]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/dir.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * nfs directory handling functions | |
7 | * | |
8 | * 10 Apr 1996 Added silly rename for unlink --okir | |
9 | * 28 Sep 1996 Improved directory cache --okir | |
10 | * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de | |
11 | * Re-implemented silly rename for unlink, newly implemented | |
12 | * silly rename for nfs_rename() following the suggestions | |
13 | * of Olaf Kirch (okir) found in this file. | |
14 | * Following Linus comments on my original hack, this version | |
15 | * depends only on the dcache stuff and doesn't touch the inode | |
16 | * layer (iput() and friends). | |
17 | * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM | |
18 | */ | |
19 | ||
ddda8e0a | 20 | #include <linux/module.h> |
1da177e4 LT |
21 | #include <linux/time.h> |
22 | #include <linux/errno.h> | |
23 | #include <linux/stat.h> | |
24 | #include <linux/fcntl.h> | |
25 | #include <linux/string.h> | |
26 | #include <linux/kernel.h> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/mm.h> | |
29 | #include <linux/sunrpc/clnt.h> | |
30 | #include <linux/nfs_fs.h> | |
31 | #include <linux/nfs_mount.h> | |
32 | #include <linux/pagemap.h> | |
873101b3 | 33 | #include <linux/pagevec.h> |
1da177e4 | 34 | #include <linux/namei.h> |
54ceac45 | 35 | #include <linux/mount.h> |
a0b8cab3 | 36 | #include <linux/swap.h> |
e8edc6e0 | 37 | #include <linux/sched.h> |
04e4bd1c | 38 | #include <linux/kmemleak.h> |
64c2ce8b | 39 | #include <linux/xattr.h> |
1da177e4 LT |
40 | |
41 | #include "delegation.h" | |
91d5b470 | 42 | #include "iostat.h" |
4c30d56e | 43 | #include "internal.h" |
cd9a1c0e | 44 | #include "fscache.h" |
1da177e4 | 45 | |
f4ce1299 TM |
46 | #include "nfstrace.h" |
47 | ||
1da177e4 LT |
48 | /* #define NFS_DEBUG_VERBOSE 1 */ |
49 | ||
50 | static int nfs_opendir(struct inode *, struct file *); | |
480c2006 | 51 | static int nfs_closedir(struct inode *, struct file *); |
23db8620 | 52 | static int nfs_readdir(struct file *, struct dir_context *); |
02c24a82 | 53 | static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); |
f0dd2136 | 54 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
11de3b11 | 55 | static void nfs_readdir_clear_array(struct page*); |
1da177e4 | 56 | |
4b6f5d20 | 57 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 58 | .llseek = nfs_llseek_dir, |
1da177e4 | 59 | .read = generic_read_dir, |
23db8620 | 60 | .iterate = nfs_readdir, |
1da177e4 | 61 | .open = nfs_opendir, |
480c2006 | 62 | .release = nfs_closedir, |
1da177e4 LT |
63 | .fsync = nfs_fsync_dir, |
64 | }; | |
65 | ||
11de3b11 TM |
66 | const struct address_space_operations nfs_dir_aops = { |
67 | .freepage = nfs_readdir_clear_array, | |
d1bacf9e BS |
68 | }; |
69 | ||
0c030806 | 70 | static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred) |
480c2006 BS |
71 | { |
72 | struct nfs_open_dir_context *ctx; | |
73 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
74 | if (ctx != NULL) { | |
8ef2ce3e | 75 | ctx->duped = 0; |
0c030806 | 76 | ctx->attr_gencount = NFS_I(dir)->attr_gencount; |
480c2006 | 77 | ctx->dir_cookie = 0; |
8ef2ce3e | 78 | ctx->dup_cookie = 0; |
480c2006 | 79 | ctx->cred = get_rpccred(cred); |
0c030806 TM |
80 | return ctx; |
81 | } | |
82 | return ERR_PTR(-ENOMEM); | |
480c2006 BS |
83 | } |
84 | ||
85 | static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx) | |
86 | { | |
87 | put_rpccred(ctx->cred); | |
88 | kfree(ctx); | |
89 | } | |
90 | ||
1da177e4 LT |
91 | /* |
92 | * Open file | |
93 | */ | |
94 | static int | |
95 | nfs_opendir(struct inode *inode, struct file *filp) | |
96 | { | |
480c2006 BS |
97 | int res = 0; |
98 | struct nfs_open_dir_context *ctx; | |
99 | struct rpc_cred *cred; | |
1da177e4 | 100 | |
6de1472f | 101 | dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); |
cc0dd2d1 CL |
102 | |
103 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 104 | |
480c2006 BS |
105 | cred = rpc_lookup_cred(); |
106 | if (IS_ERR(cred)) | |
107 | return PTR_ERR(cred); | |
0c030806 | 108 | ctx = alloc_nfs_open_dir_context(inode, cred); |
480c2006 BS |
109 | if (IS_ERR(ctx)) { |
110 | res = PTR_ERR(ctx); | |
111 | goto out; | |
112 | } | |
113 | filp->private_data = ctx; | |
f5a73672 NB |
114 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { |
115 | /* This is a mountpoint, so d_revalidate will never | |
116 | * have been called, so we need to refresh the | |
117 | * inode (for close-open consistency) ourselves. | |
118 | */ | |
119 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
120 | } | |
480c2006 BS |
121 | out: |
122 | put_rpccred(cred); | |
1da177e4 LT |
123 | return res; |
124 | } | |
125 | ||
480c2006 BS |
126 | static int |
127 | nfs_closedir(struct inode *inode, struct file *filp) | |
128 | { | |
129 | put_nfs_open_dir_context(filp->private_data); | |
130 | return 0; | |
131 | } | |
132 | ||
d1bacf9e BS |
133 | struct nfs_cache_array_entry { |
134 | u64 cookie; | |
135 | u64 ino; | |
136 | struct qstr string; | |
0b26a0bf | 137 | unsigned char d_type; |
d1bacf9e BS |
138 | }; |
139 | ||
140 | struct nfs_cache_array { | |
88b8e133 | 141 | int size; |
d1bacf9e BS |
142 | int eof_index; |
143 | u64 last_cookie; | |
144 | struct nfs_cache_array_entry array[0]; | |
145 | }; | |
146 | ||
573c4e1e | 147 | typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); |
1da177e4 LT |
148 | typedef struct { |
149 | struct file *file; | |
150 | struct page *page; | |
23db8620 | 151 | struct dir_context *ctx; |
1da177e4 | 152 | unsigned long page_index; |
f0dd2136 | 153 | u64 *dir_cookie; |
0aded708 | 154 | u64 last_cookie; |
f0dd2136 | 155 | loff_t current_index; |
1da177e4 | 156 | decode_dirent_t decode; |
d1bacf9e | 157 | |
1f4eab7e | 158 | unsigned long timestamp; |
4704f0e2 | 159 | unsigned long gencount; |
d1bacf9e BS |
160 | unsigned int cache_entry_index; |
161 | unsigned int plus:1; | |
162 | unsigned int eof:1; | |
1da177e4 LT |
163 | } nfs_readdir_descriptor_t; |
164 | ||
d1bacf9e BS |
165 | /* |
166 | * The caller is responsible for calling nfs_readdir_release_array(page) | |
1da177e4 LT |
167 | */ |
168 | static | |
d1bacf9e BS |
169 | struct nfs_cache_array *nfs_readdir_get_array(struct page *page) |
170 | { | |
8cd51a0c | 171 | void *ptr; |
d1bacf9e BS |
172 | if (page == NULL) |
173 | return ERR_PTR(-EIO); | |
8cd51a0c TM |
174 | ptr = kmap(page); |
175 | if (ptr == NULL) | |
176 | return ERR_PTR(-ENOMEM); | |
177 | return ptr; | |
d1bacf9e BS |
178 | } |
179 | ||
180 | static | |
181 | void nfs_readdir_release_array(struct page *page) | |
182 | { | |
183 | kunmap(page); | |
184 | } | |
185 | ||
186 | /* | |
187 | * we are freeing strings created by nfs_add_to_readdir_array() | |
188 | */ | |
189 | static | |
11de3b11 | 190 | void nfs_readdir_clear_array(struct page *page) |
d1bacf9e | 191 | { |
11de3b11 | 192 | struct nfs_cache_array *array; |
d1bacf9e | 193 | int i; |
8cd51a0c | 194 | |
2b86ce2d | 195 | array = kmap_atomic(page); |
d1bacf9e BS |
196 | for (i = 0; i < array->size; i++) |
197 | kfree(array->array[i].string.name); | |
2b86ce2d | 198 | kunmap_atomic(array); |
d1bacf9e BS |
199 | } |
200 | ||
201 | /* | |
202 | * the caller is responsible for freeing qstr.name | |
203 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
204 | * nfs_clear_readdir_array() | |
205 | */ | |
206 | static | |
4a201d6e | 207 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
208 | { |
209 | string->len = len; | |
210 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
211 | if (string->name == NULL) |
212 | return -ENOMEM; | |
04e4bd1c CM |
213 | /* |
214 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
215 | * in a page cache page which kmemleak does not scan. | |
216 | */ | |
217 | kmemleak_not_leak(string->name); | |
4a201d6e TM |
218 | string->hash = full_name_hash(name, len); |
219 | return 0; | |
d1bacf9e BS |
220 | } |
221 | ||
222 | static | |
223 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
224 | { | |
225 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
4a201d6e TM |
226 | struct nfs_cache_array_entry *cache_entry; |
227 | int ret; | |
228 | ||
d1bacf9e BS |
229 | if (IS_ERR(array)) |
230 | return PTR_ERR(array); | |
3020093f TM |
231 | |
232 | cache_entry = &array->array[array->size]; | |
233 | ||
234 | /* Check that this entry lies within the page bounds */ | |
8cd51a0c | 235 | ret = -ENOSPC; |
3020093f | 236 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) |
4a201d6e | 237 | goto out; |
d1bacf9e | 238 | |
4a201d6e TM |
239 | cache_entry->cookie = entry->prev_cookie; |
240 | cache_entry->ino = entry->ino; | |
0b26a0bf | 241 | cache_entry->d_type = entry->d_type; |
4a201d6e TM |
242 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); |
243 | if (ret) | |
244 | goto out; | |
d1bacf9e | 245 | array->last_cookie = entry->cookie; |
8cd51a0c | 246 | array->size++; |
47c716cb | 247 | if (entry->eof != 0) |
d1bacf9e | 248 | array->eof_index = array->size; |
4a201d6e | 249 | out: |
d1bacf9e | 250 | nfs_readdir_release_array(page); |
4a201d6e | 251 | return ret; |
d1bacf9e BS |
252 | } |
253 | ||
254 | static | |
255 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
256 | { | |
23db8620 | 257 | loff_t diff = desc->ctx->pos - desc->current_index; |
d1bacf9e BS |
258 | unsigned int index; |
259 | ||
260 | if (diff < 0) | |
261 | goto out_eof; | |
262 | if (diff >= array->size) { | |
8cd51a0c | 263 | if (array->eof_index >= 0) |
d1bacf9e | 264 | goto out_eof; |
d1bacf9e BS |
265 | return -EAGAIN; |
266 | } | |
267 | ||
268 | index = (unsigned int)diff; | |
269 | *desc->dir_cookie = array->array[index].cookie; | |
270 | desc->cache_entry_index = index; | |
d1bacf9e BS |
271 | return 0; |
272 | out_eof: | |
273 | desc->eof = 1; | |
274 | return -EBADCOOKIE; | |
275 | } | |
276 | ||
277 | static | |
278 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
279 | { | |
280 | int i; | |
8ef2ce3e | 281 | loff_t new_pos; |
d1bacf9e BS |
282 | int status = -EAGAIN; |
283 | ||
284 | for (i = 0; i < array->size; i++) { | |
d1bacf9e | 285 | if (array->array[i].cookie == *desc->dir_cookie) { |
496ad9aa | 286 | struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); |
0c030806 TM |
287 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
288 | ||
8ef2ce3e | 289 | new_pos = desc->current_index + i; |
0c030806 TM |
290 | if (ctx->attr_gencount != nfsi->attr_gencount |
291 | || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) { | |
292 | ctx->duped = 0; | |
293 | ctx->attr_gencount = nfsi->attr_gencount; | |
23db8620 | 294 | } else if (new_pos < desc->ctx->pos) { |
0c030806 TM |
295 | if (ctx->duped > 0 |
296 | && ctx->dup_cookie == *desc->dir_cookie) { | |
297 | if (printk_ratelimit()) { | |
6de1472f | 298 | pr_notice("NFS: directory %pD2 contains a readdir loop." |
0c030806 | 299 | "Please contact your server vendor. " |
374e4e3e | 300 | "The file: %s has duplicate cookie %llu\n", |
6de1472f | 301 | desc->file, |
374e4e3e | 302 | array->array[i].string.name, |
0c030806 TM |
303 | *desc->dir_cookie); |
304 | } | |
305 | status = -ELOOP; | |
306 | goto out; | |
307 | } | |
8ef2ce3e | 308 | ctx->dup_cookie = *desc->dir_cookie; |
0c030806 | 309 | ctx->duped = -1; |
8ef2ce3e | 310 | } |
23db8620 | 311 | desc->ctx->pos = new_pos; |
d1bacf9e | 312 | desc->cache_entry_index = i; |
47c716cb | 313 | return 0; |
d1bacf9e BS |
314 | } |
315 | } | |
47c716cb | 316 | if (array->eof_index >= 0) { |
8cd51a0c | 317 | status = -EBADCOOKIE; |
18fb5fe4 TM |
318 | if (*desc->dir_cookie == array->last_cookie) |
319 | desc->eof = 1; | |
8cd51a0c | 320 | } |
0c030806 | 321 | out: |
d1bacf9e BS |
322 | return status; |
323 | } | |
324 | ||
325 | static | |
326 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
327 | { | |
328 | struct nfs_cache_array *array; | |
47c716cb | 329 | int status; |
d1bacf9e BS |
330 | |
331 | array = nfs_readdir_get_array(desc->page); | |
332 | if (IS_ERR(array)) { | |
333 | status = PTR_ERR(array); | |
334 | goto out; | |
335 | } | |
336 | ||
337 | if (*desc->dir_cookie == 0) | |
338 | status = nfs_readdir_search_for_pos(array, desc); | |
339 | else | |
340 | status = nfs_readdir_search_for_cookie(array, desc); | |
341 | ||
47c716cb | 342 | if (status == -EAGAIN) { |
0aded708 | 343 | desc->last_cookie = array->last_cookie; |
e47c085a | 344 | desc->current_index += array->size; |
47c716cb TM |
345 | desc->page_index++; |
346 | } | |
d1bacf9e BS |
347 | nfs_readdir_release_array(desc->page); |
348 | out: | |
349 | return status; | |
350 | } | |
351 | ||
352 | /* Fill a page with xdr information before transferring to the cache page */ | |
353 | static | |
56e4ebf8 | 354 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 355 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 356 | { |
480c2006 BS |
357 | struct nfs_open_dir_context *ctx = file->private_data; |
358 | struct rpc_cred *cred = ctx->cred; | |
4704f0e2 | 359 | unsigned long timestamp, gencount; |
1da177e4 LT |
360 | int error; |
361 | ||
1da177e4 LT |
362 | again: |
363 | timestamp = jiffies; | |
4704f0e2 | 364 | gencount = nfs_inc_attr_generation_counter(); |
56e4ebf8 | 365 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, |
1da177e4 LT |
366 | NFS_SERVER(inode)->dtsize, desc->plus); |
367 | if (error < 0) { | |
368 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
369 | if (error == -ENOTSUPP && desc->plus) { | |
370 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 371 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 LT |
372 | desc->plus = 0; |
373 | goto again; | |
374 | } | |
375 | goto error; | |
376 | } | |
1f4eab7e | 377 | desc->timestamp = timestamp; |
4704f0e2 | 378 | desc->gencount = gencount; |
d1bacf9e BS |
379 | error: |
380 | return error; | |
1da177e4 LT |
381 | } |
382 | ||
573c4e1e CL |
383 | static int xdr_decode(nfs_readdir_descriptor_t *desc, |
384 | struct nfs_entry *entry, struct xdr_stream *xdr) | |
1da177e4 | 385 | { |
573c4e1e | 386 | int error; |
1da177e4 | 387 | |
573c4e1e CL |
388 | error = desc->decode(xdr, entry, desc->plus); |
389 | if (error) | |
390 | return error; | |
d1bacf9e BS |
391 | entry->fattr->time_start = desc->timestamp; |
392 | entry->fattr->gencount = desc->gencount; | |
393 | return 0; | |
1da177e4 LT |
394 | } |
395 | ||
d39ab9de BS |
396 | static |
397 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
398 | { | |
d39ab9de BS |
399 | if (dentry->d_inode == NULL) |
400 | goto different; | |
37a09f07 | 401 | if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0) |
d39ab9de BS |
402 | goto different; |
403 | return 1; | |
404 | different: | |
405 | return 0; | |
406 | } | |
407 | ||
d69ee9b8 | 408 | static |
23db8620 | 409 | bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) |
d69ee9b8 TM |
410 | { |
411 | if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) | |
412 | return false; | |
413 | if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) | |
414 | return true; | |
23db8620 | 415 | if (ctx->pos == 0) |
d69ee9b8 TM |
416 | return true; |
417 | return false; | |
418 | } | |
419 | ||
420 | /* | |
421 | * This function is called by the lookup code to request the use of | |
422 | * readdirplus to accelerate any future lookups in the same | |
423 | * directory. | |
424 | */ | |
425 | static | |
426 | void nfs_advise_use_readdirplus(struct inode *dir) | |
427 | { | |
428 | set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); | |
429 | } | |
430 | ||
d39ab9de BS |
431 | static |
432 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
433 | { | |
26fe5750 | 434 | struct qstr filename = QSTR_INIT(entry->name, entry->len); |
4a201d6e TM |
435 | struct dentry *dentry; |
436 | struct dentry *alias; | |
d39ab9de BS |
437 | struct inode *dir = parent->d_inode; |
438 | struct inode *inode; | |
aa9c2669 | 439 | int status; |
d39ab9de | 440 | |
4a201d6e TM |
441 | if (filename.name[0] == '.') { |
442 | if (filename.len == 1) | |
443 | return; | |
444 | if (filename.len == 2 && filename.name[1] == '.') | |
445 | return; | |
446 | } | |
447 | filename.hash = full_name_hash(filename.name, filename.len); | |
d39ab9de | 448 | |
4a201d6e | 449 | dentry = d_lookup(parent, &filename); |
d39ab9de BS |
450 | if (dentry != NULL) { |
451 | if (nfs_same_file(dentry, entry)) { | |
cda57a1e | 452 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
aa9c2669 DQ |
453 | status = nfs_refresh_inode(dentry->d_inode, entry->fattr); |
454 | if (!status) | |
455 | nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label); | |
d39ab9de BS |
456 | goto out; |
457 | } else { | |
696199f8 AV |
458 | if (d_invalidate(dentry) != 0) |
459 | goto out; | |
d39ab9de BS |
460 | dput(dentry); |
461 | } | |
462 | } | |
463 | ||
464 | dentry = d_alloc(parent, &filename); | |
4a201d6e TM |
465 | if (dentry == NULL) |
466 | return; | |
467 | ||
1775fd3e | 468 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); |
d39ab9de BS |
469 | if (IS_ERR(inode)) |
470 | goto out; | |
471 | ||
472 | alias = d_materialise_unique(dentry, inode); | |
473 | if (IS_ERR(alias)) | |
474 | goto out; | |
475 | else if (alias) { | |
476 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); | |
477 | dput(alias); | |
478 | } else | |
479 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
480 | ||
481 | out: | |
482 | dput(dentry); | |
d39ab9de BS |
483 | } |
484 | ||
d1bacf9e BS |
485 | /* Perform conversion from xdr to cache array */ |
486 | static | |
8cd51a0c | 487 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
6650239a | 488 | struct page **xdr_pages, struct page *page, unsigned int buflen) |
1da177e4 | 489 | { |
babddc72 | 490 | struct xdr_stream stream; |
f7da7a12 | 491 | struct xdr_buf buf; |
6650239a | 492 | struct page *scratch; |
99424380 | 493 | struct nfs_cache_array *array; |
5c346854 TM |
494 | unsigned int count = 0; |
495 | int status; | |
babddc72 | 496 | |
6650239a TM |
497 | scratch = alloc_page(GFP_KERNEL); |
498 | if (scratch == NULL) | |
499 | return -ENOMEM; | |
babddc72 | 500 | |
f7da7a12 | 501 | xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); |
6650239a | 502 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
99424380 BS |
503 | |
504 | do { | |
505 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
506 | if (status != 0) { |
507 | if (status == -EAGAIN) | |
508 | status = 0; | |
99424380 | 509 | break; |
8cd51a0c | 510 | } |
99424380 | 511 | |
5c346854 TM |
512 | count++; |
513 | ||
47c716cb | 514 | if (desc->plus != 0) |
d39ab9de | 515 | nfs_prime_dcache(desc->file->f_path.dentry, entry); |
8cd51a0c TM |
516 | |
517 | status = nfs_readdir_add_to_array(entry, page); | |
518 | if (status != 0) | |
519 | break; | |
99424380 BS |
520 | } while (!entry->eof); |
521 | ||
47c716cb | 522 | if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { |
99424380 | 523 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
524 | if (!IS_ERR(array)) { |
525 | array->eof_index = array->size; | |
526 | status = 0; | |
527 | nfs_readdir_release_array(page); | |
5c346854 TM |
528 | } else |
529 | status = PTR_ERR(array); | |
1da177e4 | 530 | } |
6650239a TM |
531 | |
532 | put_page(scratch); | |
8cd51a0c | 533 | return status; |
56e4ebf8 BS |
534 | } |
535 | ||
536 | static | |
537 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) | |
538 | { | |
539 | unsigned int i; | |
540 | for (i = 0; i < npages; i++) | |
541 | put_page(pages[i]); | |
542 | } | |
543 | ||
544 | static | |
545 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, | |
546 | unsigned int npages) | |
547 | { | |
56e4ebf8 BS |
548 | nfs_readdir_free_pagearray(pages, npages); |
549 | } | |
550 | ||
551 | /* | |
552 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
553 | * to nfs_readdir_free_large_page | |
554 | */ | |
555 | static | |
6650239a | 556 | int nfs_readdir_large_page(struct page **pages, unsigned int npages) |
56e4ebf8 | 557 | { |
56e4ebf8 BS |
558 | unsigned int i; |
559 | ||
560 | for (i = 0; i < npages; i++) { | |
561 | struct page *page = alloc_page(GFP_KERNEL); | |
562 | if (page == NULL) | |
563 | goto out_freepages; | |
564 | pages[i] = page; | |
565 | } | |
6650239a | 566 | return 0; |
56e4ebf8 | 567 | |
56e4ebf8 BS |
568 | out_freepages: |
569 | nfs_readdir_free_pagearray(pages, i); | |
6650239a | 570 | return -ENOMEM; |
1da177e4 LT |
571 | } |
572 | ||
d1bacf9e BS |
573 | static |
574 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 575 | { |
56e4ebf8 BS |
576 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
577 | void *pages_ptr = NULL; | |
d1bacf9e BS |
578 | struct nfs_entry entry; |
579 | struct file *file = desc->file; | |
580 | struct nfs_cache_array *array; | |
8cd51a0c | 581 | int status = -ENOMEM; |
56e4ebf8 | 582 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
583 | |
584 | entry.prev_cookie = 0; | |
0aded708 | 585 | entry.cookie = desc->last_cookie; |
d1bacf9e BS |
586 | entry.eof = 0; |
587 | entry.fh = nfs_alloc_fhandle(); | |
588 | entry.fattr = nfs_alloc_fattr(); | |
573c4e1e | 589 | entry.server = NFS_SERVER(inode); |
d1bacf9e BS |
590 | if (entry.fh == NULL || entry.fattr == NULL) |
591 | goto out; | |
00a92642 | 592 | |
14c43f76 DQ |
593 | entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); |
594 | if (IS_ERR(entry.label)) { | |
595 | status = PTR_ERR(entry.label); | |
596 | goto out; | |
597 | } | |
598 | ||
d1bacf9e | 599 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
600 | if (IS_ERR(array)) { |
601 | status = PTR_ERR(array); | |
14c43f76 | 602 | goto out_label_free; |
8cd51a0c | 603 | } |
d1bacf9e BS |
604 | memset(array, 0, sizeof(struct nfs_cache_array)); |
605 | array->eof_index = -1; | |
00a92642 | 606 | |
6650239a TM |
607 | status = nfs_readdir_large_page(pages, array_size); |
608 | if (status < 0) | |
d1bacf9e BS |
609 | goto out_release_array; |
610 | do { | |
ac396128 | 611 | unsigned int pglen; |
56e4ebf8 | 612 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 613 | |
d1bacf9e | 614 | if (status < 0) |
00a92642 | 615 | break; |
ac396128 | 616 | pglen = status; |
6650239a | 617 | status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); |
8cd51a0c TM |
618 | if (status < 0) { |
619 | if (status == -ENOSPC) | |
620 | status = 0; | |
621 | break; | |
622 | } | |
623 | } while (array->eof_index < 0); | |
d1bacf9e | 624 | |
56e4ebf8 | 625 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
d1bacf9e BS |
626 | out_release_array: |
627 | nfs_readdir_release_array(page); | |
14c43f76 DQ |
628 | out_label_free: |
629 | nfs4_label_free(entry.label); | |
d1bacf9e BS |
630 | out: |
631 | nfs_free_fattr(entry.fattr); | |
632 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
633 | return status; |
634 | } | |
635 | ||
636 | /* | |
d1bacf9e BS |
637 | * Now we cache directories properly, by converting xdr information |
638 | * to an array that can be used for lookups later. This results in | |
639 | * fewer cache pages, since we can store more information on each page. | |
640 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 641 | */ |
d1bacf9e BS |
642 | static |
643 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 644 | { |
496ad9aa | 645 | struct inode *inode = file_inode(desc->file); |
8cd51a0c | 646 | int ret; |
1da177e4 | 647 | |
8cd51a0c TM |
648 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
649 | if (ret < 0) | |
d1bacf9e BS |
650 | goto error; |
651 | SetPageUptodate(page); | |
1da177e4 | 652 | |
d1bacf9e BS |
653 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
654 | /* Should never happen */ | |
655 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 656 | } |
d1bacf9e BS |
657 | unlock_page(page); |
658 | return 0; | |
659 | error: | |
660 | unlock_page(page); | |
8cd51a0c | 661 | return ret; |
d1bacf9e | 662 | } |
1da177e4 | 663 | |
d1bacf9e BS |
664 | static |
665 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
666 | { | |
11de3b11 TM |
667 | if (!desc->page->mapping) |
668 | nfs_readdir_clear_array(desc->page); | |
d1bacf9e BS |
669 | page_cache_release(desc->page); |
670 | desc->page = NULL; | |
671 | } | |
672 | ||
673 | static | |
674 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
675 | { | |
496ad9aa | 676 | return read_cache_page(file_inode(desc->file)->i_mapping, |
d1bacf9e | 677 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
1da177e4 LT |
678 | } |
679 | ||
680 | /* | |
d1bacf9e | 681 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 682 | */ |
d1bacf9e BS |
683 | static |
684 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
685 | { | |
686 | int res; | |
687 | ||
688 | desc->page = get_cache_page(desc); | |
689 | if (IS_ERR(desc->page)) | |
690 | return PTR_ERR(desc->page); | |
691 | ||
692 | res = nfs_readdir_search_array(desc); | |
47c716cb TM |
693 | if (res != 0) |
694 | cache_page_release(desc); | |
d1bacf9e BS |
695 | return res; |
696 | } | |
697 | ||
698 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
699 | static inline |
700 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
701 | { | |
8cd51a0c | 702 | int res; |
d1bacf9e | 703 | |
0aded708 | 704 | if (desc->page_index == 0) { |
8cd51a0c | 705 | desc->current_index = 0; |
0aded708 TM |
706 | desc->last_cookie = 0; |
707 | } | |
47c716cb | 708 | do { |
d1bacf9e | 709 | res = find_cache_page(desc); |
47c716cb | 710 | } while (res == -EAGAIN); |
1da177e4 LT |
711 | return res; |
712 | } | |
713 | ||
1da177e4 LT |
714 | /* |
715 | * Once we've found the start of the dirent within a page: fill 'er up... | |
716 | */ | |
717 | static | |
23db8620 | 718 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc) |
1da177e4 LT |
719 | { |
720 | struct file *file = desc->file; | |
d1bacf9e BS |
721 | int i = 0; |
722 | int res = 0; | |
723 | struct nfs_cache_array *array = NULL; | |
8ef2ce3e BS |
724 | struct nfs_open_dir_context *ctx = file->private_data; |
725 | ||
d1bacf9e | 726 | array = nfs_readdir_get_array(desc->page); |
e7c58e97 TM |
727 | if (IS_ERR(array)) { |
728 | res = PTR_ERR(array); | |
729 | goto out; | |
730 | } | |
d1bacf9e BS |
731 | |
732 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
ece0b423 | 733 | struct nfs_cache_array_entry *ent; |
1da177e4 | 734 | |
ece0b423 | 735 | ent = &array->array[i]; |
23db8620 AV |
736 | if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, |
737 | nfs_compat_user_ino64(ent->ino), ent->d_type)) { | |
ece0b423 | 738 | desc->eof = 1; |
1da177e4 | 739 | break; |
ece0b423 | 740 | } |
23db8620 | 741 | desc->ctx->pos++; |
d1bacf9e BS |
742 | if (i < (array->size-1)) |
743 | *desc->dir_cookie = array->array[i+1].cookie; | |
744 | else | |
745 | *desc->dir_cookie = array->last_cookie; | |
0c030806 TM |
746 | if (ctx->duped != 0) |
747 | ctx->duped = 1; | |
1da177e4 | 748 | } |
47c716cb | 749 | if (array->eof_index >= 0) |
8cd51a0c | 750 | desc->eof = 1; |
d1bacf9e BS |
751 | |
752 | nfs_readdir_release_array(desc->page); | |
e7c58e97 | 753 | out: |
d1bacf9e | 754 | cache_page_release(desc); |
1e7cb3dc CL |
755 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
756 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
757 | return res; |
758 | } | |
759 | ||
760 | /* | |
761 | * If we cannot find a cookie in our cache, we suspect that this is | |
762 | * because it points to a deleted file, so we ask the server to return | |
763 | * whatever it thinks is the next entry. We then feed this to filldir. | |
764 | * If all goes well, we should then be able to find our way round the | |
765 | * cache on the next call to readdir_search_pagecache(); | |
766 | * | |
767 | * NOTE: we cannot add the anonymous page to the pagecache because | |
768 | * the data it contains might not be page aligned. Besides, | |
769 | * we should already have a complete representation of the | |
770 | * directory in the page cache by the time we get here. | |
771 | */ | |
772 | static inline | |
23db8620 | 773 | int uncached_readdir(nfs_readdir_descriptor_t *desc) |
1da177e4 | 774 | { |
1da177e4 LT |
775 | struct page *page = NULL; |
776 | int status; | |
496ad9aa | 777 | struct inode *inode = file_inode(desc->file); |
0c030806 | 778 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
1da177e4 | 779 | |
1e7cb3dc CL |
780 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
781 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
782 | |
783 | page = alloc_page(GFP_HIGHUSER); | |
784 | if (!page) { | |
785 | status = -ENOMEM; | |
786 | goto out; | |
787 | } | |
d1bacf9e | 788 | |
7a8e1dc3 | 789 | desc->page_index = 0; |
0aded708 | 790 | desc->last_cookie = *desc->dir_cookie; |
7a8e1dc3 | 791 | desc->page = page; |
0c030806 | 792 | ctx->duped = 0; |
7a8e1dc3 | 793 | |
85f8607e TM |
794 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
795 | if (status < 0) | |
1da177e4 LT |
796 | goto out_release; |
797 | ||
23db8620 | 798 | status = nfs_do_filldir(desc); |
1da177e4 | 799 | |
1da177e4 | 800 | out: |
1e7cb3dc | 801 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 802 | __func__, status); |
1da177e4 LT |
803 | return status; |
804 | out_release: | |
d1bacf9e | 805 | cache_page_release(desc); |
1da177e4 LT |
806 | goto out; |
807 | } | |
808 | ||
00a92642 OG |
809 | /* The file offset position represents the dirent entry number. A |
810 | last cookie cache takes care of the common case of reading the | |
811 | whole directory. | |
1da177e4 | 812 | */ |
23db8620 | 813 | static int nfs_readdir(struct file *file, struct dir_context *ctx) |
1da177e4 | 814 | { |
23db8620 | 815 | struct dentry *dentry = file->f_path.dentry; |
1da177e4 LT |
816 | struct inode *inode = dentry->d_inode; |
817 | nfs_readdir_descriptor_t my_desc, | |
818 | *desc = &my_desc; | |
23db8620 | 819 | struct nfs_open_dir_context *dir_ctx = file->private_data; |
07b5ce8e | 820 | int res = 0; |
1da177e4 | 821 | |
6de1472f AV |
822 | dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", |
823 | file, (long long)ctx->pos); | |
91d5b470 CL |
824 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
825 | ||
1da177e4 | 826 | /* |
23db8620 | 827 | * ctx->pos points to the dirent entry number. |
f0dd2136 | 828 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
829 | * to either find the entry with the appropriate number or |
830 | * revalidate the cookie. | |
1da177e4 LT |
831 | */ |
832 | memset(desc, 0, sizeof(*desc)); | |
833 | ||
23db8620 AV |
834 | desc->file = file; |
835 | desc->ctx = ctx; | |
480c2006 | 836 | desc->dir_cookie = &dir_ctx->dir_cookie; |
1da177e4 | 837 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
23db8620 | 838 | desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; |
1da177e4 | 839 | |
565277f6 | 840 | nfs_block_sillyrename(dentry); |
07b5ce8e SM |
841 | if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) |
842 | res = nfs_revalidate_mapping(inode, file->f_mapping); | |
fccca7fc TM |
843 | if (res < 0) |
844 | goto out; | |
845 | ||
47c716cb | 846 | do { |
1da177e4 | 847 | res = readdir_search_pagecache(desc); |
00a92642 | 848 | |
1da177e4 | 849 | if (res == -EBADCOOKIE) { |
ece0b423 | 850 | res = 0; |
1da177e4 | 851 | /* This means either end of directory */ |
d1bacf9e | 852 | if (*desc->dir_cookie && desc->eof == 0) { |
1da177e4 | 853 | /* Or that the server has 'lost' a cookie */ |
23db8620 | 854 | res = uncached_readdir(desc); |
ece0b423 | 855 | if (res == 0) |
1da177e4 LT |
856 | continue; |
857 | } | |
1da177e4 LT |
858 | break; |
859 | } | |
860 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 861 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 862 | nfs_zap_caches(inode); |
baf57a09 | 863 | desc->page_index = 0; |
1da177e4 | 864 | desc->plus = 0; |
d1bacf9e | 865 | desc->eof = 0; |
1da177e4 LT |
866 | continue; |
867 | } | |
868 | if (res < 0) | |
869 | break; | |
870 | ||
23db8620 | 871 | res = nfs_do_filldir(desc); |
ece0b423 | 872 | if (res < 0) |
1da177e4 | 873 | break; |
47c716cb | 874 | } while (!desc->eof); |
fccca7fc | 875 | out: |
565277f6 | 876 | nfs_unblock_sillyrename(dentry); |
1e7cb3dc CL |
877 | if (res > 0) |
878 | res = 0; | |
6de1472f | 879 | dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); |
1e7cb3dc | 880 | return res; |
1da177e4 LT |
881 | } |
882 | ||
965c8e59 | 883 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) |
f0dd2136 | 884 | { |
6de1472f | 885 | struct inode *inode = file_inode(filp); |
480c2006 | 886 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
b84e06c5 | 887 | |
6de1472f AV |
888 | dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", |
889 | filp, offset, whence); | |
b84e06c5 CL |
890 | |
891 | mutex_lock(&inode->i_mutex); | |
965c8e59 | 892 | switch (whence) { |
f0dd2136 TM |
893 | case 1: |
894 | offset += filp->f_pos; | |
895 | case 0: | |
896 | if (offset >= 0) | |
897 | break; | |
898 | default: | |
899 | offset = -EINVAL; | |
900 | goto out; | |
901 | } | |
902 | if (offset != filp->f_pos) { | |
903 | filp->f_pos = offset; | |
480c2006 | 904 | dir_ctx->dir_cookie = 0; |
8ef2ce3e | 905 | dir_ctx->duped = 0; |
f0dd2136 TM |
906 | } |
907 | out: | |
b84e06c5 | 908 | mutex_unlock(&inode->i_mutex); |
f0dd2136 TM |
909 | return offset; |
910 | } | |
911 | ||
1da177e4 LT |
912 | /* |
913 | * All directory operations under NFS are synchronous, so fsync() | |
914 | * is a dummy operation. | |
915 | */ | |
02c24a82 JB |
916 | static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, |
917 | int datasync) | |
1da177e4 | 918 | { |
6de1472f | 919 | struct inode *inode = file_inode(filp); |
7ea80859 | 920 | |
6de1472f | 921 | dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); |
1e7cb3dc | 922 | |
02c24a82 | 923 | mutex_lock(&inode->i_mutex); |
6de1472f | 924 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
02c24a82 | 925 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
926 | return 0; |
927 | } | |
928 | ||
bfc69a45 TM |
929 | /** |
930 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
931 | * @dir - pointer to directory inode | |
932 | * | |
933 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
934 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
935 | * on the server that might have invalidated our dcache. | |
936 | * | |
937 | * The caller should be holding dir->i_lock | |
938 | */ | |
939 | void nfs_force_lookup_revalidate(struct inode *dir) | |
940 | { | |
011935a0 | 941 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 | 942 | } |
89d77c8f | 943 | EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); |
bfc69a45 | 944 | |
1da177e4 LT |
945 | /* |
946 | * A check for whether or not the parent directory has changed. | |
947 | * In the case it has, we assume that the dentries are untrustworthy | |
948 | * and may need to be looked up again. | |
949 | */ | |
c79ba787 | 950 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
951 | { |
952 | if (IS_ROOT(dentry)) | |
953 | return 1; | |
4eec952e TM |
954 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
955 | return 0; | |
f2c77f4e TM |
956 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
957 | return 0; | |
958 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
959 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
960 | return 0; | |
961 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
962 | return 0; | |
963 | return 1; | |
1da177e4 LT |
964 | } |
965 | ||
a12802ca TM |
966 | /* |
967 | * Use intent information to check whether or not we're going to do | |
968 | * an O_EXCL create using this path component. | |
969 | */ | |
fa3c56bb | 970 | static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) |
a12802ca TM |
971 | { |
972 | if (NFS_PROTO(dir)->version == 2) | |
973 | return 0; | |
fa3c56bb | 974 | return flags & LOOKUP_EXCL; |
a12802ca TM |
975 | } |
976 | ||
1d6757fb TM |
977 | /* |
978 | * Inode and filehandle revalidation for lookups. | |
979 | * | |
980 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
981 | * or if the intent information indicates that we're about to open this | |
982 | * particular file and the "nocto" mount flag is not set. | |
983 | * | |
984 | */ | |
65a0c149 | 985 | static |
fa3c56bb | 986 | int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) |
1da177e4 LT |
987 | { |
988 | struct nfs_server *server = NFS_SERVER(inode); | |
65a0c149 | 989 | int ret; |
1da177e4 | 990 | |
36d43a43 | 991 | if (IS_AUTOMOUNT(inode)) |
4e99a1ff | 992 | return 0; |
facc3530 | 993 | /* VFS wants an on-the-wire revalidation */ |
fa3c56bb | 994 | if (flags & LOOKUP_REVAL) |
facc3530 AV |
995 | goto out_force; |
996 | /* This is an open(2) */ | |
fa3c56bb AV |
997 | if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && |
998 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) | |
facc3530 | 999 | goto out_force; |
65a0c149 TM |
1000 | out: |
1001 | return (inode->i_nlink == 0) ? -ENOENT : 0; | |
1da177e4 | 1002 | out_force: |
65a0c149 TM |
1003 | ret = __nfs_revalidate_inode(server, inode); |
1004 | if (ret != 0) | |
1005 | return ret; | |
1006 | goto out; | |
1da177e4 LT |
1007 | } |
1008 | ||
1009 | /* | |
1010 | * We judge how long we want to trust negative | |
1011 | * dentries by looking at the parent inode mtime. | |
1012 | * | |
1013 | * If parent mtime has changed, we revalidate, else we wait for a | |
1014 | * period corresponding to the parent's attribute cache timeout value. | |
1015 | */ | |
1016 | static inline | |
1017 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
fa3c56bb | 1018 | unsigned int flags) |
1da177e4 | 1019 | { |
1da177e4 | 1020 | /* Don't revalidate a negative dentry if we're creating a new file */ |
fa3c56bb | 1021 | if (flags & LOOKUP_CREATE) |
1da177e4 | 1022 | return 0; |
4eec952e TM |
1023 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1024 | return 1; | |
1da177e4 LT |
1025 | return !nfs_check_verifier(dir, dentry); |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * This is called every time the dcache has a lookup hit, | |
1030 | * and we should check whether we can really trust that | |
1031 | * lookup. | |
1032 | * | |
1033 | * NOTE! The hit can be a negative hit too, don't assume | |
1034 | * we have an inode! | |
1035 | * | |
1036 | * If the parent directory is seen to have changed, we throw out the | |
1037 | * cached dentry and do a new lookup. | |
1038 | */ | |
0b728e19 | 1039 | static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
1da177e4 LT |
1040 | { |
1041 | struct inode *dir; | |
1042 | struct inode *inode; | |
1043 | struct dentry *parent; | |
e1fb4d05 TM |
1044 | struct nfs_fh *fhandle = NULL; |
1045 | struct nfs_fattr *fattr = NULL; | |
1775fd3e | 1046 | struct nfs4_label *label = NULL; |
1da177e4 | 1047 | int error; |
1da177e4 | 1048 | |
fa3c56bb | 1049 | if (flags & LOOKUP_RCU) |
34286d66 NP |
1050 | return -ECHILD; |
1051 | ||
1da177e4 | 1052 | parent = dget_parent(dentry); |
1da177e4 | 1053 | dir = parent->d_inode; |
91d5b470 | 1054 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
1055 | inode = dentry->d_inode; |
1056 | ||
1057 | if (!inode) { | |
fa3c56bb | 1058 | if (nfs_neg_need_reval(dir, dentry, flags)) |
1da177e4 | 1059 | goto out_bad; |
d69ee9b8 | 1060 | goto out_valid_noent; |
1da177e4 LT |
1061 | } |
1062 | ||
1063 | if (is_bad_inode(inode)) { | |
6de1472f AV |
1064 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1065 | __func__, dentry); | |
1da177e4 LT |
1066 | goto out_bad; |
1067 | } | |
1068 | ||
011e2a7f | 1069 | if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) |
15860ab1 TM |
1070 | goto out_set_verifier; |
1071 | ||
1da177e4 | 1072 | /* Force a full look up iff the parent directory has changed */ |
fa3c56bb AV |
1073 | if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) { |
1074 | if (nfs_lookup_verify_inode(inode, flags)) | |
1da177e4 LT |
1075 | goto out_zap_parent; |
1076 | goto out_valid; | |
1077 | } | |
1078 | ||
1079 | if (NFS_STALE(inode)) | |
1080 | goto out_bad; | |
1081 | ||
e1fb4d05 TM |
1082 | error = -ENOMEM; |
1083 | fhandle = nfs_alloc_fhandle(); | |
1084 | fattr = nfs_alloc_fattr(); | |
1085 | if (fhandle == NULL || fattr == NULL) | |
1086 | goto out_error; | |
1087 | ||
14c43f76 DQ |
1088 | label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); |
1089 | if (IS_ERR(label)) | |
1090 | goto out_error; | |
1091 | ||
6e0d0be7 | 1092 | trace_nfs_lookup_revalidate_enter(dir, dentry, flags); |
1775fd3e | 1093 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); |
6e0d0be7 | 1094 | trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); |
1da177e4 LT |
1095 | if (error) |
1096 | goto out_bad; | |
e1fb4d05 | 1097 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1098 | goto out_bad; |
e1fb4d05 | 1099 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1100 | goto out_bad; |
1101 | ||
aa9c2669 DQ |
1102 | nfs_setsecurity(inode, fattr, label); |
1103 | ||
e1fb4d05 TM |
1104 | nfs_free_fattr(fattr); |
1105 | nfs_free_fhandle(fhandle); | |
14c43f76 DQ |
1106 | nfs4_label_free(label); |
1107 | ||
15860ab1 | 1108 | out_set_verifier: |
cf8ba45e | 1109 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1110 | out_valid: |
d69ee9b8 TM |
1111 | /* Success: notify readdir to use READDIRPLUS */ |
1112 | nfs_advise_use_readdirplus(dir); | |
1113 | out_valid_noent: | |
1da177e4 | 1114 | dput(parent); |
6de1472f AV |
1115 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", |
1116 | __func__, dentry); | |
1da177e4 LT |
1117 | return 1; |
1118 | out_zap_parent: | |
1119 | nfs_zap_caches(dir); | |
1120 | out_bad: | |
c44600c9 AV |
1121 | nfs_free_fattr(fattr); |
1122 | nfs_free_fhandle(fhandle); | |
14c43f76 | 1123 | nfs4_label_free(label); |
a1643a92 | 1124 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1125 | if (inode && S_ISDIR(inode->i_mode)) { |
1126 | /* Purge readdir caches. */ | |
1127 | nfs_zap_caches(inode); | |
a3f432bf BF |
1128 | /* |
1129 | * We can't d_drop the root of a disconnected tree: | |
1130 | * its d_hash is on the s_anon list and d_drop() would hide | |
1131 | * it from shrink_dcache_for_unmount(), leading to busy | |
1132 | * inodes on unmount and further oopses. | |
1133 | */ | |
1134 | if (IS_ROOT(dentry)) | |
d9e80b7d | 1135 | goto out_valid; |
1da177e4 | 1136 | } |
13caa9fb MS |
1137 | /* If we have submounts, don't unhash ! */ |
1138 | if (check_submounts_and_drop(dentry) != 0) | |
1139 | goto out_valid; | |
1140 | ||
1da177e4 | 1141 | dput(parent); |
6de1472f AV |
1142 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", |
1143 | __func__, dentry); | |
1da177e4 | 1144 | return 0; |
e1fb4d05 TM |
1145 | out_error: |
1146 | nfs_free_fattr(fattr); | |
1147 | nfs_free_fhandle(fhandle); | |
14c43f76 | 1148 | nfs4_label_free(label); |
e1fb4d05 | 1149 | dput(parent); |
6de1472f AV |
1150 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", |
1151 | __func__, dentry, error); | |
e1fb4d05 | 1152 | return error; |
1da177e4 LT |
1153 | } |
1154 | ||
ecf3d1f1 JL |
1155 | /* |
1156 | * A weaker form of d_revalidate for revalidating just the dentry->d_inode | |
1157 | * when we don't really care about the dentry name. This is called when a | |
1158 | * pathwalk ends on a dentry that was not found via a normal lookup in the | |
1159 | * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). | |
1160 | * | |
1161 | * In this situation, we just want to verify that the inode itself is OK | |
1162 | * since the dentry might have changed on the server. | |
1163 | */ | |
1164 | static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) | |
1165 | { | |
1166 | int error; | |
1167 | struct inode *inode = dentry->d_inode; | |
1168 | ||
1169 | /* | |
1170 | * I believe we can only get a negative dentry here in the case of a | |
1171 | * procfs-style symlink. Just assume it's correct for now, but we may | |
1172 | * eventually need to do something more here. | |
1173 | */ | |
1174 | if (!inode) { | |
6de1472f AV |
1175 | dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", |
1176 | __func__, dentry); | |
ecf3d1f1 JL |
1177 | return 1; |
1178 | } | |
1179 | ||
1180 | if (is_bad_inode(inode)) { | |
6de1472f AV |
1181 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1182 | __func__, dentry); | |
ecf3d1f1 JL |
1183 | return 0; |
1184 | } | |
1185 | ||
1186 | error = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
1187 | dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", | |
1188 | __func__, inode->i_ino, error ? "invalid" : "valid"); | |
1189 | return !error; | |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* |
1193 | * This is called from dput() when d_count is going to 0. | |
1194 | */ | |
fe15ce44 | 1195 | static int nfs_dentry_delete(const struct dentry *dentry) |
1da177e4 | 1196 | { |
6de1472f AV |
1197 | dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", |
1198 | dentry, dentry->d_flags); | |
1da177e4 | 1199 | |
77f11192 TM |
1200 | /* Unhash any dentry with a stale inode */ |
1201 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) | |
1202 | return 1; | |
1203 | ||
1da177e4 LT |
1204 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1205 | /* Unhash it, so that ->d_iput() would be called */ | |
1206 | return 1; | |
1207 | } | |
1208 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1209 | /* Unhash it, so that ancestors of killed async unlink | |
1210 | * files will be cleaned up during umount */ | |
1211 | return 1; | |
1212 | } | |
1213 | return 0; | |
1214 | ||
1215 | } | |
1216 | ||
1f018458 | 1217 | /* Ensure that we revalidate inode->i_nlink */ |
1b83d707 TM |
1218 | static void nfs_drop_nlink(struct inode *inode) |
1219 | { | |
1220 | spin_lock(&inode->i_lock); | |
1f018458 TM |
1221 | /* drop the inode if we're reasonably sure this is the last link */ |
1222 | if (inode->i_nlink == 1) | |
1223 | clear_nlink(inode); | |
1224 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; | |
1b83d707 TM |
1225 | spin_unlock(&inode->i_lock); |
1226 | } | |
1227 | ||
1da177e4 LT |
1228 | /* |
1229 | * Called when the dentry loses inode. | |
1230 | * We use it to clean up silly-renamed files. | |
1231 | */ | |
1232 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1233 | { | |
83672d39 NB |
1234 | if (S_ISDIR(inode->i_mode)) |
1235 | /* drop any readdir cache as it could easily be old */ | |
1236 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1237 | ||
1da177e4 | 1238 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
e4eff1a6 | 1239 | nfs_complete_unlink(dentry, inode); |
1f018458 | 1240 | nfs_drop_nlink(inode); |
1da177e4 | 1241 | } |
1da177e4 LT |
1242 | iput(inode); |
1243 | } | |
1244 | ||
b1942c5f AV |
1245 | static void nfs_d_release(struct dentry *dentry) |
1246 | { | |
1247 | /* free cached devname value, if it survived that far */ | |
1248 | if (unlikely(dentry->d_fsdata)) { | |
1249 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1250 | WARN_ON(1); | |
1251 | else | |
1252 | kfree(dentry->d_fsdata); | |
1253 | } | |
1254 | } | |
1255 | ||
f786aa90 | 1256 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 | 1257 | .d_revalidate = nfs_lookup_revalidate, |
ecf3d1f1 | 1258 | .d_weak_revalidate = nfs_weak_revalidate, |
1da177e4 LT |
1259 | .d_delete = nfs_dentry_delete, |
1260 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1261 | .d_automount = nfs_d_automount, |
b1942c5f | 1262 | .d_release = nfs_d_release, |
1da177e4 | 1263 | }; |
ddda8e0a | 1264 | EXPORT_SYMBOL_GPL(nfs_dentry_operations); |
1da177e4 | 1265 | |
597d9289 | 1266 | struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) |
1da177e4 LT |
1267 | { |
1268 | struct dentry *res; | |
565277f6 | 1269 | struct dentry *parent; |
1da177e4 | 1270 | struct inode *inode = NULL; |
e1fb4d05 TM |
1271 | struct nfs_fh *fhandle = NULL; |
1272 | struct nfs_fattr *fattr = NULL; | |
1775fd3e | 1273 | struct nfs4_label *label = NULL; |
1da177e4 | 1274 | int error; |
1da177e4 | 1275 | |
6de1472f | 1276 | dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); |
91d5b470 | 1277 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
1278 | |
1279 | res = ERR_PTR(-ENAMETOOLONG); | |
1280 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1281 | goto out; | |
1282 | ||
fd684071 TM |
1283 | /* |
1284 | * If we're doing an exclusive create, optimize away the lookup | |
1285 | * but don't hash the dentry. | |
1286 | */ | |
00cd8dd3 | 1287 | if (nfs_is_exclusive_create(dir, flags)) { |
fd684071 TM |
1288 | d_instantiate(dentry, NULL); |
1289 | res = NULL; | |
fc0f684c | 1290 | goto out; |
fd684071 | 1291 | } |
1da177e4 | 1292 | |
e1fb4d05 TM |
1293 | res = ERR_PTR(-ENOMEM); |
1294 | fhandle = nfs_alloc_fhandle(); | |
1295 | fattr = nfs_alloc_fattr(); | |
1296 | if (fhandle == NULL || fattr == NULL) | |
1297 | goto out; | |
1298 | ||
14c43f76 DQ |
1299 | label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); |
1300 | if (IS_ERR(label)) | |
1301 | goto out; | |
1302 | ||
565277f6 TM |
1303 | parent = dentry->d_parent; |
1304 | /* Protect against concurrent sillydeletes */ | |
6e0d0be7 | 1305 | trace_nfs_lookup_enter(dir, dentry, flags); |
565277f6 | 1306 | nfs_block_sillyrename(parent); |
1775fd3e | 1307 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); |
1da177e4 LT |
1308 | if (error == -ENOENT) |
1309 | goto no_entry; | |
1310 | if (error < 0) { | |
1311 | res = ERR_PTR(error); | |
565277f6 | 1312 | goto out_unblock_sillyrename; |
1da177e4 | 1313 | } |
1775fd3e | 1314 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
bf0c84f1 | 1315 | res = ERR_CAST(inode); |
03f28e3a | 1316 | if (IS_ERR(res)) |
565277f6 | 1317 | goto out_unblock_sillyrename; |
54ceac45 | 1318 | |
d69ee9b8 TM |
1319 | /* Success: notify readdir to use READDIRPLUS */ |
1320 | nfs_advise_use_readdirplus(dir); | |
1321 | ||
1da177e4 | 1322 | no_entry: |
54ceac45 | 1323 | res = d_materialise_unique(dentry, inode); |
9eaef27b TM |
1324 | if (res != NULL) { |
1325 | if (IS_ERR(res)) | |
565277f6 | 1326 | goto out_unblock_sillyrename; |
1da177e4 | 1327 | dentry = res; |
9eaef27b | 1328 | } |
1da177e4 | 1329 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
565277f6 TM |
1330 | out_unblock_sillyrename: |
1331 | nfs_unblock_sillyrename(parent); | |
6e0d0be7 | 1332 | trace_nfs_lookup_exit(dir, dentry, flags, error); |
14c43f76 | 1333 | nfs4_label_free(label); |
1da177e4 | 1334 | out: |
e1fb4d05 TM |
1335 | nfs_free_fattr(fattr); |
1336 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1337 | return res; |
1338 | } | |
ddda8e0a | 1339 | EXPORT_SYMBOL_GPL(nfs_lookup); |
1da177e4 | 1340 | |
89d77c8f | 1341 | #if IS_ENABLED(CONFIG_NFS_V4) |
0b728e19 | 1342 | static int nfs4_lookup_revalidate(struct dentry *, unsigned int); |
1da177e4 | 1343 | |
f786aa90 | 1344 | const struct dentry_operations nfs4_dentry_operations = { |
0ef97dcf | 1345 | .d_revalidate = nfs4_lookup_revalidate, |
1da177e4 LT |
1346 | .d_delete = nfs_dentry_delete, |
1347 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1348 | .d_automount = nfs_d_automount, |
b1942c5f | 1349 | .d_release = nfs_d_release, |
1da177e4 | 1350 | }; |
89d77c8f | 1351 | EXPORT_SYMBOL_GPL(nfs4_dentry_operations); |
1da177e4 | 1352 | |
8a5e929d AV |
1353 | static fmode_t flags_to_mode(int flags) |
1354 | { | |
1355 | fmode_t res = (__force fmode_t)flags & FMODE_EXEC; | |
1356 | if ((flags & O_ACCMODE) != O_WRONLY) | |
1357 | res |= FMODE_READ; | |
1358 | if ((flags & O_ACCMODE) != O_RDONLY) | |
1359 | res |= FMODE_WRITE; | |
1360 | return res; | |
1361 | } | |
1362 | ||
51141598 | 1363 | static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) |
cd9a1c0e | 1364 | { |
5ede7b1c | 1365 | return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); |
cd9a1c0e TM |
1366 | } |
1367 | ||
1368 | static int do_open(struct inode *inode, struct file *filp) | |
1369 | { | |
f1fe29b4 | 1370 | nfs_fscache_open_file(inode, filp); |
cd9a1c0e TM |
1371 | return 0; |
1372 | } | |
1373 | ||
d9585277 AV |
1374 | static int nfs_finish_open(struct nfs_open_context *ctx, |
1375 | struct dentry *dentry, | |
30d90494 | 1376 | struct file *file, unsigned open_flags, |
d9585277 | 1377 | int *opened) |
cd9a1c0e | 1378 | { |
0dd2b474 MS |
1379 | int err; |
1380 | ||
01c919ab MS |
1381 | if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) |
1382 | *opened |= FILE_CREATED; | |
1383 | ||
30d90494 AV |
1384 | err = finish_open(file, dentry, do_open, opened); |
1385 | if (err) | |
d9585277 | 1386 | goto out; |
30d90494 | 1387 | nfs_file_set_open_context(file, ctx); |
0dd2b474 | 1388 | |
cd9a1c0e | 1389 | out: |
d9585277 | 1390 | return err; |
cd9a1c0e TM |
1391 | } |
1392 | ||
73a79706 BS |
1393 | int nfs_atomic_open(struct inode *dir, struct dentry *dentry, |
1394 | struct file *file, unsigned open_flags, | |
1395 | umode_t mode, int *opened) | |
1da177e4 | 1396 | { |
cd9a1c0e | 1397 | struct nfs_open_context *ctx; |
0dd2b474 MS |
1398 | struct dentry *res; |
1399 | struct iattr attr = { .ia_valid = ATTR_OPEN }; | |
f46e0bd3 | 1400 | struct inode *inode; |
1472b83e | 1401 | unsigned int lookup_flags = 0; |
898f635c | 1402 | int err; |
1da177e4 | 1403 | |
0dd2b474 MS |
1404 | /* Expect a negative dentry */ |
1405 | BUG_ON(dentry->d_inode); | |
1406 | ||
6de1472f AV |
1407 | dfprintk(VFS, "NFS: atomic_open(%s/%ld), %pd\n", |
1408 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1e7cb3dc | 1409 | |
9597c13b JL |
1410 | err = nfs_check_flags(open_flags); |
1411 | if (err) | |
1412 | return err; | |
1413 | ||
0dd2b474 MS |
1414 | /* NFS only supports OPEN on regular files */ |
1415 | if ((open_flags & O_DIRECTORY)) { | |
0dd2b474 MS |
1416 | if (!d_unhashed(dentry)) { |
1417 | /* | |
1418 | * Hashed negative dentry with O_DIRECTORY: dentry was | |
1419 | * revalidated and is fine, no need to perform lookup | |
1420 | * again | |
1421 | */ | |
d9585277 | 1422 | return -ENOENT; |
0dd2b474 | 1423 | } |
1472b83e | 1424 | lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; |
1da177e4 | 1425 | goto no_open; |
02a913a7 | 1426 | } |
1da177e4 | 1427 | |
0dd2b474 | 1428 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) |
d9585277 | 1429 | return -ENAMETOOLONG; |
cd9a1c0e | 1430 | |
0dd2b474 | 1431 | if (open_flags & O_CREAT) { |
536e43d1 | 1432 | attr.ia_valid |= ATTR_MODE; |
0dd2b474 MS |
1433 | attr.ia_mode = mode & ~current_umask(); |
1434 | } | |
536e43d1 TM |
1435 | if (open_flags & O_TRUNC) { |
1436 | attr.ia_valid |= ATTR_SIZE; | |
1437 | attr.ia_size = 0; | |
cd9a1c0e TM |
1438 | } |
1439 | ||
0dd2b474 MS |
1440 | ctx = create_nfs_open_context(dentry, open_flags); |
1441 | err = PTR_ERR(ctx); | |
1442 | if (IS_ERR(ctx)) | |
d9585277 | 1443 | goto out; |
0dd2b474 | 1444 | |
6e0d0be7 | 1445 | trace_nfs_atomic_open_enter(dir, ctx, open_flags); |
f46e0bd3 | 1446 | nfs_block_sillyrename(dentry->d_parent); |
5bc2afc2 | 1447 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); |
275bb307 | 1448 | nfs_unblock_sillyrename(dentry->d_parent); |
f46e0bd3 | 1449 | if (IS_ERR(inode)) { |
0dd2b474 | 1450 | err = PTR_ERR(inode); |
6e0d0be7 | 1451 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 1452 | put_nfs_open_context(ctx); |
0dd2b474 MS |
1453 | switch (err) { |
1454 | case -ENOENT: | |
275bb307 | 1455 | d_drop(dentry); |
0dd2b474 MS |
1456 | d_add(dentry, NULL); |
1457 | break; | |
1458 | case -EISDIR: | |
1459 | case -ENOTDIR: | |
1460 | goto no_open; | |
1461 | case -ELOOP: | |
1462 | if (!(open_flags & O_NOFOLLOW)) | |
6f926b5b | 1463 | goto no_open; |
0dd2b474 | 1464 | break; |
1da177e4 | 1465 | /* case -EINVAL: */ |
0dd2b474 MS |
1466 | default: |
1467 | break; | |
1da177e4 | 1468 | } |
d9585277 | 1469 | goto out; |
cd9a1c0e | 1470 | } |
0dd2b474 | 1471 | |
275bb307 | 1472 | err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); |
6e0d0be7 | 1473 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 1474 | put_nfs_open_context(ctx); |
d9585277 AV |
1475 | out: |
1476 | return err; | |
0dd2b474 | 1477 | |
1da177e4 | 1478 | no_open: |
1472b83e | 1479 | res = nfs_lookup(dir, dentry, lookup_flags); |
0dd2b474 MS |
1480 | err = PTR_ERR(res); |
1481 | if (IS_ERR(res)) | |
d9585277 | 1482 | goto out; |
0dd2b474 | 1483 | |
e45198a6 | 1484 | return finish_no_open(file, res); |
1da177e4 | 1485 | } |
89d77c8f | 1486 | EXPORT_SYMBOL_GPL(nfs_atomic_open); |
1da177e4 | 1487 | |
0b728e19 | 1488 | static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
1da177e4 LT |
1489 | { |
1490 | struct dentry *parent = NULL; | |
657e94b6 | 1491 | struct inode *inode; |
1da177e4 | 1492 | struct inode *dir; |
50de348c | 1493 | int ret = 0; |
1da177e4 | 1494 | |
fa3c56bb | 1495 | if (flags & LOOKUP_RCU) |
657e94b6 NP |
1496 | return -ECHILD; |
1497 | ||
fa3c56bb | 1498 | if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) |
eda72afb MS |
1499 | goto no_open; |
1500 | if (d_mountpoint(dentry)) | |
5584c306 | 1501 | goto no_open; |
49f9a0fa TM |
1502 | if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) |
1503 | goto no_open; | |
2b484297 | 1504 | |
eda72afb | 1505 | inode = dentry->d_inode; |
1da177e4 LT |
1506 | parent = dget_parent(dentry); |
1507 | dir = parent->d_inode; | |
2b484297 | 1508 | |
1da177e4 LT |
1509 | /* We can't create new files in nfs_open_revalidate(), so we |
1510 | * optimize away revalidation of negative dentries. | |
1511 | */ | |
216d5d06 | 1512 | if (inode == NULL) { |
fa3c56bb | 1513 | if (!nfs_neg_need_reval(dir, dentry, flags)) |
216d5d06 | 1514 | ret = 1; |
1da177e4 | 1515 | goto out; |
216d5d06 TM |
1516 | } |
1517 | ||
1da177e4 LT |
1518 | /* NFS only supports OPEN on regular files */ |
1519 | if (!S_ISREG(inode->i_mode)) | |
5584c306 | 1520 | goto no_open_dput; |
1da177e4 | 1521 | /* We cannot do exclusive creation on a positive dentry */ |
fa3c56bb | 1522 | if (flags & LOOKUP_EXCL) |
5584c306 | 1523 | goto no_open_dput; |
1da177e4 | 1524 | |
0ef97dcf MS |
1525 | /* Let f_op->open() actually open (and revalidate) the file */ |
1526 | ret = 1; | |
536e43d1 | 1527 | |
1da177e4 LT |
1528 | out: |
1529 | dput(parent); | |
1da177e4 | 1530 | return ret; |
535918f1 | 1531 | |
5584c306 | 1532 | no_open_dput: |
1da177e4 | 1533 | dput(parent); |
5584c306 | 1534 | no_open: |
0b728e19 | 1535 | return nfs_lookup_revalidate(dentry, flags); |
c0204fd2 TM |
1536 | } |
1537 | ||
1da177e4 LT |
1538 | #endif /* CONFIG_NFSV4 */ |
1539 | ||
1da177e4 LT |
1540 | /* |
1541 | * Code common to create, mkdir, and mknod. | |
1542 | */ | |
1543 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1775fd3e DQ |
1544 | struct nfs_fattr *fattr, |
1545 | struct nfs4_label *label) | |
1da177e4 | 1546 | { |
fab728e1 TM |
1547 | struct dentry *parent = dget_parent(dentry); |
1548 | struct inode *dir = parent->d_inode; | |
1da177e4 LT |
1549 | struct inode *inode; |
1550 | int error = -EACCES; | |
1551 | ||
fab728e1 TM |
1552 | d_drop(dentry); |
1553 | ||
1da177e4 LT |
1554 | /* We may have been initialized further down */ |
1555 | if (dentry->d_inode) | |
fab728e1 | 1556 | goto out; |
1da177e4 | 1557 | if (fhandle->size == 0) { |
1775fd3e | 1558 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); |
1da177e4 | 1559 | if (error) |
fab728e1 | 1560 | goto out_error; |
1da177e4 | 1561 | } |
5724ab37 | 1562 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1563 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1564 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
1775fd3e | 1565 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); |
1da177e4 | 1566 | if (error < 0) |
fab728e1 | 1567 | goto out_error; |
1da177e4 | 1568 | } |
1775fd3e | 1569 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
03f28e3a TM |
1570 | error = PTR_ERR(inode); |
1571 | if (IS_ERR(inode)) | |
fab728e1 TM |
1572 | goto out_error; |
1573 | d_add(dentry, inode); | |
1574 | out: | |
1575 | dput(parent); | |
1da177e4 | 1576 | return 0; |
fab728e1 TM |
1577 | out_error: |
1578 | nfs_mark_for_revalidate(dir); | |
1579 | dput(parent); | |
1580 | return error; | |
1da177e4 | 1581 | } |
ddda8e0a | 1582 | EXPORT_SYMBOL_GPL(nfs_instantiate); |
1da177e4 LT |
1583 | |
1584 | /* | |
1585 | * Following a failed create operation, we drop the dentry rather | |
1586 | * than retain a negative dentry. This avoids a problem in the event | |
1587 | * that the operation succeeded on the server, but an error in the | |
1588 | * reply path made it appear to have failed. | |
1589 | */ | |
597d9289 | 1590 | int nfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 1591 | umode_t mode, bool excl) |
1da177e4 LT |
1592 | { |
1593 | struct iattr attr; | |
ebfc3b49 | 1594 | int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; |
1da177e4 | 1595 | int error; |
1da177e4 | 1596 | |
6de1472f AV |
1597 | dfprintk(VFS, "NFS: create(%s/%ld), %pd\n", |
1598 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1da177e4 LT |
1599 | |
1600 | attr.ia_mode = mode; | |
1601 | attr.ia_valid = ATTR_MODE; | |
1602 | ||
8b0ad3d4 | 1603 | trace_nfs_create_enter(dir, dentry, open_flags); |
8867fe58 | 1604 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); |
8b0ad3d4 | 1605 | trace_nfs_create_exit(dir, dentry, open_flags, error); |
1da177e4 LT |
1606 | if (error != 0) |
1607 | goto out_err; | |
1da177e4 LT |
1608 | return 0; |
1609 | out_err: | |
1da177e4 LT |
1610 | d_drop(dentry); |
1611 | return error; | |
1612 | } | |
ddda8e0a | 1613 | EXPORT_SYMBOL_GPL(nfs_create); |
1da177e4 LT |
1614 | |
1615 | /* | |
1616 | * See comments for nfs_proc_create regarding failed operations. | |
1617 | */ | |
597d9289 | 1618 | int |
1a67aafb | 1619 | nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) |
1da177e4 LT |
1620 | { |
1621 | struct iattr attr; | |
1622 | int status; | |
1623 | ||
6de1472f AV |
1624 | dfprintk(VFS, "NFS: mknod(%s/%ld), %pd\n", |
1625 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1da177e4 LT |
1626 | |
1627 | if (!new_valid_dev(rdev)) | |
1628 | return -EINVAL; | |
1629 | ||
1630 | attr.ia_mode = mode; | |
1631 | attr.ia_valid = ATTR_MODE; | |
1632 | ||
1ca42382 | 1633 | trace_nfs_mknod_enter(dir, dentry); |
1da177e4 | 1634 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1ca42382 | 1635 | trace_nfs_mknod_exit(dir, dentry, status); |
1da177e4 LT |
1636 | if (status != 0) |
1637 | goto out_err; | |
1da177e4 LT |
1638 | return 0; |
1639 | out_err: | |
1da177e4 LT |
1640 | d_drop(dentry); |
1641 | return status; | |
1642 | } | |
ddda8e0a | 1643 | EXPORT_SYMBOL_GPL(nfs_mknod); |
1da177e4 LT |
1644 | |
1645 | /* | |
1646 | * See comments for nfs_proc_create regarding failed operations. | |
1647 | */ | |
597d9289 | 1648 | int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
1649 | { |
1650 | struct iattr attr; | |
1651 | int error; | |
1652 | ||
6de1472f AV |
1653 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %pd\n", |
1654 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1da177e4 LT |
1655 | |
1656 | attr.ia_valid = ATTR_MODE; | |
1657 | attr.ia_mode = mode | S_IFDIR; | |
1658 | ||
1ca42382 | 1659 | trace_nfs_mkdir_enter(dir, dentry); |
1da177e4 | 1660 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1ca42382 | 1661 | trace_nfs_mkdir_exit(dir, dentry, error); |
1da177e4 LT |
1662 | if (error != 0) |
1663 | goto out_err; | |
1da177e4 LT |
1664 | return 0; |
1665 | out_err: | |
1666 | d_drop(dentry); | |
1da177e4 LT |
1667 | return error; |
1668 | } | |
ddda8e0a | 1669 | EXPORT_SYMBOL_GPL(nfs_mkdir); |
1da177e4 | 1670 | |
d45b9d8b TM |
1671 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1672 | { | |
1673 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) | |
1674 | d_delete(dentry); | |
1675 | } | |
1676 | ||
597d9289 | 1677 | int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
1678 | { |
1679 | int error; | |
1680 | ||
6de1472f AV |
1681 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %pd\n", |
1682 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1da177e4 | 1683 | |
1ca42382 | 1684 | trace_nfs_rmdir_enter(dir, dentry); |
ba6c0592 TM |
1685 | if (dentry->d_inode) { |
1686 | nfs_wait_on_sillyrename(dentry); | |
1687 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1688 | /* Ensure the VFS deletes this inode */ | |
1689 | switch (error) { | |
1690 | case 0: | |
1691 | clear_nlink(dentry->d_inode); | |
1692 | break; | |
1693 | case -ENOENT: | |
1694 | nfs_dentry_handle_enoent(dentry); | |
1695 | } | |
1696 | } else | |
1697 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1ca42382 | 1698 | trace_nfs_rmdir_exit(dir, dentry, error); |
1da177e4 LT |
1699 | |
1700 | return error; | |
1701 | } | |
ddda8e0a | 1702 | EXPORT_SYMBOL_GPL(nfs_rmdir); |
1da177e4 | 1703 | |
1da177e4 LT |
1704 | /* |
1705 | * Remove a file after making sure there are no pending writes, | |
1706 | * and after checking that the file has only one user. | |
1707 | * | |
1708 | * We invalidate the attribute cache and free the inode prior to the operation | |
1709 | * to avoid possible races if the server reuses the inode. | |
1710 | */ | |
1711 | static int nfs_safe_remove(struct dentry *dentry) | |
1712 | { | |
1713 | struct inode *dir = dentry->d_parent->d_inode; | |
1714 | struct inode *inode = dentry->d_inode; | |
1715 | int error = -EBUSY; | |
1716 | ||
6de1472f | 1717 | dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); |
1da177e4 LT |
1718 | |
1719 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1720 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1721 | error = 0; | |
1722 | goto out; | |
1723 | } | |
1724 | ||
1ca42382 | 1725 | trace_nfs_remove_enter(dir, dentry); |
1da177e4 | 1726 | if (inode != NULL) { |
57ec14c5 | 1727 | NFS_PROTO(inode)->return_delegation(inode); |
1da177e4 | 1728 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
1da177e4 | 1729 | if (error == 0) |
1b83d707 | 1730 | nfs_drop_nlink(inode); |
1da177e4 LT |
1731 | } else |
1732 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
d45b9d8b TM |
1733 | if (error == -ENOENT) |
1734 | nfs_dentry_handle_enoent(dentry); | |
1ca42382 | 1735 | trace_nfs_remove_exit(dir, dentry, error); |
1da177e4 LT |
1736 | out: |
1737 | return error; | |
1738 | } | |
1739 | ||
1740 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1741 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1742 | * | |
1743 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1744 | */ | |
597d9289 | 1745 | int nfs_unlink(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
1746 | { |
1747 | int error; | |
1748 | int need_rehash = 0; | |
1749 | ||
6de1472f AV |
1750 | dfprintk(VFS, "NFS: unlink(%s/%ld, %pd)\n", dir->i_sb->s_id, |
1751 | dir->i_ino, dentry); | |
1da177e4 | 1752 | |
1ca42382 | 1753 | trace_nfs_unlink_enter(dir, dentry); |
1da177e4 | 1754 | spin_lock(&dentry->d_lock); |
84d08fa8 | 1755 | if (d_count(dentry) > 1) { |
1da177e4 | 1756 | spin_unlock(&dentry->d_lock); |
ccfeb506 TM |
1757 | /* Start asynchronous writeout of the inode */ |
1758 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 | 1759 | error = nfs_sillyrename(dir, dentry); |
1ca42382 | 1760 | goto out; |
1da177e4 LT |
1761 | } |
1762 | if (!d_unhashed(dentry)) { | |
1763 | __d_drop(dentry); | |
1764 | need_rehash = 1; | |
1765 | } | |
1766 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1767 | error = nfs_safe_remove(dentry); |
d45b9d8b | 1768 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1769 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1770 | } else if (need_rehash) | |
1771 | d_rehash(dentry); | |
1ca42382 TM |
1772 | out: |
1773 | trace_nfs_unlink_exit(dir, dentry, error); | |
1da177e4 LT |
1774 | return error; |
1775 | } | |
ddda8e0a | 1776 | EXPORT_SYMBOL_GPL(nfs_unlink); |
1da177e4 | 1777 | |
873101b3 CL |
1778 | /* |
1779 | * To create a symbolic link, most file systems instantiate a new inode, | |
1780 | * add a page to it containing the path, then write it out to the disk | |
1781 | * using prepare_write/commit_write. | |
1782 | * | |
1783 | * Unfortunately the NFS client can't create the in-core inode first | |
1784 | * because it needs a file handle to create an in-core inode (see | |
1785 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1786 | * symlink request has completed on the server. | |
1787 | * | |
1788 | * So instead we allocate a raw page, copy the symname into it, then do | |
1789 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1790 | * now have a new file handle and can instantiate an in-core NFS inode | |
1791 | * and move the raw page into its mapping. | |
1792 | */ | |
597d9289 | 1793 | int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) |
1da177e4 | 1794 | { |
873101b3 CL |
1795 | struct page *page; |
1796 | char *kaddr; | |
1da177e4 | 1797 | struct iattr attr; |
873101b3 | 1798 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1799 | int error; |
1800 | ||
6de1472f AV |
1801 | dfprintk(VFS, "NFS: symlink(%s/%ld, %pd, %s)\n", dir->i_sb->s_id, |
1802 | dir->i_ino, dentry, symname); | |
1da177e4 | 1803 | |
873101b3 CL |
1804 | if (pathlen > PAGE_SIZE) |
1805 | return -ENAMETOOLONG; | |
1da177e4 | 1806 | |
873101b3 CL |
1807 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1808 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1809 | |
83d93f22 | 1810 | page = alloc_page(GFP_HIGHUSER); |
76566991 | 1811 | if (!page) |
873101b3 | 1812 | return -ENOMEM; |
873101b3 | 1813 | |
2b86ce2d | 1814 | kaddr = kmap_atomic(page); |
873101b3 CL |
1815 | memcpy(kaddr, symname, pathlen); |
1816 | if (pathlen < PAGE_SIZE) | |
1817 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
2b86ce2d | 1818 | kunmap_atomic(kaddr); |
873101b3 | 1819 | |
1ca42382 | 1820 | trace_nfs_symlink_enter(dir, dentry); |
94a6d753 | 1821 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
1ca42382 | 1822 | trace_nfs_symlink_exit(dir, dentry, error); |
873101b3 | 1823 | if (error != 0) { |
6de1472f | 1824 | dfprintk(VFS, "NFS: symlink(%s/%ld, %pd, %s) error %d\n", |
873101b3 | 1825 | dir->i_sb->s_id, dir->i_ino, |
6de1472f | 1826 | dentry, symname, error); |
1da177e4 | 1827 | d_drop(dentry); |
873101b3 | 1828 | __free_page(page); |
873101b3 CL |
1829 | return error; |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * No big deal if we can't add this page to the page cache here. | |
1834 | * READLINK will get the missing page from the server if needed. | |
1835 | */ | |
a0b8cab3 | 1836 | if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0, |
873101b3 | 1837 | GFP_KERNEL)) { |
873101b3 CL |
1838 | SetPageUptodate(page); |
1839 | unlock_page(page); | |
1840 | } else | |
1841 | __free_page(page); | |
1842 | ||
873101b3 | 1843 | return 0; |
1da177e4 | 1844 | } |
ddda8e0a | 1845 | EXPORT_SYMBOL_GPL(nfs_symlink); |
1da177e4 | 1846 | |
597d9289 | 1847 | int |
1da177e4 LT |
1848 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) |
1849 | { | |
1850 | struct inode *inode = old_dentry->d_inode; | |
1851 | int error; | |
1852 | ||
6de1472f AV |
1853 | dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", |
1854 | old_dentry, dentry); | |
1da177e4 | 1855 | |
1fd1085b | 1856 | trace_nfs_link_enter(inode, dir, dentry); |
57ec14c5 | 1857 | NFS_PROTO(inode)->return_delegation(inode); |
9a3936aa | 1858 | |
9697d234 | 1859 | d_drop(dentry); |
1da177e4 | 1860 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1861 | if (error == 0) { |
7de9c6ee | 1862 | ihold(inode); |
9697d234 | 1863 | d_add(dentry, inode); |
cf809556 | 1864 | } |
1fd1085b | 1865 | trace_nfs_link_exit(inode, dir, dentry, error); |
1da177e4 LT |
1866 | return error; |
1867 | } | |
ddda8e0a | 1868 | EXPORT_SYMBOL_GPL(nfs_link); |
1da177e4 LT |
1869 | |
1870 | /* | |
1871 | * RENAME | |
1872 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1873 | * different file handle for the same inode after a rename (e.g. when | |
1874 | * moving to a different directory). A fail-safe method to do so would | |
1875 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1876 | * rename the old file using the sillyrename stuff. This way, the original | |
1877 | * file in old_dir will go away when the last process iput()s the inode. | |
1878 | * | |
1879 | * FIXED. | |
1880 | * | |
1881 | * It actually works quite well. One needs to have the possibility for | |
1882 | * at least one ".nfs..." file in each directory the file ever gets | |
1883 | * moved or linked to which happens automagically with the new | |
1884 | * implementation that only depends on the dcache stuff instead of | |
1885 | * using the inode layer | |
1886 | * | |
1887 | * Unfortunately, things are a little more complicated than indicated | |
1888 | * above. For a cross-directory move, we want to make sure we can get | |
1889 | * rid of the old inode after the operation. This means there must be | |
1890 | * no pending writes (if it's a file), and the use count must be 1. | |
1891 | * If these conditions are met, we can drop the dentries before doing | |
1892 | * the rename. | |
1893 | */ | |
597d9289 | 1894 | int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
1da177e4 LT |
1895 | struct inode *new_dir, struct dentry *new_dentry) |
1896 | { | |
1897 | struct inode *old_inode = old_dentry->d_inode; | |
1898 | struct inode *new_inode = new_dentry->d_inode; | |
1899 | struct dentry *dentry = NULL, *rehash = NULL; | |
1900 | int error = -EBUSY; | |
1901 | ||
6de1472f AV |
1902 | dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", |
1903 | old_dentry, new_dentry, | |
84d08fa8 | 1904 | d_count(new_dentry)); |
1da177e4 | 1905 | |
70ded201 | 1906 | trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); |
1da177e4 | 1907 | /* |
28f79a1a MS |
1908 | * For non-directories, check whether the target is busy and if so, |
1909 | * make a copy of the dentry and then do a silly-rename. If the | |
1910 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1911 | * the new target. | |
1da177e4 | 1912 | */ |
27226104 MS |
1913 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1914 | /* | |
1915 | * To prevent any new references to the target during the | |
1916 | * rename, we unhash the dentry in advance. | |
1917 | */ | |
1918 | if (!d_unhashed(new_dentry)) { | |
1919 | d_drop(new_dentry); | |
1920 | rehash = new_dentry; | |
1921 | } | |
1da177e4 | 1922 | |
84d08fa8 | 1923 | if (d_count(new_dentry) > 2) { |
27226104 MS |
1924 | int err; |
1925 | ||
1926 | /* copy the target dentry's name */ | |
1927 | dentry = d_alloc(new_dentry->d_parent, | |
1928 | &new_dentry->d_name); | |
1929 | if (!dentry) | |
1930 | goto out; | |
1931 | ||
1932 | /* silly-rename the existing target ... */ | |
1933 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 1934 | if (err) |
27226104 | 1935 | goto out; |
24e93025 MS |
1936 | |
1937 | new_dentry = dentry; | |
56335936 | 1938 | rehash = NULL; |
24e93025 | 1939 | new_inode = NULL; |
27226104 | 1940 | } |
b1e4adf4 | 1941 | } |
1da177e4 | 1942 | |
57ec14c5 | 1943 | NFS_PROTO(old_inode)->return_delegation(old_inode); |
b1e4adf4 | 1944 | if (new_inode != NULL) |
57ec14c5 | 1945 | NFS_PROTO(new_inode)->return_delegation(new_inode); |
1da177e4 | 1946 | |
1da177e4 LT |
1947 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
1948 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1949 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1950 | out: |
1951 | if (rehash) | |
1952 | d_rehash(rehash); | |
70ded201 TM |
1953 | trace_nfs_rename_exit(old_dir, old_dentry, |
1954 | new_dir, new_dentry, error); | |
1da177e4 | 1955 | if (!error) { |
b1e4adf4 TM |
1956 | if (new_inode != NULL) |
1957 | nfs_drop_nlink(new_inode); | |
349457cc | 1958 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
1959 | nfs_set_verifier(new_dentry, |
1960 | nfs_save_change_attribute(new_dir)); | |
d45b9d8b TM |
1961 | } else if (error == -ENOENT) |
1962 | nfs_dentry_handle_enoent(old_dentry); | |
1da177e4 LT |
1963 | |
1964 | /* new dentry created? */ | |
1965 | if (dentry) | |
1966 | dput(dentry); | |
1da177e4 LT |
1967 | return error; |
1968 | } | |
ddda8e0a | 1969 | EXPORT_SYMBOL_GPL(nfs_rename); |
1da177e4 | 1970 | |
cfcea3e8 TM |
1971 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1972 | static LIST_HEAD(nfs_access_lru_list); | |
1973 | static atomic_long_t nfs_access_nr_entries; | |
1974 | ||
1c3c07e9 TM |
1975 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1976 | { | |
1977 | put_rpccred(entry->cred); | |
1978 | kfree(entry); | |
cfcea3e8 TM |
1979 | smp_mb__before_atomic_dec(); |
1980 | atomic_long_dec(&nfs_access_nr_entries); | |
1981 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1982 | } |
1983 | ||
1a81bb8a TM |
1984 | static void nfs_access_free_list(struct list_head *head) |
1985 | { | |
1986 | struct nfs_access_entry *cache; | |
1987 | ||
1988 | while (!list_empty(head)) { | |
1989 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
1990 | list_del(&cache->lru); | |
1991 | nfs_access_free_entry(cache); | |
1992 | } | |
1993 | } | |
1994 | ||
1ab6c499 DC |
1995 | unsigned long |
1996 | nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
979df72e TM |
1997 | { |
1998 | LIST_HEAD(head); | |
aa510da5 | 1999 | struct nfs_inode *nfsi, *next; |
979df72e | 2000 | struct nfs_access_entry *cache; |
1495f230 YH |
2001 | int nr_to_scan = sc->nr_to_scan; |
2002 | gfp_t gfp_mask = sc->gfp_mask; | |
1ab6c499 | 2003 | long freed = 0; |
979df72e | 2004 | |
61d5eb29 | 2005 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
1ab6c499 | 2006 | return SHRINK_STOP; |
9c7e7e23 | 2007 | |
a50f7951 | 2008 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 2009 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
2010 | struct inode *inode; |
2011 | ||
2012 | if (nr_to_scan-- == 0) | |
2013 | break; | |
9c7e7e23 | 2014 | inode = &nfsi->vfs_inode; |
979df72e TM |
2015 | spin_lock(&inode->i_lock); |
2016 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
2017 | goto remove_lru_entry; | |
2018 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
2019 | struct nfs_access_entry, lru); | |
2020 | list_move(&cache->lru, &head); | |
2021 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1ab6c499 | 2022 | freed++; |
979df72e TM |
2023 | if (!list_empty(&nfsi->access_cache_entry_lru)) |
2024 | list_move_tail(&nfsi->access_cache_inode_lru, | |
2025 | &nfs_access_lru_list); | |
2026 | else { | |
2027 | remove_lru_entry: | |
2028 | list_del_init(&nfsi->access_cache_inode_lru); | |
9c7e7e23 | 2029 | smp_mb__before_clear_bit(); |
979df72e | 2030 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
9c7e7e23 | 2031 | smp_mb__after_clear_bit(); |
979df72e | 2032 | } |
59844a9b | 2033 | spin_unlock(&inode->i_lock); |
979df72e TM |
2034 | } |
2035 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2036 | nfs_access_free_list(&head); |
1ab6c499 DC |
2037 | return freed; |
2038 | } | |
2039 | ||
2040 | unsigned long | |
2041 | nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
2042 | { | |
55f841ce | 2043 | return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); |
979df72e TM |
2044 | } |
2045 | ||
1a81bb8a | 2046 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2047 | { |
1c3c07e9 | 2048 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2049 | struct rb_node *n; |
1c3c07e9 TM |
2050 | struct nfs_access_entry *entry; |
2051 | ||
2052 | /* Unhook entries from the cache */ | |
2053 | while ((n = rb_first(root_node)) != NULL) { | |
2054 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2055 | rb_erase(n, root_node); | |
1a81bb8a | 2056 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2057 | } |
2058 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2059 | } |
2060 | ||
1c3c07e9 | 2061 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2062 | { |
1a81bb8a TM |
2063 | LIST_HEAD(head); |
2064 | ||
2065 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2066 | return; | |
cfcea3e8 | 2067 | /* Remove from global LRU init */ |
1a81bb8a TM |
2068 | spin_lock(&nfs_access_lru_lock); |
2069 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2070 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2071 | |
1c3c07e9 | 2072 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2073 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2074 | spin_unlock(&inode->i_lock); | |
2075 | spin_unlock(&nfs_access_lru_lock); | |
2076 | nfs_access_free_list(&head); | |
1c3c07e9 | 2077 | } |
1c606fb7 | 2078 | EXPORT_SYMBOL_GPL(nfs_access_zap_cache); |
1da177e4 | 2079 | |
1c3c07e9 TM |
2080 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2081 | { | |
2082 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2083 | struct nfs_access_entry *entry; | |
2084 | ||
2085 | while (n != NULL) { | |
2086 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2087 | ||
2088 | if (cred < entry->cred) | |
2089 | n = n->rb_left; | |
2090 | else if (cred > entry->cred) | |
2091 | n = n->rb_right; | |
2092 | else | |
2093 | return entry; | |
1da177e4 | 2094 | } |
1c3c07e9 TM |
2095 | return NULL; |
2096 | } | |
2097 | ||
af22f94a | 2098 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
1c3c07e9 TM |
2099 | { |
2100 | struct nfs_inode *nfsi = NFS_I(inode); | |
2101 | struct nfs_access_entry *cache; | |
2102 | int err = -ENOENT; | |
2103 | ||
dc59250c | 2104 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
2105 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
2106 | goto out_zap; | |
2107 | cache = nfs_access_search_rbtree(inode, cred); | |
2108 | if (cache == NULL) | |
2109 | goto out; | |
b4d2314b | 2110 | if (!nfs_have_delegated_attributes(inode) && |
64672d55 | 2111 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
1c3c07e9 TM |
2112 | goto out_stale; |
2113 | res->jiffies = cache->jiffies; | |
2114 | res->cred = cache->cred; | |
2115 | res->mask = cache->mask; | |
cfcea3e8 | 2116 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2117 | err = 0; |
2118 | out: | |
2119 | spin_unlock(&inode->i_lock); | |
2120 | return err; | |
2121 | out_stale: | |
2122 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 2123 | list_del(&cache->lru); |
1c3c07e9 TM |
2124 | spin_unlock(&inode->i_lock); |
2125 | nfs_access_free_entry(cache); | |
2126 | return -ENOENT; | |
2127 | out_zap: | |
1a81bb8a TM |
2128 | spin_unlock(&inode->i_lock); |
2129 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2130 | return -ENOENT; |
2131 | } | |
2132 | ||
2133 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2134 | { | |
cfcea3e8 TM |
2135 | struct nfs_inode *nfsi = NFS_I(inode); |
2136 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2137 | struct rb_node **p = &root_node->rb_node; |
2138 | struct rb_node *parent = NULL; | |
2139 | struct nfs_access_entry *entry; | |
2140 | ||
2141 | spin_lock(&inode->i_lock); | |
2142 | while (*p != NULL) { | |
2143 | parent = *p; | |
2144 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2145 | ||
2146 | if (set->cred < entry->cred) | |
2147 | p = &parent->rb_left; | |
2148 | else if (set->cred > entry->cred) | |
2149 | p = &parent->rb_right; | |
2150 | else | |
2151 | goto found; | |
2152 | } | |
2153 | rb_link_node(&set->rb_node, parent, p); | |
2154 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2155 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2156 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2157 | return; |
2158 | found: | |
2159 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2160 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2161 | list_del(&entry->lru); | |
1c3c07e9 TM |
2162 | spin_unlock(&inode->i_lock); |
2163 | nfs_access_free_entry(entry); | |
2164 | } | |
2165 | ||
6168f62c | 2166 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2167 | { |
2168 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2169 | if (cache == NULL) | |
2170 | return; | |
2171 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 2172 | cache->jiffies = set->jiffies; |
1c3c07e9 | 2173 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2174 | cache->mask = set->mask; |
1c3c07e9 TM |
2175 | |
2176 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
2177 | |
2178 | /* Update accounting */ | |
2179 | smp_mb__before_atomic_inc(); | |
2180 | atomic_long_inc(&nfs_access_nr_entries); | |
2181 | smp_mb__after_atomic_inc(); | |
2182 | ||
2183 | /* Add inode to global LRU list */ | |
1a81bb8a | 2184 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2185 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2186 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2187 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2188 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2189 | spin_unlock(&nfs_access_lru_lock); |
2190 | } | |
1da177e4 | 2191 | } |
6168f62c WAA |
2192 | EXPORT_SYMBOL_GPL(nfs_access_add_cache); |
2193 | ||
2194 | void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) | |
2195 | { | |
2196 | entry->mask = 0; | |
2197 | if (access_result & NFS4_ACCESS_READ) | |
2198 | entry->mask |= MAY_READ; | |
2199 | if (access_result & | |
2200 | (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) | |
2201 | entry->mask |= MAY_WRITE; | |
2202 | if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) | |
2203 | entry->mask |= MAY_EXEC; | |
2204 | } | |
2205 | EXPORT_SYMBOL_GPL(nfs_access_set_mask); | |
1da177e4 LT |
2206 | |
2207 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2208 | { | |
2209 | struct nfs_access_entry cache; | |
2210 | int status; | |
2211 | ||
f4ce1299 TM |
2212 | trace_nfs_access_enter(inode); |
2213 | ||
1da177e4 LT |
2214 | status = nfs_access_get_cached(inode, cred, &cache); |
2215 | if (status == 0) | |
f4ce1299 | 2216 | goto out_cached; |
1da177e4 LT |
2217 | |
2218 | /* Be clever: ask server to check for all possible rights */ | |
2219 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2220 | cache.cred = cred; | |
2221 | cache.jiffies = jiffies; | |
2222 | status = NFS_PROTO(inode)->access(inode, &cache); | |
a71ee337 SJ |
2223 | if (status != 0) { |
2224 | if (status == -ESTALE) { | |
2225 | nfs_zap_caches(inode); | |
2226 | if (!S_ISDIR(inode->i_mode)) | |
2227 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2228 | } | |
f4ce1299 | 2229 | goto out; |
a71ee337 | 2230 | } |
1da177e4 | 2231 | nfs_access_add_cache(inode, &cache); |
f4ce1299 TM |
2232 | out_cached: |
2233 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) | |
2234 | status = -EACCES; | |
1da177e4 | 2235 | out: |
f4ce1299 TM |
2236 | trace_nfs_access_exit(inode, status); |
2237 | return status; | |
1da177e4 LT |
2238 | } |
2239 | ||
af22f94a TM |
2240 | static int nfs_open_permission_mask(int openflags) |
2241 | { | |
2242 | int mask = 0; | |
2243 | ||
f8d9a897 WAA |
2244 | if (openflags & __FMODE_EXEC) { |
2245 | /* ONLY check exec rights */ | |
2246 | mask = MAY_EXEC; | |
2247 | } else { | |
2248 | if ((openflags & O_ACCMODE) != O_WRONLY) | |
2249 | mask |= MAY_READ; | |
2250 | if ((openflags & O_ACCMODE) != O_RDONLY) | |
2251 | mask |= MAY_WRITE; | |
2252 | } | |
2253 | ||
af22f94a TM |
2254 | return mask; |
2255 | } | |
2256 | ||
2257 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2258 | { | |
2259 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2260 | } | |
89d77c8f | 2261 | EXPORT_SYMBOL_GPL(nfs_may_open); |
af22f94a | 2262 | |
10556cb2 | 2263 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2264 | { |
2265 | struct rpc_cred *cred; | |
2266 | int res = 0; | |
2267 | ||
10556cb2 | 2268 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2269 | return -ECHILD; |
2270 | ||
91d5b470 CL |
2271 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2272 | ||
e6305c43 | 2273 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2274 | goto out; |
2275 | /* Is this sys_access() ? */ | |
9cfcac81 | 2276 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2277 | goto force_lookup; |
2278 | ||
2279 | switch (inode->i_mode & S_IFMT) { | |
2280 | case S_IFLNK: | |
2281 | goto out; | |
2282 | case S_IFREG: | |
1da177e4 LT |
2283 | break; |
2284 | case S_IFDIR: | |
2285 | /* | |
2286 | * Optimize away all write operations, since the server | |
2287 | * will check permissions when we perform the op. | |
2288 | */ | |
2289 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2290 | goto out; | |
2291 | } | |
2292 | ||
2293 | force_lookup: | |
1da177e4 LT |
2294 | if (!NFS_PROTO(inode)->access) |
2295 | goto out_notsup; | |
2296 | ||
98a8e323 | 2297 | cred = rpc_lookup_cred(); |
1da177e4 LT |
2298 | if (!IS_ERR(cred)) { |
2299 | res = nfs_do_access(inode, cred, mask); | |
2300 | put_rpccred(cred); | |
2301 | } else | |
2302 | res = PTR_ERR(cred); | |
1da177e4 | 2303 | out: |
f696a365 MS |
2304 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
2305 | res = -EACCES; | |
2306 | ||
1e7cb3dc CL |
2307 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
2308 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
2309 | return res; |
2310 | out_notsup: | |
2311 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2312 | if (res == 0) | |
2830ba7f | 2313 | res = generic_permission(inode, mask); |
1e7cb3dc | 2314 | goto out; |
1da177e4 | 2315 | } |
ddda8e0a | 2316 | EXPORT_SYMBOL_GPL(nfs_permission); |
1da177e4 LT |
2317 | |
2318 | /* | |
2319 | * Local variables: | |
2320 | * version-control: t | |
2321 | * kept-new-versions: 5 | |
2322 | * End: | |
2323 | */ |