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 | ||
b6459415 | 21 | #include <linux/compat.h> |
ddda8e0a | 22 | #include <linux/module.h> |
1da177e4 LT |
23 | #include <linux/time.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/stat.h> | |
26 | #include <linux/fcntl.h> | |
27 | #include <linux/string.h> | |
28 | #include <linux/kernel.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/mm.h> | |
31 | #include <linux/sunrpc/clnt.h> | |
32 | #include <linux/nfs_fs.h> | |
33 | #include <linux/nfs_mount.h> | |
34 | #include <linux/pagemap.h> | |
873101b3 | 35 | #include <linux/pagevec.h> |
1da177e4 | 36 | #include <linux/namei.h> |
54ceac45 | 37 | #include <linux/mount.h> |
a0b8cab3 | 38 | #include <linux/swap.h> |
e8edc6e0 | 39 | #include <linux/sched.h> |
04e4bd1c | 40 | #include <linux/kmemleak.h> |
64c2ce8b | 41 | #include <linux/xattr.h> |
830f1111 | 42 | #include <linux/hash.h> |
1da177e4 LT |
43 | |
44 | #include "delegation.h" | |
91d5b470 | 45 | #include "iostat.h" |
4c30d56e | 46 | #include "internal.h" |
cd9a1c0e | 47 | #include "fscache.h" |
1da177e4 | 48 | |
f4ce1299 TM |
49 | #include "nfstrace.h" |
50 | ||
1da177e4 LT |
51 | /* #define NFS_DEBUG_VERBOSE 1 */ |
52 | ||
53 | static int nfs_opendir(struct inode *, struct file *); | |
480c2006 | 54 | static int nfs_closedir(struct inode *, struct file *); |
23db8620 | 55 | static int nfs_readdir(struct file *, struct dir_context *); |
02c24a82 | 56 | static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); |
f0dd2136 | 57 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
ec108d3c | 58 | static void nfs_readdir_clear_array(struct folio *); |
1da177e4 | 59 | |
4b6f5d20 | 60 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 61 | .llseek = nfs_llseek_dir, |
1da177e4 | 62 | .read = generic_read_dir, |
93a6ab7b | 63 | .iterate_shared = nfs_readdir, |
1da177e4 | 64 | .open = nfs_opendir, |
480c2006 | 65 | .release = nfs_closedir, |
1da177e4 LT |
66 | .fsync = nfs_fsync_dir, |
67 | }; | |
68 | ||
11de3b11 | 69 | const struct address_space_operations nfs_dir_aops = { |
ec108d3c | 70 | .free_folio = nfs_readdir_clear_array, |
d1bacf9e BS |
71 | }; |
72 | ||
580f2367 TM |
73 | #define NFS_INIT_DTSIZE PAGE_SIZE |
74 | ||
281f31b2 TM |
75 | static struct nfs_open_dir_context * |
76 | alloc_nfs_open_dir_context(struct inode *dir) | |
480c2006 | 77 | { |
311324ad | 78 | struct nfs_inode *nfsi = NFS_I(dir); |
480c2006 | 79 | struct nfs_open_dir_context *ctx; |
281f31b2 TM |
80 | |
81 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT); | |
480c2006 | 82 | if (ctx != NULL) { |
311324ad | 83 | ctx->attr_gencount = nfsi->attr_gencount; |
580f2367 | 84 | ctx->dtsize = NFS_INIT_DTSIZE; |
311324ad | 85 | spin_lock(&dir->i_lock); |
1c341b77 TM |
86 | if (list_empty(&nfsi->open_files) && |
87 | (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) | |
ac46b3d7 TM |
88 | nfs_set_cache_invalid(dir, |
89 | NFS_INO_INVALID_DATA | | |
90 | NFS_INO_REVAL_FORCED); | |
230bc98f | 91 | list_add_tail_rcu(&ctx->list, &nfsi->open_files); |
d1e32ea3 | 92 | memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf)); |
311324ad | 93 | spin_unlock(&dir->i_lock); |
0c030806 TM |
94 | return ctx; |
95 | } | |
96 | return ERR_PTR(-ENOMEM); | |
480c2006 BS |
97 | } |
98 | ||
311324ad | 99 | static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) |
480c2006 | 100 | { |
311324ad | 101 | spin_lock(&dir->i_lock); |
230bc98f | 102 | list_del_rcu(&ctx->list); |
311324ad | 103 | spin_unlock(&dir->i_lock); |
230bc98f | 104 | kfree_rcu(ctx, rcu_head); |
480c2006 BS |
105 | } |
106 | ||
1da177e4 LT |
107 | /* |
108 | * Open file | |
109 | */ | |
110 | static int | |
111 | nfs_opendir(struct inode *inode, struct file *filp) | |
112 | { | |
480c2006 BS |
113 | int res = 0; |
114 | struct nfs_open_dir_context *ctx; | |
1da177e4 | 115 | |
6de1472f | 116 | dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); |
cc0dd2d1 CL |
117 | |
118 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 119 | |
93b8959a | 120 | ctx = alloc_nfs_open_dir_context(inode); |
480c2006 BS |
121 | if (IS_ERR(ctx)) { |
122 | res = PTR_ERR(ctx); | |
123 | goto out; | |
124 | } | |
125 | filp->private_data = ctx; | |
480c2006 | 126 | out: |
1da177e4 LT |
127 | return res; |
128 | } | |
129 | ||
480c2006 BS |
130 | static int |
131 | nfs_closedir(struct inode *inode, struct file *filp) | |
132 | { | |
a455589f | 133 | put_nfs_open_dir_context(file_inode(filp), filp->private_data); |
480c2006 BS |
134 | return 0; |
135 | } | |
136 | ||
d1bacf9e BS |
137 | struct nfs_cache_array_entry { |
138 | u64 cookie; | |
139 | u64 ino; | |
a52a8a6a TM |
140 | const char *name; |
141 | unsigned int name_len; | |
0b26a0bf | 142 | unsigned char d_type; |
d1bacf9e BS |
143 | }; |
144 | ||
145 | struct nfs_cache_array { | |
d09e673f | 146 | u64 change_attr; |
d1bacf9e | 147 | u64 last_cookie; |
b1e21c97 | 148 | unsigned int size; |
ec108d3c AS |
149 | unsigned char folio_full : 1, |
150 | folio_is_eof : 1, | |
762567b7 | 151 | cookies_are_ordered : 1; |
5601cda8 | 152 | struct nfs_cache_array_entry array[]; |
d1bacf9e BS |
153 | }; |
154 | ||
6c981eff | 155 | struct nfs_readdir_descriptor { |
1da177e4 | 156 | struct file *file; |
61f02e0a | 157 | struct folio *folio; |
23db8620 | 158 | struct dir_context *ctx; |
61f02e0a AS |
159 | pgoff_t folio_index; |
160 | pgoff_t folio_index_max; | |
2e7a4641 | 161 | u64 dir_cookie; |
0aded708 | 162 | u64 last_cookie; |
f0dd2136 | 163 | loff_t current_index; |
d1bacf9e | 164 | |
b593c09f | 165 | __be32 verf[NFS_DIR_VERIFIER_SIZE]; |
a1147b82 | 166 | unsigned long dir_verifier; |
1f4eab7e | 167 | unsigned long timestamp; |
4704f0e2 | 168 | unsigned long gencount; |
2e7a4641 | 169 | unsigned long attr_gencount; |
d1bacf9e | 170 | unsigned int cache_entry_index; |
580f2367 TM |
171 | unsigned int buffer_fills; |
172 | unsigned int dtsize; | |
b0365ccb | 173 | bool clear_cache; |
a7a3b1e9 | 174 | bool plus; |
e1d2699b | 175 | bool eob; |
a7a3b1e9 | 176 | bool eof; |
6c981eff | 177 | }; |
1da177e4 | 178 | |
580f2367 TM |
179 | static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz) |
180 | { | |
181 | struct nfs_server *server = NFS_SERVER(file_inode(desc->file)); | |
182 | unsigned int maxsize = server->dtsize; | |
183 | ||
184 | if (sz > maxsize) | |
185 | sz = maxsize; | |
186 | if (sz < NFS_MIN_FILE_IO_SIZE) | |
187 | sz = NFS_MIN_FILE_IO_SIZE; | |
188 | desc->dtsize = sz; | |
189 | } | |
190 | ||
191 | static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc) | |
192 | { | |
193 | nfs_set_dtsize(desc, desc->dtsize >> 1); | |
194 | } | |
195 | ||
196 | static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc) | |
197 | { | |
198 | nfs_set_dtsize(desc, desc->dtsize << 1); | |
199 | } | |
200 | ||
ec108d3c AS |
201 | static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie, |
202 | u64 change_attr) | |
4b310319 TM |
203 | { |
204 | struct nfs_cache_array *array; | |
205 | ||
ec108d3c | 206 | array = kmap_local_folio(folio, 0); |
d09e673f | 207 | array->change_attr = change_attr; |
1f1d4aa4 | 208 | array->last_cookie = last_cookie; |
9332cf14 | 209 | array->size = 0; |
ec108d3c AS |
210 | array->folio_full = 0; |
211 | array->folio_is_eof = 0; | |
762567b7 | 212 | array->cookies_are_ordered = 1; |
1683ed16 | 213 | kunmap_local(array); |
4b310319 TM |
214 | } |
215 | ||
d1bacf9e BS |
216 | /* |
217 | * we are freeing strings created by nfs_add_to_readdir_array() | |
218 | */ | |
ec108d3c | 219 | static void nfs_readdir_clear_array(struct folio *folio) |
d1bacf9e | 220 | { |
11de3b11 | 221 | struct nfs_cache_array *array; |
9332cf14 | 222 | unsigned int i; |
8cd51a0c | 223 | |
ec108d3c | 224 | array = kmap_local_folio(folio, 0); |
b044f645 | 225 | for (i = 0; i < array->size; i++) |
a52a8a6a | 226 | kfree(array->array[i].name); |
9332cf14 | 227 | array->size = 0; |
1683ed16 | 228 | kunmap_local(array); |
d1bacf9e BS |
229 | } |
230 | ||
ec108d3c AS |
231 | static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie, |
232 | u64 change_attr) | |
aa5dc8c4 | 233 | { |
ec108d3c AS |
234 | nfs_readdir_clear_array(folio); |
235 | nfs_readdir_folio_init_array(folio, last_cookie, change_attr); | |
aa5dc8c4 MWO |
236 | } |
237 | ||
ec108d3c AS |
238 | static struct folio * |
239 | nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags) | |
35df59d3 | 240 | { |
ec108d3c AS |
241 | struct folio *folio = folio_alloc(gfp_flags, 0); |
242 | if (folio) | |
243 | nfs_readdir_folio_init_array(folio, last_cookie, 0); | |
244 | return folio; | |
35df59d3 TM |
245 | } |
246 | ||
ec108d3c | 247 | static void nfs_readdir_folio_array_free(struct folio *folio) |
35df59d3 | 248 | { |
ec108d3c AS |
249 | if (folio) { |
250 | nfs_readdir_clear_array(folio); | |
251 | folio_put(folio); | |
35df59d3 TM |
252 | } |
253 | } | |
254 | ||
e47a62df TM |
255 | static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array) |
256 | { | |
257 | return array->size == 0 ? array->last_cookie : array->array[0].cookie; | |
258 | } | |
259 | ||
b1e21c97 TM |
260 | static void nfs_readdir_array_set_eof(struct nfs_cache_array *array) |
261 | { | |
ec108d3c AS |
262 | array->folio_is_eof = 1; |
263 | array->folio_full = 1; | |
b1e21c97 TM |
264 | } |
265 | ||
266 | static bool nfs_readdir_array_is_full(struct nfs_cache_array *array) | |
267 | { | |
ec108d3c | 268 | return array->folio_full; |
b1e21c97 TM |
269 | } |
270 | ||
d1bacf9e BS |
271 | /* |
272 | * the caller is responsible for freeing qstr.name | |
273 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
274 | * nfs_clear_readdir_array() | |
275 | */ | |
a52a8a6a | 276 | static const char *nfs_readdir_copy_name(const char *name, unsigned int len) |
d1bacf9e | 277 | { |
a52a8a6a TM |
278 | const char *ret = kmemdup_nul(name, len, GFP_KERNEL); |
279 | ||
04e4bd1c CM |
280 | /* |
281 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
282 | * in a page cache page which kmemleak does not scan. | |
283 | */ | |
a52a8a6a TM |
284 | if (ret != NULL) |
285 | kmemleak_not_leak(ret); | |
286 | return ret; | |
d1bacf9e BS |
287 | } |
288 | ||
0b2662b7 TM |
289 | static size_t nfs_readdir_array_maxentries(void) |
290 | { | |
291 | return (PAGE_SIZE - sizeof(struct nfs_cache_array)) / | |
292 | sizeof(struct nfs_cache_array_entry); | |
293 | } | |
294 | ||
b1e21c97 TM |
295 | /* |
296 | * Check that the next array entry lies entirely within the page bounds | |
297 | */ | |
298 | static int nfs_readdir_array_can_expand(struct nfs_cache_array *array) | |
299 | { | |
ec108d3c | 300 | if (array->folio_full) |
b1e21c97 | 301 | return -ENOSPC; |
0b2662b7 | 302 | if (array->size == nfs_readdir_array_maxentries()) { |
ec108d3c | 303 | array->folio_full = 1; |
b1e21c97 TM |
304 | return -ENOSPC; |
305 | } | |
4a201d6e | 306 | return 0; |
d1bacf9e BS |
307 | } |
308 | ||
ec108d3c | 309 | static int nfs_readdir_folio_array_append(struct folio *folio, |
61f02e0a AS |
310 | const struct nfs_entry *entry, |
311 | u64 *cookie) | |
d1bacf9e | 312 | { |
a52a8a6a | 313 | struct nfs_cache_array *array; |
4a201d6e | 314 | struct nfs_cache_array_entry *cache_entry; |
a52a8a6a | 315 | const char *name; |
0adf85b4 | 316 | int ret = -ENOMEM; |
4a201d6e | 317 | |
a52a8a6a | 318 | name = nfs_readdir_copy_name(entry->name, entry->len); |
3020093f | 319 | |
ec108d3c | 320 | array = kmap_atomic(folio_page(folio, 0)); |
0adf85b4 TM |
321 | if (!name) |
322 | goto out; | |
b1e21c97 | 323 | ret = nfs_readdir_array_can_expand(array); |
a52a8a6a TM |
324 | if (ret) { |
325 | kfree(name); | |
4a201d6e | 326 | goto out; |
a52a8a6a | 327 | } |
d1bacf9e | 328 | |
b1e21c97 | 329 | cache_entry = &array->array[array->size]; |
0adf85b4 | 330 | cache_entry->cookie = array->last_cookie; |
4a201d6e | 331 | cache_entry->ino = entry->ino; |
0b26a0bf | 332 | cache_entry->d_type = entry->d_type; |
a52a8a6a TM |
333 | cache_entry->name_len = entry->len; |
334 | cache_entry->name = name; | |
d1bacf9e | 335 | array->last_cookie = entry->cookie; |
762567b7 TM |
336 | if (array->last_cookie <= cache_entry->cookie) |
337 | array->cookies_are_ordered = 0; | |
8cd51a0c | 338 | array->size++; |
47c716cb | 339 | if (entry->eof != 0) |
b1e21c97 | 340 | nfs_readdir_array_set_eof(array); |
4a201d6e | 341 | out: |
0adf85b4 | 342 | *cookie = array->last_cookie; |
a52a8a6a | 343 | kunmap_atomic(array); |
4a201d6e | 344 | return ret; |
d1bacf9e BS |
345 | } |
346 | ||
f648022f TM |
347 | #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14) |
348 | /* | |
349 | * Hash algorithm allowing content addressible access to sequences | |
350 | * of directory cookies. Content is addressed by the value of the | |
351 | * cookie index of the first readdir entry in a page. | |
352 | * | |
830f1111 | 353 | * We select only the first 18 bits to avoid issues with excessive |
f648022f TM |
354 | * memory use for the page cache XArray. 18 bits should allow the caching |
355 | * of 262144 pages of sequences of readdir entries. Since each page holds | |
356 | * 127 readdir entries for a typical 64-bit system, that works out to a | |
357 | * cache of ~ 33 million entries per directory. | |
358 | */ | |
ec108d3c | 359 | static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie) |
f648022f TM |
360 | { |
361 | if (cookie == 0) | |
362 | return 0; | |
830f1111 | 363 | return hash_64(cookie, 18); |
f648022f TM |
364 | } |
365 | ||
ec108d3c AS |
366 | static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie, |
367 | u64 change_attr) | |
d09e673f | 368 | { |
ec108d3c | 369 | struct nfs_cache_array *array = kmap_local_folio(folio, 0); |
d09e673f TM |
370 | int ret = true; |
371 | ||
372 | if (array->change_attr != change_attr) | |
373 | ret = false; | |
e47a62df | 374 | if (nfs_readdir_array_index_cookie(array) != last_cookie) |
d09e673f | 375 | ret = false; |
1683ed16 | 376 | kunmap_local(array); |
d09e673f TM |
377 | return ret; |
378 | } | |
379 | ||
ec108d3c | 380 | static void nfs_readdir_folio_unlock_and_put(struct folio *folio) |
d09e673f | 381 | { |
ec108d3c AS |
382 | folio_unlock(folio); |
383 | folio_put(folio); | |
d09e673f TM |
384 | } |
385 | ||
ec108d3c AS |
386 | static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie, |
387 | u64 change_attr) | |
648a4548 | 388 | { |
ec108d3c AS |
389 | if (folio_test_uptodate(folio)) { |
390 | if (nfs_readdir_folio_validate(folio, cookie, change_attr)) | |
648a4548 | 391 | return; |
ec108d3c | 392 | nfs_readdir_clear_array(folio); |
648a4548 | 393 | } |
ec108d3c AS |
394 | nfs_readdir_folio_init_array(folio, cookie, change_attr); |
395 | folio_mark_uptodate(folio); | |
648a4548 TM |
396 | } |
397 | ||
ec108d3c AS |
398 | static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping, |
399 | u64 cookie, u64 change_attr) | |
1f1d4aa4 | 400 | { |
ec108d3c AS |
401 | pgoff_t index = nfs_readdir_folio_cookie_hash(cookie); |
402 | struct folio *folio; | |
1f1d4aa4 | 403 | |
ec108d3c AS |
404 | folio = filemap_grab_folio(mapping, index); |
405 | if (!folio) | |
d09e673f | 406 | return NULL; |
ec108d3c AS |
407 | nfs_readdir_folio_init_and_validate(folio, cookie, change_attr); |
408 | return folio; | |
1f1d4aa4 TM |
409 | } |
410 | ||
ec108d3c | 411 | static u64 nfs_readdir_folio_last_cookie(struct folio *folio) |
1f1d4aa4 TM |
412 | { |
413 | struct nfs_cache_array *array; | |
414 | u64 ret; | |
415 | ||
ec108d3c | 416 | array = kmap_local_folio(folio, 0); |
1f1d4aa4 | 417 | ret = array->last_cookie; |
1683ed16 | 418 | kunmap_local(array); |
1f1d4aa4 TM |
419 | return ret; |
420 | } | |
421 | ||
ec108d3c | 422 | static bool nfs_readdir_folio_needs_filling(struct folio *folio) |
1f1d4aa4 TM |
423 | { |
424 | struct nfs_cache_array *array; | |
425 | bool ret; | |
426 | ||
ec108d3c | 427 | array = kmap_local_folio(folio, 0); |
1f1d4aa4 | 428 | ret = !nfs_readdir_array_is_full(array); |
1683ed16 | 429 | kunmap_local(array); |
4a201d6e | 430 | return ret; |
d1bacf9e BS |
431 | } |
432 | ||
ec108d3c | 433 | static void nfs_readdir_folio_set_eof(struct folio *folio) |
b1e21c97 TM |
434 | { |
435 | struct nfs_cache_array *array; | |
436 | ||
ec108d3c | 437 | array = kmap_local_folio(folio, 0); |
b1e21c97 | 438 | nfs_readdir_array_set_eof(array); |
1683ed16 | 439 | kunmap_local(array); |
b1e21c97 TM |
440 | } |
441 | ||
ec108d3c AS |
442 | static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping, |
443 | u64 cookie, u64 change_attr) | |
3b2a09f1 | 444 | { |
ec108d3c AS |
445 | pgoff_t index = nfs_readdir_folio_cookie_hash(cookie); |
446 | struct folio *folio; | |
3b2a09f1 | 447 | |
ec108d3c AS |
448 | folio = __filemap_get_folio(mapping, index, |
449 | FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT, | |
450 | mapping_gfp_mask(mapping)); | |
451 | if (!folio) | |
b0365ccb | 452 | return NULL; |
ec108d3c AS |
453 | nfs_readdir_folio_init_and_validate(folio, cookie, change_attr); |
454 | if (nfs_readdir_folio_last_cookie(folio) != cookie) | |
455 | nfs_readdir_folio_reinit_array(folio, cookie, change_attr); | |
456 | return folio; | |
3b2a09f1 TM |
457 | } |
458 | ||
59e356a9 TM |
459 | static inline |
460 | int is_32bit_api(void) | |
461 | { | |
462 | #ifdef CONFIG_COMPAT | |
463 | return in_compat_syscall(); | |
464 | #else | |
465 | return (BITS_PER_LONG == 32); | |
466 | #endif | |
467 | } | |
468 | ||
469 | static | |
470 | bool nfs_readdir_use_cookie(const struct file *filp) | |
471 | { | |
472 | if ((filp->f_mode & FMODE_32BITHASH) || | |
473 | (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) | |
474 | return false; | |
475 | return true; | |
476 | } | |
477 | ||
c8f0523b TM |
478 | static void nfs_readdir_seek_next_array(struct nfs_cache_array *array, |
479 | struct nfs_readdir_descriptor *desc) | |
480 | { | |
ec108d3c | 481 | if (array->folio_full) { |
c8f0523b TM |
482 | desc->last_cookie = array->last_cookie; |
483 | desc->current_index += array->size; | |
484 | desc->cache_entry_index = 0; | |
61f02e0a | 485 | desc->folio_index++; |
c8f0523b | 486 | } else |
e47a62df | 487 | desc->last_cookie = nfs_readdir_array_index_cookie(array); |
c8f0523b TM |
488 | } |
489 | ||
f648022f TM |
490 | static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc) |
491 | { | |
492 | desc->current_index = 0; | |
493 | desc->last_cookie = 0; | |
61f02e0a | 494 | desc->folio_index = 0; |
f648022f TM |
495 | } |
496 | ||
6c981eff TM |
497 | static int nfs_readdir_search_for_pos(struct nfs_cache_array *array, |
498 | struct nfs_readdir_descriptor *desc) | |
d1bacf9e | 499 | { |
23db8620 | 500 | loff_t diff = desc->ctx->pos - desc->current_index; |
d1bacf9e BS |
501 | unsigned int index; |
502 | ||
503 | if (diff < 0) | |
504 | goto out_eof; | |
505 | if (diff >= array->size) { | |
ec108d3c | 506 | if (array->folio_is_eof) |
d1bacf9e | 507 | goto out_eof; |
c8f0523b | 508 | nfs_readdir_seek_next_array(array, desc); |
d1bacf9e BS |
509 | return -EAGAIN; |
510 | } | |
511 | ||
512 | index = (unsigned int)diff; | |
2e7a4641 | 513 | desc->dir_cookie = array->array[index].cookie; |
d1bacf9e | 514 | desc->cache_entry_index = index; |
d1bacf9e BS |
515 | return 0; |
516 | out_eof: | |
6089dd0d | 517 | desc->eof = true; |
d1bacf9e BS |
518 | return -EBADCOOKIE; |
519 | } | |
520 | ||
762567b7 TM |
521 | static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array, |
522 | u64 cookie) | |
523 | { | |
524 | if (!array->cookies_are_ordered) | |
525 | return true; | |
526 | /* Optimisation for monotonically increasing cookies */ | |
527 | if (cookie >= array->last_cookie) | |
528 | return false; | |
529 | if (array->size && cookie < array->array[0].cookie) | |
530 | return false; | |
531 | return true; | |
532 | } | |
533 | ||
6c981eff TM |
534 | static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, |
535 | struct nfs_readdir_descriptor *desc) | |
d1bacf9e | 536 | { |
f648022f | 537 | unsigned int i; |
d1bacf9e BS |
538 | int status = -EAGAIN; |
539 | ||
762567b7 TM |
540 | if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie)) |
541 | goto check_eof; | |
542 | ||
d1bacf9e | 543 | for (i = 0; i < array->size; i++) { |
2e7a4641 | 544 | if (array->array[i].cookie == desc->dir_cookie) { |
59e356a9 | 545 | if (nfs_readdir_use_cookie(desc->file)) |
2e7a4641 | 546 | desc->ctx->pos = desc->dir_cookie; |
59e356a9 | 547 | else |
f648022f | 548 | desc->ctx->pos = desc->current_index + i; |
d1bacf9e | 549 | desc->cache_entry_index = i; |
47c716cb | 550 | return 0; |
d1bacf9e BS |
551 | } |
552 | } | |
762567b7 | 553 | check_eof: |
ec108d3c | 554 | if (array->folio_is_eof) { |
8cd51a0c | 555 | status = -EBADCOOKIE; |
2e7a4641 | 556 | if (desc->dir_cookie == array->last_cookie) |
6089dd0d | 557 | desc->eof = true; |
c8f0523b TM |
558 | } else |
559 | nfs_readdir_seek_next_array(array, desc); | |
d1bacf9e BS |
560 | return status; |
561 | } | |
562 | ||
6c981eff | 563 | static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc) |
d1bacf9e BS |
564 | { |
565 | struct nfs_cache_array *array; | |
47c716cb | 566 | int status; |
d1bacf9e | 567 | |
61f02e0a | 568 | array = kmap_local_folio(desc->folio, 0); |
d1bacf9e | 569 | |
2e7a4641 | 570 | if (desc->dir_cookie == 0) |
d1bacf9e BS |
571 | status = nfs_readdir_search_for_pos(array, desc); |
572 | else | |
573 | status = nfs_readdir_search_for_cookie(array, desc); | |
574 | ||
1683ed16 | 575 | kunmap_local(array); |
d1bacf9e BS |
576 | return status; |
577 | } | |
578 | ||
579 | /* Fill a page with xdr information before transferring to the cache page */ | |
93b8959a | 580 | static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc, |
b593c09f TM |
581 | __be32 *verf, u64 cookie, |
582 | struct page **pages, size_t bufsize, | |
583 | __be32 *verf_res) | |
1da177e4 | 584 | { |
82e22a5e | 585 | struct inode *inode = file_inode(desc->file); |
82e22a5e TM |
586 | struct nfs_readdir_arg arg = { |
587 | .dentry = file_dentry(desc->file), | |
588 | .cred = desc->file->f_cred, | |
b593c09f | 589 | .verf = verf, |
82e22a5e TM |
590 | .cookie = cookie, |
591 | .pages = pages, | |
592 | .page_len = bufsize, | |
593 | .plus = desc->plus, | |
594 | }; | |
595 | struct nfs_readdir_res res = { | |
596 | .verf = verf_res, | |
597 | }; | |
4704f0e2 | 598 | unsigned long timestamp, gencount; |
1da177e4 LT |
599 | int error; |
600 | ||
1da177e4 LT |
601 | again: |
602 | timestamp = jiffies; | |
4704f0e2 | 603 | gencount = nfs_inc_attr_generation_counter(); |
a1147b82 | 604 | desc->dir_verifier = nfs_save_change_attribute(inode); |
82e22a5e | 605 | error = NFS_PROTO(inode)->readdir(&arg, &res); |
1da177e4 LT |
606 | if (error < 0) { |
607 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
608 | if (error == -ENOTSUPP && desc->plus) { | |
609 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
82e22a5e | 610 | desc->plus = arg.plus = false; |
1da177e4 LT |
611 | goto again; |
612 | } | |
613 | goto error; | |
614 | } | |
1f4eab7e | 615 | desc->timestamp = timestamp; |
4704f0e2 | 616 | desc->gencount = gencount; |
d1bacf9e BS |
617 | error: |
618 | return error; | |
1da177e4 LT |
619 | } |
620 | ||
6c981eff | 621 | static int xdr_decode(struct nfs_readdir_descriptor *desc, |
573c4e1e | 622 | struct nfs_entry *entry, struct xdr_stream *xdr) |
1da177e4 | 623 | { |
59e356a9 | 624 | struct inode *inode = file_inode(desc->file); |
573c4e1e | 625 | int error; |
1da177e4 | 626 | |
59e356a9 | 627 | error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus); |
573c4e1e CL |
628 | if (error) |
629 | return error; | |
d1bacf9e BS |
630 | entry->fattr->time_start = desc->timestamp; |
631 | entry->fattr->gencount = desc->gencount; | |
632 | return 0; | |
1da177e4 LT |
633 | } |
634 | ||
fa923369 TM |
635 | /* Match file and dirent using either filehandle or fileid |
636 | * Note: caller is responsible for checking the fsid | |
637 | */ | |
d39ab9de BS |
638 | static |
639 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
640 | { | |
d8fdb47f | 641 | struct inode *inode; |
fa923369 TM |
642 | struct nfs_inode *nfsi; |
643 | ||
2b0143b5 DH |
644 | if (d_really_is_negative(dentry)) |
645 | return 0; | |
fa923369 | 646 | |
d8fdb47f TM |
647 | inode = d_inode(dentry); |
648 | if (is_bad_inode(inode) || NFS_STALE(inode)) | |
649 | return 0; | |
650 | ||
651 | nfsi = NFS_I(inode); | |
7dc72d5f TM |
652 | if (entry->fattr->fileid != nfsi->fileid) |
653 | return 0; | |
654 | if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) | |
655 | return 0; | |
656 | return 1; | |
d39ab9de BS |
657 | } |
658 | ||
230bc98f TM |
659 | #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL) |
660 | ||
661 | static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx, | |
662 | unsigned int cache_hits, | |
663 | unsigned int cache_misses) | |
d69ee9b8 TM |
664 | { |
665 | if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) | |
666 | return false; | |
230bc98f TM |
667 | if (ctx->pos == 0 || |
668 | cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD) | |
d69ee9b8 TM |
669 | return true; |
670 | return false; | |
671 | } | |
672 | ||
673 | /* | |
230bc98f | 674 | * This function is called by the getattr code to request the |
63519fbc | 675 | * use of readdirplus to accelerate any future lookups in the same |
d69ee9b8 TM |
676 | * directory. |
677 | */ | |
230bc98f | 678 | void nfs_readdir_record_entry_cache_hit(struct inode *dir) |
d69ee9b8 | 679 | { |
63519fbc | 680 | struct nfs_inode *nfsi = NFS_I(dir); |
230bc98f | 681 | struct nfs_open_dir_context *ctx; |
63519fbc TM |
682 | |
683 | if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && | |
230bc98f TM |
684 | S_ISDIR(dir->i_mode)) { |
685 | rcu_read_lock(); | |
686 | list_for_each_entry_rcu (ctx, &nfsi->open_files, list) | |
687 | atomic_inc(&ctx->cache_hits); | |
688 | rcu_read_unlock(); | |
689 | } | |
d69ee9b8 TM |
690 | } |
691 | ||
311324ad TM |
692 | /* |
693 | * This function is mainly for use by nfs_getattr(). | |
694 | * | |
695 | * If this is an 'ls -l', we want to force use of readdirplus. | |
311324ad | 696 | */ |
230bc98f | 697 | void nfs_readdir_record_entry_cache_miss(struct inode *dir) |
311324ad | 698 | { |
63519fbc | 699 | struct nfs_inode *nfsi = NFS_I(dir); |
230bc98f | 700 | struct nfs_open_dir_context *ctx; |
63519fbc TM |
701 | |
702 | if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && | |
230bc98f TM |
703 | S_ISDIR(dir->i_mode)) { |
704 | rcu_read_lock(); | |
705 | list_for_each_entry_rcu (ctx, &nfsi->open_files, list) | |
706 | atomic_inc(&ctx->cache_misses); | |
707 | rcu_read_unlock(); | |
311324ad TM |
708 | } |
709 | } | |
710 | ||
0b3cc71b TM |
711 | static void nfs_lookup_advise_force_readdirplus(struct inode *dir, |
712 | unsigned int flags) | |
230bc98f | 713 | { |
2c2c3365 TM |
714 | if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) |
715 | return; | |
0b3cc71b TM |
716 | if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL)) |
717 | return; | |
230bc98f TM |
718 | nfs_readdir_record_entry_cache_miss(dir); |
719 | } | |
720 | ||
d39ab9de | 721 | static |
a1147b82 TM |
722 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry, |
723 | unsigned long dir_verifier) | |
d39ab9de | 724 | { |
26fe5750 | 725 | struct qstr filename = QSTR_INIT(entry->name, entry->len); |
9ac3d3e8 | 726 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
4a201d6e TM |
727 | struct dentry *dentry; |
728 | struct dentry *alias; | |
d39ab9de | 729 | struct inode *inode; |
aa9c2669 | 730 | int status; |
d39ab9de | 731 | |
fa923369 TM |
732 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) |
733 | return; | |
6c441c25 TM |
734 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) |
735 | return; | |
78d04af4 TM |
736 | if (filename.len == 0) |
737 | return; | |
738 | /* Validate that the name doesn't contain any illegal '\0' */ | |
739 | if (strnlen(filename.name, filename.len) != filename.len) | |
740 | return; | |
741 | /* ...or '/' */ | |
742 | if (strnchr(filename.name, filename.len, '/')) | |
743 | return; | |
4a201d6e TM |
744 | if (filename.name[0] == '.') { |
745 | if (filename.len == 1) | |
746 | return; | |
747 | if (filename.len == 2 && filename.name[1] == '.') | |
748 | return; | |
749 | } | |
8387ff25 | 750 | filename.hash = full_name_hash(parent, filename.name, filename.len); |
d39ab9de | 751 | |
4a201d6e | 752 | dentry = d_lookup(parent, &filename); |
9ac3d3e8 AV |
753 | again: |
754 | if (!dentry) { | |
755 | dentry = d_alloc_parallel(parent, &filename, &wq); | |
756 | if (IS_ERR(dentry)) | |
757 | return; | |
758 | } | |
759 | if (!d_in_lookup(dentry)) { | |
6c441c25 TM |
760 | /* Is there a mountpoint here? If so, just exit */ |
761 | if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, | |
762 | &entry->fattr->fsid)) | |
763 | goto out; | |
d39ab9de | 764 | if (nfs_same_file(dentry, entry)) { |
7dc72d5f TM |
765 | if (!entry->fh->size) |
766 | goto out; | |
a1147b82 | 767 | nfs_set_verifier(dentry, dir_verifier); |
2b0143b5 | 768 | status = nfs_refresh_inode(d_inode(dentry), entry->fattr); |
aa9c2669 | 769 | if (!status) |
dd225cb3 | 770 | nfs_setsecurity(d_inode(dentry), entry->fattr); |
eace45a1 TM |
771 | trace_nfs_readdir_lookup_revalidate(d_inode(parent), |
772 | dentry, 0, status); | |
d39ab9de BS |
773 | goto out; |
774 | } else { | |
eace45a1 TM |
775 | trace_nfs_readdir_lookup_revalidate_failed( |
776 | d_inode(parent), dentry, 0); | |
5542aa2f | 777 | d_invalidate(dentry); |
d39ab9de | 778 | dput(dentry); |
9ac3d3e8 AV |
779 | dentry = NULL; |
780 | goto again; | |
d39ab9de BS |
781 | } |
782 | } | |
7dc72d5f TM |
783 | if (!entry->fh->size) { |
784 | d_lookup_done(dentry); | |
785 | goto out; | |
786 | } | |
d39ab9de | 787 | |
cf7ab00a | 788 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); |
41d28bca | 789 | alias = d_splice_alias(inode, dentry); |
9ac3d3e8 AV |
790 | d_lookup_done(dentry); |
791 | if (alias) { | |
792 | if (IS_ERR(alias)) | |
793 | goto out; | |
794 | dput(dentry); | |
795 | dentry = alias; | |
796 | } | |
a1147b82 | 797 | nfs_set_verifier(dentry, dir_verifier); |
eace45a1 | 798 | trace_nfs_readdir_lookup(d_inode(parent), dentry, 0); |
d39ab9de BS |
799 | out: |
800 | dput(dentry); | |
d39ab9de BS |
801 | } |
802 | ||
612896ec TM |
803 | static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc, |
804 | struct nfs_entry *entry, | |
805 | struct xdr_stream *stream) | |
806 | { | |
807 | int ret; | |
808 | ||
809 | if (entry->fattr->label) | |
810 | entry->fattr->label->len = NFS4_MAXLABELLEN; | |
811 | ret = xdr_decode(desc, entry, stream); | |
812 | if (ret || !desc->plus) | |
813 | return ret; | |
814 | nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier); | |
815 | return 0; | |
816 | } | |
817 | ||
d1bacf9e | 818 | /* Perform conversion from xdr to cache array */ |
61f02e0a AS |
819 | static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc, |
820 | struct nfs_entry *entry, | |
821 | struct page **xdr_pages, unsigned int buflen, | |
822 | struct folio **arrays, size_t narrays, | |
823 | u64 change_attr) | |
1da177e4 | 824 | { |
3b2a09f1 | 825 | struct address_space *mapping = desc->file->f_mapping; |
ec108d3c | 826 | struct folio *new, *folio = *arrays; |
babddc72 | 827 | struct xdr_stream stream; |
ec108d3c | 828 | struct page *scratch; |
f7da7a12 | 829 | struct xdr_buf buf; |
0adf85b4 | 830 | u64 cookie; |
5c346854 | 831 | int status; |
babddc72 | 832 | |
6650239a TM |
833 | scratch = alloc_page(GFP_KERNEL); |
834 | if (scratch == NULL) | |
835 | return -ENOMEM; | |
babddc72 | 836 | |
f7da7a12 | 837 | xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); |
0ae4c3e8 | 838 | xdr_set_scratch_page(&stream, scratch); |
99424380 BS |
839 | |
840 | do { | |
612896ec | 841 | status = nfs_readdir_entry_decode(desc, entry, &stream); |
972bcdf2 | 842 | if (status != 0) |
99424380 | 843 | break; |
5c346854 | 844 | |
ec108d3c | 845 | status = nfs_readdir_folio_array_append(folio, entry, &cookie); |
3b2a09f1 TM |
846 | if (status != -ENOSPC) |
847 | continue; | |
848 | ||
61f02e0a | 849 | if (folio->mapping != mapping) { |
35df59d3 TM |
850 | if (!--narrays) |
851 | break; | |
ec108d3c | 852 | new = nfs_readdir_folio_array_alloc(cookie, GFP_KERNEL); |
35df59d3 TM |
853 | if (!new) |
854 | break; | |
855 | arrays++; | |
ec108d3c | 856 | *arrays = folio = new; |
35df59d3 | 857 | } else { |
ec108d3c AS |
858 | new = nfs_readdir_folio_get_next(mapping, cookie, |
859 | change_attr); | |
35df59d3 TM |
860 | if (!new) |
861 | break; | |
61f02e0a | 862 | if (folio != *arrays) |
ec108d3c AS |
863 | nfs_readdir_folio_unlock_and_put(folio); |
864 | folio = new; | |
35df59d3 | 865 | } |
61f02e0a | 866 | desc->folio_index_max++; |
ec108d3c | 867 | status = nfs_readdir_folio_array_append(folio, entry, &cookie); |
972bcdf2 | 868 | } while (!status && !entry->eof); |
99424380 | 869 | |
972bcdf2 TM |
870 | switch (status) { |
871 | case -EBADCOOKIE: | |
612896ec TM |
872 | if (!entry->eof) |
873 | break; | |
ec108d3c | 874 | nfs_readdir_folio_set_eof(folio); |
612896ec | 875 | fallthrough; |
972bcdf2 | 876 | case -EAGAIN: |
0795bf83 | 877 | status = 0; |
972bcdf2 | 878 | break; |
612896ec TM |
879 | case -ENOSPC: |
880 | status = 0; | |
881 | if (!desc->plus) | |
882 | break; | |
883 | while (!nfs_readdir_entry_decode(desc, entry, &stream)) | |
884 | ; | |
1da177e4 | 885 | } |
6650239a | 886 | |
61f02e0a | 887 | if (folio != *arrays) |
ec108d3c | 888 | nfs_readdir_folio_unlock_and_put(folio); |
3b2a09f1 | 889 | |
6650239a | 890 | put_page(scratch); |
8cd51a0c | 891 | return status; |
56e4ebf8 BS |
892 | } |
893 | ||
1a34c8c9 | 894 | static void nfs_readdir_free_pages(struct page **pages, size_t npages) |
56e4ebf8 | 895 | { |
1a34c8c9 TM |
896 | while (npages--) |
897 | put_page(pages[npages]); | |
898 | kfree(pages); | |
56e4ebf8 BS |
899 | } |
900 | ||
56e4ebf8 | 901 | /* |
bf211ca1 | 902 | * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call |
903 | * to nfs_readdir_free_pages() | |
56e4ebf8 | 904 | */ |
1a34c8c9 | 905 | static struct page **nfs_readdir_alloc_pages(size_t npages) |
56e4ebf8 | 906 | { |
1a34c8c9 TM |
907 | struct page **pages; |
908 | size_t i; | |
56e4ebf8 | 909 | |
1a34c8c9 TM |
910 | pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL); |
911 | if (!pages) | |
912 | return NULL; | |
56e4ebf8 BS |
913 | for (i = 0; i < npages; i++) { |
914 | struct page *page = alloc_page(GFP_KERNEL); | |
915 | if (page == NULL) | |
916 | goto out_freepages; | |
917 | pages[i] = page; | |
918 | } | |
1a34c8c9 | 919 | return pages; |
56e4ebf8 | 920 | |
56e4ebf8 | 921 | out_freepages: |
c7e9668e | 922 | nfs_readdir_free_pages(pages, i); |
1a34c8c9 | 923 | return NULL; |
1da177e4 LT |
924 | } |
925 | ||
6c981eff | 926 | static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc, |
35df59d3 | 927 | __be32 *verf_arg, __be32 *verf_res, |
61f02e0a | 928 | struct folio **arrays, size_t narrays) |
00a92642 | 929 | { |
f648022f | 930 | u64 change_attr; |
1a34c8c9 | 931 | struct page **pages; |
61f02e0a | 932 | struct folio *folio = *arrays; |
6b75cf9e | 933 | struct nfs_entry *entry; |
1a34c8c9 | 934 | size_t array_size; |
b593c09f | 935 | struct inode *inode = file_inode(desc->file); |
580f2367 | 936 | unsigned int dtsize = desc->dtsize; |
9ff89c25 | 937 | unsigned int pglen; |
8cd51a0c | 938 | int status = -ENOMEM; |
d1bacf9e | 939 | |
6b75cf9e TM |
940 | entry = kzalloc(sizeof(*entry), GFP_KERNEL); |
941 | if (!entry) | |
942 | return -ENOMEM; | |
ec108d3c | 943 | entry->cookie = nfs_readdir_folio_last_cookie(folio); |
6b75cf9e | 944 | entry->fh = nfs_alloc_fhandle(); |
b1db9a40 | 945 | entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); |
6b75cf9e TM |
946 | entry->server = NFS_SERVER(inode); |
947 | if (entry->fh == NULL || entry->fattr == NULL) | |
d1bacf9e | 948 | goto out; |
00a92642 | 949 | |
1a34c8c9 TM |
950 | array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT; |
951 | pages = nfs_readdir_alloc_pages(array_size); | |
952 | if (!pages) | |
b1db9a40 | 953 | goto out; |
00a92642 | 954 | |
f648022f | 955 | change_attr = inode_peek_iversion_raw(inode); |
9ff89c25 TM |
956 | status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages, |
957 | dtsize, verf_res); | |
958 | if (status < 0) | |
959 | goto free_pages; | |
ee3707ae | 960 | |
9ff89c25 TM |
961 | pglen = status; |
962 | if (pglen != 0) | |
61f02e0a AS |
963 | status = nfs_readdir_folio_filler(desc, entry, pages, pglen, |
964 | arrays, narrays, change_attr); | |
9ff89c25 | 965 | else |
ec108d3c | 966 | nfs_readdir_folio_set_eof(folio); |
9ff89c25 | 967 | desc->buffer_fills++; |
d1bacf9e | 968 | |
9ff89c25 | 969 | free_pages: |
c7e9668e | 970 | nfs_readdir_free_pages(pages, array_size); |
d1bacf9e | 971 | out: |
6b75cf9e TM |
972 | nfs_free_fattr(entry->fattr); |
973 | nfs_free_fhandle(entry->fh); | |
974 | kfree(entry); | |
00a92642 OG |
975 | return status; |
976 | } | |
977 | ||
61f02e0a | 978 | static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc) |
1da177e4 | 979 | { |
61f02e0a AS |
980 | folio_put(desc->folio); |
981 | desc->folio = NULL; | |
d1bacf9e | 982 | } |
1da177e4 | 983 | |
1f1d4aa4 | 984 | static void |
61f02e0a | 985 | nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc) |
d1bacf9e | 986 | { |
61f02e0a AS |
987 | folio_unlock(desc->folio); |
988 | nfs_readdir_folio_put(desc); | |
d1bacf9e BS |
989 | } |
990 | ||
61f02e0a AS |
991 | static struct folio * |
992 | nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc) | |
d1bacf9e | 993 | { |
f648022f TM |
994 | struct address_space *mapping = desc->file->f_mapping; |
995 | u64 change_attr = inode_peek_iversion_raw(mapping->host); | |
b0365ccb | 996 | u64 cookie = desc->last_cookie; |
ec108d3c | 997 | struct folio *folio; |
f648022f | 998 | |
ec108d3c AS |
999 | folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr); |
1000 | if (!folio) | |
b0365ccb | 1001 | return NULL; |
ec108d3c AS |
1002 | if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio)) |
1003 | nfs_readdir_folio_reinit_array(folio, cookie, change_attr); | |
1004 | return folio; | |
1da177e4 LT |
1005 | } |
1006 | ||
1007 | /* | |
d1bacf9e | 1008 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
114de382 | 1009 | * and locks the page to prevent removal from the page cache. |
1da177e4 | 1010 | */ |
6c981eff | 1011 | static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc) |
d1bacf9e | 1012 | { |
227823d2 DN |
1013 | struct inode *inode = file_inode(desc->file); |
1014 | struct nfs_inode *nfsi = NFS_I(inode); | |
b593c09f | 1015 | __be32 verf[NFS_DIR_VERIFIER_SIZE]; |
d1bacf9e BS |
1016 | int res; |
1017 | ||
61f02e0a AS |
1018 | desc->folio = nfs_readdir_folio_get_cached(desc); |
1019 | if (!desc->folio) | |
1f1d4aa4 | 1020 | return -ENOMEM; |
ec108d3c | 1021 | if (nfs_readdir_folio_needs_filling(desc->folio)) { |
580f2367 | 1022 | /* Grow the dtsize if we had to go back for more pages */ |
61f02e0a | 1023 | if (desc->folio_index == desc->folio_index_max) |
580f2367 | 1024 | nfs_grow_dtsize(desc); |
61f02e0a | 1025 | desc->folio_index_max = desc->folio_index; |
310e3187 TM |
1026 | trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf, |
1027 | desc->last_cookie, | |
61f02e0a | 1028 | desc->folio->index, desc->dtsize); |
35df59d3 | 1029 | res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf, |
61f02e0a | 1030 | &desc->folio, 1); |
9fff59ed | 1031 | if (res < 0) { |
61f02e0a | 1032 | nfs_readdir_folio_unlock_and_put_cached(desc); |
310e3187 | 1033 | trace_nfs_readdir_cache_fill_done(inode, res); |
9fff59ed TM |
1034 | if (res == -EBADCOOKIE || res == -ENOTSYNC) { |
1035 | invalidate_inode_pages2(desc->file->f_mapping); | |
f648022f | 1036 | nfs_readdir_rewind_search(desc); |
11d03d0a TM |
1037 | trace_nfs_readdir_invalidate_cache_range( |
1038 | inode, 0, MAX_LFS_FILESIZE); | |
9fff59ed TM |
1039 | return -EAGAIN; |
1040 | } | |
1041 | return res; | |
227823d2 | 1042 | } |
f892c41c TM |
1043 | /* |
1044 | * Set the cookie verifier if the page cache was empty | |
1045 | */ | |
6c34f05b TM |
1046 | if (desc->last_cookie == 0 && |
1047 | memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) { | |
f892c41c TM |
1048 | memcpy(nfsi->cookieverf, verf, |
1049 | sizeof(nfsi->cookieverf)); | |
f648022f | 1050 | invalidate_inode_pages2_range(desc->file->f_mapping, 1, |
6c34f05b | 1051 | -1); |
11d03d0a | 1052 | trace_nfs_readdir_invalidate_cache_range( |
f648022f | 1053 | inode, 1, MAX_LFS_FILESIZE); |
6c34f05b | 1054 | } |
b0365ccb | 1055 | desc->clear_cache = false; |
114de382 | 1056 | } |
1f1d4aa4 | 1057 | res = nfs_readdir_search_array(desc); |
ff81dfb5 | 1058 | if (res == 0) |
1f1d4aa4 | 1059 | return 0; |
61f02e0a | 1060 | nfs_readdir_folio_unlock_and_put_cached(desc); |
d1bacf9e BS |
1061 | return res; |
1062 | } | |
1063 | ||
1064 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
6c981eff | 1065 | static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc) |
1da177e4 | 1066 | { |
8cd51a0c | 1067 | int res; |
d1bacf9e | 1068 | |
47c716cb | 1069 | do { |
114de382 | 1070 | res = find_and_lock_cache_page(desc); |
47c716cb | 1071 | } while (res == -EAGAIN); |
1da177e4 LT |
1072 | return res; |
1073 | } | |
1074 | ||
85aa8ddc BC |
1075 | #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL) |
1076 | ||
1da177e4 LT |
1077 | /* |
1078 | * Once we've found the start of the dirent within a page: fill 'er up... | |
1079 | */ | |
13884ff2 TM |
1080 | static void nfs_do_filldir(struct nfs_readdir_descriptor *desc, |
1081 | const __be32 *verf) | |
1da177e4 LT |
1082 | { |
1083 | struct file *file = desc->file; | |
dbeaf8c9 | 1084 | struct nfs_cache_array *array; |
c8f0523b | 1085 | unsigned int i; |
85aa8ddc | 1086 | bool first_emit = !desc->dir_cookie; |
8ef2ce3e | 1087 | |
61f02e0a | 1088 | array = kmap_local_folio(desc->folio, 0); |
d1bacf9e | 1089 | for (i = desc->cache_entry_index; i < array->size; i++) { |
ece0b423 | 1090 | struct nfs_cache_array_entry *ent; |
1da177e4 | 1091 | |
ece0b423 | 1092 | ent = &array->array[i]; |
a52a8a6a | 1093 | if (!dir_emit(desc->ctx, ent->name, ent->name_len, |
23db8620 | 1094 | nfs_compat_user_ino64(ent->ino), ent->d_type)) { |
e1d2699b | 1095 | desc->eob = true; |
1da177e4 | 1096 | break; |
ece0b423 | 1097 | } |
13884ff2 | 1098 | memcpy(desc->verf, verf, sizeof(desc->verf)); |
c8f0523b | 1099 | if (i == array->size - 1) { |
2e7a4641 | 1100 | desc->dir_cookie = array->last_cookie; |
c8f0523b TM |
1101 | nfs_readdir_seek_next_array(array, desc); |
1102 | } else { | |
1103 | desc->dir_cookie = array->array[i + 1].cookie; | |
1104 | desc->last_cookie = array->array[0].cookie; | |
1105 | } | |
59e356a9 | 1106 | if (nfs_readdir_use_cookie(file)) |
2e7a4641 | 1107 | desc->ctx->pos = desc->dir_cookie; |
59e356a9 TM |
1108 | else |
1109 | desc->ctx->pos++; | |
85aa8ddc BC |
1110 | if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 1) { |
1111 | desc->eob = true; | |
1112 | break; | |
1113 | } | |
1da177e4 | 1114 | } |
ec108d3c | 1115 | if (array->folio_is_eof) |
e1d2699b | 1116 | desc->eof = !desc->eob; |
d1bacf9e | 1117 | |
c77c738c | 1118 | kunmap_local(array); |
dbeaf8c9 TM |
1119 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n", |
1120 | (unsigned long long)desc->dir_cookie); | |
1da177e4 LT |
1121 | } |
1122 | ||
1123 | /* | |
1124 | * If we cannot find a cookie in our cache, we suspect that this is | |
1125 | * because it points to a deleted file, so we ask the server to return | |
1126 | * whatever it thinks is the next entry. We then feed this to filldir. | |
1127 | * If all goes well, we should then be able to find our way round the | |
1128 | * cache on the next call to readdir_search_pagecache(); | |
1129 | * | |
1130 | * NOTE: we cannot add the anonymous page to the pagecache because | |
1131 | * the data it contains might not be page aligned. Besides, | |
1132 | * we should already have a complete representation of the | |
1133 | * directory in the page cache by the time we get here. | |
1134 | */ | |
6c981eff | 1135 | static int uncached_readdir(struct nfs_readdir_descriptor *desc) |
1da177e4 | 1136 | { |
61f02e0a | 1137 | struct folio **arrays; |
35df59d3 | 1138 | size_t i, sz = 512; |
b593c09f | 1139 | __be32 verf[NFS_DIR_VERIFIER_SIZE]; |
35df59d3 | 1140 | int status = -ENOMEM; |
1da177e4 | 1141 | |
35df59d3 | 1142 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n", |
2e7a4641 | 1143 | (unsigned long long)desc->dir_cookie); |
1da177e4 | 1144 | |
35df59d3 TM |
1145 | arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL); |
1146 | if (!arrays) | |
1147 | goto out; | |
ec108d3c | 1148 | arrays[0] = nfs_readdir_folio_array_alloc(desc->dir_cookie, GFP_KERNEL); |
35df59d3 | 1149 | if (!arrays[0]) |
1da177e4 | 1150 | goto out; |
d1bacf9e | 1151 | |
61f02e0a | 1152 | desc->folio_index = 0; |
ce292d8f | 1153 | desc->cache_entry_index = 0; |
2e7a4641 | 1154 | desc->last_cookie = desc->dir_cookie; |
61f02e0a | 1155 | desc->folio_index_max = 0; |
7a8e1dc3 | 1156 | |
310e3187 TM |
1157 | trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie, |
1158 | -1, desc->dtsize); | |
1159 | ||
35df59d3 | 1160 | status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz); |
310e3187 TM |
1161 | if (status < 0) { |
1162 | trace_nfs_readdir_uncached_done(file_inode(desc->file), status); | |
1163 | goto out_free; | |
1164 | } | |
1da177e4 | 1165 | |
e1d2699b | 1166 | for (i = 0; !desc->eob && i < sz && arrays[i]; i++) { |
61f02e0a | 1167 | desc->folio = arrays[i]; |
13884ff2 | 1168 | nfs_do_filldir(desc, verf); |
35df59d3 | 1169 | } |
61f02e0a | 1170 | desc->folio = NULL; |
1da177e4 | 1171 | |
580f2367 TM |
1172 | /* |
1173 | * Grow the dtsize if we have to go back for more pages, | |
1174 | * or shrink it if we're reading too many. | |
1175 | */ | |
1176 | if (!desc->eof) { | |
1177 | if (!desc->eob) | |
1178 | nfs_grow_dtsize(desc); | |
1179 | else if (desc->buffer_fills == 1 && | |
61f02e0a | 1180 | i < (desc->folio_index_max >> 1)) |
580f2367 TM |
1181 | nfs_shrink_dtsize(desc); |
1182 | } | |
310e3187 | 1183 | out_free: |
35df59d3 | 1184 | for (i = 0; i < sz && arrays[i]; i++) |
ec108d3c | 1185 | nfs_readdir_folio_array_free(arrays[i]); |
35df59d3 | 1186 | out: |
f648022f TM |
1187 | if (!nfs_readdir_use_cookie(desc->file)) |
1188 | nfs_readdir_rewind_search(desc); | |
61f02e0a | 1189 | desc->folio_index_max = -1; |
35df59d3 TM |
1190 | kfree(arrays); |
1191 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status); | |
1da177e4 | 1192 | return status; |
1da177e4 LT |
1193 | } |
1194 | ||
b0365ccb | 1195 | static bool nfs_readdir_handle_cache_misses(struct inode *inode, |
230bc98f | 1196 | struct nfs_readdir_descriptor *desc, |
b0365ccb TM |
1197 | unsigned int cache_misses, |
1198 | bool force_clear) | |
230bc98f | 1199 | { |
b0365ccb TM |
1200 | if (desc->ctx->pos == 0 || !desc->plus) |
1201 | return false; | |
1202 | if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear) | |
1203 | return false; | |
1204 | trace_nfs_readdir_force_readdirplus(inode); | |
1205 | return true; | |
230bc98f TM |
1206 | } |
1207 | ||
00a92642 OG |
1208 | /* The file offset position represents the dirent entry number. A |
1209 | last cookie cache takes care of the common case of reading the | |
1210 | whole directory. | |
1da177e4 | 1211 | */ |
23db8620 | 1212 | static int nfs_readdir(struct file *file, struct dir_context *ctx) |
1da177e4 | 1213 | { |
be62a1a8 | 1214 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 1215 | struct inode *inode = d_inode(dentry); |
13884ff2 | 1216 | struct nfs_inode *nfsi = NFS_I(inode); |
23db8620 | 1217 | struct nfs_open_dir_context *dir_ctx = file->private_data; |
6b75cf9e | 1218 | struct nfs_readdir_descriptor *desc; |
230bc98f | 1219 | unsigned int cache_hits, cache_misses; |
b0365ccb | 1220 | bool force_clear; |
6b75cf9e | 1221 | int res; |
1da177e4 | 1222 | |
6de1472f AV |
1223 | dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", |
1224 | file, (long long)ctx->pos); | |
91d5b470 CL |
1225 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
1226 | ||
1da177e4 | 1227 | /* |
23db8620 | 1228 | * ctx->pos points to the dirent entry number. |
f0dd2136 | 1229 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
1230 | * to either find the entry with the appropriate number or |
1231 | * revalidate the cookie. | |
1da177e4 | 1232 | */ |
d09e673f | 1233 | nfs_revalidate_mapping(inode, file->f_mapping); |
6b75cf9e TM |
1234 | |
1235 | res = -ENOMEM; | |
1236 | desc = kzalloc(sizeof(*desc), GFP_KERNEL); | |
1237 | if (!desc) | |
fccca7fc | 1238 | goto out; |
6b75cf9e TM |
1239 | desc->file = file; |
1240 | desc->ctx = ctx; | |
61f02e0a | 1241 | desc->folio_index_max = -1; |
fccca7fc | 1242 | |
2e7a4641 TM |
1243 | spin_lock(&file->f_lock); |
1244 | desc->dir_cookie = dir_ctx->dir_cookie; | |
61f02e0a | 1245 | desc->folio_index = dir_ctx->page_index; |
728dd0ab | 1246 | desc->last_cookie = dir_ctx->last_cookie; |
2e7a4641 | 1247 | desc->attr_gencount = dir_ctx->attr_gencount; |
e1d2699b | 1248 | desc->eof = dir_ctx->eof; |
580f2367 | 1249 | nfs_set_dtsize(desc, dir_ctx->dtsize); |
b593c09f | 1250 | memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf)); |
230bc98f TM |
1251 | cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0); |
1252 | cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0); | |
b0365ccb | 1253 | force_clear = dir_ctx->force_clear; |
2e7a4641 | 1254 | spin_unlock(&file->f_lock); |
fccca7fc | 1255 | |
e1d2699b TM |
1256 | if (desc->eof) { |
1257 | res = 0; | |
1258 | goto out_free; | |
1259 | } | |
1260 | ||
230bc98f | 1261 | desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses); |
b0365ccb TM |
1262 | force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses, |
1263 | force_clear); | |
1264 | desc->clear_cache = force_clear; | |
ff81dfb5 | 1265 | |
47c716cb | 1266 | do { |
1da177e4 | 1267 | res = readdir_search_pagecache(desc); |
00a92642 | 1268 | |
1da177e4 | 1269 | if (res == -EBADCOOKIE) { |
ece0b423 | 1270 | res = 0; |
1da177e4 | 1271 | /* This means either end of directory */ |
2e7a4641 | 1272 | if (desc->dir_cookie && !desc->eof) { |
1da177e4 | 1273 | /* Or that the server has 'lost' a cookie */ |
23db8620 | 1274 | res = uncached_readdir(desc); |
ece0b423 | 1275 | if (res == 0) |
1da177e4 | 1276 | continue; |
9fff59ed TM |
1277 | if (res == -EBADCOOKIE || res == -ENOTSYNC) |
1278 | res = 0; | |
1da177e4 | 1279 | } |
1da177e4 LT |
1280 | break; |
1281 | } | |
1282 | if (res == -ETOOSMALL && desc->plus) { | |
1da177e4 | 1283 | nfs_zap_caches(inode); |
a7a3b1e9 BC |
1284 | desc->plus = false; |
1285 | desc->eof = false; | |
1da177e4 LT |
1286 | continue; |
1287 | } | |
1288 | if (res < 0) | |
1289 | break; | |
1290 | ||
13884ff2 | 1291 | nfs_do_filldir(desc, nfsi->cookieverf); |
61f02e0a AS |
1292 | nfs_readdir_folio_unlock_and_put_cached(desc); |
1293 | if (desc->folio_index == desc->folio_index_max) | |
b0365ccb | 1294 | desc->clear_cache = force_clear; |
e1d2699b | 1295 | } while (!desc->eob && !desc->eof); |
2e7a4641 TM |
1296 | |
1297 | spin_lock(&file->f_lock); | |
1298 | dir_ctx->dir_cookie = desc->dir_cookie; | |
728dd0ab | 1299 | dir_ctx->last_cookie = desc->last_cookie; |
2e7a4641 | 1300 | dir_ctx->attr_gencount = desc->attr_gencount; |
61f02e0a | 1301 | dir_ctx->page_index = desc->folio_index; |
b0365ccb | 1302 | dir_ctx->force_clear = force_clear; |
e1d2699b | 1303 | dir_ctx->eof = desc->eof; |
580f2367 | 1304 | dir_ctx->dtsize = desc->dtsize; |
b593c09f | 1305 | memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf)); |
2e7a4641 | 1306 | spin_unlock(&file->f_lock); |
e1d2699b | 1307 | out_free: |
6b75cf9e TM |
1308 | kfree(desc); |
1309 | ||
fccca7fc | 1310 | out: |
6de1472f | 1311 | dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); |
1e7cb3dc | 1312 | return res; |
1da177e4 LT |
1313 | } |
1314 | ||
965c8e59 | 1315 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) |
f0dd2136 | 1316 | { |
480c2006 | 1317 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
b84e06c5 | 1318 | |
6de1472f AV |
1319 | dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", |
1320 | filp, offset, whence); | |
b84e06c5 | 1321 | |
965c8e59 | 1322 | switch (whence) { |
b2b1ff3d TM |
1323 | default: |
1324 | return -EINVAL; | |
1325 | case SEEK_SET: | |
1326 | if (offset < 0) | |
1327 | return -EINVAL; | |
83f2c45e | 1328 | spin_lock(&filp->f_lock); |
b2b1ff3d TM |
1329 | break; |
1330 | case SEEK_CUR: | |
1331 | if (offset == 0) | |
1332 | return filp->f_pos; | |
83f2c45e | 1333 | spin_lock(&filp->f_lock); |
b2b1ff3d TM |
1334 | offset += filp->f_pos; |
1335 | if (offset < 0) { | |
83f2c45e | 1336 | spin_unlock(&filp->f_lock); |
b2b1ff3d TM |
1337 | return -EINVAL; |
1338 | } | |
f0dd2136 TM |
1339 | } |
1340 | if (offset != filp->f_pos) { | |
1341 | filp->f_pos = offset; | |
9c3f4d98 | 1342 | dir_ctx->page_index = 0; |
f648022f | 1343 | if (!nfs_readdir_use_cookie(filp)) { |
59e356a9 | 1344 | dir_ctx->dir_cookie = 0; |
f648022f TM |
1345 | dir_ctx->last_cookie = 0; |
1346 | } else { | |
728dd0ab | 1347 | dir_ctx->dir_cookie = offset; |
f648022f TM |
1348 | dir_ctx->last_cookie = offset; |
1349 | } | |
e1d2699b | 1350 | dir_ctx->eof = false; |
f0dd2136 | 1351 | } |
83f2c45e | 1352 | spin_unlock(&filp->f_lock); |
f0dd2136 TM |
1353 | return offset; |
1354 | } | |
1355 | ||
1da177e4 LT |
1356 | /* |
1357 | * All directory operations under NFS are synchronous, so fsync() | |
1358 | * is a dummy operation. | |
1359 | */ | |
02c24a82 JB |
1360 | static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, |
1361 | int datasync) | |
1da177e4 | 1362 | { |
6de1472f | 1363 | dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); |
1e7cb3dc | 1364 | |
11decaf8 | 1365 | nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC); |
1da177e4 LT |
1366 | return 0; |
1367 | } | |
1368 | ||
bfc69a45 TM |
1369 | /** |
1370 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
302fad7b | 1371 | * @dir: pointer to directory inode |
bfc69a45 TM |
1372 | * |
1373 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
1374 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
1375 | * on the server that might have invalidated our dcache. | |
1376 | * | |
efeda80d TM |
1377 | * Note that we reserve bit '0' as a tag to let us know when a dentry |
1378 | * was revalidated while holding a delegation on its inode. | |
1379 | * | |
bfc69a45 TM |
1380 | * The caller should be holding dir->i_lock |
1381 | */ | |
1382 | void nfs_force_lookup_revalidate(struct inode *dir) | |
1383 | { | |
efeda80d | 1384 | NFS_I(dir)->cache_change_attribute += 2; |
bfc69a45 | 1385 | } |
89d77c8f | 1386 | EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); |
bfc69a45 | 1387 | |
efeda80d TM |
1388 | /** |
1389 | * nfs_verify_change_attribute - Detects NFS remote directory changes | |
1390 | * @dir: pointer to parent directory inode | |
1391 | * @verf: previously saved change attribute | |
1392 | * | |
1393 | * Return "false" if the verifiers doesn't match the change attribute. | |
1394 | * This would usually indicate that the directory contents have changed on | |
1395 | * the server, and that any dentries need revalidating. | |
1396 | */ | |
1397 | static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf) | |
1398 | { | |
1399 | return (verf & ~1UL) == nfs_save_change_attribute(dir); | |
1400 | } | |
1401 | ||
1402 | static void nfs_set_verifier_delegated(unsigned long *verf) | |
1403 | { | |
1404 | *verf |= 1UL; | |
1405 | } | |
1406 | ||
1407 | #if IS_ENABLED(CONFIG_NFS_V4) | |
1408 | static void nfs_unset_verifier_delegated(unsigned long *verf) | |
1409 | { | |
1410 | *verf &= ~1UL; | |
1411 | } | |
1412 | #endif /* IS_ENABLED(CONFIG_NFS_V4) */ | |
1413 | ||
1414 | static bool nfs_test_verifier_delegated(unsigned long verf) | |
1415 | { | |
1416 | return verf & 1; | |
1417 | } | |
1418 | ||
1419 | static bool nfs_verifier_is_delegated(struct dentry *dentry) | |
1420 | { | |
1421 | return nfs_test_verifier_delegated(dentry->d_time); | |
1422 | } | |
1423 | ||
1424 | static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf) | |
1425 | { | |
1426 | struct inode *inode = d_inode(dentry); | |
cec08f45 | 1427 | struct inode *dir = d_inode(dentry->d_parent); |
efeda80d | 1428 | |
cec08f45 TM |
1429 | if (!nfs_verify_change_attribute(dir, verf)) |
1430 | return; | |
efeda80d TM |
1431 | if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
1432 | nfs_set_verifier_delegated(&verf); | |
efeda80d TM |
1433 | dentry->d_time = verf; |
1434 | } | |
1435 | ||
1436 | /** | |
1437 | * nfs_set_verifier - save a parent directory verifier in the dentry | |
1438 | * @dentry: pointer to dentry | |
1439 | * @verf: verifier to save | |
1440 | * | |
1441 | * Saves the parent directory verifier in @dentry. If the inode has | |
1442 | * a delegation, we also tag the dentry as having been revalidated | |
1443 | * while holding a delegation so that we know we don't have to | |
1444 | * look it up again after a directory change. | |
1445 | */ | |
1446 | void nfs_set_verifier(struct dentry *dentry, unsigned long verf) | |
1447 | { | |
1448 | ||
1449 | spin_lock(&dentry->d_lock); | |
1450 | nfs_set_verifier_locked(dentry, verf); | |
1451 | spin_unlock(&dentry->d_lock); | |
1452 | } | |
1453 | EXPORT_SYMBOL_GPL(nfs_set_verifier); | |
1454 | ||
1455 | #if IS_ENABLED(CONFIG_NFS_V4) | |
1456 | /** | |
1457 | * nfs_clear_verifier_delegated - clear the dir verifier delegation tag | |
1458 | * @inode: pointer to inode | |
1459 | * | |
1460 | * Iterates through the dentries in the inode alias list and clears | |
1461 | * the tag used to indicate that the dentry has been revalidated | |
1462 | * while holding a delegation. | |
1463 | * This function is intended for use when the delegation is being | |
1464 | * returned or revoked. | |
1465 | */ | |
1466 | void nfs_clear_verifier_delegated(struct inode *inode) | |
1467 | { | |
1468 | struct dentry *alias; | |
1469 | ||
1470 | if (!inode) | |
1471 | return; | |
1472 | spin_lock(&inode->i_lock); | |
1473 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
1474 | spin_lock(&alias->d_lock); | |
1475 | nfs_unset_verifier_delegated(&alias->d_time); | |
1476 | spin_unlock(&alias->d_lock); | |
1477 | } | |
1478 | spin_unlock(&inode->i_lock); | |
1479 | } | |
1480 | EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated); | |
1481 | #endif /* IS_ENABLED(CONFIG_NFS_V4) */ | |
1482 | ||
8ce37abd TM |
1483 | static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry) |
1484 | { | |
1485 | if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) && | |
1486 | d_really_is_negative(dentry)) | |
1487 | return dentry->d_time == inode_peek_iversion_raw(dir); | |
1488 | return nfs_verify_change_attribute(dir, dentry->d_time); | |
1489 | } | |
1490 | ||
1da177e4 LT |
1491 | /* |
1492 | * A check for whether or not the parent directory has changed. | |
1493 | * In the case it has, we assume that the dentries are untrustworthy | |
1494 | * and may need to be looked up again. | |
912a108d | 1495 | * If rcu_walk prevents us from performing a full check, return 0. |
1da177e4 | 1496 | */ |
912a108d N |
1497 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, |
1498 | int rcu_walk) | |
1da177e4 LT |
1499 | { |
1500 | if (IS_ROOT(dentry)) | |
1501 | return 1; | |
4eec952e TM |
1502 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
1503 | return 0; | |
8ce37abd | 1504 | if (!nfs_dentry_verify_change(dir, dentry)) |
f2c77f4e TM |
1505 | return 0; |
1506 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
1cd9cb05 TM |
1507 | if (nfs_mapping_need_revalidate_inode(dir)) { |
1508 | if (rcu_walk) | |
1509 | return 0; | |
1510 | if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
1511 | return 0; | |
1512 | } | |
8ce37abd | 1513 | if (!nfs_dentry_verify_change(dir, dentry)) |
f2c77f4e TM |
1514 | return 0; |
1515 | return 1; | |
1da177e4 LT |
1516 | } |
1517 | ||
a12802ca TM |
1518 | /* |
1519 | * Use intent information to check whether or not we're going to do | |
1520 | * an O_EXCL create using this path component. | |
1521 | */ | |
fa3c56bb | 1522 | static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) |
a12802ca TM |
1523 | { |
1524 | if (NFS_PROTO(dir)->version == 2) | |
1525 | return 0; | |
fa3c56bb | 1526 | return flags & LOOKUP_EXCL; |
a12802ca TM |
1527 | } |
1528 | ||
1d6757fb TM |
1529 | /* |
1530 | * Inode and filehandle revalidation for lookups. | |
1531 | * | |
1532 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
1533 | * or if the intent information indicates that we're about to open this | |
1534 | * particular file and the "nocto" mount flag is not set. | |
1535 | * | |
1536 | */ | |
65a0c149 | 1537 | static |
fa3c56bb | 1538 | int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) |
1da177e4 LT |
1539 | { |
1540 | struct nfs_server *server = NFS_SERVER(inode); | |
65a0c149 | 1541 | int ret; |
1da177e4 | 1542 | |
36d43a43 | 1543 | if (IS_AUTOMOUNT(inode)) |
4e99a1ff | 1544 | return 0; |
47921921 TM |
1545 | |
1546 | if (flags & LOOKUP_OPEN) { | |
1547 | switch (inode->i_mode & S_IFMT) { | |
1548 | case S_IFREG: | |
1549 | /* A NFSv4 OPEN will revalidate later */ | |
1550 | if (server->caps & NFS_CAP_ATOMIC_OPEN) | |
1551 | goto out; | |
df561f66 | 1552 | fallthrough; |
47921921 TM |
1553 | case S_IFDIR: |
1554 | if (server->flags & NFS_MOUNT_NOCTO) | |
1555 | break; | |
1556 | /* NFS close-to-open cache consistency validation */ | |
1557 | goto out_force; | |
1558 | } | |
1559 | } | |
1560 | ||
facc3530 | 1561 | /* VFS wants an on-the-wire revalidation */ |
fa3c56bb | 1562 | if (flags & LOOKUP_REVAL) |
facc3530 | 1563 | goto out_force; |
65a0c149 | 1564 | out: |
43245eca OK |
1565 | if (inode->i_nlink > 0 || |
1566 | (inode->i_nlink == 0 && | |
1567 | test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags))) | |
1568 | return 0; | |
1569 | else | |
1570 | return -ESTALE; | |
1da177e4 | 1571 | out_force: |
1fa1e384 N |
1572 | if (flags & LOOKUP_RCU) |
1573 | return -ECHILD; | |
65a0c149 TM |
1574 | ret = __nfs_revalidate_inode(server, inode); |
1575 | if (ret != 0) | |
1576 | return ret; | |
1577 | goto out; | |
1da177e4 LT |
1578 | } |
1579 | ||
82e7ca13 TM |
1580 | static void nfs_mark_dir_for_revalidate(struct inode *inode) |
1581 | { | |
82e7ca13 | 1582 | spin_lock(&inode->i_lock); |
a6a361c4 | 1583 | nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); |
82e7ca13 TM |
1584 | spin_unlock(&inode->i_lock); |
1585 | } | |
1586 | ||
1da177e4 LT |
1587 | /* |
1588 | * We judge how long we want to trust negative | |
1589 | * dentries by looking at the parent inode mtime. | |
1590 | * | |
1591 | * If parent mtime has changed, we revalidate, else we wait for a | |
1592 | * period corresponding to the parent's attribute cache timeout value. | |
912a108d N |
1593 | * |
1594 | * If LOOKUP_RCU prevents us from performing a full check, return 1 | |
1595 | * suggesting a reval is needed. | |
9f6d44d4 TM |
1596 | * |
1597 | * Note that when creating a new file, or looking up a rename target, | |
1598 | * then it shouldn't be necessary to revalidate a negative dentry. | |
1da177e4 LT |
1599 | */ |
1600 | static inline | |
1601 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
fa3c56bb | 1602 | unsigned int flags) |
1da177e4 | 1603 | { |
9f6d44d4 | 1604 | if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) |
1da177e4 | 1605 | return 0; |
4eec952e TM |
1606 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1607 | return 1; | |
98ca3ee6 TM |
1608 | /* Case insensitive server? Revalidate negative dentries */ |
1609 | if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) | |
1610 | return 1; | |
912a108d | 1611 | return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); |
1da177e4 LT |
1612 | } |
1613 | ||
5ceb9d7f TM |
1614 | static int |
1615 | nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry, | |
1616 | struct inode *inode, int error) | |
1617 | { | |
1618 | switch (error) { | |
1619 | case 1: | |
2eef8a31 | 1620 | break; |
5ceb9d7f | 1621 | case 0: |
47397915 TM |
1622 | /* |
1623 | * We can't d_drop the root of a disconnected tree: | |
1624 | * its d_hash is on the s_anon list and d_drop() would hide | |
1625 | * it from shrink_dcache_for_unmount(), leading to busy | |
1626 | * inodes on unmount and further oopses. | |
1627 | */ | |
1628 | if (inode && IS_ROOT(dentry)) | |
2eef8a31 TM |
1629 | error = 1; |
1630 | break; | |
5ceb9d7f | 1631 | } |
2eef8a31 | 1632 | trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error); |
5ceb9d7f TM |
1633 | return error; |
1634 | } | |
1635 | ||
1636 | static int | |
1637 | nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry, | |
1638 | unsigned int flags) | |
1639 | { | |
1640 | int ret = 1; | |
1641 | if (nfs_neg_need_reval(dir, dentry, flags)) { | |
1642 | if (flags & LOOKUP_RCU) | |
1643 | return -ECHILD; | |
1644 | ret = 0; | |
1645 | } | |
1646 | return nfs_lookup_revalidate_done(dir, dentry, NULL, ret); | |
1647 | } | |
1648 | ||
1649 | static int | |
1650 | nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry, | |
1651 | struct inode *inode) | |
1652 | { | |
1653 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1654 | return nfs_lookup_revalidate_done(dir, dentry, inode, 1); | |
1655 | } | |
1656 | ||
0b3cc71b TM |
1657 | static int nfs_lookup_revalidate_dentry(struct inode *dir, |
1658 | struct dentry *dentry, | |
1659 | struct inode *inode, unsigned int flags) | |
5ceb9d7f TM |
1660 | { |
1661 | struct nfs_fh *fhandle; | |
1662 | struct nfs_fattr *fattr; | |
a1147b82 | 1663 | unsigned long dir_verifier; |
5ceb9d7f TM |
1664 | int ret; |
1665 | ||
0b3cc71b TM |
1666 | trace_nfs_lookup_revalidate_enter(dir, dentry, flags); |
1667 | ||
5ceb9d7f TM |
1668 | ret = -ENOMEM; |
1669 | fhandle = nfs_alloc_fhandle(); | |
9558a007 AS |
1670 | fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); |
1671 | if (fhandle == NULL || fattr == NULL) | |
5ceb9d7f TM |
1672 | goto out; |
1673 | ||
a1147b82 | 1674 | dir_verifier = nfs_save_change_attribute(dir); |
9558a007 | 1675 | ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); |
5ceb9d7f | 1676 | if (ret < 0) { |
f7b37b8b TM |
1677 | switch (ret) { |
1678 | case -ESTALE: | |
1679 | case -ENOENT: | |
5ceb9d7f | 1680 | ret = 0; |
f7b37b8b TM |
1681 | break; |
1682 | case -ETIMEDOUT: | |
1683 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) | |
1684 | ret = 1; | |
1685 | } | |
5ceb9d7f TM |
1686 | goto out; |
1687 | } | |
0b3cc71b TM |
1688 | |
1689 | /* Request help from readdirplus */ | |
1690 | nfs_lookup_advise_force_readdirplus(dir, flags); | |
1691 | ||
5ceb9d7f TM |
1692 | ret = 0; |
1693 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) | |
1694 | goto out; | |
1695 | if (nfs_refresh_inode(inode, fattr) < 0) | |
1696 | goto out; | |
1697 | ||
dd225cb3 | 1698 | nfs_setsecurity(inode, fattr); |
a1147b82 | 1699 | nfs_set_verifier(dentry, dir_verifier); |
5ceb9d7f | 1700 | |
5ceb9d7f TM |
1701 | ret = 1; |
1702 | out: | |
1703 | nfs_free_fattr(fattr); | |
1704 | nfs_free_fhandle(fhandle); | |
82e7ca13 TM |
1705 | |
1706 | /* | |
1707 | * If the lookup failed despite the dentry change attribute being | |
1708 | * a match, then we should revalidate the directory cache. | |
1709 | */ | |
8ce37abd | 1710 | if (!ret && nfs_dentry_verify_change(dir, dentry)) |
82e7ca13 | 1711 | nfs_mark_dir_for_revalidate(dir); |
5ceb9d7f TM |
1712 | return nfs_lookup_revalidate_done(dir, dentry, inode, ret); |
1713 | } | |
1714 | ||
1da177e4 LT |
1715 | /* |
1716 | * This is called every time the dcache has a lookup hit, | |
1717 | * and we should check whether we can really trust that | |
1718 | * lookup. | |
1719 | * | |
1720 | * NOTE! The hit can be a negative hit too, don't assume | |
1721 | * we have an inode! | |
1722 | * | |
1723 | * If the parent directory is seen to have changed, we throw out the | |
1724 | * cached dentry and do a new lookup. | |
1725 | */ | |
5ceb9d7f TM |
1726 | static int |
1727 | nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, | |
1728 | unsigned int flags) | |
1da177e4 | 1729 | { |
1da177e4 | 1730 | struct inode *inode; |
1da177e4 | 1731 | int error; |
1da177e4 | 1732 | |
91d5b470 | 1733 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
2b0143b5 | 1734 | inode = d_inode(dentry); |
1da177e4 | 1735 | |
5ceb9d7f TM |
1736 | if (!inode) |
1737 | return nfs_lookup_revalidate_negative(dir, dentry, flags); | |
1da177e4 LT |
1738 | |
1739 | if (is_bad_inode(inode)) { | |
6de1472f AV |
1740 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1741 | __func__, dentry); | |
1da177e4 LT |
1742 | goto out_bad; |
1743 | } | |
1744 | ||
6ca0a6f8 TM |
1745 | if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 && |
1746 | nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) | |
1747 | goto out_bad; | |
1748 | ||
efeda80d | 1749 | if (nfs_verifier_is_delegated(dentry)) |
5ceb9d7f | 1750 | return nfs_lookup_revalidate_delegated(dir, dentry, inode); |
15860ab1 | 1751 | |
1da177e4 | 1752 | /* Force a full look up iff the parent directory has changed */ |
73dd684a | 1753 | if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && |
912a108d | 1754 | nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { |
cc89684c N |
1755 | error = nfs_lookup_verify_inode(inode, flags); |
1756 | if (error) { | |
cc89684c | 1757 | if (error == -ESTALE) |
82e7ca13 | 1758 | nfs_mark_dir_for_revalidate(dir); |
5ceb9d7f | 1759 | goto out_bad; |
1fa1e384 | 1760 | } |
1da177e4 LT |
1761 | goto out_valid; |
1762 | } | |
1763 | ||
912a108d N |
1764 | if (flags & LOOKUP_RCU) |
1765 | return -ECHILD; | |
1766 | ||
1da177e4 LT |
1767 | if (NFS_STALE(inode)) |
1768 | goto out_bad; | |
1769 | ||
0b3cc71b | 1770 | return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags); |
5ceb9d7f TM |
1771 | out_valid: |
1772 | return nfs_lookup_revalidate_done(dir, dentry, inode, 1); | |
1773 | out_bad: | |
1774 | if (flags & LOOKUP_RCU) | |
1775 | return -ECHILD; | |
1776 | return nfs_lookup_revalidate_done(dir, dentry, inode, 0); | |
1777 | } | |
14c43f76 | 1778 | |
5ceb9d7f | 1779 | static int |
c7944ebb TM |
1780 | __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, |
1781 | int (*reval)(struct inode *, struct dentry *, unsigned int)) | |
5ceb9d7f TM |
1782 | { |
1783 | struct dentry *parent; | |
1784 | struct inode *dir; | |
1785 | int ret; | |
63519fbc | 1786 | |
d51ac1a8 | 1787 | if (flags & LOOKUP_RCU) { |
3c59366c N |
1788 | if (dentry->d_fsdata == NFS_FSDATA_BLOCKED) |
1789 | return -ECHILD; | |
5ceb9d7f TM |
1790 | parent = READ_ONCE(dentry->d_parent); |
1791 | dir = d_inode_rcu(parent); | |
1792 | if (!dir) | |
1793 | return -ECHILD; | |
c7944ebb | 1794 | ret = reval(dir, dentry, flags); |
6aa7de05 | 1795 | if (parent != READ_ONCE(dentry->d_parent)) |
d51ac1a8 | 1796 | return -ECHILD; |
5ceb9d7f | 1797 | } else { |
3c59366c N |
1798 | /* Wait for unlink to complete */ |
1799 | wait_var_event(&dentry->d_fsdata, | |
1800 | dentry->d_fsdata != NFS_FSDATA_BLOCKED); | |
5ceb9d7f | 1801 | parent = dget_parent(dentry); |
c7944ebb | 1802 | ret = reval(d_inode(parent), dentry, flags); |
d51ac1a8 | 1803 | dput(parent); |
1da177e4 | 1804 | } |
5ceb9d7f | 1805 | return ret; |
1da177e4 LT |
1806 | } |
1807 | ||
c7944ebb TM |
1808 | static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
1809 | { | |
1810 | return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); | |
1811 | } | |
1812 | ||
ecf3d1f1 | 1813 | /* |
2b0143b5 | 1814 | * A weaker form of d_revalidate for revalidating just the d_inode(dentry) |
ecf3d1f1 JL |
1815 | * when we don't really care about the dentry name. This is called when a |
1816 | * pathwalk ends on a dentry that was not found via a normal lookup in the | |
1817 | * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). | |
1818 | * | |
1819 | * In this situation, we just want to verify that the inode itself is OK | |
1820 | * since the dentry might have changed on the server. | |
1821 | */ | |
1822 | static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) | |
1823 | { | |
2b0143b5 | 1824 | struct inode *inode = d_inode(dentry); |
9cdd1d3f | 1825 | int error = 0; |
ecf3d1f1 JL |
1826 | |
1827 | /* | |
1828 | * I believe we can only get a negative dentry here in the case of a | |
1829 | * procfs-style symlink. Just assume it's correct for now, but we may | |
1830 | * eventually need to do something more here. | |
1831 | */ | |
1832 | if (!inode) { | |
6de1472f AV |
1833 | dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", |
1834 | __func__, dentry); | |
ecf3d1f1 JL |
1835 | return 1; |
1836 | } | |
1837 | ||
1838 | if (is_bad_inode(inode)) { | |
6de1472f AV |
1839 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1840 | __func__, dentry); | |
ecf3d1f1 JL |
1841 | return 0; |
1842 | } | |
1843 | ||
b688741c | 1844 | error = nfs_lookup_verify_inode(inode, flags); |
ecf3d1f1 JL |
1845 | dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", |
1846 | __func__, inode->i_ino, error ? "invalid" : "valid"); | |
1847 | return !error; | |
1848 | } | |
1849 | ||
1da177e4 LT |
1850 | /* |
1851 | * This is called from dput() when d_count is going to 0. | |
1852 | */ | |
fe15ce44 | 1853 | static int nfs_dentry_delete(const struct dentry *dentry) |
1da177e4 | 1854 | { |
6de1472f AV |
1855 | dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", |
1856 | dentry, dentry->d_flags); | |
1da177e4 | 1857 | |
77f11192 | 1858 | /* Unhash any dentry with a stale inode */ |
2b0143b5 | 1859 | if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) |
77f11192 TM |
1860 | return 1; |
1861 | ||
1da177e4 LT |
1862 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1863 | /* Unhash it, so that ->d_iput() would be called */ | |
1864 | return 1; | |
1865 | } | |
1751e8a6 | 1866 | if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { |
1da177e4 LT |
1867 | /* Unhash it, so that ancestors of killed async unlink |
1868 | * files will be cleaned up during umount */ | |
1869 | return 1; | |
1870 | } | |
1871 | return 0; | |
1872 | ||
1873 | } | |
1874 | ||
1f018458 | 1875 | /* Ensure that we revalidate inode->i_nlink */ |
1b83d707 TM |
1876 | static void nfs_drop_nlink(struct inode *inode) |
1877 | { | |
1878 | spin_lock(&inode->i_lock); | |
1f018458 | 1879 | /* drop the inode if we're reasonably sure this is the last link */ |
59a707b0 TM |
1880 | if (inode->i_nlink > 0) |
1881 | drop_nlink(inode); | |
1882 | NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); | |
ac46b3d7 TM |
1883 | nfs_set_cache_invalid( |
1884 | inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | | |
1301e421 | 1885 | NFS_INO_INVALID_NLINK); |
1b83d707 TM |
1886 | spin_unlock(&inode->i_lock); |
1887 | } | |
1888 | ||
1da177e4 LT |
1889 | /* |
1890 | * Called when the dentry loses inode. | |
1891 | * We use it to clean up silly-renamed files. | |
1892 | */ | |
1893 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1894 | { | |
1895 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
e4eff1a6 | 1896 | nfs_complete_unlink(dentry, inode); |
1f018458 | 1897 | nfs_drop_nlink(inode); |
1da177e4 | 1898 | } |
1da177e4 LT |
1899 | iput(inode); |
1900 | } | |
1901 | ||
b1942c5f AV |
1902 | static void nfs_d_release(struct dentry *dentry) |
1903 | { | |
1904 | /* free cached devname value, if it survived that far */ | |
1905 | if (unlikely(dentry->d_fsdata)) { | |
1906 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1907 | WARN_ON(1); | |
1908 | else | |
1909 | kfree(dentry->d_fsdata); | |
1910 | } | |
1911 | } | |
1912 | ||
f786aa90 | 1913 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 | 1914 | .d_revalidate = nfs_lookup_revalidate, |
ecf3d1f1 | 1915 | .d_weak_revalidate = nfs_weak_revalidate, |
1da177e4 LT |
1916 | .d_delete = nfs_dentry_delete, |
1917 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1918 | .d_automount = nfs_d_automount, |
b1942c5f | 1919 | .d_release = nfs_d_release, |
1da177e4 | 1920 | }; |
ddda8e0a | 1921 | EXPORT_SYMBOL_GPL(nfs_dentry_operations); |
1da177e4 | 1922 | |
597d9289 | 1923 | struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) |
1da177e4 LT |
1924 | { |
1925 | struct dentry *res; | |
1926 | struct inode *inode = NULL; | |
e1fb4d05 TM |
1927 | struct nfs_fh *fhandle = NULL; |
1928 | struct nfs_fattr *fattr = NULL; | |
a1147b82 | 1929 | unsigned long dir_verifier; |
1da177e4 | 1930 | int error; |
1da177e4 | 1931 | |
6de1472f | 1932 | dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); |
91d5b470 | 1933 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 | 1934 | |
130f9ab7 AV |
1935 | if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) |
1936 | return ERR_PTR(-ENAMETOOLONG); | |
1da177e4 | 1937 | |
fd684071 TM |
1938 | /* |
1939 | * If we're doing an exclusive create, optimize away the lookup | |
1940 | * but don't hash the dentry. | |
1941 | */ | |
9f6d44d4 | 1942 | if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) |
130f9ab7 | 1943 | return NULL; |
1da177e4 | 1944 | |
e1fb4d05 TM |
1945 | res = ERR_PTR(-ENOMEM); |
1946 | fhandle = nfs_alloc_fhandle(); | |
9558a007 | 1947 | fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir)); |
e1fb4d05 TM |
1948 | if (fhandle == NULL || fattr == NULL) |
1949 | goto out; | |
1950 | ||
a1147b82 | 1951 | dir_verifier = nfs_save_change_attribute(dir); |
6e0d0be7 | 1952 | trace_nfs_lookup_enter(dir, dentry, flags); |
9558a007 | 1953 | error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); |
8ce37abd TM |
1954 | if (error == -ENOENT) { |
1955 | if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) | |
1956 | dir_verifier = inode_peek_iversion_raw(dir); | |
1da177e4 | 1957 | goto no_entry; |
8ce37abd | 1958 | } |
1da177e4 LT |
1959 | if (error < 0) { |
1960 | res = ERR_PTR(error); | |
9558a007 | 1961 | goto out; |
1da177e4 | 1962 | } |
cf7ab00a | 1963 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
bf0c84f1 | 1964 | res = ERR_CAST(inode); |
03f28e3a | 1965 | if (IS_ERR(res)) |
9558a007 | 1966 | goto out; |
54ceac45 | 1967 | |
63519fbc | 1968 | /* Notify readdir to use READDIRPLUS */ |
0b3cc71b | 1969 | nfs_lookup_advise_force_readdirplus(dir, flags); |
d69ee9b8 | 1970 | |
1da177e4 | 1971 | no_entry: |
41d28bca | 1972 | res = d_splice_alias(inode, dentry); |
9eaef27b TM |
1973 | if (res != NULL) { |
1974 | if (IS_ERR(res)) | |
9558a007 | 1975 | goto out; |
1da177e4 | 1976 | dentry = res; |
9eaef27b | 1977 | } |
a1147b82 | 1978 | nfs_set_verifier(dentry, dir_verifier); |
1da177e4 | 1979 | out: |
9558a007 | 1980 | trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res)); |
e1fb4d05 TM |
1981 | nfs_free_fattr(fattr); |
1982 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1983 | return res; |
1984 | } | |
ddda8e0a | 1985 | EXPORT_SYMBOL_GPL(nfs_lookup); |
1da177e4 | 1986 | |
00bdadc7 TM |
1987 | void nfs_d_prune_case_insensitive_aliases(struct inode *inode) |
1988 | { | |
1989 | /* Case insensitive server? Revalidate dentries */ | |
1990 | if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE)) | |
1991 | d_prune_aliases(inode); | |
1992 | } | |
1993 | EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases); | |
1994 | ||
89d77c8f | 1995 | #if IS_ENABLED(CONFIG_NFS_V4) |
0b728e19 | 1996 | static int nfs4_lookup_revalidate(struct dentry *, unsigned int); |
1da177e4 | 1997 | |
f786aa90 | 1998 | const struct dentry_operations nfs4_dentry_operations = { |
0ef97dcf | 1999 | .d_revalidate = nfs4_lookup_revalidate, |
b688741c | 2000 | .d_weak_revalidate = nfs_weak_revalidate, |
1da177e4 LT |
2001 | .d_delete = nfs_dentry_delete, |
2002 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 2003 | .d_automount = nfs_d_automount, |
b1942c5f | 2004 | .d_release = nfs_d_release, |
1da177e4 | 2005 | }; |
89d77c8f | 2006 | EXPORT_SYMBOL_GPL(nfs4_dentry_operations); |
1da177e4 | 2007 | |
532d4def | 2008 | static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) |
cd9a1c0e | 2009 | { |
532d4def | 2010 | return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); |
cd9a1c0e TM |
2011 | } |
2012 | ||
2013 | static int do_open(struct inode *inode, struct file *filp) | |
2014 | { | |
f1fe29b4 | 2015 | nfs_fscache_open_file(inode, filp); |
cd9a1c0e TM |
2016 | return 0; |
2017 | } | |
2018 | ||
d9585277 AV |
2019 | static int nfs_finish_open(struct nfs_open_context *ctx, |
2020 | struct dentry *dentry, | |
b452a458 | 2021 | struct file *file, unsigned open_flags) |
cd9a1c0e | 2022 | { |
0dd2b474 MS |
2023 | int err; |
2024 | ||
be12af3e | 2025 | err = finish_open(file, dentry, do_open); |
30d90494 | 2026 | if (err) |
d9585277 | 2027 | goto out; |
1f24cd31 | 2028 | if (S_ISREG(file_inode(file)->i_mode)) |
eaa2b82c N |
2029 | nfs_file_set_open_context(file, ctx); |
2030 | else | |
9821421a | 2031 | err = -EOPENSTALE; |
cd9a1c0e | 2032 | out: |
d9585277 | 2033 | return err; |
cd9a1c0e TM |
2034 | } |
2035 | ||
73a79706 BS |
2036 | int nfs_atomic_open(struct inode *dir, struct dentry *dentry, |
2037 | struct file *file, unsigned open_flags, | |
44907d79 | 2038 | umode_t mode) |
1da177e4 | 2039 | { |
c94c0953 | 2040 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
cd9a1c0e | 2041 | struct nfs_open_context *ctx; |
0dd2b474 MS |
2042 | struct dentry *res; |
2043 | struct iattr attr = { .ia_valid = ATTR_OPEN }; | |
f46e0bd3 | 2044 | struct inode *inode; |
1472b83e | 2045 | unsigned int lookup_flags = 0; |
68eaba4c | 2046 | unsigned long dir_verifier; |
c94c0953 | 2047 | bool switched = false; |
73a09dd9 | 2048 | int created = 0; |
898f635c | 2049 | int err; |
1da177e4 | 2050 | |
0dd2b474 | 2051 | /* Expect a negative dentry */ |
2b0143b5 | 2052 | BUG_ON(d_inode(dentry)); |
0dd2b474 | 2053 | |
1e8968c5 | 2054 | dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", |
6de1472f | 2055 | dir->i_sb->s_id, dir->i_ino, dentry); |
1e7cb3dc | 2056 | |
9597c13b JL |
2057 | err = nfs_check_flags(open_flags); |
2058 | if (err) | |
2059 | return err; | |
2060 | ||
0dd2b474 MS |
2061 | /* NFS only supports OPEN on regular files */ |
2062 | if ((open_flags & O_DIRECTORY)) { | |
00699ad8 | 2063 | if (!d_in_lookup(dentry)) { |
0dd2b474 MS |
2064 | /* |
2065 | * Hashed negative dentry with O_DIRECTORY: dentry was | |
2066 | * revalidated and is fine, no need to perform lookup | |
2067 | * again | |
2068 | */ | |
d9585277 | 2069 | return -ENOENT; |
0dd2b474 | 2070 | } |
1472b83e | 2071 | lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; |
1da177e4 | 2072 | goto no_open; |
02a913a7 | 2073 | } |
1da177e4 | 2074 | |
0dd2b474 | 2075 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) |
d9585277 | 2076 | return -ENAMETOOLONG; |
cd9a1c0e | 2077 | |
0dd2b474 | 2078 | if (open_flags & O_CREAT) { |
dff25ddb AG |
2079 | struct nfs_server *server = NFS_SERVER(dir); |
2080 | ||
2081 | if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) | |
2082 | mode &= ~current_umask(); | |
2083 | ||
536e43d1 | 2084 | attr.ia_valid |= ATTR_MODE; |
dff25ddb | 2085 | attr.ia_mode = mode; |
0dd2b474 | 2086 | } |
536e43d1 TM |
2087 | if (open_flags & O_TRUNC) { |
2088 | attr.ia_valid |= ATTR_SIZE; | |
2089 | attr.ia_size = 0; | |
cd9a1c0e TM |
2090 | } |
2091 | ||
c94c0953 AV |
2092 | if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { |
2093 | d_drop(dentry); | |
2094 | switched = true; | |
2095 | dentry = d_alloc_parallel(dentry->d_parent, | |
2096 | &dentry->d_name, &wq); | |
2097 | if (IS_ERR(dentry)) | |
2098 | return PTR_ERR(dentry); | |
2099 | if (unlikely(!d_in_lookup(dentry))) | |
2100 | return finish_no_open(file, dentry); | |
2101 | } | |
2102 | ||
532d4def | 2103 | ctx = create_nfs_open_context(dentry, open_flags, file); |
0dd2b474 MS |
2104 | err = PTR_ERR(ctx); |
2105 | if (IS_ERR(ctx)) | |
d9585277 | 2106 | goto out; |
0dd2b474 | 2107 | |
6e0d0be7 | 2108 | trace_nfs_atomic_open_enter(dir, ctx, open_flags); |
73a09dd9 AV |
2109 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); |
2110 | if (created) | |
2111 | file->f_mode |= FMODE_CREATED; | |
f46e0bd3 | 2112 | if (IS_ERR(inode)) { |
0dd2b474 | 2113 | err = PTR_ERR(inode); |
6e0d0be7 | 2114 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 2115 | put_nfs_open_context(ctx); |
d20cb71d | 2116 | d_drop(dentry); |
0dd2b474 MS |
2117 | switch (err) { |
2118 | case -ENOENT: | |
774d9513 | 2119 | d_splice_alias(NULL, dentry); |
68eaba4c TM |
2120 | if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) |
2121 | dir_verifier = inode_peek_iversion_raw(dir); | |
2122 | else | |
2123 | dir_verifier = nfs_save_change_attribute(dir); | |
2124 | nfs_set_verifier(dentry, dir_verifier); | |
0dd2b474 MS |
2125 | break; |
2126 | case -EISDIR: | |
2127 | case -ENOTDIR: | |
2128 | goto no_open; | |
2129 | case -ELOOP: | |
2130 | if (!(open_flags & O_NOFOLLOW)) | |
6f926b5b | 2131 | goto no_open; |
0dd2b474 | 2132 | break; |
1da177e4 | 2133 | /* case -EINVAL: */ |
0dd2b474 MS |
2134 | default: |
2135 | break; | |
1da177e4 | 2136 | } |
d9585277 | 2137 | goto out; |
cd9a1c0e | 2138 | } |
5ee3d10f | 2139 | file->f_mode |= FMODE_CAN_ODIRECT; |
0dd2b474 | 2140 | |
b452a458 | 2141 | err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); |
6e0d0be7 | 2142 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 2143 | put_nfs_open_context(ctx); |
d9585277 | 2144 | out: |
c94c0953 AV |
2145 | if (unlikely(switched)) { |
2146 | d_lookup_done(dentry); | |
2147 | dput(dentry); | |
2148 | } | |
d9585277 | 2149 | return err; |
0dd2b474 | 2150 | |
1da177e4 | 2151 | no_open: |
1472b83e | 2152 | res = nfs_lookup(dir, dentry, lookup_flags); |
ac795161 TM |
2153 | if (!res) { |
2154 | inode = d_inode(dentry); | |
2155 | if ((lookup_flags & LOOKUP_DIRECTORY) && inode && | |
e0caaf75 | 2156 | !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) |
ac795161 | 2157 | res = ERR_PTR(-ENOTDIR); |
1751fc1d TM |
2158 | else if (inode && S_ISREG(inode->i_mode)) |
2159 | res = ERR_PTR(-EOPENSTALE); | |
ac795161 TM |
2160 | } else if (!IS_ERR(res)) { |
2161 | inode = d_inode(res); | |
2162 | if ((lookup_flags & LOOKUP_DIRECTORY) && inode && | |
e0caaf75 | 2163 | !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) { |
ac795161 TM |
2164 | dput(res); |
2165 | res = ERR_PTR(-ENOTDIR); | |
1751fc1d TM |
2166 | } else if (inode && S_ISREG(inode->i_mode)) { |
2167 | dput(res); | |
2168 | res = ERR_PTR(-EOPENSTALE); | |
ac795161 TM |
2169 | } |
2170 | } | |
c94c0953 AV |
2171 | if (switched) { |
2172 | d_lookup_done(dentry); | |
2173 | if (!res) | |
2174 | res = dentry; | |
2175 | else | |
2176 | dput(dentry); | |
2177 | } | |
0dd2b474 | 2178 | if (IS_ERR(res)) |
c94c0953 | 2179 | return PTR_ERR(res); |
e45198a6 | 2180 | return finish_no_open(file, res); |
1da177e4 | 2181 | } |
89d77c8f | 2182 | EXPORT_SYMBOL_GPL(nfs_atomic_open); |
1da177e4 | 2183 | |
c7944ebb TM |
2184 | static int |
2185 | nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, | |
2186 | unsigned int flags) | |
1da177e4 | 2187 | { |
657e94b6 | 2188 | struct inode *inode; |
1da177e4 | 2189 | |
fa3c56bb | 2190 | if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) |
c7944ebb | 2191 | goto full_reval; |
eda72afb | 2192 | if (d_mountpoint(dentry)) |
c7944ebb | 2193 | goto full_reval; |
2b484297 | 2194 | |
2b0143b5 | 2195 | inode = d_inode(dentry); |
2b484297 | 2196 | |
1da177e4 LT |
2197 | /* We can't create new files in nfs_open_revalidate(), so we |
2198 | * optimize away revalidation of negative dentries. | |
2199 | */ | |
c7944ebb TM |
2200 | if (inode == NULL) |
2201 | goto full_reval; | |
2202 | ||
efeda80d | 2203 | if (nfs_verifier_is_delegated(dentry)) |
c7944ebb | 2204 | return nfs_lookup_revalidate_delegated(dir, dentry, inode); |
216d5d06 | 2205 | |
1da177e4 LT |
2206 | /* NFS only supports OPEN on regular files */ |
2207 | if (!S_ISREG(inode->i_mode)) | |
c7944ebb TM |
2208 | goto full_reval; |
2209 | ||
1da177e4 | 2210 | /* We cannot do exclusive creation on a positive dentry */ |
c7944ebb TM |
2211 | if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) |
2212 | goto reval_dentry; | |
2213 | ||
2214 | /* Check if the directory changed */ | |
2215 | if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) | |
2216 | goto reval_dentry; | |
1da177e4 | 2217 | |
0ef97dcf | 2218 | /* Let f_op->open() actually open (and revalidate) the file */ |
c7944ebb TM |
2219 | return 1; |
2220 | reval_dentry: | |
2221 | if (flags & LOOKUP_RCU) | |
2222 | return -ECHILD; | |
0b3cc71b | 2223 | return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags); |
536e43d1 | 2224 | |
c7944ebb TM |
2225 | full_reval: |
2226 | return nfs_do_lookup_revalidate(dir, dentry, flags); | |
2227 | } | |
535918f1 | 2228 | |
c7944ebb TM |
2229 | static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
2230 | { | |
2231 | return __nfs_lookup_revalidate(dentry, flags, | |
2232 | nfs4_do_lookup_revalidate); | |
c0204fd2 TM |
2233 | } |
2234 | ||
1da177e4 LT |
2235 | #endif /* CONFIG_NFSV4 */ |
2236 | ||
406cd915 BC |
2237 | struct dentry * |
2238 | nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, | |
cc6f3298 | 2239 | struct nfs_fattr *fattr) |
1da177e4 | 2240 | { |
fab728e1 | 2241 | struct dentry *parent = dget_parent(dentry); |
2b0143b5 | 2242 | struct inode *dir = d_inode(parent); |
1da177e4 | 2243 | struct inode *inode; |
b0c6108e | 2244 | struct dentry *d; |
406cd915 | 2245 | int error; |
1da177e4 | 2246 | |
fab728e1 TM |
2247 | d_drop(dentry); |
2248 | ||
1da177e4 | 2249 | if (fhandle->size == 0) { |
9558a007 | 2250 | error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); |
1da177e4 | 2251 | if (error) |
fab728e1 | 2252 | goto out_error; |
1da177e4 | 2253 | } |
5724ab37 | 2254 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
2255 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
2256 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
a841b54d | 2257 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, |
2ef61e0e | 2258 | fattr, NULL); |
1da177e4 | 2259 | if (error < 0) |
fab728e1 | 2260 | goto out_error; |
1da177e4 | 2261 | } |
cf7ab00a | 2262 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
b0c6108e | 2263 | d = d_splice_alias(inode, dentry); |
fab728e1 TM |
2264 | out: |
2265 | dput(parent); | |
406cd915 | 2266 | return d; |
fab728e1 | 2267 | out_error: |
406cd915 BC |
2268 | d = ERR_PTR(error); |
2269 | goto out; | |
2270 | } | |
2271 | EXPORT_SYMBOL_GPL(nfs_add_or_obtain); | |
2272 | ||
2273 | /* | |
2274 | * Code common to create, mkdir, and mknod. | |
2275 | */ | |
2276 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
d91bfc46 | 2277 | struct nfs_fattr *fattr) |
406cd915 BC |
2278 | { |
2279 | struct dentry *d; | |
2280 | ||
cc6f3298 | 2281 | d = nfs_add_or_obtain(dentry, fhandle, fattr); |
406cd915 BC |
2282 | if (IS_ERR(d)) |
2283 | return PTR_ERR(d); | |
2284 | ||
2285 | /* Callers don't care */ | |
2286 | dput(d); | |
2287 | return 0; | |
1da177e4 | 2288 | } |
ddda8e0a | 2289 | EXPORT_SYMBOL_GPL(nfs_instantiate); |
1da177e4 LT |
2290 | |
2291 | /* | |
2292 | * Following a failed create operation, we drop the dentry rather | |
2293 | * than retain a negative dentry. This avoids a problem in the event | |
2294 | * that the operation succeeded on the server, but an error in the | |
2295 | * reply path made it appear to have failed. | |
2296 | */ | |
6c960e68 | 2297 | int nfs_create(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 2298 | struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
2299 | { |
2300 | struct iattr attr; | |
ebfc3b49 | 2301 | int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; |
1da177e4 | 2302 | int error; |
1da177e4 | 2303 | |
1e8968c5 | 2304 | dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", |
6de1472f | 2305 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 LT |
2306 | |
2307 | attr.ia_mode = mode; | |
2308 | attr.ia_valid = ATTR_MODE; | |
2309 | ||
8b0ad3d4 | 2310 | trace_nfs_create_enter(dir, dentry, open_flags); |
8867fe58 | 2311 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); |
8b0ad3d4 | 2312 | trace_nfs_create_exit(dir, dentry, open_flags, error); |
1da177e4 LT |
2313 | if (error != 0) |
2314 | goto out_err; | |
1da177e4 LT |
2315 | return 0; |
2316 | out_err: | |
1da177e4 LT |
2317 | d_drop(dentry); |
2318 | return error; | |
2319 | } | |
ddda8e0a | 2320 | EXPORT_SYMBOL_GPL(nfs_create); |
1da177e4 LT |
2321 | |
2322 | /* | |
2323 | * See comments for nfs_proc_create regarding failed operations. | |
2324 | */ | |
597d9289 | 2325 | int |
5ebb29be | 2326 | nfs_mknod(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 2327 | struct dentry *dentry, umode_t mode, dev_t rdev) |
1da177e4 LT |
2328 | { |
2329 | struct iattr attr; | |
2330 | int status; | |
2331 | ||
1e8968c5 | 2332 | dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", |
6de1472f | 2333 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 | 2334 | |
1da177e4 LT |
2335 | attr.ia_mode = mode; |
2336 | attr.ia_valid = ATTR_MODE; | |
2337 | ||
1ca42382 | 2338 | trace_nfs_mknod_enter(dir, dentry); |
1da177e4 | 2339 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1ca42382 | 2340 | trace_nfs_mknod_exit(dir, dentry, status); |
1da177e4 LT |
2341 | if (status != 0) |
2342 | goto out_err; | |
1da177e4 LT |
2343 | return 0; |
2344 | out_err: | |
1da177e4 LT |
2345 | d_drop(dentry); |
2346 | return status; | |
2347 | } | |
ddda8e0a | 2348 | EXPORT_SYMBOL_GPL(nfs_mknod); |
1da177e4 LT |
2349 | |
2350 | /* | |
2351 | * See comments for nfs_proc_create regarding failed operations. | |
2352 | */ | |
c54bd91e | 2353 | int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 2354 | struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2355 | { |
2356 | struct iattr attr; | |
2357 | int error; | |
2358 | ||
1e8968c5 | 2359 | dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", |
6de1472f | 2360 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 LT |
2361 | |
2362 | attr.ia_valid = ATTR_MODE; | |
2363 | attr.ia_mode = mode | S_IFDIR; | |
2364 | ||
1ca42382 | 2365 | trace_nfs_mkdir_enter(dir, dentry); |
1da177e4 | 2366 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1ca42382 | 2367 | trace_nfs_mkdir_exit(dir, dentry, error); |
1da177e4 LT |
2368 | if (error != 0) |
2369 | goto out_err; | |
1da177e4 LT |
2370 | return 0; |
2371 | out_err: | |
2372 | d_drop(dentry); | |
1da177e4 LT |
2373 | return error; |
2374 | } | |
ddda8e0a | 2375 | EXPORT_SYMBOL_GPL(nfs_mkdir); |
1da177e4 | 2376 | |
d45b9d8b TM |
2377 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
2378 | { | |
dc3f4198 | 2379 | if (simple_positive(dentry)) |
d45b9d8b TM |
2380 | d_delete(dentry); |
2381 | } | |
2382 | ||
9019fb39 TM |
2383 | static void nfs_dentry_remove_handle_error(struct inode *dir, |
2384 | struct dentry *dentry, int error) | |
2385 | { | |
2386 | switch (error) { | |
2387 | case -ENOENT: | |
f16857e6 N |
2388 | if (d_really_is_positive(dentry)) |
2389 | d_delete(dentry); | |
00bdadc7 TM |
2390 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
2391 | break; | |
9019fb39 | 2392 | case 0: |
00bdadc7 | 2393 | nfs_d_prune_case_insensitive_aliases(d_inode(dentry)); |
9019fb39 TM |
2394 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
2395 | } | |
2396 | } | |
2397 | ||
597d9289 | 2398 | int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
2399 | { |
2400 | int error; | |
2401 | ||
1e8968c5 | 2402 | dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", |
6de1472f | 2403 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 | 2404 | |
1ca42382 | 2405 | trace_nfs_rmdir_enter(dir, dentry); |
2b0143b5 | 2406 | if (d_really_is_positive(dentry)) { |
884be175 | 2407 | down_write(&NFS_I(d_inode(dentry))->rmdir_sem); |
ba6c0592 TM |
2408 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
2409 | /* Ensure the VFS deletes this inode */ | |
2410 | switch (error) { | |
2411 | case 0: | |
2b0143b5 | 2412 | clear_nlink(d_inode(dentry)); |
ba6c0592 TM |
2413 | break; |
2414 | case -ENOENT: | |
2415 | nfs_dentry_handle_enoent(dentry); | |
2416 | } | |
884be175 | 2417 | up_write(&NFS_I(d_inode(dentry))->rmdir_sem); |
ba6c0592 TM |
2418 | } else |
2419 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
9019fb39 | 2420 | nfs_dentry_remove_handle_error(dir, dentry, error); |
1ca42382 | 2421 | trace_nfs_rmdir_exit(dir, dentry, error); |
1da177e4 LT |
2422 | |
2423 | return error; | |
2424 | } | |
ddda8e0a | 2425 | EXPORT_SYMBOL_GPL(nfs_rmdir); |
1da177e4 | 2426 | |
1da177e4 LT |
2427 | /* |
2428 | * Remove a file after making sure there are no pending writes, | |
2429 | * and after checking that the file has only one user. | |
2430 | * | |
2431 | * We invalidate the attribute cache and free the inode prior to the operation | |
2432 | * to avoid possible races if the server reuses the inode. | |
2433 | */ | |
2434 | static int nfs_safe_remove(struct dentry *dentry) | |
2435 | { | |
2b0143b5 DH |
2436 | struct inode *dir = d_inode(dentry->d_parent); |
2437 | struct inode *inode = d_inode(dentry); | |
1da177e4 LT |
2438 | int error = -EBUSY; |
2439 | ||
6de1472f | 2440 | dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); |
1da177e4 LT |
2441 | |
2442 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
2443 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
2444 | error = 0; | |
2445 | goto out; | |
2446 | } | |
2447 | ||
1ca42382 | 2448 | trace_nfs_remove_enter(dir, dentry); |
1da177e4 | 2449 | if (inode != NULL) { |
912678db | 2450 | error = NFS_PROTO(dir)->remove(dir, dentry); |
1da177e4 | 2451 | if (error == 0) |
1b83d707 | 2452 | nfs_drop_nlink(inode); |
1da177e4 | 2453 | } else |
912678db | 2454 | error = NFS_PROTO(dir)->remove(dir, dentry); |
d45b9d8b TM |
2455 | if (error == -ENOENT) |
2456 | nfs_dentry_handle_enoent(dentry); | |
1ca42382 | 2457 | trace_nfs_remove_exit(dir, dentry, error); |
1da177e4 LT |
2458 | out: |
2459 | return error; | |
2460 | } | |
2461 | ||
2462 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
2463 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
2464 | * | |
2465 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
2466 | */ | |
597d9289 | 2467 | int nfs_unlink(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
2468 | { |
2469 | int error; | |
1da177e4 | 2470 | |
1e8968c5 | 2471 | dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, |
6de1472f | 2472 | dir->i_ino, dentry); |
1da177e4 | 2473 | |
1ca42382 | 2474 | trace_nfs_unlink_enter(dir, dentry); |
1da177e4 | 2475 | spin_lock(&dentry->d_lock); |
43245eca OK |
2476 | if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED, |
2477 | &NFS_I(d_inode(dentry))->flags)) { | |
1da177e4 | 2478 | spin_unlock(&dentry->d_lock); |
ccfeb506 | 2479 | /* Start asynchronous writeout of the inode */ |
2b0143b5 | 2480 | write_inode_now(d_inode(dentry), 0); |
1da177e4 | 2481 | error = nfs_sillyrename(dir, dentry); |
1ca42382 | 2482 | goto out; |
1da177e4 | 2483 | } |
3c59366c N |
2484 | /* We must prevent any concurrent open until the unlink |
2485 | * completes. ->d_revalidate will wait for ->d_fsdata | |
2486 | * to clear. We set it here to ensure no lookup succeeds until | |
2487 | * the unlink is complete on the server. | |
2488 | */ | |
2489 | error = -ETXTBSY; | |
2490 | if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) || | |
2067231a SK |
2491 | WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) { |
2492 | spin_unlock(&dentry->d_lock); | |
3c59366c | 2493 | goto out; |
2067231a | 2494 | } |
121affdf YZ |
2495 | /* old devname */ |
2496 | kfree(dentry->d_fsdata); | |
3c59366c N |
2497 | dentry->d_fsdata = NFS_FSDATA_BLOCKED; |
2498 | ||
1da177e4 | 2499 | spin_unlock(&dentry->d_lock); |
1da177e4 | 2500 | error = nfs_safe_remove(dentry); |
9019fb39 | 2501 | nfs_dentry_remove_handle_error(dir, dentry, error); |
3c59366c N |
2502 | dentry->d_fsdata = NULL; |
2503 | wake_up_var(&dentry->d_fsdata); | |
1ca42382 TM |
2504 | out: |
2505 | trace_nfs_unlink_exit(dir, dentry, error); | |
1da177e4 LT |
2506 | return error; |
2507 | } | |
ddda8e0a | 2508 | EXPORT_SYMBOL_GPL(nfs_unlink); |
1da177e4 | 2509 | |
873101b3 CL |
2510 | /* |
2511 | * To create a symbolic link, most file systems instantiate a new inode, | |
2512 | * add a page to it containing the path, then write it out to the disk | |
2513 | * using prepare_write/commit_write. | |
2514 | * | |
2515 | * Unfortunately the NFS client can't create the in-core inode first | |
2516 | * because it needs a file handle to create an in-core inode (see | |
2517 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
2518 | * symlink request has completed on the server. | |
2519 | * | |
2520 | * So instead we allocate a raw page, copy the symname into it, then do | |
2521 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
2522 | * now have a new file handle and can instantiate an in-core NFS inode | |
2523 | * and move the raw page into its mapping. | |
2524 | */ | |
7a77db95 | 2525 | int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 2526 | struct dentry *dentry, const char *symname) |
1da177e4 | 2527 | { |
873101b3 CL |
2528 | struct page *page; |
2529 | char *kaddr; | |
1da177e4 | 2530 | struct iattr attr; |
873101b3 | 2531 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
2532 | int error; |
2533 | ||
1e8968c5 | 2534 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, |
6de1472f | 2535 | dir->i_ino, dentry, symname); |
1da177e4 | 2536 | |
873101b3 CL |
2537 | if (pathlen > PAGE_SIZE) |
2538 | return -ENAMETOOLONG; | |
1da177e4 | 2539 | |
873101b3 CL |
2540 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
2541 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 2542 | |
e8ecde25 | 2543 | page = alloc_page(GFP_USER); |
76566991 | 2544 | if (!page) |
873101b3 | 2545 | return -ENOMEM; |
873101b3 | 2546 | |
e8ecde25 | 2547 | kaddr = page_address(page); |
873101b3 CL |
2548 | memcpy(kaddr, symname, pathlen); |
2549 | if (pathlen < PAGE_SIZE) | |
2550 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
873101b3 | 2551 | |
1ca42382 | 2552 | trace_nfs_symlink_enter(dir, dentry); |
94a6d753 | 2553 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
1ca42382 | 2554 | trace_nfs_symlink_exit(dir, dentry, error); |
873101b3 | 2555 | if (error != 0) { |
1e8968c5 | 2556 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", |
873101b3 | 2557 | dir->i_sb->s_id, dir->i_ino, |
6de1472f | 2558 | dentry, symname, error); |
1da177e4 | 2559 | d_drop(dentry); |
873101b3 | 2560 | __free_page(page); |
873101b3 CL |
2561 | return error; |
2562 | } | |
2563 | ||
342a67f0 TM |
2564 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
2565 | ||
873101b3 CL |
2566 | /* |
2567 | * No big deal if we can't add this page to the page cache here. | |
2568 | * READLINK will get the missing page from the server if needed. | |
2569 | */ | |
2b0143b5 | 2570 | if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, |
873101b3 | 2571 | GFP_KERNEL)) { |
873101b3 CL |
2572 | SetPageUptodate(page); |
2573 | unlock_page(page); | |
a0b54add RA |
2574 | /* |
2575 | * add_to_page_cache_lru() grabs an extra page refcount. | |
2576 | * Drop it here to avoid leaking this page later. | |
2577 | */ | |
09cbfeaf | 2578 | put_page(page); |
873101b3 CL |
2579 | } else |
2580 | __free_page(page); | |
2581 | ||
873101b3 | 2582 | return 0; |
1da177e4 | 2583 | } |
ddda8e0a | 2584 | EXPORT_SYMBOL_GPL(nfs_symlink); |
1da177e4 | 2585 | |
597d9289 | 2586 | int |
1da177e4 LT |
2587 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) |
2588 | { | |
2b0143b5 | 2589 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
2590 | int error; |
2591 | ||
6de1472f AV |
2592 | dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", |
2593 | old_dentry, dentry); | |
1da177e4 | 2594 | |
1fd1085b | 2595 | trace_nfs_link_enter(inode, dir, dentry); |
9697d234 | 2596 | d_drop(dentry); |
20497503 TM |
2597 | if (S_ISREG(inode->i_mode)) |
2598 | nfs_sync_inode(inode); | |
1da177e4 | 2599 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 2600 | if (error == 0) { |
342a67f0 | 2601 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
7de9c6ee | 2602 | ihold(inode); |
9697d234 | 2603 | d_add(dentry, inode); |
cf809556 | 2604 | } |
1fd1085b | 2605 | trace_nfs_link_exit(inode, dir, dentry, error); |
1da177e4 LT |
2606 | return error; |
2607 | } | |
ddda8e0a | 2608 | EXPORT_SYMBOL_GPL(nfs_link); |
1da177e4 | 2609 | |
3c59366c N |
2610 | static void |
2611 | nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data) | |
2612 | { | |
2613 | struct dentry *new_dentry = data->new_dentry; | |
2614 | ||
2615 | new_dentry->d_fsdata = NULL; | |
2616 | wake_up_var(&new_dentry->d_fsdata); | |
2617 | } | |
2618 | ||
1da177e4 LT |
2619 | /* |
2620 | * RENAME | |
2621 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
2622 | * different file handle for the same inode after a rename (e.g. when | |
2623 | * moving to a different directory). A fail-safe method to do so would | |
2624 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
2625 | * rename the old file using the sillyrename stuff. This way, the original | |
2626 | * file in old_dir will go away when the last process iput()s the inode. | |
2627 | * | |
2628 | * FIXED. | |
2629 | * | |
2630 | * It actually works quite well. One needs to have the possibility for | |
2631 | * at least one ".nfs..." file in each directory the file ever gets | |
2632 | * moved or linked to which happens automagically with the new | |
2633 | * implementation that only depends on the dcache stuff instead of | |
2634 | * using the inode layer | |
2635 | * | |
2636 | * Unfortunately, things are a little more complicated than indicated | |
2637 | * above. For a cross-directory move, we want to make sure we can get | |
2638 | * rid of the old inode after the operation. This means there must be | |
2639 | * no pending writes (if it's a file), and the use count must be 1. | |
2640 | * If these conditions are met, we can drop the dentries before doing | |
2641 | * the rename. | |
2642 | */ | |
e18275ae | 2643 | int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir, |
549c7297 CB |
2644 | struct dentry *old_dentry, struct inode *new_dir, |
2645 | struct dentry *new_dentry, unsigned int flags) | |
1da177e4 | 2646 | { |
2b0143b5 DH |
2647 | struct inode *old_inode = d_inode(old_dentry); |
2648 | struct inode *new_inode = d_inode(new_dentry); | |
3c59366c | 2649 | struct dentry *dentry = NULL; |
80a491fd | 2650 | struct rpc_task *task; |
3c59366c | 2651 | bool must_unblock = false; |
1da177e4 LT |
2652 | int error = -EBUSY; |
2653 | ||
1cd66c93 MS |
2654 | if (flags) |
2655 | return -EINVAL; | |
2656 | ||
6de1472f AV |
2657 | dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", |
2658 | old_dentry, new_dentry, | |
84d08fa8 | 2659 | d_count(new_dentry)); |
1da177e4 | 2660 | |
70ded201 | 2661 | trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); |
1da177e4 | 2662 | /* |
28f79a1a MS |
2663 | * For non-directories, check whether the target is busy and if so, |
2664 | * make a copy of the dentry and then do a silly-rename. If the | |
2665 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
2666 | * the new target. | |
1da177e4 | 2667 | */ |
27226104 | 2668 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
3c59366c N |
2669 | /* We must prevent any concurrent open until the unlink |
2670 | * completes. ->d_revalidate will wait for ->d_fsdata | |
2671 | * to clear. We set it here to ensure no lookup succeeds until | |
2672 | * the unlink is complete on the server. | |
27226104 | 2673 | */ |
3c59366c N |
2674 | error = -ETXTBSY; |
2675 | if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) || | |
2676 | WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED)) | |
2677 | goto out; | |
2678 | if (new_dentry->d_fsdata) { | |
2679 | /* old devname */ | |
2680 | kfree(new_dentry->d_fsdata); | |
2681 | new_dentry->d_fsdata = NULL; | |
d9f29500 | 2682 | } |
1da177e4 | 2683 | |
3c59366c | 2684 | spin_lock(&new_dentry->d_lock); |
84d08fa8 | 2685 | if (d_count(new_dentry) > 2) { |
27226104 MS |
2686 | int err; |
2687 | ||
3c59366c N |
2688 | spin_unlock(&new_dentry->d_lock); |
2689 | ||
27226104 MS |
2690 | /* copy the target dentry's name */ |
2691 | dentry = d_alloc(new_dentry->d_parent, | |
2692 | &new_dentry->d_name); | |
2693 | if (!dentry) | |
2694 | goto out; | |
2695 | ||
2696 | /* silly-rename the existing target ... */ | |
2697 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 2698 | if (err) |
27226104 | 2699 | goto out; |
24e93025 MS |
2700 | |
2701 | new_dentry = dentry; | |
2702 | new_inode = NULL; | |
3c59366c N |
2703 | } else { |
2704 | new_dentry->d_fsdata = NFS_FSDATA_BLOCKED; | |
2705 | must_unblock = true; | |
2706 | spin_unlock(&new_dentry->d_lock); | |
27226104 | 2707 | } |
3c59366c | 2708 | |
b1e4adf4 | 2709 | } |
1da177e4 | 2710 | |
6ff9d99b TM |
2711 | if (S_ISREG(old_inode->i_mode)) |
2712 | nfs_sync_inode(old_inode); | |
3c59366c N |
2713 | task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, |
2714 | must_unblock ? nfs_unblock_rename : NULL); | |
80a491fd JL |
2715 | if (IS_ERR(task)) { |
2716 | error = PTR_ERR(task); | |
2717 | goto out; | |
2718 | } | |
2719 | ||
2720 | error = rpc_wait_for_completion_task(task); | |
818a8dbe BC |
2721 | if (error != 0) { |
2722 | ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; | |
2723 | /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ | |
2724 | smp_wmb(); | |
2725 | } else | |
80a491fd JL |
2726 | error = task->tk_status; |
2727 | rpc_put_task(task); | |
59a707b0 TM |
2728 | /* Ensure the inode attributes are revalidated */ |
2729 | if (error == 0) { | |
2730 | spin_lock(&old_inode->i_lock); | |
2731 | NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); | |
ac46b3d7 TM |
2732 | nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE | |
2733 | NFS_INO_INVALID_CTIME | | |
2734 | NFS_INO_REVAL_FORCED); | |
59a707b0 TM |
2735 | spin_unlock(&old_inode->i_lock); |
2736 | } | |
1da177e4 | 2737 | out: |
70ded201 TM |
2738 | trace_nfs_rename_exit(old_dir, old_dentry, |
2739 | new_dir, new_dentry, error); | |
d9f29500 BC |
2740 | if (!error) { |
2741 | if (new_inode != NULL) | |
2742 | nfs_drop_nlink(new_inode); | |
2743 | /* | |
2744 | * The d_move() should be here instead of in an async RPC completion | |
2745 | * handler because we need the proper locks to move the dentry. If | |
2746 | * we're interrupted by a signal, the async RPC completion handler | |
2747 | * should mark the directories for revalidation. | |
2748 | */ | |
2749 | d_move(old_dentry, new_dentry); | |
d803224c | 2750 | nfs_set_verifier(old_dentry, |
d9f29500 BC |
2751 | nfs_save_change_attribute(new_dir)); |
2752 | } else if (error == -ENOENT) | |
2753 | nfs_dentry_handle_enoent(old_dentry); | |
2754 | ||
1da177e4 LT |
2755 | /* new dentry created? */ |
2756 | if (dentry) | |
2757 | dput(dentry); | |
1da177e4 LT |
2758 | return error; |
2759 | } | |
ddda8e0a | 2760 | EXPORT_SYMBOL_GPL(nfs_rename); |
1da177e4 | 2761 | |
cfcea3e8 TM |
2762 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
2763 | static LIST_HEAD(nfs_access_lru_list); | |
2764 | static atomic_long_t nfs_access_nr_entries; | |
2765 | ||
a8b373ee | 2766 | static unsigned long nfs_access_max_cachesize = 4*1024*1024; |
3a505845 TM |
2767 | module_param(nfs_access_max_cachesize, ulong, 0644); |
2768 | MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); | |
2769 | ||
1c3c07e9 TM |
2770 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
2771 | { | |
6238aec8 | 2772 | put_group_info(entry->group_info); |
f682a398 | 2773 | kfree_rcu(entry, rcu_head); |
4e857c58 | 2774 | smp_mb__before_atomic(); |
cfcea3e8 | 2775 | atomic_long_dec(&nfs_access_nr_entries); |
4e857c58 | 2776 | smp_mb__after_atomic(); |
1c3c07e9 TM |
2777 | } |
2778 | ||
1a81bb8a TM |
2779 | static void nfs_access_free_list(struct list_head *head) |
2780 | { | |
2781 | struct nfs_access_entry *cache; | |
2782 | ||
2783 | while (!list_empty(head)) { | |
2784 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
2785 | list_del(&cache->lru); | |
2786 | nfs_access_free_entry(cache); | |
2787 | } | |
2788 | } | |
2789 | ||
3a505845 TM |
2790 | static unsigned long |
2791 | nfs_do_access_cache_scan(unsigned int nr_to_scan) | |
979df72e TM |
2792 | { |
2793 | LIST_HEAD(head); | |
aa510da5 | 2794 | struct nfs_inode *nfsi, *next; |
979df72e | 2795 | struct nfs_access_entry *cache; |
1ab6c499 | 2796 | long freed = 0; |
979df72e | 2797 | |
a50f7951 | 2798 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 2799 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
2800 | struct inode *inode; |
2801 | ||
2802 | if (nr_to_scan-- == 0) | |
2803 | break; | |
9c7e7e23 | 2804 | inode = &nfsi->vfs_inode; |
979df72e TM |
2805 | spin_lock(&inode->i_lock); |
2806 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
2807 | goto remove_lru_entry; | |
2808 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
2809 | struct nfs_access_entry, lru); | |
2810 | list_move(&cache->lru, &head); | |
2811 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1ab6c499 | 2812 | freed++; |
979df72e TM |
2813 | if (!list_empty(&nfsi->access_cache_entry_lru)) |
2814 | list_move_tail(&nfsi->access_cache_inode_lru, | |
2815 | &nfs_access_lru_list); | |
2816 | else { | |
2817 | remove_lru_entry: | |
2818 | list_del_init(&nfsi->access_cache_inode_lru); | |
4e857c58 | 2819 | smp_mb__before_atomic(); |
979df72e | 2820 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
4e857c58 | 2821 | smp_mb__after_atomic(); |
979df72e | 2822 | } |
59844a9b | 2823 | spin_unlock(&inode->i_lock); |
979df72e TM |
2824 | } |
2825 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2826 | nfs_access_free_list(&head); |
1ab6c499 DC |
2827 | return freed; |
2828 | } | |
2829 | ||
3a505845 TM |
2830 | unsigned long |
2831 | nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
2832 | { | |
2833 | int nr_to_scan = sc->nr_to_scan; | |
2834 | gfp_t gfp_mask = sc->gfp_mask; | |
2835 | ||
2836 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) | |
2837 | return SHRINK_STOP; | |
2838 | return nfs_do_access_cache_scan(nr_to_scan); | |
2839 | } | |
2840 | ||
2841 | ||
1ab6c499 DC |
2842 | unsigned long |
2843 | nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
2844 | { | |
55f841ce | 2845 | return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); |
979df72e TM |
2846 | } |
2847 | ||
3a505845 TM |
2848 | static void |
2849 | nfs_access_cache_enforce_limit(void) | |
2850 | { | |
2851 | long nr_entries = atomic_long_read(&nfs_access_nr_entries); | |
2852 | unsigned long diff; | |
2853 | unsigned int nr_to_scan; | |
2854 | ||
2855 | if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) | |
2856 | return; | |
2857 | nr_to_scan = 100; | |
2858 | diff = nr_entries - nfs_access_max_cachesize; | |
2859 | if (diff < nr_to_scan) | |
2860 | nr_to_scan = diff; | |
2861 | nfs_do_access_cache_scan(nr_to_scan); | |
2862 | } | |
2863 | ||
1a81bb8a | 2864 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2865 | { |
1c3c07e9 | 2866 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2867 | struct rb_node *n; |
1c3c07e9 TM |
2868 | struct nfs_access_entry *entry; |
2869 | ||
2870 | /* Unhook entries from the cache */ | |
2871 | while ((n = rb_first(root_node)) != NULL) { | |
2872 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2873 | rb_erase(n, root_node); | |
1a81bb8a | 2874 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2875 | } |
2876 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2877 | } |
2878 | ||
1c3c07e9 | 2879 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2880 | { |
1a81bb8a TM |
2881 | LIST_HEAD(head); |
2882 | ||
2883 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2884 | return; | |
cfcea3e8 | 2885 | /* Remove from global LRU init */ |
1a81bb8a TM |
2886 | spin_lock(&nfs_access_lru_lock); |
2887 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2888 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2889 | |
1c3c07e9 | 2890 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2891 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2892 | spin_unlock(&inode->i_lock); | |
2893 | spin_unlock(&nfs_access_lru_lock); | |
2894 | nfs_access_free_list(&head); | |
1c3c07e9 | 2895 | } |
1c606fb7 | 2896 | EXPORT_SYMBOL_GPL(nfs_access_zap_cache); |
1da177e4 | 2897 | |
6238aec8 N |
2898 | static int access_cmp(const struct cred *a, const struct nfs_access_entry *b) |
2899 | { | |
2900 | struct group_info *ga, *gb; | |
2901 | int g; | |
2902 | ||
2903 | if (uid_lt(a->fsuid, b->fsuid)) | |
2904 | return -1; | |
2905 | if (uid_gt(a->fsuid, b->fsuid)) | |
2906 | return 1; | |
2907 | ||
2908 | if (gid_lt(a->fsgid, b->fsgid)) | |
2909 | return -1; | |
2910 | if (gid_gt(a->fsgid, b->fsgid)) | |
2911 | return 1; | |
2912 | ||
2913 | ga = a->group_info; | |
2914 | gb = b->group_info; | |
2915 | if (ga == gb) | |
2916 | return 0; | |
2917 | if (ga == NULL) | |
2918 | return -1; | |
2919 | if (gb == NULL) | |
2920 | return 1; | |
2921 | if (ga->ngroups < gb->ngroups) | |
2922 | return -1; | |
2923 | if (ga->ngroups > gb->ngroups) | |
2924 | return 1; | |
2925 | ||
2926 | for (g = 0; g < ga->ngroups; g++) { | |
2927 | if (gid_lt(ga->gid[g], gb->gid[g])) | |
2928 | return -1; | |
2929 | if (gid_gt(ga->gid[g], gb->gid[g])) | |
2930 | return 1; | |
2931 | } | |
2932 | return 0; | |
2933 | } | |
2934 | ||
b68572e0 | 2935 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) |
1c3c07e9 TM |
2936 | { |
2937 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
1c3c07e9 TM |
2938 | |
2939 | while (n != NULL) { | |
b68572e0 N |
2940 | struct nfs_access_entry *entry = |
2941 | rb_entry(n, struct nfs_access_entry, rb_node); | |
6238aec8 | 2942 | int cmp = access_cmp(cred, entry); |
1c3c07e9 | 2943 | |
b68572e0 | 2944 | if (cmp < 0) |
1c3c07e9 | 2945 | n = n->rb_left; |
b68572e0 | 2946 | else if (cmp > 0) |
1c3c07e9 TM |
2947 | n = n->rb_right; |
2948 | else | |
2949 | return entry; | |
1da177e4 | 2950 | } |
1c3c07e9 TM |
2951 | return NULL; |
2952 | } | |
2953 | ||
0eb43812 TM |
2954 | static u64 nfs_access_login_time(const struct task_struct *task, |
2955 | const struct cred *cred) | |
2956 | { | |
2957 | const struct task_struct *parent; | |
5e9a7b9c | 2958 | const struct cred *pcred; |
0eb43812 TM |
2959 | u64 ret; |
2960 | ||
2961 | rcu_read_lock(); | |
2962 | for (;;) { | |
2963 | parent = rcu_dereference(task->real_parent); | |
5e9a7b9c TM |
2964 | pcred = rcu_dereference(parent->cred); |
2965 | if (parent == task || cred_fscmp(pcred, cred) != 0) | |
0eb43812 TM |
2966 | break; |
2967 | task = parent; | |
2968 | } | |
2969 | ret = task->start_time; | |
2970 | rcu_read_unlock(); | |
2971 | return ret; | |
2972 | } | |
2973 | ||
b5e7b59c | 2974 | static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block) |
1c3c07e9 TM |
2975 | { |
2976 | struct nfs_inode *nfsi = NFS_I(inode); | |
0eb43812 | 2977 | u64 login_time = nfs_access_login_time(current, cred); |
1c3c07e9 | 2978 | struct nfs_access_entry *cache; |
57b69181 TM |
2979 | bool retry = true; |
2980 | int err; | |
1c3c07e9 | 2981 | |
dc59250c | 2982 | spin_lock(&inode->i_lock); |
57b69181 TM |
2983 | for(;;) { |
2984 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
2985 | goto out_zap; | |
2986 | cache = nfs_access_search_rbtree(inode, cred); | |
2987 | err = -ENOENT; | |
2988 | if (cache == NULL) | |
2989 | goto out; | |
2990 | /* Found an entry, is our attribute cache valid? */ | |
21c3ba7e | 2991 | if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) |
57b69181 | 2992 | break; |
5c965db8 TM |
2993 | if (!retry) |
2994 | break; | |
57b69181 TM |
2995 | err = -ECHILD; |
2996 | if (!may_block) | |
2997 | goto out; | |
57b69181 TM |
2998 | spin_unlock(&inode->i_lock); |
2999 | err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
3000 | if (err) | |
3001 | return err; | |
3002 | spin_lock(&inode->i_lock); | |
3003 | retry = false; | |
3004 | } | |
0eb43812 TM |
3005 | err = -ENOENT; |
3006 | if ((s64)(login_time - cache->timestamp) > 0) | |
3007 | goto out; | |
b5e7b59c | 3008 | *mask = cache->mask; |
cfcea3e8 | 3009 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
3010 | err = 0; |
3011 | out: | |
3012 | spin_unlock(&inode->i_lock); | |
3013 | return err; | |
1c3c07e9 | 3014 | out_zap: |
1a81bb8a TM |
3015 | spin_unlock(&inode->i_lock); |
3016 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
3017 | return -ENOENT; |
3018 | } | |
3019 | ||
b5e7b59c | 3020 | static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask) |
f682a398 N |
3021 | { |
3022 | /* Only check the most recently returned cache entry, | |
3023 | * but do it without locking. | |
3024 | */ | |
3025 | struct nfs_inode *nfsi = NFS_I(inode); | |
029085b8 | 3026 | u64 login_time = nfs_access_login_time(current, cred); |
f682a398 N |
3027 | struct nfs_access_entry *cache; |
3028 | int err = -ECHILD; | |
3029 | struct list_head *lh; | |
3030 | ||
3031 | rcu_read_lock(); | |
3032 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
3033 | goto out; | |
9f01eb5d | 3034 | lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); |
f682a398 N |
3035 | cache = list_entry(lh, struct nfs_access_entry, lru); |
3036 | if (lh == &nfsi->access_cache_entry_lru || | |
6238aec8 | 3037 | access_cmp(cred, cache) != 0) |
f682a398 N |
3038 | cache = NULL; |
3039 | if (cache == NULL) | |
3040 | goto out; | |
029085b8 CD |
3041 | if ((s64)(login_time - cache->timestamp) > 0) |
3042 | goto out; | |
21c3ba7e | 3043 | if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) |
f682a398 | 3044 | goto out; |
b5e7b59c | 3045 | *mask = cache->mask; |
21c3ba7e | 3046 | err = 0; |
f682a398 N |
3047 | out: |
3048 | rcu_read_unlock(); | |
3049 | return err; | |
3050 | } | |
3051 | ||
b5e7b59c N |
3052 | int nfs_access_get_cached(struct inode *inode, const struct cred *cred, |
3053 | u32 *mask, bool may_block) | |
d2ae4f8b FL |
3054 | { |
3055 | int status; | |
3056 | ||
b5e7b59c | 3057 | status = nfs_access_get_cached_rcu(inode, cred, mask); |
d2ae4f8b | 3058 | if (status != 0) |
b5e7b59c | 3059 | status = nfs_access_get_cached_locked(inode, cred, mask, |
d2ae4f8b FL |
3060 | may_block); |
3061 | ||
3062 | return status; | |
3063 | } | |
3064 | EXPORT_SYMBOL_GPL(nfs_access_get_cached); | |
3065 | ||
73fbb3fa N |
3066 | static void nfs_access_add_rbtree(struct inode *inode, |
3067 | struct nfs_access_entry *set, | |
3068 | const struct cred *cred) | |
1c3c07e9 | 3069 | { |
cfcea3e8 TM |
3070 | struct nfs_inode *nfsi = NFS_I(inode); |
3071 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
3072 | struct rb_node **p = &root_node->rb_node; |
3073 | struct rb_node *parent = NULL; | |
3074 | struct nfs_access_entry *entry; | |
b68572e0 | 3075 | int cmp; |
1c3c07e9 TM |
3076 | |
3077 | spin_lock(&inode->i_lock); | |
3078 | while (*p != NULL) { | |
3079 | parent = *p; | |
3080 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
6238aec8 | 3081 | cmp = access_cmp(cred, entry); |
1c3c07e9 | 3082 | |
b68572e0 | 3083 | if (cmp < 0) |
1c3c07e9 | 3084 | p = &parent->rb_left; |
b68572e0 | 3085 | else if (cmp > 0) |
1c3c07e9 TM |
3086 | p = &parent->rb_right; |
3087 | else | |
3088 | goto found; | |
3089 | } | |
3090 | rb_link_node(&set->rb_node, parent, p); | |
3091 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 3092 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 3093 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
3094 | return; |
3095 | found: | |
3096 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
3097 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
3098 | list_del(&entry->lru); | |
1c3c07e9 TM |
3099 | spin_unlock(&inode->i_lock); |
3100 | nfs_access_free_entry(entry); | |
3101 | } | |
3102 | ||
73fbb3fa N |
3103 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set, |
3104 | const struct cred *cred) | |
1c3c07e9 TM |
3105 | { |
3106 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
3107 | if (cache == NULL) | |
3108 | return; | |
3109 | RB_CLEAR_NODE(&cache->rb_node); | |
6238aec8 N |
3110 | cache->fsuid = cred->fsuid; |
3111 | cache->fsgid = cred->fsgid; | |
3112 | cache->group_info = get_group_info(cred->group_info); | |
1da177e4 | 3113 | cache->mask = set->mask; |
21fd9e87 | 3114 | cache->timestamp = ktime_get_ns(); |
1c3c07e9 | 3115 | |
f682a398 N |
3116 | /* The above field assignments must be visible |
3117 | * before this item appears on the lru. We cannot easily | |
3118 | * use rcu_assign_pointer, so just force the memory barrier. | |
3119 | */ | |
3120 | smp_wmb(); | |
73fbb3fa | 3121 | nfs_access_add_rbtree(inode, cache, cred); |
cfcea3e8 TM |
3122 | |
3123 | /* Update accounting */ | |
4e857c58 | 3124 | smp_mb__before_atomic(); |
cfcea3e8 | 3125 | atomic_long_inc(&nfs_access_nr_entries); |
4e857c58 | 3126 | smp_mb__after_atomic(); |
cfcea3e8 TM |
3127 | |
3128 | /* Add inode to global LRU list */ | |
1a81bb8a | 3129 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 3130 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
3131 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
3132 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
3133 | &nfs_access_lru_list); | |
cfcea3e8 TM |
3134 | spin_unlock(&nfs_access_lru_lock); |
3135 | } | |
3a505845 | 3136 | nfs_access_cache_enforce_limit(); |
1da177e4 | 3137 | } |
6168f62c WAA |
3138 | EXPORT_SYMBOL_GPL(nfs_access_add_cache); |
3139 | ||
3c181827 AS |
3140 | #define NFS_MAY_READ (NFS_ACCESS_READ) |
3141 | #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ | |
3142 | NFS_ACCESS_EXTEND | \ | |
3143 | NFS_ACCESS_DELETE) | |
3144 | #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ | |
3145 | NFS_ACCESS_EXTEND) | |
ecbb903c | 3146 | #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE |
3c181827 AS |
3147 | #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) |
3148 | #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) | |
15d4b73a | 3149 | static int |
ecbb903c | 3150 | nfs_access_calc_mask(u32 access_result, umode_t umode) |
15d4b73a TM |
3151 | { |
3152 | int mask = 0; | |
3153 | ||
3154 | if (access_result & NFS_MAY_READ) | |
3155 | mask |= MAY_READ; | |
ecbb903c TM |
3156 | if (S_ISDIR(umode)) { |
3157 | if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) | |
3158 | mask |= MAY_WRITE; | |
3159 | if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) | |
3160 | mask |= MAY_EXEC; | |
3161 | } else if (S_ISREG(umode)) { | |
3162 | if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) | |
3163 | mask |= MAY_WRITE; | |
3164 | if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) | |
3165 | mask |= MAY_EXEC; | |
3166 | } else if (access_result & NFS_MAY_WRITE) | |
3167 | mask |= MAY_WRITE; | |
15d4b73a TM |
3168 | return mask; |
3169 | } | |
3170 | ||
6168f62c WAA |
3171 | void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) |
3172 | { | |
bd8b2441 | 3173 | entry->mask = access_result; |
6168f62c WAA |
3174 | } |
3175 | EXPORT_SYMBOL_GPL(nfs_access_set_mask); | |
1da177e4 | 3176 | |
b68572e0 | 3177 | static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) |
1da177e4 LT |
3178 | { |
3179 | struct nfs_access_entry cache; | |
57b69181 | 3180 | bool may_block = (mask & MAY_NOT_BLOCK) == 0; |
e8194b7d | 3181 | int cache_mask = -1; |
1da177e4 LT |
3182 | int status; |
3183 | ||
f4ce1299 TM |
3184 | trace_nfs_access_enter(inode); |
3185 | ||
b5e7b59c | 3186 | status = nfs_access_get_cached(inode, cred, &cache.mask, may_block); |
1da177e4 | 3187 | if (status == 0) |
f4ce1299 | 3188 | goto out_cached; |
1da177e4 | 3189 | |
f3324a2a | 3190 | status = -ECHILD; |
57b69181 | 3191 | if (!may_block) |
f3324a2a N |
3192 | goto out; |
3193 | ||
1750d929 AS |
3194 | /* |
3195 | * Determine which access bits we want to ask for... | |
3196 | */ | |
84631f84 TM |
3197 | cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND | |
3198 | nfs_access_xattr_mask(NFS_SERVER(inode)); | |
1750d929 AS |
3199 | if (S_ISDIR(inode->i_mode)) |
3200 | cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; | |
3201 | else | |
3202 | cache.mask |= NFS_ACCESS_EXECUTE; | |
73fbb3fa | 3203 | status = NFS_PROTO(inode)->access(inode, &cache, cred); |
a71ee337 SJ |
3204 | if (status != 0) { |
3205 | if (status == -ESTALE) { | |
a71ee337 | 3206 | if (!S_ISDIR(inode->i_mode)) |
93ce4af7 TM |
3207 | nfs_set_inode_stale(inode); |
3208 | else | |
3209 | nfs_zap_caches(inode); | |
a71ee337 | 3210 | } |
f4ce1299 | 3211 | goto out; |
a71ee337 | 3212 | } |
73fbb3fa | 3213 | nfs_access_add_cache(inode, &cache, cred); |
f4ce1299 | 3214 | out_cached: |
ecbb903c | 3215 | cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); |
bd8b2441 | 3216 | if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) |
f4ce1299 | 3217 | status = -EACCES; |
1da177e4 | 3218 | out: |
e8194b7d | 3219 | trace_nfs_access_exit(inode, mask, cache_mask, status); |
f4ce1299 | 3220 | return status; |
1da177e4 LT |
3221 | } |
3222 | ||
af22f94a TM |
3223 | static int nfs_open_permission_mask(int openflags) |
3224 | { | |
3225 | int mask = 0; | |
3226 | ||
f8d9a897 WAA |
3227 | if (openflags & __FMODE_EXEC) { |
3228 | /* ONLY check exec rights */ | |
3229 | mask = MAY_EXEC; | |
3230 | } else { | |
3231 | if ((openflags & O_ACCMODE) != O_WRONLY) | |
3232 | mask |= MAY_READ; | |
3233 | if ((openflags & O_ACCMODE) != O_RDONLY) | |
3234 | mask |= MAY_WRITE; | |
3235 | } | |
3236 | ||
af22f94a TM |
3237 | return mask; |
3238 | } | |
3239 | ||
b68572e0 | 3240 | int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) |
af22f94a TM |
3241 | { |
3242 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
3243 | } | |
89d77c8f | 3244 | EXPORT_SYMBOL_GPL(nfs_may_open); |
af22f94a | 3245 | |
5c5fc09a TM |
3246 | static int nfs_execute_ok(struct inode *inode, int mask) |
3247 | { | |
3248 | struct nfs_server *server = NFS_SERVER(inode); | |
21c3ba7e | 3249 | int ret = 0; |
5c5fc09a | 3250 | |
3825827e TM |
3251 | if (S_ISDIR(inode->i_mode)) |
3252 | return 0; | |
720869eb | 3253 | if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) { |
21c3ba7e TM |
3254 | if (mask & MAY_NOT_BLOCK) |
3255 | return -ECHILD; | |
3256 | ret = __nfs_revalidate_inode(server, inode); | |
3257 | } | |
5c5fc09a TM |
3258 | if (ret == 0 && !execute_ok(inode)) |
3259 | ret = -EACCES; | |
3260 | return ret; | |
3261 | } | |
3262 | ||
4609e1f1 | 3263 | int nfs_permission(struct mnt_idmap *idmap, |
549c7297 CB |
3264 | struct inode *inode, |
3265 | int mask) | |
1da177e4 | 3266 | { |
b68572e0 | 3267 | const struct cred *cred = current_cred(); |
1da177e4 LT |
3268 | int res = 0; |
3269 | ||
91d5b470 CL |
3270 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
3271 | ||
e6305c43 | 3272 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
3273 | goto out; |
3274 | /* Is this sys_access() ? */ | |
9cfcac81 | 3275 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
3276 | goto force_lookup; |
3277 | ||
3278 | switch (inode->i_mode & S_IFMT) { | |
3279 | case S_IFLNK: | |
3280 | goto out; | |
3281 | case S_IFREG: | |
762674f8 TM |
3282 | if ((mask & MAY_OPEN) && |
3283 | nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) | |
3284 | return 0; | |
1da177e4 LT |
3285 | break; |
3286 | case S_IFDIR: | |
3287 | /* | |
3288 | * Optimize away all write operations, since the server | |
3289 | * will check permissions when we perform the op. | |
3290 | */ | |
3291 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
3292 | goto out; | |
3293 | } | |
3294 | ||
3295 | force_lookup: | |
1da177e4 LT |
3296 | if (!NFS_PROTO(inode)->access) |
3297 | goto out_notsup; | |
3298 | ||
eb095c14 | 3299 | res = nfs_do_access(inode, cred, mask); |
1da177e4 | 3300 | out: |
5c5fc09a TM |
3301 | if (!res && (mask & MAY_EXEC)) |
3302 | res = nfs_execute_ok(inode, mask); | |
f696a365 | 3303 | |
1e8968c5 | 3304 | dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", |
1e7cb3dc | 3305 | inode->i_sb->s_id, inode->i_ino, mask, res); |
1da177e4 LT |
3306 | return res; |
3307 | out_notsup: | |
d51ac1a8 N |
3308 | if (mask & MAY_NOT_BLOCK) |
3309 | return -ECHILD; | |
3310 | ||
720869eb TM |
3311 | res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE | |
3312 | NFS_INO_INVALID_OTHER); | |
1da177e4 | 3313 | if (res == 0) |
4609e1f1 | 3314 | res = generic_permission(&nop_mnt_idmap, inode, mask); |
1e7cb3dc | 3315 | goto out; |
1da177e4 | 3316 | } |
ddda8e0a | 3317 | EXPORT_SYMBOL_GPL(nfs_permission); |