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