1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/fs/nfs/dir.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * nfs directory handling functions
8 *
9 * 10 Apr 1996 Added silly rename for unlink --okir
10 * 28 Sep 1996 Improved directory cache --okir
11 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
12 * Re-implemented silly rename for unlink, newly implemented
13 * silly rename for nfs_rename() following the suggestions
14 * of Olaf Kirch (okir) found in this file.
15 * Following Linus comments on my original hack, this version
16 * depends only on the dcache stuff and doesn't touch the inode
17 * layer (iput() and friends).
18 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
19 */
20
21 #include <linux/compat.h>
22 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/errno.h>
25 #include <linux/stat.h>
26 #include <linux/fcntl.h>
27 #include <linux/string.h>
28 #include <linux/kernel.h>
29 #include <linux/slab.h>
30 #include <linux/mm.h>
31 #include <linux/sunrpc/clnt.h>
32 #include <linux/nfs_fs.h>
33 #include <linux/nfs_mount.h>
34 #include <linux/pagemap.h>
35 #include <linux/pagevec.h>
36 #include <linux/namei.h>
37 #include <linux/mount.h>
38 #include <linux/swap.h>
39 #include <linux/sched.h>
40 #include <linux/kmemleak.h>
41 #include <linux/xattr.h>
42 #include <linux/hash.h>
43
44 #include "delegation.h"
45 #include "iostat.h"
46 #include "internal.h"
47 #include "fscache.h"
48
49 #include "nfstrace.h"
50
51 /* #define NFS_DEBUG_VERBOSE 1 */
52
53 static int nfs_opendir(struct inode *, struct file *);
54 static int nfs_closedir(struct inode *, struct file *);
55 static int nfs_readdir(struct file *, struct dir_context *);
56 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
58 static void nfs_readdir_clear_array(struct folio *);
59
60 const struct file_operations nfs_dir_operations = {
61 .llseek = nfs_llseek_dir,
62 .read = generic_read_dir,
63 .iterate_shared = nfs_readdir,
64 .open = nfs_opendir,
65 .release = nfs_closedir,
66 .fsync = nfs_fsync_dir,
67 };
68
69 const struct address_space_operations nfs_dir_aops = {
70 .free_folio = nfs_readdir_clear_array,
71 };
72
73 #define NFS_INIT_DTSIZE PAGE_SIZE
74
75 static struct nfs_open_dir_context *
alloc_nfs_open_dir_context(struct inode * dir)76 alloc_nfs_open_dir_context(struct inode *dir)
77 {
78 struct nfs_inode *nfsi = NFS_I(dir);
79 struct nfs_open_dir_context *ctx;
80
81 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
82 if (ctx != NULL) {
83 ctx->attr_gencount = nfsi->attr_gencount;
84 ctx->dtsize = NFS_INIT_DTSIZE;
85 spin_lock(&dir->i_lock);
86 if (list_empty(&nfsi->open_files) &&
87 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
88 nfs_set_cache_invalid(dir,
89 NFS_INO_INVALID_DATA |
90 NFS_INO_REVAL_FORCED);
91 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
92 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
93 spin_unlock(&dir->i_lock);
94 return ctx;
95 }
96 return ERR_PTR(-ENOMEM);
97 }
98
put_nfs_open_dir_context(struct inode * dir,struct nfs_open_dir_context * ctx)99 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
100 {
101 spin_lock(&dir->i_lock);
102 list_del_rcu(&ctx->list);
103 spin_unlock(&dir->i_lock);
104 kfree_rcu(ctx, rcu_head);
105 }
106
107 /*
108 * Open file
109 */
110 static int
nfs_opendir(struct inode * inode,struct file * filp)111 nfs_opendir(struct inode *inode, struct file *filp)
112 {
113 int res = 0;
114 struct nfs_open_dir_context *ctx;
115
116 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
117
118 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
119
120 ctx = alloc_nfs_open_dir_context(inode);
121 if (IS_ERR(ctx)) {
122 res = PTR_ERR(ctx);
123 goto out;
124 }
125 filp->private_data = ctx;
126 out:
127 return res;
128 }
129
130 static int
nfs_closedir(struct inode * inode,struct file * filp)131 nfs_closedir(struct inode *inode, struct file *filp)
132 {
133 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
134 return 0;
135 }
136
137 struct nfs_cache_array_entry {
138 u64 cookie;
139 u64 ino;
140 const char *name;
141 unsigned int name_len;
142 unsigned char d_type;
143 };
144
145 struct nfs_cache_array {
146 u64 change_attr;
147 u64 last_cookie;
148 unsigned int size;
149 unsigned char folio_full : 1,
150 folio_is_eof : 1,
151 cookies_are_ordered : 1;
152 struct nfs_cache_array_entry array[];
153 };
154
155 struct nfs_readdir_descriptor {
156 struct file *file;
157 struct folio *folio;
158 struct dir_context *ctx;
159 pgoff_t folio_index;
160 pgoff_t folio_index_max;
161 u64 dir_cookie;
162 u64 last_cookie;
163 loff_t current_index;
164
165 __be32 verf[NFS_DIR_VERIFIER_SIZE];
166 unsigned long dir_verifier;
167 unsigned long timestamp;
168 unsigned long gencount;
169 unsigned long attr_gencount;
170 unsigned int cache_entry_index;
171 unsigned int buffer_fills;
172 unsigned int dtsize;
173 bool clear_cache;
174 bool plus;
175 bool eob;
176 bool eof;
177 };
178
nfs_set_dtsize(struct nfs_readdir_descriptor * desc,unsigned int sz)179 static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
180 {
181 struct nfs_server *server = NFS_SERVER(file_inode(desc->file));
182 unsigned int maxsize = server->dtsize;
183
184 if (sz > maxsize)
185 sz = maxsize;
186 if (sz < NFS_MIN_FILE_IO_SIZE)
187 sz = NFS_MIN_FILE_IO_SIZE;
188 desc->dtsize = sz;
189 }
190
nfs_shrink_dtsize(struct nfs_readdir_descriptor * desc)191 static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
192 {
193 nfs_set_dtsize(desc, desc->dtsize >> 1);
194 }
195
nfs_grow_dtsize(struct nfs_readdir_descriptor * desc)196 static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
197 {
198 nfs_set_dtsize(desc, desc->dtsize << 1);
199 }
200
nfs_readdir_folio_init_array(struct folio * folio,u64 last_cookie,u64 change_attr)201 static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie,
202 u64 change_attr)
203 {
204 struct nfs_cache_array *array;
205
206 array = kmap_local_folio(folio, 0);
207 array->change_attr = change_attr;
208 array->last_cookie = last_cookie;
209 array->size = 0;
210 array->folio_full = 0;
211 array->folio_is_eof = 0;
212 array->cookies_are_ordered = 1;
213 kunmap_local(array);
214 }
215
216 /*
217 * we are freeing strings created by nfs_add_to_readdir_array()
218 */
nfs_readdir_clear_array(struct folio * folio)219 static void nfs_readdir_clear_array(struct folio *folio)
220 {
221 struct nfs_cache_array *array;
222 unsigned int i;
223
224 array = kmap_local_folio(folio, 0);
225 for (i = 0; i < array->size; i++)
226 kfree(array->array[i].name);
227 array->size = 0;
228 kunmap_local(array);
229 }
230
nfs_readdir_folio_reinit_array(struct folio * folio,u64 last_cookie,u64 change_attr)231 static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
232 u64 change_attr)
233 {
234 nfs_readdir_clear_array(folio);
235 nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
236 }
237
238 static struct folio *
nfs_readdir_folio_array_alloc(u64 last_cookie,gfp_t gfp_flags)239 nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
240 {
241 struct folio *folio = folio_alloc(gfp_flags, 0);
242 if (folio)
243 nfs_readdir_folio_init_array(folio, last_cookie, 0);
244 return folio;
245 }
246
nfs_readdir_folio_array_free(struct folio * folio)247 static void nfs_readdir_folio_array_free(struct folio *folio)
248 {
249 if (folio) {
250 nfs_readdir_clear_array(folio);
251 folio_put(folio);
252 }
253 }
254
nfs_readdir_array_index_cookie(struct nfs_cache_array * array)255 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
256 {
257 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
258 }
259
nfs_readdir_array_set_eof(struct nfs_cache_array * array)260 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
261 {
262 array->folio_is_eof = 1;
263 array->folio_full = 1;
264 }
265
nfs_readdir_array_is_full(struct nfs_cache_array * array)266 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
267 {
268 return array->folio_full;
269 }
270
271 /*
272 * the caller is responsible for freeing qstr.name
273 * when called by nfs_readdir_add_to_array, the strings will be freed in
274 * nfs_clear_readdir_array()
275 */
nfs_readdir_copy_name(const char * name,unsigned int len)276 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
277 {
278 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
279
280 /*
281 * Avoid a kmemleak false positive. The pointer to the name is stored
282 * in a page cache page which kmemleak does not scan.
283 */
284 if (ret != NULL)
285 kmemleak_not_leak(ret);
286 return ret;
287 }
288
nfs_readdir_array_maxentries(void)289 static size_t nfs_readdir_array_maxentries(void)
290 {
291 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
292 sizeof(struct nfs_cache_array_entry);
293 }
294
295 /*
296 * Check that the next array entry lies entirely within the page bounds
297 */
nfs_readdir_array_can_expand(struct nfs_cache_array * array)298 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
299 {
300 if (array->folio_full)
301 return -ENOSPC;
302 if (array->size == nfs_readdir_array_maxentries()) {
303 array->folio_full = 1;
304 return -ENOSPC;
305 }
306 return 0;
307 }
308
nfs_readdir_folio_array_append(struct folio * folio,const struct nfs_entry * entry,u64 * cookie)309 static int nfs_readdir_folio_array_append(struct folio *folio,
310 const struct nfs_entry *entry,
311 u64 *cookie)
312 {
313 struct nfs_cache_array *array;
314 struct nfs_cache_array_entry *cache_entry;
315 const char *name;
316 int ret = -ENOMEM;
317
318 name = nfs_readdir_copy_name(entry->name, entry->len);
319
320 array = kmap_local_folio(folio, 0);
321 if (!name)
322 goto out;
323 ret = nfs_readdir_array_can_expand(array);
324 if (ret) {
325 kfree(name);
326 goto out;
327 }
328
329 cache_entry = &array->array[array->size];
330 cache_entry->cookie = array->last_cookie;
331 cache_entry->ino = entry->ino;
332 cache_entry->d_type = entry->d_type;
333 cache_entry->name_len = entry->len;
334 cache_entry->name = name;
335 array->last_cookie = entry->cookie;
336 if (array->last_cookie <= cache_entry->cookie)
337 array->cookies_are_ordered = 0;
338 array->size++;
339 if (entry->eof != 0)
340 nfs_readdir_array_set_eof(array);
341 out:
342 *cookie = array->last_cookie;
343 kunmap_local(array);
344 return ret;
345 }
346
347 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
348 /*
349 * Hash algorithm allowing content addressible access to sequences
350 * of directory cookies. Content is addressed by the value of the
351 * cookie index of the first readdir entry in a page.
352 *
353 * We select only the first 18 bits to avoid issues with excessive
354 * memory use for the page cache XArray. 18 bits should allow the caching
355 * of 262144 pages of sequences of readdir entries. Since each page holds
356 * 127 readdir entries for a typical 64-bit system, that works out to a
357 * cache of ~ 33 million entries per directory.
358 */
nfs_readdir_folio_cookie_hash(u64 cookie)359 static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
360 {
361 if (cookie == 0)
362 return 0;
363 return hash_64(cookie, 18);
364 }
365
nfs_readdir_folio_validate(struct folio * folio,u64 last_cookie,u64 change_attr)366 static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
367 u64 change_attr)
368 {
369 struct nfs_cache_array *array = kmap_local_folio(folio, 0);
370 int ret = true;
371
372 if (array->change_attr != change_attr)
373 ret = false;
374 if (nfs_readdir_array_index_cookie(array) != last_cookie)
375 ret = false;
376 kunmap_local(array);
377 return ret;
378 }
379
nfs_readdir_folio_unlock_and_put(struct folio * folio)380 static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
381 {
382 folio_unlock(folio);
383 folio_put(folio);
384 }
385
nfs_readdir_folio_init_and_validate(struct folio * folio,u64 cookie,u64 change_attr)386 static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
387 u64 change_attr)
388 {
389 if (folio_test_uptodate(folio)) {
390 if (nfs_readdir_folio_validate(folio, cookie, change_attr))
391 return;
392 nfs_readdir_clear_array(folio);
393 }
394 nfs_readdir_folio_init_array(folio, cookie, change_attr);
395 folio_mark_uptodate(folio);
396 }
397
nfs_readdir_folio_get_locked(struct address_space * mapping,u64 cookie,u64 change_attr)398 static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
399 u64 cookie, u64 change_attr)
400 {
401 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
402 struct folio *folio;
403
404 folio = filemap_grab_folio(mapping, index);
405 if (IS_ERR(folio))
406 return NULL;
407 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
408 return folio;
409 }
410
nfs_readdir_folio_last_cookie(struct folio * folio)411 static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
412 {
413 struct nfs_cache_array *array;
414 u64 ret;
415
416 array = kmap_local_folio(folio, 0);
417 ret = array->last_cookie;
418 kunmap_local(array);
419 return ret;
420 }
421
nfs_readdir_folio_needs_filling(struct folio * folio)422 static bool nfs_readdir_folio_needs_filling(struct folio *folio)
423 {
424 struct nfs_cache_array *array;
425 bool ret;
426
427 array = kmap_local_folio(folio, 0);
428 ret = !nfs_readdir_array_is_full(array);
429 kunmap_local(array);
430 return ret;
431 }
432
nfs_readdir_folio_set_eof(struct folio * folio)433 static void nfs_readdir_folio_set_eof(struct folio *folio)
434 {
435 struct nfs_cache_array *array;
436
437 array = kmap_local_folio(folio, 0);
438 nfs_readdir_array_set_eof(array);
439 kunmap_local(array);
440 }
441
nfs_readdir_folio_get_next(struct address_space * mapping,u64 cookie,u64 change_attr)442 static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
443 u64 cookie, u64 change_attr)
444 {
445 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
446 struct folio *folio;
447
448 folio = __filemap_get_folio(mapping, index,
449 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
450 mapping_gfp_mask(mapping));
451 if (IS_ERR(folio))
452 return NULL;
453 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
454 if (nfs_readdir_folio_last_cookie(folio) != cookie)
455 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
456 return folio;
457 }
458
459 static inline
is_32bit_api(void)460 int is_32bit_api(void)
461 {
462 #ifdef CONFIG_COMPAT
463 return in_compat_syscall();
464 #else
465 return (BITS_PER_LONG == 32);
466 #endif
467 }
468
469 static
nfs_readdir_use_cookie(const struct file * filp)470 bool nfs_readdir_use_cookie(const struct file *filp)
471 {
472 if ((filp->f_mode & FMODE_32BITHASH) ||
473 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
474 return false;
475 return true;
476 }
477
nfs_readdir_seek_next_array(struct nfs_cache_array * array,struct nfs_readdir_descriptor * desc)478 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
479 struct nfs_readdir_descriptor *desc)
480 {
481 if (array->folio_full) {
482 desc->last_cookie = array->last_cookie;
483 desc->current_index += array->size;
484 desc->cache_entry_index = 0;
485 desc->folio_index++;
486 } else
487 desc->last_cookie = nfs_readdir_array_index_cookie(array);
488 }
489
nfs_readdir_rewind_search(struct nfs_readdir_descriptor * desc)490 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
491 {
492 desc->current_index = 0;
493 desc->last_cookie = 0;
494 desc->folio_index = 0;
495 }
496
nfs_readdir_search_for_pos(struct nfs_cache_array * array,struct nfs_readdir_descriptor * desc)497 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
498 struct nfs_readdir_descriptor *desc)
499 {
500 loff_t diff = desc->ctx->pos - desc->current_index;
501 unsigned int index;
502
503 if (diff < 0)
504 goto out_eof;
505 if (diff >= array->size) {
506 if (array->folio_is_eof)
507 goto out_eof;
508 nfs_readdir_seek_next_array(array, desc);
509 return -EAGAIN;
510 }
511
512 index = (unsigned int)diff;
513 desc->dir_cookie = array->array[index].cookie;
514 desc->cache_entry_index = index;
515 return 0;
516 out_eof:
517 desc->eof = true;
518 return -EBADCOOKIE;
519 }
520
nfs_readdir_array_cookie_in_range(struct nfs_cache_array * array,u64 cookie)521 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
522 u64 cookie)
523 {
524 if (!array->cookies_are_ordered)
525 return true;
526 /* Optimisation for monotonically increasing cookies */
527 if (cookie >= array->last_cookie)
528 return false;
529 if (array->size && cookie < array->array[0].cookie)
530 return false;
531 return true;
532 }
533
nfs_readdir_search_for_cookie(struct nfs_cache_array * array,struct nfs_readdir_descriptor * desc)534 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
535 struct nfs_readdir_descriptor *desc)
536 {
537 unsigned int i;
538 int status = -EAGAIN;
539
540 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
541 goto check_eof;
542
543 for (i = 0; i < array->size; i++) {
544 if (array->array[i].cookie == desc->dir_cookie) {
545 if (nfs_readdir_use_cookie(desc->file))
546 desc->ctx->pos = desc->dir_cookie;
547 else
548 desc->ctx->pos = desc->current_index + i;
549 desc->cache_entry_index = i;
550 return 0;
551 }
552 }
553 check_eof:
554 if (array->folio_is_eof) {
555 status = -EBADCOOKIE;
556 if (desc->dir_cookie == array->last_cookie)
557 desc->eof = true;
558 } else
559 nfs_readdir_seek_next_array(array, desc);
560 return status;
561 }
562
nfs_readdir_search_array(struct nfs_readdir_descriptor * desc)563 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
564 {
565 struct nfs_cache_array *array;
566 int status;
567
568 array = kmap_local_folio(desc->folio, 0);
569
570 if (desc->dir_cookie == 0)
571 status = nfs_readdir_search_for_pos(array, desc);
572 else
573 status = nfs_readdir_search_for_cookie(array, desc);
574
575 kunmap_local(array);
576 return status;
577 }
578
579 /* Fill a page with xdr information before transferring to the cache page */
nfs_readdir_xdr_filler(struct nfs_readdir_descriptor * desc,__be32 * verf,u64 cookie,struct page ** pages,size_t bufsize,__be32 * verf_res)580 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
581 __be32 *verf, u64 cookie,
582 struct page **pages, size_t bufsize,
583 __be32 *verf_res)
584 {
585 struct inode *inode = file_inode(desc->file);
586 struct nfs_readdir_arg arg = {
587 .dentry = file_dentry(desc->file),
588 .cred = desc->file->f_cred,
589 .verf = verf,
590 .cookie = cookie,
591 .pages = pages,
592 .page_len = bufsize,
593 .plus = desc->plus,
594 };
595 struct nfs_readdir_res res = {
596 .verf = verf_res,
597 };
598 unsigned long timestamp, gencount;
599 int error;
600
601 again:
602 timestamp = jiffies;
603 gencount = nfs_inc_attr_generation_counter();
604 desc->dir_verifier = nfs_save_change_attribute(inode);
605 error = NFS_PROTO(inode)->readdir(&arg, &res);
606 if (error < 0) {
607 /* We requested READDIRPLUS, but the server doesn't grok it */
608 if (error == -ENOTSUPP && desc->plus) {
609 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
610 desc->plus = arg.plus = false;
611 goto again;
612 }
613 goto error;
614 }
615 desc->timestamp = timestamp;
616 desc->gencount = gencount;
617 error:
618 return error;
619 }
620
xdr_decode(struct nfs_readdir_descriptor * desc,struct nfs_entry * entry,struct xdr_stream * xdr)621 static int xdr_decode(struct nfs_readdir_descriptor *desc,
622 struct nfs_entry *entry, struct xdr_stream *xdr)
623 {
624 struct inode *inode = file_inode(desc->file);
625 int error;
626
627 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
628 if (error)
629 return error;
630 entry->fattr->time_start = desc->timestamp;
631 entry->fattr->gencount = desc->gencount;
632 return 0;
633 }
634
635 /* Match file and dirent using either filehandle or fileid
636 * Note: caller is responsible for checking the fsid
637 */
638 static
nfs_same_file(struct dentry * dentry,struct nfs_entry * entry)639 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
640 {
641 struct inode *inode;
642 struct nfs_inode *nfsi;
643
644 if (d_really_is_negative(dentry))
645 return 0;
646
647 inode = d_inode(dentry);
648 if (is_bad_inode(inode) || NFS_STALE(inode))
649 return 0;
650
651 nfsi = NFS_I(inode);
652 if (entry->fattr->fileid != nfsi->fileid)
653 return 0;
654 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
655 return 0;
656 return 1;
657 }
658
659 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
660
nfs_use_readdirplus(struct inode * dir,struct dir_context * ctx,unsigned int cache_hits,unsigned int cache_misses)661 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
662 unsigned int cache_hits,
663 unsigned int cache_misses)
664 {
665 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
666 return false;
667 if (ctx->pos == 0 ||
668 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
669 return true;
670 return false;
671 }
672
673 /*
674 * This function is called by the getattr code to request the
675 * use of readdirplus to accelerate any future lookups in the same
676 * directory.
677 */
nfs_readdir_record_entry_cache_hit(struct inode * dir)678 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
679 {
680 struct nfs_inode *nfsi = NFS_I(dir);
681 struct nfs_open_dir_context *ctx;
682
683 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
684 S_ISDIR(dir->i_mode)) {
685 rcu_read_lock();
686 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
687 atomic_inc(&ctx->cache_hits);
688 rcu_read_unlock();
689 }
690 }
691
692 /*
693 * This function is mainly for use by nfs_getattr().
694 *
695 * If this is an 'ls -l', we want to force use of readdirplus.
696 */
nfs_readdir_record_entry_cache_miss(struct inode * dir)697 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
698 {
699 struct nfs_inode *nfsi = NFS_I(dir);
700 struct nfs_open_dir_context *ctx;
701
702 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
703 S_ISDIR(dir->i_mode)) {
704 rcu_read_lock();
705 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
706 atomic_inc(&ctx->cache_misses);
707 rcu_read_unlock();
708 }
709 }
710
nfs_lookup_advise_force_readdirplus(struct inode * dir,unsigned int flags)711 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
712 unsigned int flags)
713 {
714 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
715 return;
716 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
717 return;
718 nfs_readdir_record_entry_cache_miss(dir);
719 }
720
721 static
nfs_prime_dcache(struct dentry * parent,struct nfs_entry * entry,unsigned long dir_verifier)722 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
723 unsigned long dir_verifier)
724 {
725 struct qstr filename = QSTR_INIT(entry->name, entry->len);
726 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
727 struct dentry *dentry;
728 struct dentry *alias;
729 struct inode *inode;
730 int status;
731
732 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
733 return;
734 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
735 return;
736 if (filename.len == 0)
737 return;
738 /* Validate that the name doesn't contain any illegal '\0' */
739 if (strnlen(filename.name, filename.len) != filename.len)
740 return;
741 /* ...or '/' */
742 if (strnchr(filename.name, filename.len, '/'))
743 return;
744 if (filename.name[0] == '.') {
745 if (filename.len == 1)
746 return;
747 if (filename.len == 2 && filename.name[1] == '.')
748 return;
749 }
750 filename.hash = full_name_hash(parent, filename.name, filename.len);
751
752 dentry = d_lookup(parent, &filename);
753 again:
754 if (!dentry) {
755 dentry = d_alloc_parallel(parent, &filename, &wq);
756 if (IS_ERR(dentry))
757 return;
758 }
759 if (!d_in_lookup(dentry)) {
760 /* Is there a mountpoint here? If so, just exit */
761 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
762 &entry->fattr->fsid))
763 goto out;
764 if (nfs_same_file(dentry, entry)) {
765 if (!entry->fh->size)
766 goto out;
767 nfs_set_verifier(dentry, dir_verifier);
768 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
769 if (!status)
770 nfs_setsecurity(d_inode(dentry), entry->fattr);
771 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
772 dentry, 0, status);
773 goto out;
774 } else {
775 trace_nfs_readdir_lookup_revalidate_failed(
776 d_inode(parent), dentry, 0);
777 d_invalidate(dentry);
778 dput(dentry);
779 dentry = NULL;
780 goto again;
781 }
782 }
783 if (!entry->fh->size) {
784 d_lookup_done(dentry);
785 goto out;
786 }
787
788 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
789 alias = d_splice_alias(inode, dentry);
790 d_lookup_done(dentry);
791 if (alias) {
792 if (IS_ERR(alias))
793 goto out;
794 dput(dentry);
795 dentry = alias;
796 }
797 nfs_set_verifier(dentry, dir_verifier);
798 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
799 out:
800 dput(dentry);
801 }
802
nfs_readdir_entry_decode(struct nfs_readdir_descriptor * desc,struct nfs_entry * entry,struct xdr_stream * stream)803 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
804 struct nfs_entry *entry,
805 struct xdr_stream *stream)
806 {
807 int ret;
808
809 if (entry->fattr->label)
810 entry->fattr->label->len = NFS4_MAXLABELLEN;
811 ret = xdr_decode(desc, entry, stream);
812 if (ret || !desc->plus)
813 return ret;
814 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
815 return 0;
816 }
817
818 /* Perform conversion from xdr to cache array */
nfs_readdir_folio_filler(struct nfs_readdir_descriptor * desc,struct nfs_entry * entry,struct page ** xdr_pages,unsigned int buflen,struct folio ** arrays,size_t narrays,u64 change_attr)819 static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
820 struct nfs_entry *entry,
821 struct page **xdr_pages, unsigned int buflen,
822 struct folio **arrays, size_t narrays,
823 u64 change_attr)
824 {
825 struct address_space *mapping = desc->file->f_mapping;
826 struct folio *new, *folio = *arrays;
827 struct xdr_stream stream;
828 struct page *scratch;
829 struct xdr_buf buf;
830 u64 cookie;
831 int status;
832
833 scratch = alloc_page(GFP_KERNEL);
834 if (scratch == NULL)
835 return -ENOMEM;
836
837 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
838 xdr_set_scratch_page(&stream, scratch);
839
840 do {
841 status = nfs_readdir_entry_decode(desc, entry, &stream);
842 if (status != 0)
843 break;
844
845 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
846 if (status != -ENOSPC)
847 continue;
848
849 if (folio->mapping != mapping) {
850 if (!--narrays)
851 break;
852 new = nfs_readdir_folio_array_alloc(cookie, GFP_KERNEL);
853 if (!new)
854 break;
855 arrays++;
856 *arrays = folio = new;
857 } else {
858 new = nfs_readdir_folio_get_next(mapping, cookie,
859 change_attr);
860 if (!new)
861 break;
862 if (folio != *arrays)
863 nfs_readdir_folio_unlock_and_put(folio);
864 folio = new;
865 }
866 desc->folio_index_max++;
867 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
868 } while (!status && !entry->eof);
869
870 switch (status) {
871 case -EBADCOOKIE:
872 if (!entry->eof)
873 break;
874 nfs_readdir_folio_set_eof(folio);
875 fallthrough;
876 case -EAGAIN:
877 status = 0;
878 break;
879 case -ENOSPC:
880 status = 0;
881 if (!desc->plus)
882 break;
883 while (!nfs_readdir_entry_decode(desc, entry, &stream))
884 ;
885 }
886
887 if (folio != *arrays)
888 nfs_readdir_folio_unlock_and_put(folio);
889
890 put_page(scratch);
891 return status;
892 }
893
nfs_readdir_free_pages(struct page ** pages,size_t npages)894 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
895 {
896 while (npages--)
897 put_page(pages[npages]);
898 kfree(pages);
899 }
900
901 /*
902 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
903 * to nfs_readdir_free_pages()
904 */
nfs_readdir_alloc_pages(size_t npages)905 static struct page **nfs_readdir_alloc_pages(size_t npages)
906 {
907 struct page **pages;
908 size_t i;
909
910 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
911 if (!pages)
912 return NULL;
913 for (i = 0; i < npages; i++) {
914 struct page *page = alloc_page(GFP_KERNEL);
915 if (page == NULL)
916 goto out_freepages;
917 pages[i] = page;
918 }
919 return pages;
920
921 out_freepages:
922 nfs_readdir_free_pages(pages, i);
923 return NULL;
924 }
925
nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor * desc,__be32 * verf_arg,__be32 * verf_res,struct folio ** arrays,size_t narrays)926 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
927 __be32 *verf_arg, __be32 *verf_res,
928 struct folio **arrays, size_t narrays)
929 {
930 u64 change_attr;
931 struct page **pages;
932 struct folio *folio = *arrays;
933 struct nfs_entry *entry;
934 size_t array_size;
935 struct inode *inode = file_inode(desc->file);
936 unsigned int dtsize = desc->dtsize;
937 unsigned int pglen;
938 int status = -ENOMEM;
939
940 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
941 if (!entry)
942 return -ENOMEM;
943 entry->cookie = nfs_readdir_folio_last_cookie(folio);
944 entry->fh = nfs_alloc_fhandle();
945 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
946 entry->server = NFS_SERVER(inode);
947 if (entry->fh == NULL || entry->fattr == NULL)
948 goto out;
949
950 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 pages = nfs_readdir_alloc_pages(array_size);
952 if (!pages)
953 goto out;
954
955 change_attr = inode_peek_iversion_raw(inode);
956 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
957 dtsize, verf_res);
958 if (status < 0)
959 goto free_pages;
960
961 pglen = status;
962 if (pglen != 0)
963 status = nfs_readdir_folio_filler(desc, entry, pages, pglen,
964 arrays, narrays, change_attr);
965 else
966 nfs_readdir_folio_set_eof(folio);
967 desc->buffer_fills++;
968
969 free_pages:
970 nfs_readdir_free_pages(pages, array_size);
971 out:
972 nfs_free_fattr(entry->fattr);
973 nfs_free_fhandle(entry->fh);
974 kfree(entry);
975 return status;
976 }
977
nfs_readdir_folio_put(struct nfs_readdir_descriptor * desc)978 static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
979 {
980 folio_put(desc->folio);
981 desc->folio = NULL;
982 }
983
984 static void
nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor * desc)985 nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
986 {
987 folio_unlock(desc->folio);
988 nfs_readdir_folio_put(desc);
989 }
990
991 static struct folio *
nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor * desc)992 nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
993 {
994 struct address_space *mapping = desc->file->f_mapping;
995 u64 change_attr = inode_peek_iversion_raw(mapping->host);
996 u64 cookie = desc->last_cookie;
997 struct folio *folio;
998
999 folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1000 if (!folio)
1001 return NULL;
1002 if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1003 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
1004 return folio;
1005 }
1006
1007 /*
1008 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1009 * and locks the page to prevent removal from the page cache.
1010 */
find_and_lock_cache_page(struct nfs_readdir_descriptor * desc)1011 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1012 {
1013 struct inode *inode = file_inode(desc->file);
1014 struct nfs_inode *nfsi = NFS_I(inode);
1015 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1016 int res;
1017
1018 desc->folio = nfs_readdir_folio_get_cached(desc);
1019 if (!desc->folio)
1020 return -ENOMEM;
1021 if (nfs_readdir_folio_needs_filling(desc->folio)) {
1022 /* Grow the dtsize if we had to go back for more pages */
1023 if (desc->folio_index == desc->folio_index_max)
1024 nfs_grow_dtsize(desc);
1025 desc->folio_index_max = desc->folio_index;
1026 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1027 desc->last_cookie,
1028 desc->folio->index, desc->dtsize);
1029 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1030 &desc->folio, 1);
1031 if (res < 0) {
1032 nfs_readdir_folio_unlock_and_put_cached(desc);
1033 trace_nfs_readdir_cache_fill_done(inode, res);
1034 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1035 invalidate_inode_pages2(desc->file->f_mapping);
1036 nfs_readdir_rewind_search(desc);
1037 trace_nfs_readdir_invalidate_cache_range(
1038 inode, 0, MAX_LFS_FILESIZE);
1039 return -EAGAIN;
1040 }
1041 return res;
1042 }
1043 /*
1044 * Set the cookie verifier if the page cache was empty
1045 */
1046 if (desc->last_cookie == 0 &&
1047 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1048 memcpy(nfsi->cookieverf, verf,
1049 sizeof(nfsi->cookieverf));
1050 invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1051 -1);
1052 trace_nfs_readdir_invalidate_cache_range(
1053 inode, 1, MAX_LFS_FILESIZE);
1054 }
1055 desc->clear_cache = false;
1056 }
1057 res = nfs_readdir_search_array(desc);
1058 if (res == 0)
1059 return 0;
1060 nfs_readdir_folio_unlock_and_put_cached(desc);
1061 return res;
1062 }
1063
1064 /* Search for desc->dir_cookie from the beginning of the page cache */
readdir_search_pagecache(struct nfs_readdir_descriptor * desc)1065 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1066 {
1067 int res;
1068
1069 do {
1070 res = find_and_lock_cache_page(desc);
1071 } while (res == -EAGAIN);
1072 return res;
1073 }
1074
1075 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1076
1077 /*
1078 * Once we've found the start of the dirent within a page: fill 'er up...
1079 */
nfs_do_filldir(struct nfs_readdir_descriptor * desc,const __be32 * verf)1080 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1081 const __be32 *verf)
1082 {
1083 struct file *file = desc->file;
1084 struct nfs_cache_array *array;
1085 unsigned int i;
1086 bool first_emit = !desc->dir_cookie;
1087
1088 array = kmap_local_folio(desc->folio, 0);
1089 for (i = desc->cache_entry_index; i < array->size; i++) {
1090 struct nfs_cache_array_entry *ent;
1091
1092 /*
1093 * nfs_readdir_handle_cache_misses return force clear at
1094 * (cache_misses > NFS_READDIR_CACHE_MISS_THRESHOLD) for
1095 * readdir heuristic, NFS_READDIR_CACHE_MISS_THRESHOLD + 1
1096 * entries need be emitted here.
1097 */
1098 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 2) {
1099 desc->eob = true;
1100 break;
1101 }
1102
1103 ent = &array->array[i];
1104 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1105 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1106 desc->eob = true;
1107 break;
1108 }
1109 memcpy(desc->verf, verf, sizeof(desc->verf));
1110 if (i == array->size - 1) {
1111 desc->dir_cookie = array->last_cookie;
1112 nfs_readdir_seek_next_array(array, desc);
1113 } else {
1114 desc->dir_cookie = array->array[i + 1].cookie;
1115 desc->last_cookie = array->array[0].cookie;
1116 }
1117 if (nfs_readdir_use_cookie(file))
1118 desc->ctx->pos = desc->dir_cookie;
1119 else
1120 desc->ctx->pos++;
1121 }
1122 if (array->folio_is_eof)
1123 desc->eof = !desc->eob;
1124
1125 kunmap_local(array);
1126 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1127 (unsigned long long)desc->dir_cookie);
1128 }
1129
1130 /*
1131 * If we cannot find a cookie in our cache, we suspect that this is
1132 * because it points to a deleted file, so we ask the server to return
1133 * whatever it thinks is the next entry. We then feed this to filldir.
1134 * If all goes well, we should then be able to find our way round the
1135 * cache on the next call to readdir_search_pagecache();
1136 *
1137 * NOTE: we cannot add the anonymous page to the pagecache because
1138 * the data it contains might not be page aligned. Besides,
1139 * we should already have a complete representation of the
1140 * directory in the page cache by the time we get here.
1141 */
uncached_readdir(struct nfs_readdir_descriptor * desc)1142 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1143 {
1144 struct folio **arrays;
1145 size_t i, sz = 512;
1146 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1147 int status = -ENOMEM;
1148
1149 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1150 (unsigned long long)desc->dir_cookie);
1151
1152 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1153 if (!arrays)
1154 goto out;
1155 arrays[0] = nfs_readdir_folio_array_alloc(desc->dir_cookie, GFP_KERNEL);
1156 if (!arrays[0])
1157 goto out;
1158
1159 desc->folio_index = 0;
1160 desc->cache_entry_index = 0;
1161 desc->last_cookie = desc->dir_cookie;
1162 desc->folio_index_max = 0;
1163
1164 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1165 -1, desc->dtsize);
1166
1167 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1168 if (status < 0) {
1169 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1170 goto out_free;
1171 }
1172
1173 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1174 desc->folio = arrays[i];
1175 nfs_do_filldir(desc, verf);
1176 }
1177 desc->folio = NULL;
1178
1179 /*
1180 * Grow the dtsize if we have to go back for more pages,
1181 * or shrink it if we're reading too many.
1182 */
1183 if (!desc->eof) {
1184 if (!desc->eob)
1185 nfs_grow_dtsize(desc);
1186 else if (desc->buffer_fills == 1 &&
1187 i < (desc->folio_index_max >> 1))
1188 nfs_shrink_dtsize(desc);
1189 }
1190 out_free:
1191 for (i = 0; i < sz && arrays[i]; i++)
1192 nfs_readdir_folio_array_free(arrays[i]);
1193 out:
1194 if (!nfs_readdir_use_cookie(desc->file))
1195 nfs_readdir_rewind_search(desc);
1196 desc->folio_index_max = -1;
1197 kfree(arrays);
1198 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1199 return status;
1200 }
1201
nfs_readdir_handle_cache_misses(struct inode * inode,struct nfs_readdir_descriptor * desc,unsigned int cache_misses,bool force_clear)1202 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1203 struct nfs_readdir_descriptor *desc,
1204 unsigned int cache_misses,
1205 bool force_clear)
1206 {
1207 if (desc->ctx->pos == 0 || !desc->plus)
1208 return false;
1209 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1210 return false;
1211 trace_nfs_readdir_force_readdirplus(inode);
1212 return true;
1213 }
1214
1215 /* The file offset position represents the dirent entry number. A
1216 last cookie cache takes care of the common case of reading the
1217 whole directory.
1218 */
nfs_readdir(struct file * file,struct dir_context * ctx)1219 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1220 {
1221 struct dentry *dentry = file_dentry(file);
1222 struct inode *inode = d_inode(dentry);
1223 struct nfs_inode *nfsi = NFS_I(inode);
1224 struct nfs_open_dir_context *dir_ctx = file->private_data;
1225 struct nfs_readdir_descriptor *desc;
1226 unsigned int cache_hits, cache_misses;
1227 bool force_clear;
1228 int res;
1229
1230 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1231 file, (long long)ctx->pos);
1232 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1233
1234 /*
1235 * ctx->pos points to the dirent entry number.
1236 * *desc->dir_cookie has the cookie for the next entry. We have
1237 * to either find the entry with the appropriate number or
1238 * revalidate the cookie.
1239 */
1240 nfs_revalidate_mapping(inode, file->f_mapping);
1241
1242 res = -ENOMEM;
1243 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1244 if (!desc)
1245 goto out;
1246 desc->file = file;
1247 desc->ctx = ctx;
1248 desc->folio_index_max = -1;
1249
1250 spin_lock(&file->f_lock);
1251 desc->dir_cookie = dir_ctx->dir_cookie;
1252 desc->folio_index = dir_ctx->page_index;
1253 desc->last_cookie = dir_ctx->last_cookie;
1254 desc->attr_gencount = dir_ctx->attr_gencount;
1255 desc->eof = dir_ctx->eof;
1256 nfs_set_dtsize(desc, dir_ctx->dtsize);
1257 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1258 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1259 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1260 force_clear = dir_ctx->force_clear;
1261 spin_unlock(&file->f_lock);
1262
1263 if (desc->eof) {
1264 res = 0;
1265 goto out_free;
1266 }
1267
1268 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1269 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1270 force_clear);
1271 desc->clear_cache = force_clear;
1272
1273 do {
1274 res = readdir_search_pagecache(desc);
1275
1276 if (res == -EBADCOOKIE) {
1277 res = 0;
1278 /* This means either end of directory */
1279 if (desc->dir_cookie && !desc->eof) {
1280 /* Or that the server has 'lost' a cookie */
1281 res = uncached_readdir(desc);
1282 if (res == 0)
1283 continue;
1284 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1285 res = 0;
1286 }
1287 break;
1288 }
1289 if (res == -ETOOSMALL && desc->plus) {
1290 nfs_zap_caches(inode);
1291 desc->plus = false;
1292 desc->eof = false;
1293 continue;
1294 }
1295 if (res < 0)
1296 break;
1297
1298 nfs_do_filldir(desc, nfsi->cookieverf);
1299 nfs_readdir_folio_unlock_and_put_cached(desc);
1300 if (desc->folio_index == desc->folio_index_max)
1301 desc->clear_cache = force_clear;
1302 } while (!desc->eob && !desc->eof);
1303
1304 spin_lock(&file->f_lock);
1305 dir_ctx->dir_cookie = desc->dir_cookie;
1306 dir_ctx->last_cookie = desc->last_cookie;
1307 dir_ctx->attr_gencount = desc->attr_gencount;
1308 dir_ctx->page_index = desc->folio_index;
1309 dir_ctx->force_clear = force_clear;
1310 dir_ctx->eof = desc->eof;
1311 dir_ctx->dtsize = desc->dtsize;
1312 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1313 spin_unlock(&file->f_lock);
1314 out_free:
1315 kfree(desc);
1316
1317 out:
1318 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1319 return res;
1320 }
1321
nfs_llseek_dir(struct file * filp,loff_t offset,int whence)1322 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1323 {
1324 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1325
1326 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1327 filp, offset, whence);
1328
1329 switch (whence) {
1330 default:
1331 return -EINVAL;
1332 case SEEK_SET:
1333 if (offset < 0)
1334 return -EINVAL;
1335 spin_lock(&filp->f_lock);
1336 break;
1337 case SEEK_CUR:
1338 if (offset == 0)
1339 return filp->f_pos;
1340 spin_lock(&filp->f_lock);
1341 offset += filp->f_pos;
1342 if (offset < 0) {
1343 spin_unlock(&filp->f_lock);
1344 return -EINVAL;
1345 }
1346 }
1347 if (offset != filp->f_pos) {
1348 filp->f_pos = offset;
1349 dir_ctx->page_index = 0;
1350 if (!nfs_readdir_use_cookie(filp)) {
1351 dir_ctx->dir_cookie = 0;
1352 dir_ctx->last_cookie = 0;
1353 } else {
1354 dir_ctx->dir_cookie = offset;
1355 dir_ctx->last_cookie = offset;
1356 }
1357 dir_ctx->eof = false;
1358 }
1359 spin_unlock(&filp->f_lock);
1360 return offset;
1361 }
1362
1363 /*
1364 * All directory operations under NFS are synchronous, so fsync()
1365 * is a dummy operation.
1366 */
nfs_fsync_dir(struct file * filp,loff_t start,loff_t end,int datasync)1367 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1368 int datasync)
1369 {
1370 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1371
1372 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1373 return 0;
1374 }
1375
1376 /**
1377 * nfs_force_lookup_revalidate - Mark the directory as having changed
1378 * @dir: pointer to directory inode
1379 *
1380 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1381 * full lookup on all child dentries of 'dir' whenever a change occurs
1382 * on the server that might have invalidated our dcache.
1383 *
1384 * Note that we reserve bit '0' as a tag to let us know when a dentry
1385 * was revalidated while holding a delegation on its inode.
1386 *
1387 * The caller should be holding dir->i_lock
1388 */
nfs_force_lookup_revalidate(struct inode * dir)1389 void nfs_force_lookup_revalidate(struct inode *dir)
1390 {
1391 NFS_I(dir)->cache_change_attribute += 2;
1392 }
1393 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1394
1395 /**
1396 * nfs_verify_change_attribute - Detects NFS remote directory changes
1397 * @dir: pointer to parent directory inode
1398 * @verf: previously saved change attribute
1399 *
1400 * Return "false" if the verifiers doesn't match the change attribute.
1401 * This would usually indicate that the directory contents have changed on
1402 * the server, and that any dentries need revalidating.
1403 */
nfs_verify_change_attribute(struct inode * dir,unsigned long verf)1404 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1405 {
1406 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1407 }
1408
nfs_set_verifier_delegated(unsigned long * verf)1409 static void nfs_set_verifier_delegated(unsigned long *verf)
1410 {
1411 *verf |= 1UL;
1412 }
1413
1414 #if IS_ENABLED(CONFIG_NFS_V4)
nfs_unset_verifier_delegated(unsigned long * verf)1415 static void nfs_unset_verifier_delegated(unsigned long *verf)
1416 {
1417 *verf &= ~1UL;
1418 }
1419 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1420
nfs_test_verifier_delegated(unsigned long verf)1421 static bool nfs_test_verifier_delegated(unsigned long verf)
1422 {
1423 return verf & 1;
1424 }
1425
nfs_verifier_is_delegated(struct dentry * dentry)1426 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1427 {
1428 return nfs_test_verifier_delegated(dentry->d_time);
1429 }
1430
nfs_set_verifier_locked(struct dentry * dentry,unsigned long verf)1431 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1432 {
1433 struct inode *inode = d_inode(dentry);
1434 struct inode *dir = d_inode(dentry->d_parent);
1435
1436 if (!nfs_verify_change_attribute(dir, verf))
1437 return;
1438 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1439 nfs_set_verifier_delegated(&verf);
1440 dentry->d_time = verf;
1441 }
1442
1443 /**
1444 * nfs_set_verifier - save a parent directory verifier in the dentry
1445 * @dentry: pointer to dentry
1446 * @verf: verifier to save
1447 *
1448 * Saves the parent directory verifier in @dentry. If the inode has
1449 * a delegation, we also tag the dentry as having been revalidated
1450 * while holding a delegation so that we know we don't have to
1451 * look it up again after a directory change.
1452 */
nfs_set_verifier(struct dentry * dentry,unsigned long verf)1453 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1454 {
1455
1456 spin_lock(&dentry->d_lock);
1457 nfs_set_verifier_locked(dentry, verf);
1458 spin_unlock(&dentry->d_lock);
1459 }
1460 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1461
1462 #if IS_ENABLED(CONFIG_NFS_V4)
1463 /**
1464 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1465 * @inode: pointer to inode
1466 *
1467 * Iterates through the dentries in the inode alias list and clears
1468 * the tag used to indicate that the dentry has been revalidated
1469 * while holding a delegation.
1470 * This function is intended for use when the delegation is being
1471 * returned or revoked.
1472 */
nfs_clear_verifier_delegated(struct inode * inode)1473 void nfs_clear_verifier_delegated(struct inode *inode)
1474 {
1475 struct dentry *alias;
1476
1477 if (!inode)
1478 return;
1479 spin_lock(&inode->i_lock);
1480 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1481 spin_lock(&alias->d_lock);
1482 nfs_unset_verifier_delegated(&alias->d_time);
1483 spin_unlock(&alias->d_lock);
1484 }
1485 spin_unlock(&inode->i_lock);
1486 }
1487 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1488 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1489
nfs_dentry_verify_change(struct inode * dir,struct dentry * dentry)1490 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1491 {
1492 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1493 d_really_is_negative(dentry))
1494 return dentry->d_time == inode_peek_iversion_raw(dir);
1495 return nfs_verify_change_attribute(dir, dentry->d_time);
1496 }
1497
1498 /*
1499 * A check for whether or not the parent directory has changed.
1500 * In the case it has, we assume that the dentries are untrustworthy
1501 * and may need to be looked up again.
1502 * If rcu_walk prevents us from performing a full check, return 0.
1503 */
nfs_check_verifier(struct inode * dir,struct dentry * dentry,int rcu_walk)1504 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1505 int rcu_walk)
1506 {
1507 if (IS_ROOT(dentry))
1508 return 1;
1509 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1510 return 0;
1511 if (!nfs_dentry_verify_change(dir, dentry))
1512 return 0;
1513 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1514 if (nfs_mapping_need_revalidate_inode(dir)) {
1515 if (rcu_walk)
1516 return 0;
1517 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1518 return 0;
1519 }
1520 if (!nfs_dentry_verify_change(dir, dentry))
1521 return 0;
1522 return 1;
1523 }
1524
1525 /*
1526 * Use intent information to check whether or not we're going to do
1527 * an O_EXCL create using this path component.
1528 */
nfs_is_exclusive_create(struct inode * dir,unsigned int flags)1529 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1530 {
1531 if (NFS_PROTO(dir)->version == 2)
1532 return 0;
1533 return flags & LOOKUP_EXCL;
1534 }
1535
1536 /*
1537 * Inode and filehandle revalidation for lookups.
1538 *
1539 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1540 * or if the intent information indicates that we're about to open this
1541 * particular file and the "nocto" mount flag is not set.
1542 *
1543 */
1544 static
nfs_lookup_verify_inode(struct inode * inode,unsigned int flags)1545 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1546 {
1547 struct nfs_server *server = NFS_SERVER(inode);
1548 int ret;
1549
1550 if (IS_AUTOMOUNT(inode))
1551 return 0;
1552
1553 if (flags & LOOKUP_OPEN) {
1554 switch (inode->i_mode & S_IFMT) {
1555 case S_IFREG:
1556 /* A NFSv4 OPEN will revalidate later */
1557 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1558 goto out;
1559 fallthrough;
1560 case S_IFDIR:
1561 if (server->flags & NFS_MOUNT_NOCTO)
1562 break;
1563 /* NFS close-to-open cache consistency validation */
1564 goto out_force;
1565 }
1566 }
1567
1568 /* VFS wants an on-the-wire revalidation */
1569 if (flags & LOOKUP_REVAL)
1570 goto out_force;
1571 out:
1572 if (inode->i_nlink > 0 ||
1573 (inode->i_nlink == 0 &&
1574 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1575 return 0;
1576 else
1577 return -ESTALE;
1578 out_force:
1579 if (flags & LOOKUP_RCU)
1580 return -ECHILD;
1581 ret = __nfs_revalidate_inode(server, inode);
1582 if (ret != 0)
1583 return ret;
1584 goto out;
1585 }
1586
nfs_mark_dir_for_revalidate(struct inode * inode)1587 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1588 {
1589 spin_lock(&inode->i_lock);
1590 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1591 spin_unlock(&inode->i_lock);
1592 }
1593
1594 /*
1595 * We judge how long we want to trust negative
1596 * dentries by looking at the parent inode mtime.
1597 *
1598 * If parent mtime has changed, we revalidate, else we wait for a
1599 * period corresponding to the parent's attribute cache timeout value.
1600 *
1601 * If LOOKUP_RCU prevents us from performing a full check, return 1
1602 * suggesting a reval is needed.
1603 *
1604 * Note that when creating a new file, or looking up a rename target,
1605 * then it shouldn't be necessary to revalidate a negative dentry.
1606 */
1607 static inline
nfs_neg_need_reval(struct inode * dir,struct dentry * dentry,unsigned int flags)1608 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1609 unsigned int flags)
1610 {
1611 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1612 return 0;
1613 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1614 return 1;
1615 /* Case insensitive server? Revalidate negative dentries */
1616 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1617 return 1;
1618 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1619 }
1620
1621 static int
nfs_lookup_revalidate_done(struct inode * dir,struct dentry * dentry,struct inode * inode,int error)1622 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1623 struct inode *inode, int error)
1624 {
1625 switch (error) {
1626 case 1:
1627 break;
1628 case 0:
1629 /*
1630 * We can't d_drop the root of a disconnected tree:
1631 * its d_hash is on the s_anon list and d_drop() would hide
1632 * it from shrink_dcache_for_unmount(), leading to busy
1633 * inodes on unmount and further oopses.
1634 */
1635 if (inode && IS_ROOT(dentry))
1636 error = 1;
1637 break;
1638 }
1639 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1640 return error;
1641 }
1642
1643 static int
nfs_lookup_revalidate_negative(struct inode * dir,struct dentry * dentry,unsigned int flags)1644 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1645 unsigned int flags)
1646 {
1647 int ret = 1;
1648 if (nfs_neg_need_reval(dir, dentry, flags)) {
1649 if (flags & LOOKUP_RCU)
1650 return -ECHILD;
1651 ret = 0;
1652 }
1653 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1654 }
1655
1656 static int
nfs_lookup_revalidate_delegated(struct inode * dir,struct dentry * dentry,struct inode * inode)1657 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1658 struct inode *inode)
1659 {
1660 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1661 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1662 }
1663
nfs_lookup_revalidate_dentry(struct inode * dir,struct dentry * dentry,struct inode * inode,unsigned int flags)1664 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1665 struct dentry *dentry,
1666 struct inode *inode, unsigned int flags)
1667 {
1668 struct nfs_fh *fhandle;
1669 struct nfs_fattr *fattr;
1670 unsigned long dir_verifier;
1671 int ret;
1672
1673 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1674
1675 ret = -ENOMEM;
1676 fhandle = nfs_alloc_fhandle();
1677 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1678 if (fhandle == NULL || fattr == NULL)
1679 goto out;
1680
1681 dir_verifier = nfs_save_change_attribute(dir);
1682 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1683 if (ret < 0) {
1684 switch (ret) {
1685 case -ESTALE:
1686 case -ENOENT:
1687 ret = 0;
1688 break;
1689 case -ETIMEDOUT:
1690 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1691 ret = 1;
1692 }
1693 goto out;
1694 }
1695
1696 /* Request help from readdirplus */
1697 nfs_lookup_advise_force_readdirplus(dir, flags);
1698
1699 ret = 0;
1700 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1701 goto out;
1702 if (nfs_refresh_inode(inode, fattr) < 0)
1703 goto out;
1704
1705 nfs_setsecurity(inode, fattr);
1706 nfs_set_verifier(dentry, dir_verifier);
1707
1708 ret = 1;
1709 out:
1710 nfs_free_fattr(fattr);
1711 nfs_free_fhandle(fhandle);
1712
1713 /*
1714 * If the lookup failed despite the dentry change attribute being
1715 * a match, then we should revalidate the directory cache.
1716 */
1717 if (!ret && nfs_dentry_verify_change(dir, dentry))
1718 nfs_mark_dir_for_revalidate(dir);
1719 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1720 }
1721
1722 /*
1723 * This is called every time the dcache has a lookup hit,
1724 * and we should check whether we can really trust that
1725 * lookup.
1726 *
1727 * NOTE! The hit can be a negative hit too, don't assume
1728 * we have an inode!
1729 *
1730 * If the parent directory is seen to have changed, we throw out the
1731 * cached dentry and do a new lookup.
1732 */
1733 static int
nfs_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)1734 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1735 unsigned int flags)
1736 {
1737 struct inode *inode;
1738 int error;
1739
1740 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1741 inode = d_inode(dentry);
1742
1743 if (!inode)
1744 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1745
1746 if (is_bad_inode(inode)) {
1747 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1748 __func__, dentry);
1749 goto out_bad;
1750 }
1751
1752 if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1753 nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1754 goto out_bad;
1755
1756 if (nfs_verifier_is_delegated(dentry))
1757 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1758
1759 /* Force a full look up iff the parent directory has changed */
1760 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1761 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1762 error = nfs_lookup_verify_inode(inode, flags);
1763 if (error) {
1764 if (error == -ESTALE)
1765 nfs_mark_dir_for_revalidate(dir);
1766 goto out_bad;
1767 }
1768 goto out_valid;
1769 }
1770
1771 if (flags & LOOKUP_RCU)
1772 return -ECHILD;
1773
1774 if (NFS_STALE(inode))
1775 goto out_bad;
1776
1777 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1778 out_valid:
1779 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1780 out_bad:
1781 if (flags & LOOKUP_RCU)
1782 return -ECHILD;
1783 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1784 }
1785
1786 static int
__nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags,int (* reval)(struct inode *,struct dentry *,unsigned int))1787 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1788 int (*reval)(struct inode *, struct dentry *, unsigned int))
1789 {
1790 struct dentry *parent;
1791 struct inode *dir;
1792 int ret;
1793
1794 if (flags & LOOKUP_RCU) {
1795 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1796 return -ECHILD;
1797 parent = READ_ONCE(dentry->d_parent);
1798 dir = d_inode_rcu(parent);
1799 if (!dir)
1800 return -ECHILD;
1801 ret = reval(dir, dentry, flags);
1802 if (parent != READ_ONCE(dentry->d_parent))
1803 return -ECHILD;
1804 } else {
1805 /* Wait for unlink to complete - see unblock_revalidate() */
1806 wait_var_event(&dentry->d_fsdata,
1807 smp_load_acquire(&dentry->d_fsdata)
1808 != NFS_FSDATA_BLOCKED);
1809 parent = dget_parent(dentry);
1810 ret = reval(d_inode(parent), dentry, flags);
1811 dput(parent);
1812 }
1813 return ret;
1814 }
1815
nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags)1816 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1817 {
1818 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1819 }
1820
block_revalidate(struct dentry * dentry)1821 static void block_revalidate(struct dentry *dentry)
1822 {
1823 /* old devname - just in case */
1824 kfree(dentry->d_fsdata);
1825
1826 /* Any new reference that could lead to an open
1827 * will take ->d_lock in lookup_open() -> d_lookup().
1828 * Holding this lock ensures we cannot race with
1829 * __nfs_lookup_revalidate() and removes and need
1830 * for further barriers.
1831 */
1832 lockdep_assert_held(&dentry->d_lock);
1833
1834 dentry->d_fsdata = NFS_FSDATA_BLOCKED;
1835 }
1836
unblock_revalidate(struct dentry * dentry)1837 static void unblock_revalidate(struct dentry *dentry)
1838 {
1839 /* store_release ensures wait_var_event() sees the update */
1840 smp_store_release(&dentry->d_fsdata, NULL);
1841 wake_up_var(&dentry->d_fsdata);
1842 }
1843
1844 /*
1845 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1846 * when we don't really care about the dentry name. This is called when a
1847 * pathwalk ends on a dentry that was not found via a normal lookup in the
1848 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1849 *
1850 * In this situation, we just want to verify that the inode itself is OK
1851 * since the dentry might have changed on the server.
1852 */
nfs_weak_revalidate(struct dentry * dentry,unsigned int flags)1853 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1854 {
1855 struct inode *inode = d_inode(dentry);
1856 int error = 0;
1857
1858 /*
1859 * I believe we can only get a negative dentry here in the case of a
1860 * procfs-style symlink. Just assume it's correct for now, but we may
1861 * eventually need to do something more here.
1862 */
1863 if (!inode) {
1864 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1865 __func__, dentry);
1866 return 1;
1867 }
1868
1869 if (is_bad_inode(inode)) {
1870 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1871 __func__, dentry);
1872 return 0;
1873 }
1874
1875 error = nfs_lookup_verify_inode(inode, flags);
1876 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1877 __func__, inode->i_ino, error ? "invalid" : "valid");
1878 return !error;
1879 }
1880
1881 /*
1882 * This is called from dput() when d_count is going to 0.
1883 */
nfs_dentry_delete(const struct dentry * dentry)1884 static int nfs_dentry_delete(const struct dentry *dentry)
1885 {
1886 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1887 dentry, dentry->d_flags);
1888
1889 /* Unhash any dentry with a stale inode */
1890 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1891 return 1;
1892
1893 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1894 /* Unhash it, so that ->d_iput() would be called */
1895 return 1;
1896 }
1897 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1898 /* Unhash it, so that ancestors of killed async unlink
1899 * files will be cleaned up during umount */
1900 return 1;
1901 }
1902 return 0;
1903
1904 }
1905
1906 /* Ensure that we revalidate inode->i_nlink */
nfs_drop_nlink(struct inode * inode)1907 static void nfs_drop_nlink(struct inode *inode)
1908 {
1909 spin_lock(&inode->i_lock);
1910 /* drop the inode if we're reasonably sure this is the last link */
1911 if (inode->i_nlink > 0)
1912 drop_nlink(inode);
1913 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1914 nfs_set_cache_invalid(
1915 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1916 NFS_INO_INVALID_NLINK);
1917 spin_unlock(&inode->i_lock);
1918 }
1919
1920 /*
1921 * Called when the dentry loses inode.
1922 * We use it to clean up silly-renamed files.
1923 */
nfs_dentry_iput(struct dentry * dentry,struct inode * inode)1924 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1925 {
1926 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1927 nfs_complete_unlink(dentry, inode);
1928 nfs_drop_nlink(inode);
1929 }
1930 iput(inode);
1931 }
1932
nfs_d_release(struct dentry * dentry)1933 static void nfs_d_release(struct dentry *dentry)
1934 {
1935 /* free cached devname value, if it survived that far */
1936 if (unlikely(dentry->d_fsdata)) {
1937 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1938 WARN_ON(1);
1939 else
1940 kfree(dentry->d_fsdata);
1941 }
1942 }
1943
1944 const struct dentry_operations nfs_dentry_operations = {
1945 .d_revalidate = nfs_lookup_revalidate,
1946 .d_weak_revalidate = nfs_weak_revalidate,
1947 .d_delete = nfs_dentry_delete,
1948 .d_iput = nfs_dentry_iput,
1949 .d_automount = nfs_d_automount,
1950 .d_release = nfs_d_release,
1951 };
1952 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1953
nfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)1954 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1955 {
1956 struct dentry *res;
1957 struct inode *inode = NULL;
1958 struct nfs_fh *fhandle = NULL;
1959 struct nfs_fattr *fattr = NULL;
1960 unsigned long dir_verifier;
1961 int error;
1962
1963 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1964 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1965
1966 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1967 return ERR_PTR(-ENAMETOOLONG);
1968
1969 /*
1970 * If we're doing an exclusive create, optimize away the lookup
1971 * but don't hash the dentry.
1972 */
1973 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1974 return NULL;
1975
1976 res = ERR_PTR(-ENOMEM);
1977 fhandle = nfs_alloc_fhandle();
1978 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1979 if (fhandle == NULL || fattr == NULL)
1980 goto out;
1981
1982 dir_verifier = nfs_save_change_attribute(dir);
1983 trace_nfs_lookup_enter(dir, dentry, flags);
1984 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1985 if (error == -ENOENT) {
1986 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1987 dir_verifier = inode_peek_iversion_raw(dir);
1988 goto no_entry;
1989 }
1990 if (error < 0) {
1991 res = ERR_PTR(error);
1992 goto out;
1993 }
1994 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1995 res = ERR_CAST(inode);
1996 if (IS_ERR(res))
1997 goto out;
1998
1999 /* Notify readdir to use READDIRPLUS */
2000 nfs_lookup_advise_force_readdirplus(dir, flags);
2001
2002 no_entry:
2003 res = d_splice_alias(inode, dentry);
2004 if (res != NULL) {
2005 if (IS_ERR(res))
2006 goto out;
2007 dentry = res;
2008 }
2009 nfs_set_verifier(dentry, dir_verifier);
2010 out:
2011 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
2012 nfs_free_fattr(fattr);
2013 nfs_free_fhandle(fhandle);
2014 return res;
2015 }
2016 EXPORT_SYMBOL_GPL(nfs_lookup);
2017
nfs_d_prune_case_insensitive_aliases(struct inode * inode)2018 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
2019 {
2020 /* Case insensitive server? Revalidate dentries */
2021 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
2022 d_prune_aliases(inode);
2023 }
2024 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
2025
2026 #if IS_ENABLED(CONFIG_NFS_V4)
2027 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
2028
2029 const struct dentry_operations nfs4_dentry_operations = {
2030 .d_revalidate = nfs4_lookup_revalidate,
2031 .d_weak_revalidate = nfs_weak_revalidate,
2032 .d_delete = nfs_dentry_delete,
2033 .d_iput = nfs_dentry_iput,
2034 .d_automount = nfs_d_automount,
2035 .d_release = nfs_d_release,
2036 };
2037 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2038
create_nfs_open_context(struct dentry * dentry,int open_flags,struct file * filp)2039 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2040 {
2041 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2042 }
2043
do_open(struct inode * inode,struct file * filp)2044 static int do_open(struct inode *inode, struct file *filp)
2045 {
2046 nfs_fscache_open_file(inode, filp);
2047 return 0;
2048 }
2049
nfs_finish_open(struct nfs_open_context * ctx,struct dentry * dentry,struct file * file,unsigned open_flags)2050 static int nfs_finish_open(struct nfs_open_context *ctx,
2051 struct dentry *dentry,
2052 struct file *file, unsigned open_flags)
2053 {
2054 int err;
2055
2056 err = finish_open(file, dentry, do_open);
2057 if (err)
2058 goto out;
2059 if (S_ISREG(file_inode(file)->i_mode))
2060 nfs_file_set_open_context(file, ctx);
2061 else
2062 err = -EOPENSTALE;
2063 out:
2064 return err;
2065 }
2066
nfs_atomic_open(struct inode * dir,struct dentry * dentry,struct file * file,unsigned open_flags,umode_t mode)2067 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2068 struct file *file, unsigned open_flags,
2069 umode_t mode)
2070 {
2071 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2072 struct nfs_open_context *ctx;
2073 struct dentry *res;
2074 struct iattr attr = { .ia_valid = ATTR_OPEN };
2075 struct inode *inode;
2076 unsigned int lookup_flags = 0;
2077 unsigned long dir_verifier;
2078 bool switched = false;
2079 int created = 0;
2080 int err;
2081
2082 /* Expect a negative dentry */
2083 BUG_ON(d_inode(dentry));
2084
2085 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2086 dir->i_sb->s_id, dir->i_ino, dentry);
2087
2088 err = nfs_check_flags(open_flags);
2089 if (err)
2090 return err;
2091
2092 /* NFS only supports OPEN on regular files */
2093 if ((open_flags & O_DIRECTORY)) {
2094 if (!d_in_lookup(dentry)) {
2095 /*
2096 * Hashed negative dentry with O_DIRECTORY: dentry was
2097 * revalidated and is fine, no need to perform lookup
2098 * again
2099 */
2100 return -ENOENT;
2101 }
2102 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2103 goto no_open;
2104 }
2105
2106 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2107 return -ENAMETOOLONG;
2108
2109 if (open_flags & O_CREAT) {
2110 struct nfs_server *server = NFS_SERVER(dir);
2111
2112 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2113 mode &= ~current_umask();
2114
2115 attr.ia_valid |= ATTR_MODE;
2116 attr.ia_mode = mode;
2117 }
2118 if (open_flags & O_TRUNC) {
2119 attr.ia_valid |= ATTR_SIZE;
2120 attr.ia_size = 0;
2121 }
2122
2123 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2124 d_drop(dentry);
2125 switched = true;
2126 dentry = d_alloc_parallel(dentry->d_parent,
2127 &dentry->d_name, &wq);
2128 if (IS_ERR(dentry))
2129 return PTR_ERR(dentry);
2130 if (unlikely(!d_in_lookup(dentry)))
2131 return finish_no_open(file, dentry);
2132 }
2133
2134 ctx = create_nfs_open_context(dentry, open_flags, file);
2135 err = PTR_ERR(ctx);
2136 if (IS_ERR(ctx))
2137 goto out;
2138
2139 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2140 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2141 if (created)
2142 file->f_mode |= FMODE_CREATED;
2143 if (IS_ERR(inode)) {
2144 err = PTR_ERR(inode);
2145 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2146 put_nfs_open_context(ctx);
2147 d_drop(dentry);
2148 switch (err) {
2149 case -ENOENT:
2150 d_splice_alias(NULL, dentry);
2151 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2152 dir_verifier = inode_peek_iversion_raw(dir);
2153 else
2154 dir_verifier = nfs_save_change_attribute(dir);
2155 nfs_set_verifier(dentry, dir_verifier);
2156 break;
2157 case -EISDIR:
2158 case -ENOTDIR:
2159 goto no_open;
2160 case -ELOOP:
2161 if (!(open_flags & O_NOFOLLOW))
2162 goto no_open;
2163 break;
2164 /* case -EINVAL: */
2165 default:
2166 break;
2167 }
2168 goto out;
2169 }
2170 file->f_mode |= FMODE_CAN_ODIRECT;
2171
2172 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2173 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2174 put_nfs_open_context(ctx);
2175 out:
2176 if (unlikely(switched)) {
2177 d_lookup_done(dentry);
2178 dput(dentry);
2179 }
2180 return err;
2181
2182 no_open:
2183 res = nfs_lookup(dir, dentry, lookup_flags);
2184 if (!res) {
2185 inode = d_inode(dentry);
2186 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2187 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2188 res = ERR_PTR(-ENOTDIR);
2189 else if (inode && S_ISREG(inode->i_mode))
2190 res = ERR_PTR(-EOPENSTALE);
2191 } else if (!IS_ERR(res)) {
2192 inode = d_inode(res);
2193 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2194 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2195 dput(res);
2196 res = ERR_PTR(-ENOTDIR);
2197 } else if (inode && S_ISREG(inode->i_mode)) {
2198 dput(res);
2199 res = ERR_PTR(-EOPENSTALE);
2200 }
2201 }
2202 if (switched) {
2203 d_lookup_done(dentry);
2204 if (!res)
2205 res = dentry;
2206 else
2207 dput(dentry);
2208 }
2209 if (IS_ERR(res))
2210 return PTR_ERR(res);
2211 return finish_no_open(file, res);
2212 }
2213 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2214
2215 static int
nfs4_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)2216 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2217 unsigned int flags)
2218 {
2219 struct inode *inode;
2220
2221 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2222 goto full_reval;
2223 if (d_mountpoint(dentry))
2224 goto full_reval;
2225
2226 inode = d_inode(dentry);
2227
2228 /* We can't create new files in nfs_open_revalidate(), so we
2229 * optimize away revalidation of negative dentries.
2230 */
2231 if (inode == NULL)
2232 goto full_reval;
2233
2234 if (nfs_verifier_is_delegated(dentry))
2235 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2236
2237 /* NFS only supports OPEN on regular files */
2238 if (!S_ISREG(inode->i_mode))
2239 goto full_reval;
2240
2241 /* We cannot do exclusive creation on a positive dentry */
2242 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2243 goto reval_dentry;
2244
2245 /* Check if the directory changed */
2246 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2247 goto reval_dentry;
2248
2249 /* Let f_op->open() actually open (and revalidate) the file */
2250 return 1;
2251 reval_dentry:
2252 if (flags & LOOKUP_RCU)
2253 return -ECHILD;
2254 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2255
2256 full_reval:
2257 return nfs_do_lookup_revalidate(dir, dentry, flags);
2258 }
2259
nfs4_lookup_revalidate(struct dentry * dentry,unsigned int flags)2260 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2261 {
2262 return __nfs_lookup_revalidate(dentry, flags,
2263 nfs4_do_lookup_revalidate);
2264 }
2265
2266 #endif /* CONFIG_NFSV4 */
2267
2268 struct dentry *
nfs_add_or_obtain(struct dentry * dentry,struct nfs_fh * fhandle,struct nfs_fattr * fattr)2269 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2270 struct nfs_fattr *fattr)
2271 {
2272 struct dentry *parent = dget_parent(dentry);
2273 struct inode *dir = d_inode(parent);
2274 struct inode *inode;
2275 struct dentry *d;
2276 int error;
2277
2278 d_drop(dentry);
2279
2280 if (fhandle->size == 0) {
2281 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2282 if (error)
2283 goto out_error;
2284 }
2285 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2286 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2287 struct nfs_server *server = NFS_SB(dentry->d_sb);
2288 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2289 fattr, NULL);
2290 if (error < 0)
2291 goto out_error;
2292 }
2293 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2294 d = d_splice_alias(inode, dentry);
2295 out:
2296 dput(parent);
2297 return d;
2298 out_error:
2299 d = ERR_PTR(error);
2300 goto out;
2301 }
2302 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2303
2304 /*
2305 * Code common to create, mkdir, and mknod.
2306 */
nfs_instantiate(struct dentry * dentry,struct nfs_fh * fhandle,struct nfs_fattr * fattr)2307 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2308 struct nfs_fattr *fattr)
2309 {
2310 struct dentry *d;
2311
2312 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2313 if (IS_ERR(d))
2314 return PTR_ERR(d);
2315
2316 /* Callers don't care */
2317 dput(d);
2318 return 0;
2319 }
2320 EXPORT_SYMBOL_GPL(nfs_instantiate);
2321
2322 /*
2323 * Following a failed create operation, we drop the dentry rather
2324 * than retain a negative dentry. This avoids a problem in the event
2325 * that the operation succeeded on the server, but an error in the
2326 * reply path made it appear to have failed.
2327 */
nfs_create(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)2328 int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2329 struct dentry *dentry, umode_t mode, bool excl)
2330 {
2331 struct iattr attr;
2332 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2333 int error;
2334
2335 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2336 dir->i_sb->s_id, dir->i_ino, dentry);
2337
2338 attr.ia_mode = mode;
2339 attr.ia_valid = ATTR_MODE;
2340
2341 trace_nfs_create_enter(dir, dentry, open_flags);
2342 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2343 trace_nfs_create_exit(dir, dentry, open_flags, error);
2344 if (error != 0)
2345 goto out_err;
2346 return 0;
2347 out_err:
2348 d_drop(dentry);
2349 return error;
2350 }
2351 EXPORT_SYMBOL_GPL(nfs_create);
2352
2353 /*
2354 * See comments for nfs_proc_create regarding failed operations.
2355 */
2356 int
nfs_mknod(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)2357 nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2358 struct dentry *dentry, umode_t mode, dev_t rdev)
2359 {
2360 struct iattr attr;
2361 int status;
2362
2363 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2364 dir->i_sb->s_id, dir->i_ino, dentry);
2365
2366 attr.ia_mode = mode;
2367 attr.ia_valid = ATTR_MODE;
2368
2369 trace_nfs_mknod_enter(dir, dentry);
2370 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2371 trace_nfs_mknod_exit(dir, dentry, status);
2372 if (status != 0)
2373 goto out_err;
2374 return 0;
2375 out_err:
2376 d_drop(dentry);
2377 return status;
2378 }
2379 EXPORT_SYMBOL_GPL(nfs_mknod);
2380
2381 /*
2382 * See comments for nfs_proc_create regarding failed operations.
2383 */
nfs_mkdir(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode)2384 int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2385 struct dentry *dentry, umode_t mode)
2386 {
2387 struct iattr attr;
2388 int error;
2389
2390 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2391 dir->i_sb->s_id, dir->i_ino, dentry);
2392
2393 attr.ia_valid = ATTR_MODE;
2394 attr.ia_mode = mode | S_IFDIR;
2395
2396 trace_nfs_mkdir_enter(dir, dentry);
2397 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2398 trace_nfs_mkdir_exit(dir, dentry, error);
2399 if (error != 0)
2400 goto out_err;
2401 return 0;
2402 out_err:
2403 d_drop(dentry);
2404 return error;
2405 }
2406 EXPORT_SYMBOL_GPL(nfs_mkdir);
2407
nfs_dentry_handle_enoent(struct dentry * dentry)2408 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2409 {
2410 if (simple_positive(dentry))
2411 d_delete(dentry);
2412 }
2413
nfs_dentry_remove_handle_error(struct inode * dir,struct dentry * dentry,int error)2414 static void nfs_dentry_remove_handle_error(struct inode *dir,
2415 struct dentry *dentry, int error)
2416 {
2417 switch (error) {
2418 case -ENOENT:
2419 if (d_really_is_positive(dentry))
2420 d_delete(dentry);
2421 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2422 break;
2423 case 0:
2424 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2425 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2426 }
2427 }
2428
nfs_rmdir(struct inode * dir,struct dentry * dentry)2429 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2430 {
2431 int error;
2432
2433 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2434 dir->i_sb->s_id, dir->i_ino, dentry);
2435
2436 trace_nfs_rmdir_enter(dir, dentry);
2437 if (d_really_is_positive(dentry)) {
2438 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2439 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2440 /* Ensure the VFS deletes this inode */
2441 switch (error) {
2442 case 0:
2443 clear_nlink(d_inode(dentry));
2444 break;
2445 case -ENOENT:
2446 nfs_dentry_handle_enoent(dentry);
2447 }
2448 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2449 } else
2450 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2451 nfs_dentry_remove_handle_error(dir, dentry, error);
2452 trace_nfs_rmdir_exit(dir, dentry, error);
2453
2454 return error;
2455 }
2456 EXPORT_SYMBOL_GPL(nfs_rmdir);
2457
2458 /*
2459 * Remove a file after making sure there are no pending writes,
2460 * and after checking that the file has only one user.
2461 *
2462 * We invalidate the attribute cache and free the inode prior to the operation
2463 * to avoid possible races if the server reuses the inode.
2464 */
nfs_safe_remove(struct dentry * dentry)2465 static int nfs_safe_remove(struct dentry *dentry)
2466 {
2467 struct inode *dir = d_inode(dentry->d_parent);
2468 struct inode *inode = d_inode(dentry);
2469 int error = -EBUSY;
2470
2471 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2472
2473 /* If the dentry was sillyrenamed, we simply call d_delete() */
2474 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2475 error = 0;
2476 goto out;
2477 }
2478
2479 trace_nfs_remove_enter(dir, dentry);
2480 if (inode != NULL) {
2481 error = NFS_PROTO(dir)->remove(dir, dentry);
2482 if (error == 0)
2483 nfs_drop_nlink(inode);
2484 } else
2485 error = NFS_PROTO(dir)->remove(dir, dentry);
2486 if (error == -ENOENT)
2487 nfs_dentry_handle_enoent(dentry);
2488 trace_nfs_remove_exit(dir, dentry, error);
2489 out:
2490 return error;
2491 }
2492
2493 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2494 * belongs to an active ".nfs..." file and we return -EBUSY.
2495 *
2496 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2497 */
nfs_unlink(struct inode * dir,struct dentry * dentry)2498 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2499 {
2500 int error;
2501
2502 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2503 dir->i_ino, dentry);
2504
2505 trace_nfs_unlink_enter(dir, dentry);
2506 spin_lock(&dentry->d_lock);
2507 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2508 &NFS_I(d_inode(dentry))->flags)) {
2509 spin_unlock(&dentry->d_lock);
2510 /* Start asynchronous writeout of the inode */
2511 write_inode_now(d_inode(dentry), 0);
2512 error = nfs_sillyrename(dir, dentry);
2513 goto out;
2514 }
2515 /* We must prevent any concurrent open until the unlink
2516 * completes. ->d_revalidate will wait for ->d_fsdata
2517 * to clear. We set it here to ensure no lookup succeeds until
2518 * the unlink is complete on the server.
2519 */
2520 error = -ETXTBSY;
2521 if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2522 WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2523 spin_unlock(&dentry->d_lock);
2524 goto out;
2525 }
2526 block_revalidate(dentry);
2527
2528 spin_unlock(&dentry->d_lock);
2529 error = nfs_safe_remove(dentry);
2530 nfs_dentry_remove_handle_error(dir, dentry, error);
2531 unblock_revalidate(dentry);
2532 out:
2533 trace_nfs_unlink_exit(dir, dentry, error);
2534 return error;
2535 }
2536 EXPORT_SYMBOL_GPL(nfs_unlink);
2537
2538 /*
2539 * To create a symbolic link, most file systems instantiate a new inode,
2540 * add a page to it containing the path, then write it out to the disk
2541 * using prepare_write/commit_write.
2542 *
2543 * Unfortunately the NFS client can't create the in-core inode first
2544 * because it needs a file handle to create an in-core inode (see
2545 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2546 * symlink request has completed on the server.
2547 *
2548 * So instead we allocate a raw page, copy the symname into it, then do
2549 * the SYMLINK request with the page as the buffer. If it succeeds, we
2550 * now have a new file handle and can instantiate an in-core NFS inode
2551 * and move the raw page into its mapping.
2552 */
nfs_symlink(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,const char * symname)2553 int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2554 struct dentry *dentry, const char *symname)
2555 {
2556 struct page *page;
2557 char *kaddr;
2558 struct iattr attr;
2559 unsigned int pathlen = strlen(symname);
2560 int error;
2561
2562 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2563 dir->i_ino, dentry, symname);
2564
2565 if (pathlen > PAGE_SIZE)
2566 return -ENAMETOOLONG;
2567
2568 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2569 attr.ia_valid = ATTR_MODE;
2570
2571 page = alloc_page(GFP_USER);
2572 if (!page)
2573 return -ENOMEM;
2574
2575 kaddr = page_address(page);
2576 memcpy(kaddr, symname, pathlen);
2577 if (pathlen < PAGE_SIZE)
2578 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2579
2580 trace_nfs_symlink_enter(dir, dentry);
2581 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2582 trace_nfs_symlink_exit(dir, dentry, error);
2583 if (error != 0) {
2584 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2585 dir->i_sb->s_id, dir->i_ino,
2586 dentry, symname, error);
2587 d_drop(dentry);
2588 __free_page(page);
2589 return error;
2590 }
2591
2592 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2593
2594 /*
2595 * No big deal if we can't add this page to the page cache here.
2596 * READLINK will get the missing page from the server if needed.
2597 */
2598 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2599 GFP_KERNEL)) {
2600 SetPageUptodate(page);
2601 unlock_page(page);
2602 /*
2603 * add_to_page_cache_lru() grabs an extra page refcount.
2604 * Drop it here to avoid leaking this page later.
2605 */
2606 put_page(page);
2607 } else
2608 __free_page(page);
2609
2610 return 0;
2611 }
2612 EXPORT_SYMBOL_GPL(nfs_symlink);
2613
2614 int
nfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)2615 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2616 {
2617 struct inode *inode = d_inode(old_dentry);
2618 int error;
2619
2620 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2621 old_dentry, dentry);
2622
2623 trace_nfs_link_enter(inode, dir, dentry);
2624 d_drop(dentry);
2625 if (S_ISREG(inode->i_mode))
2626 nfs_sync_inode(inode);
2627 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2628 if (error == 0) {
2629 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2630 ihold(inode);
2631 d_add(dentry, inode);
2632 }
2633 trace_nfs_link_exit(inode, dir, dentry, error);
2634 return error;
2635 }
2636 EXPORT_SYMBOL_GPL(nfs_link);
2637
2638 static void
nfs_unblock_rename(struct rpc_task * task,struct nfs_renamedata * data)2639 nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2640 {
2641 struct dentry *new_dentry = data->new_dentry;
2642
2643 unblock_revalidate(new_dentry);
2644 }
2645
2646 /*
2647 * RENAME
2648 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2649 * different file handle for the same inode after a rename (e.g. when
2650 * moving to a different directory). A fail-safe method to do so would
2651 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2652 * rename the old file using the sillyrename stuff. This way, the original
2653 * file in old_dir will go away when the last process iput()s the inode.
2654 *
2655 * FIXED.
2656 *
2657 * It actually works quite well. One needs to have the possibility for
2658 * at least one ".nfs..." file in each directory the file ever gets
2659 * moved or linked to which happens automagically with the new
2660 * implementation that only depends on the dcache stuff instead of
2661 * using the inode layer
2662 *
2663 * Unfortunately, things are a little more complicated than indicated
2664 * above. For a cross-directory move, we want to make sure we can get
2665 * rid of the old inode after the operation. This means there must be
2666 * no pending writes (if it's a file), and the use count must be 1.
2667 * If these conditions are met, we can drop the dentries before doing
2668 * the rename.
2669 */
nfs_rename(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)2670 int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2671 struct dentry *old_dentry, struct inode *new_dir,
2672 struct dentry *new_dentry, unsigned int flags)
2673 {
2674 struct inode *old_inode = d_inode(old_dentry);
2675 struct inode *new_inode = d_inode(new_dentry);
2676 struct dentry *dentry = NULL;
2677 struct rpc_task *task;
2678 bool must_unblock = false;
2679 int error = -EBUSY;
2680
2681 if (flags)
2682 return -EINVAL;
2683
2684 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2685 old_dentry, new_dentry,
2686 d_count(new_dentry));
2687
2688 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2689 /*
2690 * For non-directories, check whether the target is busy and if so,
2691 * make a copy of the dentry and then do a silly-rename. If the
2692 * silly-rename succeeds, the copied dentry is hashed and becomes
2693 * the new target.
2694 */
2695 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2696 /* We must prevent any concurrent open until the unlink
2697 * completes. ->d_revalidate will wait for ->d_fsdata
2698 * to clear. We set it here to ensure no lookup succeeds until
2699 * the unlink is complete on the server.
2700 */
2701 error = -ETXTBSY;
2702 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2703 WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2704 goto out;
2705
2706 spin_lock(&new_dentry->d_lock);
2707 if (d_count(new_dentry) > 2) {
2708 int err;
2709
2710 spin_unlock(&new_dentry->d_lock);
2711
2712 /* copy the target dentry's name */
2713 dentry = d_alloc(new_dentry->d_parent,
2714 &new_dentry->d_name);
2715 if (!dentry)
2716 goto out;
2717
2718 /* silly-rename the existing target ... */
2719 err = nfs_sillyrename(new_dir, new_dentry);
2720 if (err)
2721 goto out;
2722
2723 new_dentry = dentry;
2724 new_inode = NULL;
2725 } else {
2726 block_revalidate(new_dentry);
2727 must_unblock = true;
2728 spin_unlock(&new_dentry->d_lock);
2729 }
2730
2731 }
2732
2733 if (S_ISREG(old_inode->i_mode))
2734 nfs_sync_inode(old_inode);
2735 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2736 must_unblock ? nfs_unblock_rename : NULL);
2737 if (IS_ERR(task)) {
2738 if (must_unblock)
2739 unblock_revalidate(new_dentry);
2740 error = PTR_ERR(task);
2741 goto out;
2742 }
2743
2744 error = rpc_wait_for_completion_task(task);
2745 if (error != 0) {
2746 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2747 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2748 smp_wmb();
2749 } else
2750 error = task->tk_status;
2751 rpc_put_task(task);
2752 /* Ensure the inode attributes are revalidated */
2753 if (error == 0) {
2754 spin_lock(&old_inode->i_lock);
2755 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2756 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2757 NFS_INO_INVALID_CTIME |
2758 NFS_INO_REVAL_FORCED);
2759 spin_unlock(&old_inode->i_lock);
2760 }
2761 out:
2762 trace_nfs_rename_exit(old_dir, old_dentry,
2763 new_dir, new_dentry, error);
2764 if (!error) {
2765 if (new_inode != NULL)
2766 nfs_drop_nlink(new_inode);
2767 /*
2768 * The d_move() should be here instead of in an async RPC completion
2769 * handler because we need the proper locks to move the dentry. If
2770 * we're interrupted by a signal, the async RPC completion handler
2771 * should mark the directories for revalidation.
2772 */
2773 d_move(old_dentry, new_dentry);
2774 nfs_set_verifier(old_dentry,
2775 nfs_save_change_attribute(new_dir));
2776 } else if (error == -ENOENT)
2777 nfs_dentry_handle_enoent(old_dentry);
2778
2779 /* new dentry created? */
2780 if (dentry)
2781 dput(dentry);
2782 return error;
2783 }
2784 EXPORT_SYMBOL_GPL(nfs_rename);
2785
2786 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2787 static LIST_HEAD(nfs_access_lru_list);
2788 static atomic_long_t nfs_access_nr_entries;
2789
2790 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2791 module_param(nfs_access_max_cachesize, ulong, 0644);
2792 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2793
nfs_access_free_entry(struct nfs_access_entry * entry)2794 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2795 {
2796 put_group_info(entry->group_info);
2797 kfree_rcu(entry, rcu_head);
2798 smp_mb__before_atomic();
2799 atomic_long_dec(&nfs_access_nr_entries);
2800 smp_mb__after_atomic();
2801 }
2802
nfs_access_free_list(struct list_head * head)2803 static void nfs_access_free_list(struct list_head *head)
2804 {
2805 struct nfs_access_entry *cache;
2806
2807 while (!list_empty(head)) {
2808 cache = list_entry(head->next, struct nfs_access_entry, lru);
2809 list_del(&cache->lru);
2810 nfs_access_free_entry(cache);
2811 }
2812 }
2813
2814 static unsigned long
nfs_do_access_cache_scan(unsigned int nr_to_scan)2815 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2816 {
2817 LIST_HEAD(head);
2818 struct nfs_inode *nfsi, *next;
2819 struct nfs_access_entry *cache;
2820 long freed = 0;
2821
2822 spin_lock(&nfs_access_lru_lock);
2823 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2824 struct inode *inode;
2825
2826 if (nr_to_scan-- == 0)
2827 break;
2828 inode = &nfsi->vfs_inode;
2829 spin_lock(&inode->i_lock);
2830 if (list_empty(&nfsi->access_cache_entry_lru))
2831 goto remove_lru_entry;
2832 cache = list_entry(nfsi->access_cache_entry_lru.next,
2833 struct nfs_access_entry, lru);
2834 list_move(&cache->lru, &head);
2835 rb_erase(&cache->rb_node, &nfsi->access_cache);
2836 freed++;
2837 if (!list_empty(&nfsi->access_cache_entry_lru))
2838 list_move_tail(&nfsi->access_cache_inode_lru,
2839 &nfs_access_lru_list);
2840 else {
2841 remove_lru_entry:
2842 list_del_init(&nfsi->access_cache_inode_lru);
2843 smp_mb__before_atomic();
2844 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2845 smp_mb__after_atomic();
2846 }
2847 spin_unlock(&inode->i_lock);
2848 }
2849 spin_unlock(&nfs_access_lru_lock);
2850 nfs_access_free_list(&head);
2851 return freed;
2852 }
2853
2854 unsigned long
nfs_access_cache_scan(struct shrinker * shrink,struct shrink_control * sc)2855 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2856 {
2857 int nr_to_scan = sc->nr_to_scan;
2858 gfp_t gfp_mask = sc->gfp_mask;
2859
2860 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2861 return SHRINK_STOP;
2862 return nfs_do_access_cache_scan(nr_to_scan);
2863 }
2864
2865
2866 unsigned long
nfs_access_cache_count(struct shrinker * shrink,struct shrink_control * sc)2867 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2868 {
2869 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2870 }
2871
2872 static void
nfs_access_cache_enforce_limit(void)2873 nfs_access_cache_enforce_limit(void)
2874 {
2875 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2876 unsigned long diff;
2877 unsigned int nr_to_scan;
2878
2879 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2880 return;
2881 nr_to_scan = 100;
2882 diff = nr_entries - nfs_access_max_cachesize;
2883 if (diff < nr_to_scan)
2884 nr_to_scan = diff;
2885 nfs_do_access_cache_scan(nr_to_scan);
2886 }
2887
__nfs_access_zap_cache(struct nfs_inode * nfsi,struct list_head * head)2888 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2889 {
2890 struct rb_root *root_node = &nfsi->access_cache;
2891 struct rb_node *n;
2892 struct nfs_access_entry *entry;
2893
2894 /* Unhook entries from the cache */
2895 while ((n = rb_first(root_node)) != NULL) {
2896 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2897 rb_erase(n, root_node);
2898 list_move(&entry->lru, head);
2899 }
2900 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2901 }
2902
nfs_access_zap_cache(struct inode * inode)2903 void nfs_access_zap_cache(struct inode *inode)
2904 {
2905 LIST_HEAD(head);
2906
2907 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2908 return;
2909 /* Remove from global LRU init */
2910 spin_lock(&nfs_access_lru_lock);
2911 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2912 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2913
2914 spin_lock(&inode->i_lock);
2915 __nfs_access_zap_cache(NFS_I(inode), &head);
2916 spin_unlock(&inode->i_lock);
2917 spin_unlock(&nfs_access_lru_lock);
2918 nfs_access_free_list(&head);
2919 }
2920 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2921
access_cmp(const struct cred * a,const struct nfs_access_entry * b)2922 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2923 {
2924 struct group_info *ga, *gb;
2925 int g;
2926
2927 if (uid_lt(a->fsuid, b->fsuid))
2928 return -1;
2929 if (uid_gt(a->fsuid, b->fsuid))
2930 return 1;
2931
2932 if (gid_lt(a->fsgid, b->fsgid))
2933 return -1;
2934 if (gid_gt(a->fsgid, b->fsgid))
2935 return 1;
2936
2937 ga = a->group_info;
2938 gb = b->group_info;
2939 if (ga == gb)
2940 return 0;
2941 if (ga == NULL)
2942 return -1;
2943 if (gb == NULL)
2944 return 1;
2945 if (ga->ngroups < gb->ngroups)
2946 return -1;
2947 if (ga->ngroups > gb->ngroups)
2948 return 1;
2949
2950 for (g = 0; g < ga->ngroups; g++) {
2951 if (gid_lt(ga->gid[g], gb->gid[g]))
2952 return -1;
2953 if (gid_gt(ga->gid[g], gb->gid[g]))
2954 return 1;
2955 }
2956 return 0;
2957 }
2958
nfs_access_search_rbtree(struct inode * inode,const struct cred * cred)2959 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2960 {
2961 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2962
2963 while (n != NULL) {
2964 struct nfs_access_entry *entry =
2965 rb_entry(n, struct nfs_access_entry, rb_node);
2966 int cmp = access_cmp(cred, entry);
2967
2968 if (cmp < 0)
2969 n = n->rb_left;
2970 else if (cmp > 0)
2971 n = n->rb_right;
2972 else
2973 return entry;
2974 }
2975 return NULL;
2976 }
2977
nfs_access_login_time(const struct task_struct * task,const struct cred * cred)2978 static u64 nfs_access_login_time(const struct task_struct *task,
2979 const struct cred *cred)
2980 {
2981 const struct task_struct *parent;
2982 const struct cred *pcred;
2983 u64 ret;
2984
2985 rcu_read_lock();
2986 for (;;) {
2987 parent = rcu_dereference(task->real_parent);
2988 pcred = __task_cred(parent);
2989 if (parent == task || cred_fscmp(pcred, cred) != 0)
2990 break;
2991 task = parent;
2992 }
2993 ret = task->start_time;
2994 rcu_read_unlock();
2995 return ret;
2996 }
2997
nfs_access_get_cached_locked(struct inode * inode,const struct cred * cred,u32 * mask,bool may_block)2998 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2999 {
3000 struct nfs_inode *nfsi = NFS_I(inode);
3001 u64 login_time = nfs_access_login_time(current, cred);
3002 struct nfs_access_entry *cache;
3003 bool retry = true;
3004 int err;
3005
3006 spin_lock(&inode->i_lock);
3007 for(;;) {
3008 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3009 goto out_zap;
3010 cache = nfs_access_search_rbtree(inode, cred);
3011 err = -ENOENT;
3012 if (cache == NULL)
3013 goto out;
3014 /* Found an entry, is our attribute cache valid? */
3015 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3016 break;
3017 if (!retry)
3018 break;
3019 err = -ECHILD;
3020 if (!may_block)
3021 goto out;
3022 spin_unlock(&inode->i_lock);
3023 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3024 if (err)
3025 return err;
3026 spin_lock(&inode->i_lock);
3027 retry = false;
3028 }
3029 err = -ENOENT;
3030 if ((s64)(login_time - cache->timestamp) > 0)
3031 goto out;
3032 *mask = cache->mask;
3033 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3034 err = 0;
3035 out:
3036 spin_unlock(&inode->i_lock);
3037 return err;
3038 out_zap:
3039 spin_unlock(&inode->i_lock);
3040 nfs_access_zap_cache(inode);
3041 return -ENOENT;
3042 }
3043
nfs_access_get_cached_rcu(struct inode * inode,const struct cred * cred,u32 * mask)3044 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3045 {
3046 /* Only check the most recently returned cache entry,
3047 * but do it without locking.
3048 */
3049 struct nfs_inode *nfsi = NFS_I(inode);
3050 u64 login_time = nfs_access_login_time(current, cred);
3051 struct nfs_access_entry *cache;
3052 int err = -ECHILD;
3053 struct list_head *lh;
3054
3055 rcu_read_lock();
3056 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3057 goto out;
3058 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3059 cache = list_entry(lh, struct nfs_access_entry, lru);
3060 if (lh == &nfsi->access_cache_entry_lru ||
3061 access_cmp(cred, cache) != 0)
3062 cache = NULL;
3063 if (cache == NULL)
3064 goto out;
3065 if ((s64)(login_time - cache->timestamp) > 0)
3066 goto out;
3067 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3068 goto out;
3069 *mask = cache->mask;
3070 err = 0;
3071 out:
3072 rcu_read_unlock();
3073 return err;
3074 }
3075
nfs_access_get_cached(struct inode * inode,const struct cred * cred,u32 * mask,bool may_block)3076 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3077 u32 *mask, bool may_block)
3078 {
3079 int status;
3080
3081 status = nfs_access_get_cached_rcu(inode, cred, mask);
3082 if (status != 0)
3083 status = nfs_access_get_cached_locked(inode, cred, mask,
3084 may_block);
3085
3086 return status;
3087 }
3088 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3089
nfs_access_add_rbtree(struct inode * inode,struct nfs_access_entry * set,const struct cred * cred)3090 static void nfs_access_add_rbtree(struct inode *inode,
3091 struct nfs_access_entry *set,
3092 const struct cred *cred)
3093 {
3094 struct nfs_inode *nfsi = NFS_I(inode);
3095 struct rb_root *root_node = &nfsi->access_cache;
3096 struct rb_node **p = &root_node->rb_node;
3097 struct rb_node *parent = NULL;
3098 struct nfs_access_entry *entry;
3099 int cmp;
3100
3101 spin_lock(&inode->i_lock);
3102 while (*p != NULL) {
3103 parent = *p;
3104 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3105 cmp = access_cmp(cred, entry);
3106
3107 if (cmp < 0)
3108 p = &parent->rb_left;
3109 else if (cmp > 0)
3110 p = &parent->rb_right;
3111 else
3112 goto found;
3113 }
3114 rb_link_node(&set->rb_node, parent, p);
3115 rb_insert_color(&set->rb_node, root_node);
3116 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3117 spin_unlock(&inode->i_lock);
3118 return;
3119 found:
3120 rb_replace_node(parent, &set->rb_node, root_node);
3121 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3122 list_del(&entry->lru);
3123 spin_unlock(&inode->i_lock);
3124 nfs_access_free_entry(entry);
3125 }
3126
nfs_access_add_cache(struct inode * inode,struct nfs_access_entry * set,const struct cred * cred)3127 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3128 const struct cred *cred)
3129 {
3130 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3131 if (cache == NULL)
3132 return;
3133 RB_CLEAR_NODE(&cache->rb_node);
3134 cache->fsuid = cred->fsuid;
3135 cache->fsgid = cred->fsgid;
3136 cache->group_info = get_group_info(cred->group_info);
3137 cache->mask = set->mask;
3138 cache->timestamp = ktime_get_ns();
3139
3140 /* The above field assignments must be visible
3141 * before this item appears on the lru. We cannot easily
3142 * use rcu_assign_pointer, so just force the memory barrier.
3143 */
3144 smp_wmb();
3145 nfs_access_add_rbtree(inode, cache, cred);
3146
3147 /* Update accounting */
3148 smp_mb__before_atomic();
3149 atomic_long_inc(&nfs_access_nr_entries);
3150 smp_mb__after_atomic();
3151
3152 /* Add inode to global LRU list */
3153 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3154 spin_lock(&nfs_access_lru_lock);
3155 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3156 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3157 &nfs_access_lru_list);
3158 spin_unlock(&nfs_access_lru_lock);
3159 }
3160 nfs_access_cache_enforce_limit();
3161 }
3162 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3163
3164 #define NFS_MAY_READ (NFS_ACCESS_READ)
3165 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3166 NFS_ACCESS_EXTEND | \
3167 NFS_ACCESS_DELETE)
3168 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3169 NFS_ACCESS_EXTEND)
3170 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3171 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3172 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3173 static int
nfs_access_calc_mask(u32 access_result,umode_t umode)3174 nfs_access_calc_mask(u32 access_result, umode_t umode)
3175 {
3176 int mask = 0;
3177
3178 if (access_result & NFS_MAY_READ)
3179 mask |= MAY_READ;
3180 if (S_ISDIR(umode)) {
3181 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3182 mask |= MAY_WRITE;
3183 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3184 mask |= MAY_EXEC;
3185 } else if (S_ISREG(umode)) {
3186 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3187 mask |= MAY_WRITE;
3188 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3189 mask |= MAY_EXEC;
3190 } else if (access_result & NFS_MAY_WRITE)
3191 mask |= MAY_WRITE;
3192 return mask;
3193 }
3194
nfs_access_set_mask(struct nfs_access_entry * entry,u32 access_result)3195 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3196 {
3197 entry->mask = access_result;
3198 }
3199 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3200
nfs_do_access(struct inode * inode,const struct cred * cred,int mask)3201 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3202 {
3203 struct nfs_access_entry cache;
3204 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3205 int cache_mask = -1;
3206 int status;
3207
3208 trace_nfs_access_enter(inode);
3209
3210 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3211 if (status == 0)
3212 goto out_cached;
3213
3214 status = -ECHILD;
3215 if (!may_block)
3216 goto out;
3217
3218 /*
3219 * Determine which access bits we want to ask for...
3220 */
3221 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3222 nfs_access_xattr_mask(NFS_SERVER(inode));
3223 if (S_ISDIR(inode->i_mode))
3224 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3225 else
3226 cache.mask |= NFS_ACCESS_EXECUTE;
3227 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3228 if (status != 0) {
3229 if (status == -ESTALE) {
3230 if (!S_ISDIR(inode->i_mode))
3231 nfs_set_inode_stale(inode);
3232 else
3233 nfs_zap_caches(inode);
3234 }
3235 goto out;
3236 }
3237 nfs_access_add_cache(inode, &cache, cred);
3238 out_cached:
3239 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3240 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3241 status = -EACCES;
3242 out:
3243 trace_nfs_access_exit(inode, mask, cache_mask, status);
3244 return status;
3245 }
3246
nfs_open_permission_mask(int openflags)3247 static int nfs_open_permission_mask(int openflags)
3248 {
3249 int mask = 0;
3250
3251 if (openflags & __FMODE_EXEC) {
3252 /* ONLY check exec rights */
3253 mask = MAY_EXEC;
3254 } else {
3255 if ((openflags & O_ACCMODE) != O_WRONLY)
3256 mask |= MAY_READ;
3257 if ((openflags & O_ACCMODE) != O_RDONLY)
3258 mask |= MAY_WRITE;
3259 }
3260
3261 return mask;
3262 }
3263
nfs_may_open(struct inode * inode,const struct cred * cred,int openflags)3264 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3265 {
3266 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3267 }
3268 EXPORT_SYMBOL_GPL(nfs_may_open);
3269
nfs_execute_ok(struct inode * inode,int mask)3270 static int nfs_execute_ok(struct inode *inode, int mask)
3271 {
3272 struct nfs_server *server = NFS_SERVER(inode);
3273 int ret = 0;
3274
3275 if (S_ISDIR(inode->i_mode))
3276 return 0;
3277 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3278 if (mask & MAY_NOT_BLOCK)
3279 return -ECHILD;
3280 ret = __nfs_revalidate_inode(server, inode);
3281 }
3282 if (ret == 0 && !execute_ok(inode))
3283 ret = -EACCES;
3284 return ret;
3285 }
3286
nfs_permission(struct mnt_idmap * idmap,struct inode * inode,int mask)3287 int nfs_permission(struct mnt_idmap *idmap,
3288 struct inode *inode,
3289 int mask)
3290 {
3291 const struct cred *cred = current_cred();
3292 int res = 0;
3293
3294 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3295
3296 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3297 goto out;
3298 /* Is this sys_access() ? */
3299 if (mask & (MAY_ACCESS | MAY_CHDIR))
3300 goto force_lookup;
3301
3302 switch (inode->i_mode & S_IFMT) {
3303 case S_IFLNK:
3304 goto out;
3305 case S_IFREG:
3306 if ((mask & MAY_OPEN) &&
3307 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3308 return 0;
3309 break;
3310 case S_IFDIR:
3311 /*
3312 * Optimize away all write operations, since the server
3313 * will check permissions when we perform the op.
3314 */
3315 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3316 goto out;
3317 }
3318
3319 force_lookup:
3320 if (!NFS_PROTO(inode)->access)
3321 goto out_notsup;
3322
3323 res = nfs_do_access(inode, cred, mask);
3324 out:
3325 if (!res && (mask & MAY_EXEC))
3326 res = nfs_execute_ok(inode, mask);
3327
3328 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3329 inode->i_sb->s_id, inode->i_ino, mask, res);
3330 return res;
3331 out_notsup:
3332 if (mask & MAY_NOT_BLOCK)
3333 return -ECHILD;
3334
3335 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3336 NFS_INO_INVALID_OTHER);
3337 if (res == 0)
3338 res = generic_permission(&nop_mnt_idmap, inode, mask);
3339 goto out;
3340 }
3341 EXPORT_SYMBOL_GPL(nfs_permission);
3342