_start();
1874 if (err)
1875 goto out7;
1876
1877 err = nfs_fs_proc_init();
1878 if (err)
1879 goto out6;
1880
1881 err = nfs_init_nfspagecache();
1882 if (err)
1883 goto out5;
1884
1885 err = nfs_init_inodecache();
1886 if (err)
1887 goto out4;
1888
1889 err = nfs_init_readpagecache();
1890 if (err)
1891 goto out3;
1892
1893 err = nfs_init_writepagecache();
1894 if (err)
1895 goto out2;
1896
1897 err = nfs_init_directcache();
1898 if (err)
1899 goto out1;
1900
1901 #ifdef CONFIG_PROC_FS
1902 rpc_proc_register(&init_net, &nfs_rpcstat);
1903 #endif
1904 if ((err = register_nfs_fs()) != 0)
1905 goto out0;
1906
1907 return 0;
1908 out0:
1909 #ifdef CONFIG_PROC_FS
1910 rpc_proc_unregister(&init_net, "nfs");
1911 #endif
1912 nfs_destroy_directcache();
1913 out1:
1914 nfs_destroy_writepagecache();
1915 out2:
1916 nfs_destroy_readpagecache();
1917 out3:
1918 nfs_destroy_inodecache();
1919 out4:
1920 nfs_destroy_nfspagecache();
1921 out5:
1922 nfs_fs_proc_exit();
1923 out6:
1924 nfsiod_stop();
1925 out7:
1926 nfs_fscache_unregister();
1927 out8:
1928 unregister_pernet_subsys(&nfs_net_ops);
1929 out9:
1930 return err;
1931 }
1932
1933 static void __exit exit_nfs_fs(void)
1934 {
1935 nfs_destroy_directcache();
1936 nfs_destroy_writepagecache();
1937 nfs_destroy_readpagecache();
1938 nfs_destroy_inodecache();
1939 nfs_destroy_nfspagecache();
1940 nfs_fscache_unregister();
1941 unregister_pernet_subsys(&nfs_net_ops);
1942 #ifdef CONFIG_PROC_FS
1943 rpc_proc_unregister(&init_net, "nfs");
1944 #endif
1945 unregister_nfs_fs();
1946 nfs_fs_proc_exit();
1947 nfsiod_stop();
1948 }
1949
1950 /* Not quite true; I just maintain it */
1951 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1952 MODULE_LICENSE("GPL");
1953 module_param(enable_ino64, bool, 0644);
1954
1955 module_init(init_nfs_fs)
1956 module_exit(exit_nfs_fs)
Linux with added FPC-III support