]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/nfs/dir.c
nfs: make nfs_async_rename non-static
[mirror_ubuntu-zesty-kernel.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 71{
311324ad 72 struct nfs_inode *nfsi = NFS_I(dir);
480c2006
BS
73 struct nfs_open_dir_context *ctx;
74 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
75 if (ctx != NULL) {
8ef2ce3e 76 ctx->duped = 0;
311324ad 77 ctx->attr_gencount = nfsi->attr_gencount;
480c2006 78 ctx->dir_cookie = 0;
8ef2ce3e 79 ctx->dup_cookie = 0;
480c2006 80 ctx->cred = get_rpccred(cred);
311324ad
TM
81 spin_lock(&dir->i_lock);
82 list_add(&ctx->list, &nfsi->open_files);
83 spin_unlock(&dir->i_lock);
0c030806
TM
84 return ctx;
85 }
86 return ERR_PTR(-ENOMEM);
480c2006
BS
87}
88
311324ad 89static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
480c2006 90{
311324ad
TM
91 spin_lock(&dir->i_lock);
92 list_del(&ctx->list);
93 spin_unlock(&dir->i_lock);
480c2006
BS
94 put_rpccred(ctx->cred);
95 kfree(ctx);
96}
97
1da177e4
LT
98/*
99 * Open file
100 */
101static int
102nfs_opendir(struct inode *inode, struct file *filp)
103{
480c2006
BS
104 int res = 0;
105 struct nfs_open_dir_context *ctx;
106 struct rpc_cred *cred;
1da177e4 107
6de1472f 108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
cc0dd2d1
CL
109
110 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
1e7cb3dc 111
480c2006
BS
112 cred = rpc_lookup_cred();
113 if (IS_ERR(cred))
114 return PTR_ERR(cred);
0c030806 115 ctx = alloc_nfs_open_dir_context(inode, cred);
480c2006
BS
116 if (IS_ERR(ctx)) {
117 res = PTR_ERR(ctx);
118 goto out;
119 }
120 filp->private_data = ctx;
f5a73672
NB
121 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
122 /* This is a mountpoint, so d_revalidate will never
123 * have been called, so we need to refresh the
124 * inode (for close-open consistency) ourselves.
125 */
126 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
127 }
480c2006
BS
128out:
129 put_rpccred(cred);
1da177e4
LT
130 return res;
131}
132
480c2006
BS
133static int
134nfs_closedir(struct inode *inode, struct file *filp)
135{
311324ad 136 put_nfs_open_dir_context(filp->f_path.dentry->d_inode, filp->private_data);
480c2006
BS
137 return 0;
138}
139
d1bacf9e
BS
140struct nfs_cache_array_entry {
141 u64 cookie;
142 u64 ino;
143 struct qstr string;
0b26a0bf 144 unsigned char d_type;
d1bacf9e
BS
145};
146
147struct nfs_cache_array {
88b8e133 148 int size;
d1bacf9e
BS
149 int eof_index;
150 u64 last_cookie;
151 struct nfs_cache_array_entry array[0];
152};
153
573c4e1e 154typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
1da177e4
LT
155typedef struct {
156 struct file *file;
157 struct page *page;
23db8620 158 struct dir_context *ctx;
1da177e4 159 unsigned long page_index;
f0dd2136 160 u64 *dir_cookie;
0aded708 161 u64 last_cookie;
f0dd2136 162 loff_t current_index;
1da177e4 163 decode_dirent_t decode;
d1bacf9e 164
1f4eab7e 165 unsigned long timestamp;
4704f0e2 166 unsigned long gencount;
d1bacf9e
BS
167 unsigned int cache_entry_index;
168 unsigned int plus:1;
169 unsigned int eof:1;
1da177e4
LT
170} nfs_readdir_descriptor_t;
171
d1bacf9e
BS
172/*
173 * The caller is responsible for calling nfs_readdir_release_array(page)
1da177e4
LT
174 */
175static
d1bacf9e
BS
176struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
177{
8cd51a0c 178 void *ptr;
d1bacf9e
BS
179 if (page == NULL)
180 return ERR_PTR(-EIO);
8cd51a0c
TM
181 ptr = kmap(page);
182 if (ptr == NULL)
183 return ERR_PTR(-ENOMEM);
184 return ptr;
d1bacf9e
BS
185}
186
187static
188void nfs_readdir_release_array(struct page *page)
189{
190 kunmap(page);
191}
192
193/*
194 * we are freeing strings created by nfs_add_to_readdir_array()
195 */
196static
11de3b11 197void nfs_readdir_clear_array(struct page *page)
d1bacf9e 198{
11de3b11 199 struct nfs_cache_array *array;
d1bacf9e 200 int i;
8cd51a0c 201
2b86ce2d 202 array = kmap_atomic(page);
d1bacf9e
BS
203 for (i = 0; i < array->size; i++)
204 kfree(array->array[i].string.name);
2b86ce2d 205 kunmap_atomic(array);
d1bacf9e
BS
206}
207
208/*
209 * the caller is responsible for freeing qstr.name
210 * when called by nfs_readdir_add_to_array, the strings will be freed in
211 * nfs_clear_readdir_array()
212 */
213static
4a201d6e 214int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
d1bacf9e
BS
215{
216 string->len = len;
217 string->name = kmemdup(name, len, GFP_KERNEL);
4a201d6e
TM
218 if (string->name == NULL)
219 return -ENOMEM;
04e4bd1c
CM
220 /*
221 * Avoid a kmemleak false positive. The pointer to the name is stored
222 * in a page cache page which kmemleak does not scan.
223 */
224 kmemleak_not_leak(string->name);
4a201d6e
TM
225 string->hash = full_name_hash(name, len);
226 return 0;
d1bacf9e
BS
227}
228
229static
230int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
231{
232 struct nfs_cache_array *array = nfs_readdir_get_array(page);
4a201d6e
TM
233 struct nfs_cache_array_entry *cache_entry;
234 int ret;
235
d1bacf9e
BS
236 if (IS_ERR(array))
237 return PTR_ERR(array);
3020093f
TM
238
239 cache_entry = &array->array[array->size];
240
241 /* Check that this entry lies within the page bounds */
8cd51a0c 242 ret = -ENOSPC;
3020093f 243 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
4a201d6e 244 goto out;
d1bacf9e 245
4a201d6e
TM
246 cache_entry->cookie = entry->prev_cookie;
247 cache_entry->ino = entry->ino;
0b26a0bf 248 cache_entry->d_type = entry->d_type;
4a201d6e
TM
249 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
250 if (ret)
251 goto out;
d1bacf9e 252 array->last_cookie = entry->cookie;
8cd51a0c 253 array->size++;
47c716cb 254 if (entry->eof != 0)
d1bacf9e 255 array->eof_index = array->size;
4a201d6e 256out:
d1bacf9e 257 nfs_readdir_release_array(page);
4a201d6e 258 return ret;
d1bacf9e
BS
259}
260
261static
262int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
263{
23db8620 264 loff_t diff = desc->ctx->pos - desc->current_index;
d1bacf9e
BS
265 unsigned int index;
266
267 if (diff < 0)
268 goto out_eof;
269 if (diff >= array->size) {
8cd51a0c 270 if (array->eof_index >= 0)
d1bacf9e 271 goto out_eof;
d1bacf9e
BS
272 return -EAGAIN;
273 }
274
275 index = (unsigned int)diff;
276 *desc->dir_cookie = array->array[index].cookie;
277 desc->cache_entry_index = index;
d1bacf9e
BS
278 return 0;
279out_eof:
280 desc->eof = 1;
281 return -EBADCOOKIE;
282}
283
4db72b40
JL
284static bool
285nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
286{
287 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
288 return false;
289 smp_rmb();
290 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
291}
292
d1bacf9e
BS
293static
294int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
295{
296 int i;
8ef2ce3e 297 loff_t new_pos;
d1bacf9e
BS
298 int status = -EAGAIN;
299
300 for (i = 0; i < array->size; i++) {
d1bacf9e 301 if (array->array[i].cookie == *desc->dir_cookie) {
496ad9aa 302 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
0c030806
TM
303 struct nfs_open_dir_context *ctx = desc->file->private_data;
304
8ef2ce3e 305 new_pos = desc->current_index + i;
4db72b40
JL
306 if (ctx->attr_gencount != nfsi->attr_gencount ||
307 !nfs_readdir_inode_mapping_valid(nfsi)) {
0c030806
TM
308 ctx->duped = 0;
309 ctx->attr_gencount = nfsi->attr_gencount;
23db8620 310 } else if (new_pos < desc->ctx->pos) {
0c030806
TM
311 if (ctx->duped > 0
312 && ctx->dup_cookie == *desc->dir_cookie) {
313 if (printk_ratelimit()) {
6de1472f 314 pr_notice("NFS: directory %pD2 contains a readdir loop."
0c030806 315 "Please contact your server vendor. "
374e4e3e 316 "The file: %s has duplicate cookie %llu\n",
6de1472f 317 desc->file,
374e4e3e 318 array->array[i].string.name,
0c030806
TM
319 *desc->dir_cookie);
320 }
321 status = -ELOOP;
322 goto out;
323 }
8ef2ce3e 324 ctx->dup_cookie = *desc->dir_cookie;
0c030806 325 ctx->duped = -1;
8ef2ce3e 326 }
23db8620 327 desc->ctx->pos = new_pos;
d1bacf9e 328 desc->cache_entry_index = i;
47c716cb 329 return 0;
d1bacf9e
BS
330 }
331 }
47c716cb 332 if (array->eof_index >= 0) {
8cd51a0c 333 status = -EBADCOOKIE;
18fb5fe4
TM
334 if (*desc->dir_cookie == array->last_cookie)
335 desc->eof = 1;
8cd51a0c 336 }
0c030806 337out:
d1bacf9e
BS
338 return status;
339}
340
341static
342int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
343{
344 struct nfs_cache_array *array;
47c716cb 345 int status;
d1bacf9e
BS
346
347 array = nfs_readdir_get_array(desc->page);
348 if (IS_ERR(array)) {
349 status = PTR_ERR(array);
350 goto out;
351 }
352
353 if (*desc->dir_cookie == 0)
354 status = nfs_readdir_search_for_pos(array, desc);
355 else
356 status = nfs_readdir_search_for_cookie(array, desc);
357
47c716cb 358 if (status == -EAGAIN) {
0aded708 359 desc->last_cookie = array->last_cookie;
e47c085a 360 desc->current_index += array->size;
47c716cb
TM
361 desc->page_index++;
362 }
d1bacf9e
BS
363 nfs_readdir_release_array(desc->page);
364out:
365 return status;
366}
367
368/* Fill a page with xdr information before transferring to the cache page */
369static
56e4ebf8 370int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
d1bacf9e 371 struct nfs_entry *entry, struct file *file, struct inode *inode)
1da177e4 372{
480c2006
BS
373 struct nfs_open_dir_context *ctx = file->private_data;
374 struct rpc_cred *cred = ctx->cred;
4704f0e2 375 unsigned long timestamp, gencount;
1da177e4
LT
376 int error;
377
1da177e4
LT
378 again:
379 timestamp = jiffies;
4704f0e2 380 gencount = nfs_inc_attr_generation_counter();
56e4ebf8 381 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
1da177e4
LT
382 NFS_SERVER(inode)->dtsize, desc->plus);
383 if (error < 0) {
384 /* We requested READDIRPLUS, but the server doesn't grok it */
385 if (error == -ENOTSUPP && desc->plus) {
386 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
3a10c30a 387 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
1da177e4
LT
388 desc->plus = 0;
389 goto again;
390 }
391 goto error;
392 }
1f4eab7e 393 desc->timestamp = timestamp;
4704f0e2 394 desc->gencount = gencount;
d1bacf9e
BS
395error:
396 return error;
1da177e4
LT
397}
398
573c4e1e
CL
399static int xdr_decode(nfs_readdir_descriptor_t *desc,
400 struct nfs_entry *entry, struct xdr_stream *xdr)
1da177e4 401{
573c4e1e 402 int error;
1da177e4 403
573c4e1e
CL
404 error = desc->decode(xdr, entry, desc->plus);
405 if (error)
406 return error;
d1bacf9e
BS
407 entry->fattr->time_start = desc->timestamp;
408 entry->fattr->gencount = desc->gencount;
409 return 0;
1da177e4
LT
410}
411
d39ab9de
BS
412static
413int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
414{
d39ab9de
BS
415 if (dentry->d_inode == NULL)
416 goto different;
37a09f07 417 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
d39ab9de
BS
418 goto different;
419 return 1;
420different:
421 return 0;
422}
423
d69ee9b8 424static
23db8620 425bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
d69ee9b8
TM
426{
427 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
428 return false;
429 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
430 return true;
23db8620 431 if (ctx->pos == 0)
d69ee9b8
TM
432 return true;
433 return false;
434}
435
436/*
437 * This function is called by the lookup code to request the use of
438 * readdirplus to accelerate any future lookups in the same
439 * directory.
440 */
441static
442void nfs_advise_use_readdirplus(struct inode *dir)
443{
444 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
445}
446
311324ad
TM
447/*
448 * This function is mainly for use by nfs_getattr().
449 *
450 * If this is an 'ls -l', we want to force use of readdirplus.
451 * Do this by checking if there is an active file descriptor
452 * and calling nfs_advise_use_readdirplus, then forcing a
453 * cache flush.
454 */
455void nfs_force_use_readdirplus(struct inode *dir)
456{
457 if (!list_empty(&NFS_I(dir)->open_files)) {
458 nfs_advise_use_readdirplus(dir);
459 nfs_zap_mapping(dir, dir->i_mapping);
460 }
461}
462
d39ab9de
BS
463static
464void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
465{
26fe5750 466 struct qstr filename = QSTR_INIT(entry->name, entry->len);
4a201d6e
TM
467 struct dentry *dentry;
468 struct dentry *alias;
d39ab9de
BS
469 struct inode *dir = parent->d_inode;
470 struct inode *inode;
aa9c2669 471 int status;
d39ab9de 472
4a201d6e
TM
473 if (filename.name[0] == '.') {
474 if (filename.len == 1)
475 return;
476 if (filename.len == 2 && filename.name[1] == '.')
477 return;
478 }
479 filename.hash = full_name_hash(filename.name, filename.len);
d39ab9de 480
4a201d6e 481 dentry = d_lookup(parent, &filename);
d39ab9de
BS
482 if (dentry != NULL) {
483 if (nfs_same_file(dentry, entry)) {
cda57a1e 484 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
aa9c2669
DQ
485 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
486 if (!status)
487 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
d39ab9de
BS
488 goto out;
489 } else {
696199f8
AV
490 if (d_invalidate(dentry) != 0)
491 goto out;
d39ab9de
BS
492 dput(dentry);
493 }
494 }
495
496 dentry = d_alloc(parent, &filename);
4a201d6e
TM
497 if (dentry == NULL)
498 return;
499
1775fd3e 500 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
d39ab9de
BS
501 if (IS_ERR(inode))
502 goto out;
503
504 alias = d_materialise_unique(dentry, inode);
505 if (IS_ERR(alias))
506 goto out;
507 else if (alias) {
508 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
509 dput(alias);
510 } else
511 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
512
513out:
514 dput(dentry);
d39ab9de
BS
515}
516
d1bacf9e
BS
517/* Perform conversion from xdr to cache array */
518static
8cd51a0c 519int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
6650239a 520 struct page **xdr_pages, struct page *page, unsigned int buflen)
1da177e4 521{
babddc72 522 struct xdr_stream stream;
f7da7a12 523 struct xdr_buf buf;
6650239a 524 struct page *scratch;
99424380 525 struct nfs_cache_array *array;
5c346854
TM
526 unsigned int count = 0;
527 int status;
babddc72 528
6650239a
TM
529 scratch = alloc_page(GFP_KERNEL);
530 if (scratch == NULL)
531 return -ENOMEM;
babddc72 532
f7da7a12 533 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
6650239a 534 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
99424380
BS
535
536 do {
537 status = xdr_decode(desc, entry, &stream);
8cd51a0c
TM
538 if (status != 0) {
539 if (status == -EAGAIN)
540 status = 0;
99424380 541 break;
8cd51a0c 542 }
99424380 543
5c346854
TM
544 count++;
545
47c716cb 546 if (desc->plus != 0)
d39ab9de 547 nfs_prime_dcache(desc->file->f_path.dentry, entry);
8cd51a0c
TM
548
549 status = nfs_readdir_add_to_array(entry, page);
550 if (status != 0)
551 break;
99424380
BS
552 } while (!entry->eof);
553
47c716cb 554 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
99424380 555 array = nfs_readdir_get_array(page);
8cd51a0c
TM
556 if (!IS_ERR(array)) {
557 array->eof_index = array->size;
558 status = 0;
559 nfs_readdir_release_array(page);
5c346854
TM
560 } else
561 status = PTR_ERR(array);
1da177e4 562 }
6650239a
TM
563
564 put_page(scratch);
8cd51a0c 565 return status;
56e4ebf8
BS
566}
567
568static
569void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
570{
571 unsigned int i;
572 for (i = 0; i < npages; i++)
573 put_page(pages[i]);
574}
575
576static
577void nfs_readdir_free_large_page(void *ptr, struct page **pages,
578 unsigned int npages)
579{
56e4ebf8
BS
580 nfs_readdir_free_pagearray(pages, npages);
581}
582
583/*
584 * nfs_readdir_large_page will allocate pages that must be freed with a call
585 * to nfs_readdir_free_large_page
586 */
587static
6650239a 588int nfs_readdir_large_page(struct page **pages, unsigned int npages)
56e4ebf8 589{
56e4ebf8
BS
590 unsigned int i;
591
592 for (i = 0; i < npages; i++) {
593 struct page *page = alloc_page(GFP_KERNEL);
594 if (page == NULL)
595 goto out_freepages;
596 pages[i] = page;
597 }
6650239a 598 return 0;
56e4ebf8 599
56e4ebf8
BS
600out_freepages:
601 nfs_readdir_free_pagearray(pages, i);
6650239a 602 return -ENOMEM;
1da177e4
LT
603}
604
d1bacf9e
BS
605static
606int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
00a92642 607{
56e4ebf8
BS
608 struct page *pages[NFS_MAX_READDIR_PAGES];
609 void *pages_ptr = NULL;
d1bacf9e
BS
610 struct nfs_entry entry;
611 struct file *file = desc->file;
612 struct nfs_cache_array *array;
8cd51a0c 613 int status = -ENOMEM;
56e4ebf8 614 unsigned int array_size = ARRAY_SIZE(pages);
d1bacf9e
BS
615
616 entry.prev_cookie = 0;
0aded708 617 entry.cookie = desc->last_cookie;
d1bacf9e
BS
618 entry.eof = 0;
619 entry.fh = nfs_alloc_fhandle();
620 entry.fattr = nfs_alloc_fattr();
573c4e1e 621 entry.server = NFS_SERVER(inode);
d1bacf9e
BS
622 if (entry.fh == NULL || entry.fattr == NULL)
623 goto out;
00a92642 624
14c43f76
DQ
625 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
626 if (IS_ERR(entry.label)) {
627 status = PTR_ERR(entry.label);
628 goto out;
629 }
630
d1bacf9e 631 array = nfs_readdir_get_array(page);
8cd51a0c
TM
632 if (IS_ERR(array)) {
633 status = PTR_ERR(array);
14c43f76 634 goto out_label_free;
8cd51a0c 635 }
d1bacf9e
BS
636 memset(array, 0, sizeof(struct nfs_cache_array));
637 array->eof_index = -1;
00a92642 638
6650239a
TM
639 status = nfs_readdir_large_page(pages, array_size);
640 if (status < 0)
d1bacf9e
BS
641 goto out_release_array;
642 do {
ac396128 643 unsigned int pglen;
56e4ebf8 644 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
babddc72 645
d1bacf9e 646 if (status < 0)
00a92642 647 break;
ac396128 648 pglen = status;
6650239a 649 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
8cd51a0c
TM
650 if (status < 0) {
651 if (status == -ENOSPC)
652 status = 0;
653 break;
654 }
655 } while (array->eof_index < 0);
d1bacf9e 656
56e4ebf8 657 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
d1bacf9e
BS
658out_release_array:
659 nfs_readdir_release_array(page);
14c43f76
DQ
660out_label_free:
661 nfs4_label_free(entry.label);
d1bacf9e
BS
662out:
663 nfs_free_fattr(entry.fattr);
664 nfs_free_fhandle(entry.fh);
00a92642
OG
665 return status;
666}
667
668/*
d1bacf9e
BS
669 * Now we cache directories properly, by converting xdr information
670 * to an array that can be used for lookups later. This results in
671 * fewer cache pages, since we can store more information on each page.
672 * We only need to convert from xdr once so future lookups are much simpler
1da177e4 673 */
d1bacf9e
BS
674static
675int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
1da177e4 676{
496ad9aa 677 struct inode *inode = file_inode(desc->file);
8cd51a0c 678 int ret;
1da177e4 679
8cd51a0c
TM
680 ret = nfs_readdir_xdr_to_array(desc, page, inode);
681 if (ret < 0)
d1bacf9e
BS
682 goto error;
683 SetPageUptodate(page);
1da177e4 684
d1bacf9e
BS
685 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
686 /* Should never happen */
687 nfs_zap_mapping(inode, inode->i_mapping);
1da177e4 688 }
d1bacf9e
BS
689 unlock_page(page);
690 return 0;
691 error:
692 unlock_page(page);
8cd51a0c 693 return ret;
d1bacf9e 694}
1da177e4 695
d1bacf9e
BS
696static
697void cache_page_release(nfs_readdir_descriptor_t *desc)
698{
11de3b11
TM
699 if (!desc->page->mapping)
700 nfs_readdir_clear_array(desc->page);
d1bacf9e
BS
701 page_cache_release(desc->page);
702 desc->page = NULL;
703}
704
705static
706struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
707{
496ad9aa 708 return read_cache_page(file_inode(desc->file)->i_mapping,
d1bacf9e 709 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
1da177e4
LT
710}
711
712/*
d1bacf9e 713 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1da177e4 714 */
d1bacf9e
BS
715static
716int find_cache_page(nfs_readdir_descriptor_t *desc)
717{
718 int res;
719
720 desc->page = get_cache_page(desc);
721 if (IS_ERR(desc->page))
722 return PTR_ERR(desc->page);
723
724 res = nfs_readdir_search_array(desc);
47c716cb
TM
725 if (res != 0)
726 cache_page_release(desc);
d1bacf9e
BS
727 return res;
728}
729
730/* Search for desc->dir_cookie from the beginning of the page cache */
1da177e4
LT
731static inline
732int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
733{
8cd51a0c 734 int res;
d1bacf9e 735
0aded708 736 if (desc->page_index == 0) {
8cd51a0c 737 desc->current_index = 0;
0aded708
TM
738 desc->last_cookie = 0;
739 }
47c716cb 740 do {
d1bacf9e 741 res = find_cache_page(desc);
47c716cb 742 } while (res == -EAGAIN);
1da177e4
LT
743 return res;
744}
745
1da177e4
LT
746/*
747 * Once we've found the start of the dirent within a page: fill 'er up...
748 */
749static
23db8620 750int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
1da177e4
LT
751{
752 struct file *file = desc->file;
d1bacf9e
BS
753 int i = 0;
754 int res = 0;
755 struct nfs_cache_array *array = NULL;
8ef2ce3e
BS
756 struct nfs_open_dir_context *ctx = file->private_data;
757
d1bacf9e 758 array = nfs_readdir_get_array(desc->page);
e7c58e97
TM
759 if (IS_ERR(array)) {
760 res = PTR_ERR(array);
761 goto out;
762 }
d1bacf9e
BS
763
764 for (i = desc->cache_entry_index; i < array->size; i++) {
ece0b423 765 struct nfs_cache_array_entry *ent;
1da177e4 766
ece0b423 767 ent = &array->array[i];
23db8620
AV
768 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
769 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
ece0b423 770 desc->eof = 1;
1da177e4 771 break;
ece0b423 772 }
23db8620 773 desc->ctx->pos++;
d1bacf9e
BS
774 if (i < (array->size-1))
775 *desc->dir_cookie = array->array[i+1].cookie;
776 else
777 *desc->dir_cookie = array->last_cookie;
0c030806
TM
778 if (ctx->duped != 0)
779 ctx->duped = 1;
1da177e4 780 }
47c716cb 781 if (array->eof_index >= 0)
8cd51a0c 782 desc->eof = 1;
d1bacf9e
BS
783
784 nfs_readdir_release_array(desc->page);
e7c58e97 785out:
d1bacf9e 786 cache_page_release(desc);
1e7cb3dc
CL
787 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
788 (unsigned long long)*desc->dir_cookie, res);
1da177e4
LT
789 return res;
790}
791
792/*
793 * If we cannot find a cookie in our cache, we suspect that this is
794 * because it points to a deleted file, so we ask the server to return
795 * whatever it thinks is the next entry. We then feed this to filldir.
796 * If all goes well, we should then be able to find our way round the
797 * cache on the next call to readdir_search_pagecache();
798 *
799 * NOTE: we cannot add the anonymous page to the pagecache because
800 * the data it contains might not be page aligned. Besides,
801 * we should already have a complete representation of the
802 * directory in the page cache by the time we get here.
803 */
804static inline
23db8620 805int uncached_readdir(nfs_readdir_descriptor_t *desc)
1da177e4 806{
1da177e4
LT
807 struct page *page = NULL;
808 int status;
496ad9aa 809 struct inode *inode = file_inode(desc->file);
0c030806 810 struct nfs_open_dir_context *ctx = desc->file->private_data;
1da177e4 811
1e7cb3dc
CL
812 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
813 (unsigned long long)*desc->dir_cookie);
1da177e4
LT
814
815 page = alloc_page(GFP_HIGHUSER);
816 if (!page) {
817 status = -ENOMEM;
818 goto out;
819 }
d1bacf9e 820
7a8e1dc3 821 desc->page_index = 0;
0aded708 822 desc->last_cookie = *desc->dir_cookie;
7a8e1dc3 823 desc->page = page;
0c030806 824 ctx->duped = 0;
7a8e1dc3 825
85f8607e
TM
826 status = nfs_readdir_xdr_to_array(desc, page, inode);
827 if (status < 0)
1da177e4
LT
828 goto out_release;
829
23db8620 830 status = nfs_do_filldir(desc);
1da177e4 831
1da177e4 832 out:
1e7cb3dc 833 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
3110ff80 834 __func__, status);
1da177e4
LT
835 return status;
836 out_release:
d1bacf9e 837 cache_page_release(desc);
1da177e4
LT
838 goto out;
839}
840
311324ad
TM
841static bool nfs_dir_mapping_need_revalidate(struct inode *dir)
842{
843 struct nfs_inode *nfsi = NFS_I(dir);
844
845 if (nfs_attribute_cache_expired(dir))
846 return true;
847 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
848 return true;
849 return false;
850}
851
00a92642
OG
852/* The file offset position represents the dirent entry number. A
853 last cookie cache takes care of the common case of reading the
854 whole directory.
1da177e4 855 */
23db8620 856static int nfs_readdir(struct file *file, struct dir_context *ctx)
1da177e4 857{
23db8620 858 struct dentry *dentry = file->f_path.dentry;
1da177e4
LT
859 struct inode *inode = dentry->d_inode;
860 nfs_readdir_descriptor_t my_desc,
861 *desc = &my_desc;
23db8620 862 struct nfs_open_dir_context *dir_ctx = file->private_data;
07b5ce8e 863 int res = 0;
1da177e4 864
6de1472f
AV
865 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
866 file, (long long)ctx->pos);
91d5b470
CL
867 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
868
1da177e4 869 /*
23db8620 870 * ctx->pos points to the dirent entry number.
f0dd2136 871 * *desc->dir_cookie has the cookie for the next entry. We have
00a92642
OG
872 * to either find the entry with the appropriate number or
873 * revalidate the cookie.
1da177e4
LT
874 */
875 memset(desc, 0, sizeof(*desc));
876
23db8620
AV
877 desc->file = file;
878 desc->ctx = ctx;
480c2006 879 desc->dir_cookie = &dir_ctx->dir_cookie;
1da177e4 880 desc->decode = NFS_PROTO(inode)->decode_dirent;
23db8620 881 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
1da177e4 882
565277f6 883 nfs_block_sillyrename(dentry);
311324ad 884 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode))
07b5ce8e 885 res = nfs_revalidate_mapping(inode, file->f_mapping);
fccca7fc
TM
886 if (res < 0)
887 goto out;
888
47c716cb 889 do {
1da177e4 890 res = readdir_search_pagecache(desc);
00a92642 891
1da177e4 892 if (res == -EBADCOOKIE) {
ece0b423 893 res = 0;
1da177e4 894 /* This means either end of directory */
d1bacf9e 895 if (*desc->dir_cookie && desc->eof == 0) {
1da177e4 896 /* Or that the server has 'lost' a cookie */
23db8620 897 res = uncached_readdir(desc);
ece0b423 898 if (res == 0)
1da177e4
LT
899 continue;
900 }
1da177e4
LT
901 break;
902 }
903 if (res == -ETOOSMALL && desc->plus) {
3a10c30a 904 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
1da177e4 905 nfs_zap_caches(inode);
baf57a09 906 desc->page_index = 0;
1da177e4 907 desc->plus = 0;
d1bacf9e 908 desc->eof = 0;
1da177e4
LT
909 continue;
910 }
911 if (res < 0)
912 break;
913
23db8620 914 res = nfs_do_filldir(desc);
ece0b423 915 if (res < 0)
1da177e4 916 break;
47c716cb 917 } while (!desc->eof);
fccca7fc 918out:
565277f6 919 nfs_unblock_sillyrename(dentry);
1e7cb3dc
CL
920 if (res > 0)
921 res = 0;
6de1472f 922 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1e7cb3dc 923 return res;
1da177e4
LT
924}
925
965c8e59 926static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
f0dd2136 927{
6de1472f 928 struct inode *inode = file_inode(filp);
480c2006 929 struct nfs_open_dir_context *dir_ctx = filp->private_data;
b84e06c5 930
6de1472f
AV
931 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
932 filp, offset, whence);
b84e06c5
CL
933
934 mutex_lock(&inode->i_mutex);
965c8e59 935 switch (whence) {
f0dd2136
TM
936 case 1:
937 offset += filp->f_pos;
938 case 0:
939 if (offset >= 0)
940 break;
941 default:
942 offset = -EINVAL;
943 goto out;
944 }
945 if (offset != filp->f_pos) {
946 filp->f_pos = offset;
480c2006 947 dir_ctx->dir_cookie = 0;
8ef2ce3e 948 dir_ctx->duped = 0;
f0dd2136
TM
949 }
950out:
b84e06c5 951 mutex_unlock(&inode->i_mutex);
f0dd2136
TM
952 return offset;
953}
954
1da177e4
LT
955/*
956 * All directory operations under NFS are synchronous, so fsync()
957 * is a dummy operation.
958 */
02c24a82
JB
959static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
960 int datasync)
1da177e4 961{
6de1472f 962 struct inode *inode = file_inode(filp);
7ea80859 963
6de1472f 964 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1e7cb3dc 965
02c24a82 966 mutex_lock(&inode->i_mutex);
6de1472f 967 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
02c24a82 968 mutex_unlock(&inode->i_mutex);
1da177e4
LT
969 return 0;
970}
971
bfc69a45
TM
972/**
973 * nfs_force_lookup_revalidate - Mark the directory as having changed
974 * @dir - pointer to directory inode
975 *
976 * This forces the revalidation code in nfs_lookup_revalidate() to do a
977 * full lookup on all child dentries of 'dir' whenever a change occurs
978 * on the server that might have invalidated our dcache.
979 *
980 * The caller should be holding dir->i_lock
981 */
982void nfs_force_lookup_revalidate(struct inode *dir)
983{
011935a0 984 NFS_I(dir)->cache_change_attribute++;
bfc69a45 985}
89d77c8f 986EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
bfc69a45 987
1da177e4
LT
988/*
989 * A check for whether or not the parent directory has changed.
990 * In the case it has, we assume that the dentries are untrustworthy
991 * and may need to be looked up again.
992 */
c79ba787 993static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
1da177e4
LT
994{
995 if (IS_ROOT(dentry))
996 return 1;
4eec952e
TM
997 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
998 return 0;
f2c77f4e
TM
999 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1000 return 0;
1001 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1002 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1003 return 0;
1004 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1005 return 0;
1006 return 1;
1da177e4
LT
1007}
1008
a12802ca
TM
1009/*
1010 * Use intent information to check whether or not we're going to do
1011 * an O_EXCL create using this path component.
1012 */
fa3c56bb 1013static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
a12802ca
TM
1014{
1015 if (NFS_PROTO(dir)->version == 2)
1016 return 0;
fa3c56bb 1017 return flags & LOOKUP_EXCL;
a12802ca
TM
1018}
1019
1d6757fb
TM
1020/*
1021 * Inode and filehandle revalidation for lookups.
1022 *
1023 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1024 * or if the intent information indicates that we're about to open this
1025 * particular file and the "nocto" mount flag is not set.
1026 *
1027 */
65a0c149 1028static
fa3c56bb 1029int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1da177e4
LT
1030{
1031 struct nfs_server *server = NFS_SERVER(inode);
65a0c149 1032 int ret;
1da177e4 1033
36d43a43 1034 if (IS_AUTOMOUNT(inode))
4e99a1ff 1035 return 0;
facc3530 1036 /* VFS wants an on-the-wire revalidation */
fa3c56bb 1037 if (flags & LOOKUP_REVAL)
facc3530
AV
1038 goto out_force;
1039 /* This is an open(2) */
fa3c56bb
AV
1040 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1041 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
facc3530 1042 goto out_force;
65a0c149
TM
1043out:
1044 return (inode->i_nlink == 0) ? -ENOENT : 0;
1da177e4 1045out_force:
65a0c149
TM
1046 ret = __nfs_revalidate_inode(server, inode);
1047 if (ret != 0)
1048 return ret;
1049 goto out;
1da177e4
LT
1050}
1051
1052/*
1053 * We judge how long we want to trust negative
1054 * dentries by looking at the parent inode mtime.
1055 *
1056 * If parent mtime has changed, we revalidate, else we wait for a
1057 * period corresponding to the parent's attribute cache timeout value.
1058 */
1059static inline
1060int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
fa3c56bb 1061 unsigned int flags)
1da177e4 1062{
1da177e4 1063 /* Don't revalidate a negative dentry if we're creating a new file */
fa3c56bb 1064 if (flags & LOOKUP_CREATE)
1da177e4 1065 return 0;
4eec952e
TM
1066 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1067 return 1;
1da177e4
LT
1068 return !nfs_check_verifier(dir, dentry);
1069}
1070
1071/*
1072 * This is called every time the dcache has a lookup hit,
1073 * and we should check whether we can really trust that
1074 * lookup.
1075 *
1076 * NOTE! The hit can be a negative hit too, don't assume
1077 * we have an inode!
1078 *
1079 * If the parent directory is seen to have changed, we throw out the
1080 * cached dentry and do a new lookup.
1081 */
0b728e19 1082static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1da177e4
LT
1083{
1084 struct inode *dir;
1085 struct inode *inode;
1086 struct dentry *parent;
e1fb4d05
TM
1087 struct nfs_fh *fhandle = NULL;
1088 struct nfs_fattr *fattr = NULL;
1775fd3e 1089 struct nfs4_label *label = NULL;
1da177e4 1090 int error;
1da177e4 1091
fa3c56bb 1092 if (flags & LOOKUP_RCU)
34286d66
NP
1093 return -ECHILD;
1094
1da177e4 1095 parent = dget_parent(dentry);
1da177e4 1096 dir = parent->d_inode;
91d5b470 1097 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1da177e4
LT
1098 inode = dentry->d_inode;
1099
1100 if (!inode) {
fa3c56bb 1101 if (nfs_neg_need_reval(dir, dentry, flags))
1da177e4 1102 goto out_bad;
d69ee9b8 1103 goto out_valid_noent;
1da177e4
LT
1104 }
1105
1106 if (is_bad_inode(inode)) {
6de1472f
AV
1107 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1108 __func__, dentry);
1da177e4
LT
1109 goto out_bad;
1110 }
1111
011e2a7f 1112 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
15860ab1
TM
1113 goto out_set_verifier;
1114
1da177e4 1115 /* Force a full look up iff the parent directory has changed */
fa3c56bb
AV
1116 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1117 if (nfs_lookup_verify_inode(inode, flags))
1da177e4
LT
1118 goto out_zap_parent;
1119 goto out_valid;
1120 }
1121
1122 if (NFS_STALE(inode))
1123 goto out_bad;
1124
e1fb4d05
TM
1125 error = -ENOMEM;
1126 fhandle = nfs_alloc_fhandle();
1127 fattr = nfs_alloc_fattr();
1128 if (fhandle == NULL || fattr == NULL)
1129 goto out_error;
1130
14c43f76
DQ
1131 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1132 if (IS_ERR(label))
1133 goto out_error;
1134
6e0d0be7 1135 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1775fd3e 1136 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
6e0d0be7 1137 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1da177e4
LT
1138 if (error)
1139 goto out_bad;
e1fb4d05 1140 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1da177e4 1141 goto out_bad;
e1fb4d05 1142 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1da177e4
LT
1143 goto out_bad;
1144
aa9c2669
DQ
1145 nfs_setsecurity(inode, fattr, label);
1146
e1fb4d05
TM
1147 nfs_free_fattr(fattr);
1148 nfs_free_fhandle(fhandle);
14c43f76
DQ
1149 nfs4_label_free(label);
1150
15860ab1 1151out_set_verifier:
cf8ba45e 1152 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1da177e4 1153 out_valid:
d69ee9b8
TM
1154 /* Success: notify readdir to use READDIRPLUS */
1155 nfs_advise_use_readdirplus(dir);
1156 out_valid_noent:
1da177e4 1157 dput(parent);
6de1472f
AV
1158 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1159 __func__, dentry);
1da177e4
LT
1160 return 1;
1161out_zap_parent:
1162 nfs_zap_caches(dir);
1163 out_bad:
c44600c9
AV
1164 nfs_free_fattr(fattr);
1165 nfs_free_fhandle(fhandle);
14c43f76 1166 nfs4_label_free(label);
a1643a92 1167 nfs_mark_for_revalidate(dir);
1da177e4
LT
1168 if (inode && S_ISDIR(inode->i_mode)) {
1169 /* Purge readdir caches. */
1170 nfs_zap_caches(inode);
a3f432bf
BF
1171 /*
1172 * We can't d_drop the root of a disconnected tree:
1173 * its d_hash is on the s_anon list and d_drop() would hide
1174 * it from shrink_dcache_for_unmount(), leading to busy
1175 * inodes on unmount and further oopses.
1176 */
1177 if (IS_ROOT(dentry))
d9e80b7d 1178 goto out_valid;
1da177e4 1179 }
13caa9fb
MS
1180 /* If we have submounts, don't unhash ! */
1181 if (check_submounts_and_drop(dentry) != 0)
1182 goto out_valid;
1183
1da177e4 1184 dput(parent);
6de1472f
AV
1185 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1186 __func__, dentry);
1da177e4 1187 return 0;
e1fb4d05
TM
1188out_error:
1189 nfs_free_fattr(fattr);
1190 nfs_free_fhandle(fhandle);
14c43f76 1191 nfs4_label_free(label);
e1fb4d05 1192 dput(parent);
6de1472f
AV
1193 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1194 __func__, dentry, error);
e1fb4d05 1195 return error;
1da177e4
LT
1196}
1197
ecf3d1f1
JL
1198/*
1199 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1200 * when we don't really care about the dentry name. This is called when a
1201 * pathwalk ends on a dentry that was not found via a normal lookup in the
1202 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1203 *
1204 * In this situation, we just want to verify that the inode itself is OK
1205 * since the dentry might have changed on the server.
1206 */
1207static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1208{
1209 int error;
1210 struct inode *inode = dentry->d_inode;
1211
1212 /*
1213 * I believe we can only get a negative dentry here in the case of a
1214 * procfs-style symlink. Just assume it's correct for now, but we may
1215 * eventually need to do something more here.
1216 */
1217 if (!inode) {
6de1472f
AV
1218 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1219 __func__, dentry);
ecf3d1f1
JL
1220 return 1;
1221 }
1222
1223 if (is_bad_inode(inode)) {
6de1472f
AV
1224 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1225 __func__, dentry);
ecf3d1f1
JL
1226 return 0;
1227 }
1228
1229 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1230 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1231 __func__, inode->i_ino, error ? "invalid" : "valid");
1232 return !error;
1233}
1234
1da177e4
LT
1235/*
1236 * This is called from dput() when d_count is going to 0.
1237 */
fe15ce44 1238static int nfs_dentry_delete(const struct dentry *dentry)
1da177e4 1239{
6de1472f
AV
1240 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1241 dentry, dentry->d_flags);
1da177e4 1242
77f11192
TM
1243 /* Unhash any dentry with a stale inode */
1244 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1245 return 1;
1246
1da177e4
LT
1247 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1248 /* Unhash it, so that ->d_iput() would be called */
1249 return 1;
1250 }
1251 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1252 /* Unhash it, so that ancestors of killed async unlink
1253 * files will be cleaned up during umount */
1254 return 1;
1255 }
1256 return 0;
1257
1258}
1259
1f018458 1260/* Ensure that we revalidate inode->i_nlink */
1b83d707
TM
1261static void nfs_drop_nlink(struct inode *inode)
1262{
1263 spin_lock(&inode->i_lock);
1f018458
TM
1264 /* drop the inode if we're reasonably sure this is the last link */
1265 if (inode->i_nlink == 1)
1266 clear_nlink(inode);
1267 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1b83d707
TM
1268 spin_unlock(&inode->i_lock);
1269}
1270
1da177e4
LT
1271/*
1272 * Called when the dentry loses inode.
1273 * We use it to clean up silly-renamed files.
1274 */
1275static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1276{
83672d39
NB
1277 if (S_ISDIR(inode->i_mode))
1278 /* drop any readdir cache as it could easily be old */
1279 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1280
1da177e4 1281 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
e4eff1a6 1282 nfs_complete_unlink(dentry, inode);
1f018458 1283 nfs_drop_nlink(inode);
1da177e4 1284 }
1da177e4
LT
1285 iput(inode);
1286}
1287
b1942c5f
AV
1288static void nfs_d_release(struct dentry *dentry)
1289{
1290 /* free cached devname value, if it survived that far */
1291 if (unlikely(dentry->d_fsdata)) {
1292 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1293 WARN_ON(1);
1294 else
1295 kfree(dentry->d_fsdata);
1296 }
1297}
1298
f786aa90 1299const struct dentry_operations nfs_dentry_operations = {
1da177e4 1300 .d_revalidate = nfs_lookup_revalidate,
ecf3d1f1 1301 .d_weak_revalidate = nfs_weak_revalidate,
1da177e4
LT
1302 .d_delete = nfs_dentry_delete,
1303 .d_iput = nfs_dentry_iput,
36d43a43 1304 .d_automount = nfs_d_automount,
b1942c5f 1305 .d_release = nfs_d_release,
1da177e4 1306};
ddda8e0a 1307EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1da177e4 1308
597d9289 1309struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1da177e4
LT
1310{
1311 struct dentry *res;
565277f6 1312 struct dentry *parent;
1da177e4 1313 struct inode *inode = NULL;
e1fb4d05
TM
1314 struct nfs_fh *fhandle = NULL;
1315 struct nfs_fattr *fattr = NULL;
1775fd3e 1316 struct nfs4_label *label = NULL;
1da177e4 1317 int error;
1da177e4 1318
6de1472f 1319 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
91d5b470 1320 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1da177e4
LT
1321
1322 res = ERR_PTR(-ENAMETOOLONG);
1323 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1324 goto out;
1325
fd684071
TM
1326 /*
1327 * If we're doing an exclusive create, optimize away the lookup
1328 * but don't hash the dentry.
1329 */
00cd8dd3 1330 if (nfs_is_exclusive_create(dir, flags)) {
fd684071
TM
1331 d_instantiate(dentry, NULL);
1332 res = NULL;
fc0f684c 1333 goto out;
fd684071 1334 }
1da177e4 1335
e1fb4d05
TM
1336 res = ERR_PTR(-ENOMEM);
1337 fhandle = nfs_alloc_fhandle();
1338 fattr = nfs_alloc_fattr();
1339 if (fhandle == NULL || fattr == NULL)
1340 goto out;
1341
14c43f76
DQ
1342 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1343 if (IS_ERR(label))
1344 goto out;
1345
565277f6
TM
1346 parent = dentry->d_parent;
1347 /* Protect against concurrent sillydeletes */
6e0d0be7 1348 trace_nfs_lookup_enter(dir, dentry, flags);
565277f6 1349 nfs_block_sillyrename(parent);
1775fd3e 1350 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1da177e4
LT
1351 if (error == -ENOENT)
1352 goto no_entry;
1353 if (error < 0) {
1354 res = ERR_PTR(error);
565277f6 1355 goto out_unblock_sillyrename;
1da177e4 1356 }
1775fd3e 1357 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
bf0c84f1 1358 res = ERR_CAST(inode);
03f28e3a 1359 if (IS_ERR(res))
565277f6 1360 goto out_unblock_sillyrename;
54ceac45 1361
d69ee9b8
TM
1362 /* Success: notify readdir to use READDIRPLUS */
1363 nfs_advise_use_readdirplus(dir);
1364
1da177e4 1365no_entry:
54ceac45 1366 res = d_materialise_unique(dentry, inode);
9eaef27b
TM
1367 if (res != NULL) {
1368 if (IS_ERR(res))
565277f6 1369 goto out_unblock_sillyrename;
1da177e4 1370 dentry = res;
9eaef27b 1371 }
1da177e4 1372 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
565277f6
TM
1373out_unblock_sillyrename:
1374 nfs_unblock_sillyrename(parent);
6e0d0be7 1375 trace_nfs_lookup_exit(dir, dentry, flags, error);
14c43f76 1376 nfs4_label_free(label);
1da177e4 1377out:
e1fb4d05
TM
1378 nfs_free_fattr(fattr);
1379 nfs_free_fhandle(fhandle);
1da177e4
LT
1380 return res;
1381}
ddda8e0a 1382EXPORT_SYMBOL_GPL(nfs_lookup);
1da177e4 1383
89d77c8f 1384#if IS_ENABLED(CONFIG_NFS_V4)
0b728e19 1385static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1da177e4 1386
f786aa90 1387const struct dentry_operations nfs4_dentry_operations = {
0ef97dcf 1388 .d_revalidate = nfs4_lookup_revalidate,
1da177e4
LT
1389 .d_delete = nfs_dentry_delete,
1390 .d_iput = nfs_dentry_iput,
36d43a43 1391 .d_automount = nfs_d_automount,
b1942c5f 1392 .d_release = nfs_d_release,
1da177e4 1393};
89d77c8f 1394EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1da177e4 1395
8a5e929d
AV
1396static fmode_t flags_to_mode(int flags)
1397{
1398 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1399 if ((flags & O_ACCMODE) != O_WRONLY)
1400 res |= FMODE_READ;
1401 if ((flags & O_ACCMODE) != O_RDONLY)
1402 res |= FMODE_WRITE;
1403 return res;
1404}
1405
51141598 1406static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
cd9a1c0e 1407{
5ede7b1c 1408 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
cd9a1c0e
TM
1409}
1410
1411static int do_open(struct inode *inode, struct file *filp)
1412{
f1fe29b4 1413 nfs_fscache_open_file(inode, filp);
cd9a1c0e
TM
1414 return 0;
1415}
1416
d9585277
AV
1417static int nfs_finish_open(struct nfs_open_context *ctx,
1418 struct dentry *dentry,
30d90494 1419 struct file *file, unsigned open_flags,
d9585277 1420 int *opened)
cd9a1c0e 1421{
0dd2b474
MS
1422 int err;
1423
01c919ab
MS
1424 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1425 *opened |= FILE_CREATED;
1426
30d90494
AV
1427 err = finish_open(file, dentry, do_open, opened);
1428 if (err)
d9585277 1429 goto out;
30d90494 1430 nfs_file_set_open_context(file, ctx);
0dd2b474 1431
cd9a1c0e 1432out:
d9585277 1433 return err;
cd9a1c0e
TM
1434}
1435
73a79706
BS
1436int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1437 struct file *file, unsigned open_flags,
1438 umode_t mode, int *opened)
1da177e4 1439{
cd9a1c0e 1440 struct nfs_open_context *ctx;
0dd2b474
MS
1441 struct dentry *res;
1442 struct iattr attr = { .ia_valid = ATTR_OPEN };
f46e0bd3 1443 struct inode *inode;
1472b83e 1444 unsigned int lookup_flags = 0;
898f635c 1445 int err;
1da177e4 1446
0dd2b474
MS
1447 /* Expect a negative dentry */
1448 BUG_ON(dentry->d_inode);
1449
1e8968c5 1450 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
6de1472f 1451 dir->i_sb->s_id, dir->i_ino, dentry);
1e7cb3dc 1452
9597c13b
JL
1453 err = nfs_check_flags(open_flags);
1454 if (err)
1455 return err;
1456
0dd2b474
MS
1457 /* NFS only supports OPEN on regular files */
1458 if ((open_flags & O_DIRECTORY)) {
0dd2b474
MS
1459 if (!d_unhashed(dentry)) {
1460 /*
1461 * Hashed negative dentry with O_DIRECTORY: dentry was
1462 * revalidated and is fine, no need to perform lookup
1463 * again
1464 */
d9585277 1465 return -ENOENT;
0dd2b474 1466 }
1472b83e 1467 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1da177e4 1468 goto no_open;
02a913a7 1469 }
1da177e4 1470
0dd2b474 1471 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
d9585277 1472 return -ENAMETOOLONG;
cd9a1c0e 1473
0dd2b474 1474 if (open_flags & O_CREAT) {
536e43d1 1475 attr.ia_valid |= ATTR_MODE;
0dd2b474
MS
1476 attr.ia_mode = mode & ~current_umask();
1477 }
536e43d1
TM
1478 if (open_flags & O_TRUNC) {
1479 attr.ia_valid |= ATTR_SIZE;
1480 attr.ia_size = 0;
cd9a1c0e
TM
1481 }
1482
0dd2b474
MS
1483 ctx = create_nfs_open_context(dentry, open_flags);
1484 err = PTR_ERR(ctx);
1485 if (IS_ERR(ctx))
d9585277 1486 goto out;
0dd2b474 1487
6e0d0be7 1488 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
f46e0bd3 1489 nfs_block_sillyrename(dentry->d_parent);
5bc2afc2 1490 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
275bb307 1491 nfs_unblock_sillyrename(dentry->d_parent);
f46e0bd3 1492 if (IS_ERR(inode)) {
0dd2b474 1493 err = PTR_ERR(inode);
6e0d0be7 1494 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2d9db750 1495 put_nfs_open_context(ctx);
0dd2b474
MS
1496 switch (err) {
1497 case -ENOENT:
275bb307 1498 d_drop(dentry);
0dd2b474
MS
1499 d_add(dentry, NULL);
1500 break;
1501 case -EISDIR:
1502 case -ENOTDIR:
1503 goto no_open;
1504 case -ELOOP:
1505 if (!(open_flags & O_NOFOLLOW))
6f926b5b 1506 goto no_open;
0dd2b474 1507 break;
1da177e4 1508 /* case -EINVAL: */
0dd2b474
MS
1509 default:
1510 break;
1da177e4 1511 }
d9585277 1512 goto out;
cd9a1c0e 1513 }
0dd2b474 1514
275bb307 1515 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
6e0d0be7 1516 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2d9db750 1517 put_nfs_open_context(ctx);
d9585277
AV
1518out:
1519 return err;
0dd2b474 1520
1da177e4 1521no_open:
1472b83e 1522 res = nfs_lookup(dir, dentry, lookup_flags);
0dd2b474
MS
1523 err = PTR_ERR(res);
1524 if (IS_ERR(res))
d9585277 1525 goto out;
0dd2b474 1526
e45198a6 1527 return finish_no_open(file, res);
1da177e4 1528}
89d77c8f 1529EXPORT_SYMBOL_GPL(nfs_atomic_open);
1da177e4 1530
0b728e19 1531static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1da177e4
LT
1532{
1533 struct dentry *parent = NULL;
657e94b6 1534 struct inode *inode;
1da177e4 1535 struct inode *dir;
50de348c 1536 int ret = 0;
1da177e4 1537
fa3c56bb 1538 if (flags & LOOKUP_RCU)
657e94b6
NP
1539 return -ECHILD;
1540
fa3c56bb 1541 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
eda72afb
MS
1542 goto no_open;
1543 if (d_mountpoint(dentry))
5584c306 1544 goto no_open;
49f9a0fa
TM
1545 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1546 goto no_open;
2b484297 1547
eda72afb 1548 inode = dentry->d_inode;
1da177e4
LT
1549 parent = dget_parent(dentry);
1550 dir = parent->d_inode;
2b484297 1551
1da177e4
LT
1552 /* We can't create new files in nfs_open_revalidate(), so we
1553 * optimize away revalidation of negative dentries.
1554 */
216d5d06 1555 if (inode == NULL) {
fa3c56bb 1556 if (!nfs_neg_need_reval(dir, dentry, flags))
216d5d06 1557 ret = 1;
1da177e4 1558 goto out;
216d5d06
TM
1559 }
1560
1da177e4
LT
1561 /* NFS only supports OPEN on regular files */
1562 if (!S_ISREG(inode->i_mode))
5584c306 1563 goto no_open_dput;
1da177e4 1564 /* We cannot do exclusive creation on a positive dentry */
fa3c56bb 1565 if (flags & LOOKUP_EXCL)
5584c306 1566 goto no_open_dput;
1da177e4 1567
0ef97dcf
MS
1568 /* Let f_op->open() actually open (and revalidate) the file */
1569 ret = 1;
536e43d1 1570
1da177e4
LT
1571out:
1572 dput(parent);
1da177e4 1573 return ret;
535918f1 1574
5584c306 1575no_open_dput:
1da177e4 1576 dput(parent);
5584c306 1577no_open:
0b728e19 1578 return nfs_lookup_revalidate(dentry, flags);
c0204fd2
TM
1579}
1580
1da177e4
LT
1581#endif /* CONFIG_NFSV4 */
1582
1da177e4
LT
1583/*
1584 * Code common to create, mkdir, and mknod.
1585 */
1586int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1775fd3e
DQ
1587 struct nfs_fattr *fattr,
1588 struct nfs4_label *label)
1da177e4 1589{
fab728e1
TM
1590 struct dentry *parent = dget_parent(dentry);
1591 struct inode *dir = parent->d_inode;
1da177e4
LT
1592 struct inode *inode;
1593 int error = -EACCES;
1594
fab728e1
TM
1595 d_drop(dentry);
1596
1da177e4
LT
1597 /* We may have been initialized further down */
1598 if (dentry->d_inode)
fab728e1 1599 goto out;
1da177e4 1600 if (fhandle->size == 0) {
1775fd3e 1601 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1da177e4 1602 if (error)
fab728e1 1603 goto out_error;
1da177e4 1604 }
5724ab37 1605 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1da177e4
LT
1606 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1607 struct nfs_server *server = NFS_SB(dentry->d_sb);
1775fd3e 1608 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1da177e4 1609 if (error < 0)
fab728e1 1610 goto out_error;
1da177e4 1611 }
1775fd3e 1612 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
03f28e3a
TM
1613 error = PTR_ERR(inode);
1614 if (IS_ERR(inode))
fab728e1
TM
1615 goto out_error;
1616 d_add(dentry, inode);
1617out:
1618 dput(parent);
1da177e4 1619 return 0;
fab728e1
TM
1620out_error:
1621 nfs_mark_for_revalidate(dir);
1622 dput(parent);
1623 return error;
1da177e4 1624}
ddda8e0a 1625EXPORT_SYMBOL_GPL(nfs_instantiate);
1da177e4
LT
1626
1627/*
1628 * Following a failed create operation, we drop the dentry rather
1629 * than retain a negative dentry. This avoids a problem in the event
1630 * that the operation succeeded on the server, but an error in the
1631 * reply path made it appear to have failed.
1632 */
597d9289 1633int nfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 1634 umode_t mode, bool excl)
1da177e4
LT
1635{
1636 struct iattr attr;
ebfc3b49 1637 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1da177e4 1638 int error;
1da177e4 1639
1e8968c5 1640 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
6de1472f 1641 dir->i_sb->s_id, dir->i_ino, dentry);
1da177e4
LT
1642
1643 attr.ia_mode = mode;
1644 attr.ia_valid = ATTR_MODE;
1645
8b0ad3d4 1646 trace_nfs_create_enter(dir, dentry, open_flags);
8867fe58 1647 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
8b0ad3d4 1648 trace_nfs_create_exit(dir, dentry, open_flags, error);
1da177e4
LT
1649 if (error != 0)
1650 goto out_err;
1da177e4
LT
1651 return 0;
1652out_err:
1da177e4
LT
1653 d_drop(dentry);
1654 return error;
1655}
ddda8e0a 1656EXPORT_SYMBOL_GPL(nfs_create);
1da177e4
LT
1657
1658/*
1659 * See comments for nfs_proc_create regarding failed operations.
1660 */
597d9289 1661int
1a67aafb 1662nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1da177e4
LT
1663{
1664 struct iattr attr;
1665 int status;
1666
1e8968c5 1667 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
6de1472f 1668 dir->i_sb->s_id, dir->i_ino, dentry);
1da177e4
LT
1669
1670 if (!new_valid_dev(rdev))
1671 return -EINVAL;
1672
1673 attr.ia_mode = mode;
1674 attr.ia_valid = ATTR_MODE;
1675
1ca42382 1676 trace_nfs_mknod_enter(dir, dentry);
1da177e4 1677 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1ca42382 1678 trace_nfs_mknod_exit(dir, dentry, status);
1da177e4
LT
1679 if (status != 0)
1680 goto out_err;
1da177e4
LT
1681 return 0;
1682out_err:
1da177e4
LT
1683 d_drop(dentry);
1684 return status;
1685}
ddda8e0a 1686EXPORT_SYMBOL_GPL(nfs_mknod);
1da177e4
LT
1687
1688/*
1689 * See comments for nfs_proc_create regarding failed operations.
1690 */
597d9289 1691int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4
LT
1692{
1693 struct iattr attr;
1694 int error;
1695
1e8968c5 1696 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
6de1472f 1697 dir->i_sb->s_id, dir->i_ino, dentry);
1da177e4
LT
1698
1699 attr.ia_valid = ATTR_MODE;
1700 attr.ia_mode = mode | S_IFDIR;
1701
1ca42382 1702 trace_nfs_mkdir_enter(dir, dentry);
1da177e4 1703 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1ca42382 1704 trace_nfs_mkdir_exit(dir, dentry, error);
1da177e4
LT
1705 if (error != 0)
1706 goto out_err;
1da177e4
LT
1707 return 0;
1708out_err:
1709 d_drop(dentry);
1da177e4
LT
1710 return error;
1711}
ddda8e0a 1712EXPORT_SYMBOL_GPL(nfs_mkdir);
1da177e4 1713
d45b9d8b
TM
1714static void nfs_dentry_handle_enoent(struct dentry *dentry)
1715{
1716 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1717 d_delete(dentry);
1718}
1719
597d9289 1720int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1da177e4
LT
1721{
1722 int error;
1723
1e8968c5 1724 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
6de1472f 1725 dir->i_sb->s_id, dir->i_ino, dentry);
1da177e4 1726
1ca42382 1727 trace_nfs_rmdir_enter(dir, dentry);
ba6c0592
TM
1728 if (dentry->d_inode) {
1729 nfs_wait_on_sillyrename(dentry);
1730 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1731 /* Ensure the VFS deletes this inode */
1732 switch (error) {
1733 case 0:
1734 clear_nlink(dentry->d_inode);
1735 break;
1736 case -ENOENT:
1737 nfs_dentry_handle_enoent(dentry);
1738 }
1739 } else
1740 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1ca42382 1741 trace_nfs_rmdir_exit(dir, dentry, error);
1da177e4
LT
1742
1743 return error;
1744}
ddda8e0a 1745EXPORT_SYMBOL_GPL(nfs_rmdir);
1da177e4 1746
1da177e4
LT
1747/*
1748 * Remove a file after making sure there are no pending writes,
1749 * and after checking that the file has only one user.
1750 *
1751 * We invalidate the attribute cache and free the inode prior to the operation
1752 * to avoid possible races if the server reuses the inode.
1753 */
1754static int nfs_safe_remove(struct dentry *dentry)
1755{
1756 struct inode *dir = dentry->d_parent->d_inode;
1757 struct inode *inode = dentry->d_inode;
1758 int error = -EBUSY;
1759
6de1472f 1760 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1da177e4
LT
1761
1762 /* If the dentry was sillyrenamed, we simply call d_delete() */
1763 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1764 error = 0;
1765 goto out;
1766 }
1767
1ca42382 1768 trace_nfs_remove_enter(dir, dentry);
1da177e4 1769 if (inode != NULL) {
57ec14c5 1770 NFS_PROTO(inode)->return_delegation(inode);
1da177e4 1771 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1da177e4 1772 if (error == 0)
1b83d707 1773 nfs_drop_nlink(inode);
1da177e4
LT
1774 } else
1775 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
d45b9d8b
TM
1776 if (error == -ENOENT)
1777 nfs_dentry_handle_enoent(dentry);
1ca42382 1778 trace_nfs_remove_exit(dir, dentry, error);
1da177e4
LT
1779out:
1780 return error;
1781}
1782
1783/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1784 * belongs to an active ".nfs..." file and we return -EBUSY.
1785 *
1786 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1787 */
597d9289 1788int nfs_unlink(struct inode *dir, struct dentry *dentry)
1da177e4
LT
1789{
1790 int error;
1791 int need_rehash = 0;
1792
1e8968c5 1793 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
6de1472f 1794 dir->i_ino, dentry);
1da177e4 1795
1ca42382 1796 trace_nfs_unlink_enter(dir, dentry);
1da177e4 1797 spin_lock(&dentry->d_lock);
84d08fa8 1798 if (d_count(dentry) > 1) {
1da177e4 1799 spin_unlock(&dentry->d_lock);
ccfeb506
TM
1800 /* Start asynchronous writeout of the inode */
1801 write_inode_now(dentry->d_inode, 0);
1da177e4 1802 error = nfs_sillyrename(dir, dentry);
1ca42382 1803 goto out;
1da177e4
LT
1804 }
1805 if (!d_unhashed(dentry)) {
1806 __d_drop(dentry);
1807 need_rehash = 1;
1808 }
1809 spin_unlock(&dentry->d_lock);
1da177e4 1810 error = nfs_safe_remove(dentry);
d45b9d8b 1811 if (!error || error == -ENOENT) {
1da177e4
LT
1812 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1813 } else if (need_rehash)
1814 d_rehash(dentry);
1ca42382
TM
1815out:
1816 trace_nfs_unlink_exit(dir, dentry, error);
1da177e4
LT
1817 return error;
1818}
ddda8e0a 1819EXPORT_SYMBOL_GPL(nfs_unlink);
1da177e4 1820
873101b3
CL
1821/*
1822 * To create a symbolic link, most file systems instantiate a new inode,
1823 * add a page to it containing the path, then write it out to the disk
1824 * using prepare_write/commit_write.
1825 *
1826 * Unfortunately the NFS client can't create the in-core inode first
1827 * because it needs a file handle to create an in-core inode (see
1828 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1829 * symlink request has completed on the server.
1830 *
1831 * So instead we allocate a raw page, copy the symname into it, then do
1832 * the SYMLINK request with the page as the buffer. If it succeeds, we
1833 * now have a new file handle and can instantiate an in-core NFS inode
1834 * and move the raw page into its mapping.
1835 */
597d9289 1836int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1da177e4 1837{
873101b3
CL
1838 struct page *page;
1839 char *kaddr;
1da177e4 1840 struct iattr attr;
873101b3 1841 unsigned int pathlen = strlen(symname);
1da177e4
LT
1842 int error;
1843
1e8968c5 1844 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
6de1472f 1845 dir->i_ino, dentry, symname);
1da177e4 1846
873101b3
CL
1847 if (pathlen > PAGE_SIZE)
1848 return -ENAMETOOLONG;
1da177e4 1849
873101b3
CL
1850 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1851 attr.ia_valid = ATTR_MODE;
1da177e4 1852
83d93f22 1853 page = alloc_page(GFP_HIGHUSER);
76566991 1854 if (!page)
873101b3 1855 return -ENOMEM;
873101b3 1856
2b86ce2d 1857 kaddr = kmap_atomic(page);
873101b3
CL
1858 memcpy(kaddr, symname, pathlen);
1859 if (pathlen < PAGE_SIZE)
1860 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2b86ce2d 1861 kunmap_atomic(kaddr);
873101b3 1862
1ca42382 1863 trace_nfs_symlink_enter(dir, dentry);
94a6d753 1864 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1ca42382 1865 trace_nfs_symlink_exit(dir, dentry, error);
873101b3 1866 if (error != 0) {
1e8968c5 1867 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
873101b3 1868 dir->i_sb->s_id, dir->i_ino,
6de1472f 1869 dentry, symname, error);
1da177e4 1870 d_drop(dentry);
873101b3 1871 __free_page(page);
873101b3
CL
1872 return error;
1873 }
1874
1875 /*
1876 * No big deal if we can't add this page to the page cache here.
1877 * READLINK will get the missing page from the server if needed.
1878 */
a0b8cab3 1879 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
873101b3 1880 GFP_KERNEL)) {
873101b3
CL
1881 SetPageUptodate(page);
1882 unlock_page(page);
1883 } else
1884 __free_page(page);
1885
873101b3 1886 return 0;
1da177e4 1887}
ddda8e0a 1888EXPORT_SYMBOL_GPL(nfs_symlink);
1da177e4 1889
597d9289 1890int
1da177e4
LT
1891nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1892{
1893 struct inode *inode = old_dentry->d_inode;
1894 int error;
1895
6de1472f
AV
1896 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1897 old_dentry, dentry);
1da177e4 1898
1fd1085b 1899 trace_nfs_link_enter(inode, dir, dentry);
57ec14c5 1900 NFS_PROTO(inode)->return_delegation(inode);
9a3936aa 1901
9697d234 1902 d_drop(dentry);
1da177e4 1903 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
cf809556 1904 if (error == 0) {
7de9c6ee 1905 ihold(inode);
9697d234 1906 d_add(dentry, inode);
cf809556 1907 }
1fd1085b 1908 trace_nfs_link_exit(inode, dir, dentry, error);
1da177e4
LT
1909 return error;
1910}
ddda8e0a 1911EXPORT_SYMBOL_GPL(nfs_link);
1da177e4
LT
1912
1913/*
1914 * RENAME
1915 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1916 * different file handle for the same inode after a rename (e.g. when
1917 * moving to a different directory). A fail-safe method to do so would
1918 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1919 * rename the old file using the sillyrename stuff. This way, the original
1920 * file in old_dir will go away when the last process iput()s the inode.
1921 *
1922 * FIXED.
1923 *
1924 * It actually works quite well. One needs to have the possibility for
1925 * at least one ".nfs..." file in each directory the file ever gets
1926 * moved or linked to which happens automagically with the new
1927 * implementation that only depends on the dcache stuff instead of
1928 * using the inode layer
1929 *
1930 * Unfortunately, things are a little more complicated than indicated
1931 * above. For a cross-directory move, we want to make sure we can get
1932 * rid of the old inode after the operation. This means there must be
1933 * no pending writes (if it's a file), and the use count must be 1.
1934 * If these conditions are met, we can drop the dentries before doing
1935 * the rename.
1936 */
597d9289 1937int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1da177e4
LT
1938 struct inode *new_dir, struct dentry *new_dentry)
1939{
1940 struct inode *old_inode = old_dentry->d_inode;
1941 struct inode *new_inode = new_dentry->d_inode;
1942 struct dentry *dentry = NULL, *rehash = NULL;
1943 int error = -EBUSY;
1944
6de1472f
AV
1945 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1946 old_dentry, new_dentry,
84d08fa8 1947 d_count(new_dentry));
1da177e4 1948
70ded201 1949 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1da177e4 1950 /*
28f79a1a
MS
1951 * For non-directories, check whether the target is busy and if so,
1952 * make a copy of the dentry and then do a silly-rename. If the
1953 * silly-rename succeeds, the copied dentry is hashed and becomes
1954 * the new target.
1da177e4 1955 */
27226104
MS
1956 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1957 /*
1958 * To prevent any new references to the target during the
1959 * rename, we unhash the dentry in advance.
1960 */
1961 if (!d_unhashed(new_dentry)) {
1962 d_drop(new_dentry);
1963 rehash = new_dentry;
1964 }
1da177e4 1965
84d08fa8 1966 if (d_count(new_dentry) > 2) {
27226104
MS
1967 int err;
1968
1969 /* copy the target dentry's name */
1970 dentry = d_alloc(new_dentry->d_parent,
1971 &new_dentry->d_name);
1972 if (!dentry)
1973 goto out;
1974
1975 /* silly-rename the existing target ... */
1976 err = nfs_sillyrename(new_dir, new_dentry);
24e93025 1977 if (err)
27226104 1978 goto out;
24e93025
MS
1979
1980 new_dentry = dentry;
56335936 1981 rehash = NULL;
24e93025 1982 new_inode = NULL;
27226104 1983 }
b1e4adf4 1984 }
1da177e4 1985
57ec14c5 1986 NFS_PROTO(old_inode)->return_delegation(old_inode);
b1e4adf4 1987 if (new_inode != NULL)
57ec14c5 1988 NFS_PROTO(new_inode)->return_delegation(new_inode);
1da177e4 1989
1da177e4
LT
1990 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1991 new_dir, &new_dentry->d_name);
5ba7cc48 1992 nfs_mark_for_revalidate(old_inode);
1da177e4
LT
1993out:
1994 if (rehash)
1995 d_rehash(rehash);
70ded201
TM
1996 trace_nfs_rename_exit(old_dir, old_dentry,
1997 new_dir, new_dentry, error);
1da177e4 1998 if (!error) {
b1e4adf4
TM
1999 if (new_inode != NULL)
2000 nfs_drop_nlink(new_inode);
349457cc 2001 d_move(old_dentry, new_dentry);
8fb559f8
CL
2002 nfs_set_verifier(new_dentry,
2003 nfs_save_change_attribute(new_dir));
d45b9d8b
TM
2004 } else if (error == -ENOENT)
2005 nfs_dentry_handle_enoent(old_dentry);
1da177e4
LT
2006
2007 /* new dentry created? */
2008 if (dentry)
2009 dput(dentry);
1da177e4
LT
2010 return error;
2011}
ddda8e0a 2012EXPORT_SYMBOL_GPL(nfs_rename);
1da177e4 2013
cfcea3e8
TM
2014static DEFINE_SPINLOCK(nfs_access_lru_lock);
2015static LIST_HEAD(nfs_access_lru_list);
2016static atomic_long_t nfs_access_nr_entries;
2017
1c3c07e9
TM
2018static void nfs_access_free_entry(struct nfs_access_entry *entry)
2019{
2020 put_rpccred(entry->cred);
2021 kfree(entry);
cfcea3e8
TM
2022 smp_mb__before_atomic_dec();
2023 atomic_long_dec(&nfs_access_nr_entries);
2024 smp_mb__after_atomic_dec();
1c3c07e9
TM
2025}
2026
1a81bb8a
TM
2027static void nfs_access_free_list(struct list_head *head)
2028{
2029 struct nfs_access_entry *cache;
2030
2031 while (!list_empty(head)) {
2032 cache = list_entry(head->next, struct nfs_access_entry, lru);
2033 list_del(&cache->lru);
2034 nfs_access_free_entry(cache);
2035 }
2036}
2037
1ab6c499
DC
2038unsigned long
2039nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
979df72e
TM
2040{
2041 LIST_HEAD(head);
aa510da5 2042 struct nfs_inode *nfsi, *next;
979df72e 2043 struct nfs_access_entry *cache;
1495f230
YH
2044 int nr_to_scan = sc->nr_to_scan;
2045 gfp_t gfp_mask = sc->gfp_mask;
1ab6c499 2046 long freed = 0;
979df72e 2047
61d5eb29 2048 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1ab6c499 2049 return SHRINK_STOP;
9c7e7e23 2050
a50f7951 2051 spin_lock(&nfs_access_lru_lock);
aa510da5 2052 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
979df72e
TM
2053 struct inode *inode;
2054
2055 if (nr_to_scan-- == 0)
2056 break;
9c7e7e23 2057 inode = &nfsi->vfs_inode;
979df72e
TM
2058 spin_lock(&inode->i_lock);
2059 if (list_empty(&nfsi->access_cache_entry_lru))
2060 goto remove_lru_entry;
2061 cache = list_entry(nfsi->access_cache_entry_lru.next,
2062 struct nfs_access_entry, lru);
2063 list_move(&cache->lru, &head);
2064 rb_erase(&cache->rb_node, &nfsi->access_cache);
1ab6c499 2065 freed++;
979df72e
TM
2066 if (!list_empty(&nfsi->access_cache_entry_lru))
2067 list_move_tail(&nfsi->access_cache_inode_lru,
2068 &nfs_access_lru_list);
2069 else {
2070remove_lru_entry:
2071 list_del_init(&nfsi->access_cache_inode_lru);
9c7e7e23 2072 smp_mb__before_clear_bit();
979df72e 2073 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
9c7e7e23 2074 smp_mb__after_clear_bit();
979df72e 2075 }
59844a9b 2076 spin_unlock(&inode->i_lock);
979df72e
TM
2077 }
2078 spin_unlock(&nfs_access_lru_lock);
1a81bb8a 2079 nfs_access_free_list(&head);
1ab6c499
DC
2080 return freed;
2081}
2082
2083unsigned long
2084nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2085{
55f841ce 2086 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
979df72e
TM
2087}
2088
1a81bb8a 2089static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
1da177e4 2090{
1c3c07e9 2091 struct rb_root *root_node = &nfsi->access_cache;
1a81bb8a 2092 struct rb_node *n;
1c3c07e9
TM
2093 struct nfs_access_entry *entry;
2094
2095 /* Unhook entries from the cache */
2096 while ((n = rb_first(root_node)) != NULL) {
2097 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2098 rb_erase(n, root_node);
1a81bb8a 2099 list_move(&entry->lru, head);
1c3c07e9
TM
2100 }
2101 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1da177e4
LT
2102}
2103
1c3c07e9 2104void nfs_access_zap_cache(struct inode *inode)
1da177e4 2105{
1a81bb8a
TM
2106 LIST_HEAD(head);
2107
2108 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2109 return;
cfcea3e8 2110 /* Remove from global LRU init */
1a81bb8a
TM
2111 spin_lock(&nfs_access_lru_lock);
2112 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
cfcea3e8 2113 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
cfcea3e8 2114
1c3c07e9 2115 spin_lock(&inode->i_lock);
1a81bb8a
TM
2116 __nfs_access_zap_cache(NFS_I(inode), &head);
2117 spin_unlock(&inode->i_lock);
2118 spin_unlock(&nfs_access_lru_lock);
2119 nfs_access_free_list(&head);
1c3c07e9 2120}
1c606fb7 2121EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
1da177e4 2122
1c3c07e9
TM
2123static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2124{
2125 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2126 struct nfs_access_entry *entry;
2127
2128 while (n != NULL) {
2129 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2130
2131 if (cred < entry->cred)
2132 n = n->rb_left;
2133 else if (cred > entry->cred)
2134 n = n->rb_right;
2135 else
2136 return entry;
1da177e4 2137 }
1c3c07e9
TM
2138 return NULL;
2139}
2140
af22f94a 2141static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1c3c07e9
TM
2142{
2143 struct nfs_inode *nfsi = NFS_I(inode);
2144 struct nfs_access_entry *cache;
2145 int err = -ENOENT;
2146
dc59250c 2147 spin_lock(&inode->i_lock);
1c3c07e9
TM
2148 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2149 goto out_zap;
2150 cache = nfs_access_search_rbtree(inode, cred);
2151 if (cache == NULL)
2152 goto out;
b4d2314b 2153 if (!nfs_have_delegated_attributes(inode) &&
64672d55 2154 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1c3c07e9
TM
2155 goto out_stale;
2156 res->jiffies = cache->jiffies;
2157 res->cred = cache->cred;
2158 res->mask = cache->mask;
cfcea3e8 2159 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1c3c07e9
TM
2160 err = 0;
2161out:
2162 spin_unlock(&inode->i_lock);
2163 return err;
2164out_stale:
2165 rb_erase(&cache->rb_node, &nfsi->access_cache);
cfcea3e8 2166 list_del(&cache->lru);
1c3c07e9
TM
2167 spin_unlock(&inode->i_lock);
2168 nfs_access_free_entry(cache);
2169 return -ENOENT;
2170out_zap:
1a81bb8a
TM
2171 spin_unlock(&inode->i_lock);
2172 nfs_access_zap_cache(inode);
1c3c07e9
TM
2173 return -ENOENT;
2174}
2175
2176static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2177{
cfcea3e8
TM
2178 struct nfs_inode *nfsi = NFS_I(inode);
2179 struct rb_root *root_node = &nfsi->access_cache;
1c3c07e9
TM
2180 struct rb_node **p = &root_node->rb_node;
2181 struct rb_node *parent = NULL;
2182 struct nfs_access_entry *entry;
2183
2184 spin_lock(&inode->i_lock);
2185 while (*p != NULL) {
2186 parent = *p;
2187 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2188
2189 if (set->cred < entry->cred)
2190 p = &parent->rb_left;
2191 else if (set->cred > entry->cred)
2192 p = &parent->rb_right;
2193 else
2194 goto found;
2195 }
2196 rb_link_node(&set->rb_node, parent, p);
2197 rb_insert_color(&set->rb_node, root_node);
cfcea3e8 2198 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
dc59250c 2199 spin_unlock(&inode->i_lock);
1c3c07e9
TM
2200 return;
2201found:
2202 rb_replace_node(parent, &set->rb_node, root_node);
cfcea3e8
TM
2203 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2204 list_del(&entry->lru);
1c3c07e9
TM
2205 spin_unlock(&inode->i_lock);
2206 nfs_access_free_entry(entry);
2207}
2208
6168f62c 2209void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1c3c07e9
TM
2210{
2211 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2212 if (cache == NULL)
2213 return;
2214 RB_CLEAR_NODE(&cache->rb_node);
1da177e4 2215 cache->jiffies = set->jiffies;
1c3c07e9 2216 cache->cred = get_rpccred(set->cred);
1da177e4 2217 cache->mask = set->mask;
1c3c07e9
TM
2218
2219 nfs_access_add_rbtree(inode, cache);
cfcea3e8
TM
2220
2221 /* Update accounting */
2222 smp_mb__before_atomic_inc();
2223 atomic_long_inc(&nfs_access_nr_entries);
2224 smp_mb__after_atomic_inc();
2225
2226 /* Add inode to global LRU list */
1a81bb8a 2227 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
cfcea3e8 2228 spin_lock(&nfs_access_lru_lock);
1a81bb8a
TM
2229 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2230 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2231 &nfs_access_lru_list);
cfcea3e8
TM
2232 spin_unlock(&nfs_access_lru_lock);
2233 }
1da177e4 2234}
6168f62c
WAA
2235EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2236
2237void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2238{
2239 entry->mask = 0;
2240 if (access_result & NFS4_ACCESS_READ)
2241 entry->mask |= MAY_READ;
2242 if (access_result &
2243 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2244 entry->mask |= MAY_WRITE;
2245 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2246 entry->mask |= MAY_EXEC;
2247}
2248EXPORT_SYMBOL_GPL(nfs_access_set_mask);
1da177e4
LT
2249
2250static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2251{
2252 struct nfs_access_entry cache;
2253 int status;
2254
f4ce1299
TM
2255 trace_nfs_access_enter(inode);
2256
1da177e4
LT
2257 status = nfs_access_get_cached(inode, cred, &cache);
2258 if (status == 0)
f4ce1299 2259 goto out_cached;
1da177e4
LT
2260
2261 /* Be clever: ask server to check for all possible rights */
2262 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2263 cache.cred = cred;
2264 cache.jiffies = jiffies;
2265 status = NFS_PROTO(inode)->access(inode, &cache);
a71ee337
SJ
2266 if (status != 0) {
2267 if (status == -ESTALE) {
2268 nfs_zap_caches(inode);
2269 if (!S_ISDIR(inode->i_mode))
2270 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2271 }
f4ce1299 2272 goto out;
a71ee337 2273 }
1da177e4 2274 nfs_access_add_cache(inode, &cache);
f4ce1299
TM
2275out_cached:
2276 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2277 status = -EACCES;
1da177e4 2278out:
f4ce1299
TM
2279 trace_nfs_access_exit(inode, status);
2280 return status;
1da177e4
LT
2281}
2282
af22f94a
TM
2283static int nfs_open_permission_mask(int openflags)
2284{
2285 int mask = 0;
2286
f8d9a897
WAA
2287 if (openflags & __FMODE_EXEC) {
2288 /* ONLY check exec rights */
2289 mask = MAY_EXEC;
2290 } else {
2291 if ((openflags & O_ACCMODE) != O_WRONLY)
2292 mask |= MAY_READ;
2293 if ((openflags & O_ACCMODE) != O_RDONLY)
2294 mask |= MAY_WRITE;
2295 }
2296
af22f94a
TM
2297 return mask;
2298}
2299
2300int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2301{
2302 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2303}
89d77c8f 2304EXPORT_SYMBOL_GPL(nfs_may_open);
af22f94a 2305
10556cb2 2306int nfs_permission(struct inode *inode, int mask)
1da177e4
LT
2307{
2308 struct rpc_cred *cred;
2309 int res = 0;
2310
10556cb2 2311 if (mask & MAY_NOT_BLOCK)
b74c79e9
NP
2312 return -ECHILD;
2313
91d5b470
CL
2314 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2315
e6305c43 2316 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4
LT
2317 goto out;
2318 /* Is this sys_access() ? */
9cfcac81 2319 if (mask & (MAY_ACCESS | MAY_CHDIR))
1da177e4
LT
2320 goto force_lookup;
2321
2322 switch (inode->i_mode & S_IFMT) {
2323 case S_IFLNK:
2324 goto out;
2325 case S_IFREG:
1da177e4
LT
2326 break;
2327 case S_IFDIR:
2328 /*
2329 * Optimize away all write operations, since the server
2330 * will check permissions when we perform the op.
2331 */
2332 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2333 goto out;
2334 }
2335
2336force_lookup:
1da177e4
LT
2337 if (!NFS_PROTO(inode)->access)
2338 goto out_notsup;
2339
98a8e323 2340 cred = rpc_lookup_cred();
1da177e4
LT
2341 if (!IS_ERR(cred)) {
2342 res = nfs_do_access(inode, cred, mask);
2343 put_rpccred(cred);
2344 } else
2345 res = PTR_ERR(cred);
1da177e4 2346out:
f696a365
MS
2347 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2348 res = -EACCES;
2349
1e8968c5 2350 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
1e7cb3dc 2351 inode->i_sb->s_id, inode->i_ino, mask, res);
1da177e4
LT
2352 return res;
2353out_notsup:
2354 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2355 if (res == 0)
2830ba7f 2356 res = generic_permission(inode, mask);
1e7cb3dc 2357 goto out;
1da177e4 2358}
ddda8e0a 2359EXPORT_SYMBOL_GPL(nfs_permission);
1da177e4
LT
2360
2361/*
2362 * Local variables:
2363 * version-control: t
2364 * kept-new-versions: 5
2365 * End:
2366 */