Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/nfs/dir.c | |
4 | * | |
5 | * Copyright (C) 1992 Rick Sladkey | |
6 | * | |
7 | * nfs directory handling functions | |
8 | * | |
9 | * 10 Apr 1996 Added silly rename for unlink --okir | |
10 | * 28 Sep 1996 Improved directory cache --okir | |
11 | * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de | |
12 | * Re-implemented silly rename for unlink, newly implemented | |
13 | * silly rename for nfs_rename() following the suggestions | |
14 | * of Olaf Kirch (okir) found in this file. | |
15 | * Following Linus comments on my original hack, this version | |
16 | * depends only on the dcache stuff and doesn't touch the inode | |
17 | * layer (iput() and friends). | |
18 | * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM | |
19 | */ | |
20 | ||
ddda8e0a | 21 | #include <linux/module.h> |
1da177e4 LT |
22 | #include <linux/time.h> |
23 | #include <linux/errno.h> | |
24 | #include <linux/stat.h> | |
25 | #include <linux/fcntl.h> | |
26 | #include <linux/string.h> | |
27 | #include <linux/kernel.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/mm.h> | |
30 | #include <linux/sunrpc/clnt.h> | |
31 | #include <linux/nfs_fs.h> | |
32 | #include <linux/nfs_mount.h> | |
33 | #include <linux/pagemap.h> | |
873101b3 | 34 | #include <linux/pagevec.h> |
1da177e4 | 35 | #include <linux/namei.h> |
54ceac45 | 36 | #include <linux/mount.h> |
a0b8cab3 | 37 | #include <linux/swap.h> |
e8edc6e0 | 38 | #include <linux/sched.h> |
04e4bd1c | 39 | #include <linux/kmemleak.h> |
64c2ce8b | 40 | #include <linux/xattr.h> |
1da177e4 LT |
41 | |
42 | #include "delegation.h" | |
91d5b470 | 43 | #include "iostat.h" |
4c30d56e | 44 | #include "internal.h" |
cd9a1c0e | 45 | #include "fscache.h" |
1da177e4 | 46 | |
f4ce1299 TM |
47 | #include "nfstrace.h" |
48 | ||
1da177e4 LT |
49 | /* #define NFS_DEBUG_VERBOSE 1 */ |
50 | ||
51 | static int nfs_opendir(struct inode *, struct file *); | |
480c2006 | 52 | static int nfs_closedir(struct inode *, struct file *); |
23db8620 | 53 | static int nfs_readdir(struct file *, struct dir_context *); |
02c24a82 | 54 | static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); |
f0dd2136 | 55 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
11de3b11 | 56 | static void nfs_readdir_clear_array(struct page*); |
1da177e4 | 57 | |
4b6f5d20 | 58 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 59 | .llseek = nfs_llseek_dir, |
1da177e4 | 60 | .read = generic_read_dir, |
b044f645 | 61 | .iterate = nfs_readdir, |
1da177e4 | 62 | .open = nfs_opendir, |
480c2006 | 63 | .release = nfs_closedir, |
1da177e4 LT |
64 | .fsync = nfs_fsync_dir, |
65 | }; | |
66 | ||
11de3b11 TM |
67 | const struct address_space_operations nfs_dir_aops = { |
68 | .freepage = nfs_readdir_clear_array, | |
d1bacf9e BS |
69 | }; |
70 | ||
684f39b4 | 71 | static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, const struct cred *cred) |
480c2006 | 72 | { |
311324ad | 73 | struct nfs_inode *nfsi = NFS_I(dir); |
480c2006 BS |
74 | struct nfs_open_dir_context *ctx; |
75 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
76 | if (ctx != NULL) { | |
8ef2ce3e | 77 | ctx->duped = 0; |
311324ad | 78 | ctx->attr_gencount = nfsi->attr_gencount; |
480c2006 | 79 | ctx->dir_cookie = 0; |
8ef2ce3e | 80 | ctx->dup_cookie = 0; |
684f39b4 | 81 | ctx->cred = get_cred(cred); |
311324ad | 82 | spin_lock(&dir->i_lock); |
1c341b77 TM |
83 | if (list_empty(&nfsi->open_files) && |
84 | (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) | |
85 | nfsi->cache_validity |= NFS_INO_INVALID_DATA | | |
86 | NFS_INO_REVAL_FORCED; | |
311324ad TM |
87 | list_add(&ctx->list, &nfsi->open_files); |
88 | spin_unlock(&dir->i_lock); | |
0c030806 TM |
89 | return ctx; |
90 | } | |
91 | return ERR_PTR(-ENOMEM); | |
480c2006 BS |
92 | } |
93 | ||
311324ad | 94 | static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) |
480c2006 | 95 | { |
311324ad TM |
96 | spin_lock(&dir->i_lock); |
97 | list_del(&ctx->list); | |
98 | spin_unlock(&dir->i_lock); | |
684f39b4 | 99 | put_cred(ctx->cred); |
480c2006 BS |
100 | kfree(ctx); |
101 | } | |
102 | ||
1da177e4 LT |
103 | /* |
104 | * Open file | |
105 | */ | |
106 | static int | |
107 | nfs_opendir(struct inode *inode, struct file *filp) | |
108 | { | |
480c2006 BS |
109 | int res = 0; |
110 | struct nfs_open_dir_context *ctx; | |
1da177e4 | 111 | |
6de1472f | 112 | dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); |
cc0dd2d1 CL |
113 | |
114 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 115 | |
684f39b4 | 116 | ctx = alloc_nfs_open_dir_context(inode, current_cred()); |
480c2006 BS |
117 | if (IS_ERR(ctx)) { |
118 | res = PTR_ERR(ctx); | |
119 | goto out; | |
120 | } | |
121 | filp->private_data = ctx; | |
480c2006 | 122 | out: |
1da177e4 LT |
123 | return res; |
124 | } | |
125 | ||
480c2006 BS |
126 | static int |
127 | nfs_closedir(struct inode *inode, struct file *filp) | |
128 | { | |
a455589f | 129 | put_nfs_open_dir_context(file_inode(filp), filp->private_data); |
480c2006 BS |
130 | return 0; |
131 | } | |
132 | ||
d1bacf9e BS |
133 | struct nfs_cache_array_entry { |
134 | u64 cookie; | |
135 | u64 ino; | |
136 | struct qstr string; | |
0b26a0bf | 137 | unsigned char d_type; |
d1bacf9e BS |
138 | }; |
139 | ||
140 | struct nfs_cache_array { | |
88b8e133 | 141 | int size; |
d1bacf9e BS |
142 | int eof_index; |
143 | u64 last_cookie; | |
144 | struct nfs_cache_array_entry array[0]; | |
145 | }; | |
146 | ||
a7a3b1e9 | 147 | typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool); |
1da177e4 LT |
148 | typedef struct { |
149 | struct file *file; | |
150 | struct page *page; | |
23db8620 | 151 | struct dir_context *ctx; |
1da177e4 | 152 | unsigned long page_index; |
f0dd2136 | 153 | u64 *dir_cookie; |
0aded708 | 154 | u64 last_cookie; |
f0dd2136 | 155 | loff_t current_index; |
1da177e4 | 156 | decode_dirent_t decode; |
d1bacf9e | 157 | |
1f4eab7e | 158 | unsigned long timestamp; |
4704f0e2 | 159 | unsigned long gencount; |
d1bacf9e | 160 | unsigned int cache_entry_index; |
a7a3b1e9 BC |
161 | bool plus; |
162 | bool eof; | |
1da177e4 LT |
163 | } nfs_readdir_descriptor_t; |
164 | ||
4b310319 TM |
165 | static |
166 | void nfs_readdir_init_array(struct page *page) | |
167 | { | |
168 | struct nfs_cache_array *array; | |
169 | ||
170 | array = kmap_atomic(page); | |
171 | memset(array, 0, sizeof(struct nfs_cache_array)); | |
172 | array->eof_index = -1; | |
173 | kunmap_atomic(array); | |
174 | } | |
175 | ||
d1bacf9e BS |
176 | /* |
177 | * we are freeing strings created by nfs_add_to_readdir_array() | |
178 | */ | |
179 | static | |
11de3b11 | 180 | void nfs_readdir_clear_array(struct page *page) |
d1bacf9e | 181 | { |
11de3b11 | 182 | struct nfs_cache_array *array; |
d1bacf9e | 183 | int i; |
8cd51a0c | 184 | |
2b86ce2d | 185 | array = kmap_atomic(page); |
b044f645 BC |
186 | for (i = 0; i < array->size; i++) |
187 | kfree(array->array[i].string.name); | |
4b310319 | 188 | array->size = 0; |
2b86ce2d | 189 | kunmap_atomic(array); |
d1bacf9e BS |
190 | } |
191 | ||
192 | /* | |
193 | * the caller is responsible for freeing qstr.name | |
194 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
195 | * nfs_clear_readdir_array() | |
196 | */ | |
197 | static | |
4a201d6e | 198 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
199 | { |
200 | string->len = len; | |
201 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
202 | if (string->name == NULL) |
203 | return -ENOMEM; | |
04e4bd1c CM |
204 | /* |
205 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
206 | * in a page cache page which kmemleak does not scan. | |
207 | */ | |
208 | kmemleak_not_leak(string->name); | |
8387ff25 | 209 | string->hash = full_name_hash(NULL, name, len); |
4a201d6e | 210 | return 0; |
d1bacf9e BS |
211 | } |
212 | ||
213 | static | |
214 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
215 | { | |
0795bf83 | 216 | struct nfs_cache_array *array = kmap(page); |
4a201d6e TM |
217 | struct nfs_cache_array_entry *cache_entry; |
218 | int ret; | |
219 | ||
3020093f TM |
220 | cache_entry = &array->array[array->size]; |
221 | ||
222 | /* Check that this entry lies within the page bounds */ | |
8cd51a0c | 223 | ret = -ENOSPC; |
3020093f | 224 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) |
4a201d6e | 225 | goto out; |
d1bacf9e | 226 | |
4a201d6e TM |
227 | cache_entry->cookie = entry->prev_cookie; |
228 | cache_entry->ino = entry->ino; | |
0b26a0bf | 229 | cache_entry->d_type = entry->d_type; |
4a201d6e TM |
230 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); |
231 | if (ret) | |
232 | goto out; | |
d1bacf9e | 233 | array->last_cookie = entry->cookie; |
8cd51a0c | 234 | array->size++; |
47c716cb | 235 | if (entry->eof != 0) |
d1bacf9e | 236 | array->eof_index = array->size; |
4a201d6e | 237 | out: |
0795bf83 | 238 | kunmap(page); |
4a201d6e | 239 | return ret; |
d1bacf9e BS |
240 | } |
241 | ||
242 | static | |
243 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
244 | { | |
23db8620 | 245 | loff_t diff = desc->ctx->pos - desc->current_index; |
d1bacf9e BS |
246 | unsigned int index; |
247 | ||
248 | if (diff < 0) | |
249 | goto out_eof; | |
250 | if (diff >= array->size) { | |
8cd51a0c | 251 | if (array->eof_index >= 0) |
d1bacf9e | 252 | goto out_eof; |
d1bacf9e BS |
253 | return -EAGAIN; |
254 | } | |
255 | ||
256 | index = (unsigned int)diff; | |
257 | *desc->dir_cookie = array->array[index].cookie; | |
258 | desc->cache_entry_index = index; | |
d1bacf9e BS |
259 | return 0; |
260 | out_eof: | |
6089dd0d | 261 | desc->eof = true; |
d1bacf9e BS |
262 | return -EBADCOOKIE; |
263 | } | |
264 | ||
4db72b40 JL |
265 | static bool |
266 | nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) | |
267 | { | |
268 | if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) | |
269 | return false; | |
270 | smp_rmb(); | |
271 | return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); | |
272 | } | |
273 | ||
d1bacf9e BS |
274 | static |
275 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
276 | { | |
277 | int i; | |
8ef2ce3e | 278 | loff_t new_pos; |
d1bacf9e BS |
279 | int status = -EAGAIN; |
280 | ||
281 | for (i = 0; i < array->size; i++) { | |
d1bacf9e | 282 | if (array->array[i].cookie == *desc->dir_cookie) { |
496ad9aa | 283 | struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); |
0c030806 TM |
284 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
285 | ||
8ef2ce3e | 286 | new_pos = desc->current_index + i; |
4db72b40 JL |
287 | if (ctx->attr_gencount != nfsi->attr_gencount || |
288 | !nfs_readdir_inode_mapping_valid(nfsi)) { | |
0c030806 TM |
289 | ctx->duped = 0; |
290 | ctx->attr_gencount = nfsi->attr_gencount; | |
23db8620 | 291 | } else if (new_pos < desc->ctx->pos) { |
0c030806 TM |
292 | if (ctx->duped > 0 |
293 | && ctx->dup_cookie == *desc->dir_cookie) { | |
294 | if (printk_ratelimit()) { | |
6de1472f | 295 | pr_notice("NFS: directory %pD2 contains a readdir loop." |
0c030806 | 296 | "Please contact your server vendor. " |
9581a4ae JL |
297 | "The file: %.*s has duplicate cookie %llu\n", |
298 | desc->file, array->array[i].string.len, | |
299 | array->array[i].string.name, *desc->dir_cookie); | |
0c030806 TM |
300 | } |
301 | status = -ELOOP; | |
302 | goto out; | |
303 | } | |
8ef2ce3e | 304 | ctx->dup_cookie = *desc->dir_cookie; |
0c030806 | 305 | ctx->duped = -1; |
8ef2ce3e | 306 | } |
23db8620 | 307 | desc->ctx->pos = new_pos; |
d1bacf9e | 308 | desc->cache_entry_index = i; |
47c716cb | 309 | return 0; |
d1bacf9e BS |
310 | } |
311 | } | |
47c716cb | 312 | if (array->eof_index >= 0) { |
8cd51a0c | 313 | status = -EBADCOOKIE; |
18fb5fe4 | 314 | if (*desc->dir_cookie == array->last_cookie) |
6089dd0d | 315 | desc->eof = true; |
8cd51a0c | 316 | } |
0c030806 | 317 | out: |
d1bacf9e BS |
318 | return status; |
319 | } | |
320 | ||
321 | static | |
322 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
323 | { | |
324 | struct nfs_cache_array *array; | |
47c716cb | 325 | int status; |
d1bacf9e | 326 | |
0795bf83 | 327 | array = kmap(desc->page); |
d1bacf9e BS |
328 | |
329 | if (*desc->dir_cookie == 0) | |
330 | status = nfs_readdir_search_for_pos(array, desc); | |
331 | else | |
332 | status = nfs_readdir_search_for_cookie(array, desc); | |
333 | ||
47c716cb | 334 | if (status == -EAGAIN) { |
0aded708 | 335 | desc->last_cookie = array->last_cookie; |
e47c085a | 336 | desc->current_index += array->size; |
47c716cb TM |
337 | desc->page_index++; |
338 | } | |
0795bf83 | 339 | kunmap(desc->page); |
d1bacf9e BS |
340 | return status; |
341 | } | |
342 | ||
343 | /* Fill a page with xdr information before transferring to the cache page */ | |
344 | static | |
56e4ebf8 | 345 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 346 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 347 | { |
480c2006 | 348 | struct nfs_open_dir_context *ctx = file->private_data; |
684f39b4 | 349 | const struct cred *cred = ctx->cred; |
4704f0e2 | 350 | unsigned long timestamp, gencount; |
1da177e4 LT |
351 | int error; |
352 | ||
1da177e4 LT |
353 | again: |
354 | timestamp = jiffies; | |
4704f0e2 | 355 | gencount = nfs_inc_attr_generation_counter(); |
be62a1a8 | 356 | error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, |
1da177e4 LT |
357 | NFS_SERVER(inode)->dtsize, desc->plus); |
358 | if (error < 0) { | |
359 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
360 | if (error == -ENOTSUPP && desc->plus) { | |
361 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 362 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
a7a3b1e9 | 363 | desc->plus = false; |
1da177e4 LT |
364 | goto again; |
365 | } | |
366 | goto error; | |
367 | } | |
1f4eab7e | 368 | desc->timestamp = timestamp; |
4704f0e2 | 369 | desc->gencount = gencount; |
d1bacf9e BS |
370 | error: |
371 | return error; | |
1da177e4 LT |
372 | } |
373 | ||
573c4e1e CL |
374 | static int xdr_decode(nfs_readdir_descriptor_t *desc, |
375 | struct nfs_entry *entry, struct xdr_stream *xdr) | |
1da177e4 | 376 | { |
573c4e1e | 377 | int error; |
1da177e4 | 378 | |
573c4e1e CL |
379 | error = desc->decode(xdr, entry, desc->plus); |
380 | if (error) | |
381 | return error; | |
d1bacf9e BS |
382 | entry->fattr->time_start = desc->timestamp; |
383 | entry->fattr->gencount = desc->gencount; | |
384 | return 0; | |
1da177e4 LT |
385 | } |
386 | ||
fa923369 TM |
387 | /* Match file and dirent using either filehandle or fileid |
388 | * Note: caller is responsible for checking the fsid | |
389 | */ | |
d39ab9de BS |
390 | static |
391 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
392 | { | |
d8fdb47f | 393 | struct inode *inode; |
fa923369 TM |
394 | struct nfs_inode *nfsi; |
395 | ||
2b0143b5 DH |
396 | if (d_really_is_negative(dentry)) |
397 | return 0; | |
fa923369 | 398 | |
d8fdb47f TM |
399 | inode = d_inode(dentry); |
400 | if (is_bad_inode(inode) || NFS_STALE(inode)) | |
401 | return 0; | |
402 | ||
403 | nfsi = NFS_I(inode); | |
7dc72d5f TM |
404 | if (entry->fattr->fileid != nfsi->fileid) |
405 | return 0; | |
406 | if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) | |
407 | return 0; | |
408 | return 1; | |
d39ab9de BS |
409 | } |
410 | ||
d69ee9b8 | 411 | static |
23db8620 | 412 | bool 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 | /* | |
63519fbc TM |
424 | * This function is called by the lookup and getattr code to request the |
425 | * use of readdirplus to accelerate any future lookups in the same | |
d69ee9b8 TM |
426 | * directory. |
427 | */ | |
d69ee9b8 TM |
428 | void nfs_advise_use_readdirplus(struct inode *dir) |
429 | { | |
63519fbc TM |
430 | struct nfs_inode *nfsi = NFS_I(dir); |
431 | ||
432 | if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && | |
433 | !list_empty(&nfsi->open_files)) | |
434 | set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); | |
d69ee9b8 TM |
435 | } |
436 | ||
311324ad TM |
437 | /* |
438 | * This function is mainly for use by nfs_getattr(). | |
439 | * | |
440 | * If this is an 'ls -l', we want to force use of readdirplus. | |
441 | * Do this by checking if there is an active file descriptor | |
442 | * and calling nfs_advise_use_readdirplus, then forcing a | |
443 | * cache flush. | |
444 | */ | |
445 | void nfs_force_use_readdirplus(struct inode *dir) | |
446 | { | |
63519fbc TM |
447 | struct nfs_inode *nfsi = NFS_I(dir); |
448 | ||
449 | if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && | |
450 | !list_empty(&nfsi->open_files)) { | |
451 | set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); | |
79f687a3 | 452 | invalidate_mapping_pages(dir->i_mapping, 0, -1); |
311324ad TM |
453 | } |
454 | } | |
455 | ||
d39ab9de BS |
456 | static |
457 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
458 | { | |
26fe5750 | 459 | struct qstr filename = QSTR_INIT(entry->name, entry->len); |
9ac3d3e8 | 460 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
4a201d6e TM |
461 | struct dentry *dentry; |
462 | struct dentry *alias; | |
2b0143b5 | 463 | struct inode *dir = d_inode(parent); |
d39ab9de | 464 | struct inode *inode; |
aa9c2669 | 465 | int status; |
d39ab9de | 466 | |
fa923369 TM |
467 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) |
468 | return; | |
6c441c25 TM |
469 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) |
470 | return; | |
78d04af4 TM |
471 | if (filename.len == 0) |
472 | return; | |
473 | /* Validate that the name doesn't contain any illegal '\0' */ | |
474 | if (strnlen(filename.name, filename.len) != filename.len) | |
475 | return; | |
476 | /* ...or '/' */ | |
477 | if (strnchr(filename.name, filename.len, '/')) | |
478 | return; | |
4a201d6e TM |
479 | if (filename.name[0] == '.') { |
480 | if (filename.len == 1) | |
481 | return; | |
482 | if (filename.len == 2 && filename.name[1] == '.') | |
483 | return; | |
484 | } | |
8387ff25 | 485 | filename.hash = full_name_hash(parent, filename.name, filename.len); |
d39ab9de | 486 | |
4a201d6e | 487 | dentry = d_lookup(parent, &filename); |
9ac3d3e8 AV |
488 | again: |
489 | if (!dentry) { | |
490 | dentry = d_alloc_parallel(parent, &filename, &wq); | |
491 | if (IS_ERR(dentry)) | |
492 | return; | |
493 | } | |
494 | if (!d_in_lookup(dentry)) { | |
6c441c25 TM |
495 | /* Is there a mountpoint here? If so, just exit */ |
496 | if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, | |
497 | &entry->fattr->fsid)) | |
498 | goto out; | |
d39ab9de | 499 | if (nfs_same_file(dentry, entry)) { |
7dc72d5f TM |
500 | if (!entry->fh->size) |
501 | goto out; | |
cda57a1e | 502 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
2b0143b5 | 503 | status = nfs_refresh_inode(d_inode(dentry), entry->fattr); |
aa9c2669 | 504 | if (!status) |
2b0143b5 | 505 | nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); |
d39ab9de BS |
506 | goto out; |
507 | } else { | |
5542aa2f | 508 | d_invalidate(dentry); |
d39ab9de | 509 | dput(dentry); |
9ac3d3e8 AV |
510 | dentry = NULL; |
511 | goto again; | |
d39ab9de BS |
512 | } |
513 | } | |
7dc72d5f TM |
514 | if (!entry->fh->size) { |
515 | d_lookup_done(dentry); | |
516 | goto out; | |
517 | } | |
d39ab9de | 518 | |
1775fd3e | 519 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); |
41d28bca | 520 | alias = d_splice_alias(inode, dentry); |
9ac3d3e8 AV |
521 | d_lookup_done(dentry); |
522 | if (alias) { | |
523 | if (IS_ERR(alias)) | |
524 | goto out; | |
525 | dput(dentry); | |
526 | dentry = alias; | |
527 | } | |
528 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
d39ab9de BS |
529 | out: |
530 | dput(dentry); | |
d39ab9de BS |
531 | } |
532 | ||
d1bacf9e BS |
533 | /* Perform conversion from xdr to cache array */ |
534 | static | |
8cd51a0c | 535 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
6650239a | 536 | struct page **xdr_pages, struct page *page, unsigned int buflen) |
1da177e4 | 537 | { |
babddc72 | 538 | struct xdr_stream stream; |
f7da7a12 | 539 | struct xdr_buf buf; |
6650239a | 540 | struct page *scratch; |
99424380 | 541 | struct nfs_cache_array *array; |
5c346854 TM |
542 | unsigned int count = 0; |
543 | int status; | |
babddc72 | 544 | |
6650239a TM |
545 | scratch = alloc_page(GFP_KERNEL); |
546 | if (scratch == NULL) | |
547 | return -ENOMEM; | |
babddc72 | 548 | |
ce85cfbe BC |
549 | if (buflen == 0) |
550 | goto out_nopages; | |
551 | ||
f7da7a12 | 552 | xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); |
6650239a | 553 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
99424380 BS |
554 | |
555 | do { | |
556 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
557 | if (status != 0) { |
558 | if (status == -EAGAIN) | |
559 | status = 0; | |
99424380 | 560 | break; |
8cd51a0c | 561 | } |
99424380 | 562 | |
5c346854 TM |
563 | count++; |
564 | ||
a7a3b1e9 | 565 | if (desc->plus) |
be62a1a8 | 566 | nfs_prime_dcache(file_dentry(desc->file), entry); |
8cd51a0c | 567 | |
db531db9 | 568 | status = nfs_readdir_add_to_array(entry, page); |
8cd51a0c TM |
569 | if (status != 0) |
570 | break; | |
99424380 BS |
571 | } while (!entry->eof); |
572 | ||
ce85cfbe | 573 | out_nopages: |
47c716cb | 574 | if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { |
db531db9 | 575 | array = kmap(page); |
0795bf83 FF |
576 | array->eof_index = array->size; |
577 | status = 0; | |
db531db9 | 578 | kunmap(page); |
1da177e4 | 579 | } |
6650239a TM |
580 | |
581 | put_page(scratch); | |
8cd51a0c | 582 | return status; |
56e4ebf8 BS |
583 | } |
584 | ||
585 | static | |
c7e9668e | 586 | void nfs_readdir_free_pages(struct page **pages, unsigned int npages) |
56e4ebf8 BS |
587 | { |
588 | unsigned int i; | |
589 | for (i = 0; i < npages; i++) | |
590 | put_page(pages[i]); | |
591 | } | |
592 | ||
56e4ebf8 | 593 | /* |
bf211ca1 | 594 | * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call |
595 | * to nfs_readdir_free_pages() | |
56e4ebf8 BS |
596 | */ |
597 | static | |
c7e9668e | 598 | int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) |
56e4ebf8 | 599 | { |
56e4ebf8 BS |
600 | unsigned int i; |
601 | ||
602 | for (i = 0; i < npages; i++) { | |
603 | struct page *page = alloc_page(GFP_KERNEL); | |
604 | if (page == NULL) | |
605 | goto out_freepages; | |
606 | pages[i] = page; | |
607 | } | |
6650239a | 608 | return 0; |
56e4ebf8 | 609 | |
56e4ebf8 | 610 | out_freepages: |
c7e9668e | 611 | nfs_readdir_free_pages(pages, i); |
6650239a | 612 | return -ENOMEM; |
1da177e4 LT |
613 | } |
614 | ||
d1bacf9e BS |
615 | static |
616 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 617 | { |
56e4ebf8 | 618 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
d1bacf9e BS |
619 | struct nfs_entry entry; |
620 | struct file *file = desc->file; | |
621 | struct nfs_cache_array *array; | |
8cd51a0c | 622 | int status = -ENOMEM; |
56e4ebf8 | 623 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e | 624 | |
4b310319 TM |
625 | nfs_readdir_init_array(page); |
626 | ||
d1bacf9e | 627 | entry.prev_cookie = 0; |
0aded708 | 628 | entry.cookie = desc->last_cookie; |
d1bacf9e BS |
629 | entry.eof = 0; |
630 | entry.fh = nfs_alloc_fhandle(); | |
631 | entry.fattr = nfs_alloc_fattr(); | |
573c4e1e | 632 | entry.server = NFS_SERVER(inode); |
d1bacf9e BS |
633 | if (entry.fh == NULL || entry.fattr == NULL) |
634 | goto out; | |
00a92642 | 635 | |
14c43f76 DQ |
636 | entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); |
637 | if (IS_ERR(entry.label)) { | |
638 | status = PTR_ERR(entry.label); | |
639 | goto out; | |
640 | } | |
641 | ||
0795bf83 | 642 | array = kmap(page); |
00a92642 | 643 | |
c7e9668e | 644 | status = nfs_readdir_alloc_pages(pages, array_size); |
6650239a | 645 | if (status < 0) |
d1bacf9e BS |
646 | goto out_release_array; |
647 | do { | |
ac396128 | 648 | unsigned int pglen; |
56e4ebf8 | 649 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 650 | |
d1bacf9e | 651 | if (status < 0) |
00a92642 | 652 | break; |
ac396128 | 653 | pglen = status; |
6650239a | 654 | status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); |
8cd51a0c TM |
655 | if (status < 0) { |
656 | if (status == -ENOSPC) | |
657 | status = 0; | |
658 | break; | |
659 | } | |
660 | } while (array->eof_index < 0); | |
d1bacf9e | 661 | |
c7e9668e | 662 | nfs_readdir_free_pages(pages, array_size); |
d1bacf9e | 663 | out_release_array: |
0795bf83 | 664 | kunmap(page); |
14c43f76 | 665 | nfs4_label_free(entry.label); |
d1bacf9e BS |
666 | out: |
667 | nfs_free_fattr(entry.fattr); | |
668 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
669 | return status; |
670 | } | |
671 | ||
672 | /* | |
d1bacf9e BS |
673 | * Now we cache directories properly, by converting xdr information |
674 | * to an array that can be used for lookups later. This results in | |
675 | * fewer cache pages, since we can store more information on each page. | |
676 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 677 | */ |
d1bacf9e | 678 | static |
a46126cc | 679 | int nfs_readdir_filler(void *data, struct page* page) |
1da177e4 | 680 | { |
a46126cc | 681 | nfs_readdir_descriptor_t *desc = data; |
496ad9aa | 682 | struct inode *inode = file_inode(desc->file); |
8cd51a0c | 683 | int ret; |
1da177e4 | 684 | |
db531db9 MK |
685 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
686 | if (ret < 0) | |
687 | goto error; | |
d1bacf9e | 688 | SetPageUptodate(page); |
1da177e4 | 689 | |
d1bacf9e BS |
690 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
691 | /* Should never happen */ | |
692 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 693 | } |
d1bacf9e BS |
694 | unlock_page(page); |
695 | return 0; | |
696 | error: | |
4b310319 | 697 | nfs_readdir_clear_array(page); |
d1bacf9e | 698 | unlock_page(page); |
8cd51a0c | 699 | return ret; |
d1bacf9e | 700 | } |
1da177e4 | 701 | |
d1bacf9e BS |
702 | static |
703 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
704 | { | |
09cbfeaf | 705 | put_page(desc->page); |
d1bacf9e BS |
706 | desc->page = NULL; |
707 | } | |
708 | ||
709 | static | |
710 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
711 | { | |
a46126cc CH |
712 | return read_cache_page(desc->file->f_mapping, desc->page_index, |
713 | nfs_readdir_filler, desc); | |
1da177e4 LT |
714 | } |
715 | ||
716 | /* | |
d1bacf9e | 717 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
114de382 | 718 | * and locks the page to prevent removal from the page cache. |
1da177e4 | 719 | */ |
d1bacf9e | 720 | static |
114de382 | 721 | int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc) |
d1bacf9e BS |
722 | { |
723 | int res; | |
724 | ||
725 | desc->page = get_cache_page(desc); | |
726 | if (IS_ERR(desc->page)) | |
727 | return PTR_ERR(desc->page); | |
114de382 | 728 | res = lock_page_killable(desc->page); |
47c716cb | 729 | if (res != 0) |
114de382 TM |
730 | goto error; |
731 | res = -EAGAIN; | |
732 | if (desc->page->mapping != NULL) { | |
733 | res = nfs_readdir_search_array(desc); | |
734 | if (res == 0) | |
735 | return 0; | |
736 | } | |
737 | unlock_page(desc->page); | |
738 | error: | |
739 | cache_page_release(desc); | |
d1bacf9e BS |
740 | return res; |
741 | } | |
742 | ||
743 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
744 | static inline |
745 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
746 | { | |
8cd51a0c | 747 | int res; |
d1bacf9e | 748 | |
0aded708 | 749 | if (desc->page_index == 0) { |
8cd51a0c | 750 | desc->current_index = 0; |
0aded708 TM |
751 | desc->last_cookie = 0; |
752 | } | |
47c716cb | 753 | do { |
114de382 | 754 | res = find_and_lock_cache_page(desc); |
47c716cb | 755 | } while (res == -EAGAIN); |
1da177e4 LT |
756 | return res; |
757 | } | |
758 | ||
1da177e4 LT |
759 | /* |
760 | * Once we've found the start of the dirent within a page: fill 'er up... | |
761 | */ | |
762 | static | |
23db8620 | 763 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc) |
1da177e4 LT |
764 | { |
765 | struct file *file = desc->file; | |
d1bacf9e BS |
766 | int i = 0; |
767 | int res = 0; | |
768 | struct nfs_cache_array *array = NULL; | |
8ef2ce3e BS |
769 | struct nfs_open_dir_context *ctx = file->private_data; |
770 | ||
0795bf83 | 771 | array = kmap(desc->page); |
d1bacf9e | 772 | for (i = desc->cache_entry_index; i < array->size; i++) { |
ece0b423 | 773 | struct nfs_cache_array_entry *ent; |
1da177e4 | 774 | |
ece0b423 | 775 | ent = &array->array[i]; |
23db8620 AV |
776 | if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, |
777 | nfs_compat_user_ino64(ent->ino), ent->d_type)) { | |
6089dd0d | 778 | desc->eof = true; |
1da177e4 | 779 | break; |
ece0b423 | 780 | } |
23db8620 | 781 | desc->ctx->pos++; |
d1bacf9e BS |
782 | if (i < (array->size-1)) |
783 | *desc->dir_cookie = array->array[i+1].cookie; | |
784 | else | |
785 | *desc->dir_cookie = array->last_cookie; | |
0c030806 TM |
786 | if (ctx->duped != 0) |
787 | ctx->duped = 1; | |
1da177e4 | 788 | } |
47c716cb | 789 | if (array->eof_index >= 0) |
6089dd0d | 790 | desc->eof = true; |
d1bacf9e | 791 | |
0795bf83 | 792 | kunmap(desc->page); |
1e7cb3dc CL |
793 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
794 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
795 | return res; |
796 | } | |
797 | ||
798 | /* | |
799 | * If we cannot find a cookie in our cache, we suspect that this is | |
800 | * because it points to a deleted file, so we ask the server to return | |
801 | * whatever it thinks is the next entry. We then feed this to filldir. | |
802 | * If all goes well, we should then be able to find our way round the | |
803 | * cache on the next call to readdir_search_pagecache(); | |
804 | * | |
805 | * NOTE: we cannot add the anonymous page to the pagecache because | |
806 | * the data it contains might not be page aligned. Besides, | |
807 | * we should already have a complete representation of the | |
808 | * directory in the page cache by the time we get here. | |
809 | */ | |
810 | static inline | |
23db8620 | 811 | int uncached_readdir(nfs_readdir_descriptor_t *desc) |
1da177e4 | 812 | { |
1da177e4 LT |
813 | struct page *page = NULL; |
814 | int status; | |
496ad9aa | 815 | struct inode *inode = file_inode(desc->file); |
0c030806 | 816 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
1da177e4 | 817 | |
1e7cb3dc CL |
818 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
819 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
820 | |
821 | page = alloc_page(GFP_HIGHUSER); | |
822 | if (!page) { | |
823 | status = -ENOMEM; | |
824 | goto out; | |
825 | } | |
d1bacf9e | 826 | |
7a8e1dc3 | 827 | desc->page_index = 0; |
0aded708 | 828 | desc->last_cookie = *desc->dir_cookie; |
7a8e1dc3 | 829 | desc->page = page; |
0c030806 | 830 | ctx->duped = 0; |
7a8e1dc3 | 831 | |
85f8607e TM |
832 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
833 | if (status < 0) | |
1da177e4 LT |
834 | goto out_release; |
835 | ||
23db8620 | 836 | status = nfs_do_filldir(desc); |
1da177e4 | 837 | |
114de382 TM |
838 | out_release: |
839 | nfs_readdir_clear_array(desc->page); | |
840 | cache_page_release(desc); | |
1da177e4 | 841 | out: |
1e7cb3dc | 842 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 843 | __func__, status); |
1da177e4 | 844 | return status; |
1da177e4 LT |
845 | } |
846 | ||
00a92642 OG |
847 | /* The file offset position represents the dirent entry number. A |
848 | last cookie cache takes care of the common case of reading the | |
849 | whole directory. | |
1da177e4 | 850 | */ |
23db8620 | 851 | static int nfs_readdir(struct file *file, struct dir_context *ctx) |
1da177e4 | 852 | { |
be62a1a8 | 853 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 854 | struct inode *inode = d_inode(dentry); |
1da177e4 LT |
855 | nfs_readdir_descriptor_t my_desc, |
856 | *desc = &my_desc; | |
23db8620 | 857 | struct nfs_open_dir_context *dir_ctx = file->private_data; |
07b5ce8e | 858 | int res = 0; |
1da177e4 | 859 | |
6de1472f AV |
860 | dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", |
861 | file, (long long)ctx->pos); | |
91d5b470 CL |
862 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
863 | ||
1da177e4 | 864 | /* |
23db8620 | 865 | * ctx->pos points to the dirent entry number. |
f0dd2136 | 866 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
867 | * to either find the entry with the appropriate number or |
868 | * revalidate the cookie. | |
1da177e4 LT |
869 | */ |
870 | memset(desc, 0, sizeof(*desc)); | |
871 | ||
23db8620 AV |
872 | desc->file = file; |
873 | desc->ctx = ctx; | |
480c2006 | 874 | desc->dir_cookie = &dir_ctx->dir_cookie; |
1da177e4 | 875 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
a7a3b1e9 | 876 | desc->plus = nfs_use_readdirplus(inode, ctx); |
1da177e4 | 877 | |
79f687a3 | 878 | if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) |
07b5ce8e | 879 | res = nfs_revalidate_mapping(inode, file->f_mapping); |
fccca7fc TM |
880 | if (res < 0) |
881 | goto out; | |
882 | ||
47c716cb | 883 | do { |
1da177e4 | 884 | res = readdir_search_pagecache(desc); |
00a92642 | 885 | |
1da177e4 | 886 | if (res == -EBADCOOKIE) { |
ece0b423 | 887 | res = 0; |
1da177e4 | 888 | /* This means either end of directory */ |
6089dd0d | 889 | if (*desc->dir_cookie && !desc->eof) { |
1da177e4 | 890 | /* Or that the server has 'lost' a cookie */ |
23db8620 | 891 | res = uncached_readdir(desc); |
ece0b423 | 892 | if (res == 0) |
1da177e4 LT |
893 | continue; |
894 | } | |
1da177e4 LT |
895 | break; |
896 | } | |
897 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 898 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 899 | nfs_zap_caches(inode); |
baf57a09 | 900 | desc->page_index = 0; |
a7a3b1e9 BC |
901 | desc->plus = false; |
902 | desc->eof = false; | |
1da177e4 LT |
903 | continue; |
904 | } | |
905 | if (res < 0) | |
906 | break; | |
907 | ||
23db8620 | 908 | res = nfs_do_filldir(desc); |
114de382 TM |
909 | unlock_page(desc->page); |
910 | cache_page_release(desc); | |
ece0b423 | 911 | if (res < 0) |
1da177e4 | 912 | break; |
47c716cb | 913 | } while (!desc->eof); |
fccca7fc | 914 | out: |
1e7cb3dc CL |
915 | if (res > 0) |
916 | res = 0; | |
6de1472f | 917 | dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); |
1e7cb3dc | 918 | return res; |
1da177e4 LT |
919 | } |
920 | ||
965c8e59 | 921 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) |
f0dd2136 | 922 | { |
b044f645 | 923 | struct inode *inode = file_inode(filp); |
480c2006 | 924 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
b84e06c5 | 925 | |
6de1472f AV |
926 | dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", |
927 | filp, offset, whence); | |
b84e06c5 | 928 | |
965c8e59 | 929 | switch (whence) { |
b2b1ff3d TM |
930 | default: |
931 | return -EINVAL; | |
932 | case SEEK_SET: | |
933 | if (offset < 0) | |
934 | return -EINVAL; | |
935 | inode_lock(inode); | |
936 | break; | |
937 | case SEEK_CUR: | |
938 | if (offset == 0) | |
939 | return filp->f_pos; | |
940 | inode_lock(inode); | |
941 | offset += filp->f_pos; | |
942 | if (offset < 0) { | |
943 | inode_unlock(inode); | |
944 | return -EINVAL; | |
945 | } | |
f0dd2136 TM |
946 | } |
947 | if (offset != filp->f_pos) { | |
948 | filp->f_pos = offset; | |
480c2006 | 949 | dir_ctx->dir_cookie = 0; |
8ef2ce3e | 950 | dir_ctx->duped = 0; |
f0dd2136 | 951 | } |
b044f645 | 952 | inode_unlock(inode); |
f0dd2136 TM |
953 | return offset; |
954 | } | |
955 | ||
1da177e4 LT |
956 | /* |
957 | * All directory operations under NFS are synchronous, so fsync() | |
958 | * is a dummy operation. | |
959 | */ | |
02c24a82 JB |
960 | static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, |
961 | int datasync) | |
1da177e4 | 962 | { |
6de1472f | 963 | struct inode *inode = file_inode(filp); |
7ea80859 | 964 | |
6de1472f | 965 | dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); |
1e7cb3dc | 966 | |
5955102c | 967 | inode_lock(inode); |
6de1472f | 968 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
5955102c | 969 | inode_unlock(inode); |
1da177e4 LT |
970 | return 0; |
971 | } | |
972 | ||
bfc69a45 TM |
973 | /** |
974 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
302fad7b | 975 | * @dir: pointer to directory inode |
bfc69a45 TM |
976 | * |
977 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
978 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
979 | * on the server that might have invalidated our dcache. | |
980 | * | |
981 | * The caller should be holding dir->i_lock | |
982 | */ | |
983 | void nfs_force_lookup_revalidate(struct inode *dir) | |
984 | { | |
011935a0 | 985 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 | 986 | } |
89d77c8f | 987 | EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); |
bfc69a45 | 988 | |
1da177e4 LT |
989 | /* |
990 | * A check for whether or not the parent directory has changed. | |
991 | * In the case it has, we assume that the dentries are untrustworthy | |
992 | * and may need to be looked up again. | |
912a108d | 993 | * If rcu_walk prevents us from performing a full check, return 0. |
1da177e4 | 994 | */ |
912a108d N |
995 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, |
996 | int rcu_walk) | |
1da177e4 LT |
997 | { |
998 | if (IS_ROOT(dentry)) | |
999 | return 1; | |
4eec952e TM |
1000 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
1001 | return 0; | |
f2c77f4e TM |
1002 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
1003 | return 0; | |
1004 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
1cd9cb05 TM |
1005 | if (nfs_mapping_need_revalidate_inode(dir)) { |
1006 | if (rcu_walk) | |
1007 | return 0; | |
1008 | if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
1009 | return 0; | |
1010 | } | |
f2c77f4e TM |
1011 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
1012 | return 0; | |
1013 | return 1; | |
1da177e4 LT |
1014 | } |
1015 | ||
a12802ca TM |
1016 | /* |
1017 | * Use intent information to check whether or not we're going to do | |
1018 | * an O_EXCL create using this path component. | |
1019 | */ | |
fa3c56bb | 1020 | static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) |
a12802ca TM |
1021 | { |
1022 | if (NFS_PROTO(dir)->version == 2) | |
1023 | return 0; | |
fa3c56bb | 1024 | return flags & LOOKUP_EXCL; |
a12802ca TM |
1025 | } |
1026 | ||
1d6757fb TM |
1027 | /* |
1028 | * Inode and filehandle revalidation for lookups. | |
1029 | * | |
1030 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
1031 | * or if the intent information indicates that we're about to open this | |
1032 | * particular file and the "nocto" mount flag is not set. | |
1033 | * | |
1034 | */ | |
65a0c149 | 1035 | static |
fa3c56bb | 1036 | int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) |
1da177e4 LT |
1037 | { |
1038 | struct nfs_server *server = NFS_SERVER(inode); | |
65a0c149 | 1039 | int ret; |
1da177e4 | 1040 | |
36d43a43 | 1041 | if (IS_AUTOMOUNT(inode)) |
4e99a1ff | 1042 | return 0; |
47921921 TM |
1043 | |
1044 | if (flags & LOOKUP_OPEN) { | |
1045 | switch (inode->i_mode & S_IFMT) { | |
1046 | case S_IFREG: | |
1047 | /* A NFSv4 OPEN will revalidate later */ | |
1048 | if (server->caps & NFS_CAP_ATOMIC_OPEN) | |
1049 | goto out; | |
1050 | /* Fallthrough */ | |
1051 | case S_IFDIR: | |
1052 | if (server->flags & NFS_MOUNT_NOCTO) | |
1053 | break; | |
1054 | /* NFS close-to-open cache consistency validation */ | |
1055 | goto out_force; | |
1056 | } | |
1057 | } | |
1058 | ||
facc3530 | 1059 | /* VFS wants an on-the-wire revalidation */ |
fa3c56bb | 1060 | if (flags & LOOKUP_REVAL) |
facc3530 | 1061 | goto out_force; |
65a0c149 | 1062 | out: |
a61246c9 | 1063 | return (inode->i_nlink == 0) ? -ESTALE : 0; |
1da177e4 | 1064 | out_force: |
1fa1e384 N |
1065 | if (flags & LOOKUP_RCU) |
1066 | return -ECHILD; | |
65a0c149 TM |
1067 | ret = __nfs_revalidate_inode(server, inode); |
1068 | if (ret != 0) | |
1069 | return ret; | |
1070 | goto out; | |
1da177e4 LT |
1071 | } |
1072 | ||
1073 | /* | |
1074 | * We judge how long we want to trust negative | |
1075 | * dentries by looking at the parent inode mtime. | |
1076 | * | |
1077 | * If parent mtime has changed, we revalidate, else we wait for a | |
1078 | * period corresponding to the parent's attribute cache timeout value. | |
912a108d N |
1079 | * |
1080 | * If LOOKUP_RCU prevents us from performing a full check, return 1 | |
1081 | * suggesting a reval is needed. | |
9f6d44d4 TM |
1082 | * |
1083 | * Note that when creating a new file, or looking up a rename target, | |
1084 | * then it shouldn't be necessary to revalidate a negative dentry. | |
1da177e4 LT |
1085 | */ |
1086 | static inline | |
1087 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
fa3c56bb | 1088 | unsigned int flags) |
1da177e4 | 1089 | { |
9f6d44d4 | 1090 | if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) |
1da177e4 | 1091 | return 0; |
4eec952e TM |
1092 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1093 | return 1; | |
912a108d | 1094 | return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); |
1da177e4 LT |
1095 | } |
1096 | ||
5ceb9d7f TM |
1097 | static int |
1098 | nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry, | |
1099 | struct inode *inode, int error) | |
1100 | { | |
1101 | switch (error) { | |
1102 | case 1: | |
1103 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", | |
1104 | __func__, dentry); | |
1105 | return 1; | |
1106 | case 0: | |
1107 | nfs_mark_for_revalidate(dir); | |
1108 | if (inode && S_ISDIR(inode->i_mode)) { | |
1109 | /* Purge readdir caches. */ | |
1110 | nfs_zap_caches(inode); | |
1111 | /* | |
1112 | * We can't d_drop the root of a disconnected tree: | |
1113 | * its d_hash is on the s_anon list and d_drop() would hide | |
1114 | * it from shrink_dcache_for_unmount(), leading to busy | |
1115 | * inodes on unmount and further oopses. | |
1116 | */ | |
1117 | if (IS_ROOT(dentry)) | |
1118 | return 1; | |
1119 | } | |
1120 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", | |
1121 | __func__, dentry); | |
1122 | return 0; | |
1123 | } | |
1124 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", | |
1125 | __func__, dentry, error); | |
1126 | return error; | |
1127 | } | |
1128 | ||
1129 | static int | |
1130 | nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry, | |
1131 | unsigned int flags) | |
1132 | { | |
1133 | int ret = 1; | |
1134 | if (nfs_neg_need_reval(dir, dentry, flags)) { | |
1135 | if (flags & LOOKUP_RCU) | |
1136 | return -ECHILD; | |
1137 | ret = 0; | |
1138 | } | |
1139 | return nfs_lookup_revalidate_done(dir, dentry, NULL, ret); | |
1140 | } | |
1141 | ||
1142 | static int | |
1143 | nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry, | |
1144 | struct inode *inode) | |
1145 | { | |
1146 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1147 | return nfs_lookup_revalidate_done(dir, dentry, inode, 1); | |
1148 | } | |
1149 | ||
1150 | static int | |
1151 | nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry, | |
1152 | struct inode *inode) | |
1153 | { | |
1154 | struct nfs_fh *fhandle; | |
1155 | struct nfs_fattr *fattr; | |
1156 | struct nfs4_label *label; | |
1157 | int ret; | |
1158 | ||
1159 | ret = -ENOMEM; | |
1160 | fhandle = nfs_alloc_fhandle(); | |
1161 | fattr = nfs_alloc_fattr(); | |
1162 | label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); | |
1163 | if (fhandle == NULL || fattr == NULL || IS_ERR(label)) | |
1164 | goto out; | |
1165 | ||
f7b37b8b | 1166 | ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); |
5ceb9d7f | 1167 | if (ret < 0) { |
f7b37b8b TM |
1168 | switch (ret) { |
1169 | case -ESTALE: | |
1170 | case -ENOENT: | |
5ceb9d7f | 1171 | ret = 0; |
f7b37b8b TM |
1172 | break; |
1173 | case -ETIMEDOUT: | |
1174 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) | |
1175 | ret = 1; | |
1176 | } | |
5ceb9d7f TM |
1177 | goto out; |
1178 | } | |
1179 | ret = 0; | |
1180 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) | |
1181 | goto out; | |
1182 | if (nfs_refresh_inode(inode, fattr) < 0) | |
1183 | goto out; | |
1184 | ||
1185 | nfs_setsecurity(inode, fattr, label); | |
1186 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1187 | ||
1188 | /* set a readdirplus hint that we had a cache miss */ | |
1189 | nfs_force_use_readdirplus(dir); | |
1190 | ret = 1; | |
1191 | out: | |
1192 | nfs_free_fattr(fattr); | |
1193 | nfs_free_fhandle(fhandle); | |
1194 | nfs4_label_free(label); | |
1195 | return nfs_lookup_revalidate_done(dir, dentry, inode, ret); | |
1196 | } | |
1197 | ||
1da177e4 LT |
1198 | /* |
1199 | * This is called every time the dcache has a lookup hit, | |
1200 | * and we should check whether we can really trust that | |
1201 | * lookup. | |
1202 | * | |
1203 | * NOTE! The hit can be a negative hit too, don't assume | |
1204 | * we have an inode! | |
1205 | * | |
1206 | * If the parent directory is seen to have changed, we throw out the | |
1207 | * cached dentry and do a new lookup. | |
1208 | */ | |
5ceb9d7f TM |
1209 | static int |
1210 | nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, | |
1211 | unsigned int flags) | |
1da177e4 | 1212 | { |
1da177e4 | 1213 | struct inode *inode; |
1da177e4 | 1214 | int error; |
1da177e4 | 1215 | |
91d5b470 | 1216 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
2b0143b5 | 1217 | inode = d_inode(dentry); |
1da177e4 | 1218 | |
5ceb9d7f TM |
1219 | if (!inode) |
1220 | return nfs_lookup_revalidate_negative(dir, dentry, flags); | |
1da177e4 LT |
1221 | |
1222 | if (is_bad_inode(inode)) { | |
6de1472f AV |
1223 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1224 | __func__, dentry); | |
1da177e4 LT |
1225 | goto out_bad; |
1226 | } | |
1227 | ||
011e2a7f | 1228 | if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) |
5ceb9d7f | 1229 | return nfs_lookup_revalidate_delegated(dir, dentry, inode); |
15860ab1 | 1230 | |
1da177e4 | 1231 | /* Force a full look up iff the parent directory has changed */ |
73dd684a | 1232 | if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && |
912a108d | 1233 | nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { |
cc89684c N |
1234 | error = nfs_lookup_verify_inode(inode, flags); |
1235 | if (error) { | |
cc89684c | 1236 | if (error == -ESTALE) |
5ceb9d7f TM |
1237 | nfs_zap_caches(dir); |
1238 | goto out_bad; | |
1fa1e384 | 1239 | } |
63519fbc | 1240 | nfs_advise_use_readdirplus(dir); |
1da177e4 LT |
1241 | goto out_valid; |
1242 | } | |
1243 | ||
912a108d N |
1244 | if (flags & LOOKUP_RCU) |
1245 | return -ECHILD; | |
1246 | ||
1da177e4 LT |
1247 | if (NFS_STALE(inode)) |
1248 | goto out_bad; | |
1249 | ||
6e0d0be7 | 1250 | trace_nfs_lookup_revalidate_enter(dir, dentry, flags); |
5ceb9d7f | 1251 | error = nfs_lookup_revalidate_dentry(dir, dentry, inode); |
6e0d0be7 | 1252 | trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); |
5ceb9d7f TM |
1253 | return error; |
1254 | out_valid: | |
1255 | return nfs_lookup_revalidate_done(dir, dentry, inode, 1); | |
1256 | out_bad: | |
1257 | if (flags & LOOKUP_RCU) | |
1258 | return -ECHILD; | |
1259 | return nfs_lookup_revalidate_done(dir, dentry, inode, 0); | |
1260 | } | |
14c43f76 | 1261 | |
5ceb9d7f | 1262 | static int |
c7944ebb TM |
1263 | __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, |
1264 | int (*reval)(struct inode *, struct dentry *, unsigned int)) | |
5ceb9d7f TM |
1265 | { |
1266 | struct dentry *parent; | |
1267 | struct inode *dir; | |
1268 | int ret; | |
63519fbc | 1269 | |
d51ac1a8 | 1270 | if (flags & LOOKUP_RCU) { |
5ceb9d7f TM |
1271 | parent = READ_ONCE(dentry->d_parent); |
1272 | dir = d_inode_rcu(parent); | |
1273 | if (!dir) | |
1274 | return -ECHILD; | |
c7944ebb | 1275 | ret = reval(dir, dentry, flags); |
6aa7de05 | 1276 | if (parent != READ_ONCE(dentry->d_parent)) |
d51ac1a8 | 1277 | return -ECHILD; |
5ceb9d7f TM |
1278 | } else { |
1279 | parent = dget_parent(dentry); | |
c7944ebb | 1280 | ret = reval(d_inode(parent), dentry, flags); |
d51ac1a8 | 1281 | dput(parent); |
1da177e4 | 1282 | } |
5ceb9d7f | 1283 | return ret; |
1da177e4 LT |
1284 | } |
1285 | ||
c7944ebb TM |
1286 | static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
1287 | { | |
1288 | return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); | |
1289 | } | |
1290 | ||
ecf3d1f1 | 1291 | /* |
2b0143b5 | 1292 | * A weaker form of d_revalidate for revalidating just the d_inode(dentry) |
ecf3d1f1 JL |
1293 | * when we don't really care about the dentry name. This is called when a |
1294 | * pathwalk ends on a dentry that was not found via a normal lookup in the | |
1295 | * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). | |
1296 | * | |
1297 | * In this situation, we just want to verify that the inode itself is OK | |
1298 | * since the dentry might have changed on the server. | |
1299 | */ | |
1300 | static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) | |
1301 | { | |
2b0143b5 | 1302 | struct inode *inode = d_inode(dentry); |
9cdd1d3f | 1303 | int error = 0; |
ecf3d1f1 JL |
1304 | |
1305 | /* | |
1306 | * I believe we can only get a negative dentry here in the case of a | |
1307 | * procfs-style symlink. Just assume it's correct for now, but we may | |
1308 | * eventually need to do something more here. | |
1309 | */ | |
1310 | if (!inode) { | |
6de1472f AV |
1311 | dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", |
1312 | __func__, dentry); | |
ecf3d1f1 JL |
1313 | return 1; |
1314 | } | |
1315 | ||
1316 | if (is_bad_inode(inode)) { | |
6de1472f AV |
1317 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1318 | __func__, dentry); | |
ecf3d1f1 JL |
1319 | return 0; |
1320 | } | |
1321 | ||
b688741c | 1322 | error = nfs_lookup_verify_inode(inode, flags); |
ecf3d1f1 JL |
1323 | dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", |
1324 | __func__, inode->i_ino, error ? "invalid" : "valid"); | |
1325 | return !error; | |
1326 | } | |
1327 | ||
1da177e4 LT |
1328 | /* |
1329 | * This is called from dput() when d_count is going to 0. | |
1330 | */ | |
fe15ce44 | 1331 | static int nfs_dentry_delete(const struct dentry *dentry) |
1da177e4 | 1332 | { |
6de1472f AV |
1333 | dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", |
1334 | dentry, dentry->d_flags); | |
1da177e4 | 1335 | |
77f11192 | 1336 | /* Unhash any dentry with a stale inode */ |
2b0143b5 | 1337 | if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) |
77f11192 TM |
1338 | return 1; |
1339 | ||
1da177e4 LT |
1340 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1341 | /* Unhash it, so that ->d_iput() would be called */ | |
1342 | return 1; | |
1343 | } | |
1751e8a6 | 1344 | if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { |
1da177e4 LT |
1345 | /* Unhash it, so that ancestors of killed async unlink |
1346 | * files will be cleaned up during umount */ | |
1347 | return 1; | |
1348 | } | |
1349 | return 0; | |
1350 | ||
1351 | } | |
1352 | ||
1f018458 | 1353 | /* Ensure that we revalidate inode->i_nlink */ |
1b83d707 TM |
1354 | static void nfs_drop_nlink(struct inode *inode) |
1355 | { | |
1356 | spin_lock(&inode->i_lock); | |
1f018458 | 1357 | /* drop the inode if we're reasonably sure this is the last link */ |
59a707b0 TM |
1358 | if (inode->i_nlink > 0) |
1359 | drop_nlink(inode); | |
1360 | NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); | |
16e14375 TM |
1361 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE |
1362 | | NFS_INO_INVALID_CTIME | |
59a707b0 TM |
1363 | | NFS_INO_INVALID_OTHER |
1364 | | NFS_INO_REVAL_FORCED; | |
1b83d707 TM |
1365 | spin_unlock(&inode->i_lock); |
1366 | } | |
1367 | ||
1da177e4 LT |
1368 | /* |
1369 | * Called when the dentry loses inode. | |
1370 | * We use it to clean up silly-renamed files. | |
1371 | */ | |
1372 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1373 | { | |
83672d39 NB |
1374 | if (S_ISDIR(inode->i_mode)) |
1375 | /* drop any readdir cache as it could easily be old */ | |
1376 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1377 | ||
1da177e4 | 1378 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
e4eff1a6 | 1379 | nfs_complete_unlink(dentry, inode); |
1f018458 | 1380 | nfs_drop_nlink(inode); |
1da177e4 | 1381 | } |
1da177e4 LT |
1382 | iput(inode); |
1383 | } | |
1384 | ||
b1942c5f AV |
1385 | static void nfs_d_release(struct dentry *dentry) |
1386 | { | |
1387 | /* free cached devname value, if it survived that far */ | |
1388 | if (unlikely(dentry->d_fsdata)) { | |
1389 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1390 | WARN_ON(1); | |
1391 | else | |
1392 | kfree(dentry->d_fsdata); | |
1393 | } | |
1394 | } | |
1395 | ||
f786aa90 | 1396 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 | 1397 | .d_revalidate = nfs_lookup_revalidate, |
ecf3d1f1 | 1398 | .d_weak_revalidate = nfs_weak_revalidate, |
1da177e4 LT |
1399 | .d_delete = nfs_dentry_delete, |
1400 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1401 | .d_automount = nfs_d_automount, |
b1942c5f | 1402 | .d_release = nfs_d_release, |
1da177e4 | 1403 | }; |
ddda8e0a | 1404 | EXPORT_SYMBOL_GPL(nfs_dentry_operations); |
1da177e4 | 1405 | |
597d9289 | 1406 | struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) |
1da177e4 LT |
1407 | { |
1408 | struct dentry *res; | |
1409 | struct inode *inode = NULL; | |
e1fb4d05 TM |
1410 | struct nfs_fh *fhandle = NULL; |
1411 | struct nfs_fattr *fattr = NULL; | |
1775fd3e | 1412 | struct nfs4_label *label = NULL; |
1da177e4 | 1413 | int error; |
1da177e4 | 1414 | |
6de1472f | 1415 | dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); |
91d5b470 | 1416 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 | 1417 | |
130f9ab7 AV |
1418 | if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) |
1419 | return ERR_PTR(-ENAMETOOLONG); | |
1da177e4 | 1420 | |
fd684071 TM |
1421 | /* |
1422 | * If we're doing an exclusive create, optimize away the lookup | |
1423 | * but don't hash the dentry. | |
1424 | */ | |
9f6d44d4 | 1425 | if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) |
130f9ab7 | 1426 | return NULL; |
1da177e4 | 1427 | |
e1fb4d05 TM |
1428 | res = ERR_PTR(-ENOMEM); |
1429 | fhandle = nfs_alloc_fhandle(); | |
1430 | fattr = nfs_alloc_fattr(); | |
1431 | if (fhandle == NULL || fattr == NULL) | |
1432 | goto out; | |
1433 | ||
14c43f76 DQ |
1434 | label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); |
1435 | if (IS_ERR(label)) | |
1436 | goto out; | |
1437 | ||
6e0d0be7 | 1438 | trace_nfs_lookup_enter(dir, dentry, flags); |
f7b37b8b | 1439 | error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); |
1da177e4 LT |
1440 | if (error == -ENOENT) |
1441 | goto no_entry; | |
1442 | if (error < 0) { | |
1443 | res = ERR_PTR(error); | |
bf130914 | 1444 | goto out_label; |
1da177e4 | 1445 | } |
1775fd3e | 1446 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
bf0c84f1 | 1447 | res = ERR_CAST(inode); |
03f28e3a | 1448 | if (IS_ERR(res)) |
bf130914 | 1449 | goto out_label; |
54ceac45 | 1450 | |
63519fbc TM |
1451 | /* Notify readdir to use READDIRPLUS */ |
1452 | nfs_force_use_readdirplus(dir); | |
d69ee9b8 | 1453 | |
1da177e4 | 1454 | no_entry: |
41d28bca | 1455 | res = d_splice_alias(inode, dentry); |
9eaef27b TM |
1456 | if (res != NULL) { |
1457 | if (IS_ERR(res)) | |
bf130914 | 1458 | goto out_label; |
1da177e4 | 1459 | dentry = res; |
9eaef27b | 1460 | } |
1da177e4 | 1461 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
bf130914 | 1462 | out_label: |
6e0d0be7 | 1463 | trace_nfs_lookup_exit(dir, dentry, flags, error); |
14c43f76 | 1464 | nfs4_label_free(label); |
1da177e4 | 1465 | out: |
e1fb4d05 TM |
1466 | nfs_free_fattr(fattr); |
1467 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1468 | return res; |
1469 | } | |
ddda8e0a | 1470 | EXPORT_SYMBOL_GPL(nfs_lookup); |
1da177e4 | 1471 | |
89d77c8f | 1472 | #if IS_ENABLED(CONFIG_NFS_V4) |
0b728e19 | 1473 | static int nfs4_lookup_revalidate(struct dentry *, unsigned int); |
1da177e4 | 1474 | |
f786aa90 | 1475 | const struct dentry_operations nfs4_dentry_operations = { |
0ef97dcf | 1476 | .d_revalidate = nfs4_lookup_revalidate, |
b688741c | 1477 | .d_weak_revalidate = nfs_weak_revalidate, |
1da177e4 LT |
1478 | .d_delete = nfs_dentry_delete, |
1479 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1480 | .d_automount = nfs_d_automount, |
b1942c5f | 1481 | .d_release = nfs_d_release, |
1da177e4 | 1482 | }; |
89d77c8f | 1483 | EXPORT_SYMBOL_GPL(nfs4_dentry_operations); |
1da177e4 | 1484 | |
8a5e929d AV |
1485 | static fmode_t flags_to_mode(int flags) |
1486 | { | |
1487 | fmode_t res = (__force fmode_t)flags & FMODE_EXEC; | |
1488 | if ((flags & O_ACCMODE) != O_WRONLY) | |
1489 | res |= FMODE_READ; | |
1490 | if ((flags & O_ACCMODE) != O_RDONLY) | |
1491 | res |= FMODE_WRITE; | |
1492 | return res; | |
1493 | } | |
1494 | ||
532d4def | 1495 | static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) |
cd9a1c0e | 1496 | { |
532d4def | 1497 | return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); |
cd9a1c0e TM |
1498 | } |
1499 | ||
1500 | static int do_open(struct inode *inode, struct file *filp) | |
1501 | { | |
f1fe29b4 | 1502 | nfs_fscache_open_file(inode, filp); |
cd9a1c0e TM |
1503 | return 0; |
1504 | } | |
1505 | ||
d9585277 AV |
1506 | static int nfs_finish_open(struct nfs_open_context *ctx, |
1507 | struct dentry *dentry, | |
b452a458 | 1508 | struct file *file, unsigned open_flags) |
cd9a1c0e | 1509 | { |
0dd2b474 MS |
1510 | int err; |
1511 | ||
be12af3e | 1512 | err = finish_open(file, dentry, do_open); |
30d90494 | 1513 | if (err) |
d9585277 | 1514 | goto out; |
eaa2b82c N |
1515 | if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) |
1516 | nfs_file_set_open_context(file, ctx); | |
1517 | else | |
9821421a | 1518 | err = -EOPENSTALE; |
cd9a1c0e | 1519 | out: |
d9585277 | 1520 | return err; |
cd9a1c0e TM |
1521 | } |
1522 | ||
73a79706 BS |
1523 | int nfs_atomic_open(struct inode *dir, struct dentry *dentry, |
1524 | struct file *file, unsigned open_flags, | |
44907d79 | 1525 | umode_t mode) |
1da177e4 | 1526 | { |
c94c0953 | 1527 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
cd9a1c0e | 1528 | struct nfs_open_context *ctx; |
0dd2b474 MS |
1529 | struct dentry *res; |
1530 | struct iattr attr = { .ia_valid = ATTR_OPEN }; | |
f46e0bd3 | 1531 | struct inode *inode; |
1472b83e | 1532 | unsigned int lookup_flags = 0; |
c94c0953 | 1533 | bool switched = false; |
73a09dd9 | 1534 | int created = 0; |
898f635c | 1535 | int err; |
1da177e4 | 1536 | |
0dd2b474 | 1537 | /* Expect a negative dentry */ |
2b0143b5 | 1538 | BUG_ON(d_inode(dentry)); |
0dd2b474 | 1539 | |
1e8968c5 | 1540 | dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", |
6de1472f | 1541 | dir->i_sb->s_id, dir->i_ino, dentry); |
1e7cb3dc | 1542 | |
9597c13b JL |
1543 | err = nfs_check_flags(open_flags); |
1544 | if (err) | |
1545 | return err; | |
1546 | ||
0dd2b474 MS |
1547 | /* NFS only supports OPEN on regular files */ |
1548 | if ((open_flags & O_DIRECTORY)) { | |
00699ad8 | 1549 | if (!d_in_lookup(dentry)) { |
0dd2b474 MS |
1550 | /* |
1551 | * Hashed negative dentry with O_DIRECTORY: dentry was | |
1552 | * revalidated and is fine, no need to perform lookup | |
1553 | * again | |
1554 | */ | |
d9585277 | 1555 | return -ENOENT; |
0dd2b474 | 1556 | } |
1472b83e | 1557 | lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; |
1da177e4 | 1558 | goto no_open; |
02a913a7 | 1559 | } |
1da177e4 | 1560 | |
0dd2b474 | 1561 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) |
d9585277 | 1562 | return -ENAMETOOLONG; |
cd9a1c0e | 1563 | |
0dd2b474 | 1564 | if (open_flags & O_CREAT) { |
dff25ddb AG |
1565 | struct nfs_server *server = NFS_SERVER(dir); |
1566 | ||
1567 | if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) | |
1568 | mode &= ~current_umask(); | |
1569 | ||
536e43d1 | 1570 | attr.ia_valid |= ATTR_MODE; |
dff25ddb | 1571 | attr.ia_mode = mode; |
0dd2b474 | 1572 | } |
536e43d1 TM |
1573 | if (open_flags & O_TRUNC) { |
1574 | attr.ia_valid |= ATTR_SIZE; | |
1575 | attr.ia_size = 0; | |
cd9a1c0e TM |
1576 | } |
1577 | ||
c94c0953 AV |
1578 | if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { |
1579 | d_drop(dentry); | |
1580 | switched = true; | |
1581 | dentry = d_alloc_parallel(dentry->d_parent, | |
1582 | &dentry->d_name, &wq); | |
1583 | if (IS_ERR(dentry)) | |
1584 | return PTR_ERR(dentry); | |
1585 | if (unlikely(!d_in_lookup(dentry))) | |
1586 | return finish_no_open(file, dentry); | |
1587 | } | |
1588 | ||
532d4def | 1589 | ctx = create_nfs_open_context(dentry, open_flags, file); |
0dd2b474 MS |
1590 | err = PTR_ERR(ctx); |
1591 | if (IS_ERR(ctx)) | |
d9585277 | 1592 | goto out; |
0dd2b474 | 1593 | |
6e0d0be7 | 1594 | trace_nfs_atomic_open_enter(dir, ctx, open_flags); |
73a09dd9 AV |
1595 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); |
1596 | if (created) | |
1597 | file->f_mode |= FMODE_CREATED; | |
f46e0bd3 | 1598 | if (IS_ERR(inode)) { |
0dd2b474 | 1599 | err = PTR_ERR(inode); |
6e0d0be7 | 1600 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 1601 | put_nfs_open_context(ctx); |
d20cb71d | 1602 | d_drop(dentry); |
0dd2b474 MS |
1603 | switch (err) { |
1604 | case -ENOENT: | |
774d9513 | 1605 | d_splice_alias(NULL, dentry); |
809fd143 | 1606 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
0dd2b474 MS |
1607 | break; |
1608 | case -EISDIR: | |
1609 | case -ENOTDIR: | |
1610 | goto no_open; | |
1611 | case -ELOOP: | |
1612 | if (!(open_flags & O_NOFOLLOW)) | |
6f926b5b | 1613 | goto no_open; |
0dd2b474 | 1614 | break; |
1da177e4 | 1615 | /* case -EINVAL: */ |
0dd2b474 MS |
1616 | default: |
1617 | break; | |
1da177e4 | 1618 | } |
d9585277 | 1619 | goto out; |
cd9a1c0e | 1620 | } |
0dd2b474 | 1621 | |
b452a458 | 1622 | err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); |
6e0d0be7 | 1623 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 1624 | put_nfs_open_context(ctx); |
d9585277 | 1625 | out: |
c94c0953 AV |
1626 | if (unlikely(switched)) { |
1627 | d_lookup_done(dentry); | |
1628 | dput(dentry); | |
1629 | } | |
d9585277 | 1630 | return err; |
0dd2b474 | 1631 | |
1da177e4 | 1632 | no_open: |
1472b83e | 1633 | res = nfs_lookup(dir, dentry, lookup_flags); |
c94c0953 AV |
1634 | if (switched) { |
1635 | d_lookup_done(dentry); | |
1636 | if (!res) | |
1637 | res = dentry; | |
1638 | else | |
1639 | dput(dentry); | |
1640 | } | |
0dd2b474 | 1641 | if (IS_ERR(res)) |
c94c0953 | 1642 | return PTR_ERR(res); |
e45198a6 | 1643 | return finish_no_open(file, res); |
1da177e4 | 1644 | } |
89d77c8f | 1645 | EXPORT_SYMBOL_GPL(nfs_atomic_open); |
1da177e4 | 1646 | |
c7944ebb TM |
1647 | static int |
1648 | nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, | |
1649 | unsigned int flags) | |
1da177e4 | 1650 | { |
657e94b6 | 1651 | struct inode *inode; |
1da177e4 | 1652 | |
fa3c56bb | 1653 | if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) |
c7944ebb | 1654 | goto full_reval; |
eda72afb | 1655 | if (d_mountpoint(dentry)) |
c7944ebb | 1656 | goto full_reval; |
2b484297 | 1657 | |
2b0143b5 | 1658 | inode = d_inode(dentry); |
2b484297 | 1659 | |
1da177e4 LT |
1660 | /* We can't create new files in nfs_open_revalidate(), so we |
1661 | * optimize away revalidation of negative dentries. | |
1662 | */ | |
c7944ebb TM |
1663 | if (inode == NULL) |
1664 | goto full_reval; | |
1665 | ||
1666 | if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) | |
1667 | return nfs_lookup_revalidate_delegated(dir, dentry, inode); | |
216d5d06 | 1668 | |
1da177e4 LT |
1669 | /* NFS only supports OPEN on regular files */ |
1670 | if (!S_ISREG(inode->i_mode)) | |
c7944ebb TM |
1671 | goto full_reval; |
1672 | ||
1da177e4 | 1673 | /* We cannot do exclusive creation on a positive dentry */ |
c7944ebb TM |
1674 | if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) |
1675 | goto reval_dentry; | |
1676 | ||
1677 | /* Check if the directory changed */ | |
1678 | if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) | |
1679 | goto reval_dentry; | |
1da177e4 | 1680 | |
0ef97dcf | 1681 | /* Let f_op->open() actually open (and revalidate) the file */ |
c7944ebb TM |
1682 | return 1; |
1683 | reval_dentry: | |
1684 | if (flags & LOOKUP_RCU) | |
1685 | return -ECHILD; | |
42f72cf3 | 1686 | return nfs_lookup_revalidate_dentry(dir, dentry, inode); |
536e43d1 | 1687 | |
c7944ebb TM |
1688 | full_reval: |
1689 | return nfs_do_lookup_revalidate(dir, dentry, flags); | |
1690 | } | |
535918f1 | 1691 | |
c7944ebb TM |
1692 | static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
1693 | { | |
1694 | return __nfs_lookup_revalidate(dentry, flags, | |
1695 | nfs4_do_lookup_revalidate); | |
c0204fd2 TM |
1696 | } |
1697 | ||
1da177e4 LT |
1698 | #endif /* CONFIG_NFSV4 */ |
1699 | ||
406cd915 BC |
1700 | struct dentry * |
1701 | nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, | |
1775fd3e DQ |
1702 | struct nfs_fattr *fattr, |
1703 | struct nfs4_label *label) | |
1da177e4 | 1704 | { |
fab728e1 | 1705 | struct dentry *parent = dget_parent(dentry); |
2b0143b5 | 1706 | struct inode *dir = d_inode(parent); |
1da177e4 | 1707 | struct inode *inode; |
b0c6108e | 1708 | struct dentry *d; |
406cd915 | 1709 | int error; |
1da177e4 | 1710 | |
fab728e1 TM |
1711 | d_drop(dentry); |
1712 | ||
1da177e4 | 1713 | if (fhandle->size == 0) { |
f7b37b8b | 1714 | error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL); |
1da177e4 | 1715 | if (error) |
fab728e1 | 1716 | goto out_error; |
1da177e4 | 1717 | } |
5724ab37 | 1718 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1719 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1720 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
a841b54d TM |
1721 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, |
1722 | fattr, NULL, NULL); | |
1da177e4 | 1723 | if (error < 0) |
fab728e1 | 1724 | goto out_error; |
1da177e4 | 1725 | } |
1775fd3e | 1726 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
b0c6108e | 1727 | d = d_splice_alias(inode, dentry); |
fab728e1 TM |
1728 | out: |
1729 | dput(parent); | |
406cd915 | 1730 | return d; |
fab728e1 TM |
1731 | out_error: |
1732 | nfs_mark_for_revalidate(dir); | |
406cd915 BC |
1733 | d = ERR_PTR(error); |
1734 | goto out; | |
1735 | } | |
1736 | EXPORT_SYMBOL_GPL(nfs_add_or_obtain); | |
1737 | ||
1738 | /* | |
1739 | * Code common to create, mkdir, and mknod. | |
1740 | */ | |
1741 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1742 | struct nfs_fattr *fattr, | |
1743 | struct nfs4_label *label) | |
1744 | { | |
1745 | struct dentry *d; | |
1746 | ||
1747 | d = nfs_add_or_obtain(dentry, fhandle, fattr, label); | |
1748 | if (IS_ERR(d)) | |
1749 | return PTR_ERR(d); | |
1750 | ||
1751 | /* Callers don't care */ | |
1752 | dput(d); | |
1753 | return 0; | |
1da177e4 | 1754 | } |
ddda8e0a | 1755 | EXPORT_SYMBOL_GPL(nfs_instantiate); |
1da177e4 LT |
1756 | |
1757 | /* | |
1758 | * Following a failed create operation, we drop the dentry rather | |
1759 | * than retain a negative dentry. This avoids a problem in the event | |
1760 | * that the operation succeeded on the server, but an error in the | |
1761 | * reply path made it appear to have failed. | |
1762 | */ | |
597d9289 | 1763 | int nfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 1764 | umode_t mode, bool excl) |
1da177e4 LT |
1765 | { |
1766 | struct iattr attr; | |
ebfc3b49 | 1767 | int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; |
1da177e4 | 1768 | int error; |
1da177e4 | 1769 | |
1e8968c5 | 1770 | dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", |
6de1472f | 1771 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 LT |
1772 | |
1773 | attr.ia_mode = mode; | |
1774 | attr.ia_valid = ATTR_MODE; | |
1775 | ||
8b0ad3d4 | 1776 | trace_nfs_create_enter(dir, dentry, open_flags); |
8867fe58 | 1777 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); |
8b0ad3d4 | 1778 | trace_nfs_create_exit(dir, dentry, open_flags, error); |
1da177e4 LT |
1779 | if (error != 0) |
1780 | goto out_err; | |
1da177e4 LT |
1781 | return 0; |
1782 | out_err: | |
1da177e4 LT |
1783 | d_drop(dentry); |
1784 | return error; | |
1785 | } | |
ddda8e0a | 1786 | EXPORT_SYMBOL_GPL(nfs_create); |
1da177e4 LT |
1787 | |
1788 | /* | |
1789 | * See comments for nfs_proc_create regarding failed operations. | |
1790 | */ | |
597d9289 | 1791 | int |
1a67aafb | 1792 | nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) |
1da177e4 LT |
1793 | { |
1794 | struct iattr attr; | |
1795 | int status; | |
1796 | ||
1e8968c5 | 1797 | dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", |
6de1472f | 1798 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 | 1799 | |
1da177e4 LT |
1800 | attr.ia_mode = mode; |
1801 | attr.ia_valid = ATTR_MODE; | |
1802 | ||
1ca42382 | 1803 | trace_nfs_mknod_enter(dir, dentry); |
1da177e4 | 1804 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1ca42382 | 1805 | trace_nfs_mknod_exit(dir, dentry, status); |
1da177e4 LT |
1806 | if (status != 0) |
1807 | goto out_err; | |
1da177e4 LT |
1808 | return 0; |
1809 | out_err: | |
1da177e4 LT |
1810 | d_drop(dentry); |
1811 | return status; | |
1812 | } | |
ddda8e0a | 1813 | EXPORT_SYMBOL_GPL(nfs_mknod); |
1da177e4 LT |
1814 | |
1815 | /* | |
1816 | * See comments for nfs_proc_create regarding failed operations. | |
1817 | */ | |
597d9289 | 1818 | int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
1819 | { |
1820 | struct iattr attr; | |
1821 | int error; | |
1822 | ||
1e8968c5 | 1823 | dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", |
6de1472f | 1824 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 LT |
1825 | |
1826 | attr.ia_valid = ATTR_MODE; | |
1827 | attr.ia_mode = mode | S_IFDIR; | |
1828 | ||
1ca42382 | 1829 | trace_nfs_mkdir_enter(dir, dentry); |
1da177e4 | 1830 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1ca42382 | 1831 | trace_nfs_mkdir_exit(dir, dentry, error); |
1da177e4 LT |
1832 | if (error != 0) |
1833 | goto out_err; | |
1da177e4 LT |
1834 | return 0; |
1835 | out_err: | |
1836 | d_drop(dentry); | |
1da177e4 LT |
1837 | return error; |
1838 | } | |
ddda8e0a | 1839 | EXPORT_SYMBOL_GPL(nfs_mkdir); |
1da177e4 | 1840 | |
d45b9d8b TM |
1841 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1842 | { | |
dc3f4198 | 1843 | if (simple_positive(dentry)) |
d45b9d8b TM |
1844 | d_delete(dentry); |
1845 | } | |
1846 | ||
597d9289 | 1847 | int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
1848 | { |
1849 | int error; | |
1850 | ||
1e8968c5 | 1851 | dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", |
6de1472f | 1852 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 | 1853 | |
1ca42382 | 1854 | trace_nfs_rmdir_enter(dir, dentry); |
2b0143b5 | 1855 | if (d_really_is_positive(dentry)) { |
884be175 | 1856 | down_write(&NFS_I(d_inode(dentry))->rmdir_sem); |
ba6c0592 TM |
1857 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1858 | /* Ensure the VFS deletes this inode */ | |
1859 | switch (error) { | |
1860 | case 0: | |
2b0143b5 | 1861 | clear_nlink(d_inode(dentry)); |
ba6c0592 TM |
1862 | break; |
1863 | case -ENOENT: | |
1864 | nfs_dentry_handle_enoent(dentry); | |
1865 | } | |
884be175 | 1866 | up_write(&NFS_I(d_inode(dentry))->rmdir_sem); |
ba6c0592 TM |
1867 | } else |
1868 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1ca42382 | 1869 | trace_nfs_rmdir_exit(dir, dentry, error); |
1da177e4 LT |
1870 | |
1871 | return error; | |
1872 | } | |
ddda8e0a | 1873 | EXPORT_SYMBOL_GPL(nfs_rmdir); |
1da177e4 | 1874 | |
1da177e4 LT |
1875 | /* |
1876 | * Remove a file after making sure there are no pending writes, | |
1877 | * and after checking that the file has only one user. | |
1878 | * | |
1879 | * We invalidate the attribute cache and free the inode prior to the operation | |
1880 | * to avoid possible races if the server reuses the inode. | |
1881 | */ | |
1882 | static int nfs_safe_remove(struct dentry *dentry) | |
1883 | { | |
2b0143b5 DH |
1884 | struct inode *dir = d_inode(dentry->d_parent); |
1885 | struct inode *inode = d_inode(dentry); | |
1da177e4 LT |
1886 | int error = -EBUSY; |
1887 | ||
6de1472f | 1888 | dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); |
1da177e4 LT |
1889 | |
1890 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1891 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1892 | error = 0; | |
1893 | goto out; | |
1894 | } | |
1895 | ||
1ca42382 | 1896 | trace_nfs_remove_enter(dir, dentry); |
1da177e4 | 1897 | if (inode != NULL) { |
912678db | 1898 | error = NFS_PROTO(dir)->remove(dir, dentry); |
1da177e4 | 1899 | if (error == 0) |
1b83d707 | 1900 | nfs_drop_nlink(inode); |
1da177e4 | 1901 | } else |
912678db | 1902 | error = NFS_PROTO(dir)->remove(dir, dentry); |
d45b9d8b TM |
1903 | if (error == -ENOENT) |
1904 | nfs_dentry_handle_enoent(dentry); | |
1ca42382 | 1905 | trace_nfs_remove_exit(dir, dentry, error); |
1da177e4 LT |
1906 | out: |
1907 | return error; | |
1908 | } | |
1909 | ||
1910 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1911 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1912 | * | |
1913 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1914 | */ | |
597d9289 | 1915 | int nfs_unlink(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
1916 | { |
1917 | int error; | |
1918 | int need_rehash = 0; | |
1919 | ||
1e8968c5 | 1920 | dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, |
6de1472f | 1921 | dir->i_ino, dentry); |
1da177e4 | 1922 | |
1ca42382 | 1923 | trace_nfs_unlink_enter(dir, dentry); |
1da177e4 | 1924 | spin_lock(&dentry->d_lock); |
84d08fa8 | 1925 | if (d_count(dentry) > 1) { |
1da177e4 | 1926 | spin_unlock(&dentry->d_lock); |
ccfeb506 | 1927 | /* Start asynchronous writeout of the inode */ |
2b0143b5 | 1928 | write_inode_now(d_inode(dentry), 0); |
1da177e4 | 1929 | error = nfs_sillyrename(dir, dentry); |
1ca42382 | 1930 | goto out; |
1da177e4 LT |
1931 | } |
1932 | if (!d_unhashed(dentry)) { | |
1933 | __d_drop(dentry); | |
1934 | need_rehash = 1; | |
1935 | } | |
1936 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1937 | error = nfs_safe_remove(dentry); |
d45b9d8b | 1938 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1939 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1940 | } else if (need_rehash) | |
1941 | d_rehash(dentry); | |
1ca42382 TM |
1942 | out: |
1943 | trace_nfs_unlink_exit(dir, dentry, error); | |
1da177e4 LT |
1944 | return error; |
1945 | } | |
ddda8e0a | 1946 | EXPORT_SYMBOL_GPL(nfs_unlink); |
1da177e4 | 1947 | |
873101b3 CL |
1948 | /* |
1949 | * To create a symbolic link, most file systems instantiate a new inode, | |
1950 | * add a page to it containing the path, then write it out to the disk | |
1951 | * using prepare_write/commit_write. | |
1952 | * | |
1953 | * Unfortunately the NFS client can't create the in-core inode first | |
1954 | * because it needs a file handle to create an in-core inode (see | |
1955 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1956 | * symlink request has completed on the server. | |
1957 | * | |
1958 | * So instead we allocate a raw page, copy the symname into it, then do | |
1959 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1960 | * now have a new file handle and can instantiate an in-core NFS inode | |
1961 | * and move the raw page into its mapping. | |
1962 | */ | |
597d9289 | 1963 | int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) |
1da177e4 | 1964 | { |
873101b3 CL |
1965 | struct page *page; |
1966 | char *kaddr; | |
1da177e4 | 1967 | struct iattr attr; |
873101b3 | 1968 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1969 | int error; |
1970 | ||
1e8968c5 | 1971 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, |
6de1472f | 1972 | dir->i_ino, dentry, symname); |
1da177e4 | 1973 | |
873101b3 CL |
1974 | if (pathlen > PAGE_SIZE) |
1975 | return -ENAMETOOLONG; | |
1da177e4 | 1976 | |
873101b3 CL |
1977 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1978 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1979 | |
e8ecde25 | 1980 | page = alloc_page(GFP_USER); |
76566991 | 1981 | if (!page) |
873101b3 | 1982 | return -ENOMEM; |
873101b3 | 1983 | |
e8ecde25 | 1984 | kaddr = page_address(page); |
873101b3 CL |
1985 | memcpy(kaddr, symname, pathlen); |
1986 | if (pathlen < PAGE_SIZE) | |
1987 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
873101b3 | 1988 | |
1ca42382 | 1989 | trace_nfs_symlink_enter(dir, dentry); |
94a6d753 | 1990 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
1ca42382 | 1991 | trace_nfs_symlink_exit(dir, dentry, error); |
873101b3 | 1992 | if (error != 0) { |
1e8968c5 | 1993 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", |
873101b3 | 1994 | dir->i_sb->s_id, dir->i_ino, |
6de1472f | 1995 | dentry, symname, error); |
1da177e4 | 1996 | d_drop(dentry); |
873101b3 | 1997 | __free_page(page); |
873101b3 CL |
1998 | return error; |
1999 | } | |
2000 | ||
2001 | /* | |
2002 | * No big deal if we can't add this page to the page cache here. | |
2003 | * READLINK will get the missing page from the server if needed. | |
2004 | */ | |
2b0143b5 | 2005 | if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, |
873101b3 | 2006 | GFP_KERNEL)) { |
873101b3 CL |
2007 | SetPageUptodate(page); |
2008 | unlock_page(page); | |
a0b54add RA |
2009 | /* |
2010 | * add_to_page_cache_lru() grabs an extra page refcount. | |
2011 | * Drop it here to avoid leaking this page later. | |
2012 | */ | |
09cbfeaf | 2013 | put_page(page); |
873101b3 CL |
2014 | } else |
2015 | __free_page(page); | |
2016 | ||
873101b3 | 2017 | return 0; |
1da177e4 | 2018 | } |
ddda8e0a | 2019 | EXPORT_SYMBOL_GPL(nfs_symlink); |
1da177e4 | 2020 | |
597d9289 | 2021 | int |
1da177e4 LT |
2022 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) |
2023 | { | |
2b0143b5 | 2024 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
2025 | int error; |
2026 | ||
6de1472f AV |
2027 | dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", |
2028 | old_dentry, dentry); | |
1da177e4 | 2029 | |
1fd1085b | 2030 | trace_nfs_link_enter(inode, dir, dentry); |
9697d234 | 2031 | d_drop(dentry); |
1da177e4 | 2032 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 2033 | if (error == 0) { |
7de9c6ee | 2034 | ihold(inode); |
9697d234 | 2035 | d_add(dentry, inode); |
cf809556 | 2036 | } |
1fd1085b | 2037 | trace_nfs_link_exit(inode, dir, dentry, error); |
1da177e4 LT |
2038 | return error; |
2039 | } | |
ddda8e0a | 2040 | EXPORT_SYMBOL_GPL(nfs_link); |
1da177e4 LT |
2041 | |
2042 | /* | |
2043 | * RENAME | |
2044 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
2045 | * different file handle for the same inode after a rename (e.g. when | |
2046 | * moving to a different directory). A fail-safe method to do so would | |
2047 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
2048 | * rename the old file using the sillyrename stuff. This way, the original | |
2049 | * file in old_dir will go away when the last process iput()s the inode. | |
2050 | * | |
2051 | * FIXED. | |
2052 | * | |
2053 | * It actually works quite well. One needs to have the possibility for | |
2054 | * at least one ".nfs..." file in each directory the file ever gets | |
2055 | * moved or linked to which happens automagically with the new | |
2056 | * implementation that only depends on the dcache stuff instead of | |
2057 | * using the inode layer | |
2058 | * | |
2059 | * Unfortunately, things are a little more complicated than indicated | |
2060 | * above. For a cross-directory move, we want to make sure we can get | |
2061 | * rid of the old inode after the operation. This means there must be | |
2062 | * no pending writes (if it's a file), and the use count must be 1. | |
2063 | * If these conditions are met, we can drop the dentries before doing | |
2064 | * the rename. | |
2065 | */ | |
597d9289 | 2066 | int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
1cd66c93 MS |
2067 | struct inode *new_dir, struct dentry *new_dentry, |
2068 | unsigned int flags) | |
1da177e4 | 2069 | { |
2b0143b5 DH |
2070 | struct inode *old_inode = d_inode(old_dentry); |
2071 | struct inode *new_inode = d_inode(new_dentry); | |
d9f29500 | 2072 | struct dentry *dentry = NULL, *rehash = NULL; |
80a491fd | 2073 | struct rpc_task *task; |
1da177e4 LT |
2074 | int error = -EBUSY; |
2075 | ||
1cd66c93 MS |
2076 | if (flags) |
2077 | return -EINVAL; | |
2078 | ||
6de1472f AV |
2079 | dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", |
2080 | old_dentry, new_dentry, | |
84d08fa8 | 2081 | d_count(new_dentry)); |
1da177e4 | 2082 | |
70ded201 | 2083 | trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); |
1da177e4 | 2084 | /* |
28f79a1a MS |
2085 | * For non-directories, check whether the target is busy and if so, |
2086 | * make a copy of the dentry and then do a silly-rename. If the | |
2087 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
2088 | * the new target. | |
1da177e4 | 2089 | */ |
27226104 MS |
2090 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
2091 | /* | |
2092 | * To prevent any new references to the target during the | |
2093 | * rename, we unhash the dentry in advance. | |
2094 | */ | |
d9f29500 | 2095 | if (!d_unhashed(new_dentry)) { |
27226104 | 2096 | d_drop(new_dentry); |
d9f29500 BC |
2097 | rehash = new_dentry; |
2098 | } | |
1da177e4 | 2099 | |
84d08fa8 | 2100 | if (d_count(new_dentry) > 2) { |
27226104 MS |
2101 | int err; |
2102 | ||
2103 | /* copy the target dentry's name */ | |
2104 | dentry = d_alloc(new_dentry->d_parent, | |
2105 | &new_dentry->d_name); | |
2106 | if (!dentry) | |
2107 | goto out; | |
2108 | ||
2109 | /* silly-rename the existing target ... */ | |
2110 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 2111 | if (err) |
27226104 | 2112 | goto out; |
24e93025 MS |
2113 | |
2114 | new_dentry = dentry; | |
d9f29500 | 2115 | rehash = NULL; |
24e93025 | 2116 | new_inode = NULL; |
27226104 | 2117 | } |
b1e4adf4 | 2118 | } |
1da177e4 | 2119 | |
d9f29500 | 2120 | task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); |
80a491fd JL |
2121 | if (IS_ERR(task)) { |
2122 | error = PTR_ERR(task); | |
2123 | goto out; | |
2124 | } | |
2125 | ||
2126 | error = rpc_wait_for_completion_task(task); | |
818a8dbe BC |
2127 | if (error != 0) { |
2128 | ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; | |
2129 | /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ | |
2130 | smp_wmb(); | |
2131 | } else | |
80a491fd JL |
2132 | error = task->tk_status; |
2133 | rpc_put_task(task); | |
59a707b0 TM |
2134 | /* Ensure the inode attributes are revalidated */ |
2135 | if (error == 0) { | |
2136 | spin_lock(&old_inode->i_lock); | |
2137 | NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); | |
2138 | NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE | |
2139 | | NFS_INO_INVALID_CTIME | |
2140 | | NFS_INO_REVAL_FORCED; | |
2141 | spin_unlock(&old_inode->i_lock); | |
2142 | } | |
1da177e4 | 2143 | out: |
d9f29500 BC |
2144 | if (rehash) |
2145 | d_rehash(rehash); | |
70ded201 TM |
2146 | trace_nfs_rename_exit(old_dir, old_dentry, |
2147 | new_dir, new_dentry, error); | |
d9f29500 BC |
2148 | if (!error) { |
2149 | if (new_inode != NULL) | |
2150 | nfs_drop_nlink(new_inode); | |
2151 | /* | |
2152 | * The d_move() should be here instead of in an async RPC completion | |
2153 | * handler because we need the proper locks to move the dentry. If | |
2154 | * we're interrupted by a signal, the async RPC completion handler | |
2155 | * should mark the directories for revalidation. | |
2156 | */ | |
2157 | d_move(old_dentry, new_dentry); | |
d803224c | 2158 | nfs_set_verifier(old_dentry, |
d9f29500 BC |
2159 | nfs_save_change_attribute(new_dir)); |
2160 | } else if (error == -ENOENT) | |
2161 | nfs_dentry_handle_enoent(old_dentry); | |
2162 | ||
1da177e4 LT |
2163 | /* new dentry created? */ |
2164 | if (dentry) | |
2165 | dput(dentry); | |
1da177e4 LT |
2166 | return error; |
2167 | } | |
ddda8e0a | 2168 | EXPORT_SYMBOL_GPL(nfs_rename); |
1da177e4 | 2169 | |
cfcea3e8 TM |
2170 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
2171 | static LIST_HEAD(nfs_access_lru_list); | |
2172 | static atomic_long_t nfs_access_nr_entries; | |
2173 | ||
3a505845 TM |
2174 | static unsigned long nfs_access_max_cachesize = ULONG_MAX; |
2175 | module_param(nfs_access_max_cachesize, ulong, 0644); | |
2176 | MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); | |
2177 | ||
1c3c07e9 TM |
2178 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
2179 | { | |
b68572e0 | 2180 | put_cred(entry->cred); |
f682a398 | 2181 | kfree_rcu(entry, rcu_head); |
4e857c58 | 2182 | smp_mb__before_atomic(); |
cfcea3e8 | 2183 | atomic_long_dec(&nfs_access_nr_entries); |
4e857c58 | 2184 | smp_mb__after_atomic(); |
1c3c07e9 TM |
2185 | } |
2186 | ||
1a81bb8a TM |
2187 | static void nfs_access_free_list(struct list_head *head) |
2188 | { | |
2189 | struct nfs_access_entry *cache; | |
2190 | ||
2191 | while (!list_empty(head)) { | |
2192 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
2193 | list_del(&cache->lru); | |
2194 | nfs_access_free_entry(cache); | |
2195 | } | |
2196 | } | |
2197 | ||
3a505845 TM |
2198 | static unsigned long |
2199 | nfs_do_access_cache_scan(unsigned int nr_to_scan) | |
979df72e TM |
2200 | { |
2201 | LIST_HEAD(head); | |
aa510da5 | 2202 | struct nfs_inode *nfsi, *next; |
979df72e | 2203 | struct nfs_access_entry *cache; |
1ab6c499 | 2204 | long freed = 0; |
979df72e | 2205 | |
a50f7951 | 2206 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 2207 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
2208 | struct inode *inode; |
2209 | ||
2210 | if (nr_to_scan-- == 0) | |
2211 | break; | |
9c7e7e23 | 2212 | inode = &nfsi->vfs_inode; |
979df72e TM |
2213 | spin_lock(&inode->i_lock); |
2214 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
2215 | goto remove_lru_entry; | |
2216 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
2217 | struct nfs_access_entry, lru); | |
2218 | list_move(&cache->lru, &head); | |
2219 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1ab6c499 | 2220 | freed++; |
979df72e TM |
2221 | if (!list_empty(&nfsi->access_cache_entry_lru)) |
2222 | list_move_tail(&nfsi->access_cache_inode_lru, | |
2223 | &nfs_access_lru_list); | |
2224 | else { | |
2225 | remove_lru_entry: | |
2226 | list_del_init(&nfsi->access_cache_inode_lru); | |
4e857c58 | 2227 | smp_mb__before_atomic(); |
979df72e | 2228 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
4e857c58 | 2229 | smp_mb__after_atomic(); |
979df72e | 2230 | } |
59844a9b | 2231 | spin_unlock(&inode->i_lock); |
979df72e TM |
2232 | } |
2233 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2234 | nfs_access_free_list(&head); |
1ab6c499 DC |
2235 | return freed; |
2236 | } | |
2237 | ||
3a505845 TM |
2238 | unsigned long |
2239 | nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
2240 | { | |
2241 | int nr_to_scan = sc->nr_to_scan; | |
2242 | gfp_t gfp_mask = sc->gfp_mask; | |
2243 | ||
2244 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) | |
2245 | return SHRINK_STOP; | |
2246 | return nfs_do_access_cache_scan(nr_to_scan); | |
2247 | } | |
2248 | ||
2249 | ||
1ab6c499 DC |
2250 | unsigned long |
2251 | nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
2252 | { | |
55f841ce | 2253 | return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); |
979df72e TM |
2254 | } |
2255 | ||
3a505845 TM |
2256 | static void |
2257 | nfs_access_cache_enforce_limit(void) | |
2258 | { | |
2259 | long nr_entries = atomic_long_read(&nfs_access_nr_entries); | |
2260 | unsigned long diff; | |
2261 | unsigned int nr_to_scan; | |
2262 | ||
2263 | if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) | |
2264 | return; | |
2265 | nr_to_scan = 100; | |
2266 | diff = nr_entries - nfs_access_max_cachesize; | |
2267 | if (diff < nr_to_scan) | |
2268 | nr_to_scan = diff; | |
2269 | nfs_do_access_cache_scan(nr_to_scan); | |
2270 | } | |
2271 | ||
1a81bb8a | 2272 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2273 | { |
1c3c07e9 | 2274 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2275 | struct rb_node *n; |
1c3c07e9 TM |
2276 | struct nfs_access_entry *entry; |
2277 | ||
2278 | /* Unhook entries from the cache */ | |
2279 | while ((n = rb_first(root_node)) != NULL) { | |
2280 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2281 | rb_erase(n, root_node); | |
1a81bb8a | 2282 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2283 | } |
2284 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2285 | } |
2286 | ||
1c3c07e9 | 2287 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2288 | { |
1a81bb8a TM |
2289 | LIST_HEAD(head); |
2290 | ||
2291 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2292 | return; | |
cfcea3e8 | 2293 | /* Remove from global LRU init */ |
1a81bb8a TM |
2294 | spin_lock(&nfs_access_lru_lock); |
2295 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2296 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2297 | |
1c3c07e9 | 2298 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2299 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2300 | spin_unlock(&inode->i_lock); | |
2301 | spin_unlock(&nfs_access_lru_lock); | |
2302 | nfs_access_free_list(&head); | |
1c3c07e9 | 2303 | } |
1c606fb7 | 2304 | EXPORT_SYMBOL_GPL(nfs_access_zap_cache); |
1da177e4 | 2305 | |
b68572e0 | 2306 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) |
1c3c07e9 TM |
2307 | { |
2308 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
1c3c07e9 TM |
2309 | |
2310 | while (n != NULL) { | |
b68572e0 N |
2311 | struct nfs_access_entry *entry = |
2312 | rb_entry(n, struct nfs_access_entry, rb_node); | |
2313 | int cmp = cred_fscmp(cred, entry->cred); | |
1c3c07e9 | 2314 | |
b68572e0 | 2315 | if (cmp < 0) |
1c3c07e9 | 2316 | n = n->rb_left; |
b68572e0 | 2317 | else if (cmp > 0) |
1c3c07e9 TM |
2318 | n = n->rb_right; |
2319 | else | |
2320 | return entry; | |
1da177e4 | 2321 | } |
1c3c07e9 TM |
2322 | return NULL; |
2323 | } | |
2324 | ||
b68572e0 | 2325 | static int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block) |
1c3c07e9 TM |
2326 | { |
2327 | struct nfs_inode *nfsi = NFS_I(inode); | |
2328 | struct nfs_access_entry *cache; | |
57b69181 TM |
2329 | bool retry = true; |
2330 | int err; | |
1c3c07e9 | 2331 | |
dc59250c | 2332 | spin_lock(&inode->i_lock); |
57b69181 TM |
2333 | for(;;) { |
2334 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
2335 | goto out_zap; | |
2336 | cache = nfs_access_search_rbtree(inode, cred); | |
2337 | err = -ENOENT; | |
2338 | if (cache == NULL) | |
2339 | goto out; | |
2340 | /* Found an entry, is our attribute cache valid? */ | |
21c3ba7e | 2341 | if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) |
57b69181 | 2342 | break; |
5c965db8 TM |
2343 | if (!retry) |
2344 | break; | |
57b69181 TM |
2345 | err = -ECHILD; |
2346 | if (!may_block) | |
2347 | goto out; | |
57b69181 TM |
2348 | spin_unlock(&inode->i_lock); |
2349 | err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2350 | if (err) | |
2351 | return err; | |
2352 | spin_lock(&inode->i_lock); | |
2353 | retry = false; | |
2354 | } | |
1c3c07e9 TM |
2355 | res->cred = cache->cred; |
2356 | res->mask = cache->mask; | |
cfcea3e8 | 2357 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2358 | err = 0; |
2359 | out: | |
2360 | spin_unlock(&inode->i_lock); | |
2361 | return err; | |
1c3c07e9 | 2362 | out_zap: |
1a81bb8a TM |
2363 | spin_unlock(&inode->i_lock); |
2364 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2365 | return -ENOENT; |
2366 | } | |
2367 | ||
b68572e0 | 2368 | static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res) |
f682a398 N |
2369 | { |
2370 | /* Only check the most recently returned cache entry, | |
2371 | * but do it without locking. | |
2372 | */ | |
2373 | struct nfs_inode *nfsi = NFS_I(inode); | |
2374 | struct nfs_access_entry *cache; | |
2375 | int err = -ECHILD; | |
2376 | struct list_head *lh; | |
2377 | ||
2378 | rcu_read_lock(); | |
2379 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
2380 | goto out; | |
2381 | lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); | |
2382 | cache = list_entry(lh, struct nfs_access_entry, lru); | |
2383 | if (lh == &nfsi->access_cache_entry_lru || | |
9a206de2 | 2384 | cred_fscmp(cred, cache->cred) != 0) |
f682a398 N |
2385 | cache = NULL; |
2386 | if (cache == NULL) | |
2387 | goto out; | |
21c3ba7e | 2388 | if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) |
f682a398 | 2389 | goto out; |
f682a398 N |
2390 | res->cred = cache->cred; |
2391 | res->mask = cache->mask; | |
21c3ba7e | 2392 | err = 0; |
f682a398 N |
2393 | out: |
2394 | rcu_read_unlock(); | |
2395 | return err; | |
2396 | } | |
2397 | ||
1c3c07e9 TM |
2398 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) |
2399 | { | |
cfcea3e8 TM |
2400 | struct nfs_inode *nfsi = NFS_I(inode); |
2401 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2402 | struct rb_node **p = &root_node->rb_node; |
2403 | struct rb_node *parent = NULL; | |
2404 | struct nfs_access_entry *entry; | |
b68572e0 | 2405 | int cmp; |
1c3c07e9 TM |
2406 | |
2407 | spin_lock(&inode->i_lock); | |
2408 | while (*p != NULL) { | |
2409 | parent = *p; | |
2410 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
b68572e0 | 2411 | cmp = cred_fscmp(set->cred, entry->cred); |
1c3c07e9 | 2412 | |
b68572e0 | 2413 | if (cmp < 0) |
1c3c07e9 | 2414 | p = &parent->rb_left; |
b68572e0 | 2415 | else if (cmp > 0) |
1c3c07e9 TM |
2416 | p = &parent->rb_right; |
2417 | else | |
2418 | goto found; | |
2419 | } | |
2420 | rb_link_node(&set->rb_node, parent, p); | |
2421 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2422 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2423 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2424 | return; |
2425 | found: | |
2426 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2427 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2428 | list_del(&entry->lru); | |
1c3c07e9 TM |
2429 | spin_unlock(&inode->i_lock); |
2430 | nfs_access_free_entry(entry); | |
2431 | } | |
2432 | ||
6168f62c | 2433 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2434 | { |
2435 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2436 | if (cache == NULL) | |
2437 | return; | |
2438 | RB_CLEAR_NODE(&cache->rb_node); | |
b68572e0 | 2439 | cache->cred = get_cred(set->cred); |
1da177e4 | 2440 | cache->mask = set->mask; |
1c3c07e9 | 2441 | |
f682a398 N |
2442 | /* The above field assignments must be visible |
2443 | * before this item appears on the lru. We cannot easily | |
2444 | * use rcu_assign_pointer, so just force the memory barrier. | |
2445 | */ | |
2446 | smp_wmb(); | |
1c3c07e9 | 2447 | nfs_access_add_rbtree(inode, cache); |
cfcea3e8 TM |
2448 | |
2449 | /* Update accounting */ | |
4e857c58 | 2450 | smp_mb__before_atomic(); |
cfcea3e8 | 2451 | atomic_long_inc(&nfs_access_nr_entries); |
4e857c58 | 2452 | smp_mb__after_atomic(); |
cfcea3e8 TM |
2453 | |
2454 | /* Add inode to global LRU list */ | |
1a81bb8a | 2455 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2456 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2457 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2458 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2459 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2460 | spin_unlock(&nfs_access_lru_lock); |
2461 | } | |
3a505845 | 2462 | nfs_access_cache_enforce_limit(); |
1da177e4 | 2463 | } |
6168f62c WAA |
2464 | EXPORT_SYMBOL_GPL(nfs_access_add_cache); |
2465 | ||
3c181827 AS |
2466 | #define NFS_MAY_READ (NFS_ACCESS_READ) |
2467 | #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ | |
2468 | NFS_ACCESS_EXTEND | \ | |
2469 | NFS_ACCESS_DELETE) | |
2470 | #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ | |
2471 | NFS_ACCESS_EXTEND) | |
ecbb903c | 2472 | #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE |
3c181827 AS |
2473 | #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) |
2474 | #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) | |
15d4b73a | 2475 | static int |
ecbb903c | 2476 | nfs_access_calc_mask(u32 access_result, umode_t umode) |
15d4b73a TM |
2477 | { |
2478 | int mask = 0; | |
2479 | ||
2480 | if (access_result & NFS_MAY_READ) | |
2481 | mask |= MAY_READ; | |
ecbb903c TM |
2482 | if (S_ISDIR(umode)) { |
2483 | if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) | |
2484 | mask |= MAY_WRITE; | |
2485 | if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) | |
2486 | mask |= MAY_EXEC; | |
2487 | } else if (S_ISREG(umode)) { | |
2488 | if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) | |
2489 | mask |= MAY_WRITE; | |
2490 | if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) | |
2491 | mask |= MAY_EXEC; | |
2492 | } else if (access_result & NFS_MAY_WRITE) | |
2493 | mask |= MAY_WRITE; | |
15d4b73a TM |
2494 | return mask; |
2495 | } | |
2496 | ||
6168f62c WAA |
2497 | void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) |
2498 | { | |
bd8b2441 | 2499 | entry->mask = access_result; |
6168f62c WAA |
2500 | } |
2501 | EXPORT_SYMBOL_GPL(nfs_access_set_mask); | |
1da177e4 | 2502 | |
b68572e0 | 2503 | static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) |
1da177e4 LT |
2504 | { |
2505 | struct nfs_access_entry cache; | |
57b69181 | 2506 | bool may_block = (mask & MAY_NOT_BLOCK) == 0; |
e8194b7d | 2507 | int cache_mask = -1; |
1da177e4 LT |
2508 | int status; |
2509 | ||
f4ce1299 TM |
2510 | trace_nfs_access_enter(inode); |
2511 | ||
f682a398 N |
2512 | status = nfs_access_get_cached_rcu(inode, cred, &cache); |
2513 | if (status != 0) | |
57b69181 | 2514 | status = nfs_access_get_cached(inode, cred, &cache, may_block); |
1da177e4 | 2515 | if (status == 0) |
f4ce1299 | 2516 | goto out_cached; |
1da177e4 | 2517 | |
f3324a2a | 2518 | status = -ECHILD; |
57b69181 | 2519 | if (!may_block) |
f3324a2a N |
2520 | goto out; |
2521 | ||
1750d929 AS |
2522 | /* |
2523 | * Determine which access bits we want to ask for... | |
2524 | */ | |
2525 | cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; | |
2526 | if (S_ISDIR(inode->i_mode)) | |
2527 | cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; | |
2528 | else | |
2529 | cache.mask |= NFS_ACCESS_EXECUTE; | |
1da177e4 | 2530 | cache.cred = cred; |
1da177e4 | 2531 | status = NFS_PROTO(inode)->access(inode, &cache); |
a71ee337 SJ |
2532 | if (status != 0) { |
2533 | if (status == -ESTALE) { | |
2534 | nfs_zap_caches(inode); | |
2535 | if (!S_ISDIR(inode->i_mode)) | |
2536 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2537 | } | |
f4ce1299 | 2538 | goto out; |
a71ee337 | 2539 | } |
1da177e4 | 2540 | nfs_access_add_cache(inode, &cache); |
f4ce1299 | 2541 | out_cached: |
ecbb903c | 2542 | cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); |
bd8b2441 | 2543 | if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) |
f4ce1299 | 2544 | status = -EACCES; |
1da177e4 | 2545 | out: |
e8194b7d | 2546 | trace_nfs_access_exit(inode, mask, cache_mask, status); |
f4ce1299 | 2547 | return status; |
1da177e4 LT |
2548 | } |
2549 | ||
af22f94a TM |
2550 | static int nfs_open_permission_mask(int openflags) |
2551 | { | |
2552 | int mask = 0; | |
2553 | ||
f8d9a897 WAA |
2554 | if (openflags & __FMODE_EXEC) { |
2555 | /* ONLY check exec rights */ | |
2556 | mask = MAY_EXEC; | |
2557 | } else { | |
2558 | if ((openflags & O_ACCMODE) != O_WRONLY) | |
2559 | mask |= MAY_READ; | |
2560 | if ((openflags & O_ACCMODE) != O_RDONLY) | |
2561 | mask |= MAY_WRITE; | |
2562 | } | |
2563 | ||
af22f94a TM |
2564 | return mask; |
2565 | } | |
2566 | ||
b68572e0 | 2567 | int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) |
af22f94a TM |
2568 | { |
2569 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2570 | } | |
89d77c8f | 2571 | EXPORT_SYMBOL_GPL(nfs_may_open); |
af22f94a | 2572 | |
5c5fc09a TM |
2573 | static int nfs_execute_ok(struct inode *inode, int mask) |
2574 | { | |
2575 | struct nfs_server *server = NFS_SERVER(inode); | |
21c3ba7e | 2576 | int ret = 0; |
5c5fc09a | 2577 | |
3825827e TM |
2578 | if (S_ISDIR(inode->i_mode)) |
2579 | return 0; | |
cf834027 | 2580 | if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) { |
21c3ba7e TM |
2581 | if (mask & MAY_NOT_BLOCK) |
2582 | return -ECHILD; | |
2583 | ret = __nfs_revalidate_inode(server, inode); | |
2584 | } | |
5c5fc09a TM |
2585 | if (ret == 0 && !execute_ok(inode)) |
2586 | ret = -EACCES; | |
2587 | return ret; | |
2588 | } | |
2589 | ||
10556cb2 | 2590 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 | 2591 | { |
b68572e0 | 2592 | const struct cred *cred = current_cred(); |
1da177e4 LT |
2593 | int res = 0; |
2594 | ||
91d5b470 CL |
2595 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2596 | ||
e6305c43 | 2597 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2598 | goto out; |
2599 | /* Is this sys_access() ? */ | |
9cfcac81 | 2600 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2601 | goto force_lookup; |
2602 | ||
2603 | switch (inode->i_mode & S_IFMT) { | |
2604 | case S_IFLNK: | |
2605 | goto out; | |
2606 | case S_IFREG: | |
762674f8 TM |
2607 | if ((mask & MAY_OPEN) && |
2608 | nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) | |
2609 | return 0; | |
1da177e4 LT |
2610 | break; |
2611 | case S_IFDIR: | |
2612 | /* | |
2613 | * Optimize away all write operations, since the server | |
2614 | * will check permissions when we perform the op. | |
2615 | */ | |
2616 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2617 | goto out; | |
2618 | } | |
2619 | ||
2620 | force_lookup: | |
1da177e4 LT |
2621 | if (!NFS_PROTO(inode)->access) |
2622 | goto out_notsup; | |
2623 | ||
f3324a2a N |
2624 | /* Always try fast lookups first */ |
2625 | rcu_read_lock(); | |
b68572e0 | 2626 | res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); |
f3324a2a N |
2627 | rcu_read_unlock(); |
2628 | if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { | |
2629 | /* Fast lookup failed, try the slow way */ | |
b68572e0 | 2630 | res = nfs_do_access(inode, cred, mask); |
f3324a2a | 2631 | } |
1da177e4 | 2632 | out: |
5c5fc09a TM |
2633 | if (!res && (mask & MAY_EXEC)) |
2634 | res = nfs_execute_ok(inode, mask); | |
f696a365 | 2635 | |
1e8968c5 | 2636 | dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", |
1e7cb3dc | 2637 | inode->i_sb->s_id, inode->i_ino, mask, res); |
1da177e4 LT |
2638 | return res; |
2639 | out_notsup: | |
d51ac1a8 N |
2640 | if (mask & MAY_NOT_BLOCK) |
2641 | return -ECHILD; | |
2642 | ||
1da177e4 LT |
2643 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); |
2644 | if (res == 0) | |
2830ba7f | 2645 | res = generic_permission(inode, mask); |
1e7cb3dc | 2646 | goto out; |
1da177e4 | 2647 | } |
ddda8e0a | 2648 | EXPORT_SYMBOL_GPL(nfs_permission); |
1da177e4 LT |
2649 | |
2650 | /* | |
2651 | * Local variables: | |
2652 | * version-control: t | |
2653 | * kept-new-versions: 5 | |
2654 | * End: | |
2655 | */ |