]>
Commit | Line | Data |
---|---|---|
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 | ||
20 | #include <linux/module.h> | |
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> | |
33 | #include <linux/pagevec.h> | |
34 | #include <linux/namei.h> | |
35 | #include <linux/mount.h> | |
36 | #include <linux/swap.h> | |
37 | #include <linux/sched.h> | |
38 | #include <linux/kmemleak.h> | |
39 | #include <linux/xattr.h> | |
40 | ||
41 | #include "delegation.h" | |
42 | #include "iostat.h" | |
43 | #include "internal.h" | |
44 | #include "fscache.h" | |
45 | ||
46 | #include "nfstrace.h" | |
47 | ||
48 | /* #define NFS_DEBUG_VERBOSE 1 */ | |
49 | ||
50 | static int nfs_opendir(struct inode *, struct file *); | |
51 | static int nfs_closedir(struct inode *, struct file *); | |
52 | static int nfs_readdir(struct file *, struct dir_context *); | |
53 | static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); | |
54 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); | |
55 | static void nfs_readdir_clear_array(struct page*); | |
56 | ||
57 | const struct file_operations nfs_dir_operations = { | |
58 | .llseek = nfs_llseek_dir, | |
59 | .read = generic_read_dir, | |
60 | .iterate_shared = nfs_readdir, | |
61 | .open = nfs_opendir, | |
62 | .release = nfs_closedir, | |
63 | .fsync = nfs_fsync_dir, | |
64 | }; | |
65 | ||
66 | const struct address_space_operations nfs_dir_aops = { | |
67 | .freepage = nfs_readdir_clear_array, | |
68 | }; | |
69 | ||
70 | static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred) | |
71 | { | |
72 | struct nfs_inode *nfsi = NFS_I(dir); | |
73 | struct nfs_open_dir_context *ctx; | |
74 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
75 | if (ctx != NULL) { | |
76 | ctx->duped = 0; | |
77 | ctx->attr_gencount = nfsi->attr_gencount; | |
78 | ctx->dir_cookie = 0; | |
79 | ctx->dup_cookie = 0; | |
80 | ctx->cred = get_rpccred(cred); | |
81 | spin_lock(&dir->i_lock); | |
82 | list_add(&ctx->list, &nfsi->open_files); | |
83 | spin_unlock(&dir->i_lock); | |
84 | return ctx; | |
85 | } | |
86 | return ERR_PTR(-ENOMEM); | |
87 | } | |
88 | ||
89 | static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) | |
90 | { | |
91 | spin_lock(&dir->i_lock); | |
92 | list_del(&ctx->list); | |
93 | spin_unlock(&dir->i_lock); | |
94 | put_rpccred(ctx->cred); | |
95 | kfree(ctx); | |
96 | } | |
97 | ||
98 | /* | |
99 | * Open file | |
100 | */ | |
101 | static int | |
102 | nfs_opendir(struct inode *inode, struct file *filp) | |
103 | { | |
104 | int res = 0; | |
105 | struct nfs_open_dir_context *ctx; | |
106 | struct rpc_cred *cred; | |
107 | ||
108 | dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); | |
109 | ||
110 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
111 | ||
112 | cred = rpc_lookup_cred(); | |
113 | if (IS_ERR(cred)) | |
114 | return PTR_ERR(cred); | |
115 | ctx = alloc_nfs_open_dir_context(inode, cred); | |
116 | if (IS_ERR(ctx)) { | |
117 | res = PTR_ERR(ctx); | |
118 | goto out; | |
119 | } | |
120 | filp->private_data = ctx; | |
121 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { | |
122 | /* This is a mountpoint, so d_revalidate will never | |
123 | * have been called, so we need to refresh the | |
124 | * inode (for close-open consistency) ourselves. | |
125 | */ | |
126 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
127 | } | |
128 | out: | |
129 | put_rpccred(cred); | |
130 | return res; | |
131 | } | |
132 | ||
133 | static int | |
134 | nfs_closedir(struct inode *inode, struct file *filp) | |
135 | { | |
136 | put_nfs_open_dir_context(file_inode(filp), filp->private_data); | |
137 | return 0; | |
138 | } | |
139 | ||
140 | struct nfs_cache_array_entry { | |
141 | u64 cookie; | |
142 | u64 ino; | |
143 | struct qstr string; | |
144 | unsigned char d_type; | |
145 | }; | |
146 | ||
147 | struct nfs_cache_array { | |
148 | atomic_t refcount; | |
149 | int size; | |
150 | int eof_index; | |
151 | u64 last_cookie; | |
152 | struct nfs_cache_array_entry array[0]; | |
153 | }; | |
154 | ||
155 | typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); | |
156 | typedef struct { | |
157 | struct file *file; | |
158 | struct page *page; | |
159 | struct dir_context *ctx; | |
160 | unsigned long page_index; | |
161 | u64 *dir_cookie; | |
162 | u64 last_cookie; | |
163 | loff_t current_index; | |
164 | decode_dirent_t decode; | |
165 | ||
166 | unsigned long timestamp; | |
167 | unsigned long gencount; | |
168 | unsigned int cache_entry_index; | |
169 | unsigned int plus:1; | |
170 | unsigned int eof:1; | |
171 | } nfs_readdir_descriptor_t; | |
172 | ||
173 | /* | |
174 | * The caller is responsible for calling nfs_readdir_release_array(page) | |
175 | */ | |
176 | static | |
177 | struct nfs_cache_array *nfs_readdir_get_array(struct page *page) | |
178 | { | |
179 | void *ptr; | |
180 | if (page == NULL) | |
181 | return ERR_PTR(-EIO); | |
182 | ptr = kmap(page); | |
183 | if (ptr == NULL) | |
184 | return ERR_PTR(-ENOMEM); | |
185 | return ptr; | |
186 | } | |
187 | ||
188 | static | |
189 | void nfs_readdir_release_array(struct page *page) | |
190 | { | |
191 | kunmap(page); | |
192 | } | |
193 | ||
194 | /* | |
195 | * we are freeing strings created by nfs_add_to_readdir_array() | |
196 | */ | |
197 | static | |
198 | void nfs_readdir_clear_array(struct page *page) | |
199 | { | |
200 | struct nfs_cache_array *array; | |
201 | int i; | |
202 | ||
203 | array = kmap_atomic(page); | |
204 | if (atomic_dec_and_test(&array->refcount)) | |
205 | for (i = 0; i < array->size; i++) | |
206 | kfree(array->array[i].string.name); | |
207 | kunmap_atomic(array); | |
208 | } | |
209 | ||
210 | static bool grab_page(struct page *page) | |
211 | { | |
212 | struct nfs_cache_array *array = kmap_atomic(page); | |
213 | bool res = atomic_inc_not_zero(&array->refcount); | |
214 | kunmap_atomic(array); | |
215 | return res; | |
216 | } | |
217 | ||
218 | /* | |
219 | * the caller is responsible for freeing qstr.name | |
220 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
221 | * nfs_clear_readdir_array() | |
222 | */ | |
223 | static | |
224 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) | |
225 | { | |
226 | string->len = len; | |
227 | string->name = kmemdup(name, len, GFP_KERNEL); | |
228 | if (string->name == NULL) | |
229 | return -ENOMEM; | |
230 | /* | |
231 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
232 | * in a page cache page which kmemleak does not scan. | |
233 | */ | |
234 | kmemleak_not_leak(string->name); | |
235 | string->hash = full_name_hash(name, len); | |
236 | return 0; | |
237 | } | |
238 | ||
239 | static | |
240 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
241 | { | |
242 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
243 | struct nfs_cache_array_entry *cache_entry; | |
244 | int ret; | |
245 | ||
246 | if (IS_ERR(array)) | |
247 | return PTR_ERR(array); | |
248 | ||
249 | cache_entry = &array->array[array->size]; | |
250 | ||
251 | /* Check that this entry lies within the page bounds */ | |
252 | ret = -ENOSPC; | |
253 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) | |
254 | goto out; | |
255 | ||
256 | cache_entry->cookie = entry->prev_cookie; | |
257 | cache_entry->ino = entry->ino; | |
258 | cache_entry->d_type = entry->d_type; | |
259 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); | |
260 | if (ret) | |
261 | goto out; | |
262 | array->last_cookie = entry->cookie; | |
263 | array->size++; | |
264 | if (entry->eof != 0) | |
265 | array->eof_index = array->size; | |
266 | out: | |
267 | nfs_readdir_release_array(page); | |
268 | return ret; | |
269 | } | |
270 | ||
271 | static | |
272 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
273 | { | |
274 | loff_t diff = desc->ctx->pos - desc->current_index; | |
275 | unsigned int index; | |
276 | ||
277 | if (diff < 0) | |
278 | goto out_eof; | |
279 | if (diff >= array->size) { | |
280 | if (array->eof_index >= 0) | |
281 | goto out_eof; | |
282 | return -EAGAIN; | |
283 | } | |
284 | ||
285 | index = (unsigned int)diff; | |
286 | *desc->dir_cookie = array->array[index].cookie; | |
287 | desc->cache_entry_index = index; | |
288 | return 0; | |
289 | out_eof: | |
290 | desc->eof = 1; | |
291 | return -EBADCOOKIE; | |
292 | } | |
293 | ||
294 | static bool | |
295 | nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) | |
296 | { | |
297 | if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) | |
298 | return false; | |
299 | smp_rmb(); | |
300 | return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); | |
301 | } | |
302 | ||
303 | static | |
304 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
305 | { | |
306 | int i; | |
307 | loff_t new_pos; | |
308 | int status = -EAGAIN; | |
309 | ||
310 | for (i = 0; i < array->size; i++) { | |
311 | if (array->array[i].cookie == *desc->dir_cookie) { | |
312 | struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); | |
313 | struct nfs_open_dir_context *ctx = desc->file->private_data; | |
314 | ||
315 | new_pos = desc->current_index + i; | |
316 | if (ctx->attr_gencount != nfsi->attr_gencount || | |
317 | !nfs_readdir_inode_mapping_valid(nfsi)) { | |
318 | ctx->duped = 0; | |
319 | ctx->attr_gencount = nfsi->attr_gencount; | |
320 | } else if (new_pos < desc->ctx->pos) { | |
321 | if (ctx->duped > 0 | |
322 | && ctx->dup_cookie == *desc->dir_cookie) { | |
323 | if (printk_ratelimit()) { | |
324 | pr_notice("NFS: directory %pD2 contains a readdir loop." | |
325 | "Please contact your server vendor. " | |
326 | "The file: %.*s has duplicate cookie %llu\n", | |
327 | desc->file, array->array[i].string.len, | |
328 | array->array[i].string.name, *desc->dir_cookie); | |
329 | } | |
330 | status = -ELOOP; | |
331 | goto out; | |
332 | } | |
333 | ctx->dup_cookie = *desc->dir_cookie; | |
334 | ctx->duped = -1; | |
335 | } | |
336 | desc->ctx->pos = new_pos; | |
337 | desc->cache_entry_index = i; | |
338 | return 0; | |
339 | } | |
340 | } | |
341 | if (array->eof_index >= 0) { | |
342 | status = -EBADCOOKIE; | |
343 | if (*desc->dir_cookie == array->last_cookie) | |
344 | desc->eof = 1; | |
345 | } | |
346 | out: | |
347 | return status; | |
348 | } | |
349 | ||
350 | static | |
351 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
352 | { | |
353 | struct nfs_cache_array *array; | |
354 | int status; | |
355 | ||
356 | array = nfs_readdir_get_array(desc->page); | |
357 | if (IS_ERR(array)) { | |
358 | status = PTR_ERR(array); | |
359 | goto out; | |
360 | } | |
361 | ||
362 | if (*desc->dir_cookie == 0) | |
363 | status = nfs_readdir_search_for_pos(array, desc); | |
364 | else | |
365 | status = nfs_readdir_search_for_cookie(array, desc); | |
366 | ||
367 | if (status == -EAGAIN) { | |
368 | desc->last_cookie = array->last_cookie; | |
369 | desc->current_index += array->size; | |
370 | desc->page_index++; | |
371 | } | |
372 | nfs_readdir_release_array(desc->page); | |
373 | out: | |
374 | return status; | |
375 | } | |
376 | ||
377 | /* Fill a page with xdr information before transferring to the cache page */ | |
378 | static | |
379 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, | |
380 | struct nfs_entry *entry, struct file *file, struct inode *inode) | |
381 | { | |
382 | struct nfs_open_dir_context *ctx = file->private_data; | |
383 | struct rpc_cred *cred = ctx->cred; | |
384 | unsigned long timestamp, gencount; | |
385 | int error; | |
386 | ||
387 | again: | |
388 | timestamp = jiffies; | |
389 | gencount = nfs_inc_attr_generation_counter(); | |
390 | error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, | |
391 | NFS_SERVER(inode)->dtsize, desc->plus); | |
392 | if (error < 0) { | |
393 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
394 | if (error == -ENOTSUPP && desc->plus) { | |
395 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
396 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); | |
397 | desc->plus = 0; | |
398 | goto again; | |
399 | } | |
400 | goto error; | |
401 | } | |
402 | desc->timestamp = timestamp; | |
403 | desc->gencount = gencount; | |
404 | error: | |
405 | return error; | |
406 | } | |
407 | ||
408 | static int xdr_decode(nfs_readdir_descriptor_t *desc, | |
409 | struct nfs_entry *entry, struct xdr_stream *xdr) | |
410 | { | |
411 | int error; | |
412 | ||
413 | error = desc->decode(xdr, entry, desc->plus); | |
414 | if (error) | |
415 | return error; | |
416 | entry->fattr->time_start = desc->timestamp; | |
417 | entry->fattr->gencount = desc->gencount; | |
418 | return 0; | |
419 | } | |
420 | ||
421 | /* Match file and dirent using either filehandle or fileid | |
422 | * Note: caller is responsible for checking the fsid | |
423 | */ | |
424 | static | |
425 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
426 | { | |
427 | struct nfs_inode *nfsi; | |
428 | ||
429 | if (d_really_is_negative(dentry)) | |
430 | return 0; | |
431 | ||
432 | nfsi = NFS_I(d_inode(dentry)); | |
433 | if (entry->fattr->fileid == nfsi->fileid) | |
434 | return 1; | |
435 | if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0) | |
436 | return 1; | |
437 | return 0; | |
438 | } | |
439 | ||
440 | static | |
441 | bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) | |
442 | { | |
443 | if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) | |
444 | return false; | |
445 | if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) | |
446 | return true; | |
447 | if (ctx->pos == 0) | |
448 | return true; | |
449 | return false; | |
450 | } | |
451 | ||
452 | /* | |
453 | * This function is called by the lookup code to request the use of | |
454 | * readdirplus to accelerate any future lookups in the same | |
455 | * directory. | |
456 | */ | |
457 | static | |
458 | void nfs_advise_use_readdirplus(struct inode *dir) | |
459 | { | |
460 | set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); | |
461 | } | |
462 | ||
463 | /* | |
464 | * This function is mainly for use by nfs_getattr(). | |
465 | * | |
466 | * If this is an 'ls -l', we want to force use of readdirplus. | |
467 | * Do this by checking if there is an active file descriptor | |
468 | * and calling nfs_advise_use_readdirplus, then forcing a | |
469 | * cache flush. | |
470 | */ | |
471 | void nfs_force_use_readdirplus(struct inode *dir) | |
472 | { | |
473 | if (!list_empty(&NFS_I(dir)->open_files)) { | |
474 | nfs_advise_use_readdirplus(dir); | |
475 | nfs_zap_mapping(dir, dir->i_mapping); | |
476 | } | |
477 | } | |
478 | ||
479 | static | |
480 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
481 | { | |
482 | struct qstr filename = QSTR_INIT(entry->name, entry->len); | |
483 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); | |
484 | struct dentry *dentry; | |
485 | struct dentry *alias; | |
486 | struct inode *dir = d_inode(parent); | |
487 | struct inode *inode; | |
488 | int status; | |
489 | ||
490 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) | |
491 | return; | |
492 | if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) | |
493 | return; | |
494 | if (filename.name[0] == '.') { | |
495 | if (filename.len == 1) | |
496 | return; | |
497 | if (filename.len == 2 && filename.name[1] == '.') | |
498 | return; | |
499 | } | |
500 | filename.hash = full_name_hash(filename.name, filename.len); | |
501 | ||
502 | dentry = d_lookup(parent, &filename); | |
503 | again: | |
504 | if (!dentry) { | |
505 | dentry = d_alloc_parallel(parent, &filename, &wq); | |
506 | if (IS_ERR(dentry)) | |
507 | return; | |
508 | } | |
509 | if (!d_in_lookup(dentry)) { | |
510 | /* Is there a mountpoint here? If so, just exit */ | |
511 | if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, | |
512 | &entry->fattr->fsid)) | |
513 | goto out; | |
514 | if (nfs_same_file(dentry, entry)) { | |
515 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
516 | status = nfs_refresh_inode(d_inode(dentry), entry->fattr); | |
517 | if (!status) | |
518 | nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); | |
519 | goto out; | |
520 | } else { | |
521 | d_invalidate(dentry); | |
522 | dput(dentry); | |
523 | dentry = NULL; | |
524 | goto again; | |
525 | } | |
526 | } | |
527 | ||
528 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); | |
529 | alias = d_splice_alias(inode, dentry); | |
530 | d_lookup_done(dentry); | |
531 | if (alias) { | |
532 | if (IS_ERR(alias)) | |
533 | goto out; | |
534 | dput(dentry); | |
535 | dentry = alias; | |
536 | } | |
537 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
538 | out: | |
539 | dput(dentry); | |
540 | } | |
541 | ||
542 | /* Perform conversion from xdr to cache array */ | |
543 | static | |
544 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, | |
545 | struct page **xdr_pages, struct page *page, unsigned int buflen) | |
546 | { | |
547 | struct xdr_stream stream; | |
548 | struct xdr_buf buf; | |
549 | struct page *scratch; | |
550 | struct nfs_cache_array *array; | |
551 | unsigned int count = 0; | |
552 | int status; | |
553 | ||
554 | scratch = alloc_page(GFP_KERNEL); | |
555 | if (scratch == NULL) | |
556 | return -ENOMEM; | |
557 | ||
558 | if (buflen == 0) | |
559 | goto out_nopages; | |
560 | ||
561 | xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); | |
562 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); | |
563 | ||
564 | do { | |
565 | status = xdr_decode(desc, entry, &stream); | |
566 | if (status != 0) { | |
567 | if (status == -EAGAIN) | |
568 | status = 0; | |
569 | break; | |
570 | } | |
571 | ||
572 | count++; | |
573 | ||
574 | if (desc->plus != 0) | |
575 | nfs_prime_dcache(file_dentry(desc->file), entry); | |
576 | ||
577 | status = nfs_readdir_add_to_array(entry, page); | |
578 | if (status != 0) | |
579 | break; | |
580 | } while (!entry->eof); | |
581 | ||
582 | out_nopages: | |
583 | if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { | |
584 | array = nfs_readdir_get_array(page); | |
585 | if (!IS_ERR(array)) { | |
586 | array->eof_index = array->size; | |
587 | status = 0; | |
588 | nfs_readdir_release_array(page); | |
589 | } else | |
590 | status = PTR_ERR(array); | |
591 | } | |
592 | ||
593 | put_page(scratch); | |
594 | return status; | |
595 | } | |
596 | ||
597 | static | |
598 | void nfs_readdir_free_pages(struct page **pages, unsigned int npages) | |
599 | { | |
600 | unsigned int i; | |
601 | for (i = 0; i < npages; i++) | |
602 | put_page(pages[i]); | |
603 | } | |
604 | ||
605 | /* | |
606 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
607 | * to nfs_readdir_free_pagearray | |
608 | */ | |
609 | static | |
610 | int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) | |
611 | { | |
612 | unsigned int i; | |
613 | ||
614 | for (i = 0; i < npages; i++) { | |
615 | struct page *page = alloc_page(GFP_KERNEL); | |
616 | if (page == NULL) | |
617 | goto out_freepages; | |
618 | pages[i] = page; | |
619 | } | |
620 | return 0; | |
621 | ||
622 | out_freepages: | |
623 | nfs_readdir_free_pages(pages, i); | |
624 | return -ENOMEM; | |
625 | } | |
626 | ||
627 | static | |
628 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
629 | { | |
630 | struct page *pages[NFS_MAX_READDIR_PAGES]; | |
631 | struct nfs_entry entry; | |
632 | struct file *file = desc->file; | |
633 | struct nfs_cache_array *array; | |
634 | int status = -ENOMEM; | |
635 | unsigned int array_size = ARRAY_SIZE(pages); | |
636 | ||
637 | entry.prev_cookie = 0; | |
638 | entry.cookie = desc->last_cookie; | |
639 | entry.eof = 0; | |
640 | entry.fh = nfs_alloc_fhandle(); | |
641 | entry.fattr = nfs_alloc_fattr(); | |
642 | entry.server = NFS_SERVER(inode); | |
643 | if (entry.fh == NULL || entry.fattr == NULL) | |
644 | goto out; | |
645 | ||
646 | entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); | |
647 | if (IS_ERR(entry.label)) { | |
648 | status = PTR_ERR(entry.label); | |
649 | goto out; | |
650 | } | |
651 | ||
652 | array = nfs_readdir_get_array(page); | |
653 | if (IS_ERR(array)) { | |
654 | status = PTR_ERR(array); | |
655 | goto out_label_free; | |
656 | } | |
657 | memset(array, 0, sizeof(struct nfs_cache_array)); | |
658 | atomic_set(&array->refcount, 1); | |
659 | array->eof_index = -1; | |
660 | ||
661 | status = nfs_readdir_alloc_pages(pages, array_size); | |
662 | if (status < 0) | |
663 | goto out_release_array; | |
664 | do { | |
665 | unsigned int pglen; | |
666 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); | |
667 | ||
668 | if (status < 0) | |
669 | break; | |
670 | pglen = status; | |
671 | status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); | |
672 | if (status < 0) { | |
673 | if (status == -ENOSPC) | |
674 | status = 0; | |
675 | break; | |
676 | } | |
677 | } while (array->eof_index < 0); | |
678 | ||
679 | nfs_readdir_free_pages(pages, array_size); | |
680 | out_release_array: | |
681 | nfs_readdir_release_array(page); | |
682 | out_label_free: | |
683 | nfs4_label_free(entry.label); | |
684 | out: | |
685 | nfs_free_fattr(entry.fattr); | |
686 | nfs_free_fhandle(entry.fh); | |
687 | return status; | |
688 | } | |
689 | ||
690 | /* | |
691 | * Now we cache directories properly, by converting xdr information | |
692 | * to an array that can be used for lookups later. This results in | |
693 | * fewer cache pages, since we can store more information on each page. | |
694 | * We only need to convert from xdr once so future lookups are much simpler | |
695 | */ | |
696 | static | |
697 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
698 | { | |
699 | struct inode *inode = file_inode(desc->file); | |
700 | int ret; | |
701 | ||
702 | ret = nfs_readdir_xdr_to_array(desc, page, inode); | |
703 | if (ret < 0) | |
704 | goto error; | |
705 | SetPageUptodate(page); | |
706 | ||
707 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { | |
708 | /* Should never happen */ | |
709 | nfs_zap_mapping(inode, inode->i_mapping); | |
710 | } | |
711 | unlock_page(page); | |
712 | return 0; | |
713 | error: | |
714 | unlock_page(page); | |
715 | return ret; | |
716 | } | |
717 | ||
718 | static | |
719 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
720 | { | |
721 | nfs_readdir_clear_array(desc->page); | |
722 | put_page(desc->page); | |
723 | desc->page = NULL; | |
724 | } | |
725 | ||
726 | static | |
727 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
728 | { | |
729 | struct page *page; | |
730 | ||
731 | for (;;) { | |
732 | page = read_cache_page(file_inode(desc->file)->i_mapping, | |
733 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); | |
734 | if (IS_ERR(page) || grab_page(page)) | |
735 | break; | |
736 | put_page(page); | |
737 | } | |
738 | return page; | |
739 | } | |
740 | ||
741 | /* | |
742 | * Returns 0 if desc->dir_cookie was found on page desc->page_index | |
743 | */ | |
744 | static | |
745 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
746 | { | |
747 | int res; | |
748 | ||
749 | desc->page = get_cache_page(desc); | |
750 | if (IS_ERR(desc->page)) | |
751 | return PTR_ERR(desc->page); | |
752 | ||
753 | res = nfs_readdir_search_array(desc); | |
754 | if (res != 0) | |
755 | cache_page_release(desc); | |
756 | return res; | |
757 | } | |
758 | ||
759 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
760 | static inline | |
761 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
762 | { | |
763 | int res; | |
764 | ||
765 | if (desc->page_index == 0) { | |
766 | desc->current_index = 0; | |
767 | desc->last_cookie = 0; | |
768 | } | |
769 | do { | |
770 | res = find_cache_page(desc); | |
771 | } while (res == -EAGAIN); | |
772 | return res; | |
773 | } | |
774 | ||
775 | /* | |
776 | * Once we've found the start of the dirent within a page: fill 'er up... | |
777 | */ | |
778 | static | |
779 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc) | |
780 | { | |
781 | struct file *file = desc->file; | |
782 | int i = 0; | |
783 | int res = 0; | |
784 | struct nfs_cache_array *array = NULL; | |
785 | struct nfs_open_dir_context *ctx = file->private_data; | |
786 | ||
787 | array = nfs_readdir_get_array(desc->page); | |
788 | if (IS_ERR(array)) { | |
789 | res = PTR_ERR(array); | |
790 | goto out; | |
791 | } | |
792 | ||
793 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
794 | struct nfs_cache_array_entry *ent; | |
795 | ||
796 | ent = &array->array[i]; | |
797 | if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, | |
798 | nfs_compat_user_ino64(ent->ino), ent->d_type)) { | |
799 | desc->eof = 1; | |
800 | break; | |
801 | } | |
802 | desc->ctx->pos++; | |
803 | if (i < (array->size-1)) | |
804 | *desc->dir_cookie = array->array[i+1].cookie; | |
805 | else | |
806 | *desc->dir_cookie = array->last_cookie; | |
807 | if (ctx->duped != 0) | |
808 | ctx->duped = 1; | |
809 | } | |
810 | if (array->eof_index >= 0) | |
811 | desc->eof = 1; | |
812 | ||
813 | nfs_readdir_release_array(desc->page); | |
814 | out: | |
815 | cache_page_release(desc); | |
816 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", | |
817 | (unsigned long long)*desc->dir_cookie, res); | |
818 | return res; | |
819 | } | |
820 | ||
821 | /* | |
822 | * If we cannot find a cookie in our cache, we suspect that this is | |
823 | * because it points to a deleted file, so we ask the server to return | |
824 | * whatever it thinks is the next entry. We then feed this to filldir. | |
825 | * If all goes well, we should then be able to find our way round the | |
826 | * cache on the next call to readdir_search_pagecache(); | |
827 | * | |
828 | * NOTE: we cannot add the anonymous page to the pagecache because | |
829 | * the data it contains might not be page aligned. Besides, | |
830 | * we should already have a complete representation of the | |
831 | * directory in the page cache by the time we get here. | |
832 | */ | |
833 | static inline | |
834 | int uncached_readdir(nfs_readdir_descriptor_t *desc) | |
835 | { | |
836 | struct page *page = NULL; | |
837 | int status; | |
838 | struct inode *inode = file_inode(desc->file); | |
839 | struct nfs_open_dir_context *ctx = desc->file->private_data; | |
840 | ||
841 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", | |
842 | (unsigned long long)*desc->dir_cookie); | |
843 | ||
844 | page = alloc_page(GFP_HIGHUSER); | |
845 | if (!page) { | |
846 | status = -ENOMEM; | |
847 | goto out; | |
848 | } | |
849 | ||
850 | desc->page_index = 0; | |
851 | desc->last_cookie = *desc->dir_cookie; | |
852 | desc->page = page; | |
853 | ctx->duped = 0; | |
854 | ||
855 | status = nfs_readdir_xdr_to_array(desc, page, inode); | |
856 | if (status < 0) | |
857 | goto out_release; | |
858 | ||
859 | status = nfs_do_filldir(desc); | |
860 | ||
861 | out: | |
862 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", | |
863 | __func__, status); | |
864 | return status; | |
865 | out_release: | |
866 | cache_page_release(desc); | |
867 | goto out; | |
868 | } | |
869 | ||
870 | static bool nfs_dir_mapping_need_revalidate(struct inode *dir) | |
871 | { | |
872 | struct nfs_inode *nfsi = NFS_I(dir); | |
873 | ||
874 | if (nfs_attribute_cache_expired(dir)) | |
875 | return true; | |
876 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) | |
877 | return true; | |
878 | return false; | |
879 | } | |
880 | ||
881 | /* The file offset position represents the dirent entry number. A | |
882 | last cookie cache takes care of the common case of reading the | |
883 | whole directory. | |
884 | */ | |
885 | static int nfs_readdir(struct file *file, struct dir_context *ctx) | |
886 | { | |
887 | struct dentry *dentry = file_dentry(file); | |
888 | struct inode *inode = d_inode(dentry); | |
889 | nfs_readdir_descriptor_t my_desc, | |
890 | *desc = &my_desc; | |
891 | struct nfs_open_dir_context *dir_ctx = file->private_data; | |
892 | int res = 0; | |
893 | ||
894 | dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", | |
895 | file, (long long)ctx->pos); | |
896 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); | |
897 | ||
898 | /* | |
899 | * ctx->pos points to the dirent entry number. | |
900 | * *desc->dir_cookie has the cookie for the next entry. We have | |
901 | * to either find the entry with the appropriate number or | |
902 | * revalidate the cookie. | |
903 | */ | |
904 | memset(desc, 0, sizeof(*desc)); | |
905 | ||
906 | desc->file = file; | |
907 | desc->ctx = ctx; | |
908 | desc->dir_cookie = &dir_ctx->dir_cookie; | |
909 | desc->decode = NFS_PROTO(inode)->decode_dirent; | |
910 | desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; | |
911 | ||
912 | if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode)) | |
913 | res = nfs_revalidate_mapping(inode, file->f_mapping); | |
914 | if (res < 0) | |
915 | goto out; | |
916 | ||
917 | do { | |
918 | res = readdir_search_pagecache(desc); | |
919 | ||
920 | if (res == -EBADCOOKIE) { | |
921 | res = 0; | |
922 | /* This means either end of directory */ | |
923 | if (*desc->dir_cookie && desc->eof == 0) { | |
924 | /* Or that the server has 'lost' a cookie */ | |
925 | res = uncached_readdir(desc); | |
926 | if (res == 0) | |
927 | continue; | |
928 | } | |
929 | break; | |
930 | } | |
931 | if (res == -ETOOSMALL && desc->plus) { | |
932 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); | |
933 | nfs_zap_caches(inode); | |
934 | desc->page_index = 0; | |
935 | desc->plus = 0; | |
936 | desc->eof = 0; | |
937 | continue; | |
938 | } | |
939 | if (res < 0) | |
940 | break; | |
941 | ||
942 | res = nfs_do_filldir(desc); | |
943 | if (res < 0) | |
944 | break; | |
945 | } while (!desc->eof); | |
946 | out: | |
947 | if (res > 0) | |
948 | res = 0; | |
949 | dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); | |
950 | return res; | |
951 | } | |
952 | ||
953 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) | |
954 | { | |
955 | struct nfs_open_dir_context *dir_ctx = filp->private_data; | |
956 | ||
957 | dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", | |
958 | filp, offset, whence); | |
959 | ||
960 | switch (whence) { | |
961 | case 1: | |
962 | offset += filp->f_pos; | |
963 | case 0: | |
964 | if (offset >= 0) | |
965 | break; | |
966 | default: | |
967 | return -EINVAL; | |
968 | } | |
969 | if (offset != filp->f_pos) { | |
970 | filp->f_pos = offset; | |
971 | dir_ctx->dir_cookie = 0; | |
972 | dir_ctx->duped = 0; | |
973 | } | |
974 | return offset; | |
975 | } | |
976 | ||
977 | /* | |
978 | * All directory operations under NFS are synchronous, so fsync() | |
979 | * is a dummy operation. | |
980 | */ | |
981 | static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, | |
982 | int datasync) | |
983 | { | |
984 | struct inode *inode = file_inode(filp); | |
985 | ||
986 | dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); | |
987 | ||
988 | inode_lock(inode); | |
989 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); | |
990 | inode_unlock(inode); | |
991 | return 0; | |
992 | } | |
993 | ||
994 | /** | |
995 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
996 | * @dir - pointer to directory inode | |
997 | * | |
998 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
999 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
1000 | * on the server that might have invalidated our dcache. | |
1001 | * | |
1002 | * The caller should be holding dir->i_lock | |
1003 | */ | |
1004 | void nfs_force_lookup_revalidate(struct inode *dir) | |
1005 | { | |
1006 | NFS_I(dir)->cache_change_attribute++; | |
1007 | } | |
1008 | EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); | |
1009 | ||
1010 | /* | |
1011 | * A check for whether or not the parent directory has changed. | |
1012 | * In the case it has, we assume that the dentries are untrustworthy | |
1013 | * and may need to be looked up again. | |
1014 | * If rcu_walk prevents us from performing a full check, return 0. | |
1015 | */ | |
1016 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, | |
1017 | int rcu_walk) | |
1018 | { | |
1019 | int ret; | |
1020 | ||
1021 | if (IS_ROOT(dentry)) | |
1022 | return 1; | |
1023 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) | |
1024 | return 0; | |
1025 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
1026 | return 0; | |
1027 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
1028 | if (rcu_walk) | |
1029 | ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir); | |
1030 | else | |
1031 | ret = nfs_revalidate_inode(NFS_SERVER(dir), dir); | |
1032 | if (ret < 0) | |
1033 | return 0; | |
1034 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
1035 | return 0; | |
1036 | return 1; | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Use intent information to check whether or not we're going to do | |
1041 | * an O_EXCL create using this path component. | |
1042 | */ | |
1043 | static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) | |
1044 | { | |
1045 | if (NFS_PROTO(dir)->version == 2) | |
1046 | return 0; | |
1047 | return flags & LOOKUP_EXCL; | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * Inode and filehandle revalidation for lookups. | |
1052 | * | |
1053 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
1054 | * or if the intent information indicates that we're about to open this | |
1055 | * particular file and the "nocto" mount flag is not set. | |
1056 | * | |
1057 | */ | |
1058 | static | |
1059 | int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) | |
1060 | { | |
1061 | struct nfs_server *server = NFS_SERVER(inode); | |
1062 | int ret; | |
1063 | ||
1064 | if (IS_AUTOMOUNT(inode)) | |
1065 | return 0; | |
1066 | /* VFS wants an on-the-wire revalidation */ | |
1067 | if (flags & LOOKUP_REVAL) | |
1068 | goto out_force; | |
1069 | /* This is an open(2) */ | |
1070 | if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && | |
1071 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) | |
1072 | goto out_force; | |
1073 | out: | |
1074 | return (inode->i_nlink == 0) ? -ENOENT : 0; | |
1075 | out_force: | |
1076 | if (flags & LOOKUP_RCU) | |
1077 | return -ECHILD; | |
1078 | ret = __nfs_revalidate_inode(server, inode); | |
1079 | if (ret != 0) | |
1080 | return ret; | |
1081 | goto out; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * We judge how long we want to trust negative | |
1086 | * dentries by looking at the parent inode mtime. | |
1087 | * | |
1088 | * If parent mtime has changed, we revalidate, else we wait for a | |
1089 | * period corresponding to the parent's attribute cache timeout value. | |
1090 | * | |
1091 | * If LOOKUP_RCU prevents us from performing a full check, return 1 | |
1092 | * suggesting a reval is needed. | |
1093 | */ | |
1094 | static inline | |
1095 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
1096 | unsigned int flags) | |
1097 | { | |
1098 | /* Don't revalidate a negative dentry if we're creating a new file */ | |
1099 | if (flags & LOOKUP_CREATE) | |
1100 | return 0; | |
1101 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) | |
1102 | return 1; | |
1103 | return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); | |
1104 | } | |
1105 | ||
1106 | /* | |
1107 | * This is called every time the dcache has a lookup hit, | |
1108 | * and we should check whether we can really trust that | |
1109 | * lookup. | |
1110 | * | |
1111 | * NOTE! The hit can be a negative hit too, don't assume | |
1112 | * we have an inode! | |
1113 | * | |
1114 | * If the parent directory is seen to have changed, we throw out the | |
1115 | * cached dentry and do a new lookup. | |
1116 | */ | |
1117 | static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) | |
1118 | { | |
1119 | struct inode *dir; | |
1120 | struct inode *inode; | |
1121 | struct dentry *parent; | |
1122 | struct nfs_fh *fhandle = NULL; | |
1123 | struct nfs_fattr *fattr = NULL; | |
1124 | struct nfs4_label *label = NULL; | |
1125 | int error; | |
1126 | ||
1127 | if (flags & LOOKUP_RCU) { | |
1128 | parent = ACCESS_ONCE(dentry->d_parent); | |
1129 | dir = d_inode_rcu(parent); | |
1130 | if (!dir) | |
1131 | return -ECHILD; | |
1132 | } else { | |
1133 | parent = dget_parent(dentry); | |
1134 | dir = d_inode(parent); | |
1135 | } | |
1136 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); | |
1137 | inode = d_inode(dentry); | |
1138 | ||
1139 | if (!inode) { | |
1140 | if (nfs_neg_need_reval(dir, dentry, flags)) { | |
1141 | if (flags & LOOKUP_RCU) | |
1142 | return -ECHILD; | |
1143 | goto out_bad; | |
1144 | } | |
1145 | goto out_valid_noent; | |
1146 | } | |
1147 | ||
1148 | if (is_bad_inode(inode)) { | |
1149 | if (flags & LOOKUP_RCU) | |
1150 | return -ECHILD; | |
1151 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", | |
1152 | __func__, dentry); | |
1153 | goto out_bad; | |
1154 | } | |
1155 | ||
1156 | if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) | |
1157 | goto out_set_verifier; | |
1158 | ||
1159 | /* Force a full look up iff the parent directory has changed */ | |
1160 | if (!nfs_is_exclusive_create(dir, flags) && | |
1161 | nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { | |
1162 | ||
1163 | if (nfs_lookup_verify_inode(inode, flags)) { | |
1164 | if (flags & LOOKUP_RCU) | |
1165 | return -ECHILD; | |
1166 | goto out_zap_parent; | |
1167 | } | |
1168 | goto out_valid; | |
1169 | } | |
1170 | ||
1171 | if (flags & LOOKUP_RCU) | |
1172 | return -ECHILD; | |
1173 | ||
1174 | if (NFS_STALE(inode)) | |
1175 | goto out_bad; | |
1176 | ||
1177 | error = -ENOMEM; | |
1178 | fhandle = nfs_alloc_fhandle(); | |
1179 | fattr = nfs_alloc_fattr(); | |
1180 | if (fhandle == NULL || fattr == NULL) | |
1181 | goto out_error; | |
1182 | ||
1183 | label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); | |
1184 | if (IS_ERR(label)) | |
1185 | goto out_error; | |
1186 | ||
1187 | trace_nfs_lookup_revalidate_enter(dir, dentry, flags); | |
1188 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); | |
1189 | trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); | |
1190 | if (error) | |
1191 | goto out_bad; | |
1192 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) | |
1193 | goto out_bad; | |
1194 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) | |
1195 | goto out_bad; | |
1196 | ||
1197 | nfs_setsecurity(inode, fattr, label); | |
1198 | ||
1199 | nfs_free_fattr(fattr); | |
1200 | nfs_free_fhandle(fhandle); | |
1201 | nfs4_label_free(label); | |
1202 | ||
1203 | out_set_verifier: | |
1204 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1205 | out_valid: | |
1206 | /* Success: notify readdir to use READDIRPLUS */ | |
1207 | nfs_advise_use_readdirplus(dir); | |
1208 | out_valid_noent: | |
1209 | if (flags & LOOKUP_RCU) { | |
1210 | if (parent != ACCESS_ONCE(dentry->d_parent)) | |
1211 | return -ECHILD; | |
1212 | } else | |
1213 | dput(parent); | |
1214 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", | |
1215 | __func__, dentry); | |
1216 | return 1; | |
1217 | out_zap_parent: | |
1218 | nfs_zap_caches(dir); | |
1219 | out_bad: | |
1220 | WARN_ON(flags & LOOKUP_RCU); | |
1221 | nfs_free_fattr(fattr); | |
1222 | nfs_free_fhandle(fhandle); | |
1223 | nfs4_label_free(label); | |
1224 | nfs_mark_for_revalidate(dir); | |
1225 | if (inode && S_ISDIR(inode->i_mode)) { | |
1226 | /* Purge readdir caches. */ | |
1227 | nfs_zap_caches(inode); | |
1228 | /* | |
1229 | * We can't d_drop the root of a disconnected tree: | |
1230 | * its d_hash is on the s_anon list and d_drop() would hide | |
1231 | * it from shrink_dcache_for_unmount(), leading to busy | |
1232 | * inodes on unmount and further oopses. | |
1233 | */ | |
1234 | if (IS_ROOT(dentry)) | |
1235 | goto out_valid; | |
1236 | } | |
1237 | dput(parent); | |
1238 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", | |
1239 | __func__, dentry); | |
1240 | return 0; | |
1241 | out_error: | |
1242 | WARN_ON(flags & LOOKUP_RCU); | |
1243 | nfs_free_fattr(fattr); | |
1244 | nfs_free_fhandle(fhandle); | |
1245 | nfs4_label_free(label); | |
1246 | dput(parent); | |
1247 | dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", | |
1248 | __func__, dentry, error); | |
1249 | return error; | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * A weaker form of d_revalidate for revalidating just the d_inode(dentry) | |
1254 | * when we don't really care about the dentry name. This is called when a | |
1255 | * pathwalk ends on a dentry that was not found via a normal lookup in the | |
1256 | * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). | |
1257 | * | |
1258 | * In this situation, we just want to verify that the inode itself is OK | |
1259 | * since the dentry might have changed on the server. | |
1260 | */ | |
1261 | static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) | |
1262 | { | |
1263 | int error; | |
1264 | struct inode *inode = d_inode(dentry); | |
1265 | ||
1266 | /* | |
1267 | * I believe we can only get a negative dentry here in the case of a | |
1268 | * procfs-style symlink. Just assume it's correct for now, but we may | |
1269 | * eventually need to do something more here. | |
1270 | */ | |
1271 | if (!inode) { | |
1272 | dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", | |
1273 | __func__, dentry); | |
1274 | return 1; | |
1275 | } | |
1276 | ||
1277 | if (is_bad_inode(inode)) { | |
1278 | dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", | |
1279 | __func__, dentry); | |
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | error = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
1284 | dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", | |
1285 | __func__, inode->i_ino, error ? "invalid" : "valid"); | |
1286 | return !error; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * This is called from dput() when d_count is going to 0. | |
1291 | */ | |
1292 | static int nfs_dentry_delete(const struct dentry *dentry) | |
1293 | { | |
1294 | dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", | |
1295 | dentry, dentry->d_flags); | |
1296 | ||
1297 | /* Unhash any dentry with a stale inode */ | |
1298 | if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) | |
1299 | return 1; | |
1300 | ||
1301 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1302 | /* Unhash it, so that ->d_iput() would be called */ | |
1303 | return 1; | |
1304 | } | |
1305 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1306 | /* Unhash it, so that ancestors of killed async unlink | |
1307 | * files will be cleaned up during umount */ | |
1308 | return 1; | |
1309 | } | |
1310 | return 0; | |
1311 | ||
1312 | } | |
1313 | ||
1314 | /* Ensure that we revalidate inode->i_nlink */ | |
1315 | static void nfs_drop_nlink(struct inode *inode) | |
1316 | { | |
1317 | spin_lock(&inode->i_lock); | |
1318 | /* drop the inode if we're reasonably sure this is the last link */ | |
1319 | if (inode->i_nlink == 1) | |
1320 | clear_nlink(inode); | |
1321 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; | |
1322 | spin_unlock(&inode->i_lock); | |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * Called when the dentry loses inode. | |
1327 | * We use it to clean up silly-renamed files. | |
1328 | */ | |
1329 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1330 | { | |
1331 | if (S_ISDIR(inode->i_mode)) | |
1332 | /* drop any readdir cache as it could easily be old */ | |
1333 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1334 | ||
1335 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1336 | nfs_complete_unlink(dentry, inode); | |
1337 | nfs_drop_nlink(inode); | |
1338 | } | |
1339 | iput(inode); | |
1340 | } | |
1341 | ||
1342 | static void nfs_d_release(struct dentry *dentry) | |
1343 | { | |
1344 | /* free cached devname value, if it survived that far */ | |
1345 | if (unlikely(dentry->d_fsdata)) { | |
1346 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1347 | WARN_ON(1); | |
1348 | else | |
1349 | kfree(dentry->d_fsdata); | |
1350 | } | |
1351 | } | |
1352 | ||
1353 | const struct dentry_operations nfs_dentry_operations = { | |
1354 | .d_revalidate = nfs_lookup_revalidate, | |
1355 | .d_weak_revalidate = nfs_weak_revalidate, | |
1356 | .d_delete = nfs_dentry_delete, | |
1357 | .d_iput = nfs_dentry_iput, | |
1358 | .d_automount = nfs_d_automount, | |
1359 | .d_release = nfs_d_release, | |
1360 | }; | |
1361 | EXPORT_SYMBOL_GPL(nfs_dentry_operations); | |
1362 | ||
1363 | struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) | |
1364 | { | |
1365 | struct dentry *res; | |
1366 | struct dentry *parent; | |
1367 | struct inode *inode = NULL; | |
1368 | struct nfs_fh *fhandle = NULL; | |
1369 | struct nfs_fattr *fattr = NULL; | |
1370 | struct nfs4_label *label = NULL; | |
1371 | int error; | |
1372 | ||
1373 | dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); | |
1374 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); | |
1375 | ||
1376 | if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) | |
1377 | return ERR_PTR(-ENAMETOOLONG); | |
1378 | ||
1379 | /* | |
1380 | * If we're doing an exclusive create, optimize away the lookup | |
1381 | * but don't hash the dentry. | |
1382 | */ | |
1383 | if (nfs_is_exclusive_create(dir, flags)) | |
1384 | return NULL; | |
1385 | ||
1386 | res = ERR_PTR(-ENOMEM); | |
1387 | fhandle = nfs_alloc_fhandle(); | |
1388 | fattr = nfs_alloc_fattr(); | |
1389 | if (fhandle == NULL || fattr == NULL) | |
1390 | goto out; | |
1391 | ||
1392 | label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); | |
1393 | if (IS_ERR(label)) | |
1394 | goto out; | |
1395 | ||
1396 | parent = dentry->d_parent; | |
1397 | /* Protect against concurrent sillydeletes */ | |
1398 | trace_nfs_lookup_enter(dir, dentry, flags); | |
1399 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); | |
1400 | if (error == -ENOENT) | |
1401 | goto no_entry; | |
1402 | if (error < 0) { | |
1403 | res = ERR_PTR(error); | |
1404 | goto out_unblock_sillyrename; | |
1405 | } | |
1406 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); | |
1407 | res = ERR_CAST(inode); | |
1408 | if (IS_ERR(res)) | |
1409 | goto out_unblock_sillyrename; | |
1410 | ||
1411 | /* Success: notify readdir to use READDIRPLUS */ | |
1412 | nfs_advise_use_readdirplus(dir); | |
1413 | ||
1414 | no_entry: | |
1415 | res = d_splice_alias(inode, dentry); | |
1416 | if (res != NULL) { | |
1417 | if (IS_ERR(res)) | |
1418 | goto out_unblock_sillyrename; | |
1419 | dentry = res; | |
1420 | } | |
1421 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1422 | out_unblock_sillyrename: | |
1423 | trace_nfs_lookup_exit(dir, dentry, flags, error); | |
1424 | nfs4_label_free(label); | |
1425 | out: | |
1426 | nfs_free_fattr(fattr); | |
1427 | nfs_free_fhandle(fhandle); | |
1428 | return res; | |
1429 | } | |
1430 | EXPORT_SYMBOL_GPL(nfs_lookup); | |
1431 | ||
1432 | #if IS_ENABLED(CONFIG_NFS_V4) | |
1433 | static int nfs4_lookup_revalidate(struct dentry *, unsigned int); | |
1434 | ||
1435 | const struct dentry_operations nfs4_dentry_operations = { | |
1436 | .d_revalidate = nfs4_lookup_revalidate, | |
1437 | .d_delete = nfs_dentry_delete, | |
1438 | .d_iput = nfs_dentry_iput, | |
1439 | .d_automount = nfs_d_automount, | |
1440 | .d_release = nfs_d_release, | |
1441 | }; | |
1442 | EXPORT_SYMBOL_GPL(nfs4_dentry_operations); | |
1443 | ||
1444 | static fmode_t flags_to_mode(int flags) | |
1445 | { | |
1446 | fmode_t res = (__force fmode_t)flags & FMODE_EXEC; | |
1447 | if ((flags & O_ACCMODE) != O_WRONLY) | |
1448 | res |= FMODE_READ; | |
1449 | if ((flags & O_ACCMODE) != O_RDONLY) | |
1450 | res |= FMODE_WRITE; | |
1451 | return res; | |
1452 | } | |
1453 | ||
1454 | static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) | |
1455 | { | |
1456 | return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); | |
1457 | } | |
1458 | ||
1459 | static int do_open(struct inode *inode, struct file *filp) | |
1460 | { | |
1461 | nfs_fscache_open_file(inode, filp); | |
1462 | return 0; | |
1463 | } | |
1464 | ||
1465 | static int nfs_finish_open(struct nfs_open_context *ctx, | |
1466 | struct dentry *dentry, | |
1467 | struct file *file, unsigned open_flags, | |
1468 | int *opened) | |
1469 | { | |
1470 | int err; | |
1471 | ||
1472 | err = finish_open(file, dentry, do_open, opened); | |
1473 | if (err) | |
1474 | goto out; | |
1475 | nfs_file_set_open_context(file, ctx); | |
1476 | ||
1477 | out: | |
1478 | return err; | |
1479 | } | |
1480 | ||
1481 | int nfs_atomic_open(struct inode *dir, struct dentry *dentry, | |
1482 | struct file *file, unsigned open_flags, | |
1483 | umode_t mode, int *opened) | |
1484 | { | |
1485 | struct nfs_open_context *ctx; | |
1486 | struct dentry *res; | |
1487 | struct iattr attr = { .ia_valid = ATTR_OPEN }; | |
1488 | struct inode *inode; | |
1489 | unsigned int lookup_flags = 0; | |
1490 | int err; | |
1491 | ||
1492 | /* Expect a negative dentry */ | |
1493 | BUG_ON(d_inode(dentry)); | |
1494 | ||
1495 | dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", | |
1496 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1497 | ||
1498 | err = nfs_check_flags(open_flags); | |
1499 | if (err) | |
1500 | return err; | |
1501 | ||
1502 | /* NFS only supports OPEN on regular files */ | |
1503 | if ((open_flags & O_DIRECTORY)) { | |
1504 | if (!d_unhashed(dentry)) { | |
1505 | /* | |
1506 | * Hashed negative dentry with O_DIRECTORY: dentry was | |
1507 | * revalidated and is fine, no need to perform lookup | |
1508 | * again | |
1509 | */ | |
1510 | return -ENOENT; | |
1511 | } | |
1512 | lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; | |
1513 | goto no_open; | |
1514 | } | |
1515 | ||
1516 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1517 | return -ENAMETOOLONG; | |
1518 | ||
1519 | if (open_flags & O_CREAT) { | |
1520 | attr.ia_valid |= ATTR_MODE; | |
1521 | attr.ia_mode = mode & ~current_umask(); | |
1522 | } | |
1523 | if (open_flags & O_TRUNC) { | |
1524 | attr.ia_valid |= ATTR_SIZE; | |
1525 | attr.ia_size = 0; | |
1526 | } | |
1527 | ||
1528 | ctx = create_nfs_open_context(dentry, open_flags); | |
1529 | err = PTR_ERR(ctx); | |
1530 | if (IS_ERR(ctx)) | |
1531 | goto out; | |
1532 | ||
1533 | trace_nfs_atomic_open_enter(dir, ctx, open_flags); | |
1534 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); | |
1535 | if (IS_ERR(inode)) { | |
1536 | err = PTR_ERR(inode); | |
1537 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); | |
1538 | put_nfs_open_context(ctx); | |
1539 | switch (err) { | |
1540 | case -ENOENT: | |
1541 | d_drop(dentry); | |
1542 | d_add(dentry, NULL); | |
1543 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1544 | break; | |
1545 | case -EISDIR: | |
1546 | case -ENOTDIR: | |
1547 | goto no_open; | |
1548 | case -ELOOP: | |
1549 | if (!(open_flags & O_NOFOLLOW)) | |
1550 | goto no_open; | |
1551 | break; | |
1552 | /* case -EINVAL: */ | |
1553 | default: | |
1554 | break; | |
1555 | } | |
1556 | goto out; | |
1557 | } | |
1558 | ||
1559 | err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); | |
1560 | trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); | |
1561 | put_nfs_open_context(ctx); | |
1562 | out: | |
1563 | return err; | |
1564 | ||
1565 | no_open: | |
1566 | res = nfs_lookup(dir, dentry, lookup_flags); | |
1567 | err = PTR_ERR(res); | |
1568 | if (IS_ERR(res)) | |
1569 | goto out; | |
1570 | ||
1571 | return finish_no_open(file, res); | |
1572 | } | |
1573 | EXPORT_SYMBOL_GPL(nfs_atomic_open); | |
1574 | ||
1575 | static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) | |
1576 | { | |
1577 | struct inode *inode; | |
1578 | int ret = 0; | |
1579 | ||
1580 | if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) | |
1581 | goto no_open; | |
1582 | if (d_mountpoint(dentry)) | |
1583 | goto no_open; | |
1584 | if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) | |
1585 | goto no_open; | |
1586 | ||
1587 | inode = d_inode(dentry); | |
1588 | ||
1589 | /* We can't create new files in nfs_open_revalidate(), so we | |
1590 | * optimize away revalidation of negative dentries. | |
1591 | */ | |
1592 | if (inode == NULL) { | |
1593 | struct dentry *parent; | |
1594 | struct inode *dir; | |
1595 | ||
1596 | if (flags & LOOKUP_RCU) { | |
1597 | parent = ACCESS_ONCE(dentry->d_parent); | |
1598 | dir = d_inode_rcu(parent); | |
1599 | if (!dir) | |
1600 | return -ECHILD; | |
1601 | } else { | |
1602 | parent = dget_parent(dentry); | |
1603 | dir = d_inode(parent); | |
1604 | } | |
1605 | if (!nfs_neg_need_reval(dir, dentry, flags)) | |
1606 | ret = 1; | |
1607 | else if (flags & LOOKUP_RCU) | |
1608 | ret = -ECHILD; | |
1609 | if (!(flags & LOOKUP_RCU)) | |
1610 | dput(parent); | |
1611 | else if (parent != ACCESS_ONCE(dentry->d_parent)) | |
1612 | return -ECHILD; | |
1613 | goto out; | |
1614 | } | |
1615 | ||
1616 | /* NFS only supports OPEN on regular files */ | |
1617 | if (!S_ISREG(inode->i_mode)) | |
1618 | goto no_open; | |
1619 | /* We cannot do exclusive creation on a positive dentry */ | |
1620 | if (flags & LOOKUP_EXCL) | |
1621 | goto no_open; | |
1622 | ||
1623 | /* Let f_op->open() actually open (and revalidate) the file */ | |
1624 | ret = 1; | |
1625 | ||
1626 | out: | |
1627 | return ret; | |
1628 | ||
1629 | no_open: | |
1630 | return nfs_lookup_revalidate(dentry, flags); | |
1631 | } | |
1632 | ||
1633 | #endif /* CONFIG_NFSV4 */ | |
1634 | ||
1635 | /* | |
1636 | * Code common to create, mkdir, and mknod. | |
1637 | */ | |
1638 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1639 | struct nfs_fattr *fattr, | |
1640 | struct nfs4_label *label) | |
1641 | { | |
1642 | struct dentry *parent = dget_parent(dentry); | |
1643 | struct inode *dir = d_inode(parent); | |
1644 | struct inode *inode; | |
1645 | int error = -EACCES; | |
1646 | ||
1647 | d_drop(dentry); | |
1648 | ||
1649 | /* We may have been initialized further down */ | |
1650 | if (d_really_is_positive(dentry)) | |
1651 | goto out; | |
1652 | if (fhandle->size == 0) { | |
1653 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); | |
1654 | if (error) | |
1655 | goto out_error; | |
1656 | } | |
1657 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1658 | if (!(fattr->valid & NFS_ATTR_FATTR)) { | |
1659 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
1660 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); | |
1661 | if (error < 0) | |
1662 | goto out_error; | |
1663 | } | |
1664 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); | |
1665 | error = PTR_ERR(inode); | |
1666 | if (IS_ERR(inode)) | |
1667 | goto out_error; | |
1668 | d_add(dentry, inode); | |
1669 | out: | |
1670 | dput(parent); | |
1671 | return 0; | |
1672 | out_error: | |
1673 | nfs_mark_for_revalidate(dir); | |
1674 | dput(parent); | |
1675 | return error; | |
1676 | } | |
1677 | EXPORT_SYMBOL_GPL(nfs_instantiate); | |
1678 | ||
1679 | /* | |
1680 | * Following a failed create operation, we drop the dentry rather | |
1681 | * than retain a negative dentry. This avoids a problem in the event | |
1682 | * that the operation succeeded on the server, but an error in the | |
1683 | * reply path made it appear to have failed. | |
1684 | */ | |
1685 | int nfs_create(struct inode *dir, struct dentry *dentry, | |
1686 | umode_t mode, bool excl) | |
1687 | { | |
1688 | struct iattr attr; | |
1689 | int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; | |
1690 | int error; | |
1691 | ||
1692 | dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", | |
1693 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1694 | ||
1695 | attr.ia_mode = mode; | |
1696 | attr.ia_valid = ATTR_MODE; | |
1697 | ||
1698 | trace_nfs_create_enter(dir, dentry, open_flags); | |
1699 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); | |
1700 | trace_nfs_create_exit(dir, dentry, open_flags, error); | |
1701 | if (error != 0) | |
1702 | goto out_err; | |
1703 | return 0; | |
1704 | out_err: | |
1705 | d_drop(dentry); | |
1706 | return error; | |
1707 | } | |
1708 | EXPORT_SYMBOL_GPL(nfs_create); | |
1709 | ||
1710 | /* | |
1711 | * See comments for nfs_proc_create regarding failed operations. | |
1712 | */ | |
1713 | int | |
1714 | nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) | |
1715 | { | |
1716 | struct iattr attr; | |
1717 | int status; | |
1718 | ||
1719 | dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", | |
1720 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1721 | ||
1722 | attr.ia_mode = mode; | |
1723 | attr.ia_valid = ATTR_MODE; | |
1724 | ||
1725 | trace_nfs_mknod_enter(dir, dentry); | |
1726 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); | |
1727 | trace_nfs_mknod_exit(dir, dentry, status); | |
1728 | if (status != 0) | |
1729 | goto out_err; | |
1730 | return 0; | |
1731 | out_err: | |
1732 | d_drop(dentry); | |
1733 | return status; | |
1734 | } | |
1735 | EXPORT_SYMBOL_GPL(nfs_mknod); | |
1736 | ||
1737 | /* | |
1738 | * See comments for nfs_proc_create regarding failed operations. | |
1739 | */ | |
1740 | int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) | |
1741 | { | |
1742 | struct iattr attr; | |
1743 | int error; | |
1744 | ||
1745 | dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", | |
1746 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1747 | ||
1748 | attr.ia_valid = ATTR_MODE; | |
1749 | attr.ia_mode = mode | S_IFDIR; | |
1750 | ||
1751 | trace_nfs_mkdir_enter(dir, dentry); | |
1752 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); | |
1753 | trace_nfs_mkdir_exit(dir, dentry, error); | |
1754 | if (error != 0) | |
1755 | goto out_err; | |
1756 | return 0; | |
1757 | out_err: | |
1758 | d_drop(dentry); | |
1759 | return error; | |
1760 | } | |
1761 | EXPORT_SYMBOL_GPL(nfs_mkdir); | |
1762 | ||
1763 | static void nfs_dentry_handle_enoent(struct dentry *dentry) | |
1764 | { | |
1765 | if (simple_positive(dentry)) | |
1766 | d_delete(dentry); | |
1767 | } | |
1768 | ||
1769 | int nfs_rmdir(struct inode *dir, struct dentry *dentry) | |
1770 | { | |
1771 | int error; | |
1772 | ||
1773 | dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", | |
1774 | dir->i_sb->s_id, dir->i_ino, dentry); | |
1775 | ||
1776 | trace_nfs_rmdir_enter(dir, dentry); | |
1777 | if (d_really_is_positive(dentry)) { | |
1778 | down_write(&NFS_I(d_inode(dentry))->rmdir_sem); | |
1779 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1780 | /* Ensure the VFS deletes this inode */ | |
1781 | switch (error) { | |
1782 | case 0: | |
1783 | clear_nlink(d_inode(dentry)); | |
1784 | break; | |
1785 | case -ENOENT: | |
1786 | nfs_dentry_handle_enoent(dentry); | |
1787 | } | |
1788 | up_write(&NFS_I(d_inode(dentry))->rmdir_sem); | |
1789 | } else | |
1790 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1791 | trace_nfs_rmdir_exit(dir, dentry, error); | |
1792 | ||
1793 | return error; | |
1794 | } | |
1795 | EXPORT_SYMBOL_GPL(nfs_rmdir); | |
1796 | ||
1797 | /* | |
1798 | * Remove a file after making sure there are no pending writes, | |
1799 | * and after checking that the file has only one user. | |
1800 | * | |
1801 | * We invalidate the attribute cache and free the inode prior to the operation | |
1802 | * to avoid possible races if the server reuses the inode. | |
1803 | */ | |
1804 | static int nfs_safe_remove(struct dentry *dentry) | |
1805 | { | |
1806 | struct inode *dir = d_inode(dentry->d_parent); | |
1807 | struct inode *inode = d_inode(dentry); | |
1808 | int error = -EBUSY; | |
1809 | ||
1810 | dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); | |
1811 | ||
1812 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1813 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1814 | error = 0; | |
1815 | goto out; | |
1816 | } | |
1817 | ||
1818 | trace_nfs_remove_enter(dir, dentry); | |
1819 | if (inode != NULL) { | |
1820 | NFS_PROTO(inode)->return_delegation(inode); | |
1821 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
1822 | if (error == 0) | |
1823 | nfs_drop_nlink(inode); | |
1824 | } else | |
1825 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
1826 | if (error == -ENOENT) | |
1827 | nfs_dentry_handle_enoent(dentry); | |
1828 | trace_nfs_remove_exit(dir, dentry, error); | |
1829 | out: | |
1830 | return error; | |
1831 | } | |
1832 | ||
1833 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1834 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1835 | * | |
1836 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1837 | */ | |
1838 | int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1839 | { | |
1840 | int error; | |
1841 | int need_rehash = 0; | |
1842 | ||
1843 | dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, | |
1844 | dir->i_ino, dentry); | |
1845 | ||
1846 | trace_nfs_unlink_enter(dir, dentry); | |
1847 | spin_lock(&dentry->d_lock); | |
1848 | if (d_count(dentry) > 1) { | |
1849 | spin_unlock(&dentry->d_lock); | |
1850 | /* Start asynchronous writeout of the inode */ | |
1851 | write_inode_now(d_inode(dentry), 0); | |
1852 | error = nfs_sillyrename(dir, dentry); | |
1853 | goto out; | |
1854 | } | |
1855 | if (!d_unhashed(dentry)) { | |
1856 | __d_drop(dentry); | |
1857 | need_rehash = 1; | |
1858 | } | |
1859 | spin_unlock(&dentry->d_lock); | |
1860 | error = nfs_safe_remove(dentry); | |
1861 | if (!error || error == -ENOENT) { | |
1862 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1863 | } else if (need_rehash) | |
1864 | d_rehash(dentry); | |
1865 | out: | |
1866 | trace_nfs_unlink_exit(dir, dentry, error); | |
1867 | return error; | |
1868 | } | |
1869 | EXPORT_SYMBOL_GPL(nfs_unlink); | |
1870 | ||
1871 | /* | |
1872 | * To create a symbolic link, most file systems instantiate a new inode, | |
1873 | * add a page to it containing the path, then write it out to the disk | |
1874 | * using prepare_write/commit_write. | |
1875 | * | |
1876 | * Unfortunately the NFS client can't create the in-core inode first | |
1877 | * because it needs a file handle to create an in-core inode (see | |
1878 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1879 | * symlink request has completed on the server. | |
1880 | * | |
1881 | * So instead we allocate a raw page, copy the symname into it, then do | |
1882 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1883 | * now have a new file handle and can instantiate an in-core NFS inode | |
1884 | * and move the raw page into its mapping. | |
1885 | */ | |
1886 | int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1887 | { | |
1888 | struct page *page; | |
1889 | char *kaddr; | |
1890 | struct iattr attr; | |
1891 | unsigned int pathlen = strlen(symname); | |
1892 | int error; | |
1893 | ||
1894 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, | |
1895 | dir->i_ino, dentry, symname); | |
1896 | ||
1897 | if (pathlen > PAGE_SIZE) | |
1898 | return -ENAMETOOLONG; | |
1899 | ||
1900 | attr.ia_mode = S_IFLNK | S_IRWXUGO; | |
1901 | attr.ia_valid = ATTR_MODE; | |
1902 | ||
1903 | page = alloc_page(GFP_USER); | |
1904 | if (!page) | |
1905 | return -ENOMEM; | |
1906 | ||
1907 | kaddr = page_address(page); | |
1908 | memcpy(kaddr, symname, pathlen); | |
1909 | if (pathlen < PAGE_SIZE) | |
1910 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
1911 | ||
1912 | trace_nfs_symlink_enter(dir, dentry); | |
1913 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); | |
1914 | trace_nfs_symlink_exit(dir, dentry, error); | |
1915 | if (error != 0) { | |
1916 | dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", | |
1917 | dir->i_sb->s_id, dir->i_ino, | |
1918 | dentry, symname, error); | |
1919 | d_drop(dentry); | |
1920 | __free_page(page); | |
1921 | return error; | |
1922 | } | |
1923 | ||
1924 | /* | |
1925 | * No big deal if we can't add this page to the page cache here. | |
1926 | * READLINK will get the missing page from the server if needed. | |
1927 | */ | |
1928 | if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, | |
1929 | GFP_KERNEL)) { | |
1930 | SetPageUptodate(page); | |
1931 | unlock_page(page); | |
1932 | /* | |
1933 | * add_to_page_cache_lru() grabs an extra page refcount. | |
1934 | * Drop it here to avoid leaking this page later. | |
1935 | */ | |
1936 | put_page(page); | |
1937 | } else | |
1938 | __free_page(page); | |
1939 | ||
1940 | return 0; | |
1941 | } | |
1942 | EXPORT_SYMBOL_GPL(nfs_symlink); | |
1943 | ||
1944 | int | |
1945 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1946 | { | |
1947 | struct inode *inode = d_inode(old_dentry); | |
1948 | int error; | |
1949 | ||
1950 | dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", | |
1951 | old_dentry, dentry); | |
1952 | ||
1953 | trace_nfs_link_enter(inode, dir, dentry); | |
1954 | NFS_PROTO(inode)->return_delegation(inode); | |
1955 | ||
1956 | d_drop(dentry); | |
1957 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); | |
1958 | if (error == 0) { | |
1959 | ihold(inode); | |
1960 | d_add(dentry, inode); | |
1961 | } | |
1962 | trace_nfs_link_exit(inode, dir, dentry, error); | |
1963 | return error; | |
1964 | } | |
1965 | EXPORT_SYMBOL_GPL(nfs_link); | |
1966 | ||
1967 | /* | |
1968 | * RENAME | |
1969 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1970 | * different file handle for the same inode after a rename (e.g. when | |
1971 | * moving to a different directory). A fail-safe method to do so would | |
1972 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1973 | * rename the old file using the sillyrename stuff. This way, the original | |
1974 | * file in old_dir will go away when the last process iput()s the inode. | |
1975 | * | |
1976 | * FIXED. | |
1977 | * | |
1978 | * It actually works quite well. One needs to have the possibility for | |
1979 | * at least one ".nfs..." file in each directory the file ever gets | |
1980 | * moved or linked to which happens automagically with the new | |
1981 | * implementation that only depends on the dcache stuff instead of | |
1982 | * using the inode layer | |
1983 | * | |
1984 | * Unfortunately, things are a little more complicated than indicated | |
1985 | * above. For a cross-directory move, we want to make sure we can get | |
1986 | * rid of the old inode after the operation. This means there must be | |
1987 | * no pending writes (if it's a file), and the use count must be 1. | |
1988 | * If these conditions are met, we can drop the dentries before doing | |
1989 | * the rename. | |
1990 | */ | |
1991 | int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1992 | struct inode *new_dir, struct dentry *new_dentry) | |
1993 | { | |
1994 | struct inode *old_inode = d_inode(old_dentry); | |
1995 | struct inode *new_inode = d_inode(new_dentry); | |
1996 | struct dentry *dentry = NULL, *rehash = NULL; | |
1997 | struct rpc_task *task; | |
1998 | int error = -EBUSY; | |
1999 | ||
2000 | dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", | |
2001 | old_dentry, new_dentry, | |
2002 | d_count(new_dentry)); | |
2003 | ||
2004 | trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); | |
2005 | /* | |
2006 | * For non-directories, check whether the target is busy and if so, | |
2007 | * make a copy of the dentry and then do a silly-rename. If the | |
2008 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
2009 | * the new target. | |
2010 | */ | |
2011 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { | |
2012 | /* | |
2013 | * To prevent any new references to the target during the | |
2014 | * rename, we unhash the dentry in advance. | |
2015 | */ | |
2016 | if (!d_unhashed(new_dentry)) { | |
2017 | d_drop(new_dentry); | |
2018 | rehash = new_dentry; | |
2019 | } | |
2020 | ||
2021 | if (d_count(new_dentry) > 2) { | |
2022 | int err; | |
2023 | ||
2024 | /* copy the target dentry's name */ | |
2025 | dentry = d_alloc(new_dentry->d_parent, | |
2026 | &new_dentry->d_name); | |
2027 | if (!dentry) | |
2028 | goto out; | |
2029 | ||
2030 | /* silly-rename the existing target ... */ | |
2031 | err = nfs_sillyrename(new_dir, new_dentry); | |
2032 | if (err) | |
2033 | goto out; | |
2034 | ||
2035 | new_dentry = dentry; | |
2036 | rehash = NULL; | |
2037 | new_inode = NULL; | |
2038 | } | |
2039 | } | |
2040 | ||
2041 | NFS_PROTO(old_inode)->return_delegation(old_inode); | |
2042 | if (new_inode != NULL) | |
2043 | NFS_PROTO(new_inode)->return_delegation(new_inode); | |
2044 | ||
2045 | task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); | |
2046 | if (IS_ERR(task)) { | |
2047 | error = PTR_ERR(task); | |
2048 | goto out; | |
2049 | } | |
2050 | ||
2051 | error = rpc_wait_for_completion_task(task); | |
2052 | if (error == 0) | |
2053 | error = task->tk_status; | |
2054 | rpc_put_task(task); | |
2055 | nfs_mark_for_revalidate(old_inode); | |
2056 | out: | |
2057 | if (rehash) | |
2058 | d_rehash(rehash); | |
2059 | trace_nfs_rename_exit(old_dir, old_dentry, | |
2060 | new_dir, new_dentry, error); | |
2061 | if (!error) { | |
2062 | if (new_inode != NULL) | |
2063 | nfs_drop_nlink(new_inode); | |
2064 | d_move(old_dentry, new_dentry); | |
2065 | nfs_set_verifier(new_dentry, | |
2066 | nfs_save_change_attribute(new_dir)); | |
2067 | } else if (error == -ENOENT) | |
2068 | nfs_dentry_handle_enoent(old_dentry); | |
2069 | ||
2070 | /* new dentry created? */ | |
2071 | if (dentry) | |
2072 | dput(dentry); | |
2073 | return error; | |
2074 | } | |
2075 | EXPORT_SYMBOL_GPL(nfs_rename); | |
2076 | ||
2077 | static DEFINE_SPINLOCK(nfs_access_lru_lock); | |
2078 | static LIST_HEAD(nfs_access_lru_list); | |
2079 | static atomic_long_t nfs_access_nr_entries; | |
2080 | ||
2081 | static unsigned long nfs_access_max_cachesize = ULONG_MAX; | |
2082 | module_param(nfs_access_max_cachesize, ulong, 0644); | |
2083 | MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); | |
2084 | ||
2085 | static void nfs_access_free_entry(struct nfs_access_entry *entry) | |
2086 | { | |
2087 | put_rpccred(entry->cred); | |
2088 | kfree_rcu(entry, rcu_head); | |
2089 | smp_mb__before_atomic(); | |
2090 | atomic_long_dec(&nfs_access_nr_entries); | |
2091 | smp_mb__after_atomic(); | |
2092 | } | |
2093 | ||
2094 | static void nfs_access_free_list(struct list_head *head) | |
2095 | { | |
2096 | struct nfs_access_entry *cache; | |
2097 | ||
2098 | while (!list_empty(head)) { | |
2099 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
2100 | list_del(&cache->lru); | |
2101 | nfs_access_free_entry(cache); | |
2102 | } | |
2103 | } | |
2104 | ||
2105 | static unsigned long | |
2106 | nfs_do_access_cache_scan(unsigned int nr_to_scan) | |
2107 | { | |
2108 | LIST_HEAD(head); | |
2109 | struct nfs_inode *nfsi, *next; | |
2110 | struct nfs_access_entry *cache; | |
2111 | long freed = 0; | |
2112 | ||
2113 | spin_lock(&nfs_access_lru_lock); | |
2114 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { | |
2115 | struct inode *inode; | |
2116 | ||
2117 | if (nr_to_scan-- == 0) | |
2118 | break; | |
2119 | inode = &nfsi->vfs_inode; | |
2120 | spin_lock(&inode->i_lock); | |
2121 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
2122 | goto remove_lru_entry; | |
2123 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
2124 | struct nfs_access_entry, lru); | |
2125 | list_move(&cache->lru, &head); | |
2126 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
2127 | freed++; | |
2128 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
2129 | list_move_tail(&nfsi->access_cache_inode_lru, | |
2130 | &nfs_access_lru_list); | |
2131 | else { | |
2132 | remove_lru_entry: | |
2133 | list_del_init(&nfsi->access_cache_inode_lru); | |
2134 | smp_mb__before_atomic(); | |
2135 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); | |
2136 | smp_mb__after_atomic(); | |
2137 | } | |
2138 | spin_unlock(&inode->i_lock); | |
2139 | } | |
2140 | spin_unlock(&nfs_access_lru_lock); | |
2141 | nfs_access_free_list(&head); | |
2142 | return freed; | |
2143 | } | |
2144 | ||
2145 | unsigned long | |
2146 | nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
2147 | { | |
2148 | int nr_to_scan = sc->nr_to_scan; | |
2149 | gfp_t gfp_mask = sc->gfp_mask; | |
2150 | ||
2151 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) | |
2152 | return SHRINK_STOP; | |
2153 | return nfs_do_access_cache_scan(nr_to_scan); | |
2154 | } | |
2155 | ||
2156 | ||
2157 | unsigned long | |
2158 | nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
2159 | { | |
2160 | return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); | |
2161 | } | |
2162 | ||
2163 | static void | |
2164 | nfs_access_cache_enforce_limit(void) | |
2165 | { | |
2166 | long nr_entries = atomic_long_read(&nfs_access_nr_entries); | |
2167 | unsigned long diff; | |
2168 | unsigned int nr_to_scan; | |
2169 | ||
2170 | if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) | |
2171 | return; | |
2172 | nr_to_scan = 100; | |
2173 | diff = nr_entries - nfs_access_max_cachesize; | |
2174 | if (diff < nr_to_scan) | |
2175 | nr_to_scan = diff; | |
2176 | nfs_do_access_cache_scan(nr_to_scan); | |
2177 | } | |
2178 | ||
2179 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) | |
2180 | { | |
2181 | struct rb_root *root_node = &nfsi->access_cache; | |
2182 | struct rb_node *n; | |
2183 | struct nfs_access_entry *entry; | |
2184 | ||
2185 | /* Unhook entries from the cache */ | |
2186 | while ((n = rb_first(root_node)) != NULL) { | |
2187 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2188 | rb_erase(n, root_node); | |
2189 | list_move(&entry->lru, head); | |
2190 | } | |
2191 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
2192 | } | |
2193 | ||
2194 | void nfs_access_zap_cache(struct inode *inode) | |
2195 | { | |
2196 | LIST_HEAD(head); | |
2197 | ||
2198 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2199 | return; | |
2200 | /* Remove from global LRU init */ | |
2201 | spin_lock(&nfs_access_lru_lock); | |
2202 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
2203 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); | |
2204 | ||
2205 | spin_lock(&inode->i_lock); | |
2206 | __nfs_access_zap_cache(NFS_I(inode), &head); | |
2207 | spin_unlock(&inode->i_lock); | |
2208 | spin_unlock(&nfs_access_lru_lock); | |
2209 | nfs_access_free_list(&head); | |
2210 | } | |
2211 | EXPORT_SYMBOL_GPL(nfs_access_zap_cache); | |
2212 | ||
2213 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) | |
2214 | { | |
2215 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2216 | struct nfs_access_entry *entry; | |
2217 | ||
2218 | while (n != NULL) { | |
2219 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2220 | ||
2221 | if (cred < entry->cred) | |
2222 | n = n->rb_left; | |
2223 | else if (cred > entry->cred) | |
2224 | n = n->rb_right; | |
2225 | else | |
2226 | return entry; | |
2227 | } | |
2228 | return NULL; | |
2229 | } | |
2230 | ||
2231 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) | |
2232 | { | |
2233 | struct nfs_inode *nfsi = NFS_I(inode); | |
2234 | struct nfs_access_entry *cache; | |
2235 | int err = -ENOENT; | |
2236 | ||
2237 | spin_lock(&inode->i_lock); | |
2238 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
2239 | goto out_zap; | |
2240 | cache = nfs_access_search_rbtree(inode, cred); | |
2241 | if (cache == NULL) | |
2242 | goto out; | |
2243 | if (!nfs_have_delegated_attributes(inode) && | |
2244 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) | |
2245 | goto out_stale; | |
2246 | res->jiffies = cache->jiffies; | |
2247 | res->cred = cache->cred; | |
2248 | res->mask = cache->mask; | |
2249 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); | |
2250 | err = 0; | |
2251 | out: | |
2252 | spin_unlock(&inode->i_lock); | |
2253 | return err; | |
2254 | out_stale: | |
2255 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
2256 | list_del(&cache->lru); | |
2257 | spin_unlock(&inode->i_lock); | |
2258 | nfs_access_free_entry(cache); | |
2259 | return -ENOENT; | |
2260 | out_zap: | |
2261 | spin_unlock(&inode->i_lock); | |
2262 | nfs_access_zap_cache(inode); | |
2263 | return -ENOENT; | |
2264 | } | |
2265 | ||
2266 | static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) | |
2267 | { | |
2268 | /* Only check the most recently returned cache entry, | |
2269 | * but do it without locking. | |
2270 | */ | |
2271 | struct nfs_inode *nfsi = NFS_I(inode); | |
2272 | struct nfs_access_entry *cache; | |
2273 | int err = -ECHILD; | |
2274 | struct list_head *lh; | |
2275 | ||
2276 | rcu_read_lock(); | |
2277 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) | |
2278 | goto out; | |
2279 | lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); | |
2280 | cache = list_entry(lh, struct nfs_access_entry, lru); | |
2281 | if (lh == &nfsi->access_cache_entry_lru || | |
2282 | cred != cache->cred) | |
2283 | cache = NULL; | |
2284 | if (cache == NULL) | |
2285 | goto out; | |
2286 | if (!nfs_have_delegated_attributes(inode) && | |
2287 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) | |
2288 | goto out; | |
2289 | res->jiffies = cache->jiffies; | |
2290 | res->cred = cache->cred; | |
2291 | res->mask = cache->mask; | |
2292 | err = 0; | |
2293 | out: | |
2294 | rcu_read_unlock(); | |
2295 | return err; | |
2296 | } | |
2297 | ||
2298 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2299 | { | |
2300 | struct nfs_inode *nfsi = NFS_I(inode); | |
2301 | struct rb_root *root_node = &nfsi->access_cache; | |
2302 | struct rb_node **p = &root_node->rb_node; | |
2303 | struct rb_node *parent = NULL; | |
2304 | struct nfs_access_entry *entry; | |
2305 | ||
2306 | spin_lock(&inode->i_lock); | |
2307 | while (*p != NULL) { | |
2308 | parent = *p; | |
2309 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2310 | ||
2311 | if (set->cred < entry->cred) | |
2312 | p = &parent->rb_left; | |
2313 | else if (set->cred > entry->cred) | |
2314 | p = &parent->rb_right; | |
2315 | else | |
2316 | goto found; | |
2317 | } | |
2318 | rb_link_node(&set->rb_node, parent, p); | |
2319 | rb_insert_color(&set->rb_node, root_node); | |
2320 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); | |
2321 | spin_unlock(&inode->i_lock); | |
2322 | return; | |
2323 | found: | |
2324 | rb_replace_node(parent, &set->rb_node, root_node); | |
2325 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); | |
2326 | list_del(&entry->lru); | |
2327 | spin_unlock(&inode->i_lock); | |
2328 | nfs_access_free_entry(entry); | |
2329 | } | |
2330 | ||
2331 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) | |
2332 | { | |
2333 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2334 | if (cache == NULL) | |
2335 | return; | |
2336 | RB_CLEAR_NODE(&cache->rb_node); | |
2337 | cache->jiffies = set->jiffies; | |
2338 | cache->cred = get_rpccred(set->cred); | |
2339 | cache->mask = set->mask; | |
2340 | ||
2341 | /* The above field assignments must be visible | |
2342 | * before this item appears on the lru. We cannot easily | |
2343 | * use rcu_assign_pointer, so just force the memory barrier. | |
2344 | */ | |
2345 | smp_wmb(); | |
2346 | nfs_access_add_rbtree(inode, cache); | |
2347 | ||
2348 | /* Update accounting */ | |
2349 | smp_mb__before_atomic(); | |
2350 | atomic_long_inc(&nfs_access_nr_entries); | |
2351 | smp_mb__after_atomic(); | |
2352 | ||
2353 | /* Add inode to global LRU list */ | |
2354 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { | |
2355 | spin_lock(&nfs_access_lru_lock); | |
2356 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
2357 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2358 | &nfs_access_lru_list); | |
2359 | spin_unlock(&nfs_access_lru_lock); | |
2360 | } | |
2361 | nfs_access_cache_enforce_limit(); | |
2362 | } | |
2363 | EXPORT_SYMBOL_GPL(nfs_access_add_cache); | |
2364 | ||
2365 | void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) | |
2366 | { | |
2367 | entry->mask = 0; | |
2368 | if (access_result & NFS4_ACCESS_READ) | |
2369 | entry->mask |= MAY_READ; | |
2370 | if (access_result & | |
2371 | (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) | |
2372 | entry->mask |= MAY_WRITE; | |
2373 | if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) | |
2374 | entry->mask |= MAY_EXEC; | |
2375 | } | |
2376 | EXPORT_SYMBOL_GPL(nfs_access_set_mask); | |
2377 | ||
2378 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2379 | { | |
2380 | struct nfs_access_entry cache; | |
2381 | int status; | |
2382 | ||
2383 | trace_nfs_access_enter(inode); | |
2384 | ||
2385 | status = nfs_access_get_cached_rcu(inode, cred, &cache); | |
2386 | if (status != 0) | |
2387 | status = nfs_access_get_cached(inode, cred, &cache); | |
2388 | if (status == 0) | |
2389 | goto out_cached; | |
2390 | ||
2391 | status = -ECHILD; | |
2392 | if (mask & MAY_NOT_BLOCK) | |
2393 | goto out; | |
2394 | ||
2395 | /* Be clever: ask server to check for all possible rights */ | |
2396 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2397 | cache.cred = cred; | |
2398 | cache.jiffies = jiffies; | |
2399 | status = NFS_PROTO(inode)->access(inode, &cache); | |
2400 | if (status != 0) { | |
2401 | if (status == -ESTALE) { | |
2402 | nfs_zap_caches(inode); | |
2403 | if (!S_ISDIR(inode->i_mode)) | |
2404 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2405 | } | |
2406 | goto out; | |
2407 | } | |
2408 | nfs_access_add_cache(inode, &cache); | |
2409 | out_cached: | |
2410 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) | |
2411 | status = -EACCES; | |
2412 | out: | |
2413 | trace_nfs_access_exit(inode, status); | |
2414 | return status; | |
2415 | } | |
2416 | ||
2417 | static int nfs_open_permission_mask(int openflags) | |
2418 | { | |
2419 | int mask = 0; | |
2420 | ||
2421 | if (openflags & __FMODE_EXEC) { | |
2422 | /* ONLY check exec rights */ | |
2423 | mask = MAY_EXEC; | |
2424 | } else { | |
2425 | if ((openflags & O_ACCMODE) != O_WRONLY) | |
2426 | mask |= MAY_READ; | |
2427 | if ((openflags & O_ACCMODE) != O_RDONLY) | |
2428 | mask |= MAY_WRITE; | |
2429 | } | |
2430 | ||
2431 | return mask; | |
2432 | } | |
2433 | ||
2434 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2435 | { | |
2436 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2437 | } | |
2438 | EXPORT_SYMBOL_GPL(nfs_may_open); | |
2439 | ||
2440 | static int nfs_execute_ok(struct inode *inode, int mask) | |
2441 | { | |
2442 | struct nfs_server *server = NFS_SERVER(inode); | |
2443 | int ret; | |
2444 | ||
2445 | if (mask & MAY_NOT_BLOCK) | |
2446 | ret = nfs_revalidate_inode_rcu(server, inode); | |
2447 | else | |
2448 | ret = nfs_revalidate_inode(server, inode); | |
2449 | if (ret == 0 && !execute_ok(inode)) | |
2450 | ret = -EACCES; | |
2451 | return ret; | |
2452 | } | |
2453 | ||
2454 | int nfs_permission(struct inode *inode, int mask) | |
2455 | { | |
2456 | struct rpc_cred *cred; | |
2457 | int res = 0; | |
2458 | ||
2459 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); | |
2460 | ||
2461 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) | |
2462 | goto out; | |
2463 | /* Is this sys_access() ? */ | |
2464 | if (mask & (MAY_ACCESS | MAY_CHDIR)) | |
2465 | goto force_lookup; | |
2466 | ||
2467 | switch (inode->i_mode & S_IFMT) { | |
2468 | case S_IFLNK: | |
2469 | goto out; | |
2470 | case S_IFREG: | |
2471 | if ((mask & MAY_OPEN) && | |
2472 | nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) | |
2473 | return 0; | |
2474 | break; | |
2475 | case S_IFDIR: | |
2476 | /* | |
2477 | * Optimize away all write operations, since the server | |
2478 | * will check permissions when we perform the op. | |
2479 | */ | |
2480 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2481 | goto out; | |
2482 | } | |
2483 | ||
2484 | force_lookup: | |
2485 | if (!NFS_PROTO(inode)->access) | |
2486 | goto out_notsup; | |
2487 | ||
2488 | /* Always try fast lookups first */ | |
2489 | rcu_read_lock(); | |
2490 | cred = rpc_lookup_cred_nonblock(); | |
2491 | if (!IS_ERR(cred)) | |
2492 | res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); | |
2493 | else | |
2494 | res = PTR_ERR(cred); | |
2495 | rcu_read_unlock(); | |
2496 | if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { | |
2497 | /* Fast lookup failed, try the slow way */ | |
2498 | cred = rpc_lookup_cred(); | |
2499 | if (!IS_ERR(cred)) { | |
2500 | res = nfs_do_access(inode, cred, mask); | |
2501 | put_rpccred(cred); | |
2502 | } else | |
2503 | res = PTR_ERR(cred); | |
2504 | } | |
2505 | out: | |
2506 | if (!res && (mask & MAY_EXEC)) | |
2507 | res = nfs_execute_ok(inode, mask); | |
2508 | ||
2509 | dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", | |
2510 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
2511 | return res; | |
2512 | out_notsup: | |
2513 | if (mask & MAY_NOT_BLOCK) | |
2514 | return -ECHILD; | |
2515 | ||
2516 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2517 | if (res == 0) | |
2518 | res = generic_permission(inode, mask); | |
2519 | goto out; | |
2520 | } | |
2521 | EXPORT_SYMBOL_GPL(nfs_permission); | |
2522 | ||
2523 | /* | |
2524 | * Local variables: | |
2525 | * version-control: t | |
2526 | * kept-new-versions: 5 | |
2527 | * End: | |
2528 | */ |