]>
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, |
b044f645 | 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 | 71 | { |
311324ad | 72 | struct nfs_inode *nfsi = NFS_I(dir); |
480c2006 BS |
73 | struct nfs_open_dir_context *ctx; |
74 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
75 | if (ctx != NULL) { | |
8ef2ce3e | 76 | ctx->duped = 0; |
311324ad | 77 | ctx->attr_gencount = nfsi->attr_gencount; |
480c2006 | 78 | ctx->dir_cookie = 0; |
8ef2ce3e | 79 | ctx->dup_cookie = 0; |
480c2006 | 80 | ctx->cred = get_rpccred(cred); |
311324ad TM |
81 | spin_lock(&dir->i_lock); |
82 | list_add(&ctx->list, &nfsi->open_files); | |
83 | spin_unlock(&dir->i_lock); | |
0c030806 TM |
84 | return ctx; |
85 | } | |
86 | return ERR_PTR(-ENOMEM); | |
480c2006 BS |
87 | } |
88 | ||
311324ad | 89 | static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) |
480c2006 | 90 | { |
311324ad TM |
91 | spin_lock(&dir->i_lock); |
92 | list_del(&ctx->list); | |
93 | spin_unlock(&dir->i_lock); | |
480c2006 BS |
94 | put_rpccred(ctx->cred); |
95 | kfree(ctx); | |
96 | } | |
97 | ||
1da177e4 LT |
98 | /* |
99 | * Open file | |
100 | */ | |
101 | static int | |
102 | nfs_opendir(struct inode *inode, struct file *filp) | |
103 | { | |
480c2006 BS |
104 | int res = 0; |
105 | struct nfs_open_dir_context *ctx; | |
106 | struct rpc_cred *cred; | |
1da177e4 | 107 | |
6de1472f | 108 | dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); |
cc0dd2d1 CL |
109 | |
110 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 111 | |
480c2006 BS |
112 | cred = rpc_lookup_cred(); |
113 | if (IS_ERR(cred)) | |
114 | return PTR_ERR(cred); | |
0c030806 | 115 | ctx = alloc_nfs_open_dir_context(inode, cred); |
480c2006 BS |
116 | if (IS_ERR(ctx)) { |
117 | res = PTR_ERR(ctx); | |
118 | goto out; | |
119 | } | |
120 | filp->private_data = ctx; | |
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 | { | |
a455589f | 129 | put_nfs_open_dir_context(file_inode(filp), filp->private_data); |
480c2006 BS |
130 | return 0; |
131 | } | |
132 | ||
d1bacf9e BS |
133 | struct nfs_cache_array_entry { |
134 | u64 cookie; | |
135 | u64 ino; | |
136 | struct qstr string; | |
0b26a0bf | 137 | unsigned char d_type; |
d1bacf9e BS |
138 | }; |
139 | ||
140 | struct nfs_cache_array { | |
88b8e133 | 141 | int size; |
d1bacf9e BS |
142 | int eof_index; |
143 | u64 last_cookie; | |
144 | struct nfs_cache_array_entry array[0]; | |
145 | }; | |
146 | ||
a7a3b1e9 | 147 | typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool); |
1da177e4 LT |
148 | typedef struct { |
149 | struct file *file; | |
150 | struct page *page; | |
23db8620 | 151 | struct dir_context *ctx; |
1da177e4 | 152 | unsigned long page_index; |
f0dd2136 | 153 | u64 *dir_cookie; |
0aded708 | 154 | u64 last_cookie; |
f0dd2136 | 155 | loff_t current_index; |
1da177e4 | 156 | decode_dirent_t decode; |
d1bacf9e | 157 | |
1f4eab7e | 158 | unsigned long timestamp; |
4704f0e2 | 159 | unsigned long gencount; |
d1bacf9e | 160 | unsigned int cache_entry_index; |
a7a3b1e9 BC |
161 | bool plus; |
162 | bool eof; | |
1da177e4 LT |
163 | } nfs_readdir_descriptor_t; |
164 | ||
d1bacf9e BS |
165 | /* |
166 | * we are freeing strings created by nfs_add_to_readdir_array() | |
167 | */ | |
168 | static | |
11de3b11 | 169 | void nfs_readdir_clear_array(struct page *page) |
d1bacf9e | 170 | { |
11de3b11 | 171 | struct nfs_cache_array *array; |
d1bacf9e | 172 | int i; |
8cd51a0c | 173 | |
2b86ce2d | 174 | array = kmap_atomic(page); |
b044f645 BC |
175 | for (i = 0; i < array->size; i++) |
176 | kfree(array->array[i].string.name); | |
2b86ce2d | 177 | kunmap_atomic(array); |
d1bacf9e BS |
178 | } |
179 | ||
180 | /* | |
181 | * the caller is responsible for freeing qstr.name | |
182 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
183 | * nfs_clear_readdir_array() | |
184 | */ | |
185 | static | |
4a201d6e | 186 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
187 | { |
188 | string->len = len; | |
189 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
190 | if (string->name == NULL) |
191 | return -ENOMEM; | |
04e4bd1c CM |
192 | /* |
193 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
194 | * in a page cache page which kmemleak does not scan. | |
195 | */ | |
196 | kmemleak_not_leak(string->name); | |
8387ff25 | 197 | string->hash = full_name_hash(NULL, name, len); |
4a201d6e | 198 | return 0; |
d1bacf9e BS |
199 | } |
200 | ||
201 | static | |
202 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
203 | { | |
0795bf83 | 204 | struct nfs_cache_array *array = kmap(page); |
4a201d6e TM |
205 | struct nfs_cache_array_entry *cache_entry; |
206 | int ret; | |
207 | ||
3020093f TM |
208 | cache_entry = &array->array[array->size]; |
209 | ||
210 | /* Check that this entry lies within the page bounds */ | |
8cd51a0c | 211 | ret = -ENOSPC; |
3020093f | 212 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) |
4a201d6e | 213 | goto out; |
d1bacf9e | 214 | |
4a201d6e TM |
215 | cache_entry->cookie = entry->prev_cookie; |
216 | cache_entry->ino = entry->ino; | |
0b26a0bf | 217 | cache_entry->d_type = entry->d_type; |
4a201d6e TM |
218 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); |
219 | if (ret) | |
220 | goto out; | |
d1bacf9e | 221 | array->last_cookie = entry->cookie; |
8cd51a0c | 222 | array->size++; |
47c716cb | 223 | if (entry->eof != 0) |
d1bacf9e | 224 | array->eof_index = array->size; |
4a201d6e | 225 | out: |
0795bf83 | 226 | kunmap(page); |
4a201d6e | 227 | return ret; |
d1bacf9e BS |
228 | } |
229 | ||
230 | static | |
231 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
232 | { | |
23db8620 | 233 | loff_t diff = desc->ctx->pos - desc->current_index; |
d1bacf9e BS |
234 | unsigned int index; |
235 | ||
236 | if (diff < 0) | |
237 | goto out_eof; | |
238 | if (diff >= array->size) { | |
8cd51a0c | 239 | if (array->eof_index >= 0) |
d1bacf9e | 240 | goto out_eof; |
d1bacf9e BS |
241 | return -EAGAIN; |
242 | } | |
243 | ||
244 | index = (unsigned int)diff; | |
245 | *desc->dir_cookie = array->array[index].cookie; | |
246 | desc->cache_entry_index = index; | |
d1bacf9e BS |
247 | return 0; |
248 | out_eof: | |
6089dd0d | 249 | desc->eof = true; |
d1bacf9e BS |
250 | return -EBADCOOKIE; |
251 | } | |
252 | ||
4db72b40 JL |
253 | static bool |
254 | nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) | |
255 | { | |
256 | if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) | |
257 | return false; | |
258 | smp_rmb(); | |
259 | return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); | |
260 | } | |
261 | ||
d1bacf9e BS |
262 | static |
263 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
264 | { | |
265 | int i; | |
8ef2ce3e | 266 | loff_t new_pos; |
d1bacf9e BS |
267 | int status = -EAGAIN; |
268 | ||
269 | for (i = 0; i < array->size; i++) { | |
d1bacf9e | 270 | if (array->array[i].cookie == *desc->dir_cookie) { |
496ad9aa | 271 | struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); |
0c030806 TM |
272 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
273 | ||
8ef2ce3e | 274 | new_pos = desc->current_index + i; |
4db72b40 JL |
275 | if (ctx->attr_gencount != nfsi->attr_gencount || |
276 | !nfs_readdir_inode_mapping_valid(nfsi)) { | |
0c030806 TM |
277 | ctx->duped = 0; |
278 | ctx->attr_gencount = nfsi->attr_gencount; | |
23db8620 | 279 | } else if (new_pos < desc->ctx->pos) { |
0c030806 TM |
280 | if (ctx->duped > 0 |
281 | && ctx->dup_cookie == *desc->dir_cookie) { | |
282 | if (printk_ratelimit()) { | |
6de1472f | 283 | pr_notice("NFS: directory %pD2 contains a readdir loop." |
0c030806 | 284 | "Please contact your server vendor. " |
9581a4ae JL |
285 | "The file: %.*s has duplicate cookie %llu\n", |
286 | desc->file, array->array[i].string.len, | |
287 | array->array[i].string.name, *desc->dir_cookie); | |
0c030806 TM |
288 | } |
289 | status = -ELOOP; | |
290 | goto out; | |
291 | } | |
8ef2ce3e | 292 | ctx->dup_cookie = *desc->dir_cookie; |
0c030806 | 293 | ctx->duped = -1; |
8ef2ce3e | 294 | } |
23db8620 | 295 | desc->ctx->pos = new_pos; |
d1bacf9e | 296 | desc->cache_entry_index = i; |
47c716cb | 297 | return 0; |
d1bacf9e BS |
298 | } |
299 | } | |
47c716cb | 300 | if (array->eof_index >= 0) { |
8cd51a0c | 301 | status = -EBADCOOKIE; |
18fb5fe4 | 302 | if (*desc->dir_cookie == array->last_cookie) |
6089dd0d | 303 | desc->eof = true; |
8cd51a0c | 304 | } |
0c030806 | 305 | out: |
d1bacf9e BS |
306 | return status; |
307 | } | |
308 | ||
309 | static | |
310 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
311 | { | |
312 | struct nfs_cache_array *array; | |
47c716cb | 313 | int status; |
d1bacf9e | 314 | |
0795bf83 | 315 | array = kmap(desc->page); |
d1bacf9e BS |
316 | |
317 | if (*desc->dir_cookie == 0) | |
318 | status = nfs_readdir_search_for_pos(array, desc); | |
319 | else | |
320 | status = nfs_readdir_search_for_cookie(array, desc); | |
321 | ||
47c716cb | 322 | if (status == -EAGAIN) { |
0aded708 | 323 | desc->last_cookie = array->last_cookie; |
e47c085a | 324 | desc->current_index += array->size; |
47c716cb TM |
325 | desc->page_index++; |
326 | } | |
0795bf83 | 327 | kunmap(desc->page); |
d1bacf9e BS |
328 | return status; |
329 | } | |
330 | ||
331 | /* Fill a page with xdr information before transferring to the cache page */ | |
332 | static | |
56e4ebf8 | 333 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 334 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 335 | { |
480c2006 BS |
336 | struct nfs_open_dir_context *ctx = file->private_data; |
337 | struct rpc_cred *cred = ctx->cred; | |
4704f0e2 | 338 | unsigned long timestamp, gencount; |
1da177e4 LT |
339 | int error; |
340 | ||
1da177e4 LT |
341 | again: |
342 | timestamp = jiffies; | |
4704f0e2 | 343 | gencount = nfs_inc_attr_generation_counter(); |
be62a1a8 | 344 | error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, |
1da177e4 LT |
345 | NFS_SERVER(inode)->dtsize, desc->plus); |
346 | if (error < 0) { | |
347 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
348 | if (error == -ENOTSUPP && desc->plus) { | |
349 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 350 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
a7a3b1e9 | 351 | desc->plus = false; |
1da177e4 LT |
352 | goto again; |
353 | } | |
354 | goto error; | |
355 | } | |
1f4eab7e | 356 | desc->timestamp = timestamp; |
4704f0e2 | 357 | desc->gencount = gencount; |
d1bacf9e BS |
358 | error: |
359 | return error; | |
1da177e4 LT |
360 | } |
361 | ||
573c4e1e CL |
362 | static int xdr_decode(nfs_readdir_descriptor_t *desc, |
363 | struct nfs_entry *entry, struct xdr_stream *xdr) | |
1da177e4 | 364 | { |
573c4e1e | 365 | int error; |
1da177e4 | 366 | |
573c4e1e CL |
367 | error = desc->decode(xdr, entry, desc->plus); |
368 | if (error) | |
369 | return error; | |
d1bacf9e BS |
370 | entry->fattr->time_start = desc->timestamp; |
371 | entry->fattr->gencount = desc->gencount; | |
372 | return 0; | |
1da177e4 LT |
373 | } |
374 | ||
fa923369 TM |
375 | /* Match file and dirent using either filehandle or fileid |
376 | * Note: caller is responsible for checking the fsid | |
377 | */ | |
d39ab9de BS |
378 | static |
379 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
380 | { | |
d8fdb47f | 381 | struct inode *inode; |
fa923369 TM |
382 | struct nfs_inode *nfsi; |
383 | ||
2b0143b5 DH |
384 | if (d_really_is_negative(dentry)) |
385 | return 0; | |
fa923369 | 386 | |
d8fdb47f TM |
387 | inode = d_inode(dentry); |
388 | if (is_bad_inode(inode) || NFS_STALE(inode)) | |
389 | return 0; | |
390 | ||
391 | nfsi = NFS_I(inode); | |
7dc72d5f TM |
392 | if (entry->fattr->fileid != nfsi->fileid) |
393 | return 0; | |
394 | if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) | |
395 | return 0; | |
396 | return 1; | |
d39ab9de BS |
397 | } |
398 | ||
d69ee9b8 | 399 | static |
23db8620 | 400 | bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) |
d69ee9b8 TM |
401 | { |
402 | if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) | |
403 | return false; | |
404 | if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) | |
405 | return true; | |
23db8620 | 406 | if (ctx->pos == 0) |
d69ee9b8 TM |
407 | return true; |
408 | return false; | |
409 | } | |
410 | ||
411 | /* | |
63519fbc TM |
412 | * This function is called by the lookup and getattr code to request the |
413 | * use of readdirplus to accelerate any future lookups in the same | |
d69ee9b8 TM |
414 | * directory. |
415 | */ | |
d69ee9b8 TM |
416 | void nfs_advise_use_readdirplus(struct inode *dir) |
417 | { | |
63519fbc TM |
418 | struct nfs_inode *nfsi = NFS_I(dir); |
419 | ||
420 | if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && | |
421 | !list_empty(&nfsi->open_files)) | |
422 | set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); | |
d69ee9b8 TM |
423 | } |
424 | ||
311324ad TM |
425 | /* |
426 | * This function is mainly for use by nfs_getattr(). | |
427 | * | |
428 | * If this is an 'ls -l', we want to force use of readdirplus. | |
429 | * Do this by checking if there is an active file descriptor | |
430 | * and calling nfs_advise_use_readdirplus, then forcing a | |
431 | * cache flush. | |
432 | */ | |
433 | void nfs_force_use_readdirplus(struct inode *dir) | |
434 | { | |
63519fbc TM |
435 | struct nfs_inode *nfsi = NFS_I(dir); |
436 | ||
437 | if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && | |
438 | !list_empty(&nfsi->open_files)) { | |
439 | set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); | |
79f687a3 | 440 | invalidate_mapping_pages(dir->i_mapping, 0, -1); |
311324ad TM |
441 | } |
442 | } | |
443 | ||
d39ab9de BS |
444 | static |
445 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
446 | { | |
26fe5750 | 447 | struct qstr filename = QSTR_INIT(entry->name, entry->len); |
9ac3d3e8 | 448 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
4a201d6e TM |
449 | struct dentry *dentry; |
450 | struct dentry *alias; | |
2b0143b5 | 451 | struct inode *dir = d_inode(parent); |
d39ab9de | 452 | struct inode *inode; |
aa9c2669 | 453 | int status; |
d39ab9de | 454 | |
fa923369 TM |
455 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) |
456 | return; | |
6c441c25 TM |
457 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) |
458 | return; | |
78d04af4 TM |
459 | if (filename.len == 0) |
460 | return; | |
461 | /* Validate that the name doesn't contain any illegal '\0' */ | |
462 | if (strnlen(filename.name, filename.len) != filename.len) | |
463 | return; | |
464 | /* ...or '/' */ | |
465 | if (strnchr(filename.name, filename.len, '/')) | |
466 | return; | |
4a201d6e TM |
467 | if (filename.name[0] == '.') { |
468 | if (filename.len == 1) | |
469 | return; | |
470 | if (filename.len == 2 && filename.name[1] == '.') | |
471 | return; | |
472 | } | |
8387ff25 | 473 | filename.hash = full_name_hash(parent, filename.name, filename.len); |
d39ab9de | 474 | |
4a201d6e | 475 | dentry = d_lookup(parent, &filename); |
9ac3d3e8 AV |
476 | again: |
477 | if (!dentry) { | |
478 | dentry = d_alloc_parallel(parent, &filename, &wq); | |
479 | if (IS_ERR(dentry)) | |
480 | return; | |
481 | } | |
482 | if (!d_in_lookup(dentry)) { | |
6c441c25 TM |
483 | /* Is there a mountpoint here? If so, just exit */ |
484 | if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, | |
485 | &entry->fattr->fsid)) | |
486 | goto out; | |
d39ab9de | 487 | if (nfs_same_file(dentry, entry)) { |
7dc72d5f TM |
488 | if (!entry->fh->size) |
489 | goto out; | |
cda57a1e | 490 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
2b0143b5 | 491 | status = nfs_refresh_inode(d_inode(dentry), entry->fattr); |
aa9c2669 | 492 | if (!status) |
2b0143b5 | 493 | nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); |
d39ab9de BS |
494 | goto out; |
495 | } else { | |
5542aa2f | 496 | d_invalidate(dentry); |
d39ab9de | 497 | dput(dentry); |
9ac3d3e8 AV |
498 | dentry = NULL; |
499 | goto again; | |
d39ab9de BS |
500 | } |
501 | } | |
7dc72d5f TM |
502 | if (!entry->fh->size) { |
503 | d_lookup_done(dentry); | |
504 | goto out; | |
505 | } | |
d39ab9de | 506 | |
1775fd3e | 507 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); |
41d28bca | 508 | alias = d_splice_alias(inode, dentry); |
9ac3d3e8 AV |
509 | d_lookup_done(dentry); |
510 | if (alias) { | |
511 | if (IS_ERR(alias)) | |
512 | goto out; | |
513 | dput(dentry); | |
514 | dentry = alias; | |
515 | } | |
516 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
d39ab9de BS |
517 | out: |
518 | dput(dentry); | |
d39ab9de BS |
519 | } |
520 | ||
d1bacf9e BS |
521 | /* Perform conversion from xdr to cache array */ |
522 | static | |
8cd51a0c | 523 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
6650239a | 524 | struct page **xdr_pages, struct page *page, unsigned int buflen) |
1da177e4 | 525 | { |
babddc72 | 526 | struct xdr_stream stream; |
f7da7a12 | 527 | struct xdr_buf buf; |
6650239a | 528 | struct page *scratch; |
99424380 | 529 | struct nfs_cache_array *array; |
5c346854 TM |
530 | unsigned int count = 0; |
531 | int status; | |
babddc72 | 532 | |
6650239a TM |
533 | scratch = alloc_page(GFP_KERNEL); |
534 | if (scratch == NULL) | |
535 | return -ENOMEM; | |
babddc72 | 536 | |
ce85cfbe BC |
537 | if (buflen == 0) |
538 | goto out_nopages; | |
539 | ||
f7da7a12 | 540 | xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); |
6650239a | 541 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
99424380 BS |
542 | |
543 | do { | |
544 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
545 | if (status != 0) { |
546 | if (status == -EAGAIN) | |
547 | status = 0; | |
99424380 | 548 | break; |
8cd51a0c | 549 | } |
99424380 | 550 | |
5c346854 TM |
551 | count++; |
552 | ||
a7a3b1e9 | 553 | if (desc->plus) |
be62a1a8 | 554 | nfs_prime_dcache(file_dentry(desc->file), entry); |
8cd51a0c TM |
555 | |
556 | status = nfs_readdir_add_to_array(entry, page); | |
557 | if (status != 0) | |
558 | break; | |
99424380 BS |
559 | } while (!entry->eof); |
560 | ||
ce85cfbe | 561 | out_nopages: |
47c716cb | 562 | if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { |
0795bf83 FF |
563 | array = kmap(page); |
564 | array->eof_index = array->size; | |
565 | status = 0; | |
566 | kunmap(page); | |
1da177e4 | 567 | } |
6650239a TM |
568 | |
569 | put_page(scratch); | |
8cd51a0c | 570 | return status; |
56e4ebf8 BS |
571 | } |
572 | ||
573 | static | |
c7e9668e | 574 | void nfs_readdir_free_pages(struct page **pages, unsigned int npages) |
56e4ebf8 BS |
575 | { |
576 | unsigned int i; | |
577 | for (i = 0; i < npages; i++) | |
578 | put_page(pages[i]); | |
579 | } | |
580 | ||
56e4ebf8 BS |
581 | /* |
582 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
0b936e37 | 583 | * to nfs_readdir_free_pagearray |
56e4ebf8 BS |
584 | */ |
585 | static | |
c7e9668e | 586 | int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) |
56e4ebf8 | 587 | { |
56e4ebf8 BS |
588 | unsigned int i; |
589 | ||
590 | for (i = 0; i < npages; i++) { | |
591 | struct page *page = alloc_page(GFP_KERNEL); | |
592 | if (page == NULL) | |
593 | goto out_freepages; | |
594 | pages[i] = page; | |
595 | } | |
6650239a | 596 | return 0; |
56e4ebf8 | 597 | |
56e4ebf8 | 598 | out_freepages: |
c7e9668e | 599 | nfs_readdir_free_pages(pages, i); |
6650239a | 600 | return -ENOMEM; |
1da177e4 LT |
601 | } |
602 | ||
d1bacf9e BS |
603 | static |
604 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 605 | { |
56e4ebf8 | 606 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
d1bacf9e BS |
607 | struct nfs_entry entry; |
608 | struct file *file = desc->file; | |
609 | struct nfs_cache_array *array; | |
8cd51a0c | 610 | int status = -ENOMEM; |
56e4ebf8 | 611 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
612 | |
613 | entry.prev_cookie = 0; | |
0aded708 | 614 | entry.cookie = desc->last_cookie; |
d1bacf9e BS |
615 | entry.eof = 0; |
616 | entry.fh = nfs_alloc_fhandle(); | |
617 | entry.fattr = nfs_alloc_fattr(); | |
573c4e1e | 618 | entry.server = NFS_SERVER(inode); |
d1bacf9e BS |
619 | if (entry.fh == NULL || entry.fattr == NULL) |
620 | goto out; | |
00a92642 | 621 | |
14c43f76 DQ |
622 | entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); |
623 | if (IS_ERR(entry.label)) { | |
624 | status = PTR_ERR(entry.label); | |
625 | goto out; | |
626 | } | |
627 | ||
0795bf83 | 628 | array = kmap(page); |
d1bacf9e BS |
629 | memset(array, 0, sizeof(struct nfs_cache_array)); |
630 | array->eof_index = -1; | |
00a92642 | 631 | |
c7e9668e | 632 | status = nfs_readdir_alloc_pages(pages, array_size); |
6650239a | 633 | if (status < 0) |
d1bacf9e BS |
634 | goto out_release_array; |
635 | do { | |
ac396128 | 636 | unsigned int pglen; |
56e4ebf8 | 637 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 638 | |
d1bacf9e | 639 | if (status < 0) |
00a92642 | 640 | break; |
ac396128 | 641 | pglen = status; |
6650239a | 642 | status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); |
8cd51a0c TM |
643 | if (status < 0) { |
644 | if (status == -ENOSPC) | |
645 | status = 0; | |
646 | break; | |
647 | } | |
648 | } while (array->eof_index < 0); | |
d1bacf9e | 649 | |
c7e9668e | 650 | nfs_readdir_free_pages(pages, array_size); |
d1bacf9e | 651 | out_release_array: |
0795bf83 | 652 | kunmap(page); |
14c43f76 | 653 | nfs4_label_free(entry.label); |
d1bacf9e BS |
654 | out: |
655 | nfs_free_fattr(entry.fattr); | |
656 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
657 | return status; |
658 | } | |
659 | ||
660 | /* | |
d1bacf9e BS |
661 | * Now we cache directories properly, by converting xdr information |
662 | * to an array that can be used for lookups later. This results in | |
663 | * fewer cache pages, since we can store more information on each page. | |
664 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 665 | */ |
d1bacf9e BS |
666 | static |
667 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 668 | { |
496ad9aa | 669 | struct inode *inode = file_inode(desc->file); |
8cd51a0c | 670 | int ret; |
1da177e4 | 671 | |
8cd51a0c TM |
672 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
673 | if (ret < 0) | |
d1bacf9e BS |
674 | goto error; |
675 | SetPageUptodate(page); | |
1da177e4 | 676 | |
d1bacf9e BS |
677 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
678 | /* Should never happen */ | |
679 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 680 | } |
d1bacf9e BS |
681 | unlock_page(page); |
682 | return 0; | |
683 | error: | |
684 | unlock_page(page); | |
8cd51a0c | 685 | return ret; |
d1bacf9e | 686 | } |
1da177e4 | 687 | |
d1bacf9e BS |
688 | static |
689 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
690 | { | |
b044f645 BC |
691 | if (!desc->page->mapping) |
692 | nfs_readdir_clear_array(desc->page); | |
09cbfeaf | 693 | put_page(desc->page); |
d1bacf9e BS |
694 | desc->page = NULL; |
695 | } | |
696 | ||
697 | static | |
698 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
699 | { | |
b044f645 | 700 | return read_cache_page(desc->file->f_mapping, |
d1bacf9e | 701 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
1da177e4 LT |
702 | } |
703 | ||
704 | /* | |
d1bacf9e | 705 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 706 | */ |
d1bacf9e BS |
707 | static |
708 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
709 | { | |
710 | int res; | |
711 | ||
712 | desc->page = get_cache_page(desc); | |
713 | if (IS_ERR(desc->page)) | |
714 | return PTR_ERR(desc->page); | |
715 | ||
716 | res = nfs_readdir_search_array(desc); | |
47c716cb TM |
717 | if (res != 0) |
718 | cache_page_release(desc); | |
d1bacf9e BS |
719 | return res; |
720 | } | |
721 | ||
722 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
723 | static inline |
724 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
725 | { | |
8cd51a0c | 726 | int res; |
d1bacf9e | 727 | |
0aded708 | 728 | if (desc->page_index == 0) { |
8cd51a0c | 729 | desc->current_index = 0; |
0aded708 TM |
730 | desc->last_cookie = 0; |
731 | } | |
47c716cb | 732 | do { |
d1bacf9e | 733 | res = find_cache_page(desc); |
47c716cb | 734 | } while (res == -EAGAIN); |
1da177e4 LT |
735 | return res; |
736 | } | |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Once we've found the start of the dirent within a page: fill 'er up... | |
740 | */ | |
741 | static | |
23db8620 | 742 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc) |
1da177e4 LT |
743 | { |
744 | struct file *file = desc->file; | |
d1bacf9e BS |
745 | int i = 0; |
746 | int res = 0; | |
747 | struct nfs_cache_array *array = NULL; | |
8ef2ce3e BS |
748 | struct nfs_open_dir_context *ctx = file->private_data; |
749 | ||
0795bf83 | 750 | array = kmap(desc->page); |
d1bacf9e | 751 | for (i = desc->cache_entry_index; i < array->size; i++) { |
ece0b423 | 752 | struct nfs_cache_array_entry *ent; |
1da177e4 | 753 | |
ece0b423 | 754 | ent = &array->array[i]; |
23db8620 AV |
755 | if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, |
756 | nfs_compat_user_ino64(ent->ino), ent->d_type)) { | |
6089dd0d | 757 | desc->eof = true; |
1da177e4 | 758 | break; |
ece0b423 | 759 | } |
23db8620 | 760 | desc->ctx->pos++; |
d1bacf9e BS |
761 | if (i < (array->size-1)) |
762 | *desc->dir_cookie = array->array[i+1].cookie; | |
763 | else | |
764 | *desc->dir_cookie = array->last_cookie; | |
0c030806 TM |
765 | if (ctx->duped != 0) |
766 | ctx->duped = 1; | |
1da177e4 | 767 | } |
47c716cb | 768 | if (array->eof_index >= 0) |
6089dd0d | 769 | desc->eof = true; |
d1bacf9e | 770 | |
0795bf83 | 771 | kunmap(desc->page); |
d1bacf9e | 772 | cache_page_release(desc); |
1e7cb3dc CL |
773 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
774 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
775 | return res; |
776 | } | |
777 | ||
778 | /* | |
779 | * If we cannot find a cookie in our cache, we suspect that this is | |
780 | * because it points to a deleted file, so we ask the server to return | |
781 | * whatever it thinks is the next entry. We then feed this to filldir. | |
782 | * If all goes well, we should then be able to find our way round the | |
783 | * cache on the next call to readdir_search_pagecache(); | |
784 | * | |
785 | * NOTE: we cannot add the anonymous page to the pagecache because | |
786 | * the data it contains might not be page aligned. Besides, | |
787 | * we should already have a complete representation of the | |
788 | * directory in the page cache by the time we get here. | |
789 | */ | |
790 | static inline | |
23db8620 | 791 | int uncached_readdir(nfs_readdir_descriptor_t *desc) |
1da177e4 | 792 | { |
1da177e4 LT |
793 | struct page *page = NULL; |
794 | int status; | |
496ad9aa | 795 | struct inode *inode = file_inode(desc->file); |
0c030806 | 796 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
1da177e4 | 797 | |
1e7cb3dc CL |
798 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
799 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
800 | |
801 | page = alloc_page(GFP_HIGHUSER); | |
802 | if (!page) { | |
803 | status = -ENOMEM; | |
804 | goto out; | |
805 | } | |
d1bacf9e | 806 | |
7a8e1dc3 | 807 | desc->page_index = 0; |
0aded708 | 808 | desc->last_cookie = *desc->dir_cookie; |
7a8e1dc3 | 809 | desc->page = page; |
0c030806 | 810 | ctx->duped = 0; |
7a8e1dc3 | 811 | |
85f8607e TM |
812 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
813 | if (status < 0) | |
1da177e4 LT |
814 | goto out_release; |
815 | ||
23db8620 | 816 | status = nfs_do_filldir(desc); |
1da177e4 | 817 | |
1da177e4 | 818 | out: |
1e7cb3dc | 819 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 820 | __func__, status); |
1da177e4 LT |
821 | return status; |
822 | out_release: | |
d1bacf9e | 823 | cache_page_release(desc); |
1da177e4 LT |
824 | goto out; |
825 | } | |
826 | ||
00a92642 OG |
827 | /* The file offset position represents the dirent entry number. A |
828 | last cookie cache takes care of the common case of reading the | |
829 | whole directory. | |
1da177e4 | 830 | */ |
23db8620 | 831 | static int nfs_readdir(struct file *file, struct dir_context *ctx) |
1da177e4 | 832 | { |
be62a1a8 | 833 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 834 | struct inode *inode = d_inode(dentry); |
1da177e4 LT |
835 | nfs_readdir_descriptor_t my_desc, |
836 | *desc = &my_desc; | |
23db8620 | 837 | struct nfs_open_dir_context *dir_ctx = file->private_data; |
07b5ce8e | 838 | int res = 0; |
1da177e4 | 839 | |
6de1472f AV |
840 | dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", |
841 | file, (long long)ctx->pos); | |
91d5b470 CL |
842 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
843 | ||
1da177e4 | 844 | /* |
23db8620 | 845 | * ctx->pos points to the dirent entry number. |
f0dd2136 | 846 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
847 | * to either find the entry with the appropriate number or |
848 | * revalidate the cookie. | |
1da177e4 LT |
849 | */ |
850 | memset(desc, 0, sizeof(*desc)); | |
851 | ||
23db8620 AV |
852 | desc->file = file; |
853 | desc->ctx = ctx; | |
480c2006 | 854 | desc->dir_cookie = &dir_ctx->dir_cookie; |
1da177e4 | 855 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
a7a3b1e9 | 856 | desc->plus = nfs_use_readdirplus(inode, ctx); |
1da177e4 | 857 | |
79f687a3 | 858 | if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) |
07b5ce8e | 859 | res = nfs_revalidate_mapping(inode, file->f_mapping); |
fccca7fc TM |
860 | if (res < 0) |
861 | goto out; | |
862 | ||
47c716cb | 863 | do { |
1da177e4 | 864 | res = readdir_search_pagecache(desc); |
00a92642 | 865 | |
1da177e4 | 866 | if (res == -EBADCOOKIE) { |
ece0b423 | 867 | res = 0; |
1da177e4 | 868 | /* This means either end of directory */ |
6089dd0d | 869 | if (*desc->dir_cookie && !desc->eof) { |
1da177e4 | 870 | /* Or that the server has 'lost' a cookie */ |
23db8620 | 871 | res = uncached_readdir(desc); |
ece0b423 | 872 | if (res == 0) |
1da177e4 LT |
873 | continue; |
874 | } | |
1da177e4 LT |
875 | break; |
876 | } | |
877 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 878 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 879 | nfs_zap_caches(inode); |
baf57a09 | 880 | desc->page_index = 0; |
a7a3b1e9 BC |
881 | desc->plus = false; |
882 | desc->eof = false; | |
1da177e4 LT |
883 | continue; |
884 | } | |
885 | if (res < 0) | |
886 | break; | |
887 | ||
23db8620 | 888 | res = nfs_do_filldir(desc); |
ece0b423 | 889 | if (res < 0) |
1da177e4 | 890 | break; |
47c716cb | 891 | } while (!desc->eof); |
fccca7fc | 892 | out: |
1e7cb3dc CL |
893 | if (res > 0) |
894 | res = 0; | |
6de1472f | 895 | dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); |
1e7cb3dc | 896 | return res; |
1da177e4 LT |
897 | } |
898 | ||
965c8e59 | 899 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) |
f0dd2136 | 900 | { |
b044f645 | 901 | struct inode *inode = file_inode(filp); |
480c2006 | 902 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
b84e06c5 | 903 | |
6de1472f AV |
904 | dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", |
905 | filp, offset, whence); | |
b84e06c5 | 906 | |
b044f645 | 907 | inode_lock(inode); |
965c8e59 | 908 | switch (whence) { |
f0dd2136 TM |
909 | case 1: |
910 | offset += filp->f_pos; | |
911 | case 0: | |
912 | if (offset >= 0) | |
913 | break; | |
914 | default: | |
b044f645 BC |
915 | offset = -EINVAL; |
916 | goto out; | |
f0dd2136 TM |
917 | } |
918 | if (offset != filp->f_pos) { | |
919 | filp->f_pos = offset; | |
480c2006 | 920 | dir_ctx->dir_cookie = 0; |
8ef2ce3e | 921 | dir_ctx->duped = 0; |
f0dd2136 | 922 | } |
b044f645 BC |
923 | out: |
924 | inode_unlock(inode); | |
f0dd2136 TM |
925 | return offset; |
926 | } | |
927 | ||
1da177e4 LT |
928 | /* |
929 | * All directory operations under NFS are synchronous, so fsync() | |
930 | * is a dummy operation. | |
931 | */ | |
02c24a82 JB |
932 | static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, |
933 | int datasync) | |
1da177e4 | 934 | { |
6de1472f | 935 | struct inode *inode = file_inode(filp); |
7ea80859 | 936 | |
6de1472f | 937 | dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); |
1e7cb3dc | 938 | |
5955102c | 939 | inode_lock(inode); |
6de1472f | 940 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
5955102c | 941 | inode_unlock(inode); |
1da177e4 LT |
942 | return 0; |
943 | } | |
944 | ||
bfc69a45 TM |
945 | /** |
946 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
947 | * @dir - pointer to directory inode | |
948 | * | |
949 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
950 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
951 | * on the server that might have invalidated our dcache. | |
952 | * | |
953 | * The caller should be holding dir->i_lock | |
954 | */ | |
955 | void nfs_force_lookup_revalidate(struct inode *dir) | |
956 | { | |
011935a0 | 957 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 | 958 | } |
89d77c8f | 959 | EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); |
bfc69a45 | 960 | |
1da177e4 LT |
961 | /* |
962 | * A check for whether or not the parent directory has changed. | |
963 | * In the case it has, we assume that the dentries are untrustworthy | |
964 | * and may need to be looked up again. | |
912a108d | 965 | * If rcu_walk prevents us from performing a full check, return 0. |
1da177e4 | 966 | */ |
912a108d N |
967 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, |
968 | int rcu_walk) | |
1da177e4 LT |
969 | { |
970 | if (IS_ROOT(dentry)) | |
971 | return 1; | |
4eec952e TM |
972 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
973 | return 0; | |
f2c77f4e TM |
974 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
975 | return 0; | |
976 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
1cd9cb05 TM |
977 | if (nfs_mapping_need_revalidate_inode(dir)) { |
978 | if (rcu_walk) | |
979 | return 0; | |
980 | if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
981 | return 0; | |
982 | } | |
f2c77f4e TM |
983 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
984 | return 0; | |
985 | return 1; | |
1da177e4 LT |
986 | } |
987 | ||
a12802ca TM |
988 | /* |
989 | * Use intent information to check whether or not we're going to do | |
990 | * an O_EXCL create using this path component. | |
991 | */ | |
fa3c56bb | 992 | static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) |
a12802ca TM |
993 | { |
994 | if (NFS_PROTO(dir)->version == 2) | |
995 | return 0; | |
fa3c56bb | 996 | return flags & LOOKUP_EXCL; |
a12802ca TM |
997 | } |
998 | ||
1d6757fb TM |
999 | /* |
1000 | * Inode and filehandle revalidation for lookups. | |
1001 | * | |
1002 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
1003 | * or if the intent information indicates that we're about to open this | |
1004 | * particular file and the "nocto" mount flag is not set. | |
1005 | * | |
1006 | */ | |
65a0c149 | 1007 | static |
fa3c56bb | 1008 | int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) |
1da177e4 LT |
1009 | { |
1010 | struct nfs_server *server = NFS_SERVER(inode); | |
65a0c149 | 1011 | int ret; |
1da177e4 | 1012 | |
36d43a43 | 1013 | if (IS_AUTOMOUNT(inode)) |
4e99a1ff | 1014 | return 0; |
facc3530 | 1015 | /* VFS wants an on-the-wire revalidation */ |
fa3c56bb | 1016 | if (flags & LOOKUP_REVAL) |
facc3530 AV |
1017 | goto out_force; |
1018 | /* This is an open(2) */ | |
fa3c56bb AV |
1019 | if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && |
1020 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) | |
facc3530 | 1021 | goto out_force; |
65a0c149 TM |
1022 | out: |
1023 | return (inode->i_nlink == 0) ? -ENOENT : 0; | |
1da177e4 | 1024 | out_force: |
1fa1e384 N |
1025 | if (flags & LOOKUP_RCU) |
1026 | return -ECHILD; | |
65a0c149 TM |
1027 | ret = __nfs_revalidate_inode(server, inode); |
1028 | if (ret != 0) | |
1029 | return ret; | |
1030 | goto out; | |
1da177e4 LT |
1031 | } |
1032 | ||
1033 | /* | |
1034 | * We judge how long we want to trust negative | |
1035 | * dentries by looking at the parent inode mtime. | |
1036 | * | |
1037 | * If parent mtime has changed, we revalidate, else we wait for a | |
1038 | * period corresponding to the parent's attribute cache timeout value. | |
912a108d N |
1039 | * |
1040 | * If LOOKUP_RCU prevents us from performing a full check, return 1 | |
1041 | * suggesting a reval is needed. | |
1da177e4 LT |
1042 | */ |
1043 | static inline | |
1044 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
fa3c56bb | 1045 | unsigned int flags) |
1da177e4 | 1046 | { |
1da177e4 | 1047 | /* Don't revalidate a negative dentry if we're creating a new file */ |
fa3c56bb | 1048 | if (flags & LOOKUP_CREATE) |
1da177e4 | 1049 | return 0; |
4eec952e TM |
1050 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1051 | return 1; | |
912a108d | 1052 | return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); |
1da177e4 LT |
1053 | } |
1054 | ||
1055 | /* | |
1056 | * This is called every time the dcache has a lookup hit, | |
1057 | * and we should check whether we can really trust that | |
1058 | * lookup. | |
1059 | * | |
1060 | * NOTE! The hit can be a negative hit too, don't assume | |
1061 | * we have an inode! | |
1062 | * | |
1063 | * If the parent directory is seen to have changed, we throw out the | |
1064 | * cached dentry and do a new lookup. | |
1065 | */ | |
0b728e19 | 1066 | static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
1da177e4 LT |
1067 | { |
1068 | struct inode *dir; | |
1069 | struct inode *inode; | |
1070 | struct dentry *parent; | |
e1fb4d05 TM |
1071 | struct nfs_fh *fhandle = NULL; |
1072 | struct nfs_fattr *fattr = NULL; | |
1775fd3e | 1073 | struct nfs4_label *label = NULL; |
1da177e4 | 1074 | int error; |
1da177e4 | 1075 | |
d51ac1a8 | 1076 | if (flags & LOOKUP_RCU) { |
6aa7de05 | 1077 | parent = READ_ONCE(dentry->d_parent); |
2b0143b5 | 1078 | dir = d_inode_rcu(parent); |
d51ac1a8 N |
1079 | if (!dir) |
1080 | return -ECHILD; | |
1081 | } else { | |
1082 | parent = dget_parent(dentry); | |
2b0143b5 | 1083 | dir = d_inode(parent); |
d51ac1a8 | 1084 | } |
91d5b470 | 1085 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
2b0143b5 | 1086 | inode = d_inode(dentry); |
1da177e4 LT |
1087 | |
1088 | if (!inode) { | |
912a108d N |
1089 | if (nfs_neg_need_reval(dir, dentry, flags)) { |
1090 | if (flags & LOOKUP_RCU) | |
1091 | return -ECHILD; | |
1da177e4 | 1092 | goto out_bad; |
912a108d | 1093 | } |
63519fbc | 1094 | goto out_valid; |
1da177e4 LT |
1095 | } |
1096 | ||
1097 | if (is_bad_inode(inode)) { | |
d51ac1a8 N |
1098 | if (flags & LOOKUP_RCU) |
1099 | return -ECHILD; | |
6de1472f AV |
1100 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1101 | __func__, dentry); | |
1da177e4 LT |
1102 | goto out_bad; |
1103 | } | |
1104 | ||
011e2a7f | 1105 | if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) |
15860ab1 TM |
1106 | goto out_set_verifier; |
1107 | ||
1da177e4 | 1108 | /* Force a full look up iff the parent directory has changed */ |
912a108d N |
1109 | if (!nfs_is_exclusive_create(dir, flags) && |
1110 | nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { | |
cc89684c N |
1111 | error = nfs_lookup_verify_inode(inode, flags); |
1112 | if (error) { | |
1fa1e384 N |
1113 | if (flags & LOOKUP_RCU) |
1114 | return -ECHILD; | |
cc89684c N |
1115 | if (error == -ESTALE) |
1116 | goto out_zap_parent; | |
1117 | goto out_error; | |
1fa1e384 | 1118 | } |
63519fbc | 1119 | nfs_advise_use_readdirplus(dir); |
1da177e4 LT |
1120 | goto out_valid; |
1121 | } | |
1122 | ||
912a108d N |
1123 | if (flags & LOOKUP_RCU) |
1124 | return -ECHILD; | |
1125 | ||
1da177e4 LT |
1126 | if (NFS_STALE(inode)) |
1127 | goto out_bad; | |
1128 | ||
e1fb4d05 TM |
1129 | error = -ENOMEM; |
1130 | fhandle = nfs_alloc_fhandle(); | |
1131 | fattr = nfs_alloc_fattr(); | |
1132 | if (fhandle == NULL || fattr == NULL) | |
1133 | goto out_error; | |
1134 | ||
14c43f76 DQ |
1135 | label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); |
1136 | if (IS_ERR(label)) | |
1137 | goto out_error; | |
1138 | ||
6e0d0be7 | 1139 | trace_nfs_lookup_revalidate_enter(dir, dentry, flags); |
1775fd3e | 1140 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); |
6e0d0be7 | 1141 | trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); |
cc89684c | 1142 | if (error == -ESTALE || error == -ENOENT) |
1da177e4 | 1143 | goto out_bad; |
cc89684c N |
1144 | if (error) |
1145 | goto out_error; | |
e1fb4d05 | 1146 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1147 | goto out_bad; |
e1fb4d05 | 1148 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1149 | goto out_bad; |
1150 | ||
aa9c2669 DQ |
1151 | nfs_setsecurity(inode, fattr, label); |
1152 | ||
e1fb4d05 TM |
1153 | nfs_free_fattr(fattr); |
1154 | nfs_free_fhandle(fhandle); | |
14c43f76 DQ |
1155 | nfs4_label_free(label); |
1156 | ||
63519fbc TM |
1157 | /* set a readdirplus hint that we had a cache miss */ |
1158 | nfs_force_use_readdirplus(dir); | |
1159 | ||
15860ab1 | 1160 | out_set_verifier: |
cf8ba45e | 1161 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1162 | out_valid: |
d51ac1a8 | 1163 | if (flags & LOOKUP_RCU) { |
6aa7de05 | 1164 | if (parent != READ_ONCE(dentry->d_parent)) |
d51ac1a8 N |
1165 | return -ECHILD; |
1166 | } else | |
1167 | dput(parent); | |
6de1472f AV |
1168 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", |
1169 | __func__, dentry); | |
1da177e4 LT |
1170 | return 1; |
1171 | out_zap_parent: | |
1172 | nfs_zap_caches(dir); | |
1173 | out_bad: | |
d51ac1a8 | 1174 | WARN_ON(flags & LOOKUP_RCU); |
c44600c9 AV |
1175 | nfs_free_fattr(fattr); |
1176 | nfs_free_fhandle(fhandle); | |
14c43f76 | 1177 | nfs4_label_free(label); |
a1643a92 | 1178 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1179 | if (inode && S_ISDIR(inode->i_mode)) { |
1180 | /* Purge readdir caches. */ | |
1181 | nfs_zap_caches(inode); | |
a3f432bf BF |
1182 | /* |
1183 | * We can't d_drop the root of a disconnected tree: | |
1184 | * its d_hash is on the s_anon list and d_drop() would hide | |
1185 | * it from shrink_dcache_for_unmount(), leading to busy | |
1186 | * inodes on unmount and further oopses. | |
1187 | */ | |
1188 | if (IS_ROOT(dentry)) | |
d9e80b7d | 1189 | goto out_valid; |
1da177e4 | 1190 | } |
1da177e4 | 1191 | dput(parent); |
6de1472f AV |
1192 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", |
1193 | __func__, dentry); | |
1da177e4 | 1194 | return 0; |
e1fb4d05 | 1195 | out_error: |
d51ac1a8 | 1196 | WARN_ON(flags & LOOKUP_RCU); |
e1fb4d05 TM |
1197 | nfs_free_fattr(fattr); |
1198 | nfs_free_fhandle(fhandle); | |
14c43f76 | 1199 | nfs4_label_free(label); |
e1fb4d05 | 1200 | dput(parent); |
6de1472f AV |
1201 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", |
1202 | __func__, dentry, error); | |
e1fb4d05 | 1203 | return error; |
1da177e4 LT |
1204 | } |
1205 | ||
ecf3d1f1 | 1206 | /* |
2b0143b5 | 1207 | * A weaker form of d_revalidate for revalidating just the d_inode(dentry) |
ecf3d1f1 JL |
1208 | * when we don't really care about the dentry name. This is called when a |
1209 | * pathwalk ends on a dentry that was not found via a normal lookup in the | |
1210 | * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). | |
1211 | * | |
1212 | * In this situation, we just want to verify that the inode itself is OK | |
1213 | * since the dentry might have changed on the server. | |
1214 | */ | |
1215 | static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) | |
1216 | { | |
2b0143b5 | 1217 | struct inode *inode = d_inode(dentry); |
9cdd1d3f | 1218 | int error = 0; |
ecf3d1f1 JL |
1219 | |
1220 | /* | |
1221 | * I believe we can only get a negative dentry here in the case of a | |
1222 | * procfs-style symlink. Just assume it's correct for now, but we may | |
1223 | * eventually need to do something more here. | |
1224 | */ | |
1225 | if (!inode) { | |
6de1472f AV |
1226 | dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", |
1227 | __func__, dentry); | |
ecf3d1f1 JL |
1228 | return 1; |
1229 | } | |
1230 | ||
1231 | if (is_bad_inode(inode)) { | |
6de1472f AV |
1232 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", |
1233 | __func__, dentry); | |
ecf3d1f1 JL |
1234 | return 0; |
1235 | } | |
1236 | ||
b688741c | 1237 | error = nfs_lookup_verify_inode(inode, flags); |
ecf3d1f1 JL |
1238 | dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", |
1239 | __func__, inode->i_ino, error ? "invalid" : "valid"); | |
1240 | return !error; | |
1241 | } | |
1242 | ||
1da177e4 LT |
1243 | /* |
1244 | * This is called from dput() when d_count is going to 0. | |
1245 | */ | |
fe15ce44 | 1246 | static int nfs_dentry_delete(const struct dentry *dentry) |
1da177e4 | 1247 | { |
6de1472f AV |
1248 | dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", |
1249 | dentry, dentry->d_flags); | |
1da177e4 | 1250 | |
77f11192 | 1251 | /* Unhash any dentry with a stale inode */ |
2b0143b5 | 1252 | if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) |
77f11192 TM |
1253 | return 1; |
1254 | ||
1da177e4 LT |
1255 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1256 | /* Unhash it, so that ->d_iput() would be called */ | |
1257 | return 1; | |
1258 | } | |
1751e8a6 | 1259 | if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { |
1da177e4 LT |
1260 | /* Unhash it, so that ancestors of killed async unlink |
1261 | * files will be cleaned up during umount */ | |
1262 | return 1; | |
1263 | } | |
1264 | return 0; | |
1265 | ||
1266 | } | |
1267 | ||
1f018458 | 1268 | /* Ensure that we revalidate inode->i_nlink */ |
1b83d707 TM |
1269 | static void nfs_drop_nlink(struct inode *inode) |
1270 | { | |
1271 | spin_lock(&inode->i_lock); | |
1f018458 TM |
1272 | /* drop the inode if we're reasonably sure this is the last link */ |
1273 | if (inode->i_nlink == 1) | |
1274 | clear_nlink(inode); | |
16e14375 TM |
1275 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE |
1276 | | NFS_INO_INVALID_CTIME | |
1277 | | NFS_INO_INVALID_OTHER; | |
1b83d707 TM |
1278 | spin_unlock(&inode->i_lock); |
1279 | } | |
1280 | ||
1da177e4 LT |
1281 | /* |
1282 | * Called when the dentry loses inode. | |
1283 | * We use it to clean up silly-renamed files. | |
1284 | */ | |
1285 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1286 | { | |
83672d39 NB |
1287 | if (S_ISDIR(inode->i_mode)) |
1288 | /* drop any readdir cache as it could easily be old */ | |
1289 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1290 | ||
1da177e4 | 1291 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
e4eff1a6 | 1292 | nfs_complete_unlink(dentry, inode); |
1f018458 | 1293 | nfs_drop_nlink(inode); |
1da177e4 | 1294 | } |
1da177e4 LT |
1295 | iput(inode); |
1296 | } | |
1297 | ||
b1942c5f AV |
1298 | static void nfs_d_release(struct dentry *dentry) |
1299 | { | |
1300 | /* free cached devname value, if it survived that far */ | |
1301 | if (unlikely(dentry->d_fsdata)) { | |
1302 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1303 | WARN_ON(1); | |
1304 | else | |
1305 | kfree(dentry->d_fsdata); | |
1306 | } | |
1307 | } | |
1308 | ||
f786aa90 | 1309 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 | 1310 | .d_revalidate = nfs_lookup_revalidate, |
ecf3d1f1 | 1311 | .d_weak_revalidate = nfs_weak_revalidate, |
1da177e4 LT |
1312 | .d_delete = nfs_dentry_delete, |
1313 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1314 | .d_automount = nfs_d_automount, |
b1942c5f | 1315 | .d_release = nfs_d_release, |
1da177e4 | 1316 | }; |
ddda8e0a | 1317 | EXPORT_SYMBOL_GPL(nfs_dentry_operations); |
1da177e4 | 1318 | |
597d9289 | 1319 | struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) |
1da177e4 LT |
1320 | { |
1321 | struct dentry *res; | |
1322 | struct inode *inode = NULL; | |
e1fb4d05 TM |
1323 | struct nfs_fh *fhandle = NULL; |
1324 | struct nfs_fattr *fattr = NULL; | |
1775fd3e | 1325 | struct nfs4_label *label = NULL; |
1da177e4 | 1326 | int error; |
1da177e4 | 1327 | |
6de1472f | 1328 | dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); |
91d5b470 | 1329 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 | 1330 | |
130f9ab7 AV |
1331 | if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) |
1332 | return ERR_PTR(-ENAMETOOLONG); | |
1da177e4 | 1333 | |
fd684071 TM |
1334 | /* |
1335 | * If we're doing an exclusive create, optimize away the lookup | |
1336 | * but don't hash the dentry. | |
1337 | */ | |
130f9ab7 AV |
1338 | if (nfs_is_exclusive_create(dir, flags)) |
1339 | return NULL; | |
1da177e4 | 1340 | |
e1fb4d05 TM |
1341 | res = ERR_PTR(-ENOMEM); |
1342 | fhandle = nfs_alloc_fhandle(); | |
1343 | fattr = nfs_alloc_fattr(); | |
1344 | if (fhandle == NULL || fattr == NULL) | |
1345 | goto out; | |
1346 | ||
14c43f76 DQ |
1347 | label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); |
1348 | if (IS_ERR(label)) | |
1349 | goto out; | |
1350 | ||
6e0d0be7 | 1351 | trace_nfs_lookup_enter(dir, dentry, flags); |
1775fd3e | 1352 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); |
1da177e4 LT |
1353 | if (error == -ENOENT) |
1354 | goto no_entry; | |
1355 | if (error < 0) { | |
1356 | res = ERR_PTR(error); | |
bf130914 | 1357 | goto out_label; |
1da177e4 | 1358 | } |
1775fd3e | 1359 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
bf0c84f1 | 1360 | res = ERR_CAST(inode); |
03f28e3a | 1361 | if (IS_ERR(res)) |
bf130914 | 1362 | goto out_label; |
54ceac45 | 1363 | |
63519fbc TM |
1364 | /* Notify readdir to use READDIRPLUS */ |
1365 | nfs_force_use_readdirplus(dir); | |
d69ee9b8 | 1366 | |
1da177e4 | 1367 | no_entry: |
41d28bca | 1368 | res = d_splice_alias(inode, dentry); |
9eaef27b TM |
1369 | if (res != NULL) { |
1370 | if (IS_ERR(res)) | |
bf130914 | 1371 | goto out_label; |
1da177e4 | 1372 | dentry = res; |
9eaef27b | 1373 | } |
1da177e4 | 1374 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
bf130914 | 1375 | out_label: |
6e0d0be7 | 1376 | trace_nfs_lookup_exit(dir, dentry, flags, error); |
14c43f76 | 1377 | nfs4_label_free(label); |
1da177e4 | 1378 | out: |
e1fb4d05 TM |
1379 | nfs_free_fattr(fattr); |
1380 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1381 | return res; |
1382 | } | |
ddda8e0a | 1383 | EXPORT_SYMBOL_GPL(nfs_lookup); |
1da177e4 | 1384 | |
89d77c8f | 1385 | #if IS_ENABLED(CONFIG_NFS_V4) |
0b728e19 | 1386 | static int nfs4_lookup_revalidate(struct dentry *, unsigned int); |
1da177e4 | 1387 | |
f786aa90 | 1388 | const struct dentry_operations nfs4_dentry_operations = { |
0ef97dcf | 1389 | .d_revalidate = nfs4_lookup_revalidate, |
b688741c | 1390 | .d_weak_revalidate = nfs_weak_revalidate, |
1da177e4 LT |
1391 | .d_delete = nfs_dentry_delete, |
1392 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1393 | .d_automount = nfs_d_automount, |
b1942c5f | 1394 | .d_release = nfs_d_release, |
1da177e4 | 1395 | }; |
89d77c8f | 1396 | EXPORT_SYMBOL_GPL(nfs4_dentry_operations); |
1da177e4 | 1397 | |
8a5e929d AV |
1398 | static fmode_t flags_to_mode(int flags) |
1399 | { | |
1400 | fmode_t res = (__force fmode_t)flags & FMODE_EXEC; | |
1401 | if ((flags & O_ACCMODE) != O_WRONLY) | |
1402 | res |= FMODE_READ; | |
1403 | if ((flags & O_ACCMODE) != O_RDONLY) | |
1404 | res |= FMODE_WRITE; | |
1405 | return res; | |
1406 | } | |
1407 | ||
532d4def | 1408 | static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) |
cd9a1c0e | 1409 | { |
532d4def | 1410 | return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); |
cd9a1c0e TM |
1411 | } |
1412 | ||
1413 | static int do_open(struct inode *inode, struct file *filp) | |
1414 | { | |
f1fe29b4 | 1415 | nfs_fscache_open_file(inode, filp); |
cd9a1c0e TM |
1416 | return 0; |
1417 | } | |
1418 | ||
d9585277 AV |
1419 | static int nfs_finish_open(struct nfs_open_context *ctx, |
1420 | struct dentry *dentry, | |
30d90494 | 1421 | struct file *file, unsigned open_flags, |
d9585277 | 1422 | int *opened) |
cd9a1c0e | 1423 | { |
0dd2b474 MS |
1424 | int err; |
1425 | ||
30d90494 AV |
1426 | err = finish_open(file, dentry, do_open, opened); |
1427 | if (err) | |
d9585277 | 1428 | goto out; |
eaa2b82c N |
1429 | if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) |
1430 | nfs_file_set_open_context(file, ctx); | |
1431 | else | |
1432 | err = -ESTALE; | |
cd9a1c0e | 1433 | out: |
d9585277 | 1434 | return err; |
cd9a1c0e TM |
1435 | } |
1436 | ||
73a79706 BS |
1437 | int nfs_atomic_open(struct inode *dir, struct dentry *dentry, |
1438 | struct file *file, unsigned open_flags, | |
1439 | umode_t mode, int *opened) | |
1da177e4 | 1440 | { |
c94c0953 | 1441 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); |
cd9a1c0e | 1442 | struct nfs_open_context *ctx; |
0dd2b474 MS |
1443 | struct dentry *res; |
1444 | struct iattr attr = { .ia_valid = ATTR_OPEN }; | |
f46e0bd3 | 1445 | struct inode *inode; |
1472b83e | 1446 | unsigned int lookup_flags = 0; |
c94c0953 | 1447 | bool switched = false; |
898f635c | 1448 | int err; |
1da177e4 | 1449 | |
0dd2b474 | 1450 | /* Expect a negative dentry */ |
2b0143b5 | 1451 | BUG_ON(d_inode(dentry)); |
0dd2b474 | 1452 | |
1e8968c5 | 1453 | dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", |
6de1472f | 1454 | dir->i_sb->s_id, dir->i_ino, dentry); |
1e7cb3dc | 1455 | |
9597c13b JL |
1456 | err = nfs_check_flags(open_flags); |
1457 | if (err) | |
1458 | return err; | |
1459 | ||
0dd2b474 MS |
1460 | /* NFS only supports OPEN on regular files */ |
1461 | if ((open_flags & O_DIRECTORY)) { | |
00699ad8 | 1462 | if (!d_in_lookup(dentry)) { |
0dd2b474 MS |
1463 | /* |
1464 | * Hashed negative dentry with O_DIRECTORY: dentry was | |
1465 | * revalidated and is fine, no need to perform lookup | |
1466 | * again | |
1467 | */ | |
d9585277 | 1468 | return -ENOENT; |
0dd2b474 | 1469 | } |
1472b83e | 1470 | lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; |
1da177e4 | 1471 | goto no_open; |
02a913a7 | 1472 | } |
1da177e4 | 1473 | |
0dd2b474 | 1474 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) |
d9585277 | 1475 | return -ENAMETOOLONG; |
cd9a1c0e | 1476 | |
0dd2b474 | 1477 | if (open_flags & O_CREAT) { |
dff25ddb AG |
1478 | struct nfs_server *server = NFS_SERVER(dir); |
1479 | ||
1480 | if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) | |
1481 | mode &= ~current_umask(); | |
1482 | ||
536e43d1 | 1483 | attr.ia_valid |= ATTR_MODE; |
dff25ddb | 1484 | attr.ia_mode = mode; |
0dd2b474 | 1485 | } |
536e43d1 TM |
1486 | if (open_flags & O_TRUNC) { |
1487 | attr.ia_valid |= ATTR_SIZE; | |
1488 | attr.ia_size = 0; | |
cd9a1c0e TM |
1489 | } |
1490 | ||
c94c0953 AV |
1491 | if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { |
1492 | d_drop(dentry); | |
1493 | switched = true; | |
1494 | dentry = d_alloc_parallel(dentry->d_parent, | |
1495 | &dentry->d_name, &wq); | |
1496 | if (IS_ERR(dentry)) | |
1497 | return PTR_ERR(dentry); | |
1498 | if (unlikely(!d_in_lookup(dentry))) | |
1499 | return finish_no_open(file, dentry); | |
1500 | } | |
1501 | ||
532d4def | 1502 | ctx = create_nfs_open_context(dentry, open_flags, file); |
0dd2b474 MS |
1503 | err = PTR_ERR(ctx); |
1504 | if (IS_ERR(ctx)) | |
d9585277 | 1505 | goto out; |
0dd2b474 | 1506 | |
6e0d0be7 | 1507 | trace_nfs_atomic_open_enter(dir, ctx, open_flags); |
5bc2afc2 | 1508 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); |
f46e0bd3 | 1509 | if (IS_ERR(inode)) { |
0dd2b474 | 1510 | err = PTR_ERR(inode); |
6e0d0be7 | 1511 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 1512 | put_nfs_open_context(ctx); |
d20cb71d | 1513 | d_drop(dentry); |
0dd2b474 MS |
1514 | switch (err) { |
1515 | case -ENOENT: | |
774d9513 | 1516 | d_splice_alias(NULL, dentry); |
809fd143 | 1517 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
0dd2b474 MS |
1518 | break; |
1519 | case -EISDIR: | |
1520 | case -ENOTDIR: | |
1521 | goto no_open; | |
1522 | case -ELOOP: | |
1523 | if (!(open_flags & O_NOFOLLOW)) | |
6f926b5b | 1524 | goto no_open; |
0dd2b474 | 1525 | break; |
1da177e4 | 1526 | /* case -EINVAL: */ |
0dd2b474 MS |
1527 | default: |
1528 | break; | |
1da177e4 | 1529 | } |
d9585277 | 1530 | goto out; |
cd9a1c0e | 1531 | } |
0dd2b474 | 1532 | |
275bb307 | 1533 | err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); |
6e0d0be7 | 1534 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); |
2d9db750 | 1535 | put_nfs_open_context(ctx); |
d9585277 | 1536 | out: |
c94c0953 AV |
1537 | if (unlikely(switched)) { |
1538 | d_lookup_done(dentry); | |
1539 | dput(dentry); | |
1540 | } | |
d9585277 | 1541 | return err; |
0dd2b474 | 1542 | |
1da177e4 | 1543 | no_open: |
1472b83e | 1544 | res = nfs_lookup(dir, dentry, lookup_flags); |
c94c0953 AV |
1545 | if (switched) { |
1546 | d_lookup_done(dentry); | |
1547 | if (!res) | |
1548 | res = dentry; | |
1549 | else | |
1550 | dput(dentry); | |
1551 | } | |
0dd2b474 | 1552 | if (IS_ERR(res)) |
c94c0953 | 1553 | return PTR_ERR(res); |
e45198a6 | 1554 | return finish_no_open(file, res); |
1da177e4 | 1555 | } |
89d77c8f | 1556 | EXPORT_SYMBOL_GPL(nfs_atomic_open); |
1da177e4 | 1557 | |
0b728e19 | 1558 | static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) |
1da177e4 | 1559 | { |
657e94b6 | 1560 | struct inode *inode; |
50de348c | 1561 | int ret = 0; |
1da177e4 | 1562 | |
fa3c56bb | 1563 | if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) |
eda72afb MS |
1564 | goto no_open; |
1565 | if (d_mountpoint(dentry)) | |
5584c306 | 1566 | goto no_open; |
49f9a0fa TM |
1567 | if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) |
1568 | goto no_open; | |
2b484297 | 1569 | |
2b0143b5 | 1570 | inode = d_inode(dentry); |
2b484297 | 1571 | |
1da177e4 LT |
1572 | /* We can't create new files in nfs_open_revalidate(), so we |
1573 | * optimize away revalidation of negative dentries. | |
1574 | */ | |
216d5d06 | 1575 | if (inode == NULL) { |
49317a7f N |
1576 | struct dentry *parent; |
1577 | struct inode *dir; | |
1578 | ||
912a108d | 1579 | if (flags & LOOKUP_RCU) { |
6aa7de05 | 1580 | parent = READ_ONCE(dentry->d_parent); |
2b0143b5 | 1581 | dir = d_inode_rcu(parent); |
912a108d N |
1582 | if (!dir) |
1583 | return -ECHILD; | |
1584 | } else { | |
1585 | parent = dget_parent(dentry); | |
2b0143b5 | 1586 | dir = d_inode(parent); |
912a108d | 1587 | } |
fa3c56bb | 1588 | if (!nfs_neg_need_reval(dir, dentry, flags)) |
216d5d06 | 1589 | ret = 1; |
912a108d N |
1590 | else if (flags & LOOKUP_RCU) |
1591 | ret = -ECHILD; | |
1592 | if (!(flags & LOOKUP_RCU)) | |
1593 | dput(parent); | |
6aa7de05 | 1594 | else if (parent != READ_ONCE(dentry->d_parent)) |
912a108d | 1595 | return -ECHILD; |
1da177e4 | 1596 | goto out; |
216d5d06 TM |
1597 | } |
1598 | ||
1da177e4 LT |
1599 | /* NFS only supports OPEN on regular files */ |
1600 | if (!S_ISREG(inode->i_mode)) | |
49317a7f | 1601 | goto no_open; |
1da177e4 | 1602 | /* We cannot do exclusive creation on a positive dentry */ |
fa3c56bb | 1603 | if (flags & LOOKUP_EXCL) |
49317a7f | 1604 | goto no_open; |
1da177e4 | 1605 | |
0ef97dcf MS |
1606 | /* Let f_op->open() actually open (and revalidate) the file */ |
1607 | ret = 1; | |
536e43d1 | 1608 | |
1da177e4 | 1609 | out: |
1da177e4 | 1610 | return ret; |
535918f1 | 1611 | |
5584c306 | 1612 | no_open: |
0b728e19 | 1613 | return nfs_lookup_revalidate(dentry, flags); |
c0204fd2 TM |
1614 | } |
1615 | ||
1da177e4 LT |
1616 | #endif /* CONFIG_NFSV4 */ |
1617 | ||
1da177e4 LT |
1618 | /* |
1619 | * Code common to create, mkdir, and mknod. | |
1620 | */ | |
1621 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1775fd3e DQ |
1622 | struct nfs_fattr *fattr, |
1623 | struct nfs4_label *label) | |
1da177e4 | 1624 | { |
fab728e1 | 1625 | struct dentry *parent = dget_parent(dentry); |
2b0143b5 | 1626 | struct inode *dir = d_inode(parent); |
1da177e4 LT |
1627 | struct inode *inode; |
1628 | int error = -EACCES; | |
1629 | ||
fab728e1 TM |
1630 | d_drop(dentry); |
1631 | ||
1da177e4 | 1632 | /* We may have been initialized further down */ |
2b0143b5 | 1633 | if (d_really_is_positive(dentry)) |
fab728e1 | 1634 | goto out; |
1da177e4 | 1635 | if (fhandle->size == 0) { |
1775fd3e | 1636 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); |
1da177e4 | 1637 | if (error) |
fab728e1 | 1638 | goto out_error; |
1da177e4 | 1639 | } |
5724ab37 | 1640 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1641 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1642 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
1775fd3e | 1643 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); |
1da177e4 | 1644 | if (error < 0) |
fab728e1 | 1645 | goto out_error; |
1da177e4 | 1646 | } |
1775fd3e | 1647 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); |
03f28e3a TM |
1648 | error = PTR_ERR(inode); |
1649 | if (IS_ERR(inode)) | |
fab728e1 TM |
1650 | goto out_error; |
1651 | d_add(dentry, inode); | |
1652 | out: | |
1653 | dput(parent); | |
1da177e4 | 1654 | return 0; |
fab728e1 TM |
1655 | out_error: |
1656 | nfs_mark_for_revalidate(dir); | |
1657 | dput(parent); | |
1658 | return error; | |
1da177e4 | 1659 | } |
ddda8e0a | 1660 | EXPORT_SYMBOL_GPL(nfs_instantiate); |
1da177e4 LT |
1661 | |
1662 | /* | |
1663 | * Following a failed create operation, we drop the dentry rather | |
1664 | * than retain a negative dentry. This avoids a problem in the event | |
1665 | * that the operation succeeded on the server, but an error in the | |
1666 | * reply path made it appear to have failed. | |
1667 | */ | |
597d9289 | 1668 | int nfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 1669 | umode_t mode, bool excl) |
1da177e4 LT |
1670 | { |
1671 | struct iattr attr; | |
ebfc3b49 | 1672 | int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; |
1da177e4 | 1673 | int error; |
1da177e4 | 1674 | |
1e8968c5 | 1675 | dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", |
6de1472f | 1676 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 LT |
1677 | |
1678 | attr.ia_mode = mode; | |
1679 | attr.ia_valid = ATTR_MODE; | |
1680 | ||
8b0ad3d4 | 1681 | trace_nfs_create_enter(dir, dentry, open_flags); |
8867fe58 | 1682 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); |
8b0ad3d4 | 1683 | trace_nfs_create_exit(dir, dentry, open_flags, error); |
1da177e4 LT |
1684 | if (error != 0) |
1685 | goto out_err; | |
1da177e4 LT |
1686 | return 0; |
1687 | out_err: | |
1da177e4 LT |
1688 | d_drop(dentry); |
1689 | return error; | |
1690 | } | |
ddda8e0a | 1691 | EXPORT_SYMBOL_GPL(nfs_create); |
1da177e4 LT |
1692 | |
1693 | /* | |
1694 | * See comments for nfs_proc_create regarding failed operations. | |
1695 | */ | |
597d9289 | 1696 | int |
1a67aafb | 1697 | nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) |
1da177e4 LT |
1698 | { |
1699 | struct iattr attr; | |
1700 | int status; | |
1701 | ||
1e8968c5 | 1702 | dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", |
6de1472f | 1703 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 | 1704 | |
1da177e4 LT |
1705 | attr.ia_mode = mode; |
1706 | attr.ia_valid = ATTR_MODE; | |
1707 | ||
1ca42382 | 1708 | trace_nfs_mknod_enter(dir, dentry); |
1da177e4 | 1709 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1ca42382 | 1710 | trace_nfs_mknod_exit(dir, dentry, status); |
1da177e4 LT |
1711 | if (status != 0) |
1712 | goto out_err; | |
1da177e4 LT |
1713 | return 0; |
1714 | out_err: | |
1da177e4 LT |
1715 | d_drop(dentry); |
1716 | return status; | |
1717 | } | |
ddda8e0a | 1718 | EXPORT_SYMBOL_GPL(nfs_mknod); |
1da177e4 LT |
1719 | |
1720 | /* | |
1721 | * See comments for nfs_proc_create regarding failed operations. | |
1722 | */ | |
597d9289 | 1723 | int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
1724 | { |
1725 | struct iattr attr; | |
1726 | int error; | |
1727 | ||
1e8968c5 | 1728 | dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", |
6de1472f | 1729 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 LT |
1730 | |
1731 | attr.ia_valid = ATTR_MODE; | |
1732 | attr.ia_mode = mode | S_IFDIR; | |
1733 | ||
1ca42382 | 1734 | trace_nfs_mkdir_enter(dir, dentry); |
1da177e4 | 1735 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1ca42382 | 1736 | trace_nfs_mkdir_exit(dir, dentry, error); |
1da177e4 LT |
1737 | if (error != 0) |
1738 | goto out_err; | |
1da177e4 LT |
1739 | return 0; |
1740 | out_err: | |
1741 | d_drop(dentry); | |
1da177e4 LT |
1742 | return error; |
1743 | } | |
ddda8e0a | 1744 | EXPORT_SYMBOL_GPL(nfs_mkdir); |
1da177e4 | 1745 | |
d45b9d8b TM |
1746 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1747 | { | |
dc3f4198 | 1748 | if (simple_positive(dentry)) |
d45b9d8b TM |
1749 | d_delete(dentry); |
1750 | } | |
1751 | ||
597d9289 | 1752 | int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
1753 | { |
1754 | int error; | |
1755 | ||
1e8968c5 | 1756 | dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", |
6de1472f | 1757 | dir->i_sb->s_id, dir->i_ino, dentry); |
1da177e4 | 1758 | |
1ca42382 | 1759 | trace_nfs_rmdir_enter(dir, dentry); |
2b0143b5 | 1760 | if (d_really_is_positive(dentry)) { |
884be175 | 1761 | down_write(&NFS_I(d_inode(dentry))->rmdir_sem); |
ba6c0592 TM |
1762 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1763 | /* Ensure the VFS deletes this inode */ | |
1764 | switch (error) { | |
1765 | case 0: | |
2b0143b5 | 1766 | clear_nlink(d_inode(dentry)); |
ba6c0592 TM |
1767 | break; |
1768 | case -ENOENT: | |
1769 | nfs_dentry_handle_enoent(dentry); | |
1770 | } | |
884be175 | 1771 | up_write(&NFS_I(d_inode(dentry))->rmdir_sem); |
ba6c0592 TM |
1772 | } else |
1773 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1ca42382 | 1774 | trace_nfs_rmdir_exit(dir, dentry, error); |
1da177e4 LT |
1775 | |
1776 | return error; | |
1777 | } | |
ddda8e0a | 1778 | EXPORT_SYMBOL_GPL(nfs_rmdir); |
1da177e4 | 1779 | |
1da177e4 LT |
1780 | /* |
1781 | * Remove a file after making sure there are no pending writes, | |
1782 | * and after checking that the file has only one user. | |
1783 | * | |
1784 | * We invalidate the attribute cache and free the inode prior to the operation | |
1785 | * to avoid possible races if the server reuses the inode. | |
1786 | */ | |
1787 | static int nfs_safe_remove(struct dentry *dentry) | |
1788 | { | |
2b0143b5 DH |
1789 | struct inode *dir = d_inode(dentry->d_parent); |
1790 | struct inode *inode = d_inode(dentry); | |
1da177e4 LT |
1791 | int error = -EBUSY; |
1792 | ||
6de1472f | 1793 | dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); |
1da177e4 LT |
1794 | |
1795 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1796 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1797 | error = 0; | |
1798 | goto out; | |
1799 | } | |
1800 | ||
1ca42382 | 1801 | trace_nfs_remove_enter(dir, dentry); |
1da177e4 | 1802 | if (inode != NULL) { |
912678db | 1803 | error = NFS_PROTO(dir)->remove(dir, dentry); |
1da177e4 | 1804 | if (error == 0) |
1b83d707 | 1805 | nfs_drop_nlink(inode); |
1da177e4 | 1806 | } else |
912678db | 1807 | error = NFS_PROTO(dir)->remove(dir, dentry); |
d45b9d8b TM |
1808 | if (error == -ENOENT) |
1809 | nfs_dentry_handle_enoent(dentry); | |
1ca42382 | 1810 | trace_nfs_remove_exit(dir, dentry, error); |
1da177e4 LT |
1811 | out: |
1812 | return error; | |
1813 | } | |
1814 | ||
1815 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1816 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1817 | * | |
1818 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1819 | */ | |
597d9289 | 1820 | int nfs_unlink(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
1821 | { |
1822 | int error; | |
1823 | int need_rehash = 0; | |
1824 | ||
1e8968c5 | 1825 | dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, |
6de1472f | 1826 | dir->i_ino, dentry); |
1da177e4 | 1827 | |
1ca42382 | 1828 | trace_nfs_unlink_enter(dir, dentry); |
1da177e4 | 1829 | spin_lock(&dentry->d_lock); |
84d08fa8 | 1830 | if (d_count(dentry) > 1) { |
1da177e4 | 1831 | spin_unlock(&dentry->d_lock); |
ccfeb506 | 1832 | /* Start asynchronous writeout of the inode */ |
2b0143b5 | 1833 | write_inode_now(d_inode(dentry), 0); |
1da177e4 | 1834 | error = nfs_sillyrename(dir, dentry); |
1ca42382 | 1835 | goto out; |
1da177e4 LT |
1836 | } |
1837 | if (!d_unhashed(dentry)) { | |
1838 | __d_drop(dentry); | |
1839 | need_rehash = 1; | |
1840 | } | |
1841 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1842 | error = nfs_safe_remove(dentry); |
d45b9d8b | 1843 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1844 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1845 | } else if (need_rehash) | |
1846 | d_rehash(dentry); | |
1ca42382 TM |
1847 | out: |
1848 | trace_nfs_unlink_exit(dir, dentry, error); | |
1da177e4 LT |
1849 | return error; |
1850 | } | |
ddda8e0a | 1851 | EXPORT_SYMBOL_GPL(nfs_unlink); |
1da177e4 | 1852 | |
873101b3 CL |
1853 | /* |
1854 | * To create a symbolic link, most file systems instantiate a new inode, | |
1855 | * add a page to it containing the path, then write it out to the disk | |
1856 | * using prepare_write/commit_write. | |
1857 | * | |
1858 | * Unfortunately the NFS client can't create the in-core inode first | |
1859 | * because it needs a file handle to create an in-core inode (see | |
1860 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1861 | * symlink request has completed on the server. | |
1862 | * | |
1863 | * So instead we allocate a raw page, copy the symname into it, then do | |
1864 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1865 | * now have a new file handle and can instantiate an in-core NFS inode | |
1866 | * and move the raw page into its mapping. | |
1867 | */ | |
597d9289 | 1868 | int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) |
1da177e4 | 1869 | { |
873101b3 CL |
1870 | struct page *page; |
1871 | char *kaddr; | |
1da177e4 | 1872 | struct iattr attr; |
873101b3 | 1873 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1874 | int error; |
1875 | ||
1e8968c5 | 1876 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, |
6de1472f | 1877 | dir->i_ino, dentry, symname); |
1da177e4 | 1878 | |
873101b3 CL |
1879 | if (pathlen > PAGE_SIZE) |
1880 | return -ENAMETOOLONG; | |
1da177e4 | 1881 | |
873101b3 CL |
1882 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1883 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1884 | |
e8ecde25 | 1885 | page = alloc_page(GFP_USER); |
76566991 | 1886 | if (!page) |
873101b3 | 1887 | return -ENOMEM; |
873101b3 | 1888 | |
e8ecde25 | 1889 | kaddr = page_address(page); |
873101b3 CL |
1890 | memcpy(kaddr, symname, pathlen); |
1891 | if (pathlen < PAGE_SIZE) | |
1892 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
873101b3 | 1893 | |
1ca42382 | 1894 | trace_nfs_symlink_enter(dir, dentry); |
94a6d753 | 1895 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
1ca42382 | 1896 | trace_nfs_symlink_exit(dir, dentry, error); |
873101b3 | 1897 | if (error != 0) { |
1e8968c5 | 1898 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", |
873101b3 | 1899 | dir->i_sb->s_id, dir->i_ino, |
6de1472f | 1900 | dentry, symname, error); |
1da177e4 | 1901 | d_drop(dentry); |
873101b3 | 1902 | __free_page(page); |
873101b3 CL |
1903 | return error; |
1904 | } | |
1905 | ||
1906 | /* | |
1907 | * No big deal if we can't add this page to the page cache here. | |
1908 | * READLINK will get the missing page from the server if needed. | |
1909 | */ | |
2b0143b5 | 1910 | if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, |
873101b3 | 1911 | GFP_KERNEL)) { |
873101b3 CL |
1912 | SetPageUptodate(page); |
1913 | unlock_page(page); | |
a0b54add RA |
1914 | /* |
1915 | * add_to_page_cache_lru() grabs an extra page refcount. | |
1916 | * Drop it here to avoid leaking this page later. | |
1917 | */ | |
09cbfeaf | 1918 | put_page(page); |
873101b3 CL |
1919 | } else |
1920 | __free_page(page); | |
1921 | ||
873101b3 | 1922 | return 0; |
1da177e4 | 1923 | } |
ddda8e0a | 1924 | EXPORT_SYMBOL_GPL(nfs_symlink); |
1da177e4 | 1925 | |
597d9289 | 1926 | int |
1da177e4 LT |
1927 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) |
1928 | { | |
2b0143b5 | 1929 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
1930 | int error; |
1931 | ||
6de1472f AV |
1932 | dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", |
1933 | old_dentry, dentry); | |
1da177e4 | 1934 | |
1fd1085b | 1935 | trace_nfs_link_enter(inode, dir, dentry); |
9697d234 | 1936 | d_drop(dentry); |
1da177e4 | 1937 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1938 | if (error == 0) { |
7de9c6ee | 1939 | ihold(inode); |
9697d234 | 1940 | d_add(dentry, inode); |
cf809556 | 1941 | } |
1fd1085b | 1942 | trace_nfs_link_exit(inode, dir, dentry, error); |
1da177e4 LT |
1943 | return error; |
1944 | } | |
ddda8e0a | 1945 | EXPORT_SYMBOL_GPL(nfs_link); |
1da177e4 LT |
1946 | |
1947 | /* | |
1948 | * RENAME | |
1949 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1950 | * different file handle for the same inode after a rename (e.g. when | |
1951 | * moving to a different directory). A fail-safe method to do so would | |
1952 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1953 | * rename the old file using the sillyrename stuff. This way, the original | |
1954 | * file in old_dir will go away when the last process iput()s the inode. | |
1955 | * | |
1956 | * FIXED. | |
1957 | * | |
1958 | * It actually works quite well. One needs to have the possibility for | |
1959 | * at least one ".nfs..." file in each directory the file ever gets | |
1960 | * moved or linked to which happens automagically with the new | |
1961 | * implementation that only depends on the dcache stuff instead of | |
1962 | * using the inode layer | |
1963 | * | |
1964 | * Unfortunately, things are a little more complicated than indicated | |
1965 | * above. For a cross-directory move, we want to make sure we can get | |
1966 | * rid of the old inode after the operation. This means there must be | |
1967 | * no pending writes (if it's a file), and the use count must be 1. | |
1968 | * If these conditions are met, we can drop the dentries before doing | |
1969 | * the rename. | |
1970 | */ | |
597d9289 | 1971 | int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
1cd66c93 MS |
1972 | struct inode *new_dir, struct dentry *new_dentry, |
1973 | unsigned int flags) | |
1da177e4 | 1974 | { |
2b0143b5 DH |
1975 | struct inode *old_inode = d_inode(old_dentry); |
1976 | struct inode *new_inode = d_inode(new_dentry); | |
d9f29500 | 1977 | struct dentry *dentry = NULL, *rehash = NULL; |
80a491fd | 1978 | struct rpc_task *task; |
1da177e4 LT |
1979 | int error = -EBUSY; |
1980 | ||
1cd66c93 MS |
1981 | if (flags) |
1982 | return -EINVAL; | |
1983 | ||
6de1472f AV |
1984 | dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", |
1985 | old_dentry, new_dentry, | |
84d08fa8 | 1986 | d_count(new_dentry)); |
1da177e4 | 1987 | |
70ded201 | 1988 | trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); |
1da177e4 | 1989 | /* |
28f79a1a MS |
1990 | * For non-directories, check whether the target is busy and if so, |
1991 | * make a copy of the dentry and then do a silly-rename. If the | |
1992 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1993 | * the new target. | |
1da177e4 | 1994 | */ |
27226104 MS |
1995 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1996 | /* | |
1997 | * To prevent any new references to the target during the | |
1998 | * rename, we unhash the dentry in advance. | |
1999 | */ | |
d9f29500 | 2000 | if (!d_unhashed(new_dentry)) { |
27226104 | 2001 | d_drop(new_dentry); |
d9f29500 BC |
2002 | rehash = new_dentry; |
2003 | } | |
1da177e4 | 2004 | |
84d08fa8 | 2005 | if (d_count(new_dentry) > 2) { |
27226104 MS |
2006 | int err; |
2007 | ||
2008 | /* copy the target dentry's name */ | |
2009 | dentry = d_alloc(new_dentry->d_parent, | |
2010 | &new_dentry->d_name); | |
2011 | if (!dentry) | |
2012 | goto out; | |
2013 | ||
2014 | /* silly-rename the existing target ... */ | |
2015 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 2016 | if (err) |
27226104 | 2017 | goto out; |
24e93025 MS |
2018 | |
2019 | new_dentry = dentry; | |
d9f29500 | 2020 | rehash = NULL; |
24e93025 | 2021 | new_inode = NULL; |
27226104 | 2022 | } |
b1e4adf4 | 2023 | } |
1da177e4 | 2024 | |
d9f29500 | 2025 | task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); |
80a491fd JL |
2026 | if (IS_ERR(task)) { |
2027 | error = PTR_ERR(task); | |
2028 | goto out; | |
2029 | } | |
2030 | ||
2031 | error = rpc_wait_for_completion_task(task); | |
818a8dbe BC |
2032 | if (error != 0) { |
2033 | ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; | |
2034 | /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ | |
2035 | smp_wmb(); | |
2036 | } else | |
80a491fd JL |
2037 | error = task->tk_status; |
2038 | rpc_put_task(task); | |
d9f29500 | 2039 | nfs_mark_for_revalidate(old_inode); |
1da177e4 | 2040 | out: |
d9f29500 BC |
2041 | if (rehash) |
2042 | d_rehash(rehash); | |
70ded201 TM |
2043 | trace_nfs_rename_exit(old_dir, old_dentry, |
2044 | new_dir, new_dentry, error); | |
d9f29500 BC |
2045 | if (!error) { |
2046 | if (new_inode != NULL) | |
2047 | nfs_drop_nlink(new_inode); | |
2048 | /* | |
2049 | * The d_move() should be here instead of in an async RPC completion | |
2050 | * handler because we need the proper locks to move the dentry. If | |
2051 | * we're interrupted by a signal, the async RPC completion handler | |
2052 | * should mark the directories for revalidation. | |
2053 | */ | |
2054 | d_move(old_dentry, new_dentry); | |
d803224c | 2055 | nfs_set_verifier(old_dentry, |
d9f29500 BC |
2056 | nfs_save_change_attribute(new_dir)); |
2057 | } else if (error == -ENOENT) | |
2058 | nfs_dentry_handle_enoent(old_dentry); | |
2059 | ||
1da177e4 LT |
2060 | /* new dentry created? */ |
2061 | if (dentry) | |
2062 | dput(dentry); | |
1da177e4 LT |
2063 | return error; |
2064 | } | |
ddda8e0a | 2065 | EXPORT_SYMBOL_GPL(nfs_rename); |
1da177e4 | 2066 | |
cfcea3e8 TM |
2067 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
2068 | static LIST_HEAD(nfs_access_lru_list); | |
2069 | static atomic_long_t nfs_access_nr_entries; | |
2070 | ||
3a505845 TM |
2071 | static unsigned long nfs_access_max_cachesize = ULONG_MAX; |
2072 | module_param(nfs_access_max_cachesize, ulong, 0644); | |
2073 | MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); | |
2074 | ||
1c3c07e9 TM |
2075 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
2076 | { | |
2077 | put_rpccred(entry->cred); | |
f682a398 | 2078 | kfree_rcu(entry, rcu_head); |
4e857c58 | 2079 | smp_mb__before_atomic(); |
cfcea3e8 | 2080 | atomic_long_dec(&nfs_access_nr_entries); |
4e857c58 | 2081 | smp_mb__after_atomic(); |
1c3c07e9 TM |
2082 | } |
2083 | ||
1a81bb8a TM |
2084 | static void nfs_access_free_list(struct list_head *head) |
2085 | { | |
2086 | struct nfs_access_entry *cache; | |
2087 | ||
2088 | while (!list_empty(head)) { | |
2089 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
2090 | list_del(&cache->lru); | |
2091 | nfs_access_free_entry(cache); | |
2092 | } | |
2093 | } | |
2094 | ||
3a505845 TM |
2095 | static unsigned long |
2096 | nfs_do_access_cache_scan(unsigned int nr_to_scan) | |
979df72e TM |
2097 | { |
2098 | LIST_HEAD(head); | |
aa510da5 | 2099 | struct nfs_inode *nfsi, *next; |
979df72e | 2100 | struct nfs_access_entry *cache; |
1ab6c499 | 2101 | long freed = 0; |
979df72e | 2102 | |
a50f7951 | 2103 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 2104 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
2105 | struct inode *inode; |
2106 | ||
2107 | if (nr_to_scan-- == 0) | |
2108 | break; | |
9c7e7e23 | 2109 | inode = &nfsi->vfs_inode; |
979df72e TM |
2110 | spin_lock(&inode->i_lock); |
2111 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
2112 | goto remove_lru_entry; | |
2113 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
2114 | struct nfs_access_entry, lru); | |
2115 | list_move(&cache->lru, &head); | |
2116 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1ab6c499 | 2117 | freed++; |
979df72e TM |
2118 | if (!list_empty(&nfsi->access_cache_entry_lru)) |
2119 | list_move_tail(&nfsi->access_cache_inode_lru, | |
2120 | &nfs_access_lru_list); | |
2121 | else { | |
2122 | remove_lru_entry: | |
2123 | list_del_init(&nfsi->access_cache_inode_lru); | |
4e857c58 | 2124 | smp_mb__before_atomic(); |
979df72e | 2125 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
4e857c58 | 2126 | smp_mb__after_atomic(); |
979df72e | 2127 | } |
59844a9b | 2128 | spin_unlock(&inode->i_lock); |
979df72e TM |
2129 | } |
2130 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2131 | nfs_access_free_list(&head); |
1ab6c499 DC |
2132 | return freed; |
2133 | } | |
2134 | ||
3a505845 TM |
2135 | unsigned long |
2136 | nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
2137 | { | |
2138 | int nr_to_scan = sc->nr_to_scan; | |
2139 | gfp_t gfp_mask = sc->gfp_mask; | |
2140 | ||
2141 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) | |
2142 | return SHRINK_STOP; | |
2143 | return nfs_do_access_cache_scan(nr_to_scan); | |
2144 | } | |
2145 | ||
2146 | ||
1ab6c499 DC |
2147 | unsigned long |
2148 | nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
2149 | { | |
55f841ce | 2150 | return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); |
979df72e TM |
2151 | } |
2152 | ||
3a505845 TM |
2153 | static void |
2154 | nfs_access_cache_enforce_limit(void) | |
2155 | { | |
2156 | long nr_entries = atomic_long_read(&nfs_access_nr_entries); | |
2157 | unsigned long diff; | |
2158 | unsigned int nr_to_scan; | |
2159 | ||
2160 | if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) | |
2161 | return; | |
2162 | nr_to_scan = 100; | |
2163 | diff = nr_entries - nfs_access_max_cachesize; | |
2164 | if (diff < nr_to_scan) | |
2165 | nr_to_scan = diff; | |
2166 | nfs_do_access_cache_scan(nr_to_scan); | |
2167 | } | |
2168 | ||
1a81bb8a | 2169 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2170 | { |
1c3c07e9 | 2171 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2172 | struct rb_node *n; |
1c3c07e9 TM |
2173 | struct nfs_access_entry *entry; |
2174 | ||
2175 | /* Unhook entries from the cache */ | |
2176 | while ((n = rb_first(root_node)) != NULL) { | |
2177 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2178 | rb_erase(n, root_node); | |
1a81bb8a | 2179 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2180 | } |
2181 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2182 | } |
2183 | ||
1c3c07e9 | 2184 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2185 | { |
1a81bb8a TM |
2186 | LIST_HEAD(head); |
2187 | ||
2188 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2189 | return; | |
cfcea3e8 | 2190 | /* Remove from global LRU init */ |
1a81bb8a TM |
2191 | spin_lock(&nfs_access_lru_lock); |
2192 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2193 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2194 | |
1c3c07e9 | 2195 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2196 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2197 | spin_unlock(&inode->i_lock); | |
2198 | spin_unlock(&nfs_access_lru_lock); | |
2199 | nfs_access_free_list(&head); | |
1c3c07e9 | 2200 | } |
1c606fb7 | 2201 | EXPORT_SYMBOL_GPL(nfs_access_zap_cache); |
1da177e4 | 2202 | |
1c3c07e9 TM |
2203 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2204 | { | |
2205 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2206 | struct nfs_access_entry *entry; | |
2207 | ||
2208 | while (n != NULL) { | |
2209 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2210 | ||
2211 | if (cred < entry->cred) | |
2212 | n = n->rb_left; | |
2213 | else if (cred > entry->cred) | |
2214 | n = n->rb_right; | |
2215 | else | |
2216 | return entry; | |
1da177e4 | 2217 | } |
1c3c07e9 TM |
2218 | return NULL; |
2219 | } | |
2220 | ||
57b69181 | 2221 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block) |
1c3c07e9 TM |
2222 | { |
2223 | struct nfs_inode *nfsi = NFS_I(inode); | |
2224 | struct nfs_access_entry *cache; | |
57b69181 TM |
2225 | bool retry = true; |
2226 | int err; | |
1c3c07e9 | 2227 | |
dc59250c | 2228 | spin_lock(&inode->i_lock); |
57b69181 TM |
2229 | for(;;) { |
2230 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
2231 | goto out_zap; | |
2232 | cache = nfs_access_search_rbtree(inode, cred); | |
2233 | err = -ENOENT; | |
2234 | if (cache == NULL) | |
2235 | goto out; | |
2236 | /* Found an entry, is our attribute cache valid? */ | |
21c3ba7e | 2237 | if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) |
57b69181 TM |
2238 | break; |
2239 | err = -ECHILD; | |
2240 | if (!may_block) | |
2241 | goto out; | |
2242 | if (!retry) | |
2243 | goto out_zap; | |
2244 | spin_unlock(&inode->i_lock); | |
2245 | err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2246 | if (err) | |
2247 | return err; | |
2248 | spin_lock(&inode->i_lock); | |
2249 | retry = false; | |
2250 | } | |
1c3c07e9 TM |
2251 | res->cred = cache->cred; |
2252 | res->mask = cache->mask; | |
cfcea3e8 | 2253 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2254 | err = 0; |
2255 | out: | |
2256 | spin_unlock(&inode->i_lock); | |
2257 | return err; | |
1c3c07e9 | 2258 | out_zap: |
1a81bb8a TM |
2259 | spin_unlock(&inode->i_lock); |
2260 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2261 | return -ENOENT; |
2262 | } | |
2263 | ||
f682a398 N |
2264 | static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
2265 | { | |
2266 | /* Only check the most recently returned cache entry, | |
2267 | * but do it without locking. | |
2268 | */ | |
2269 | struct nfs_inode *nfsi = NFS_I(inode); | |
2270 | struct nfs_access_entry *cache; | |
2271 | int err = -ECHILD; | |
2272 | struct list_head *lh; | |
2273 | ||
2274 | rcu_read_lock(); | |
2275 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
2276 | goto out; | |
2277 | lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); | |
2278 | cache = list_entry(lh, struct nfs_access_entry, lru); | |
2279 | if (lh == &nfsi->access_cache_entry_lru || | |
2280 | cred != cache->cred) | |
2281 | cache = NULL; | |
2282 | if (cache == NULL) | |
2283 | goto out; | |
21c3ba7e | 2284 | if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) |
f682a398 | 2285 | goto out; |
f682a398 N |
2286 | res->cred = cache->cred; |
2287 | res->mask = cache->mask; | |
21c3ba7e | 2288 | err = 0; |
f682a398 N |
2289 | out: |
2290 | rcu_read_unlock(); | |
2291 | return err; | |
2292 | } | |
2293 | ||
1c3c07e9 TM |
2294 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) |
2295 | { | |
cfcea3e8 TM |
2296 | struct nfs_inode *nfsi = NFS_I(inode); |
2297 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2298 | struct rb_node **p = &root_node->rb_node; |
2299 | struct rb_node *parent = NULL; | |
2300 | struct nfs_access_entry *entry; | |
2301 | ||
2302 | spin_lock(&inode->i_lock); | |
2303 | while (*p != NULL) { | |
2304 | parent = *p; | |
2305 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2306 | ||
2307 | if (set->cred < entry->cred) | |
2308 | p = &parent->rb_left; | |
2309 | else if (set->cred > entry->cred) | |
2310 | p = &parent->rb_right; | |
2311 | else | |
2312 | goto found; | |
2313 | } | |
2314 | rb_link_node(&set->rb_node, parent, p); | |
2315 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2316 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2317 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2318 | return; |
2319 | found: | |
2320 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2321 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2322 | list_del(&entry->lru); | |
1c3c07e9 TM |
2323 | spin_unlock(&inode->i_lock); |
2324 | nfs_access_free_entry(entry); | |
2325 | } | |
2326 | ||
6168f62c | 2327 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2328 | { |
2329 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2330 | if (cache == NULL) | |
2331 | return; | |
2332 | RB_CLEAR_NODE(&cache->rb_node); | |
1c3c07e9 | 2333 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2334 | cache->mask = set->mask; |
1c3c07e9 | 2335 | |
f682a398 N |
2336 | /* The above field assignments must be visible |
2337 | * before this item appears on the lru. We cannot easily | |
2338 | * use rcu_assign_pointer, so just force the memory barrier. | |
2339 | */ | |
2340 | smp_wmb(); | |
1c3c07e9 | 2341 | nfs_access_add_rbtree(inode, cache); |
cfcea3e8 TM |
2342 | |
2343 | /* Update accounting */ | |
4e857c58 | 2344 | smp_mb__before_atomic(); |
cfcea3e8 | 2345 | atomic_long_inc(&nfs_access_nr_entries); |
4e857c58 | 2346 | smp_mb__after_atomic(); |
cfcea3e8 TM |
2347 | |
2348 | /* Add inode to global LRU list */ | |
1a81bb8a | 2349 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2350 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2351 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2352 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2353 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2354 | spin_unlock(&nfs_access_lru_lock); |
2355 | } | |
3a505845 | 2356 | nfs_access_cache_enforce_limit(); |
1da177e4 | 2357 | } |
6168f62c WAA |
2358 | EXPORT_SYMBOL_GPL(nfs_access_add_cache); |
2359 | ||
3c181827 AS |
2360 | #define NFS_MAY_READ (NFS_ACCESS_READ) |
2361 | #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ | |
2362 | NFS_ACCESS_EXTEND | \ | |
2363 | NFS_ACCESS_DELETE) | |
2364 | #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ | |
2365 | NFS_ACCESS_EXTEND) | |
ecbb903c | 2366 | #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE |
3c181827 AS |
2367 | #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) |
2368 | #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) | |
15d4b73a | 2369 | static int |
ecbb903c | 2370 | nfs_access_calc_mask(u32 access_result, umode_t umode) |
15d4b73a TM |
2371 | { |
2372 | int mask = 0; | |
2373 | ||
2374 | if (access_result & NFS_MAY_READ) | |
2375 | mask |= MAY_READ; | |
ecbb903c TM |
2376 | if (S_ISDIR(umode)) { |
2377 | if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) | |
2378 | mask |= MAY_WRITE; | |
2379 | if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) | |
2380 | mask |= MAY_EXEC; | |
2381 | } else if (S_ISREG(umode)) { | |
2382 | if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) | |
2383 | mask |= MAY_WRITE; | |
2384 | if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) | |
2385 | mask |= MAY_EXEC; | |
2386 | } else if (access_result & NFS_MAY_WRITE) | |
2387 | mask |= MAY_WRITE; | |
15d4b73a TM |
2388 | return mask; |
2389 | } | |
2390 | ||
6168f62c WAA |
2391 | void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) |
2392 | { | |
bd8b2441 | 2393 | entry->mask = access_result; |
6168f62c WAA |
2394 | } |
2395 | EXPORT_SYMBOL_GPL(nfs_access_set_mask); | |
1da177e4 LT |
2396 | |
2397 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2398 | { | |
2399 | struct nfs_access_entry cache; | |
57b69181 | 2400 | bool may_block = (mask & MAY_NOT_BLOCK) == 0; |
bd8b2441 | 2401 | int cache_mask; |
1da177e4 LT |
2402 | int status; |
2403 | ||
f4ce1299 TM |
2404 | trace_nfs_access_enter(inode); |
2405 | ||
f682a398 N |
2406 | status = nfs_access_get_cached_rcu(inode, cred, &cache); |
2407 | if (status != 0) | |
57b69181 | 2408 | status = nfs_access_get_cached(inode, cred, &cache, may_block); |
1da177e4 | 2409 | if (status == 0) |
f4ce1299 | 2410 | goto out_cached; |
1da177e4 | 2411 | |
f3324a2a | 2412 | status = -ECHILD; |
57b69181 | 2413 | if (!may_block) |
f3324a2a N |
2414 | goto out; |
2415 | ||
1750d929 AS |
2416 | /* |
2417 | * Determine which access bits we want to ask for... | |
2418 | */ | |
2419 | cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; | |
2420 | if (S_ISDIR(inode->i_mode)) | |
2421 | cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; | |
2422 | else | |
2423 | cache.mask |= NFS_ACCESS_EXECUTE; | |
1da177e4 | 2424 | cache.cred = cred; |
1da177e4 | 2425 | status = NFS_PROTO(inode)->access(inode, &cache); |
a71ee337 SJ |
2426 | if (status != 0) { |
2427 | if (status == -ESTALE) { | |
2428 | nfs_zap_caches(inode); | |
2429 | if (!S_ISDIR(inode->i_mode)) | |
2430 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2431 | } | |
f4ce1299 | 2432 | goto out; |
a71ee337 | 2433 | } |
1da177e4 | 2434 | nfs_access_add_cache(inode, &cache); |
f4ce1299 | 2435 | out_cached: |
ecbb903c | 2436 | cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); |
bd8b2441 | 2437 | if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) |
f4ce1299 | 2438 | status = -EACCES; |
1da177e4 | 2439 | out: |
f4ce1299 TM |
2440 | trace_nfs_access_exit(inode, status); |
2441 | return status; | |
1da177e4 LT |
2442 | } |
2443 | ||
af22f94a TM |
2444 | static int nfs_open_permission_mask(int openflags) |
2445 | { | |
2446 | int mask = 0; | |
2447 | ||
f8d9a897 WAA |
2448 | if (openflags & __FMODE_EXEC) { |
2449 | /* ONLY check exec rights */ | |
2450 | mask = MAY_EXEC; | |
2451 | } else { | |
2452 | if ((openflags & O_ACCMODE) != O_WRONLY) | |
2453 | mask |= MAY_READ; | |
2454 | if ((openflags & O_ACCMODE) != O_RDONLY) | |
2455 | mask |= MAY_WRITE; | |
2456 | } | |
2457 | ||
af22f94a TM |
2458 | return mask; |
2459 | } | |
2460 | ||
2461 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2462 | { | |
2463 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2464 | } | |
89d77c8f | 2465 | EXPORT_SYMBOL_GPL(nfs_may_open); |
af22f94a | 2466 | |
5c5fc09a TM |
2467 | static int nfs_execute_ok(struct inode *inode, int mask) |
2468 | { | |
2469 | struct nfs_server *server = NFS_SERVER(inode); | |
21c3ba7e | 2470 | int ret = 0; |
5c5fc09a | 2471 | |
21c3ba7e TM |
2472 | if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) { |
2473 | if (mask & MAY_NOT_BLOCK) | |
2474 | return -ECHILD; | |
2475 | ret = __nfs_revalidate_inode(server, inode); | |
2476 | } | |
5c5fc09a TM |
2477 | if (ret == 0 && !execute_ok(inode)) |
2478 | ret = -EACCES; | |
2479 | return ret; | |
2480 | } | |
2481 | ||
10556cb2 | 2482 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2483 | { |
2484 | struct rpc_cred *cred; | |
2485 | int res = 0; | |
2486 | ||
91d5b470 CL |
2487 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2488 | ||
e6305c43 | 2489 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2490 | goto out; |
2491 | /* Is this sys_access() ? */ | |
9cfcac81 | 2492 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2493 | goto force_lookup; |
2494 | ||
2495 | switch (inode->i_mode & S_IFMT) { | |
2496 | case S_IFLNK: | |
2497 | goto out; | |
2498 | case S_IFREG: | |
762674f8 TM |
2499 | if ((mask & MAY_OPEN) && |
2500 | nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) | |
2501 | return 0; | |
1da177e4 LT |
2502 | break; |
2503 | case S_IFDIR: | |
2504 | /* | |
2505 | * Optimize away all write operations, since the server | |
2506 | * will check permissions when we perform the op. | |
2507 | */ | |
2508 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2509 | goto out; | |
2510 | } | |
2511 | ||
2512 | force_lookup: | |
1da177e4 LT |
2513 | if (!NFS_PROTO(inode)->access) |
2514 | goto out_notsup; | |
2515 | ||
f3324a2a N |
2516 | /* Always try fast lookups first */ |
2517 | rcu_read_lock(); | |
2518 | cred = rpc_lookup_cred_nonblock(); | |
2519 | if (!IS_ERR(cred)) | |
2520 | res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); | |
2521 | else | |
1da177e4 | 2522 | res = PTR_ERR(cred); |
f3324a2a N |
2523 | rcu_read_unlock(); |
2524 | if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { | |
2525 | /* Fast lookup failed, try the slow way */ | |
2526 | cred = rpc_lookup_cred(); | |
2527 | if (!IS_ERR(cred)) { | |
2528 | res = nfs_do_access(inode, cred, mask); | |
2529 | put_rpccred(cred); | |
2530 | } else | |
2531 | res = PTR_ERR(cred); | |
2532 | } | |
1da177e4 | 2533 | out: |
5c5fc09a TM |
2534 | if (!res && (mask & MAY_EXEC)) |
2535 | res = nfs_execute_ok(inode, mask); | |
f696a365 | 2536 | |
1e8968c5 | 2537 | dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", |
1e7cb3dc | 2538 | inode->i_sb->s_id, inode->i_ino, mask, res); |
1da177e4 LT |
2539 | return res; |
2540 | out_notsup: | |
d51ac1a8 N |
2541 | if (mask & MAY_NOT_BLOCK) |
2542 | return -ECHILD; | |
2543 | ||
1da177e4 LT |
2544 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); |
2545 | if (res == 0) | |
2830ba7f | 2546 | res = generic_permission(inode, mask); |
1e7cb3dc | 2547 | goto out; |
1da177e4 | 2548 | } |
ddda8e0a | 2549 | EXPORT_SYMBOL_GPL(nfs_permission); |
1da177e4 LT |
2550 | |
2551 | /* | |
2552 | * Local variables: | |
2553 | * version-control: t | |
2554 | * kept-new-versions: 5 | |
2555 | * End: | |
2556 | */ |