NFS: Add tracepoints for debugging NFS rename and sillyrename issues
[linux-block.git] / fs / nfs / dir.c
CommitLineData
1da177e4
LT
1/*
2 * linux/fs/nfs/dir.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs directory handling functions
7 *
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
18 */
19
ddda8e0a 20#include <linux/module.h>
1da177e4
LT
21#include <linux/time.h>
22#include <linux/errno.h>
23#include <linux/stat.h>
24#include <linux/fcntl.h>
25#include <linux/string.h>
26#include <linux/kernel.h>
27#include <linux/slab.h>
28#include <linux/mm.h>
29#include <linux/sunrpc/clnt.h>
30#include <linux/nfs_fs.h>
31#include <linux/nfs_mount.h>
32#include <linux/pagemap.h>
873101b3 33#include <linux/pagevec.h>
1da177e4 34#include <linux/namei.h>
54ceac45 35#include <linux/mount.h>
a0b8cab3 36#include <linux/swap.h>
e8edc6e0 37#include <linux/sched.h>
04e4bd1c 38#include <linux/kmemleak.h>
64c2ce8b 39#include <linux/xattr.h>
1da177e4
LT
40
41#include "delegation.h"
91d5b470 42#include "iostat.h"
4c30d56e 43#include "internal.h"
cd9a1c0e 44#include "fscache.h"
1da177e4 45
f4ce1299
TM
46#include "nfstrace.h"
47
1da177e4
LT
48/* #define NFS_DEBUG_VERBOSE 1 */
49
50static int nfs_opendir(struct inode *, struct file *);
480c2006 51static int nfs_closedir(struct inode *, struct file *);
23db8620 52static int nfs_readdir(struct file *, struct dir_context *);
02c24a82 53static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
f0dd2136 54static loff_t nfs_llseek_dir(struct file *, loff_t, int);
11de3b11 55static void nfs_readdir_clear_array(struct page*);
1da177e4 56
4b6f5d20 57const struct file_operations nfs_dir_operations = {
f0dd2136 58 .llseek = nfs_llseek_dir,
1da177e4 59 .read = generic_read_dir,
23db8620 60 .iterate = nfs_readdir,
1da177e4 61 .open = nfs_opendir,
480c2006 62 .release = nfs_closedir,
1da177e4
LT
63 .fsync = nfs_fsync_dir,
64};
65
11de3b11
TM
66const struct address_space_operations nfs_dir_aops = {
67 .freepage = nfs_readdir_clear_array,
d1bacf9e
BS
68};
69
0c030806 70static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
480c2006
BS
71{
72 struct nfs_open_dir_context *ctx;
73 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
74 if (ctx != NULL) {
8ef2ce3e 75 ctx->duped = 0;
0c030806 76 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
480c2006 77 ctx->dir_cookie = 0;
8ef2ce3e 78 ctx->dup_cookie = 0;
480c2006 79 ctx->cred = get_rpccred(cred);
0c030806
TM
80 return ctx;
81 }
82 return ERR_PTR(-ENOMEM);
480c2006
BS
83}
84
85static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
86{
87 put_rpccred(ctx->cred);
88 kfree(ctx);
89}
90
1da177e4
LT
91/*
92 * Open file
93 */
94static int
95nfs_opendir(struct inode *inode, struct file *filp)
96{
480c2006
BS
97 int res = 0;
98 struct nfs_open_dir_context *ctx;
99 struct rpc_cred *cred;
1da177e4 100
6da24bc9 101 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
cc0dd2d1
CL
102 filp->f_path.dentry->d_parent->d_name.name,
103 filp->f_path.dentry->d_name.name);
104
105 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
1e7cb3dc 106
480c2006
BS
107 cred = rpc_lookup_cred();
108 if (IS_ERR(cred))
109 return PTR_ERR(cred);
0c030806 110 ctx = alloc_nfs_open_dir_context(inode, cred);
480c2006
BS
111 if (IS_ERR(ctx)) {
112 res = PTR_ERR(ctx);
113 goto out;
114 }
115 filp->private_data = ctx;
f5a73672
NB
116 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
117 /* This is a mountpoint, so d_revalidate will never
118 * have been called, so we need to refresh the
119 * inode (for close-open consistency) ourselves.
120 */
121 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
122 }
480c2006
BS
123out:
124 put_rpccred(cred);
1da177e4
LT
125 return res;
126}
127
480c2006
BS
128static int
129nfs_closedir(struct inode *inode, struct file *filp)
130{
131 put_nfs_open_dir_context(filp->private_data);
132 return 0;
133}
134
d1bacf9e
BS
135struct nfs_cache_array_entry {
136 u64 cookie;
137 u64 ino;
138 struct qstr string;
0b26a0bf 139 unsigned char d_type;
d1bacf9e
BS
140};
141
142struct nfs_cache_array {
88b8e133 143 int size;
d1bacf9e
BS
144 int eof_index;
145 u64 last_cookie;
146 struct nfs_cache_array_entry array[0];
147};
148
573c4e1e 149typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
1da177e4
LT
150typedef struct {
151 struct file *file;
152 struct page *page;
23db8620 153 struct dir_context *ctx;
1da177e4 154 unsigned long page_index;
f0dd2136 155 u64 *dir_cookie;
0aded708 156 u64 last_cookie;
f0dd2136 157 loff_t current_index;
1da177e4 158 decode_dirent_t decode;
d1bacf9e 159
1f4eab7e 160 unsigned long timestamp;
4704f0e2 161 unsigned long gencount;
d1bacf9e
BS
162 unsigned int cache_entry_index;
163 unsigned int plus:1;
164 unsigned int eof:1;
1da177e4
LT
165} nfs_readdir_descriptor_t;
166
d1bacf9e
BS
167/*
168 * The caller is responsible for calling nfs_readdir_release_array(page)
1da177e4
LT
169 */
170static
d1bacf9e
BS
171struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
172{
8cd51a0c 173 void *ptr;
d1bacf9e
BS
174 if (page == NULL)
175 return ERR_PTR(-EIO);
8cd51a0c
TM
176 ptr = kmap(page);
177 if (ptr == NULL)
178 return ERR_PTR(-ENOMEM);
179 return ptr;
d1bacf9e
BS
180}
181
182static
183void nfs_readdir_release_array(struct page *page)
184{
185 kunmap(page);
186}
187
188/*
189 * we are freeing strings created by nfs_add_to_readdir_array()
190 */
191static
11de3b11 192void nfs_readdir_clear_array(struct page *page)
d1bacf9e 193{
11de3b11 194 struct nfs_cache_array *array;
d1bacf9e 195 int i;
8cd51a0c 196
2b86ce2d 197 array = kmap_atomic(page);
d1bacf9e
BS
198 for (i = 0; i < array->size; i++)
199 kfree(array->array[i].string.name);
2b86ce2d 200 kunmap_atomic(array);
d1bacf9e
BS
201}
202
203/*
204 * the caller is responsible for freeing qstr.name
205 * when called by nfs_readdir_add_to_array, the strings will be freed in
206 * nfs_clear_readdir_array()
207 */
208static
4a201d6e 209int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
d1bacf9e
BS
210{
211 string->len = len;
212 string->name = kmemdup(name, len, GFP_KERNEL);
4a201d6e
TM
213 if (string->name == NULL)
214 return -ENOMEM;
04e4bd1c
CM
215 /*
216 * Avoid a kmemleak false positive. The pointer to the name is stored
217 * in a page cache page which kmemleak does not scan.
218 */
219 kmemleak_not_leak(string->name);
4a201d6e
TM
220 string->hash = full_name_hash(name, len);
221 return 0;
d1bacf9e
BS
222}
223
224static
225int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
226{
227 struct nfs_cache_array *array = nfs_readdir_get_array(page);
4a201d6e
TM
228 struct nfs_cache_array_entry *cache_entry;
229 int ret;
230
d1bacf9e
BS
231 if (IS_ERR(array))
232 return PTR_ERR(array);
3020093f
TM
233
234 cache_entry = &array->array[array->size];
235
236 /* Check that this entry lies within the page bounds */
8cd51a0c 237 ret = -ENOSPC;
3020093f 238 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
4a201d6e 239 goto out;
d1bacf9e 240
4a201d6e
TM
241 cache_entry->cookie = entry->prev_cookie;
242 cache_entry->ino = entry->ino;
0b26a0bf 243 cache_entry->d_type = entry->d_type;
4a201d6e
TM
244 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
245 if (ret)
246 goto out;
d1bacf9e 247 array->last_cookie = entry->cookie;
8cd51a0c 248 array->size++;
47c716cb 249 if (entry->eof != 0)
d1bacf9e 250 array->eof_index = array->size;
4a201d6e 251out:
d1bacf9e 252 nfs_readdir_release_array(page);
4a201d6e 253 return ret;
d1bacf9e
BS
254}
255
256static
257int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
258{
23db8620 259 loff_t diff = desc->ctx->pos - desc->current_index;
d1bacf9e
BS
260 unsigned int index;
261
262 if (diff < 0)
263 goto out_eof;
264 if (diff >= array->size) {
8cd51a0c 265 if (array->eof_index >= 0)
d1bacf9e 266 goto out_eof;
d1bacf9e
BS
267 return -EAGAIN;
268 }
269
270 index = (unsigned int)diff;
271 *desc->dir_cookie = array->array[index].cookie;
272 desc->cache_entry_index = index;
d1bacf9e
BS
273 return 0;
274out_eof:
275 desc->eof = 1;
276 return -EBADCOOKIE;
277}
278
279static
280int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
281{
282 int i;
8ef2ce3e 283 loff_t new_pos;
d1bacf9e
BS
284 int status = -EAGAIN;
285
286 for (i = 0; i < array->size; i++) {
d1bacf9e 287 if (array->array[i].cookie == *desc->dir_cookie) {
496ad9aa 288 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
0c030806
TM
289 struct nfs_open_dir_context *ctx = desc->file->private_data;
290
8ef2ce3e 291 new_pos = desc->current_index + i;
0c030806
TM
292 if (ctx->attr_gencount != nfsi->attr_gencount
293 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
294 ctx->duped = 0;
295 ctx->attr_gencount = nfsi->attr_gencount;
23db8620 296 } else if (new_pos < desc->ctx->pos) {
0c030806
TM
297 if (ctx->duped > 0
298 && ctx->dup_cookie == *desc->dir_cookie) {
299 if (printk_ratelimit()) {
300 pr_notice("NFS: directory %s/%s contains a readdir loop."
301 "Please contact your server vendor. "
374e4e3e 302 "The file: %s has duplicate cookie %llu\n",
0c030806
TM
303 desc->file->f_dentry->d_parent->d_name.name,
304 desc->file->f_dentry->d_name.name,
374e4e3e 305 array->array[i].string.name,
0c030806
TM
306 *desc->dir_cookie);
307 }
308 status = -ELOOP;
309 goto out;
310 }
8ef2ce3e 311 ctx->dup_cookie = *desc->dir_cookie;
0c030806 312 ctx->duped = -1;
8ef2ce3e 313 }
23db8620 314 desc->ctx->pos = new_pos;
d1bacf9e 315 desc->cache_entry_index = i;
47c716cb 316 return 0;
d1bacf9e
BS
317 }
318 }
47c716cb 319 if (array->eof_index >= 0) {
8cd51a0c 320 status = -EBADCOOKIE;
18fb5fe4
TM
321 if (*desc->dir_cookie == array->last_cookie)
322 desc->eof = 1;
8cd51a0c 323 }
0c030806 324out:
d1bacf9e
BS
325 return status;
326}
327
328static
329int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
330{
331 struct nfs_cache_array *array;
47c716cb 332 int status;
d1bacf9e
BS
333
334 array = nfs_readdir_get_array(desc->page);
335 if (IS_ERR(array)) {
336 status = PTR_ERR(array);
337 goto out;
338 }
339
340 if (*desc->dir_cookie == 0)
341 status = nfs_readdir_search_for_pos(array, desc);
342 else
343 status = nfs_readdir_search_for_cookie(array, desc);
344
47c716cb 345 if (status == -EAGAIN) {
0aded708 346 desc->last_cookie = array->last_cookie;
e47c085a 347 desc->current_index += array->size;
47c716cb
TM
348 desc->page_index++;
349 }
d1bacf9e
BS
350 nfs_readdir_release_array(desc->page);
351out:
352 return status;
353}
354
355/* Fill a page with xdr information before transferring to the cache page */
356static
56e4ebf8 357int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
d1bacf9e 358 struct nfs_entry *entry, struct file *file, struct inode *inode)
1da177e4 359{
480c2006
BS
360 struct nfs_open_dir_context *ctx = file->private_data;
361 struct rpc_cred *cred = ctx->cred;
4704f0e2 362 unsigned long timestamp, gencount;
1da177e4
LT
363 int error;
364
1da177e4
LT
365 again:
366 timestamp = jiffies;
4704f0e2 367 gencount = nfs_inc_attr_generation_counter();
56e4ebf8 368 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
1da177e4
LT
369 NFS_SERVER(inode)->dtsize, desc->plus);
370 if (error < 0) {
371 /* We requested READDIRPLUS, but the server doesn't grok it */
372 if (error == -ENOTSUPP && desc->plus) {
373 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
3a10c30a 374 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
1da177e4
LT
375 desc->plus = 0;
376 goto again;
377 }
378 goto error;
379 }
1f4eab7e 380 desc->timestamp = timestamp;
4704f0e2 381 desc->gencount = gencount;
d1bacf9e
BS
382error:
383 return error;
1da177e4
LT
384}
385
573c4e1e
CL
386static int xdr_decode(nfs_readdir_descriptor_t *desc,
387 struct nfs_entry *entry, struct xdr_stream *xdr)
1da177e4 388{
573c4e1e 389 int error;
1da177e4 390
573c4e1e
CL
391 error = desc->decode(xdr, entry, desc->plus);
392 if (error)
393 return error;
d1bacf9e
BS
394 entry->fattr->time_start = desc->timestamp;
395 entry->fattr->gencount = desc->gencount;
396 return 0;
1da177e4
LT
397}
398
d39ab9de
BS
399static
400int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
401{
d39ab9de
BS
402 if (dentry->d_inode == NULL)
403 goto different;
37a09f07 404 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
d39ab9de
BS
405 goto different;
406 return 1;
407different:
408 return 0;
409}
410
d69ee9b8 411static
23db8620 412bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
d69ee9b8
TM
413{
414 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
415 return false;
416 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
417 return true;
23db8620 418 if (ctx->pos == 0)
d69ee9b8
TM
419 return true;
420 return false;
421}
422
423/*
424 * This function is called by the lookup code to request the use of
425 * readdirplus to accelerate any future lookups in the same
426 * directory.
427 */
428static
429void nfs_advise_use_readdirplus(struct inode *dir)
430{
431 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
432}
433
d39ab9de
BS
434static
435void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
436{
26fe5750 437 struct qstr filename = QSTR_INIT(entry->name, entry->len);
4a201d6e
TM
438 struct dentry *dentry;
439 struct dentry *alias;
d39ab9de
BS
440 struct inode *dir = parent->d_inode;
441 struct inode *inode;
aa9c2669 442 int status;
d39ab9de 443
4a201d6e
TM
444 if (filename.name[0] == '.') {
445 if (filename.len == 1)
446 return;
447 if (filename.len == 2 && filename.name[1] == '.')
448 return;
449 }
450 filename.hash = full_name_hash(filename.name, filename.len);
d39ab9de 451
4a201d6e 452 dentry = d_lookup(parent, &filename);
d39ab9de
BS
453 if (dentry != NULL) {
454 if (nfs_same_file(dentry, entry)) {
cda57a1e 455 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
aa9c2669
DQ
456 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
457 if (!status)
458 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
d39ab9de
BS
459 goto out;
460 } else {
696199f8
AV
461 if (d_invalidate(dentry) != 0)
462 goto out;
d39ab9de
BS
463 dput(dentry);
464 }
465 }
466
467 dentry = d_alloc(parent, &filename);
4a201d6e
TM
468 if (dentry == NULL)
469 return;
470
1775fd3e 471 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
d39ab9de
BS
472 if (IS_ERR(inode))
473 goto out;
474
475 alias = d_materialise_unique(dentry, inode);
476 if (IS_ERR(alias))
477 goto out;
478 else if (alias) {
479 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
480 dput(alias);
481 } else
482 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
483
484out:
485 dput(dentry);
d39ab9de
BS
486}
487
d1bacf9e
BS
488/* Perform conversion from xdr to cache array */
489static
8cd51a0c 490int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
6650239a 491 struct page **xdr_pages, struct page *page, unsigned int buflen)
1da177e4 492{
babddc72 493 struct xdr_stream stream;
f7da7a12 494 struct xdr_buf buf;
6650239a 495 struct page *scratch;
99424380 496 struct nfs_cache_array *array;
5c346854
TM
497 unsigned int count = 0;
498 int status;
babddc72 499
6650239a
TM
500 scratch = alloc_page(GFP_KERNEL);
501 if (scratch == NULL)
502 return -ENOMEM;
babddc72 503
f7da7a12 504 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
6650239a 505 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
99424380
BS
506
507 do {
508 status = xdr_decode(desc, entry, &stream);
8cd51a0c
TM
509 if (status != 0) {
510 if (status == -EAGAIN)
511 status = 0;
99424380 512 break;
8cd51a0c 513 }
99424380 514
5c346854
TM
515 count++;
516
47c716cb 517 if (desc->plus != 0)
d39ab9de 518 nfs_prime_dcache(desc->file->f_path.dentry, entry);
8cd51a0c
TM
519
520 status = nfs_readdir_add_to_array(entry, page);
521 if (status != 0)
522 break;
99424380
BS
523 } while (!entry->eof);
524
47c716cb 525 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
99424380 526 array = nfs_readdir_get_array(page);
8cd51a0c
TM
527 if (!IS_ERR(array)) {
528 array->eof_index = array->size;
529 status = 0;
530 nfs_readdir_release_array(page);
5c346854
TM
531 } else
532 status = PTR_ERR(array);
1da177e4 533 }
6650239a
TM
534
535 put_page(scratch);
8cd51a0c 536 return status;
56e4ebf8
BS
537}
538
539static
540void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
541{
542 unsigned int i;
543 for (i = 0; i < npages; i++)
544 put_page(pages[i]);
545}
546
547static
548void nfs_readdir_free_large_page(void *ptr, struct page **pages,
549 unsigned int npages)
550{
56e4ebf8
BS
551 nfs_readdir_free_pagearray(pages, npages);
552}
553
554/*
555 * nfs_readdir_large_page will allocate pages that must be freed with a call
556 * to nfs_readdir_free_large_page
557 */
558static
6650239a 559int nfs_readdir_large_page(struct page **pages, unsigned int npages)
56e4ebf8 560{
56e4ebf8
BS
561 unsigned int i;
562
563 for (i = 0; i < npages; i++) {
564 struct page *page = alloc_page(GFP_KERNEL);
565 if (page == NULL)
566 goto out_freepages;
567 pages[i] = page;
568 }
6650239a 569 return 0;
56e4ebf8 570
56e4ebf8
BS
571out_freepages:
572 nfs_readdir_free_pagearray(pages, i);
6650239a 573 return -ENOMEM;
1da177e4
LT
574}
575
d1bacf9e
BS
576static
577int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
00a92642 578{
56e4ebf8
BS
579 struct page *pages[NFS_MAX_READDIR_PAGES];
580 void *pages_ptr = NULL;
d1bacf9e
BS
581 struct nfs_entry entry;
582 struct file *file = desc->file;
583 struct nfs_cache_array *array;
8cd51a0c 584 int status = -ENOMEM;
56e4ebf8 585 unsigned int array_size = ARRAY_SIZE(pages);
d1bacf9e
BS
586
587 entry.prev_cookie = 0;
0aded708 588 entry.cookie = desc->last_cookie;
d1bacf9e
BS
589 entry.eof = 0;
590 entry.fh = nfs_alloc_fhandle();
591 entry.fattr = nfs_alloc_fattr();
573c4e1e 592 entry.server = NFS_SERVER(inode);
d1bacf9e
BS
593 if (entry.fh == NULL || entry.fattr == NULL)
594 goto out;
00a92642 595
14c43f76
DQ
596 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
597 if (IS_ERR(entry.label)) {
598 status = PTR_ERR(entry.label);
599 goto out;
600 }
601
d1bacf9e 602 array = nfs_readdir_get_array(page);
8cd51a0c
TM
603 if (IS_ERR(array)) {
604 status = PTR_ERR(array);
14c43f76 605 goto out_label_free;
8cd51a0c 606 }
d1bacf9e
BS
607 memset(array, 0, sizeof(struct nfs_cache_array));
608 array->eof_index = -1;
00a92642 609
6650239a
TM
610 status = nfs_readdir_large_page(pages, array_size);
611 if (status < 0)
d1bacf9e
BS
612 goto out_release_array;
613 do {
ac396128 614 unsigned int pglen;
56e4ebf8 615 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
babddc72 616
d1bacf9e 617 if (status < 0)
00a92642 618 break;
ac396128 619 pglen = status;
6650239a 620 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
8cd51a0c
TM
621 if (status < 0) {
622 if (status == -ENOSPC)
623 status = 0;
624 break;
625 }
626 } while (array->eof_index < 0);
d1bacf9e 627
56e4ebf8 628 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
d1bacf9e
BS
629out_release_array:
630 nfs_readdir_release_array(page);
14c43f76
DQ
631out_label_free:
632 nfs4_label_free(entry.label);
d1bacf9e
BS
633out:
634 nfs_free_fattr(entry.fattr);
635 nfs_free_fhandle(entry.fh);
00a92642
OG
636 return status;
637}
638
639/*
d1bacf9e
BS
640 * Now we cache directories properly, by converting xdr information
641 * to an array that can be used for lookups later. This results in
642 * fewer cache pages, since we can store more information on each page.
643 * We only need to convert from xdr once so future lookups are much simpler
1da177e4 644 */
d1bacf9e
BS
645static
646int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
1da177e4 647{
496ad9aa 648 struct inode *inode = file_inode(desc->file);
8cd51a0c 649 int ret;
1da177e4 650
8cd51a0c
TM
651 ret = nfs_readdir_xdr_to_array(desc, page, inode);
652 if (ret < 0)
d1bacf9e
BS
653 goto error;
654 SetPageUptodate(page);
1da177e4 655
d1bacf9e
BS
656 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
657 /* Should never happen */
658 nfs_zap_mapping(inode, inode->i_mapping);
1da177e4 659 }
d1bacf9e
BS
660 unlock_page(page);
661 return 0;
662 error:
663 unlock_page(page);
8cd51a0c 664 return ret;
d1bacf9e 665}
1da177e4 666
d1bacf9e
BS
667static
668void cache_page_release(nfs_readdir_descriptor_t *desc)
669{
11de3b11
TM
670 if (!desc->page->mapping)
671 nfs_readdir_clear_array(desc->page);
d1bacf9e
BS
672 page_cache_release(desc->page);
673 desc->page = NULL;
674}
675
676static
677struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
678{
496ad9aa 679 return read_cache_page(file_inode(desc->file)->i_mapping,
d1bacf9e 680 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
1da177e4
LT
681}
682
683/*
d1bacf9e 684 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1da177e4 685 */
d1bacf9e
BS
686static
687int find_cache_page(nfs_readdir_descriptor_t *desc)
688{
689 int res;
690
691 desc->page = get_cache_page(desc);
692 if (IS_ERR(desc->page))
693 return PTR_ERR(desc->page);
694
695 res = nfs_readdir_search_array(desc);
47c716cb
TM
696 if (res != 0)
697 cache_page_release(desc);
d1bacf9e
BS
698 return res;
699}
700
701/* Search for desc->dir_cookie from the beginning of the page cache */
1da177e4
LT
702static inline
703int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
704{
8cd51a0c 705 int res;
d1bacf9e 706
0aded708 707 if (desc->page_index == 0) {
8cd51a0c 708 desc->current_index = 0;
0aded708
TM
709 desc->last_cookie = 0;
710 }
47c716cb 711 do {
d1bacf9e 712 res = find_cache_page(desc);
47c716cb 713 } while (res == -EAGAIN);
1da177e4
LT
714 return res;
715}
716
1da177e4
LT
717/*
718 * Once we've found the start of the dirent within a page: fill 'er up...
719 */
720static
23db8620 721int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
1da177e4
LT
722{
723 struct file *file = desc->file;
d1bacf9e
BS
724 int i = 0;
725 int res = 0;
726 struct nfs_cache_array *array = NULL;
8ef2ce3e
BS
727 struct nfs_open_dir_context *ctx = file->private_data;
728
d1bacf9e 729 array = nfs_readdir_get_array(desc->page);
e7c58e97
TM
730 if (IS_ERR(array)) {
731 res = PTR_ERR(array);
732 goto out;
733 }
d1bacf9e
BS
734
735 for (i = desc->cache_entry_index; i < array->size; i++) {
ece0b423 736 struct nfs_cache_array_entry *ent;
1da177e4 737
ece0b423 738 ent = &array->array[i];
23db8620
AV
739 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
740 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
ece0b423 741 desc->eof = 1;
1da177e4 742 break;
ece0b423 743 }
23db8620 744 desc->ctx->pos++;
d1bacf9e
BS
745 if (i < (array->size-1))
746 *desc->dir_cookie = array->array[i+1].cookie;
747 else
748 *desc->dir_cookie = array->last_cookie;
0c030806
TM
749 if (ctx->duped != 0)
750 ctx->duped = 1;
1da177e4 751 }
47c716cb 752 if (array->eof_index >= 0)
8cd51a0c 753 desc->eof = 1;
d1bacf9e
BS
754
755 nfs_readdir_release_array(desc->page);
e7c58e97 756out:
d1bacf9e 757 cache_page_release(desc);
1e7cb3dc
CL
758 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
759 (unsigned long long)*desc->dir_cookie, res);
1da177e4
LT
760 return res;
761}
762
763/*
764 * If we cannot find a cookie in our cache, we suspect that this is
765 * because it points to a deleted file, so we ask the server to return
766 * whatever it thinks is the next entry. We then feed this to filldir.
767 * If all goes well, we should then be able to find our way round the
768 * cache on the next call to readdir_search_pagecache();
769 *
770 * NOTE: we cannot add the anonymous page to the pagecache because
771 * the data it contains might not be page aligned. Besides,
772 * we should already have a complete representation of the
773 * directory in the page cache by the time we get here.
774 */
775static inline
23db8620 776int uncached_readdir(nfs_readdir_descriptor_t *desc)
1da177e4 777{
1da177e4
LT
778 struct page *page = NULL;
779 int status;
496ad9aa 780 struct inode *inode = file_inode(desc->file);
0c030806 781 struct nfs_open_dir_context *ctx = desc->file->private_data;
1da177e4 782
1e7cb3dc
CL
783 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
784 (unsigned long long)*desc->dir_cookie);
1da177e4
LT
785
786 page = alloc_page(GFP_HIGHUSER);
787 if (!page) {
788 status = -ENOMEM;
789 goto out;
790 }
d1bacf9e 791
7a8e1dc3 792 desc->page_index = 0;
0aded708 793 desc->last_cookie = *desc->dir_cookie;
7a8e1dc3 794 desc->page = page;
0c030806 795 ctx->duped = 0;
7a8e1dc3 796
85f8607e
TM
797 status = nfs_readdir_xdr_to_array(desc, page, inode);
798 if (status < 0)
1da177e4
LT
799 goto out_release;
800
23db8620 801 status = nfs_do_filldir(desc);
1da177e4 802
1da177e4 803 out:
1e7cb3dc 804 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
3110ff80 805 __func__, status);
1da177e4
LT
806 return status;
807 out_release:
d1bacf9e 808 cache_page_release(desc);
1da177e4
LT
809 goto out;
810}
811
00a92642
OG
812/* The file offset position represents the dirent entry number. A
813 last cookie cache takes care of the common case of reading the
814 whole directory.
1da177e4 815 */
23db8620 816static int nfs_readdir(struct file *file, struct dir_context *ctx)
1da177e4 817{
23db8620 818 struct dentry *dentry = file->f_path.dentry;
1da177e4
LT
819 struct inode *inode = dentry->d_inode;
820 nfs_readdir_descriptor_t my_desc,
821 *desc = &my_desc;
23db8620 822 struct nfs_open_dir_context *dir_ctx = file->private_data;
07b5ce8e 823 int res = 0;
1da177e4 824
6da24bc9 825 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
1e7cb3dc 826 dentry->d_parent->d_name.name, dentry->d_name.name,
23db8620 827 (long long)ctx->pos);
91d5b470
CL
828 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
829
1da177e4 830 /*
23db8620 831 * ctx->pos points to the dirent entry number.
f0dd2136 832 * *desc->dir_cookie has the cookie for the next entry. We have
00a92642
OG
833 * to either find the entry with the appropriate number or
834 * revalidate the cookie.
1da177e4
LT
835 */
836 memset(desc, 0, sizeof(*desc));
837
23db8620
AV
838 desc->file = file;
839 desc->ctx = ctx;
480c2006 840 desc->dir_cookie = &dir_ctx->dir_cookie;
1da177e4 841 desc->decode = NFS_PROTO(inode)->decode_dirent;
23db8620 842 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
1da177e4 843
565277f6 844 nfs_block_sillyrename(dentry);
07b5ce8e
SM
845 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
846 res = nfs_revalidate_mapping(inode, file->f_mapping);
fccca7fc
TM
847 if (res < 0)
848 goto out;
849
47c716cb 850 do {
1da177e4 851 res = readdir_search_pagecache(desc);
00a92642 852
1da177e4 853 if (res == -EBADCOOKIE) {
ece0b423 854 res = 0;
1da177e4 855 /* This means either end of directory */
d1bacf9e 856 if (*desc->dir_cookie && desc->eof == 0) {
1da177e4 857 /* Or that the server has 'lost' a cookie */
23db8620 858 res = uncached_readdir(desc);
ece0b423 859 if (res == 0)
1da177e4
LT
860 continue;
861 }
1da177e4
LT
862 break;
863 }
864 if (res == -ETOOSMALL && desc->plus) {
3a10c30a 865 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
1da177e4 866 nfs_zap_caches(inode);
baf57a09 867 desc->page_index = 0;
1da177e4 868 desc->plus = 0;
d1bacf9e 869 desc->eof = 0;
1da177e4
LT
870 continue;
871 }
872 if (res < 0)
873 break;
874
23db8620 875 res = nfs_do_filldir(desc);
ece0b423 876 if (res < 0)
1da177e4 877 break;
47c716cb 878 } while (!desc->eof);
fccca7fc 879out:
565277f6 880 nfs_unblock_sillyrename(dentry);
1e7cb3dc
CL
881 if (res > 0)
882 res = 0;
aa49b4cf 883 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
1e7cb3dc
CL
884 dentry->d_parent->d_name.name, dentry->d_name.name,
885 res);
886 return res;
1da177e4
LT
887}
888
965c8e59 889static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
f0dd2136 890{
b84e06c5
CL
891 struct dentry *dentry = filp->f_path.dentry;
892 struct inode *inode = dentry->d_inode;
480c2006 893 struct nfs_open_dir_context *dir_ctx = filp->private_data;
b84e06c5 894
6da24bc9 895 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
b84e06c5
CL
896 dentry->d_parent->d_name.name,
897 dentry->d_name.name,
965c8e59 898 offset, whence);
b84e06c5
CL
899
900 mutex_lock(&inode->i_mutex);
965c8e59 901 switch (whence) {
f0dd2136
TM
902 case 1:
903 offset += filp->f_pos;
904 case 0:
905 if (offset >= 0)
906 break;
907 default:
908 offset = -EINVAL;
909 goto out;
910 }
911 if (offset != filp->f_pos) {
912 filp->f_pos = offset;
480c2006 913 dir_ctx->dir_cookie = 0;
8ef2ce3e 914 dir_ctx->duped = 0;
f0dd2136
TM
915 }
916out:
b84e06c5 917 mutex_unlock(&inode->i_mutex);
f0dd2136
TM
918 return offset;
919}
920
1da177e4
LT
921/*
922 * All directory operations under NFS are synchronous, so fsync()
923 * is a dummy operation.
924 */
02c24a82
JB
925static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
926 int datasync)
1da177e4 927{
7ea80859 928 struct dentry *dentry = filp->f_path.dentry;
02c24a82 929 struct inode *inode = dentry->d_inode;
7ea80859 930
6da24bc9 931 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
1e7cb3dc
CL
932 dentry->d_parent->d_name.name, dentry->d_name.name,
933 datasync);
934
02c24a82 935 mutex_lock(&inode->i_mutex);
54917786 936 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
02c24a82 937 mutex_unlock(&inode->i_mutex);
1da177e4
LT
938 return 0;
939}
940
bfc69a45
TM
941/**
942 * nfs_force_lookup_revalidate - Mark the directory as having changed
943 * @dir - pointer to directory inode
944 *
945 * This forces the revalidation code in nfs_lookup_revalidate() to do a
946 * full lookup on all child dentries of 'dir' whenever a change occurs
947 * on the server that might have invalidated our dcache.
948 *
949 * The caller should be holding dir->i_lock
950 */
951void nfs_force_lookup_revalidate(struct inode *dir)
952{
011935a0 953 NFS_I(dir)->cache_change_attribute++;
bfc69a45 954}
89d77c8f 955EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
bfc69a45 956
1da177e4
LT
957/*
958 * A check for whether or not the parent directory has changed.
959 * In the case it has, we assume that the dentries are untrustworthy
960 * and may need to be looked up again.
961 */
c79ba787 962static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
1da177e4
LT
963{
964 if (IS_ROOT(dentry))
965 return 1;
4eec952e
TM
966 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
967 return 0;
f2c77f4e
TM
968 if (!nfs_verify_change_attribute(dir, dentry->d_time))
969 return 0;
970 /* Revalidate nfsi->cache_change_attribute before we declare a match */
971 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
972 return 0;
973 if (!nfs_verify_change_attribute(dir, dentry->d_time))
974 return 0;
975 return 1;
1da177e4
LT
976}
977
a12802ca
TM
978/*
979 * Use intent information to check whether or not we're going to do
980 * an O_EXCL create using this path component.
981 */
fa3c56bb 982static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
a12802ca
TM
983{
984 if (NFS_PROTO(dir)->version == 2)
985 return 0;
fa3c56bb 986 return flags & LOOKUP_EXCL;
a12802ca
TM
987}
988
1d6757fb
TM
989/*
990 * Inode and filehandle revalidation for lookups.
991 *
992 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
993 * or if the intent information indicates that we're about to open this
994 * particular file and the "nocto" mount flag is not set.
995 *
996 */
65a0c149 997static
fa3c56bb 998int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1da177e4
LT
999{
1000 struct nfs_server *server = NFS_SERVER(inode);
65a0c149 1001 int ret;
1da177e4 1002
36d43a43 1003 if (IS_AUTOMOUNT(inode))
4e99a1ff 1004 return 0;
facc3530 1005 /* VFS wants an on-the-wire revalidation */
fa3c56bb 1006 if (flags & LOOKUP_REVAL)
facc3530
AV
1007 goto out_force;
1008 /* This is an open(2) */
fa3c56bb
AV
1009 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1010 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
facc3530 1011 goto out_force;
65a0c149
TM
1012out:
1013 return (inode->i_nlink == 0) ? -ENOENT : 0;
1da177e4 1014out_force:
65a0c149
TM
1015 ret = __nfs_revalidate_inode(server, inode);
1016 if (ret != 0)
1017 return ret;
1018 goto out;
1da177e4
LT
1019}
1020
1021/*
1022 * We judge how long we want to trust negative
1023 * dentries by looking at the parent inode mtime.
1024 *
1025 * If parent mtime has changed, we revalidate, else we wait for a
1026 * period corresponding to the parent's attribute cache timeout value.
1027 */
1028static inline
1029int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
fa3c56bb 1030 unsigned int flags)
1da177e4 1031{
1da177e4 1032 /* Don't revalidate a negative dentry if we're creating a new file */
fa3c56bb 1033 if (flags & LOOKUP_CREATE)
1da177e4 1034 return 0;
4eec952e
TM
1035 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1036 return 1;
1da177e4
LT
1037 return !nfs_check_verifier(dir, dentry);
1038}
1039
1040/*
1041 * This is called every time the dcache has a lookup hit,
1042 * and we should check whether we can really trust that
1043 * lookup.
1044 *
1045 * NOTE! The hit can be a negative hit too, don't assume
1046 * we have an inode!
1047 *
1048 * If the parent directory is seen to have changed, we throw out the
1049 * cached dentry and do a new lookup.
1050 */
0b728e19 1051static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1da177e4
LT
1052{
1053 struct inode *dir;
1054 struct inode *inode;
1055 struct dentry *parent;
e1fb4d05
TM
1056 struct nfs_fh *fhandle = NULL;
1057 struct nfs_fattr *fattr = NULL;
1775fd3e 1058 struct nfs4_label *label = NULL;
1da177e4 1059 int error;
1da177e4 1060
fa3c56bb 1061 if (flags & LOOKUP_RCU)
34286d66
NP
1062 return -ECHILD;
1063
1da177e4 1064 parent = dget_parent(dentry);
1da177e4 1065 dir = parent->d_inode;
91d5b470 1066 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1da177e4
LT
1067 inode = dentry->d_inode;
1068
1069 if (!inode) {
fa3c56bb 1070 if (nfs_neg_need_reval(dir, dentry, flags))
1da177e4 1071 goto out_bad;
d69ee9b8 1072 goto out_valid_noent;
1da177e4
LT
1073 }
1074
1075 if (is_bad_inode(inode)) {
1e7cb3dc 1076 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
3110ff80 1077 __func__, dentry->d_parent->d_name.name,
1e7cb3dc 1078 dentry->d_name.name);
1da177e4
LT
1079 goto out_bad;
1080 }
1081
011e2a7f 1082 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
15860ab1
TM
1083 goto out_set_verifier;
1084
1da177e4 1085 /* Force a full look up iff the parent directory has changed */
fa3c56bb
AV
1086 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1087 if (nfs_lookup_verify_inode(inode, flags))
1da177e4
LT
1088 goto out_zap_parent;
1089 goto out_valid;
1090 }
1091
1092 if (NFS_STALE(inode))
1093 goto out_bad;
1094
e1fb4d05
TM
1095 error = -ENOMEM;
1096 fhandle = nfs_alloc_fhandle();
1097 fattr = nfs_alloc_fattr();
1098 if (fhandle == NULL || fattr == NULL)
1099 goto out_error;
1100
14c43f76
DQ
1101 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1102 if (IS_ERR(label))
1103 goto out_error;
1104
6e0d0be7 1105 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1775fd3e 1106 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
6e0d0be7 1107 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1da177e4
LT
1108 if (error)
1109 goto out_bad;
e1fb4d05 1110 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1da177e4 1111 goto out_bad;
e1fb4d05 1112 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1da177e4
LT
1113 goto out_bad;
1114
aa9c2669
DQ
1115 nfs_setsecurity(inode, fattr, label);
1116
e1fb4d05
TM
1117 nfs_free_fattr(fattr);
1118 nfs_free_fhandle(fhandle);
14c43f76
DQ
1119 nfs4_label_free(label);
1120
15860ab1 1121out_set_verifier:
cf8ba45e 1122 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1da177e4 1123 out_valid:
d69ee9b8
TM
1124 /* Success: notify readdir to use READDIRPLUS */
1125 nfs_advise_use_readdirplus(dir);
1126 out_valid_noent:
1da177e4 1127 dput(parent);
1e7cb3dc 1128 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
3110ff80 1129 __func__, dentry->d_parent->d_name.name,
1e7cb3dc 1130 dentry->d_name.name);
1da177e4
LT
1131 return 1;
1132out_zap_parent:
1133 nfs_zap_caches(dir);
1134 out_bad:
c44600c9
AV
1135 nfs_free_fattr(fattr);
1136 nfs_free_fhandle(fhandle);
14c43f76 1137 nfs4_label_free(label);
a1643a92 1138 nfs_mark_for_revalidate(dir);
1da177e4
LT
1139 if (inode && S_ISDIR(inode->i_mode)) {
1140 /* Purge readdir caches. */
1141 nfs_zap_caches(inode);
1142 /* If we have submounts, don't unhash ! */
1143 if (have_submounts(dentry))
1144 goto out_valid;
d9e80b7d
AV
1145 if (dentry->d_flags & DCACHE_DISCONNECTED)
1146 goto out_valid;
1da177e4
LT
1147 shrink_dcache_parent(dentry);
1148 }
1149 d_drop(dentry);
1da177e4 1150 dput(parent);
1e7cb3dc 1151 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
3110ff80 1152 __func__, dentry->d_parent->d_name.name,
1e7cb3dc 1153 dentry->d_name.name);
1da177e4 1154 return 0;
e1fb4d05
TM
1155out_error:
1156 nfs_free_fattr(fattr);
1157 nfs_free_fhandle(fhandle);
14c43f76 1158 nfs4_label_free(label);
e1fb4d05
TM
1159 dput(parent);
1160 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1161 __func__, dentry->d_parent->d_name.name,
1162 dentry->d_name.name, error);
1163 return error;
1da177e4
LT
1164}
1165
ecf3d1f1
JL
1166/*
1167 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1168 * when we don't really care about the dentry name. This is called when a
1169 * pathwalk ends on a dentry that was not found via a normal lookup in the
1170 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1171 *
1172 * In this situation, we just want to verify that the inode itself is OK
1173 * since the dentry might have changed on the server.
1174 */
1175static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1176{
1177 int error;
1178 struct inode *inode = dentry->d_inode;
1179
1180 /*
1181 * I believe we can only get a negative dentry here in the case of a
1182 * procfs-style symlink. Just assume it's correct for now, but we may
1183 * eventually need to do something more here.
1184 */
1185 if (!inode) {
1186 dfprintk(LOOKUPCACHE, "%s: %s/%s has negative inode\n",
1187 __func__, dentry->d_parent->d_name.name,
1188 dentry->d_name.name);
1189 return 1;
1190 }
1191
1192 if (is_bad_inode(inode)) {
1193 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1194 __func__, dentry->d_parent->d_name.name,
1195 dentry->d_name.name);
1196 return 0;
1197 }
1198
1199 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1200 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1201 __func__, inode->i_ino, error ? "invalid" : "valid");
1202 return !error;
1203}
1204
1da177e4
LT
1205/*
1206 * This is called from dput() when d_count is going to 0.
1207 */
fe15ce44 1208static int nfs_dentry_delete(const struct dentry *dentry)
1da177e4
LT
1209{
1210 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1211 dentry->d_parent->d_name.name, dentry->d_name.name,
1212 dentry->d_flags);
1213
77f11192
TM
1214 /* Unhash any dentry with a stale inode */
1215 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1216 return 1;
1217
1da177e4
LT
1218 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1219 /* Unhash it, so that ->d_iput() would be called */
1220 return 1;
1221 }
1222 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1223 /* Unhash it, so that ancestors of killed async unlink
1224 * files will be cleaned up during umount */
1225 return 1;
1226 }
1227 return 0;
1228
1229}
1230
1f018458 1231/* Ensure that we revalidate inode->i_nlink */
1b83d707
TM
1232static void nfs_drop_nlink(struct inode *inode)
1233{
1234 spin_lock(&inode->i_lock);
1f018458
TM
1235 /* drop the inode if we're reasonably sure this is the last link */
1236 if (inode->i_nlink == 1)
1237 clear_nlink(inode);
1238 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1b83d707
TM
1239 spin_unlock(&inode->i_lock);
1240}
1241
1da177e4
LT
1242/*
1243 * Called when the dentry loses inode.
1244 * We use it to clean up silly-renamed files.
1245 */
1246static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1247{
83672d39
NB
1248 if (S_ISDIR(inode->i_mode))
1249 /* drop any readdir cache as it could easily be old */
1250 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1251
1da177e4 1252 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
e4eff1a6 1253 nfs_complete_unlink(dentry, inode);
1f018458 1254 nfs_drop_nlink(inode);
1da177e4 1255 }
1da177e4
LT
1256 iput(inode);
1257}
1258
b1942c5f
AV
1259static void nfs_d_release(struct dentry *dentry)
1260{
1261 /* free cached devname value, if it survived that far */
1262 if (unlikely(dentry->d_fsdata)) {
1263 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1264 WARN_ON(1);
1265 else
1266 kfree(dentry->d_fsdata);
1267 }
1268}
1269
f786aa90 1270const struct dentry_operations nfs_dentry_operations = {
1da177e4 1271 .d_revalidate = nfs_lookup_revalidate,
ecf3d1f1 1272 .d_weak_revalidate = nfs_weak_revalidate,
1da177e4
LT
1273 .d_delete = nfs_dentry_delete,
1274 .d_iput = nfs_dentry_iput,
36d43a43 1275 .d_automount = nfs_d_automount,
b1942c5f 1276 .d_release = nfs_d_release,
1da177e4 1277};
ddda8e0a 1278EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1da177e4 1279
597d9289 1280struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1da177e4
LT
1281{
1282 struct dentry *res;
565277f6 1283 struct dentry *parent;
1da177e4 1284 struct inode *inode = NULL;
e1fb4d05
TM
1285 struct nfs_fh *fhandle = NULL;
1286 struct nfs_fattr *fattr = NULL;
1775fd3e 1287 struct nfs4_label *label = NULL;
1da177e4 1288 int error;
1da177e4
LT
1289
1290 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1291 dentry->d_parent->d_name.name, dentry->d_name.name);
91d5b470 1292 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1da177e4
LT
1293
1294 res = ERR_PTR(-ENAMETOOLONG);
1295 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1296 goto out;
1297
fd684071
TM
1298 /*
1299 * If we're doing an exclusive create, optimize away the lookup
1300 * but don't hash the dentry.
1301 */
00cd8dd3 1302 if (nfs_is_exclusive_create(dir, flags)) {
fd684071
TM
1303 d_instantiate(dentry, NULL);
1304 res = NULL;
fc0f684c 1305 goto out;
fd684071 1306 }
1da177e4 1307
e1fb4d05
TM
1308 res = ERR_PTR(-ENOMEM);
1309 fhandle = nfs_alloc_fhandle();
1310 fattr = nfs_alloc_fattr();
1311 if (fhandle == NULL || fattr == NULL)
1312 goto out;
1313
14c43f76
DQ
1314 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1315 if (IS_ERR(label))
1316 goto out;
1317
565277f6
TM
1318 parent = dentry->d_parent;
1319 /* Protect against concurrent sillydeletes */
6e0d0be7 1320 trace_nfs_lookup_enter(dir, dentry, flags);
565277f6 1321 nfs_block_sillyrename(parent);
1775fd3e 1322 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1da177e4
LT
1323 if (error == -ENOENT)
1324 goto no_entry;
1325 if (error < 0) {
1326 res = ERR_PTR(error);
565277f6 1327 goto out_unblock_sillyrename;
1da177e4 1328 }
1775fd3e 1329 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
bf0c84f1 1330 res = ERR_CAST(inode);
03f28e3a 1331 if (IS_ERR(res))
565277f6 1332 goto out_unblock_sillyrename;
54ceac45 1333
d69ee9b8
TM
1334 /* Success: notify readdir to use READDIRPLUS */
1335 nfs_advise_use_readdirplus(dir);
1336
1da177e4 1337no_entry:
54ceac45 1338 res = d_materialise_unique(dentry, inode);
9eaef27b
TM
1339 if (res != NULL) {
1340 if (IS_ERR(res))
565277f6 1341 goto out_unblock_sillyrename;
1da177e4 1342 dentry = res;
9eaef27b 1343 }
1da177e4 1344 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
565277f6
TM
1345out_unblock_sillyrename:
1346 nfs_unblock_sillyrename(parent);
6e0d0be7 1347 trace_nfs_lookup_exit(dir, dentry, flags, error);
14c43f76 1348 nfs4_label_free(label);
1da177e4 1349out:
e1fb4d05
TM
1350 nfs_free_fattr(fattr);
1351 nfs_free_fhandle(fhandle);
1da177e4
LT
1352 return res;
1353}
ddda8e0a 1354EXPORT_SYMBOL_GPL(nfs_lookup);
1da177e4 1355
89d77c8f 1356#if IS_ENABLED(CONFIG_NFS_V4)
0b728e19 1357static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1da177e4 1358
f786aa90 1359const struct dentry_operations nfs4_dentry_operations = {
0ef97dcf 1360 .d_revalidate = nfs4_lookup_revalidate,
1da177e4
LT
1361 .d_delete = nfs_dentry_delete,
1362 .d_iput = nfs_dentry_iput,
36d43a43 1363 .d_automount = nfs_d_automount,
b1942c5f 1364 .d_release = nfs_d_release,
1da177e4 1365};
89d77c8f 1366EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1da177e4 1367
8a5e929d
AV
1368static fmode_t flags_to_mode(int flags)
1369{
1370 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1371 if ((flags & O_ACCMODE) != O_WRONLY)
1372 res |= FMODE_READ;
1373 if ((flags & O_ACCMODE) != O_RDONLY)
1374 res |= FMODE_WRITE;
1375 return res;
1376}
1377
51141598 1378static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
cd9a1c0e 1379{
5ede7b1c 1380 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
cd9a1c0e
TM
1381}
1382
1383static int do_open(struct inode *inode, struct file *filp)
1384{
1385 nfs_fscache_set_inode_cookie(inode, filp);
1386 return 0;
1387}
1388
d9585277
AV
1389static int nfs_finish_open(struct nfs_open_context *ctx,
1390 struct dentry *dentry,
30d90494 1391 struct file *file, unsigned open_flags,
d9585277 1392 int *opened)
cd9a1c0e 1393{
0dd2b474
MS
1394 int err;
1395
30d90494
AV
1396 err = finish_open(file, dentry, do_open, opened);
1397 if (err)
d9585277 1398 goto out;
30d90494 1399 nfs_file_set_open_context(file, ctx);
0dd2b474 1400
cd9a1c0e
TM
1401out:
1402 put_nfs_open_context(ctx);
d9585277 1403 return err;
cd9a1c0e
TM
1404}
1405
73a79706
BS
1406int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1407 struct file *file, unsigned open_flags,
1408 umode_t mode, int *opened)
1da177e4 1409{
cd9a1c0e 1410 struct nfs_open_context *ctx;
0dd2b474
MS
1411 struct dentry *res;
1412 struct iattr attr = { .ia_valid = ATTR_OPEN };
f46e0bd3 1413 struct inode *inode;
1472b83e 1414 unsigned int lookup_flags = 0;
898f635c 1415 int err;
1da177e4 1416
0dd2b474
MS
1417 /* Expect a negative dentry */
1418 BUG_ON(dentry->d_inode);
1419
1420 dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1e7cb3dc
CL
1421 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1422
9597c13b
JL
1423 err = nfs_check_flags(open_flags);
1424 if (err)
1425 return err;
1426
0dd2b474
MS
1427 /* NFS only supports OPEN on regular files */
1428 if ((open_flags & O_DIRECTORY)) {
0dd2b474
MS
1429 if (!d_unhashed(dentry)) {
1430 /*
1431 * Hashed negative dentry with O_DIRECTORY: dentry was
1432 * revalidated and is fine, no need to perform lookup
1433 * again
1434 */
d9585277 1435 return -ENOENT;
0dd2b474 1436 }
1472b83e 1437 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1da177e4 1438 goto no_open;
02a913a7 1439 }
1da177e4 1440
0dd2b474 1441 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
d9585277 1442 return -ENAMETOOLONG;
cd9a1c0e 1443
0dd2b474 1444 if (open_flags & O_CREAT) {
536e43d1 1445 attr.ia_valid |= ATTR_MODE;
0dd2b474
MS
1446 attr.ia_mode = mode & ~current_umask();
1447 }
536e43d1
TM
1448 if (open_flags & O_TRUNC) {
1449 attr.ia_valid |= ATTR_SIZE;
1450 attr.ia_size = 0;
cd9a1c0e
TM
1451 }
1452
0dd2b474
MS
1453 ctx = create_nfs_open_context(dentry, open_flags);
1454 err = PTR_ERR(ctx);
1455 if (IS_ERR(ctx))
d9585277 1456 goto out;
0dd2b474 1457
6e0d0be7 1458 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
f46e0bd3 1459 nfs_block_sillyrename(dentry->d_parent);
2b484297 1460 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
275bb307 1461 nfs_unblock_sillyrename(dentry->d_parent);
f46e0bd3 1462 if (IS_ERR(inode)) {
cd9a1c0e 1463 put_nfs_open_context(ctx);
0dd2b474 1464 err = PTR_ERR(inode);
6e0d0be7 1465 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
0dd2b474
MS
1466 switch (err) {
1467 case -ENOENT:
275bb307 1468 d_drop(dentry);
0dd2b474
MS
1469 d_add(dentry, NULL);
1470 break;
1471 case -EISDIR:
1472 case -ENOTDIR:
1473 goto no_open;
1474 case -ELOOP:
1475 if (!(open_flags & O_NOFOLLOW))
6f926b5b 1476 goto no_open;
0dd2b474 1477 break;
1da177e4 1478 /* case -EINVAL: */
0dd2b474
MS
1479 default:
1480 break;
1da177e4 1481 }
d9585277 1482 goto out;
cd9a1c0e 1483 }
0dd2b474 1484
275bb307 1485 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
6e0d0be7 1486 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
d9585277
AV
1487out:
1488 return err;
0dd2b474 1489
1da177e4 1490no_open:
1472b83e 1491 res = nfs_lookup(dir, dentry, lookup_flags);
0dd2b474
MS
1492 err = PTR_ERR(res);
1493 if (IS_ERR(res))
d9585277 1494 goto out;
0dd2b474 1495
e45198a6 1496 return finish_no_open(file, res);
1da177e4 1497}
89d77c8f 1498EXPORT_SYMBOL_GPL(nfs_atomic_open);
1da177e4 1499
0b728e19 1500static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1da177e4
LT
1501{
1502 struct dentry *parent = NULL;
657e94b6 1503 struct inode *inode;
1da177e4 1504 struct inode *dir;
50de348c 1505 int ret = 0;
1da177e4 1506
fa3c56bb 1507 if (flags & LOOKUP_RCU)
657e94b6
NP
1508 return -ECHILD;
1509
fa3c56bb 1510 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
eda72afb
MS
1511 goto no_open;
1512 if (d_mountpoint(dentry))
5584c306 1513 goto no_open;
49f9a0fa
TM
1514 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1515 goto no_open;
2b484297 1516
eda72afb 1517 inode = dentry->d_inode;
1da177e4
LT
1518 parent = dget_parent(dentry);
1519 dir = parent->d_inode;
2b484297 1520
1da177e4
LT
1521 /* We can't create new files in nfs_open_revalidate(), so we
1522 * optimize away revalidation of negative dentries.
1523 */
216d5d06 1524 if (inode == NULL) {
fa3c56bb 1525 if (!nfs_neg_need_reval(dir, dentry, flags))
216d5d06 1526 ret = 1;
1da177e4 1527 goto out;
216d5d06
TM
1528 }
1529
1da177e4
LT
1530 /* NFS only supports OPEN on regular files */
1531 if (!S_ISREG(inode->i_mode))
5584c306 1532 goto no_open_dput;
1da177e4 1533 /* We cannot do exclusive creation on a positive dentry */
fa3c56bb 1534 if (flags & LOOKUP_EXCL)
5584c306 1535 goto no_open_dput;
1da177e4 1536
0ef97dcf
MS
1537 /* Let f_op->open() actually open (and revalidate) the file */
1538 ret = 1;
536e43d1 1539
1da177e4
LT
1540out:
1541 dput(parent);
1da177e4 1542 return ret;
535918f1 1543
5584c306 1544no_open_dput:
1da177e4 1545 dput(parent);
5584c306 1546no_open:
0b728e19 1547 return nfs_lookup_revalidate(dentry, flags);
c0204fd2
TM
1548}
1549
1da177e4
LT
1550#endif /* CONFIG_NFSV4 */
1551
1da177e4
LT
1552/*
1553 * Code common to create, mkdir, and mknod.
1554 */
1555int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1775fd3e
DQ
1556 struct nfs_fattr *fattr,
1557 struct nfs4_label *label)
1da177e4 1558{
fab728e1
TM
1559 struct dentry *parent = dget_parent(dentry);
1560 struct inode *dir = parent->d_inode;
1da177e4
LT
1561 struct inode *inode;
1562 int error = -EACCES;
1563
fab728e1
TM
1564 d_drop(dentry);
1565
1da177e4
LT
1566 /* We may have been initialized further down */
1567 if (dentry->d_inode)
fab728e1 1568 goto out;
1da177e4 1569 if (fhandle->size == 0) {
1775fd3e 1570 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1da177e4 1571 if (error)
fab728e1 1572 goto out_error;
1da177e4 1573 }
5724ab37 1574 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1da177e4
LT
1575 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1576 struct nfs_server *server = NFS_SB(dentry->d_sb);
1775fd3e 1577 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1da177e4 1578 if (error < 0)
fab728e1 1579 goto out_error;
1da177e4 1580 }
1775fd3e 1581 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
03f28e3a
TM
1582 error = PTR_ERR(inode);
1583 if (IS_ERR(inode))
fab728e1
TM
1584 goto out_error;
1585 d_add(dentry, inode);
1586out:
1587 dput(parent);
1da177e4 1588 return 0;
fab728e1
TM
1589out_error:
1590 nfs_mark_for_revalidate(dir);
1591 dput(parent);
1592 return error;
1da177e4 1593}
ddda8e0a 1594EXPORT_SYMBOL_GPL(nfs_instantiate);
1da177e4
LT
1595
1596/*
1597 * Following a failed create operation, we drop the dentry rather
1598 * than retain a negative dentry. This avoids a problem in the event
1599 * that the operation succeeded on the server, but an error in the
1600 * reply path made it appear to have failed.
1601 */
597d9289 1602int nfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 1603 umode_t mode, bool excl)
1da177e4
LT
1604{
1605 struct iattr attr;
ebfc3b49 1606 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1da177e4 1607 int error;
1da177e4 1608
1e7cb3dc
CL
1609 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1610 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1da177e4
LT
1611
1612 attr.ia_mode = mode;
1613 attr.ia_valid = ATTR_MODE;
1614
8b0ad3d4 1615 trace_nfs_create_enter(dir, dentry, open_flags);
8867fe58 1616 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
8b0ad3d4 1617 trace_nfs_create_exit(dir, dentry, open_flags, error);
1da177e4
LT
1618 if (error != 0)
1619 goto out_err;
1da177e4
LT
1620 return 0;
1621out_err:
1da177e4
LT
1622 d_drop(dentry);
1623 return error;
1624}
ddda8e0a 1625EXPORT_SYMBOL_GPL(nfs_create);
1da177e4
LT
1626
1627/*
1628 * See comments for nfs_proc_create regarding failed operations.
1629 */
597d9289 1630int
1a67aafb 1631nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1da177e4
LT
1632{
1633 struct iattr attr;
1634 int status;
1635
1e7cb3dc
CL
1636 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1637 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1da177e4
LT
1638
1639 if (!new_valid_dev(rdev))
1640 return -EINVAL;
1641
1642 attr.ia_mode = mode;
1643 attr.ia_valid = ATTR_MODE;
1644
1ca42382 1645 trace_nfs_mknod_enter(dir, dentry);
1da177e4 1646 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1ca42382 1647 trace_nfs_mknod_exit(dir, dentry, status);
1da177e4
LT
1648 if (status != 0)
1649 goto out_err;
1da177e4
LT
1650 return 0;
1651out_err:
1da177e4
LT
1652 d_drop(dentry);
1653 return status;
1654}
ddda8e0a 1655EXPORT_SYMBOL_GPL(nfs_mknod);
1da177e4
LT
1656
1657/*
1658 * See comments for nfs_proc_create regarding failed operations.
1659 */
597d9289 1660int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4
LT
1661{
1662 struct iattr attr;
1663 int error;
1664
1e7cb3dc
CL
1665 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1666 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1da177e4
LT
1667
1668 attr.ia_valid = ATTR_MODE;
1669 attr.ia_mode = mode | S_IFDIR;
1670
1ca42382 1671 trace_nfs_mkdir_enter(dir, dentry);
1da177e4 1672 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1ca42382 1673 trace_nfs_mkdir_exit(dir, dentry, error);
1da177e4
LT
1674 if (error != 0)
1675 goto out_err;
1da177e4
LT
1676 return 0;
1677out_err:
1678 d_drop(dentry);
1da177e4
LT
1679 return error;
1680}
ddda8e0a 1681EXPORT_SYMBOL_GPL(nfs_mkdir);
1da177e4 1682
d45b9d8b
TM
1683static void nfs_dentry_handle_enoent(struct dentry *dentry)
1684{
1685 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1686 d_delete(dentry);
1687}
1688
597d9289 1689int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1da177e4
LT
1690{
1691 int error;
1692
1e7cb3dc
CL
1693 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1694 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1da177e4 1695
1ca42382 1696 trace_nfs_rmdir_enter(dir, dentry);
1da177e4
LT
1697 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1698 /* Ensure the VFS deletes this inode */
1699 if (error == 0 && dentry->d_inode != NULL)
ce71ec36 1700 clear_nlink(dentry->d_inode);
d45b9d8b
TM
1701 else if (error == -ENOENT)
1702 nfs_dentry_handle_enoent(dentry);
1ca42382 1703 trace_nfs_rmdir_exit(dir, dentry, error);
1da177e4
LT
1704
1705 return error;
1706}
ddda8e0a 1707EXPORT_SYMBOL_GPL(nfs_rmdir);
1da177e4 1708
1da177e4
LT
1709/*
1710 * Remove a file after making sure there are no pending writes,
1711 * and after checking that the file has only one user.
1712 *
1713 * We invalidate the attribute cache and free the inode prior to the operation
1714 * to avoid possible races if the server reuses the inode.
1715 */
1716static int nfs_safe_remove(struct dentry *dentry)
1717{
1718 struct inode *dir = dentry->d_parent->d_inode;
1719 struct inode *inode = dentry->d_inode;
1720 int error = -EBUSY;
1721
1722 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1723 dentry->d_parent->d_name.name, dentry->d_name.name);
1724
1725 /* If the dentry was sillyrenamed, we simply call d_delete() */
1726 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1727 error = 0;
1728 goto out;
1729 }
1730
1ca42382 1731 trace_nfs_remove_enter(dir, dentry);
1da177e4 1732 if (inode != NULL) {
57ec14c5 1733 NFS_PROTO(inode)->return_delegation(inode);
1da177e4 1734 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1da177e4 1735 if (error == 0)
1b83d707 1736 nfs_drop_nlink(inode);
1da177e4
LT
1737 } else
1738 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
d45b9d8b
TM
1739 if (error == -ENOENT)
1740 nfs_dentry_handle_enoent(dentry);
1ca42382 1741 trace_nfs_remove_exit(dir, dentry, error);
1da177e4
LT
1742out:
1743 return error;
1744}
1745
1746/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1747 * belongs to an active ".nfs..." file and we return -EBUSY.
1748 *
1749 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1750 */
597d9289 1751int nfs_unlink(struct inode *dir, struct dentry *dentry)
1da177e4
LT
1752{
1753 int error;
1754 int need_rehash = 0;
1755
1756 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1757 dir->i_ino, dentry->d_name.name);
1758
1ca42382 1759 trace_nfs_unlink_enter(dir, dentry);
1da177e4 1760 spin_lock(&dentry->d_lock);
84d08fa8 1761 if (d_count(dentry) > 1) {
1da177e4 1762 spin_unlock(&dentry->d_lock);
ccfeb506
TM
1763 /* Start asynchronous writeout of the inode */
1764 write_inode_now(dentry->d_inode, 0);
1da177e4 1765 error = nfs_sillyrename(dir, dentry);
1ca42382 1766 goto out;
1da177e4
LT
1767 }
1768 if (!d_unhashed(dentry)) {
1769 __d_drop(dentry);
1770 need_rehash = 1;
1771 }
1772 spin_unlock(&dentry->d_lock);
1da177e4 1773 error = nfs_safe_remove(dentry);
d45b9d8b 1774 if (!error || error == -ENOENT) {
1da177e4
LT
1775 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1776 } else if (need_rehash)
1777 d_rehash(dentry);
1ca42382
TM
1778out:
1779 trace_nfs_unlink_exit(dir, dentry, error);
1da177e4
LT
1780 return error;
1781}
ddda8e0a 1782EXPORT_SYMBOL_GPL(nfs_unlink);
1da177e4 1783
873101b3
CL
1784/*
1785 * To create a symbolic link, most file systems instantiate a new inode,
1786 * add a page to it containing the path, then write it out to the disk
1787 * using prepare_write/commit_write.
1788 *
1789 * Unfortunately the NFS client can't create the in-core inode first
1790 * because it needs a file handle to create an in-core inode (see
1791 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1792 * symlink request has completed on the server.
1793 *
1794 * So instead we allocate a raw page, copy the symname into it, then do
1795 * the SYMLINK request with the page as the buffer. If it succeeds, we
1796 * now have a new file handle and can instantiate an in-core NFS inode
1797 * and move the raw page into its mapping.
1798 */
597d9289 1799int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1da177e4 1800{
873101b3
CL
1801 struct page *page;
1802 char *kaddr;
1da177e4 1803 struct iattr attr;
873101b3 1804 unsigned int pathlen = strlen(symname);
1da177e4
LT
1805 int error;
1806
1807 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1808 dir->i_ino, dentry->d_name.name, symname);
1809
873101b3
CL
1810 if (pathlen > PAGE_SIZE)
1811 return -ENAMETOOLONG;
1da177e4 1812
873101b3
CL
1813 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1814 attr.ia_valid = ATTR_MODE;
1da177e4 1815
83d93f22 1816 page = alloc_page(GFP_HIGHUSER);
76566991 1817 if (!page)
873101b3 1818 return -ENOMEM;
873101b3 1819
2b86ce2d 1820 kaddr = kmap_atomic(page);
873101b3
CL
1821 memcpy(kaddr, symname, pathlen);
1822 if (pathlen < PAGE_SIZE)
1823 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2b86ce2d 1824 kunmap_atomic(kaddr);
873101b3 1825
1ca42382 1826 trace_nfs_symlink_enter(dir, dentry);
94a6d753 1827 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1ca42382 1828 trace_nfs_symlink_exit(dir, dentry, error);
873101b3
CL
1829 if (error != 0) {
1830 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1831 dir->i_sb->s_id, dir->i_ino,
1832 dentry->d_name.name, symname, error);
1da177e4 1833 d_drop(dentry);
873101b3 1834 __free_page(page);
873101b3
CL
1835 return error;
1836 }
1837
1838 /*
1839 * No big deal if we can't add this page to the page cache here.
1840 * READLINK will get the missing page from the server if needed.
1841 */
a0b8cab3 1842 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
873101b3 1843 GFP_KERNEL)) {
873101b3
CL
1844 SetPageUptodate(page);
1845 unlock_page(page);
1846 } else
1847 __free_page(page);
1848
873101b3 1849 return 0;
1da177e4 1850}
ddda8e0a 1851EXPORT_SYMBOL_GPL(nfs_symlink);
1da177e4 1852
597d9289 1853int
1da177e4
LT
1854nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1855{
1856 struct inode *inode = old_dentry->d_inode;
1857 int error;
1858
1859 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1860 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1861 dentry->d_parent->d_name.name, dentry->d_name.name);
1862
57ec14c5 1863 NFS_PROTO(inode)->return_delegation(inode);
9a3936aa 1864
9697d234 1865 d_drop(dentry);
1da177e4 1866 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
cf809556 1867 if (error == 0) {
7de9c6ee 1868 ihold(inode);
9697d234 1869 d_add(dentry, inode);
cf809556 1870 }
1da177e4
LT
1871 return error;
1872}
ddda8e0a 1873EXPORT_SYMBOL_GPL(nfs_link);
1da177e4
LT
1874
1875/*
1876 * RENAME
1877 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1878 * different file handle for the same inode after a rename (e.g. when
1879 * moving to a different directory). A fail-safe method to do so would
1880 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1881 * rename the old file using the sillyrename stuff. This way, the original
1882 * file in old_dir will go away when the last process iput()s the inode.
1883 *
1884 * FIXED.
1885 *
1886 * It actually works quite well. One needs to have the possibility for
1887 * at least one ".nfs..." file in each directory the file ever gets
1888 * moved or linked to which happens automagically with the new
1889 * implementation that only depends on the dcache stuff instead of
1890 * using the inode layer
1891 *
1892 * Unfortunately, things are a little more complicated than indicated
1893 * above. For a cross-directory move, we want to make sure we can get
1894 * rid of the old inode after the operation. This means there must be
1895 * no pending writes (if it's a file), and the use count must be 1.
1896 * If these conditions are met, we can drop the dentries before doing
1897 * the rename.
1898 */
597d9289 1899int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1da177e4
LT
1900 struct inode *new_dir, struct dentry *new_dentry)
1901{
1902 struct inode *old_inode = old_dentry->d_inode;
1903 struct inode *new_inode = new_dentry->d_inode;
1904 struct dentry *dentry = NULL, *rehash = NULL;
1905 int error = -EBUSY;
1906
1da177e4
LT
1907 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1908 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1909 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
84d08fa8 1910 d_count(new_dentry));
1da177e4 1911
70ded201 1912 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1da177e4 1913 /*
28f79a1a
MS
1914 * For non-directories, check whether the target is busy and if so,
1915 * make a copy of the dentry and then do a silly-rename. If the
1916 * silly-rename succeeds, the copied dentry is hashed and becomes
1917 * the new target.
1da177e4 1918 */
27226104
MS
1919 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1920 /*
1921 * To prevent any new references to the target during the
1922 * rename, we unhash the dentry in advance.
1923 */
1924 if (!d_unhashed(new_dentry)) {
1925 d_drop(new_dentry);
1926 rehash = new_dentry;
1927 }
1da177e4 1928
84d08fa8 1929 if (d_count(new_dentry) > 2) {
27226104
MS
1930 int err;
1931
1932 /* copy the target dentry's name */
1933 dentry = d_alloc(new_dentry->d_parent,
1934 &new_dentry->d_name);
1935 if (!dentry)
1936 goto out;
1937
1938 /* silly-rename the existing target ... */
1939 err = nfs_sillyrename(new_dir, new_dentry);
24e93025 1940 if (err)
27226104 1941 goto out;
24e93025
MS
1942
1943 new_dentry = dentry;
56335936 1944 rehash = NULL;
24e93025 1945 new_inode = NULL;
27226104 1946 }
b1e4adf4 1947 }
1da177e4 1948
57ec14c5 1949 NFS_PROTO(old_inode)->return_delegation(old_inode);
b1e4adf4 1950 if (new_inode != NULL)
57ec14c5 1951 NFS_PROTO(new_inode)->return_delegation(new_inode);
1da177e4 1952
1da177e4
LT
1953 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1954 new_dir, &new_dentry->d_name);
5ba7cc48 1955 nfs_mark_for_revalidate(old_inode);
1da177e4
LT
1956out:
1957 if (rehash)
1958 d_rehash(rehash);
70ded201
TM
1959 trace_nfs_rename_exit(old_dir, old_dentry,
1960 new_dir, new_dentry, error);
1da177e4 1961 if (!error) {
b1e4adf4
TM
1962 if (new_inode != NULL)
1963 nfs_drop_nlink(new_inode);
349457cc 1964 d_move(old_dentry, new_dentry);
8fb559f8
CL
1965 nfs_set_verifier(new_dentry,
1966 nfs_save_change_attribute(new_dir));
d45b9d8b
TM
1967 } else if (error == -ENOENT)
1968 nfs_dentry_handle_enoent(old_dentry);
1da177e4
LT
1969
1970 /* new dentry created? */
1971 if (dentry)
1972 dput(dentry);
1da177e4
LT
1973 return error;
1974}
ddda8e0a 1975EXPORT_SYMBOL_GPL(nfs_rename);
1da177e4 1976
cfcea3e8
TM
1977static DEFINE_SPINLOCK(nfs_access_lru_lock);
1978static LIST_HEAD(nfs_access_lru_list);
1979static atomic_long_t nfs_access_nr_entries;
1980
1c3c07e9
TM
1981static void nfs_access_free_entry(struct nfs_access_entry *entry)
1982{
1983 put_rpccred(entry->cred);
1984 kfree(entry);
cfcea3e8
TM
1985 smp_mb__before_atomic_dec();
1986 atomic_long_dec(&nfs_access_nr_entries);
1987 smp_mb__after_atomic_dec();
1c3c07e9
TM
1988}
1989
1a81bb8a
TM
1990static void nfs_access_free_list(struct list_head *head)
1991{
1992 struct nfs_access_entry *cache;
1993
1994 while (!list_empty(head)) {
1995 cache = list_entry(head->next, struct nfs_access_entry, lru);
1996 list_del(&cache->lru);
1997 nfs_access_free_entry(cache);
1998 }
1999}
2000
1495f230
YH
2001int nfs_access_cache_shrinker(struct shrinker *shrink,
2002 struct shrink_control *sc)
979df72e
TM
2003{
2004 LIST_HEAD(head);
aa510da5 2005 struct nfs_inode *nfsi, *next;
979df72e 2006 struct nfs_access_entry *cache;
1495f230
YH
2007 int nr_to_scan = sc->nr_to_scan;
2008 gfp_t gfp_mask = sc->gfp_mask;
979df72e 2009
61d5eb29
TM
2010 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2011 return (nr_to_scan == 0) ? 0 : -1;
9c7e7e23 2012
a50f7951 2013 spin_lock(&nfs_access_lru_lock);
aa510da5 2014 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
979df72e
TM
2015 struct inode *inode;
2016
2017 if (nr_to_scan-- == 0)
2018 break;
9c7e7e23 2019 inode = &nfsi->vfs_inode;
979df72e
TM
2020 spin_lock(&inode->i_lock);
2021 if (list_empty(&nfsi->access_cache_entry_lru))
2022 goto remove_lru_entry;
2023 cache = list_entry(nfsi->access_cache_entry_lru.next,
2024 struct nfs_access_entry, lru);
2025 list_move(&cache->lru, &head);
2026 rb_erase(&cache->rb_node, &nfsi->access_cache);
2027 if (!list_empty(&nfsi->access_cache_entry_lru))
2028 list_move_tail(&nfsi->access_cache_inode_lru,
2029 &nfs_access_lru_list);
2030 else {
2031remove_lru_entry:
2032 list_del_init(&nfsi->access_cache_inode_lru);
9c7e7e23 2033 smp_mb__before_clear_bit();
979df72e 2034 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
9c7e7e23 2035 smp_mb__after_clear_bit();
979df72e 2036 }
59844a9b 2037 spin_unlock(&inode->i_lock);
979df72e
TM
2038 }
2039 spin_unlock(&nfs_access_lru_lock);
1a81bb8a 2040 nfs_access_free_list(&head);
979df72e
TM
2041 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2042}
2043
1a81bb8a 2044static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
1da177e4 2045{
1c3c07e9 2046 struct rb_root *root_node = &nfsi->access_cache;
1a81bb8a 2047 struct rb_node *n;
1c3c07e9
TM
2048 struct nfs_access_entry *entry;
2049
2050 /* Unhook entries from the cache */
2051 while ((n = rb_first(root_node)) != NULL) {
2052 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2053 rb_erase(n, root_node);
1a81bb8a 2054 list_move(&entry->lru, head);
1c3c07e9
TM
2055 }
2056 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1da177e4
LT
2057}
2058
1c3c07e9 2059void nfs_access_zap_cache(struct inode *inode)
1da177e4 2060{
1a81bb8a
TM
2061 LIST_HEAD(head);
2062
2063 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2064 return;
cfcea3e8 2065 /* Remove from global LRU init */
1a81bb8a
TM
2066 spin_lock(&nfs_access_lru_lock);
2067 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
cfcea3e8 2068 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
cfcea3e8 2069
1c3c07e9 2070 spin_lock(&inode->i_lock);
1a81bb8a
TM
2071 __nfs_access_zap_cache(NFS_I(inode), &head);
2072 spin_unlock(&inode->i_lock);
2073 spin_unlock(&nfs_access_lru_lock);
2074 nfs_access_free_list(&head);
1c3c07e9 2075}
1c606fb7 2076EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
1da177e4 2077
1c3c07e9
TM
2078static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2079{
2080 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2081 struct nfs_access_entry *entry;
2082
2083 while (n != NULL) {
2084 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2085
2086 if (cred < entry->cred)
2087 n = n->rb_left;
2088 else if (cred > entry->cred)
2089 n = n->rb_right;
2090 else
2091 return entry;
1da177e4 2092 }
1c3c07e9
TM
2093 return NULL;
2094}
2095
af22f94a 2096static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1c3c07e9
TM
2097{
2098 struct nfs_inode *nfsi = NFS_I(inode);
2099 struct nfs_access_entry *cache;
2100 int err = -ENOENT;
2101
dc59250c 2102 spin_lock(&inode->i_lock);
1c3c07e9
TM
2103 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2104 goto out_zap;
2105 cache = nfs_access_search_rbtree(inode, cred);
2106 if (cache == NULL)
2107 goto out;
b4d2314b 2108 if (!nfs_have_delegated_attributes(inode) &&
64672d55 2109 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1c3c07e9
TM
2110 goto out_stale;
2111 res->jiffies = cache->jiffies;
2112 res->cred = cache->cred;
2113 res->mask = cache->mask;
cfcea3e8 2114 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1c3c07e9
TM
2115 err = 0;
2116out:
2117 spin_unlock(&inode->i_lock);
2118 return err;
2119out_stale:
2120 rb_erase(&cache->rb_node, &nfsi->access_cache);
cfcea3e8 2121 list_del(&cache->lru);
1c3c07e9
TM
2122 spin_unlock(&inode->i_lock);
2123 nfs_access_free_entry(cache);
2124 return -ENOENT;
2125out_zap:
1a81bb8a
TM
2126 spin_unlock(&inode->i_lock);
2127 nfs_access_zap_cache(inode);
1c3c07e9
TM
2128 return -ENOENT;
2129}
2130
2131static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2132{
cfcea3e8
TM
2133 struct nfs_inode *nfsi = NFS_I(inode);
2134 struct rb_root *root_node = &nfsi->access_cache;
1c3c07e9
TM
2135 struct rb_node **p = &root_node->rb_node;
2136 struct rb_node *parent = NULL;
2137 struct nfs_access_entry *entry;
2138
2139 spin_lock(&inode->i_lock);
2140 while (*p != NULL) {
2141 parent = *p;
2142 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2143
2144 if (set->cred < entry->cred)
2145 p = &parent->rb_left;
2146 else if (set->cred > entry->cred)
2147 p = &parent->rb_right;
2148 else
2149 goto found;
2150 }
2151 rb_link_node(&set->rb_node, parent, p);
2152 rb_insert_color(&set->rb_node, root_node);
cfcea3e8 2153 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
dc59250c 2154 spin_unlock(&inode->i_lock);
1c3c07e9
TM
2155 return;
2156found:
2157 rb_replace_node(parent, &set->rb_node, root_node);
cfcea3e8
TM
2158 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2159 list_del(&entry->lru);
1c3c07e9
TM
2160 spin_unlock(&inode->i_lock);
2161 nfs_access_free_entry(entry);
2162}
2163
6168f62c 2164void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1c3c07e9
TM
2165{
2166 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2167 if (cache == NULL)
2168 return;
2169 RB_CLEAR_NODE(&cache->rb_node);
1da177e4 2170 cache->jiffies = set->jiffies;
1c3c07e9 2171 cache->cred = get_rpccred(set->cred);
1da177e4 2172 cache->mask = set->mask;
1c3c07e9
TM
2173
2174 nfs_access_add_rbtree(inode, cache);
cfcea3e8
TM
2175
2176 /* Update accounting */
2177 smp_mb__before_atomic_inc();
2178 atomic_long_inc(&nfs_access_nr_entries);
2179 smp_mb__after_atomic_inc();
2180
2181 /* Add inode to global LRU list */
1a81bb8a 2182 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
cfcea3e8 2183 spin_lock(&nfs_access_lru_lock);
1a81bb8a
TM
2184 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2185 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2186 &nfs_access_lru_list);
cfcea3e8
TM
2187 spin_unlock(&nfs_access_lru_lock);
2188 }
1da177e4 2189}
6168f62c
WAA
2190EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2191
2192void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2193{
2194 entry->mask = 0;
2195 if (access_result & NFS4_ACCESS_READ)
2196 entry->mask |= MAY_READ;
2197 if (access_result &
2198 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2199 entry->mask |= MAY_WRITE;
2200 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2201 entry->mask |= MAY_EXEC;
2202}
2203EXPORT_SYMBOL_GPL(nfs_access_set_mask);
1da177e4
LT
2204
2205static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2206{
2207 struct nfs_access_entry cache;
2208 int status;
2209
f4ce1299
TM
2210 trace_nfs_access_enter(inode);
2211
1da177e4
LT
2212 status = nfs_access_get_cached(inode, cred, &cache);
2213 if (status == 0)
f4ce1299 2214 goto out_cached;
1da177e4
LT
2215
2216 /* Be clever: ask server to check for all possible rights */
2217 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2218 cache.cred = cred;
2219 cache.jiffies = jiffies;
2220 status = NFS_PROTO(inode)->access(inode, &cache);
a71ee337
SJ
2221 if (status != 0) {
2222 if (status == -ESTALE) {
2223 nfs_zap_caches(inode);
2224 if (!S_ISDIR(inode->i_mode))
2225 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2226 }
f4ce1299 2227 goto out;
a71ee337 2228 }
1da177e4 2229 nfs_access_add_cache(inode, &cache);
f4ce1299
TM
2230out_cached:
2231 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2232 status = -EACCES;
1da177e4 2233out:
f4ce1299
TM
2234 trace_nfs_access_exit(inode, status);
2235 return status;
1da177e4
LT
2236}
2237
af22f94a
TM
2238static int nfs_open_permission_mask(int openflags)
2239{
2240 int mask = 0;
2241
f8d9a897
WAA
2242 if (openflags & __FMODE_EXEC) {
2243 /* ONLY check exec rights */
2244 mask = MAY_EXEC;
2245 } else {
2246 if ((openflags & O_ACCMODE) != O_WRONLY)
2247 mask |= MAY_READ;
2248 if ((openflags & O_ACCMODE) != O_RDONLY)
2249 mask |= MAY_WRITE;
2250 }
2251
af22f94a
TM
2252 return mask;
2253}
2254
2255int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2256{
2257 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2258}
89d77c8f 2259EXPORT_SYMBOL_GPL(nfs_may_open);
af22f94a 2260
10556cb2 2261int nfs_permission(struct inode *inode, int mask)
1da177e4
LT
2262{
2263 struct rpc_cred *cred;
2264 int res = 0;
2265
10556cb2 2266 if (mask & MAY_NOT_BLOCK)
b74c79e9
NP
2267 return -ECHILD;
2268
91d5b470
CL
2269 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2270
e6305c43 2271 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4
LT
2272 goto out;
2273 /* Is this sys_access() ? */
9cfcac81 2274 if (mask & (MAY_ACCESS | MAY_CHDIR))
1da177e4
LT
2275 goto force_lookup;
2276
2277 switch (inode->i_mode & S_IFMT) {
2278 case S_IFLNK:
2279 goto out;
2280 case S_IFREG:
1da177e4
LT
2281 break;
2282 case S_IFDIR:
2283 /*
2284 * Optimize away all write operations, since the server
2285 * will check permissions when we perform the op.
2286 */
2287 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2288 goto out;
2289 }
2290
2291force_lookup:
1da177e4
LT
2292 if (!NFS_PROTO(inode)->access)
2293 goto out_notsup;
2294
98a8e323 2295 cred = rpc_lookup_cred();
1da177e4
LT
2296 if (!IS_ERR(cred)) {
2297 res = nfs_do_access(inode, cred, mask);
2298 put_rpccred(cred);
2299 } else
2300 res = PTR_ERR(cred);
1da177e4 2301out:
f696a365
MS
2302 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2303 res = -EACCES;
2304
1e7cb3dc
CL
2305 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2306 inode->i_sb->s_id, inode->i_ino, mask, res);
1da177e4
LT
2307 return res;
2308out_notsup:
2309 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2310 if (res == 0)
2830ba7f 2311 res = generic_permission(inode, mask);
1e7cb3dc 2312 goto out;
1da177e4 2313}
ddda8e0a 2314EXPORT_SYMBOL_GPL(nfs_permission);
1da177e4
LT
2315
2316/*
2317 * Local variables:
2318 * version-control: t
2319 * kept-new-versions: 5
2320 * End:
2321 */