Loading...
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/module.h>
22#include <linux/time.h>
23#include <linux/errno.h>
24#include <linux/stat.h>
25#include <linux/fcntl.h>
26#include <linux/string.h>
27#include <linux/kernel.h>
28#include <linux/slab.h>
29#include <linux/mm.h>
30#include <linux/sunrpc/clnt.h>
31#include <linux/nfs_fs.h>
32#include <linux/nfs_mount.h>
33#include <linux/pagemap.h>
34#include <linux/pagevec.h>
35#include <linux/namei.h>
36#include <linux/mount.h>
37#include <linux/swap.h>
38#include <linux/sched.h>
39#include <linux/kmemleak.h>
40#include <linux/xattr.h>
41
42#include "delegation.h"
43#include "iostat.h"
44#include "internal.h"
45#include "fscache.h"
46
47#include "nfstrace.h"
48
49/* #define NFS_DEBUG_VERBOSE 1 */
50
51static int nfs_opendir(struct inode *, struct file *);
52static int nfs_closedir(struct inode *, struct file *);
53static int nfs_readdir(struct file *, struct dir_context *);
54static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
55static loff_t nfs_llseek_dir(struct file *, loff_t, int);
56static void nfs_readdir_clear_array(struct page*);
57
58const struct file_operations nfs_dir_operations = {
59 .llseek = nfs_llseek_dir,
60 .read = generic_read_dir,
61 .iterate_shared = nfs_readdir,
62 .open = nfs_opendir,
63 .release = nfs_closedir,
64 .fsync = nfs_fsync_dir,
65};
66
67const struct address_space_operations nfs_dir_aops = {
68 .freepage = nfs_readdir_clear_array,
69};
70
71static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir)
72{
73 struct nfs_inode *nfsi = NFS_I(dir);
74 struct nfs_open_dir_context *ctx;
75 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
76 if (ctx != NULL) {
77 ctx->duped = 0;
78 ctx->attr_gencount = nfsi->attr_gencount;
79 ctx->dir_cookie = 0;
80 ctx->dup_cookie = 0;
81 spin_lock(&dir->i_lock);
82 if (list_empty(&nfsi->open_files) &&
83 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
84 nfs_set_cache_invalid(dir,
85 NFS_INO_INVALID_DATA |
86 NFS_INO_REVAL_FORCED);
87 list_add(&ctx->list, &nfsi->open_files);
88 spin_unlock(&dir->i_lock);
89 return ctx;
90 }
91 return ERR_PTR(-ENOMEM);
92}
93
94static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
95{
96 spin_lock(&dir->i_lock);
97 list_del(&ctx->list);
98 spin_unlock(&dir->i_lock);
99 kfree(ctx);
100}
101
102/*
103 * Open file
104 */
105static int
106nfs_opendir(struct inode *inode, struct file *filp)
107{
108 int res = 0;
109 struct nfs_open_dir_context *ctx;
110
111 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
112
113 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
114
115 ctx = alloc_nfs_open_dir_context(inode);
116 if (IS_ERR(ctx)) {
117 res = PTR_ERR(ctx);
118 goto out;
119 }
120 filp->private_data = ctx;
121out:
122 return res;
123}
124
125static int
126nfs_closedir(struct inode *inode, struct file *filp)
127{
128 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
129 return 0;
130}
131
132struct nfs_cache_array_entry {
133 u64 cookie;
134 u64 ino;
135 const char *name;
136 unsigned int name_len;
137 unsigned char d_type;
138};
139
140struct nfs_cache_array {
141 u64 last_cookie;
142 unsigned int size;
143 unsigned char page_full : 1,
144 page_is_eof : 1,
145 cookies_are_ordered : 1;
146 struct nfs_cache_array_entry array[];
147};
148
149struct nfs_readdir_descriptor {
150 struct file *file;
151 struct page *page;
152 struct dir_context *ctx;
153 pgoff_t page_index;
154 u64 dir_cookie;
155 u64 last_cookie;
156 u64 dup_cookie;
157 loff_t current_index;
158 loff_t prev_index;
159
160 __be32 verf[NFS_DIR_VERIFIER_SIZE];
161 unsigned long dir_verifier;
162 unsigned long timestamp;
163 unsigned long gencount;
164 unsigned long attr_gencount;
165 unsigned int cache_entry_index;
166 signed char duped;
167 bool plus;
168 bool eof;
169};
170
171static void nfs_readdir_array_init(struct nfs_cache_array *array)
172{
173 memset(array, 0, sizeof(struct nfs_cache_array));
174}
175
176static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie)
177{
178 struct nfs_cache_array *array;
179
180 array = kmap_atomic(page);
181 nfs_readdir_array_init(array);
182 array->last_cookie = last_cookie;
183 array->cookies_are_ordered = 1;
184 kunmap_atomic(array);
185}
186
187/*
188 * we are freeing strings created by nfs_add_to_readdir_array()
189 */
190static
191void nfs_readdir_clear_array(struct page *page)
192{
193 struct nfs_cache_array *array;
194 int i;
195
196 array = kmap_atomic(page);
197 for (i = 0; i < array->size; i++)
198 kfree(array->array[i].name);
199 nfs_readdir_array_init(array);
200 kunmap_atomic(array);
201}
202
203static struct page *
204nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
205{
206 struct page *page = alloc_page(gfp_flags);
207 if (page)
208 nfs_readdir_page_init_array(page, last_cookie);
209 return page;
210}
211
212static void nfs_readdir_page_array_free(struct page *page)
213{
214 if (page) {
215 nfs_readdir_clear_array(page);
216 put_page(page);
217 }
218}
219
220static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
221{
222 array->page_is_eof = 1;
223 array->page_full = 1;
224}
225
226static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
227{
228 return array->page_full;
229}
230
231/*
232 * the caller is responsible for freeing qstr.name
233 * when called by nfs_readdir_add_to_array, the strings will be freed in
234 * nfs_clear_readdir_array()
235 */
236static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
237{
238 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
239
240 /*
241 * Avoid a kmemleak false positive. The pointer to the name is stored
242 * in a page cache page which kmemleak does not scan.
243 */
244 if (ret != NULL)
245 kmemleak_not_leak(ret);
246 return ret;
247}
248
249/*
250 * Check that the next array entry lies entirely within the page bounds
251 */
252static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
253{
254 struct nfs_cache_array_entry *cache_entry;
255
256 if (array->page_full)
257 return -ENOSPC;
258 cache_entry = &array->array[array->size + 1];
259 if ((char *)cache_entry - (char *)array > PAGE_SIZE) {
260 array->page_full = 1;
261 return -ENOSPC;
262 }
263 return 0;
264}
265
266static
267int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
268{
269 struct nfs_cache_array *array;
270 struct nfs_cache_array_entry *cache_entry;
271 const char *name;
272 int ret;
273
274 name = nfs_readdir_copy_name(entry->name, entry->len);
275 if (!name)
276 return -ENOMEM;
277
278 array = kmap_atomic(page);
279 ret = nfs_readdir_array_can_expand(array);
280 if (ret) {
281 kfree(name);
282 goto out;
283 }
284
285 cache_entry = &array->array[array->size];
286 cache_entry->cookie = entry->prev_cookie;
287 cache_entry->ino = entry->ino;
288 cache_entry->d_type = entry->d_type;
289 cache_entry->name_len = entry->len;
290 cache_entry->name = name;
291 array->last_cookie = entry->cookie;
292 if (array->last_cookie <= cache_entry->cookie)
293 array->cookies_are_ordered = 0;
294 array->size++;
295 if (entry->eof != 0)
296 nfs_readdir_array_set_eof(array);
297out:
298 kunmap_atomic(array);
299 return ret;
300}
301
302static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
303 pgoff_t index, u64 last_cookie)
304{
305 struct page *page;
306
307 page = grab_cache_page(mapping, index);
308 if (page && !PageUptodate(page)) {
309 nfs_readdir_page_init_array(page, last_cookie);
310 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0)
311 nfs_zap_mapping(mapping->host, mapping);
312 SetPageUptodate(page);
313 }
314
315 return page;
316}
317
318static u64 nfs_readdir_page_last_cookie(struct page *page)
319{
320 struct nfs_cache_array *array;
321 u64 ret;
322
323 array = kmap_atomic(page);
324 ret = array->last_cookie;
325 kunmap_atomic(array);
326 return ret;
327}
328
329static bool nfs_readdir_page_needs_filling(struct page *page)
330{
331 struct nfs_cache_array *array;
332 bool ret;
333
334 array = kmap_atomic(page);
335 ret = !nfs_readdir_array_is_full(array);
336 kunmap_atomic(array);
337 return ret;
338}
339
340static void nfs_readdir_page_set_eof(struct page *page)
341{
342 struct nfs_cache_array *array;
343
344 array = kmap_atomic(page);
345 nfs_readdir_array_set_eof(array);
346 kunmap_atomic(array);
347}
348
349static void nfs_readdir_page_unlock_and_put(struct page *page)
350{
351 unlock_page(page);
352 put_page(page);
353}
354
355static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
356 pgoff_t index, u64 cookie)
357{
358 struct page *page;
359
360 page = nfs_readdir_page_get_locked(mapping, index, cookie);
361 if (page) {
362 if (nfs_readdir_page_last_cookie(page) == cookie)
363 return page;
364 nfs_readdir_page_unlock_and_put(page);
365 }
366 return NULL;
367}
368
369static inline
370int is_32bit_api(void)
371{
372#ifdef CONFIG_COMPAT
373 return in_compat_syscall();
374#else
375 return (BITS_PER_LONG == 32);
376#endif
377}
378
379static
380bool nfs_readdir_use_cookie(const struct file *filp)
381{
382 if ((filp->f_mode & FMODE_32BITHASH) ||
383 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
384 return false;
385 return true;
386}
387
388static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
389 struct nfs_readdir_descriptor *desc)
390{
391 loff_t diff = desc->ctx->pos - desc->current_index;
392 unsigned int index;
393
394 if (diff < 0)
395 goto out_eof;
396 if (diff >= array->size) {
397 if (array->page_is_eof)
398 goto out_eof;
399 return -EAGAIN;
400 }
401
402 index = (unsigned int)diff;
403 desc->dir_cookie = array->array[index].cookie;
404 desc->cache_entry_index = index;
405 return 0;
406out_eof:
407 desc->eof = true;
408 return -EBADCOOKIE;
409}
410
411static bool
412nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
413{
414 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
415 return false;
416 smp_rmb();
417 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
418}
419
420static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
421 u64 cookie)
422{
423 if (!array->cookies_are_ordered)
424 return true;
425 /* Optimisation for monotonically increasing cookies */
426 if (cookie >= array->last_cookie)
427 return false;
428 if (array->size && cookie < array->array[0].cookie)
429 return false;
430 return true;
431}
432
433static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
434 struct nfs_readdir_descriptor *desc)
435{
436 int i;
437 loff_t new_pos;
438 int status = -EAGAIN;
439
440 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
441 goto check_eof;
442
443 for (i = 0; i < array->size; i++) {
444 if (array->array[i].cookie == desc->dir_cookie) {
445 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
446
447 new_pos = desc->current_index + i;
448 if (desc->attr_gencount != nfsi->attr_gencount ||
449 !nfs_readdir_inode_mapping_valid(nfsi)) {
450 desc->duped = 0;
451 desc->attr_gencount = nfsi->attr_gencount;
452 } else if (new_pos < desc->prev_index) {
453 if (desc->duped > 0
454 && desc->dup_cookie == desc->dir_cookie) {
455 if (printk_ratelimit()) {
456 pr_notice("NFS: directory %pD2 contains a readdir loop."
457 "Please contact your server vendor. "
458 "The file: %s has duplicate cookie %llu\n",
459 desc->file, array->array[i].name, desc->dir_cookie);
460 }
461 status = -ELOOP;
462 goto out;
463 }
464 desc->dup_cookie = desc->dir_cookie;
465 desc->duped = -1;
466 }
467 if (nfs_readdir_use_cookie(desc->file))
468 desc->ctx->pos = desc->dir_cookie;
469 else
470 desc->ctx->pos = new_pos;
471 desc->prev_index = new_pos;
472 desc->cache_entry_index = i;
473 return 0;
474 }
475 }
476check_eof:
477 if (array->page_is_eof) {
478 status = -EBADCOOKIE;
479 if (desc->dir_cookie == array->last_cookie)
480 desc->eof = true;
481 }
482out:
483 return status;
484}
485
486static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
487{
488 struct nfs_cache_array *array;
489 int status;
490
491 array = kmap_atomic(desc->page);
492
493 if (desc->dir_cookie == 0)
494 status = nfs_readdir_search_for_pos(array, desc);
495 else
496 status = nfs_readdir_search_for_cookie(array, desc);
497
498 if (status == -EAGAIN) {
499 desc->last_cookie = array->last_cookie;
500 desc->current_index += array->size;
501 desc->page_index++;
502 }
503 kunmap_atomic(array);
504 return status;
505}
506
507/* Fill a page with xdr information before transferring to the cache page */
508static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
509 __be32 *verf, u64 cookie,
510 struct page **pages, size_t bufsize,
511 __be32 *verf_res)
512{
513 struct inode *inode = file_inode(desc->file);
514 struct nfs_readdir_arg arg = {
515 .dentry = file_dentry(desc->file),
516 .cred = desc->file->f_cred,
517 .verf = verf,
518 .cookie = cookie,
519 .pages = pages,
520 .page_len = bufsize,
521 .plus = desc->plus,
522 };
523 struct nfs_readdir_res res = {
524 .verf = verf_res,
525 };
526 unsigned long timestamp, gencount;
527 int error;
528
529 again:
530 timestamp = jiffies;
531 gencount = nfs_inc_attr_generation_counter();
532 desc->dir_verifier = nfs_save_change_attribute(inode);
533 error = NFS_PROTO(inode)->readdir(&arg, &res);
534 if (error < 0) {
535 /* We requested READDIRPLUS, but the server doesn't grok it */
536 if (error == -ENOTSUPP && desc->plus) {
537 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
538 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
539 desc->plus = arg.plus = false;
540 goto again;
541 }
542 goto error;
543 }
544 desc->timestamp = timestamp;
545 desc->gencount = gencount;
546error:
547 return error;
548}
549
550static int xdr_decode(struct nfs_readdir_descriptor *desc,
551 struct nfs_entry *entry, struct xdr_stream *xdr)
552{
553 struct inode *inode = file_inode(desc->file);
554 int error;
555
556 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
557 if (error)
558 return error;
559 entry->fattr->time_start = desc->timestamp;
560 entry->fattr->gencount = desc->gencount;
561 return 0;
562}
563
564/* Match file and dirent using either filehandle or fileid
565 * Note: caller is responsible for checking the fsid
566 */
567static
568int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
569{
570 struct inode *inode;
571 struct nfs_inode *nfsi;
572
573 if (d_really_is_negative(dentry))
574 return 0;
575
576 inode = d_inode(dentry);
577 if (is_bad_inode(inode) || NFS_STALE(inode))
578 return 0;
579
580 nfsi = NFS_I(inode);
581 if (entry->fattr->fileid != nfsi->fileid)
582 return 0;
583 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
584 return 0;
585 return 1;
586}
587
588static
589bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
590{
591 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
592 return false;
593 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
594 return true;
595 if (ctx->pos == 0)
596 return true;
597 return false;
598}
599
600/*
601 * This function is called by the lookup and getattr code to request the
602 * use of readdirplus to accelerate any future lookups in the same
603 * directory.
604 */
605void nfs_advise_use_readdirplus(struct inode *dir)
606{
607 struct nfs_inode *nfsi = NFS_I(dir);
608
609 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
610 !list_empty(&nfsi->open_files))
611 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
612}
613
614/*
615 * This function is mainly for use by nfs_getattr().
616 *
617 * If this is an 'ls -l', we want to force use of readdirplus.
618 * Do this by checking if there is an active file descriptor
619 * and calling nfs_advise_use_readdirplus, then forcing a
620 * cache flush.
621 */
622void nfs_force_use_readdirplus(struct inode *dir)
623{
624 struct nfs_inode *nfsi = NFS_I(dir);
625
626 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
627 !list_empty(&nfsi->open_files)) {
628 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
629 invalidate_mapping_pages(dir->i_mapping,
630 nfsi->page_index + 1, -1);
631 }
632}
633
634static
635void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
636 unsigned long dir_verifier)
637{
638 struct qstr filename = QSTR_INIT(entry->name, entry->len);
639 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
640 struct dentry *dentry;
641 struct dentry *alias;
642 struct inode *inode;
643 int status;
644
645 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
646 return;
647 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
648 return;
649 if (filename.len == 0)
650 return;
651 /* Validate that the name doesn't contain any illegal '\0' */
652 if (strnlen(filename.name, filename.len) != filename.len)
653 return;
654 /* ...or '/' */
655 if (strnchr(filename.name, filename.len, '/'))
656 return;
657 if (filename.name[0] == '.') {
658 if (filename.len == 1)
659 return;
660 if (filename.len == 2 && filename.name[1] == '.')
661 return;
662 }
663 filename.hash = full_name_hash(parent, filename.name, filename.len);
664
665 dentry = d_lookup(parent, &filename);
666again:
667 if (!dentry) {
668 dentry = d_alloc_parallel(parent, &filename, &wq);
669 if (IS_ERR(dentry))
670 return;
671 }
672 if (!d_in_lookup(dentry)) {
673 /* Is there a mountpoint here? If so, just exit */
674 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
675 &entry->fattr->fsid))
676 goto out;
677 if (nfs_same_file(dentry, entry)) {
678 if (!entry->fh->size)
679 goto out;
680 nfs_set_verifier(dentry, dir_verifier);
681 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
682 if (!status)
683 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
684 goto out;
685 } else {
686 d_invalidate(dentry);
687 dput(dentry);
688 dentry = NULL;
689 goto again;
690 }
691 }
692 if (!entry->fh->size) {
693 d_lookup_done(dentry);
694 goto out;
695 }
696
697 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
698 alias = d_splice_alias(inode, dentry);
699 d_lookup_done(dentry);
700 if (alias) {
701 if (IS_ERR(alias))
702 goto out;
703 dput(dentry);
704 dentry = alias;
705 }
706 nfs_set_verifier(dentry, dir_verifier);
707out:
708 dput(dentry);
709}
710
711/* Perform conversion from xdr to cache array */
712static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
713 struct nfs_entry *entry,
714 struct page **xdr_pages,
715 unsigned int buflen,
716 struct page **arrays,
717 size_t narrays)
718{
719 struct address_space *mapping = desc->file->f_mapping;
720 struct xdr_stream stream;
721 struct xdr_buf buf;
722 struct page *scratch, *new, *page = *arrays;
723 int status;
724
725 scratch = alloc_page(GFP_KERNEL);
726 if (scratch == NULL)
727 return -ENOMEM;
728
729 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
730 xdr_set_scratch_page(&stream, scratch);
731
732 do {
733 if (entry->label)
734 entry->label->len = NFS4_MAXLABELLEN;
735
736 status = xdr_decode(desc, entry, &stream);
737 if (status != 0)
738 break;
739
740 if (desc->plus)
741 nfs_prime_dcache(file_dentry(desc->file), entry,
742 desc->dir_verifier);
743
744 status = nfs_readdir_add_to_array(entry, page);
745 if (status != -ENOSPC)
746 continue;
747
748 if (page->mapping != mapping) {
749 if (!--narrays)
750 break;
751 new = nfs_readdir_page_array_alloc(entry->prev_cookie,
752 GFP_KERNEL);
753 if (!new)
754 break;
755 arrays++;
756 *arrays = page = new;
757 } else {
758 new = nfs_readdir_page_get_next(mapping,
759 page->index + 1,
760 entry->prev_cookie);
761 if (!new)
762 break;
763 if (page != *arrays)
764 nfs_readdir_page_unlock_and_put(page);
765 page = new;
766 }
767 status = nfs_readdir_add_to_array(entry, page);
768 } while (!status && !entry->eof);
769
770 switch (status) {
771 case -EBADCOOKIE:
772 if (entry->eof) {
773 nfs_readdir_page_set_eof(page);
774 status = 0;
775 }
776 break;
777 case -ENOSPC:
778 case -EAGAIN:
779 status = 0;
780 break;
781 }
782
783 if (page != *arrays)
784 nfs_readdir_page_unlock_and_put(page);
785
786 put_page(scratch);
787 return status;
788}
789
790static void nfs_readdir_free_pages(struct page **pages, size_t npages)
791{
792 while (npages--)
793 put_page(pages[npages]);
794 kfree(pages);
795}
796
797/*
798 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
799 * to nfs_readdir_free_pages()
800 */
801static struct page **nfs_readdir_alloc_pages(size_t npages)
802{
803 struct page **pages;
804 size_t i;
805
806 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
807 if (!pages)
808 return NULL;
809 for (i = 0; i < npages; i++) {
810 struct page *page = alloc_page(GFP_KERNEL);
811 if (page == NULL)
812 goto out_freepages;
813 pages[i] = page;
814 }
815 return pages;
816
817out_freepages:
818 nfs_readdir_free_pages(pages, i);
819 return NULL;
820}
821
822static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
823 __be32 *verf_arg, __be32 *verf_res,
824 struct page **arrays, size_t narrays)
825{
826 struct page **pages;
827 struct page *page = *arrays;
828 struct nfs_entry *entry;
829 size_t array_size;
830 struct inode *inode = file_inode(desc->file);
831 size_t dtsize = NFS_SERVER(inode)->dtsize;
832 int status = -ENOMEM;
833
834 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
835 if (!entry)
836 return -ENOMEM;
837 entry->cookie = nfs_readdir_page_last_cookie(page);
838 entry->fh = nfs_alloc_fhandle();
839 entry->fattr = nfs_alloc_fattr();
840 entry->server = NFS_SERVER(inode);
841 if (entry->fh == NULL || entry->fattr == NULL)
842 goto out;
843
844 entry->label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
845 if (IS_ERR(entry->label)) {
846 status = PTR_ERR(entry->label);
847 goto out;
848 }
849
850 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
851 pages = nfs_readdir_alloc_pages(array_size);
852 if (!pages)
853 goto out_release_label;
854
855 do {
856 unsigned int pglen;
857 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie,
858 pages, dtsize,
859 verf_res);
860 if (status < 0)
861 break;
862
863 pglen = status;
864 if (pglen == 0) {
865 nfs_readdir_page_set_eof(page);
866 break;
867 }
868
869 verf_arg = verf_res;
870
871 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
872 arrays, narrays);
873 } while (!status && nfs_readdir_page_needs_filling(page));
874
875 nfs_readdir_free_pages(pages, array_size);
876out_release_label:
877 nfs4_label_free(entry->label);
878out:
879 nfs_free_fattr(entry->fattr);
880 nfs_free_fhandle(entry->fh);
881 kfree(entry);
882 return status;
883}
884
885static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
886{
887 put_page(desc->page);
888 desc->page = NULL;
889}
890
891static void
892nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
893{
894 unlock_page(desc->page);
895 nfs_readdir_page_put(desc);
896}
897
898static struct page *
899nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
900{
901 return nfs_readdir_page_get_locked(desc->file->f_mapping,
902 desc->page_index,
903 desc->last_cookie);
904}
905
906/*
907 * Returns 0 if desc->dir_cookie was found on page desc->page_index
908 * and locks the page to prevent removal from the page cache.
909 */
910static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
911{
912 struct inode *inode = file_inode(desc->file);
913 struct nfs_inode *nfsi = NFS_I(inode);
914 __be32 verf[NFS_DIR_VERIFIER_SIZE];
915 int res;
916
917 desc->page = nfs_readdir_page_get_cached(desc);
918 if (!desc->page)
919 return -ENOMEM;
920 if (nfs_readdir_page_needs_filling(desc->page)) {
921 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
922 &desc->page, 1);
923 if (res < 0) {
924 nfs_readdir_page_unlock_and_put_cached(desc);
925 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
926 invalidate_inode_pages2(desc->file->f_mapping);
927 desc->page_index = 0;
928 return -EAGAIN;
929 }
930 return res;
931 }
932 /*
933 * Set the cookie verifier if the page cache was empty
934 */
935 if (desc->page_index == 0)
936 memcpy(nfsi->cookieverf, verf,
937 sizeof(nfsi->cookieverf));
938 }
939 res = nfs_readdir_search_array(desc);
940 if (res == 0) {
941 nfsi->page_index = desc->page_index;
942 return 0;
943 }
944 nfs_readdir_page_unlock_and_put_cached(desc);
945 return res;
946}
947
948static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc)
949{
950 struct address_space *mapping = desc->file->f_mapping;
951 struct inode *dir = file_inode(desc->file);
952 unsigned int dtsize = NFS_SERVER(dir)->dtsize;
953 loff_t size = i_size_read(dir);
954
955 /*
956 * Default to uncached readdir if the page cache is empty, and
957 * we're looking for a non-zero cookie in a large directory.
958 */
959 return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize;
960}
961
962/* Search for desc->dir_cookie from the beginning of the page cache */
963static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
964{
965 int res;
966
967 if (nfs_readdir_dont_search_cache(desc))
968 return -EBADCOOKIE;
969
970 do {
971 if (desc->page_index == 0) {
972 desc->current_index = 0;
973 desc->prev_index = 0;
974 desc->last_cookie = 0;
975 }
976 res = find_and_lock_cache_page(desc);
977 } while (res == -EAGAIN);
978 return res;
979}
980
981/*
982 * Once we've found the start of the dirent within a page: fill 'er up...
983 */
984static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
985 const __be32 *verf)
986{
987 struct file *file = desc->file;
988 struct nfs_cache_array *array;
989 unsigned int i = 0;
990
991 array = kmap(desc->page);
992 for (i = desc->cache_entry_index; i < array->size; i++) {
993 struct nfs_cache_array_entry *ent;
994
995 ent = &array->array[i];
996 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
997 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
998 desc->eof = true;
999 break;
1000 }
1001 memcpy(desc->verf, verf, sizeof(desc->verf));
1002 if (i < (array->size-1))
1003 desc->dir_cookie = array->array[i+1].cookie;
1004 else
1005 desc->dir_cookie = array->last_cookie;
1006 if (nfs_readdir_use_cookie(file))
1007 desc->ctx->pos = desc->dir_cookie;
1008 else
1009 desc->ctx->pos++;
1010 if (desc->duped != 0)
1011 desc->duped = 1;
1012 }
1013 if (array->page_is_eof)
1014 desc->eof = true;
1015
1016 kunmap(desc->page);
1017 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1018 (unsigned long long)desc->dir_cookie);
1019}
1020
1021/*
1022 * If we cannot find a cookie in our cache, we suspect that this is
1023 * because it points to a deleted file, so we ask the server to return
1024 * whatever it thinks is the next entry. We then feed this to filldir.
1025 * If all goes well, we should then be able to find our way round the
1026 * cache on the next call to readdir_search_pagecache();
1027 *
1028 * NOTE: we cannot add the anonymous page to the pagecache because
1029 * the data it contains might not be page aligned. Besides,
1030 * we should already have a complete representation of the
1031 * directory in the page cache by the time we get here.
1032 */
1033static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1034{
1035 struct page **arrays;
1036 size_t i, sz = 512;
1037 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1038 int status = -ENOMEM;
1039
1040 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1041 (unsigned long long)desc->dir_cookie);
1042
1043 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1044 if (!arrays)
1045 goto out;
1046 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1047 if (!arrays[0])
1048 goto out;
1049
1050 desc->page_index = 0;
1051 desc->last_cookie = desc->dir_cookie;
1052 desc->duped = 0;
1053
1054 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1055
1056 for (i = 0; !desc->eof && i < sz && arrays[i]; i++) {
1057 desc->page = arrays[i];
1058 nfs_do_filldir(desc, verf);
1059 }
1060 desc->page = NULL;
1061
1062
1063 for (i = 0; i < sz && arrays[i]; i++)
1064 nfs_readdir_page_array_free(arrays[i]);
1065out:
1066 kfree(arrays);
1067 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1068 return status;
1069}
1070
1071/* The file offset position represents the dirent entry number. A
1072 last cookie cache takes care of the common case of reading the
1073 whole directory.
1074 */
1075static int nfs_readdir(struct file *file, struct dir_context *ctx)
1076{
1077 struct dentry *dentry = file_dentry(file);
1078 struct inode *inode = d_inode(dentry);
1079 struct nfs_inode *nfsi = NFS_I(inode);
1080 struct nfs_open_dir_context *dir_ctx = file->private_data;
1081 struct nfs_readdir_descriptor *desc;
1082 int res;
1083
1084 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1085 file, (long long)ctx->pos);
1086 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1087
1088 /*
1089 * ctx->pos points to the dirent entry number.
1090 * *desc->dir_cookie has the cookie for the next entry. We have
1091 * to either find the entry with the appropriate number or
1092 * revalidate the cookie.
1093 */
1094 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) {
1095 res = nfs_revalidate_mapping(inode, file->f_mapping);
1096 if (res < 0)
1097 goto out;
1098 }
1099
1100 res = -ENOMEM;
1101 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1102 if (!desc)
1103 goto out;
1104 desc->file = file;
1105 desc->ctx = ctx;
1106 desc->plus = nfs_use_readdirplus(inode, ctx);
1107
1108 spin_lock(&file->f_lock);
1109 desc->dir_cookie = dir_ctx->dir_cookie;
1110 desc->dup_cookie = dir_ctx->dup_cookie;
1111 desc->duped = dir_ctx->duped;
1112 desc->attr_gencount = dir_ctx->attr_gencount;
1113 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1114 spin_unlock(&file->f_lock);
1115
1116 do {
1117 res = readdir_search_pagecache(desc);
1118
1119 if (res == -EBADCOOKIE) {
1120 res = 0;
1121 /* This means either end of directory */
1122 if (desc->dir_cookie && !desc->eof) {
1123 /* Or that the server has 'lost' a cookie */
1124 res = uncached_readdir(desc);
1125 if (res == 0)
1126 continue;
1127 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1128 res = 0;
1129 }
1130 break;
1131 }
1132 if (res == -ETOOSMALL && desc->plus) {
1133 clear_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
1134 nfs_zap_caches(inode);
1135 desc->page_index = 0;
1136 desc->plus = false;
1137 desc->eof = false;
1138 continue;
1139 }
1140 if (res < 0)
1141 break;
1142
1143 nfs_do_filldir(desc, nfsi->cookieverf);
1144 nfs_readdir_page_unlock_and_put_cached(desc);
1145 } while (!desc->eof);
1146
1147 spin_lock(&file->f_lock);
1148 dir_ctx->dir_cookie = desc->dir_cookie;
1149 dir_ctx->dup_cookie = desc->dup_cookie;
1150 dir_ctx->duped = desc->duped;
1151 dir_ctx->attr_gencount = desc->attr_gencount;
1152 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1153 spin_unlock(&file->f_lock);
1154
1155 kfree(desc);
1156
1157out:
1158 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1159 return res;
1160}
1161
1162static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1163{
1164 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1165
1166 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1167 filp, offset, whence);
1168
1169 switch (whence) {
1170 default:
1171 return -EINVAL;
1172 case SEEK_SET:
1173 if (offset < 0)
1174 return -EINVAL;
1175 spin_lock(&filp->f_lock);
1176 break;
1177 case SEEK_CUR:
1178 if (offset == 0)
1179 return filp->f_pos;
1180 spin_lock(&filp->f_lock);
1181 offset += filp->f_pos;
1182 if (offset < 0) {
1183 spin_unlock(&filp->f_lock);
1184 return -EINVAL;
1185 }
1186 }
1187 if (offset != filp->f_pos) {
1188 filp->f_pos = offset;
1189 if (nfs_readdir_use_cookie(filp))
1190 dir_ctx->dir_cookie = offset;
1191 else
1192 dir_ctx->dir_cookie = 0;
1193 if (offset == 0)
1194 memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf));
1195 dir_ctx->duped = 0;
1196 }
1197 spin_unlock(&filp->f_lock);
1198 return offset;
1199}
1200
1201/*
1202 * All directory operations under NFS are synchronous, so fsync()
1203 * is a dummy operation.
1204 */
1205static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1206 int datasync)
1207{
1208 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1209
1210 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1211 return 0;
1212}
1213
1214/**
1215 * nfs_force_lookup_revalidate - Mark the directory as having changed
1216 * @dir: pointer to directory inode
1217 *
1218 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1219 * full lookup on all child dentries of 'dir' whenever a change occurs
1220 * on the server that might have invalidated our dcache.
1221 *
1222 * Note that we reserve bit '0' as a tag to let us know when a dentry
1223 * was revalidated while holding a delegation on its inode.
1224 *
1225 * The caller should be holding dir->i_lock
1226 */
1227void nfs_force_lookup_revalidate(struct inode *dir)
1228{
1229 NFS_I(dir)->cache_change_attribute += 2;
1230}
1231EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1232
1233/**
1234 * nfs_verify_change_attribute - Detects NFS remote directory changes
1235 * @dir: pointer to parent directory inode
1236 * @verf: previously saved change attribute
1237 *
1238 * Return "false" if the verifiers doesn't match the change attribute.
1239 * This would usually indicate that the directory contents have changed on
1240 * the server, and that any dentries need revalidating.
1241 */
1242static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1243{
1244 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1245}
1246
1247static void nfs_set_verifier_delegated(unsigned long *verf)
1248{
1249 *verf |= 1UL;
1250}
1251
1252#if IS_ENABLED(CONFIG_NFS_V4)
1253static void nfs_unset_verifier_delegated(unsigned long *verf)
1254{
1255 *verf &= ~1UL;
1256}
1257#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1258
1259static bool nfs_test_verifier_delegated(unsigned long verf)
1260{
1261 return verf & 1;
1262}
1263
1264static bool nfs_verifier_is_delegated(struct dentry *dentry)
1265{
1266 return nfs_test_verifier_delegated(dentry->d_time);
1267}
1268
1269static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1270{
1271 struct inode *inode = d_inode(dentry);
1272
1273 if (!nfs_verifier_is_delegated(dentry) &&
1274 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf))
1275 goto out;
1276 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1277 nfs_set_verifier_delegated(&verf);
1278out:
1279 dentry->d_time = verf;
1280}
1281
1282/**
1283 * nfs_set_verifier - save a parent directory verifier in the dentry
1284 * @dentry: pointer to dentry
1285 * @verf: verifier to save
1286 *
1287 * Saves the parent directory verifier in @dentry. If the inode has
1288 * a delegation, we also tag the dentry as having been revalidated
1289 * while holding a delegation so that we know we don't have to
1290 * look it up again after a directory change.
1291 */
1292void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1293{
1294
1295 spin_lock(&dentry->d_lock);
1296 nfs_set_verifier_locked(dentry, verf);
1297 spin_unlock(&dentry->d_lock);
1298}
1299EXPORT_SYMBOL_GPL(nfs_set_verifier);
1300
1301#if IS_ENABLED(CONFIG_NFS_V4)
1302/**
1303 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1304 * @inode: pointer to inode
1305 *
1306 * Iterates through the dentries in the inode alias list and clears
1307 * the tag used to indicate that the dentry has been revalidated
1308 * while holding a delegation.
1309 * This function is intended for use when the delegation is being
1310 * returned or revoked.
1311 */
1312void nfs_clear_verifier_delegated(struct inode *inode)
1313{
1314 struct dentry *alias;
1315
1316 if (!inode)
1317 return;
1318 spin_lock(&inode->i_lock);
1319 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1320 spin_lock(&alias->d_lock);
1321 nfs_unset_verifier_delegated(&alias->d_time);
1322 spin_unlock(&alias->d_lock);
1323 }
1324 spin_unlock(&inode->i_lock);
1325}
1326EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1327#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1328
1329/*
1330 * A check for whether or not the parent directory has changed.
1331 * In the case it has, we assume that the dentries are untrustworthy
1332 * and may need to be looked up again.
1333 * If rcu_walk prevents us from performing a full check, return 0.
1334 */
1335static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1336 int rcu_walk)
1337{
1338 if (IS_ROOT(dentry))
1339 return 1;
1340 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1341 return 0;
1342 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1343 return 0;
1344 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1345 if (nfs_mapping_need_revalidate_inode(dir)) {
1346 if (rcu_walk)
1347 return 0;
1348 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1349 return 0;
1350 }
1351 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1352 return 0;
1353 return 1;
1354}
1355
1356/*
1357 * Use intent information to check whether or not we're going to do
1358 * an O_EXCL create using this path component.
1359 */
1360static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1361{
1362 if (NFS_PROTO(dir)->version == 2)
1363 return 0;
1364 return flags & LOOKUP_EXCL;
1365}
1366
1367/*
1368 * Inode and filehandle revalidation for lookups.
1369 *
1370 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1371 * or if the intent information indicates that we're about to open this
1372 * particular file and the "nocto" mount flag is not set.
1373 *
1374 */
1375static
1376int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1377{
1378 struct nfs_server *server = NFS_SERVER(inode);
1379 int ret;
1380
1381 if (IS_AUTOMOUNT(inode))
1382 return 0;
1383
1384 if (flags & LOOKUP_OPEN) {
1385 switch (inode->i_mode & S_IFMT) {
1386 case S_IFREG:
1387 /* A NFSv4 OPEN will revalidate later */
1388 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1389 goto out;
1390 fallthrough;
1391 case S_IFDIR:
1392 if (server->flags & NFS_MOUNT_NOCTO)
1393 break;
1394 /* NFS close-to-open cache consistency validation */
1395 goto out_force;
1396 }
1397 }
1398
1399 /* VFS wants an on-the-wire revalidation */
1400 if (flags & LOOKUP_REVAL)
1401 goto out_force;
1402out:
1403 return (inode->i_nlink == 0) ? -ESTALE : 0;
1404out_force:
1405 if (flags & LOOKUP_RCU)
1406 return -ECHILD;
1407 ret = __nfs_revalidate_inode(server, inode);
1408 if (ret != 0)
1409 return ret;
1410 goto out;
1411}
1412
1413static void nfs_mark_dir_for_revalidate(struct inode *inode)
1414{
1415 spin_lock(&inode->i_lock);
1416 nfs_set_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE);
1417 spin_unlock(&inode->i_lock);
1418}
1419
1420/*
1421 * We judge how long we want to trust negative
1422 * dentries by looking at the parent inode mtime.
1423 *
1424 * If parent mtime has changed, we revalidate, else we wait for a
1425 * period corresponding to the parent's attribute cache timeout value.
1426 *
1427 * If LOOKUP_RCU prevents us from performing a full check, return 1
1428 * suggesting a reval is needed.
1429 *
1430 * Note that when creating a new file, or looking up a rename target,
1431 * then it shouldn't be necessary to revalidate a negative dentry.
1432 */
1433static inline
1434int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1435 unsigned int flags)
1436{
1437 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1438 return 0;
1439 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1440 return 1;
1441 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1442}
1443
1444static int
1445nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1446 struct inode *inode, int error)
1447{
1448 switch (error) {
1449 case 1:
1450 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1451 __func__, dentry);
1452 return 1;
1453 case 0:
1454 /*
1455 * We can't d_drop the root of a disconnected tree:
1456 * its d_hash is on the s_anon list and d_drop() would hide
1457 * it from shrink_dcache_for_unmount(), leading to busy
1458 * inodes on unmount and further oopses.
1459 */
1460 if (inode && IS_ROOT(dentry))
1461 return 1;
1462 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1463 __func__, dentry);
1464 return 0;
1465 }
1466 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1467 __func__, dentry, error);
1468 return error;
1469}
1470
1471static int
1472nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1473 unsigned int flags)
1474{
1475 int ret = 1;
1476 if (nfs_neg_need_reval(dir, dentry, flags)) {
1477 if (flags & LOOKUP_RCU)
1478 return -ECHILD;
1479 ret = 0;
1480 }
1481 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1482}
1483
1484static int
1485nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1486 struct inode *inode)
1487{
1488 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1489 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1490}
1491
1492static int
1493nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1494 struct inode *inode)
1495{
1496 struct nfs_fh *fhandle;
1497 struct nfs_fattr *fattr;
1498 struct nfs4_label *label;
1499 unsigned long dir_verifier;
1500 int ret;
1501
1502 ret = -ENOMEM;
1503 fhandle = nfs_alloc_fhandle();
1504 fattr = nfs_alloc_fattr();
1505 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1506 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1507 goto out;
1508
1509 dir_verifier = nfs_save_change_attribute(dir);
1510 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1511 if (ret < 0) {
1512 switch (ret) {
1513 case -ESTALE:
1514 case -ENOENT:
1515 ret = 0;
1516 break;
1517 case -ETIMEDOUT:
1518 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1519 ret = 1;
1520 }
1521 goto out;
1522 }
1523 ret = 0;
1524 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1525 goto out;
1526 if (nfs_refresh_inode(inode, fattr) < 0)
1527 goto out;
1528
1529 nfs_setsecurity(inode, fattr, label);
1530 nfs_set_verifier(dentry, dir_verifier);
1531
1532 /* set a readdirplus hint that we had a cache miss */
1533 nfs_force_use_readdirplus(dir);
1534 ret = 1;
1535out:
1536 nfs_free_fattr(fattr);
1537 nfs_free_fhandle(fhandle);
1538 nfs4_label_free(label);
1539
1540 /*
1541 * If the lookup failed despite the dentry change attribute being
1542 * a match, then we should revalidate the directory cache.
1543 */
1544 if (!ret && nfs_verify_change_attribute(dir, dentry->d_time))
1545 nfs_mark_dir_for_revalidate(dir);
1546 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1547}
1548
1549/*
1550 * This is called every time the dcache has a lookup hit,
1551 * and we should check whether we can really trust that
1552 * lookup.
1553 *
1554 * NOTE! The hit can be a negative hit too, don't assume
1555 * we have an inode!
1556 *
1557 * If the parent directory is seen to have changed, we throw out the
1558 * cached dentry and do a new lookup.
1559 */
1560static int
1561nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1562 unsigned int flags)
1563{
1564 struct inode *inode;
1565 int error;
1566
1567 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1568 inode = d_inode(dentry);
1569
1570 if (!inode)
1571 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1572
1573 if (is_bad_inode(inode)) {
1574 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1575 __func__, dentry);
1576 goto out_bad;
1577 }
1578
1579 if (nfs_verifier_is_delegated(dentry))
1580 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1581
1582 /* Force a full look up iff the parent directory has changed */
1583 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1584 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1585 error = nfs_lookup_verify_inode(inode, flags);
1586 if (error) {
1587 if (error == -ESTALE)
1588 nfs_mark_dir_for_revalidate(dir);
1589 goto out_bad;
1590 }
1591 nfs_advise_use_readdirplus(dir);
1592 goto out_valid;
1593 }
1594
1595 if (flags & LOOKUP_RCU)
1596 return -ECHILD;
1597
1598 if (NFS_STALE(inode))
1599 goto out_bad;
1600
1601 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1602 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1603 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1604 return error;
1605out_valid:
1606 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1607out_bad:
1608 if (flags & LOOKUP_RCU)
1609 return -ECHILD;
1610 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1611}
1612
1613static int
1614__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1615 int (*reval)(struct inode *, struct dentry *, unsigned int))
1616{
1617 struct dentry *parent;
1618 struct inode *dir;
1619 int ret;
1620
1621 if (flags & LOOKUP_RCU) {
1622 parent = READ_ONCE(dentry->d_parent);
1623 dir = d_inode_rcu(parent);
1624 if (!dir)
1625 return -ECHILD;
1626 ret = reval(dir, dentry, flags);
1627 if (parent != READ_ONCE(dentry->d_parent))
1628 return -ECHILD;
1629 } else {
1630 parent = dget_parent(dentry);
1631 ret = reval(d_inode(parent), dentry, flags);
1632 dput(parent);
1633 }
1634 return ret;
1635}
1636
1637static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1638{
1639 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1640}
1641
1642/*
1643 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1644 * when we don't really care about the dentry name. This is called when a
1645 * pathwalk ends on a dentry that was not found via a normal lookup in the
1646 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1647 *
1648 * In this situation, we just want to verify that the inode itself is OK
1649 * since the dentry might have changed on the server.
1650 */
1651static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1652{
1653 struct inode *inode = d_inode(dentry);
1654 int error = 0;
1655
1656 /*
1657 * I believe we can only get a negative dentry here in the case of a
1658 * procfs-style symlink. Just assume it's correct for now, but we may
1659 * eventually need to do something more here.
1660 */
1661 if (!inode) {
1662 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1663 __func__, dentry);
1664 return 1;
1665 }
1666
1667 if (is_bad_inode(inode)) {
1668 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1669 __func__, dentry);
1670 return 0;
1671 }
1672
1673 error = nfs_lookup_verify_inode(inode, flags);
1674 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1675 __func__, inode->i_ino, error ? "invalid" : "valid");
1676 return !error;
1677}
1678
1679/*
1680 * This is called from dput() when d_count is going to 0.
1681 */
1682static int nfs_dentry_delete(const struct dentry *dentry)
1683{
1684 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1685 dentry, dentry->d_flags);
1686
1687 /* Unhash any dentry with a stale inode */
1688 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1689 return 1;
1690
1691 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1692 /* Unhash it, so that ->d_iput() would be called */
1693 return 1;
1694 }
1695 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1696 /* Unhash it, so that ancestors of killed async unlink
1697 * files will be cleaned up during umount */
1698 return 1;
1699 }
1700 return 0;
1701
1702}
1703
1704/* Ensure that we revalidate inode->i_nlink */
1705static void nfs_drop_nlink(struct inode *inode)
1706{
1707 spin_lock(&inode->i_lock);
1708 /* drop the inode if we're reasonably sure this is the last link */
1709 if (inode->i_nlink > 0)
1710 drop_nlink(inode);
1711 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1712 nfs_set_cache_invalid(
1713 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1714 NFS_INO_INVALID_NLINK);
1715 spin_unlock(&inode->i_lock);
1716}
1717
1718/*
1719 * Called when the dentry loses inode.
1720 * We use it to clean up silly-renamed files.
1721 */
1722static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1723{
1724 if (S_ISDIR(inode->i_mode))
1725 /* drop any readdir cache as it could easily be old */
1726 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1727
1728 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1729 nfs_complete_unlink(dentry, inode);
1730 nfs_drop_nlink(inode);
1731 }
1732 iput(inode);
1733}
1734
1735static void nfs_d_release(struct dentry *dentry)
1736{
1737 /* free cached devname value, if it survived that far */
1738 if (unlikely(dentry->d_fsdata)) {
1739 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1740 WARN_ON(1);
1741 else
1742 kfree(dentry->d_fsdata);
1743 }
1744}
1745
1746const struct dentry_operations nfs_dentry_operations = {
1747 .d_revalidate = nfs_lookup_revalidate,
1748 .d_weak_revalidate = nfs_weak_revalidate,
1749 .d_delete = nfs_dentry_delete,
1750 .d_iput = nfs_dentry_iput,
1751 .d_automount = nfs_d_automount,
1752 .d_release = nfs_d_release,
1753};
1754EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1755
1756struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1757{
1758 struct dentry *res;
1759 struct inode *inode = NULL;
1760 struct nfs_fh *fhandle = NULL;
1761 struct nfs_fattr *fattr = NULL;
1762 struct nfs4_label *label = NULL;
1763 unsigned long dir_verifier;
1764 int error;
1765
1766 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1767 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1768
1769 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1770 return ERR_PTR(-ENAMETOOLONG);
1771
1772 /*
1773 * If we're doing an exclusive create, optimize away the lookup
1774 * but don't hash the dentry.
1775 */
1776 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1777 return NULL;
1778
1779 res = ERR_PTR(-ENOMEM);
1780 fhandle = nfs_alloc_fhandle();
1781 fattr = nfs_alloc_fattr();
1782 if (fhandle == NULL || fattr == NULL)
1783 goto out;
1784
1785 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1786 if (IS_ERR(label))
1787 goto out;
1788
1789 dir_verifier = nfs_save_change_attribute(dir);
1790 trace_nfs_lookup_enter(dir, dentry, flags);
1791 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1792 if (error == -ENOENT)
1793 goto no_entry;
1794 if (error < 0) {
1795 res = ERR_PTR(error);
1796 goto out_label;
1797 }
1798 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1799 res = ERR_CAST(inode);
1800 if (IS_ERR(res))
1801 goto out_label;
1802
1803 /* Notify readdir to use READDIRPLUS */
1804 nfs_force_use_readdirplus(dir);
1805
1806no_entry:
1807 res = d_splice_alias(inode, dentry);
1808 if (res != NULL) {
1809 if (IS_ERR(res))
1810 goto out_label;
1811 dentry = res;
1812 }
1813 nfs_set_verifier(dentry, dir_verifier);
1814out_label:
1815 trace_nfs_lookup_exit(dir, dentry, flags, error);
1816 nfs4_label_free(label);
1817out:
1818 nfs_free_fattr(fattr);
1819 nfs_free_fhandle(fhandle);
1820 return res;
1821}
1822EXPORT_SYMBOL_GPL(nfs_lookup);
1823
1824#if IS_ENABLED(CONFIG_NFS_V4)
1825static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1826
1827const struct dentry_operations nfs4_dentry_operations = {
1828 .d_revalidate = nfs4_lookup_revalidate,
1829 .d_weak_revalidate = nfs_weak_revalidate,
1830 .d_delete = nfs_dentry_delete,
1831 .d_iput = nfs_dentry_iput,
1832 .d_automount = nfs_d_automount,
1833 .d_release = nfs_d_release,
1834};
1835EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1836
1837static fmode_t flags_to_mode(int flags)
1838{
1839 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1840 if ((flags & O_ACCMODE) != O_WRONLY)
1841 res |= FMODE_READ;
1842 if ((flags & O_ACCMODE) != O_RDONLY)
1843 res |= FMODE_WRITE;
1844 return res;
1845}
1846
1847static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1848{
1849 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1850}
1851
1852static int do_open(struct inode *inode, struct file *filp)
1853{
1854 nfs_fscache_open_file(inode, filp);
1855 return 0;
1856}
1857
1858static int nfs_finish_open(struct nfs_open_context *ctx,
1859 struct dentry *dentry,
1860 struct file *file, unsigned open_flags)
1861{
1862 int err;
1863
1864 err = finish_open(file, dentry, do_open);
1865 if (err)
1866 goto out;
1867 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1868 nfs_file_set_open_context(file, ctx);
1869 else
1870 err = -EOPENSTALE;
1871out:
1872 return err;
1873}
1874
1875int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1876 struct file *file, unsigned open_flags,
1877 umode_t mode)
1878{
1879 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1880 struct nfs_open_context *ctx;
1881 struct dentry *res;
1882 struct iattr attr = { .ia_valid = ATTR_OPEN };
1883 struct inode *inode;
1884 unsigned int lookup_flags = 0;
1885 bool switched = false;
1886 int created = 0;
1887 int err;
1888
1889 /* Expect a negative dentry */
1890 BUG_ON(d_inode(dentry));
1891
1892 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1893 dir->i_sb->s_id, dir->i_ino, dentry);
1894
1895 err = nfs_check_flags(open_flags);
1896 if (err)
1897 return err;
1898
1899 /* NFS only supports OPEN on regular files */
1900 if ((open_flags & O_DIRECTORY)) {
1901 if (!d_in_lookup(dentry)) {
1902 /*
1903 * Hashed negative dentry with O_DIRECTORY: dentry was
1904 * revalidated and is fine, no need to perform lookup
1905 * again
1906 */
1907 return -ENOENT;
1908 }
1909 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1910 goto no_open;
1911 }
1912
1913 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1914 return -ENAMETOOLONG;
1915
1916 if (open_flags & O_CREAT) {
1917 struct nfs_server *server = NFS_SERVER(dir);
1918
1919 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1920 mode &= ~current_umask();
1921
1922 attr.ia_valid |= ATTR_MODE;
1923 attr.ia_mode = mode;
1924 }
1925 if (open_flags & O_TRUNC) {
1926 attr.ia_valid |= ATTR_SIZE;
1927 attr.ia_size = 0;
1928 }
1929
1930 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1931 d_drop(dentry);
1932 switched = true;
1933 dentry = d_alloc_parallel(dentry->d_parent,
1934 &dentry->d_name, &wq);
1935 if (IS_ERR(dentry))
1936 return PTR_ERR(dentry);
1937 if (unlikely(!d_in_lookup(dentry)))
1938 return finish_no_open(file, dentry);
1939 }
1940
1941 ctx = create_nfs_open_context(dentry, open_flags, file);
1942 err = PTR_ERR(ctx);
1943 if (IS_ERR(ctx))
1944 goto out;
1945
1946 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1947 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1948 if (created)
1949 file->f_mode |= FMODE_CREATED;
1950 if (IS_ERR(inode)) {
1951 err = PTR_ERR(inode);
1952 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1953 put_nfs_open_context(ctx);
1954 d_drop(dentry);
1955 switch (err) {
1956 case -ENOENT:
1957 d_splice_alias(NULL, dentry);
1958 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1959 break;
1960 case -EISDIR:
1961 case -ENOTDIR:
1962 goto no_open;
1963 case -ELOOP:
1964 if (!(open_flags & O_NOFOLLOW))
1965 goto no_open;
1966 break;
1967 /* case -EINVAL: */
1968 default:
1969 break;
1970 }
1971 goto out;
1972 }
1973
1974 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1975 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1976 put_nfs_open_context(ctx);
1977out:
1978 if (unlikely(switched)) {
1979 d_lookup_done(dentry);
1980 dput(dentry);
1981 }
1982 return err;
1983
1984no_open:
1985 res = nfs_lookup(dir, dentry, lookup_flags);
1986 if (switched) {
1987 d_lookup_done(dentry);
1988 if (!res)
1989 res = dentry;
1990 else
1991 dput(dentry);
1992 }
1993 if (IS_ERR(res))
1994 return PTR_ERR(res);
1995 return finish_no_open(file, res);
1996}
1997EXPORT_SYMBOL_GPL(nfs_atomic_open);
1998
1999static int
2000nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2001 unsigned int flags)
2002{
2003 struct inode *inode;
2004
2005 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2006 goto full_reval;
2007 if (d_mountpoint(dentry))
2008 goto full_reval;
2009
2010 inode = d_inode(dentry);
2011
2012 /* We can't create new files in nfs_open_revalidate(), so we
2013 * optimize away revalidation of negative dentries.
2014 */
2015 if (inode == NULL)
2016 goto full_reval;
2017
2018 if (nfs_verifier_is_delegated(dentry))
2019 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2020
2021 /* NFS only supports OPEN on regular files */
2022 if (!S_ISREG(inode->i_mode))
2023 goto full_reval;
2024
2025 /* We cannot do exclusive creation on a positive dentry */
2026 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2027 goto reval_dentry;
2028
2029 /* Check if the directory changed */
2030 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2031 goto reval_dentry;
2032
2033 /* Let f_op->open() actually open (and revalidate) the file */
2034 return 1;
2035reval_dentry:
2036 if (flags & LOOKUP_RCU)
2037 return -ECHILD;
2038 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
2039
2040full_reval:
2041 return nfs_do_lookup_revalidate(dir, dentry, flags);
2042}
2043
2044static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2045{
2046 return __nfs_lookup_revalidate(dentry, flags,
2047 nfs4_do_lookup_revalidate);
2048}
2049
2050#endif /* CONFIG_NFSV4 */
2051
2052struct dentry *
2053nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2054 struct nfs_fattr *fattr,
2055 struct nfs4_label *label)
2056{
2057 struct dentry *parent = dget_parent(dentry);
2058 struct inode *dir = d_inode(parent);
2059 struct inode *inode;
2060 struct dentry *d;
2061 int error;
2062
2063 d_drop(dentry);
2064
2065 if (fhandle->size == 0) {
2066 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
2067 if (error)
2068 goto out_error;
2069 }
2070 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2071 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2072 struct nfs_server *server = NFS_SB(dentry->d_sb);
2073 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2074 fattr, NULL, NULL);
2075 if (error < 0)
2076 goto out_error;
2077 }
2078 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
2079 d = d_splice_alias(inode, dentry);
2080out:
2081 dput(parent);
2082 return d;
2083out_error:
2084 d = ERR_PTR(error);
2085 goto out;
2086}
2087EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2088
2089/*
2090 * Code common to create, mkdir, and mknod.
2091 */
2092int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2093 struct nfs_fattr *fattr,
2094 struct nfs4_label *label)
2095{
2096 struct dentry *d;
2097
2098 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
2099 if (IS_ERR(d))
2100 return PTR_ERR(d);
2101
2102 /* Callers don't care */
2103 dput(d);
2104 return 0;
2105}
2106EXPORT_SYMBOL_GPL(nfs_instantiate);
2107
2108/*
2109 * Following a failed create operation, we drop the dentry rather
2110 * than retain a negative dentry. This avoids a problem in the event
2111 * that the operation succeeded on the server, but an error in the
2112 * reply path made it appear to have failed.
2113 */
2114int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2115 struct dentry *dentry, umode_t mode, bool excl)
2116{
2117 struct iattr attr;
2118 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2119 int error;
2120
2121 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2122 dir->i_sb->s_id, dir->i_ino, dentry);
2123
2124 attr.ia_mode = mode;
2125 attr.ia_valid = ATTR_MODE;
2126
2127 trace_nfs_create_enter(dir, dentry, open_flags);
2128 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2129 trace_nfs_create_exit(dir, dentry, open_flags, error);
2130 if (error != 0)
2131 goto out_err;
2132 return 0;
2133out_err:
2134 d_drop(dentry);
2135 return error;
2136}
2137EXPORT_SYMBOL_GPL(nfs_create);
2138
2139/*
2140 * See comments for nfs_proc_create regarding failed operations.
2141 */
2142int
2143nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2144 struct dentry *dentry, umode_t mode, dev_t rdev)
2145{
2146 struct iattr attr;
2147 int status;
2148
2149 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2150 dir->i_sb->s_id, dir->i_ino, dentry);
2151
2152 attr.ia_mode = mode;
2153 attr.ia_valid = ATTR_MODE;
2154
2155 trace_nfs_mknod_enter(dir, dentry);
2156 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2157 trace_nfs_mknod_exit(dir, dentry, status);
2158 if (status != 0)
2159 goto out_err;
2160 return 0;
2161out_err:
2162 d_drop(dentry);
2163 return status;
2164}
2165EXPORT_SYMBOL_GPL(nfs_mknod);
2166
2167/*
2168 * See comments for nfs_proc_create regarding failed operations.
2169 */
2170int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2171 struct dentry *dentry, umode_t mode)
2172{
2173 struct iattr attr;
2174 int error;
2175
2176 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2177 dir->i_sb->s_id, dir->i_ino, dentry);
2178
2179 attr.ia_valid = ATTR_MODE;
2180 attr.ia_mode = mode | S_IFDIR;
2181
2182 trace_nfs_mkdir_enter(dir, dentry);
2183 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2184 trace_nfs_mkdir_exit(dir, dentry, error);
2185 if (error != 0)
2186 goto out_err;
2187 return 0;
2188out_err:
2189 d_drop(dentry);
2190 return error;
2191}
2192EXPORT_SYMBOL_GPL(nfs_mkdir);
2193
2194static void nfs_dentry_handle_enoent(struct dentry *dentry)
2195{
2196 if (simple_positive(dentry))
2197 d_delete(dentry);
2198}
2199
2200int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2201{
2202 int error;
2203
2204 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2205 dir->i_sb->s_id, dir->i_ino, dentry);
2206
2207 trace_nfs_rmdir_enter(dir, dentry);
2208 if (d_really_is_positive(dentry)) {
2209 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2210 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2211 /* Ensure the VFS deletes this inode */
2212 switch (error) {
2213 case 0:
2214 clear_nlink(d_inode(dentry));
2215 break;
2216 case -ENOENT:
2217 nfs_dentry_handle_enoent(dentry);
2218 }
2219 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2220 } else
2221 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2222 trace_nfs_rmdir_exit(dir, dentry, error);
2223
2224 return error;
2225}
2226EXPORT_SYMBOL_GPL(nfs_rmdir);
2227
2228/*
2229 * Remove a file after making sure there are no pending writes,
2230 * and after checking that the file has only one user.
2231 *
2232 * We invalidate the attribute cache and free the inode prior to the operation
2233 * to avoid possible races if the server reuses the inode.
2234 */
2235static int nfs_safe_remove(struct dentry *dentry)
2236{
2237 struct inode *dir = d_inode(dentry->d_parent);
2238 struct inode *inode = d_inode(dentry);
2239 int error = -EBUSY;
2240
2241 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2242
2243 /* If the dentry was sillyrenamed, we simply call d_delete() */
2244 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2245 error = 0;
2246 goto out;
2247 }
2248
2249 trace_nfs_remove_enter(dir, dentry);
2250 if (inode != NULL) {
2251 error = NFS_PROTO(dir)->remove(dir, dentry);
2252 if (error == 0)
2253 nfs_drop_nlink(inode);
2254 } else
2255 error = NFS_PROTO(dir)->remove(dir, dentry);
2256 if (error == -ENOENT)
2257 nfs_dentry_handle_enoent(dentry);
2258 trace_nfs_remove_exit(dir, dentry, error);
2259out:
2260 return error;
2261}
2262
2263/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2264 * belongs to an active ".nfs..." file and we return -EBUSY.
2265 *
2266 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2267 */
2268int nfs_unlink(struct inode *dir, struct dentry *dentry)
2269{
2270 int error;
2271 int need_rehash = 0;
2272
2273 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2274 dir->i_ino, dentry);
2275
2276 trace_nfs_unlink_enter(dir, dentry);
2277 spin_lock(&dentry->d_lock);
2278 if (d_count(dentry) > 1) {
2279 spin_unlock(&dentry->d_lock);
2280 /* Start asynchronous writeout of the inode */
2281 write_inode_now(d_inode(dentry), 0);
2282 error = nfs_sillyrename(dir, dentry);
2283 goto out;
2284 }
2285 if (!d_unhashed(dentry)) {
2286 __d_drop(dentry);
2287 need_rehash = 1;
2288 }
2289 spin_unlock(&dentry->d_lock);
2290 error = nfs_safe_remove(dentry);
2291 if (!error || error == -ENOENT) {
2292 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2293 } else if (need_rehash)
2294 d_rehash(dentry);
2295out:
2296 trace_nfs_unlink_exit(dir, dentry, error);
2297 return error;
2298}
2299EXPORT_SYMBOL_GPL(nfs_unlink);
2300
2301/*
2302 * To create a symbolic link, most file systems instantiate a new inode,
2303 * add a page to it containing the path, then write it out to the disk
2304 * using prepare_write/commit_write.
2305 *
2306 * Unfortunately the NFS client can't create the in-core inode first
2307 * because it needs a file handle to create an in-core inode (see
2308 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2309 * symlink request has completed on the server.
2310 *
2311 * So instead we allocate a raw page, copy the symname into it, then do
2312 * the SYMLINK request with the page as the buffer. If it succeeds, we
2313 * now have a new file handle and can instantiate an in-core NFS inode
2314 * and move the raw page into its mapping.
2315 */
2316int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2317 struct dentry *dentry, const char *symname)
2318{
2319 struct page *page;
2320 char *kaddr;
2321 struct iattr attr;
2322 unsigned int pathlen = strlen(symname);
2323 int error;
2324
2325 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2326 dir->i_ino, dentry, symname);
2327
2328 if (pathlen > PAGE_SIZE)
2329 return -ENAMETOOLONG;
2330
2331 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2332 attr.ia_valid = ATTR_MODE;
2333
2334 page = alloc_page(GFP_USER);
2335 if (!page)
2336 return -ENOMEM;
2337
2338 kaddr = page_address(page);
2339 memcpy(kaddr, symname, pathlen);
2340 if (pathlen < PAGE_SIZE)
2341 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2342
2343 trace_nfs_symlink_enter(dir, dentry);
2344 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2345 trace_nfs_symlink_exit(dir, dentry, error);
2346 if (error != 0) {
2347 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2348 dir->i_sb->s_id, dir->i_ino,
2349 dentry, symname, error);
2350 d_drop(dentry);
2351 __free_page(page);
2352 return error;
2353 }
2354
2355 /*
2356 * No big deal if we can't add this page to the page cache here.
2357 * READLINK will get the missing page from the server if needed.
2358 */
2359 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2360 GFP_KERNEL)) {
2361 SetPageUptodate(page);
2362 unlock_page(page);
2363 /*
2364 * add_to_page_cache_lru() grabs an extra page refcount.
2365 * Drop it here to avoid leaking this page later.
2366 */
2367 put_page(page);
2368 } else
2369 __free_page(page);
2370
2371 return 0;
2372}
2373EXPORT_SYMBOL_GPL(nfs_symlink);
2374
2375int
2376nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2377{
2378 struct inode *inode = d_inode(old_dentry);
2379 int error;
2380
2381 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2382 old_dentry, dentry);
2383
2384 trace_nfs_link_enter(inode, dir, dentry);
2385 d_drop(dentry);
2386 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2387 if (error == 0) {
2388 ihold(inode);
2389 d_add(dentry, inode);
2390 }
2391 trace_nfs_link_exit(inode, dir, dentry, error);
2392 return error;
2393}
2394EXPORT_SYMBOL_GPL(nfs_link);
2395
2396/*
2397 * RENAME
2398 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2399 * different file handle for the same inode after a rename (e.g. when
2400 * moving to a different directory). A fail-safe method to do so would
2401 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2402 * rename the old file using the sillyrename stuff. This way, the original
2403 * file in old_dir will go away when the last process iput()s the inode.
2404 *
2405 * FIXED.
2406 *
2407 * It actually works quite well. One needs to have the possibility for
2408 * at least one ".nfs..." file in each directory the file ever gets
2409 * moved or linked to which happens automagically with the new
2410 * implementation that only depends on the dcache stuff instead of
2411 * using the inode layer
2412 *
2413 * Unfortunately, things are a little more complicated than indicated
2414 * above. For a cross-directory move, we want to make sure we can get
2415 * rid of the old inode after the operation. This means there must be
2416 * no pending writes (if it's a file), and the use count must be 1.
2417 * If these conditions are met, we can drop the dentries before doing
2418 * the rename.
2419 */
2420int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2421 struct dentry *old_dentry, struct inode *new_dir,
2422 struct dentry *new_dentry, unsigned int flags)
2423{
2424 struct inode *old_inode = d_inode(old_dentry);
2425 struct inode *new_inode = d_inode(new_dentry);
2426 struct dentry *dentry = NULL, *rehash = NULL;
2427 struct rpc_task *task;
2428 int error = -EBUSY;
2429
2430 if (flags)
2431 return -EINVAL;
2432
2433 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2434 old_dentry, new_dentry,
2435 d_count(new_dentry));
2436
2437 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2438 /*
2439 * For non-directories, check whether the target is busy and if so,
2440 * make a copy of the dentry and then do a silly-rename. If the
2441 * silly-rename succeeds, the copied dentry is hashed and becomes
2442 * the new target.
2443 */
2444 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2445 /*
2446 * To prevent any new references to the target during the
2447 * rename, we unhash the dentry in advance.
2448 */
2449 if (!d_unhashed(new_dentry)) {
2450 d_drop(new_dentry);
2451 rehash = new_dentry;
2452 }
2453
2454 if (d_count(new_dentry) > 2) {
2455 int err;
2456
2457 /* copy the target dentry's name */
2458 dentry = d_alloc(new_dentry->d_parent,
2459 &new_dentry->d_name);
2460 if (!dentry)
2461 goto out;
2462
2463 /* silly-rename the existing target ... */
2464 err = nfs_sillyrename(new_dir, new_dentry);
2465 if (err)
2466 goto out;
2467
2468 new_dentry = dentry;
2469 rehash = NULL;
2470 new_inode = NULL;
2471 }
2472 }
2473
2474 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2475 if (IS_ERR(task)) {
2476 error = PTR_ERR(task);
2477 goto out;
2478 }
2479
2480 error = rpc_wait_for_completion_task(task);
2481 if (error != 0) {
2482 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2483 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2484 smp_wmb();
2485 } else
2486 error = task->tk_status;
2487 rpc_put_task(task);
2488 /* Ensure the inode attributes are revalidated */
2489 if (error == 0) {
2490 spin_lock(&old_inode->i_lock);
2491 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2492 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2493 NFS_INO_INVALID_CTIME |
2494 NFS_INO_REVAL_FORCED);
2495 spin_unlock(&old_inode->i_lock);
2496 }
2497out:
2498 if (rehash)
2499 d_rehash(rehash);
2500 trace_nfs_rename_exit(old_dir, old_dentry,
2501 new_dir, new_dentry, error);
2502 if (!error) {
2503 if (new_inode != NULL)
2504 nfs_drop_nlink(new_inode);
2505 /*
2506 * The d_move() should be here instead of in an async RPC completion
2507 * handler because we need the proper locks to move the dentry. If
2508 * we're interrupted by a signal, the async RPC completion handler
2509 * should mark the directories for revalidation.
2510 */
2511 d_move(old_dentry, new_dentry);
2512 nfs_set_verifier(old_dentry,
2513 nfs_save_change_attribute(new_dir));
2514 } else if (error == -ENOENT)
2515 nfs_dentry_handle_enoent(old_dentry);
2516
2517 /* new dentry created? */
2518 if (dentry)
2519 dput(dentry);
2520 return error;
2521}
2522EXPORT_SYMBOL_GPL(nfs_rename);
2523
2524static DEFINE_SPINLOCK(nfs_access_lru_lock);
2525static LIST_HEAD(nfs_access_lru_list);
2526static atomic_long_t nfs_access_nr_entries;
2527
2528static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2529module_param(nfs_access_max_cachesize, ulong, 0644);
2530MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2531
2532static void nfs_access_free_entry(struct nfs_access_entry *entry)
2533{
2534 put_cred(entry->cred);
2535 kfree_rcu(entry, rcu_head);
2536 smp_mb__before_atomic();
2537 atomic_long_dec(&nfs_access_nr_entries);
2538 smp_mb__after_atomic();
2539}
2540
2541static void nfs_access_free_list(struct list_head *head)
2542{
2543 struct nfs_access_entry *cache;
2544
2545 while (!list_empty(head)) {
2546 cache = list_entry(head->next, struct nfs_access_entry, lru);
2547 list_del(&cache->lru);
2548 nfs_access_free_entry(cache);
2549 }
2550}
2551
2552static unsigned long
2553nfs_do_access_cache_scan(unsigned int nr_to_scan)
2554{
2555 LIST_HEAD(head);
2556 struct nfs_inode *nfsi, *next;
2557 struct nfs_access_entry *cache;
2558 long freed = 0;
2559
2560 spin_lock(&nfs_access_lru_lock);
2561 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2562 struct inode *inode;
2563
2564 if (nr_to_scan-- == 0)
2565 break;
2566 inode = &nfsi->vfs_inode;
2567 spin_lock(&inode->i_lock);
2568 if (list_empty(&nfsi->access_cache_entry_lru))
2569 goto remove_lru_entry;
2570 cache = list_entry(nfsi->access_cache_entry_lru.next,
2571 struct nfs_access_entry, lru);
2572 list_move(&cache->lru, &head);
2573 rb_erase(&cache->rb_node, &nfsi->access_cache);
2574 freed++;
2575 if (!list_empty(&nfsi->access_cache_entry_lru))
2576 list_move_tail(&nfsi->access_cache_inode_lru,
2577 &nfs_access_lru_list);
2578 else {
2579remove_lru_entry:
2580 list_del_init(&nfsi->access_cache_inode_lru);
2581 smp_mb__before_atomic();
2582 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2583 smp_mb__after_atomic();
2584 }
2585 spin_unlock(&inode->i_lock);
2586 }
2587 spin_unlock(&nfs_access_lru_lock);
2588 nfs_access_free_list(&head);
2589 return freed;
2590}
2591
2592unsigned long
2593nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2594{
2595 int nr_to_scan = sc->nr_to_scan;
2596 gfp_t gfp_mask = sc->gfp_mask;
2597
2598 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2599 return SHRINK_STOP;
2600 return nfs_do_access_cache_scan(nr_to_scan);
2601}
2602
2603
2604unsigned long
2605nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2606{
2607 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2608}
2609
2610static void
2611nfs_access_cache_enforce_limit(void)
2612{
2613 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2614 unsigned long diff;
2615 unsigned int nr_to_scan;
2616
2617 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2618 return;
2619 nr_to_scan = 100;
2620 diff = nr_entries - nfs_access_max_cachesize;
2621 if (diff < nr_to_scan)
2622 nr_to_scan = diff;
2623 nfs_do_access_cache_scan(nr_to_scan);
2624}
2625
2626static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2627{
2628 struct rb_root *root_node = &nfsi->access_cache;
2629 struct rb_node *n;
2630 struct nfs_access_entry *entry;
2631
2632 /* Unhook entries from the cache */
2633 while ((n = rb_first(root_node)) != NULL) {
2634 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2635 rb_erase(n, root_node);
2636 list_move(&entry->lru, head);
2637 }
2638 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2639}
2640
2641void nfs_access_zap_cache(struct inode *inode)
2642{
2643 LIST_HEAD(head);
2644
2645 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2646 return;
2647 /* Remove from global LRU init */
2648 spin_lock(&nfs_access_lru_lock);
2649 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2650 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2651
2652 spin_lock(&inode->i_lock);
2653 __nfs_access_zap_cache(NFS_I(inode), &head);
2654 spin_unlock(&inode->i_lock);
2655 spin_unlock(&nfs_access_lru_lock);
2656 nfs_access_free_list(&head);
2657}
2658EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2659
2660static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2661{
2662 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2663
2664 while (n != NULL) {
2665 struct nfs_access_entry *entry =
2666 rb_entry(n, struct nfs_access_entry, rb_node);
2667 int cmp = cred_fscmp(cred, entry->cred);
2668
2669 if (cmp < 0)
2670 n = n->rb_left;
2671 else if (cmp > 0)
2672 n = n->rb_right;
2673 else
2674 return entry;
2675 }
2676 return NULL;
2677}
2678
2679static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2680{
2681 struct nfs_inode *nfsi = NFS_I(inode);
2682 struct nfs_access_entry *cache;
2683 bool retry = true;
2684 int err;
2685
2686 spin_lock(&inode->i_lock);
2687 for(;;) {
2688 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2689 goto out_zap;
2690 cache = nfs_access_search_rbtree(inode, cred);
2691 err = -ENOENT;
2692 if (cache == NULL)
2693 goto out;
2694 /* Found an entry, is our attribute cache valid? */
2695 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2696 break;
2697 if (!retry)
2698 break;
2699 err = -ECHILD;
2700 if (!may_block)
2701 goto out;
2702 spin_unlock(&inode->i_lock);
2703 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2704 if (err)
2705 return err;
2706 spin_lock(&inode->i_lock);
2707 retry = false;
2708 }
2709 res->cred = cache->cred;
2710 res->mask = cache->mask;
2711 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2712 err = 0;
2713out:
2714 spin_unlock(&inode->i_lock);
2715 return err;
2716out_zap:
2717 spin_unlock(&inode->i_lock);
2718 nfs_access_zap_cache(inode);
2719 return -ENOENT;
2720}
2721
2722static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2723{
2724 /* Only check the most recently returned cache entry,
2725 * but do it without locking.
2726 */
2727 struct nfs_inode *nfsi = NFS_I(inode);
2728 struct nfs_access_entry *cache;
2729 int err = -ECHILD;
2730 struct list_head *lh;
2731
2732 rcu_read_lock();
2733 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2734 goto out;
2735 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2736 cache = list_entry(lh, struct nfs_access_entry, lru);
2737 if (lh == &nfsi->access_cache_entry_lru ||
2738 cred_fscmp(cred, cache->cred) != 0)
2739 cache = NULL;
2740 if (cache == NULL)
2741 goto out;
2742 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2743 goto out;
2744 res->cred = cache->cred;
2745 res->mask = cache->mask;
2746 err = 0;
2747out:
2748 rcu_read_unlock();
2749 return err;
2750}
2751
2752int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct
2753nfs_access_entry *res, bool may_block)
2754{
2755 int status;
2756
2757 status = nfs_access_get_cached_rcu(inode, cred, res);
2758 if (status != 0)
2759 status = nfs_access_get_cached_locked(inode, cred, res,
2760 may_block);
2761
2762 return status;
2763}
2764EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2765
2766static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2767{
2768 struct nfs_inode *nfsi = NFS_I(inode);
2769 struct rb_root *root_node = &nfsi->access_cache;
2770 struct rb_node **p = &root_node->rb_node;
2771 struct rb_node *parent = NULL;
2772 struct nfs_access_entry *entry;
2773 int cmp;
2774
2775 spin_lock(&inode->i_lock);
2776 while (*p != NULL) {
2777 parent = *p;
2778 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2779 cmp = cred_fscmp(set->cred, entry->cred);
2780
2781 if (cmp < 0)
2782 p = &parent->rb_left;
2783 else if (cmp > 0)
2784 p = &parent->rb_right;
2785 else
2786 goto found;
2787 }
2788 rb_link_node(&set->rb_node, parent, p);
2789 rb_insert_color(&set->rb_node, root_node);
2790 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2791 spin_unlock(&inode->i_lock);
2792 return;
2793found:
2794 rb_replace_node(parent, &set->rb_node, root_node);
2795 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2796 list_del(&entry->lru);
2797 spin_unlock(&inode->i_lock);
2798 nfs_access_free_entry(entry);
2799}
2800
2801void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2802{
2803 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2804 if (cache == NULL)
2805 return;
2806 RB_CLEAR_NODE(&cache->rb_node);
2807 cache->cred = get_cred(set->cred);
2808 cache->mask = set->mask;
2809
2810 /* The above field assignments must be visible
2811 * before this item appears on the lru. We cannot easily
2812 * use rcu_assign_pointer, so just force the memory barrier.
2813 */
2814 smp_wmb();
2815 nfs_access_add_rbtree(inode, cache);
2816
2817 /* Update accounting */
2818 smp_mb__before_atomic();
2819 atomic_long_inc(&nfs_access_nr_entries);
2820 smp_mb__after_atomic();
2821
2822 /* Add inode to global LRU list */
2823 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2824 spin_lock(&nfs_access_lru_lock);
2825 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2826 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2827 &nfs_access_lru_list);
2828 spin_unlock(&nfs_access_lru_lock);
2829 }
2830 nfs_access_cache_enforce_limit();
2831}
2832EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2833
2834#define NFS_MAY_READ (NFS_ACCESS_READ)
2835#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2836 NFS_ACCESS_EXTEND | \
2837 NFS_ACCESS_DELETE)
2838#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2839 NFS_ACCESS_EXTEND)
2840#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2841#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2842#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2843static int
2844nfs_access_calc_mask(u32 access_result, umode_t umode)
2845{
2846 int mask = 0;
2847
2848 if (access_result & NFS_MAY_READ)
2849 mask |= MAY_READ;
2850 if (S_ISDIR(umode)) {
2851 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2852 mask |= MAY_WRITE;
2853 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2854 mask |= MAY_EXEC;
2855 } else if (S_ISREG(umode)) {
2856 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2857 mask |= MAY_WRITE;
2858 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2859 mask |= MAY_EXEC;
2860 } else if (access_result & NFS_MAY_WRITE)
2861 mask |= MAY_WRITE;
2862 return mask;
2863}
2864
2865void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2866{
2867 entry->mask = access_result;
2868}
2869EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2870
2871static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2872{
2873 struct nfs_access_entry cache;
2874 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2875 int cache_mask = -1;
2876 int status;
2877
2878 trace_nfs_access_enter(inode);
2879
2880 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2881 if (status == 0)
2882 goto out_cached;
2883
2884 status = -ECHILD;
2885 if (!may_block)
2886 goto out;
2887
2888 /*
2889 * Determine which access bits we want to ask for...
2890 */
2891 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2892 if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2893 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2894 NFS_ACCESS_XALIST;
2895 }
2896 if (S_ISDIR(inode->i_mode))
2897 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2898 else
2899 cache.mask |= NFS_ACCESS_EXECUTE;
2900 cache.cred = cred;
2901 status = NFS_PROTO(inode)->access(inode, &cache);
2902 if (status != 0) {
2903 if (status == -ESTALE) {
2904 if (!S_ISDIR(inode->i_mode))
2905 nfs_set_inode_stale(inode);
2906 else
2907 nfs_zap_caches(inode);
2908 }
2909 goto out;
2910 }
2911 nfs_access_add_cache(inode, &cache);
2912out_cached:
2913 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2914 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2915 status = -EACCES;
2916out:
2917 trace_nfs_access_exit(inode, mask, cache_mask, status);
2918 return status;
2919}
2920
2921static int nfs_open_permission_mask(int openflags)
2922{
2923 int mask = 0;
2924
2925 if (openflags & __FMODE_EXEC) {
2926 /* ONLY check exec rights */
2927 mask = MAY_EXEC;
2928 } else {
2929 if ((openflags & O_ACCMODE) != O_WRONLY)
2930 mask |= MAY_READ;
2931 if ((openflags & O_ACCMODE) != O_RDONLY)
2932 mask |= MAY_WRITE;
2933 }
2934
2935 return mask;
2936}
2937
2938int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2939{
2940 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2941}
2942EXPORT_SYMBOL_GPL(nfs_may_open);
2943
2944static int nfs_execute_ok(struct inode *inode, int mask)
2945{
2946 struct nfs_server *server = NFS_SERVER(inode);
2947 int ret = 0;
2948
2949 if (S_ISDIR(inode->i_mode))
2950 return 0;
2951 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
2952 if (mask & MAY_NOT_BLOCK)
2953 return -ECHILD;
2954 ret = __nfs_revalidate_inode(server, inode);
2955 }
2956 if (ret == 0 && !execute_ok(inode))
2957 ret = -EACCES;
2958 return ret;
2959}
2960
2961int nfs_permission(struct user_namespace *mnt_userns,
2962 struct inode *inode,
2963 int mask)
2964{
2965 const struct cred *cred = current_cred();
2966 int res = 0;
2967
2968 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2969
2970 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2971 goto out;
2972 /* Is this sys_access() ? */
2973 if (mask & (MAY_ACCESS | MAY_CHDIR))
2974 goto force_lookup;
2975
2976 switch (inode->i_mode & S_IFMT) {
2977 case S_IFLNK:
2978 goto out;
2979 case S_IFREG:
2980 if ((mask & MAY_OPEN) &&
2981 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2982 return 0;
2983 break;
2984 case S_IFDIR:
2985 /*
2986 * Optimize away all write operations, since the server
2987 * will check permissions when we perform the op.
2988 */
2989 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2990 goto out;
2991 }
2992
2993force_lookup:
2994 if (!NFS_PROTO(inode)->access)
2995 goto out_notsup;
2996
2997 res = nfs_do_access(inode, cred, mask);
2998out:
2999 if (!res && (mask & MAY_EXEC))
3000 res = nfs_execute_ok(inode, mask);
3001
3002 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3003 inode->i_sb->s_id, inode->i_ino, mask, res);
3004 return res;
3005out_notsup:
3006 if (mask & MAY_NOT_BLOCK)
3007 return -ECHILD;
3008
3009 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3010 NFS_INO_INVALID_OTHER);
3011 if (res == 0)
3012 res = generic_permission(&init_user_ns, inode, mask);
3013 goto out;
3014}
3015EXPORT_SYMBOL_GPL(nfs_permission);
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
53static int nfs_opendir(struct inode *, struct file *);
54static int nfs_closedir(struct inode *, struct file *);
55static int nfs_readdir(struct file *, struct dir_context *);
56static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
57static loff_t nfs_llseek_dir(struct file *, loff_t, int);
58static void nfs_readdir_clear_array(struct folio *);
59
60const 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
69const struct address_space_operations nfs_dir_aops = {
70 .free_folio = nfs_readdir_clear_array,
71};
72
73#define NFS_INIT_DTSIZE PAGE_SIZE
74
75static struct nfs_open_dir_context *
76alloc_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
99static 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 */
110static int
111nfs_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;
126out:
127 return res;
128}
129
130static int
131nfs_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
137struct 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
145struct nfs_cache_array {
146 u64 change_attr;
147 u64 last_cookie;
148 unsigned int size;
149 unsigned char folio_full : 1,
150 folio_is_eof : 1,
151 cookies_are_ordered : 1;
152 struct nfs_cache_array_entry array[];
153};
154
155struct nfs_readdir_descriptor {
156 struct file *file;
157 struct folio *folio;
158 struct dir_context *ctx;
159 pgoff_t folio_index;
160 pgoff_t folio_index_max;
161 u64 dir_cookie;
162 u64 last_cookie;
163 loff_t current_index;
164
165 __be32 verf[NFS_DIR_VERIFIER_SIZE];
166 unsigned long dir_verifier;
167 unsigned long timestamp;
168 unsigned long gencount;
169 unsigned long attr_gencount;
170 unsigned int cache_entry_index;
171 unsigned int buffer_fills;
172 unsigned int dtsize;
173 bool clear_cache;
174 bool plus;
175 bool eob;
176 bool eof;
177};
178
179static 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
191static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
192{
193 nfs_set_dtsize(desc, desc->dtsize >> 1);
194}
195
196static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
197{
198 nfs_set_dtsize(desc, desc->dtsize << 1);
199}
200
201static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie,
202 u64 change_attr)
203{
204 struct nfs_cache_array *array;
205
206 array = kmap_local_folio(folio, 0);
207 array->change_attr = change_attr;
208 array->last_cookie = last_cookie;
209 array->size = 0;
210 array->folio_full = 0;
211 array->folio_is_eof = 0;
212 array->cookies_are_ordered = 1;
213 kunmap_local(array);
214}
215
216/*
217 * we are freeing strings created by nfs_add_to_readdir_array()
218 */
219static void nfs_readdir_clear_array(struct folio *folio)
220{
221 struct nfs_cache_array *array;
222 unsigned int i;
223
224 array = kmap_local_folio(folio, 0);
225 for (i = 0; i < array->size; i++)
226 kfree(array->array[i].name);
227 array->size = 0;
228 kunmap_local(array);
229}
230
231static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
232 u64 change_attr)
233{
234 nfs_readdir_clear_array(folio);
235 nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
236}
237
238static struct folio *
239nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
240{
241 struct folio *folio = folio_alloc(gfp_flags, 0);
242 if (folio)
243 nfs_readdir_folio_init_array(folio, last_cookie, 0);
244 return folio;
245}
246
247static void nfs_readdir_folio_array_free(struct folio *folio)
248{
249 if (folio) {
250 nfs_readdir_clear_array(folio);
251 folio_put(folio);
252 }
253}
254
255static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
256{
257 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
258}
259
260static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
261{
262 array->folio_is_eof = 1;
263 array->folio_full = 1;
264}
265
266static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
267{
268 return array->folio_full;
269}
270
271/*
272 * the caller is responsible for freeing qstr.name
273 * when called by nfs_readdir_add_to_array, the strings will be freed in
274 * nfs_clear_readdir_array()
275 */
276static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
277{
278 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
279
280 /*
281 * Avoid a kmemleak false positive. The pointer to the name is stored
282 * in a page cache page which kmemleak does not scan.
283 */
284 if (ret != NULL)
285 kmemleak_not_leak(ret);
286 return ret;
287}
288
289static size_t nfs_readdir_array_maxentries(void)
290{
291 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
292 sizeof(struct nfs_cache_array_entry);
293}
294
295/*
296 * Check that the next array entry lies entirely within the page bounds
297 */
298static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
299{
300 if (array->folio_full)
301 return -ENOSPC;
302 if (array->size == nfs_readdir_array_maxentries()) {
303 array->folio_full = 1;
304 return -ENOSPC;
305 }
306 return 0;
307}
308
309static int nfs_readdir_folio_array_append(struct folio *folio,
310 const struct nfs_entry *entry,
311 u64 *cookie)
312{
313 struct nfs_cache_array *array;
314 struct nfs_cache_array_entry *cache_entry;
315 const char *name;
316 int ret = -ENOMEM;
317
318 name = nfs_readdir_copy_name(entry->name, entry->len);
319
320 array = kmap_local_folio(folio, 0);
321 if (!name)
322 goto out;
323 ret = nfs_readdir_array_can_expand(array);
324 if (ret) {
325 kfree(name);
326 goto out;
327 }
328
329 cache_entry = &array->array[array->size];
330 cache_entry->cookie = array->last_cookie;
331 cache_entry->ino = entry->ino;
332 cache_entry->d_type = entry->d_type;
333 cache_entry->name_len = entry->len;
334 cache_entry->name = name;
335 array->last_cookie = entry->cookie;
336 if (array->last_cookie <= cache_entry->cookie)
337 array->cookies_are_ordered = 0;
338 array->size++;
339 if (entry->eof != 0)
340 nfs_readdir_array_set_eof(array);
341out:
342 *cookie = array->last_cookie;
343 kunmap_local(array);
344 return ret;
345}
346
347#define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
348/*
349 * Hash algorithm allowing content addressible access to sequences
350 * of directory cookies. Content is addressed by the value of the
351 * cookie index of the first readdir entry in a page.
352 *
353 * We select only the first 18 bits to avoid issues with excessive
354 * memory use for the page cache XArray. 18 bits should allow the caching
355 * of 262144 pages of sequences of readdir entries. Since each page holds
356 * 127 readdir entries for a typical 64-bit system, that works out to a
357 * cache of ~ 33 million entries per directory.
358 */
359static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
360{
361 if (cookie == 0)
362 return 0;
363 return hash_64(cookie, 18);
364}
365
366static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
367 u64 change_attr)
368{
369 struct nfs_cache_array *array = kmap_local_folio(folio, 0);
370 int ret = true;
371
372 if (array->change_attr != change_attr)
373 ret = false;
374 if (nfs_readdir_array_index_cookie(array) != last_cookie)
375 ret = false;
376 kunmap_local(array);
377 return ret;
378}
379
380static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
381{
382 folio_unlock(folio);
383 folio_put(folio);
384}
385
386static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
387 u64 change_attr)
388{
389 if (folio_test_uptodate(folio)) {
390 if (nfs_readdir_folio_validate(folio, cookie, change_attr))
391 return;
392 nfs_readdir_clear_array(folio);
393 }
394 nfs_readdir_folio_init_array(folio, cookie, change_attr);
395 folio_mark_uptodate(folio);
396}
397
398static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
399 u64 cookie, u64 change_attr)
400{
401 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
402 struct folio *folio;
403
404 folio = filemap_grab_folio(mapping, index);
405 if (IS_ERR(folio))
406 return NULL;
407 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
408 return folio;
409}
410
411static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
412{
413 struct nfs_cache_array *array;
414 u64 ret;
415
416 array = kmap_local_folio(folio, 0);
417 ret = array->last_cookie;
418 kunmap_local(array);
419 return ret;
420}
421
422static bool nfs_readdir_folio_needs_filling(struct folio *folio)
423{
424 struct nfs_cache_array *array;
425 bool ret;
426
427 array = kmap_local_folio(folio, 0);
428 ret = !nfs_readdir_array_is_full(array);
429 kunmap_local(array);
430 return ret;
431}
432
433static void nfs_readdir_folio_set_eof(struct folio *folio)
434{
435 struct nfs_cache_array *array;
436
437 array = kmap_local_folio(folio, 0);
438 nfs_readdir_array_set_eof(array);
439 kunmap_local(array);
440}
441
442static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
443 u64 cookie, u64 change_attr)
444{
445 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
446 struct folio *folio;
447
448 folio = __filemap_get_folio(mapping, index,
449 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
450 mapping_gfp_mask(mapping));
451 if (IS_ERR(folio))
452 return NULL;
453 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
454 if (nfs_readdir_folio_last_cookie(folio) != cookie)
455 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
456 return folio;
457}
458
459static inline
460int is_32bit_api(void)
461{
462#ifdef CONFIG_COMPAT
463 return in_compat_syscall();
464#else
465 return (BITS_PER_LONG == 32);
466#endif
467}
468
469static
470bool nfs_readdir_use_cookie(const struct file *filp)
471{
472 if ((filp->f_mode & FMODE_32BITHASH) ||
473 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
474 return false;
475 return true;
476}
477
478static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
479 struct nfs_readdir_descriptor *desc)
480{
481 if (array->folio_full) {
482 desc->last_cookie = array->last_cookie;
483 desc->current_index += array->size;
484 desc->cache_entry_index = 0;
485 desc->folio_index++;
486 } else
487 desc->last_cookie = nfs_readdir_array_index_cookie(array);
488}
489
490static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
491{
492 desc->current_index = 0;
493 desc->last_cookie = 0;
494 desc->folio_index = 0;
495}
496
497static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
498 struct nfs_readdir_descriptor *desc)
499{
500 loff_t diff = desc->ctx->pos - desc->current_index;
501 unsigned int index;
502
503 if (diff < 0)
504 goto out_eof;
505 if (diff >= array->size) {
506 if (array->folio_is_eof)
507 goto out_eof;
508 nfs_readdir_seek_next_array(array, desc);
509 return -EAGAIN;
510 }
511
512 index = (unsigned int)diff;
513 desc->dir_cookie = array->array[index].cookie;
514 desc->cache_entry_index = index;
515 return 0;
516out_eof:
517 desc->eof = true;
518 return -EBADCOOKIE;
519}
520
521static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
522 u64 cookie)
523{
524 if (!array->cookies_are_ordered)
525 return true;
526 /* Optimisation for monotonically increasing cookies */
527 if (cookie >= array->last_cookie)
528 return false;
529 if (array->size && cookie < array->array[0].cookie)
530 return false;
531 return true;
532}
533
534static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
535 struct nfs_readdir_descriptor *desc)
536{
537 unsigned int i;
538 int status = -EAGAIN;
539
540 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
541 goto check_eof;
542
543 for (i = 0; i < array->size; i++) {
544 if (array->array[i].cookie == desc->dir_cookie) {
545 if (nfs_readdir_use_cookie(desc->file))
546 desc->ctx->pos = desc->dir_cookie;
547 else
548 desc->ctx->pos = desc->current_index + i;
549 desc->cache_entry_index = i;
550 return 0;
551 }
552 }
553check_eof:
554 if (array->folio_is_eof) {
555 status = -EBADCOOKIE;
556 if (desc->dir_cookie == array->last_cookie)
557 desc->eof = true;
558 } else
559 nfs_readdir_seek_next_array(array, desc);
560 return status;
561}
562
563static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
564{
565 struct nfs_cache_array *array;
566 int status;
567
568 array = kmap_local_folio(desc->folio, 0);
569
570 if (desc->dir_cookie == 0)
571 status = nfs_readdir_search_for_pos(array, desc);
572 else
573 status = nfs_readdir_search_for_cookie(array, desc);
574
575 kunmap_local(array);
576 return status;
577}
578
579/* Fill a page with xdr information before transferring to the cache page */
580static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
581 __be32 *verf, u64 cookie,
582 struct page **pages, size_t bufsize,
583 __be32 *verf_res)
584{
585 struct inode *inode = file_inode(desc->file);
586 struct nfs_readdir_arg arg = {
587 .dentry = file_dentry(desc->file),
588 .cred = desc->file->f_cred,
589 .verf = verf,
590 .cookie = cookie,
591 .pages = pages,
592 .page_len = bufsize,
593 .plus = desc->plus,
594 };
595 struct nfs_readdir_res res = {
596 .verf = verf_res,
597 };
598 unsigned long timestamp, gencount;
599 int error;
600
601 again:
602 timestamp = jiffies;
603 gencount = nfs_inc_attr_generation_counter();
604 desc->dir_verifier = nfs_save_change_attribute(inode);
605 error = NFS_PROTO(inode)->readdir(&arg, &res);
606 if (error < 0) {
607 /* We requested READDIRPLUS, but the server doesn't grok it */
608 if (error == -ENOTSUPP && desc->plus) {
609 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
610 desc->plus = arg.plus = false;
611 goto again;
612 }
613 goto error;
614 }
615 desc->timestamp = timestamp;
616 desc->gencount = gencount;
617error:
618 return error;
619}
620
621static int xdr_decode(struct nfs_readdir_descriptor *desc,
622 struct nfs_entry *entry, struct xdr_stream *xdr)
623{
624 struct inode *inode = file_inode(desc->file);
625 int error;
626
627 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
628 if (error)
629 return error;
630 entry->fattr->time_start = desc->timestamp;
631 entry->fattr->gencount = desc->gencount;
632 return 0;
633}
634
635/* Match file and dirent using either filehandle or fileid
636 * Note: caller is responsible for checking the fsid
637 */
638static
639int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
640{
641 struct inode *inode;
642 struct nfs_inode *nfsi;
643
644 if (d_really_is_negative(dentry))
645 return 0;
646
647 inode = d_inode(dentry);
648 if (is_bad_inode(inode) || NFS_STALE(inode))
649 return 0;
650
651 nfsi = NFS_I(inode);
652 if (entry->fattr->fileid != nfsi->fileid)
653 return 0;
654 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
655 return 0;
656 return 1;
657}
658
659#define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
660
661static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
662 unsigned int cache_hits,
663 unsigned int cache_misses)
664{
665 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
666 return false;
667 if (ctx->pos == 0 ||
668 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
669 return true;
670 return false;
671}
672
673/*
674 * This function is called by the getattr code to request the
675 * use of readdirplus to accelerate any future lookups in the same
676 * directory.
677 */
678void nfs_readdir_record_entry_cache_hit(struct inode *dir)
679{
680 struct nfs_inode *nfsi = NFS_I(dir);
681 struct nfs_open_dir_context *ctx;
682
683 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
684 S_ISDIR(dir->i_mode)) {
685 rcu_read_lock();
686 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
687 atomic_inc(&ctx->cache_hits);
688 rcu_read_unlock();
689 }
690}
691
692/*
693 * This function is mainly for use by nfs_getattr().
694 *
695 * If this is an 'ls -l', we want to force use of readdirplus.
696 */
697void nfs_readdir_record_entry_cache_miss(struct inode *dir)
698{
699 struct nfs_inode *nfsi = NFS_I(dir);
700 struct nfs_open_dir_context *ctx;
701
702 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
703 S_ISDIR(dir->i_mode)) {
704 rcu_read_lock();
705 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
706 atomic_inc(&ctx->cache_misses);
707 rcu_read_unlock();
708 }
709}
710
711static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
712 unsigned int flags)
713{
714 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
715 return;
716 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
717 return;
718 nfs_readdir_record_entry_cache_miss(dir);
719}
720
721static
722void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
723 unsigned long dir_verifier)
724{
725 struct qstr filename = QSTR_INIT(entry->name, entry->len);
726 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
727 struct dentry *dentry;
728 struct dentry *alias;
729 struct inode *inode;
730 int status;
731
732 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
733 return;
734 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
735 return;
736 if (filename.len == 0)
737 return;
738 /* Validate that the name doesn't contain any illegal '\0' */
739 if (strnlen(filename.name, filename.len) != filename.len)
740 return;
741 /* ...or '/' */
742 if (strnchr(filename.name, filename.len, '/'))
743 return;
744 if (filename.name[0] == '.') {
745 if (filename.len == 1)
746 return;
747 if (filename.len == 2 && filename.name[1] == '.')
748 return;
749 }
750 filename.hash = full_name_hash(parent, filename.name, filename.len);
751
752 dentry = d_lookup(parent, &filename);
753again:
754 if (!dentry) {
755 dentry = d_alloc_parallel(parent, &filename, &wq);
756 if (IS_ERR(dentry))
757 return;
758 }
759 if (!d_in_lookup(dentry)) {
760 /* Is there a mountpoint here? If so, just exit */
761 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
762 &entry->fattr->fsid))
763 goto out;
764 if (nfs_same_file(dentry, entry)) {
765 if (!entry->fh->size)
766 goto out;
767 nfs_set_verifier(dentry, dir_verifier);
768 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
769 if (!status)
770 nfs_setsecurity(d_inode(dentry), entry->fattr);
771 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
772 dentry, 0, status);
773 goto out;
774 } else {
775 trace_nfs_readdir_lookup_revalidate_failed(
776 d_inode(parent), dentry, 0);
777 d_invalidate(dentry);
778 dput(dentry);
779 dentry = NULL;
780 goto again;
781 }
782 }
783 if (!entry->fh->size) {
784 d_lookup_done(dentry);
785 goto out;
786 }
787
788 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
789 alias = d_splice_alias(inode, dentry);
790 d_lookup_done(dentry);
791 if (alias) {
792 if (IS_ERR(alias))
793 goto out;
794 dput(dentry);
795 dentry = alias;
796 }
797 nfs_set_verifier(dentry, dir_verifier);
798 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
799out:
800 dput(dentry);
801}
802
803static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
804 struct nfs_entry *entry,
805 struct xdr_stream *stream)
806{
807 int ret;
808
809 if (entry->fattr->label)
810 entry->fattr->label->len = NFS4_MAXLABELLEN;
811 ret = xdr_decode(desc, entry, stream);
812 if (ret || !desc->plus)
813 return ret;
814 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
815 return 0;
816}
817
818/* Perform conversion from xdr to cache array */
819static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
820 struct nfs_entry *entry,
821 struct page **xdr_pages, unsigned int buflen,
822 struct folio **arrays, size_t narrays,
823 u64 change_attr)
824{
825 struct address_space *mapping = desc->file->f_mapping;
826 struct folio *new, *folio = *arrays;
827 struct xdr_stream stream;
828 struct page *scratch;
829 struct xdr_buf buf;
830 u64 cookie;
831 int status;
832
833 scratch = alloc_page(GFP_KERNEL);
834 if (scratch == NULL)
835 return -ENOMEM;
836
837 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
838 xdr_set_scratch_page(&stream, scratch);
839
840 do {
841 status = nfs_readdir_entry_decode(desc, entry, &stream);
842 if (status != 0)
843 break;
844
845 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
846 if (status != -ENOSPC)
847 continue;
848
849 if (folio->mapping != mapping) {
850 if (!--narrays)
851 break;
852 new = nfs_readdir_folio_array_alloc(cookie, GFP_KERNEL);
853 if (!new)
854 break;
855 arrays++;
856 *arrays = folio = new;
857 } else {
858 new = nfs_readdir_folio_get_next(mapping, cookie,
859 change_attr);
860 if (!new)
861 break;
862 if (folio != *arrays)
863 nfs_readdir_folio_unlock_and_put(folio);
864 folio = new;
865 }
866 desc->folio_index_max++;
867 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
868 } while (!status && !entry->eof);
869
870 switch (status) {
871 case -EBADCOOKIE:
872 if (!entry->eof)
873 break;
874 nfs_readdir_folio_set_eof(folio);
875 fallthrough;
876 case -EAGAIN:
877 status = 0;
878 break;
879 case -ENOSPC:
880 status = 0;
881 if (!desc->plus)
882 break;
883 while (!nfs_readdir_entry_decode(desc, entry, &stream))
884 ;
885 }
886
887 if (folio != *arrays)
888 nfs_readdir_folio_unlock_and_put(folio);
889
890 put_page(scratch);
891 return status;
892}
893
894static void nfs_readdir_free_pages(struct page **pages, size_t npages)
895{
896 while (npages--)
897 put_page(pages[npages]);
898 kfree(pages);
899}
900
901/*
902 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
903 * to nfs_readdir_free_pages()
904 */
905static struct page **nfs_readdir_alloc_pages(size_t npages)
906{
907 struct page **pages;
908 size_t i;
909
910 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
911 if (!pages)
912 return NULL;
913 for (i = 0; i < npages; i++) {
914 struct page *page = alloc_page(GFP_KERNEL);
915 if (page == NULL)
916 goto out_freepages;
917 pages[i] = page;
918 }
919 return pages;
920
921out_freepages:
922 nfs_readdir_free_pages(pages, i);
923 return NULL;
924}
925
926static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
927 __be32 *verf_arg, __be32 *verf_res,
928 struct folio **arrays, size_t narrays)
929{
930 u64 change_attr;
931 struct page **pages;
932 struct folio *folio = *arrays;
933 struct nfs_entry *entry;
934 size_t array_size;
935 struct inode *inode = file_inode(desc->file);
936 unsigned int dtsize = desc->dtsize;
937 unsigned int pglen;
938 int status = -ENOMEM;
939
940 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
941 if (!entry)
942 return -ENOMEM;
943 entry->cookie = nfs_readdir_folio_last_cookie(folio);
944 entry->fh = nfs_alloc_fhandle();
945 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
946 entry->server = NFS_SERVER(inode);
947 if (entry->fh == NULL || entry->fattr == NULL)
948 goto out;
949
950 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 pages = nfs_readdir_alloc_pages(array_size);
952 if (!pages)
953 goto out;
954
955 change_attr = inode_peek_iversion_raw(inode);
956 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
957 dtsize, verf_res);
958 if (status < 0)
959 goto free_pages;
960
961 pglen = status;
962 if (pglen != 0)
963 status = nfs_readdir_folio_filler(desc, entry, pages, pglen,
964 arrays, narrays, change_attr);
965 else
966 nfs_readdir_folio_set_eof(folio);
967 desc->buffer_fills++;
968
969free_pages:
970 nfs_readdir_free_pages(pages, array_size);
971out:
972 nfs_free_fattr(entry->fattr);
973 nfs_free_fhandle(entry->fh);
974 kfree(entry);
975 return status;
976}
977
978static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
979{
980 folio_put(desc->folio);
981 desc->folio = NULL;
982}
983
984static void
985nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
986{
987 folio_unlock(desc->folio);
988 nfs_readdir_folio_put(desc);
989}
990
991static struct folio *
992nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
993{
994 struct address_space *mapping = desc->file->f_mapping;
995 u64 change_attr = inode_peek_iversion_raw(mapping->host);
996 u64 cookie = desc->last_cookie;
997 struct folio *folio;
998
999 folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1000 if (!folio)
1001 return NULL;
1002 if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1003 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
1004 return folio;
1005}
1006
1007/*
1008 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1009 * and locks the page to prevent removal from the page cache.
1010 */
1011static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1012{
1013 struct inode *inode = file_inode(desc->file);
1014 struct nfs_inode *nfsi = NFS_I(inode);
1015 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1016 int res;
1017
1018 desc->folio = nfs_readdir_folio_get_cached(desc);
1019 if (!desc->folio)
1020 return -ENOMEM;
1021 if (nfs_readdir_folio_needs_filling(desc->folio)) {
1022 /* Grow the dtsize if we had to go back for more pages */
1023 if (desc->folio_index == desc->folio_index_max)
1024 nfs_grow_dtsize(desc);
1025 desc->folio_index_max = desc->folio_index;
1026 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1027 desc->last_cookie,
1028 desc->folio->index, desc->dtsize);
1029 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1030 &desc->folio, 1);
1031 if (res < 0) {
1032 nfs_readdir_folio_unlock_and_put_cached(desc);
1033 trace_nfs_readdir_cache_fill_done(inode, res);
1034 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1035 invalidate_inode_pages2(desc->file->f_mapping);
1036 nfs_readdir_rewind_search(desc);
1037 trace_nfs_readdir_invalidate_cache_range(
1038 inode, 0, MAX_LFS_FILESIZE);
1039 return -EAGAIN;
1040 }
1041 return res;
1042 }
1043 /*
1044 * Set the cookie verifier if the page cache was empty
1045 */
1046 if (desc->last_cookie == 0 &&
1047 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1048 memcpy(nfsi->cookieverf, verf,
1049 sizeof(nfsi->cookieverf));
1050 invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1051 -1);
1052 trace_nfs_readdir_invalidate_cache_range(
1053 inode, 1, MAX_LFS_FILESIZE);
1054 }
1055 desc->clear_cache = false;
1056 }
1057 res = nfs_readdir_search_array(desc);
1058 if (res == 0)
1059 return 0;
1060 nfs_readdir_folio_unlock_and_put_cached(desc);
1061 return res;
1062}
1063
1064/* Search for desc->dir_cookie from the beginning of the page cache */
1065static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1066{
1067 int res;
1068
1069 do {
1070 res = find_and_lock_cache_page(desc);
1071 } while (res == -EAGAIN);
1072 return res;
1073}
1074
1075#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1076
1077/*
1078 * Once we've found the start of the dirent within a page: fill 'er up...
1079 */
1080static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1081 const __be32 *verf)
1082{
1083 struct file *file = desc->file;
1084 struct nfs_cache_array *array;
1085 unsigned int i;
1086 bool first_emit = !desc->dir_cookie;
1087
1088 array = kmap_local_folio(desc->folio, 0);
1089 for (i = desc->cache_entry_index; i < array->size; i++) {
1090 struct nfs_cache_array_entry *ent;
1091
1092 /*
1093 * nfs_readdir_handle_cache_misses return force clear at
1094 * (cache_misses > NFS_READDIR_CACHE_MISS_THRESHOLD) for
1095 * readdir heuristic, NFS_READDIR_CACHE_MISS_THRESHOLD + 1
1096 * entries need be emitted here.
1097 */
1098 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 2) {
1099 desc->eob = true;
1100 break;
1101 }
1102
1103 ent = &array->array[i];
1104 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1105 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1106 desc->eob = true;
1107 break;
1108 }
1109 memcpy(desc->verf, verf, sizeof(desc->verf));
1110 if (i == array->size - 1) {
1111 desc->dir_cookie = array->last_cookie;
1112 nfs_readdir_seek_next_array(array, desc);
1113 } else {
1114 desc->dir_cookie = array->array[i + 1].cookie;
1115 desc->last_cookie = array->array[0].cookie;
1116 }
1117 if (nfs_readdir_use_cookie(file))
1118 desc->ctx->pos = desc->dir_cookie;
1119 else
1120 desc->ctx->pos++;
1121 }
1122 if (array->folio_is_eof)
1123 desc->eof = !desc->eob;
1124
1125 kunmap_local(array);
1126 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1127 (unsigned long long)desc->dir_cookie);
1128}
1129
1130/*
1131 * If we cannot find a cookie in our cache, we suspect that this is
1132 * because it points to a deleted file, so we ask the server to return
1133 * whatever it thinks is the next entry. We then feed this to filldir.
1134 * If all goes well, we should then be able to find our way round the
1135 * cache on the next call to readdir_search_pagecache();
1136 *
1137 * NOTE: we cannot add the anonymous page to the pagecache because
1138 * the data it contains might not be page aligned. Besides,
1139 * we should already have a complete representation of the
1140 * directory in the page cache by the time we get here.
1141 */
1142static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1143{
1144 struct folio **arrays;
1145 size_t i, sz = 512;
1146 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1147 int status = -ENOMEM;
1148
1149 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1150 (unsigned long long)desc->dir_cookie);
1151
1152 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1153 if (!arrays)
1154 goto out;
1155 arrays[0] = nfs_readdir_folio_array_alloc(desc->dir_cookie, GFP_KERNEL);
1156 if (!arrays[0])
1157 goto out;
1158
1159 desc->folio_index = 0;
1160 desc->cache_entry_index = 0;
1161 desc->last_cookie = desc->dir_cookie;
1162 desc->folio_index_max = 0;
1163
1164 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1165 -1, desc->dtsize);
1166
1167 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1168 if (status < 0) {
1169 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1170 goto out_free;
1171 }
1172
1173 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1174 desc->folio = arrays[i];
1175 nfs_do_filldir(desc, verf);
1176 }
1177 desc->folio = NULL;
1178
1179 /*
1180 * Grow the dtsize if we have to go back for more pages,
1181 * or shrink it if we're reading too many.
1182 */
1183 if (!desc->eof) {
1184 if (!desc->eob)
1185 nfs_grow_dtsize(desc);
1186 else if (desc->buffer_fills == 1 &&
1187 i < (desc->folio_index_max >> 1))
1188 nfs_shrink_dtsize(desc);
1189 }
1190out_free:
1191 for (i = 0; i < sz && arrays[i]; i++)
1192 nfs_readdir_folio_array_free(arrays[i]);
1193out:
1194 if (!nfs_readdir_use_cookie(desc->file))
1195 nfs_readdir_rewind_search(desc);
1196 desc->folio_index_max = -1;
1197 kfree(arrays);
1198 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1199 return status;
1200}
1201
1202static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1203 struct nfs_readdir_descriptor *desc,
1204 unsigned int cache_misses,
1205 bool force_clear)
1206{
1207 if (desc->ctx->pos == 0 || !desc->plus)
1208 return false;
1209 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1210 return false;
1211 trace_nfs_readdir_force_readdirplus(inode);
1212 return true;
1213}
1214
1215/* The file offset position represents the dirent entry number. A
1216 last cookie cache takes care of the common case of reading the
1217 whole directory.
1218 */
1219static int nfs_readdir(struct file *file, struct dir_context *ctx)
1220{
1221 struct dentry *dentry = file_dentry(file);
1222 struct inode *inode = d_inode(dentry);
1223 struct nfs_inode *nfsi = NFS_I(inode);
1224 struct nfs_open_dir_context *dir_ctx = file->private_data;
1225 struct nfs_readdir_descriptor *desc;
1226 unsigned int cache_hits, cache_misses;
1227 bool force_clear;
1228 int res;
1229
1230 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1231 file, (long long)ctx->pos);
1232 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1233
1234 /*
1235 * ctx->pos points to the dirent entry number.
1236 * *desc->dir_cookie has the cookie for the next entry. We have
1237 * to either find the entry with the appropriate number or
1238 * revalidate the cookie.
1239 */
1240 nfs_revalidate_mapping(inode, file->f_mapping);
1241
1242 res = -ENOMEM;
1243 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1244 if (!desc)
1245 goto out;
1246 desc->file = file;
1247 desc->ctx = ctx;
1248 desc->folio_index_max = -1;
1249
1250 spin_lock(&file->f_lock);
1251 desc->dir_cookie = dir_ctx->dir_cookie;
1252 desc->folio_index = dir_ctx->page_index;
1253 desc->last_cookie = dir_ctx->last_cookie;
1254 desc->attr_gencount = dir_ctx->attr_gencount;
1255 desc->eof = dir_ctx->eof;
1256 nfs_set_dtsize(desc, dir_ctx->dtsize);
1257 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1258 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1259 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1260 force_clear = dir_ctx->force_clear;
1261 spin_unlock(&file->f_lock);
1262
1263 if (desc->eof) {
1264 res = 0;
1265 goto out_free;
1266 }
1267
1268 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1269 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1270 force_clear);
1271 desc->clear_cache = force_clear;
1272
1273 do {
1274 res = readdir_search_pagecache(desc);
1275
1276 if (res == -EBADCOOKIE) {
1277 res = 0;
1278 /* This means either end of directory */
1279 if (desc->dir_cookie && !desc->eof) {
1280 /* Or that the server has 'lost' a cookie */
1281 res = uncached_readdir(desc);
1282 if (res == 0)
1283 continue;
1284 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1285 res = 0;
1286 }
1287 break;
1288 }
1289 if (res == -ETOOSMALL && desc->plus) {
1290 nfs_zap_caches(inode);
1291 desc->plus = false;
1292 desc->eof = false;
1293 continue;
1294 }
1295 if (res < 0)
1296 break;
1297
1298 nfs_do_filldir(desc, nfsi->cookieverf);
1299 nfs_readdir_folio_unlock_and_put_cached(desc);
1300 if (desc->folio_index == desc->folio_index_max)
1301 desc->clear_cache = force_clear;
1302 } while (!desc->eob && !desc->eof);
1303
1304 spin_lock(&file->f_lock);
1305 dir_ctx->dir_cookie = desc->dir_cookie;
1306 dir_ctx->last_cookie = desc->last_cookie;
1307 dir_ctx->attr_gencount = desc->attr_gencount;
1308 dir_ctx->page_index = desc->folio_index;
1309 dir_ctx->force_clear = force_clear;
1310 dir_ctx->eof = desc->eof;
1311 dir_ctx->dtsize = desc->dtsize;
1312 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1313 spin_unlock(&file->f_lock);
1314out_free:
1315 kfree(desc);
1316
1317out:
1318 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1319 return res;
1320}
1321
1322static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1323{
1324 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1325
1326 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1327 filp, offset, whence);
1328
1329 switch (whence) {
1330 default:
1331 return -EINVAL;
1332 case SEEK_SET:
1333 if (offset < 0)
1334 return -EINVAL;
1335 spin_lock(&filp->f_lock);
1336 break;
1337 case SEEK_CUR:
1338 if (offset == 0)
1339 return filp->f_pos;
1340 spin_lock(&filp->f_lock);
1341 offset += filp->f_pos;
1342 if (offset < 0) {
1343 spin_unlock(&filp->f_lock);
1344 return -EINVAL;
1345 }
1346 }
1347 if (offset != filp->f_pos) {
1348 filp->f_pos = offset;
1349 dir_ctx->page_index = 0;
1350 if (!nfs_readdir_use_cookie(filp)) {
1351 dir_ctx->dir_cookie = 0;
1352 dir_ctx->last_cookie = 0;
1353 } else {
1354 dir_ctx->dir_cookie = offset;
1355 dir_ctx->last_cookie = offset;
1356 }
1357 dir_ctx->eof = false;
1358 }
1359 spin_unlock(&filp->f_lock);
1360 return offset;
1361}
1362
1363/*
1364 * All directory operations under NFS are synchronous, so fsync()
1365 * is a dummy operation.
1366 */
1367static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1368 int datasync)
1369{
1370 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1371
1372 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1373 return 0;
1374}
1375
1376/**
1377 * nfs_force_lookup_revalidate - Mark the directory as having changed
1378 * @dir: pointer to directory inode
1379 *
1380 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1381 * full lookup on all child dentries of 'dir' whenever a change occurs
1382 * on the server that might have invalidated our dcache.
1383 *
1384 * Note that we reserve bit '0' as a tag to let us know when a dentry
1385 * was revalidated while holding a delegation on its inode.
1386 *
1387 * The caller should be holding dir->i_lock
1388 */
1389void nfs_force_lookup_revalidate(struct inode *dir)
1390{
1391 NFS_I(dir)->cache_change_attribute += 2;
1392}
1393EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1394
1395/**
1396 * nfs_verify_change_attribute - Detects NFS remote directory changes
1397 * @dir: pointer to parent directory inode
1398 * @verf: previously saved change attribute
1399 *
1400 * Return "false" if the verifiers doesn't match the change attribute.
1401 * This would usually indicate that the directory contents have changed on
1402 * the server, and that any dentries need revalidating.
1403 */
1404static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1405{
1406 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1407}
1408
1409static void nfs_set_verifier_delegated(unsigned long *verf)
1410{
1411 *verf |= 1UL;
1412}
1413
1414#if IS_ENABLED(CONFIG_NFS_V4)
1415static void nfs_unset_verifier_delegated(unsigned long *verf)
1416{
1417 *verf &= ~1UL;
1418}
1419#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1420
1421static bool nfs_test_verifier_delegated(unsigned long verf)
1422{
1423 return verf & 1;
1424}
1425
1426static bool nfs_verifier_is_delegated(struct dentry *dentry)
1427{
1428 return nfs_test_verifier_delegated(dentry->d_time);
1429}
1430
1431static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1432{
1433 struct inode *inode = d_inode(dentry);
1434 struct inode *dir = d_inode_rcu(dentry->d_parent);
1435
1436 if (!dir || !nfs_verify_change_attribute(dir, verf))
1437 return;
1438 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1439 nfs_set_verifier_delegated(&verf);
1440 dentry->d_time = verf;
1441}
1442
1443/**
1444 * nfs_set_verifier - save a parent directory verifier in the dentry
1445 * @dentry: pointer to dentry
1446 * @verf: verifier to save
1447 *
1448 * Saves the parent directory verifier in @dentry. If the inode has
1449 * a delegation, we also tag the dentry as having been revalidated
1450 * while holding a delegation so that we know we don't have to
1451 * look it up again after a directory change.
1452 */
1453void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1454{
1455
1456 spin_lock(&dentry->d_lock);
1457 nfs_set_verifier_locked(dentry, verf);
1458 spin_unlock(&dentry->d_lock);
1459}
1460EXPORT_SYMBOL_GPL(nfs_set_verifier);
1461
1462#if IS_ENABLED(CONFIG_NFS_V4)
1463/**
1464 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1465 * @inode: pointer to inode
1466 *
1467 * Iterates through the dentries in the inode alias list and clears
1468 * the tag used to indicate that the dentry has been revalidated
1469 * while holding a delegation.
1470 * This function is intended for use when the delegation is being
1471 * returned or revoked.
1472 */
1473void nfs_clear_verifier_delegated(struct inode *inode)
1474{
1475 struct dentry *alias;
1476
1477 if (!inode)
1478 return;
1479 spin_lock(&inode->i_lock);
1480 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1481 spin_lock(&alias->d_lock);
1482 nfs_unset_verifier_delegated(&alias->d_time);
1483 spin_unlock(&alias->d_lock);
1484 }
1485 spin_unlock(&inode->i_lock);
1486}
1487EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1488#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1489
1490static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1491{
1492 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1493 d_really_is_negative(dentry))
1494 return dentry->d_time == inode_peek_iversion_raw(dir);
1495 return nfs_verify_change_attribute(dir, dentry->d_time);
1496}
1497
1498/*
1499 * A check for whether or not the parent directory has changed.
1500 * In the case it has, we assume that the dentries are untrustworthy
1501 * and may need to be looked up again.
1502 * If rcu_walk prevents us from performing a full check, return 0.
1503 */
1504static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1505 int rcu_walk)
1506{
1507 if (IS_ROOT(dentry))
1508 return 1;
1509 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1510 return 0;
1511 if (!nfs_dentry_verify_change(dir, dentry))
1512 return 0;
1513 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1514 if (nfs_mapping_need_revalidate_inode(dir)) {
1515 if (rcu_walk)
1516 return 0;
1517 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1518 return 0;
1519 }
1520 if (!nfs_dentry_verify_change(dir, dentry))
1521 return 0;
1522 return 1;
1523}
1524
1525/*
1526 * Use intent information to check whether or not we're going to do
1527 * an O_EXCL create using this path component.
1528 */
1529static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1530{
1531 if (NFS_PROTO(dir)->version == 2)
1532 return 0;
1533 return flags & LOOKUP_EXCL;
1534}
1535
1536/*
1537 * Inode and filehandle revalidation for lookups.
1538 *
1539 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1540 * or if the intent information indicates that we're about to open this
1541 * particular file and the "nocto" mount flag is not set.
1542 *
1543 */
1544static
1545int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1546{
1547 struct nfs_server *server = NFS_SERVER(inode);
1548 int ret;
1549
1550 if (IS_AUTOMOUNT(inode))
1551 return 0;
1552
1553 if (flags & LOOKUP_OPEN) {
1554 switch (inode->i_mode & S_IFMT) {
1555 case S_IFREG:
1556 /* A NFSv4 OPEN will revalidate later */
1557 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1558 goto out;
1559 fallthrough;
1560 case S_IFDIR:
1561 if (server->flags & NFS_MOUNT_NOCTO)
1562 break;
1563 /* NFS close-to-open cache consistency validation */
1564 goto out_force;
1565 }
1566 }
1567
1568 /* VFS wants an on-the-wire revalidation */
1569 if (flags & LOOKUP_REVAL)
1570 goto out_force;
1571out:
1572 if (inode->i_nlink > 0 ||
1573 (inode->i_nlink == 0 &&
1574 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1575 return 0;
1576 else
1577 return -ESTALE;
1578out_force:
1579 if (flags & LOOKUP_RCU)
1580 return -ECHILD;
1581 ret = __nfs_revalidate_inode(server, inode);
1582 if (ret != 0)
1583 return ret;
1584 goto out;
1585}
1586
1587static void nfs_mark_dir_for_revalidate(struct inode *inode)
1588{
1589 spin_lock(&inode->i_lock);
1590 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1591 spin_unlock(&inode->i_lock);
1592}
1593
1594/*
1595 * We judge how long we want to trust negative
1596 * dentries by looking at the parent inode mtime.
1597 *
1598 * If parent mtime has changed, we revalidate, else we wait for a
1599 * period corresponding to the parent's attribute cache timeout value.
1600 *
1601 * If LOOKUP_RCU prevents us from performing a full check, return 1
1602 * suggesting a reval is needed.
1603 *
1604 * Note that when creating a new file, or looking up a rename target,
1605 * then it shouldn't be necessary to revalidate a negative dentry.
1606 */
1607static inline
1608int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1609 unsigned int flags)
1610{
1611 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1612 return 0;
1613 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1614 return 1;
1615 /* Case insensitive server? Revalidate negative dentries */
1616 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1617 return 1;
1618 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1619}
1620
1621static int
1622nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1623 struct inode *inode, int error)
1624{
1625 switch (error) {
1626 case 1:
1627 break;
1628 case 0:
1629 /*
1630 * We can't d_drop the root of a disconnected tree:
1631 * its d_hash is on the s_anon list and d_drop() would hide
1632 * it from shrink_dcache_for_unmount(), leading to busy
1633 * inodes on unmount and further oopses.
1634 */
1635 if (inode && IS_ROOT(dentry))
1636 error = 1;
1637 break;
1638 }
1639 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1640 return error;
1641}
1642
1643static int
1644nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1645 unsigned int flags)
1646{
1647 int ret = 1;
1648 if (nfs_neg_need_reval(dir, dentry, flags)) {
1649 if (flags & LOOKUP_RCU)
1650 return -ECHILD;
1651 ret = 0;
1652 }
1653 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1654}
1655
1656static int
1657nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1658 struct inode *inode)
1659{
1660 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1661 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1662}
1663
1664static int nfs_lookup_revalidate_dentry(struct inode *dir,
1665 struct dentry *dentry,
1666 struct inode *inode, unsigned int flags)
1667{
1668 struct nfs_fh *fhandle;
1669 struct nfs_fattr *fattr;
1670 unsigned long dir_verifier;
1671 int ret;
1672
1673 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1674
1675 ret = -ENOMEM;
1676 fhandle = nfs_alloc_fhandle();
1677 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1678 if (fhandle == NULL || fattr == NULL)
1679 goto out;
1680
1681 dir_verifier = nfs_save_change_attribute(dir);
1682 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1683 if (ret < 0) {
1684 switch (ret) {
1685 case -ESTALE:
1686 case -ENOENT:
1687 ret = 0;
1688 break;
1689 case -ETIMEDOUT:
1690 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1691 ret = 1;
1692 }
1693 goto out;
1694 }
1695
1696 /* Request help from readdirplus */
1697 nfs_lookup_advise_force_readdirplus(dir, flags);
1698
1699 ret = 0;
1700 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1701 goto out;
1702 if (nfs_refresh_inode(inode, fattr) < 0)
1703 goto out;
1704
1705 nfs_setsecurity(inode, fattr);
1706 nfs_set_verifier(dentry, dir_verifier);
1707
1708 ret = 1;
1709out:
1710 nfs_free_fattr(fattr);
1711 nfs_free_fhandle(fhandle);
1712
1713 /*
1714 * If the lookup failed despite the dentry change attribute being
1715 * a match, then we should revalidate the directory cache.
1716 */
1717 if (!ret && nfs_dentry_verify_change(dir, dentry))
1718 nfs_mark_dir_for_revalidate(dir);
1719 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1720}
1721
1722/*
1723 * This is called every time the dcache has a lookup hit,
1724 * and we should check whether we can really trust that
1725 * lookup.
1726 *
1727 * NOTE! The hit can be a negative hit too, don't assume
1728 * we have an inode!
1729 *
1730 * If the parent directory is seen to have changed, we throw out the
1731 * cached dentry and do a new lookup.
1732 */
1733static int
1734nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1735 unsigned int flags)
1736{
1737 struct inode *inode;
1738 int error;
1739
1740 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1741 inode = d_inode(dentry);
1742
1743 if (!inode)
1744 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1745
1746 if (is_bad_inode(inode)) {
1747 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1748 __func__, dentry);
1749 goto out_bad;
1750 }
1751
1752 if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1753 nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1754 goto out_bad;
1755
1756 if (nfs_verifier_is_delegated(dentry))
1757 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1758
1759 /* Force a full look up iff the parent directory has changed */
1760 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1761 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1762 error = nfs_lookup_verify_inode(inode, flags);
1763 if (error) {
1764 if (error == -ESTALE)
1765 nfs_mark_dir_for_revalidate(dir);
1766 goto out_bad;
1767 }
1768 goto out_valid;
1769 }
1770
1771 if (flags & LOOKUP_RCU)
1772 return -ECHILD;
1773
1774 if (NFS_STALE(inode))
1775 goto out_bad;
1776
1777 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1778out_valid:
1779 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1780out_bad:
1781 if (flags & LOOKUP_RCU)
1782 return -ECHILD;
1783 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1784}
1785
1786static int
1787__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1788 int (*reval)(struct inode *, struct dentry *, unsigned int))
1789{
1790 struct dentry *parent;
1791 struct inode *dir;
1792 int ret;
1793
1794 if (flags & LOOKUP_RCU) {
1795 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1796 return -ECHILD;
1797 parent = READ_ONCE(dentry->d_parent);
1798 dir = d_inode_rcu(parent);
1799 if (!dir)
1800 return -ECHILD;
1801 ret = reval(dir, dentry, flags);
1802 if (parent != READ_ONCE(dentry->d_parent))
1803 return -ECHILD;
1804 } else {
1805 /* Wait for unlink to complete */
1806 wait_var_event(&dentry->d_fsdata,
1807 dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1808 parent = dget_parent(dentry);
1809 ret = reval(d_inode(parent), dentry, flags);
1810 dput(parent);
1811 }
1812 return ret;
1813}
1814
1815static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1816{
1817 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1818}
1819
1820/*
1821 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1822 * when we don't really care about the dentry name. This is called when a
1823 * pathwalk ends on a dentry that was not found via a normal lookup in the
1824 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1825 *
1826 * In this situation, we just want to verify that the inode itself is OK
1827 * since the dentry might have changed on the server.
1828 */
1829static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1830{
1831 struct inode *inode = d_inode(dentry);
1832 int error = 0;
1833
1834 /*
1835 * I believe we can only get a negative dentry here in the case of a
1836 * procfs-style symlink. Just assume it's correct for now, but we may
1837 * eventually need to do something more here.
1838 */
1839 if (!inode) {
1840 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1841 __func__, dentry);
1842 return 1;
1843 }
1844
1845 if (is_bad_inode(inode)) {
1846 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1847 __func__, dentry);
1848 return 0;
1849 }
1850
1851 error = nfs_lookup_verify_inode(inode, flags);
1852 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1853 __func__, inode->i_ino, error ? "invalid" : "valid");
1854 return !error;
1855}
1856
1857/*
1858 * This is called from dput() when d_count is going to 0.
1859 */
1860static int nfs_dentry_delete(const struct dentry *dentry)
1861{
1862 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1863 dentry, dentry->d_flags);
1864
1865 /* Unhash any dentry with a stale inode */
1866 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1867 return 1;
1868
1869 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1870 /* Unhash it, so that ->d_iput() would be called */
1871 return 1;
1872 }
1873 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1874 /* Unhash it, so that ancestors of killed async unlink
1875 * files will be cleaned up during umount */
1876 return 1;
1877 }
1878 return 0;
1879
1880}
1881
1882/* Ensure that we revalidate inode->i_nlink */
1883static void nfs_drop_nlink(struct inode *inode)
1884{
1885 spin_lock(&inode->i_lock);
1886 /* drop the inode if we're reasonably sure this is the last link */
1887 if (inode->i_nlink > 0)
1888 drop_nlink(inode);
1889 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1890 nfs_set_cache_invalid(
1891 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1892 NFS_INO_INVALID_NLINK);
1893 spin_unlock(&inode->i_lock);
1894}
1895
1896/*
1897 * Called when the dentry loses inode.
1898 * We use it to clean up silly-renamed files.
1899 */
1900static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1901{
1902 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1903 nfs_complete_unlink(dentry, inode);
1904 nfs_drop_nlink(inode);
1905 }
1906 iput(inode);
1907}
1908
1909static void nfs_d_release(struct dentry *dentry)
1910{
1911 /* free cached devname value, if it survived that far */
1912 if (unlikely(dentry->d_fsdata)) {
1913 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1914 WARN_ON(1);
1915 else
1916 kfree(dentry->d_fsdata);
1917 }
1918}
1919
1920const struct dentry_operations nfs_dentry_operations = {
1921 .d_revalidate = nfs_lookup_revalidate,
1922 .d_weak_revalidate = nfs_weak_revalidate,
1923 .d_delete = nfs_dentry_delete,
1924 .d_iput = nfs_dentry_iput,
1925 .d_automount = nfs_d_automount,
1926 .d_release = nfs_d_release,
1927};
1928EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1929
1930struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1931{
1932 struct dentry *res;
1933 struct inode *inode = NULL;
1934 struct nfs_fh *fhandle = NULL;
1935 struct nfs_fattr *fattr = NULL;
1936 unsigned long dir_verifier;
1937 int error;
1938
1939 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1940 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1941
1942 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1943 return ERR_PTR(-ENAMETOOLONG);
1944
1945 /*
1946 * If we're doing an exclusive create, optimize away the lookup
1947 * but don't hash the dentry.
1948 */
1949 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1950 return NULL;
1951
1952 res = ERR_PTR(-ENOMEM);
1953 fhandle = nfs_alloc_fhandle();
1954 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1955 if (fhandle == NULL || fattr == NULL)
1956 goto out;
1957
1958 dir_verifier = nfs_save_change_attribute(dir);
1959 trace_nfs_lookup_enter(dir, dentry, flags);
1960 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1961 if (error == -ENOENT) {
1962 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1963 dir_verifier = inode_peek_iversion_raw(dir);
1964 goto no_entry;
1965 }
1966 if (error < 0) {
1967 res = ERR_PTR(error);
1968 goto out;
1969 }
1970 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1971 res = ERR_CAST(inode);
1972 if (IS_ERR(res))
1973 goto out;
1974
1975 /* Notify readdir to use READDIRPLUS */
1976 nfs_lookup_advise_force_readdirplus(dir, flags);
1977
1978no_entry:
1979 res = d_splice_alias(inode, dentry);
1980 if (res != NULL) {
1981 if (IS_ERR(res))
1982 goto out;
1983 dentry = res;
1984 }
1985 nfs_set_verifier(dentry, dir_verifier);
1986out:
1987 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1988 nfs_free_fattr(fattr);
1989 nfs_free_fhandle(fhandle);
1990 return res;
1991}
1992EXPORT_SYMBOL_GPL(nfs_lookup);
1993
1994void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1995{
1996 /* Case insensitive server? Revalidate dentries */
1997 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1998 d_prune_aliases(inode);
1999}
2000EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
2001
2002#if IS_ENABLED(CONFIG_NFS_V4)
2003static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
2004
2005const struct dentry_operations nfs4_dentry_operations = {
2006 .d_revalidate = nfs4_lookup_revalidate,
2007 .d_weak_revalidate = nfs_weak_revalidate,
2008 .d_delete = nfs_dentry_delete,
2009 .d_iput = nfs_dentry_iput,
2010 .d_automount = nfs_d_automount,
2011 .d_release = nfs_d_release,
2012};
2013EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2014
2015static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2016{
2017 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2018}
2019
2020static int do_open(struct inode *inode, struct file *filp)
2021{
2022 nfs_fscache_open_file(inode, filp);
2023 return 0;
2024}
2025
2026static int nfs_finish_open(struct nfs_open_context *ctx,
2027 struct dentry *dentry,
2028 struct file *file, unsigned open_flags)
2029{
2030 int err;
2031
2032 err = finish_open(file, dentry, do_open);
2033 if (err)
2034 goto out;
2035 if (S_ISREG(file_inode(file)->i_mode))
2036 nfs_file_set_open_context(file, ctx);
2037 else
2038 err = -EOPENSTALE;
2039out:
2040 return err;
2041}
2042
2043int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2044 struct file *file, unsigned open_flags,
2045 umode_t mode)
2046{
2047 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2048 struct nfs_open_context *ctx;
2049 struct dentry *res;
2050 struct iattr attr = { .ia_valid = ATTR_OPEN };
2051 struct inode *inode;
2052 unsigned int lookup_flags = 0;
2053 unsigned long dir_verifier;
2054 bool switched = false;
2055 int created = 0;
2056 int err;
2057
2058 /* Expect a negative dentry */
2059 BUG_ON(d_inode(dentry));
2060
2061 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2062 dir->i_sb->s_id, dir->i_ino, dentry);
2063
2064 err = nfs_check_flags(open_flags);
2065 if (err)
2066 return err;
2067
2068 /* NFS only supports OPEN on regular files */
2069 if ((open_flags & O_DIRECTORY)) {
2070 if (!d_in_lookup(dentry)) {
2071 /*
2072 * Hashed negative dentry with O_DIRECTORY: dentry was
2073 * revalidated and is fine, no need to perform lookup
2074 * again
2075 */
2076 return -ENOENT;
2077 }
2078 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2079 goto no_open;
2080 }
2081
2082 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2083 return -ENAMETOOLONG;
2084
2085 if (open_flags & O_CREAT) {
2086 struct nfs_server *server = NFS_SERVER(dir);
2087
2088 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2089 mode &= ~current_umask();
2090
2091 attr.ia_valid |= ATTR_MODE;
2092 attr.ia_mode = mode;
2093 }
2094 if (open_flags & O_TRUNC) {
2095 attr.ia_valid |= ATTR_SIZE;
2096 attr.ia_size = 0;
2097 }
2098
2099 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2100 d_drop(dentry);
2101 switched = true;
2102 dentry = d_alloc_parallel(dentry->d_parent,
2103 &dentry->d_name, &wq);
2104 if (IS_ERR(dentry))
2105 return PTR_ERR(dentry);
2106 if (unlikely(!d_in_lookup(dentry)))
2107 return finish_no_open(file, dentry);
2108 }
2109
2110 ctx = create_nfs_open_context(dentry, open_flags, file);
2111 err = PTR_ERR(ctx);
2112 if (IS_ERR(ctx))
2113 goto out;
2114
2115 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2116 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2117 if (created)
2118 file->f_mode |= FMODE_CREATED;
2119 if (IS_ERR(inode)) {
2120 err = PTR_ERR(inode);
2121 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2122 put_nfs_open_context(ctx);
2123 d_drop(dentry);
2124 switch (err) {
2125 case -ENOENT:
2126 d_splice_alias(NULL, dentry);
2127 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2128 dir_verifier = inode_peek_iversion_raw(dir);
2129 else
2130 dir_verifier = nfs_save_change_attribute(dir);
2131 nfs_set_verifier(dentry, dir_verifier);
2132 break;
2133 case -EISDIR:
2134 case -ENOTDIR:
2135 goto no_open;
2136 case -ELOOP:
2137 if (!(open_flags & O_NOFOLLOW))
2138 goto no_open;
2139 break;
2140 /* case -EINVAL: */
2141 default:
2142 break;
2143 }
2144 goto out;
2145 }
2146 file->f_mode |= FMODE_CAN_ODIRECT;
2147
2148 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2149 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2150 put_nfs_open_context(ctx);
2151out:
2152 if (unlikely(switched)) {
2153 d_lookup_done(dentry);
2154 dput(dentry);
2155 }
2156 return err;
2157
2158no_open:
2159 res = nfs_lookup(dir, dentry, lookup_flags);
2160 if (!res) {
2161 inode = d_inode(dentry);
2162 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2163 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2164 res = ERR_PTR(-ENOTDIR);
2165 else if (inode && S_ISREG(inode->i_mode))
2166 res = ERR_PTR(-EOPENSTALE);
2167 } else if (!IS_ERR(res)) {
2168 inode = d_inode(res);
2169 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2170 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2171 dput(res);
2172 res = ERR_PTR(-ENOTDIR);
2173 } else if (inode && S_ISREG(inode->i_mode)) {
2174 dput(res);
2175 res = ERR_PTR(-EOPENSTALE);
2176 }
2177 }
2178 if (switched) {
2179 d_lookup_done(dentry);
2180 if (!res)
2181 res = dentry;
2182 else
2183 dput(dentry);
2184 }
2185 if (IS_ERR(res))
2186 return PTR_ERR(res);
2187 return finish_no_open(file, res);
2188}
2189EXPORT_SYMBOL_GPL(nfs_atomic_open);
2190
2191static int
2192nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2193 unsigned int flags)
2194{
2195 struct inode *inode;
2196
2197 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
2198
2199 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2200 goto full_reval;
2201 if (d_mountpoint(dentry))
2202 goto full_reval;
2203
2204 inode = d_inode(dentry);
2205
2206 /* We can't create new files in nfs_open_revalidate(), so we
2207 * optimize away revalidation of negative dentries.
2208 */
2209 if (inode == NULL)
2210 goto full_reval;
2211
2212 if (nfs_verifier_is_delegated(dentry))
2213 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2214
2215 /* NFS only supports OPEN on regular files */
2216 if (!S_ISREG(inode->i_mode))
2217 goto full_reval;
2218
2219 /* We cannot do exclusive creation on a positive dentry */
2220 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2221 goto reval_dentry;
2222
2223 /* Check if the directory changed */
2224 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2225 goto reval_dentry;
2226
2227 /* Let f_op->open() actually open (and revalidate) the file */
2228 return 1;
2229reval_dentry:
2230 if (flags & LOOKUP_RCU)
2231 return -ECHILD;
2232 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2233
2234full_reval:
2235 return nfs_do_lookup_revalidate(dir, dentry, flags);
2236}
2237
2238static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2239{
2240 return __nfs_lookup_revalidate(dentry, flags,
2241 nfs4_do_lookup_revalidate);
2242}
2243
2244#endif /* CONFIG_NFSV4 */
2245
2246struct dentry *
2247nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2248 struct nfs_fattr *fattr)
2249{
2250 struct dentry *parent = dget_parent(dentry);
2251 struct inode *dir = d_inode(parent);
2252 struct inode *inode;
2253 struct dentry *d;
2254 int error;
2255
2256 d_drop(dentry);
2257
2258 if (fhandle->size == 0) {
2259 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2260 if (error)
2261 goto out_error;
2262 }
2263 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2264 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2265 struct nfs_server *server = NFS_SB(dentry->d_sb);
2266 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2267 fattr, NULL);
2268 if (error < 0)
2269 goto out_error;
2270 }
2271 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2272 d = d_splice_alias(inode, dentry);
2273out:
2274 dput(parent);
2275 return d;
2276out_error:
2277 d = ERR_PTR(error);
2278 goto out;
2279}
2280EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2281
2282/*
2283 * Code common to create, mkdir, and mknod.
2284 */
2285int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2286 struct nfs_fattr *fattr)
2287{
2288 struct dentry *d;
2289
2290 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2291 if (IS_ERR(d))
2292 return PTR_ERR(d);
2293
2294 /* Callers don't care */
2295 dput(d);
2296 return 0;
2297}
2298EXPORT_SYMBOL_GPL(nfs_instantiate);
2299
2300/*
2301 * Following a failed create operation, we drop the dentry rather
2302 * than retain a negative dentry. This avoids a problem in the event
2303 * that the operation succeeded on the server, but an error in the
2304 * reply path made it appear to have failed.
2305 */
2306int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2307 struct dentry *dentry, umode_t mode, bool excl)
2308{
2309 struct iattr attr;
2310 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2311 int error;
2312
2313 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2314 dir->i_sb->s_id, dir->i_ino, dentry);
2315
2316 attr.ia_mode = mode;
2317 attr.ia_valid = ATTR_MODE;
2318
2319 trace_nfs_create_enter(dir, dentry, open_flags);
2320 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2321 trace_nfs_create_exit(dir, dentry, open_flags, error);
2322 if (error != 0)
2323 goto out_err;
2324 return 0;
2325out_err:
2326 d_drop(dentry);
2327 return error;
2328}
2329EXPORT_SYMBOL_GPL(nfs_create);
2330
2331/*
2332 * See comments for nfs_proc_create regarding failed operations.
2333 */
2334int
2335nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2336 struct dentry *dentry, umode_t mode, dev_t rdev)
2337{
2338 struct iattr attr;
2339 int status;
2340
2341 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2342 dir->i_sb->s_id, dir->i_ino, dentry);
2343
2344 attr.ia_mode = mode;
2345 attr.ia_valid = ATTR_MODE;
2346
2347 trace_nfs_mknod_enter(dir, dentry);
2348 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2349 trace_nfs_mknod_exit(dir, dentry, status);
2350 if (status != 0)
2351 goto out_err;
2352 return 0;
2353out_err:
2354 d_drop(dentry);
2355 return status;
2356}
2357EXPORT_SYMBOL_GPL(nfs_mknod);
2358
2359/*
2360 * See comments for nfs_proc_create regarding failed operations.
2361 */
2362int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2363 struct dentry *dentry, umode_t mode)
2364{
2365 struct iattr attr;
2366 int error;
2367
2368 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2369 dir->i_sb->s_id, dir->i_ino, dentry);
2370
2371 attr.ia_valid = ATTR_MODE;
2372 attr.ia_mode = mode | S_IFDIR;
2373
2374 trace_nfs_mkdir_enter(dir, dentry);
2375 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2376 trace_nfs_mkdir_exit(dir, dentry, error);
2377 if (error != 0)
2378 goto out_err;
2379 return 0;
2380out_err:
2381 d_drop(dentry);
2382 return error;
2383}
2384EXPORT_SYMBOL_GPL(nfs_mkdir);
2385
2386static void nfs_dentry_handle_enoent(struct dentry *dentry)
2387{
2388 if (simple_positive(dentry))
2389 d_delete(dentry);
2390}
2391
2392static void nfs_dentry_remove_handle_error(struct inode *dir,
2393 struct dentry *dentry, int error)
2394{
2395 switch (error) {
2396 case -ENOENT:
2397 if (d_really_is_positive(dentry))
2398 d_delete(dentry);
2399 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2400 break;
2401 case 0:
2402 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2403 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2404 }
2405}
2406
2407int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2408{
2409 int error;
2410
2411 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2412 dir->i_sb->s_id, dir->i_ino, dentry);
2413
2414 trace_nfs_rmdir_enter(dir, dentry);
2415 if (d_really_is_positive(dentry)) {
2416 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2417 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2418 /* Ensure the VFS deletes this inode */
2419 switch (error) {
2420 case 0:
2421 clear_nlink(d_inode(dentry));
2422 break;
2423 case -ENOENT:
2424 nfs_dentry_handle_enoent(dentry);
2425 }
2426 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2427 } else
2428 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2429 nfs_dentry_remove_handle_error(dir, dentry, error);
2430 trace_nfs_rmdir_exit(dir, dentry, error);
2431
2432 return error;
2433}
2434EXPORT_SYMBOL_GPL(nfs_rmdir);
2435
2436/*
2437 * Remove a file after making sure there are no pending writes,
2438 * and after checking that the file has only one user.
2439 *
2440 * We invalidate the attribute cache and free the inode prior to the operation
2441 * to avoid possible races if the server reuses the inode.
2442 */
2443static int nfs_safe_remove(struct dentry *dentry)
2444{
2445 struct inode *dir = d_inode(dentry->d_parent);
2446 struct inode *inode = d_inode(dentry);
2447 int error = -EBUSY;
2448
2449 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2450
2451 /* If the dentry was sillyrenamed, we simply call d_delete() */
2452 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2453 error = 0;
2454 goto out;
2455 }
2456
2457 trace_nfs_remove_enter(dir, dentry);
2458 if (inode != NULL) {
2459 error = NFS_PROTO(dir)->remove(dir, dentry);
2460 if (error == 0)
2461 nfs_drop_nlink(inode);
2462 } else
2463 error = NFS_PROTO(dir)->remove(dir, dentry);
2464 if (error == -ENOENT)
2465 nfs_dentry_handle_enoent(dentry);
2466 trace_nfs_remove_exit(dir, dentry, error);
2467out:
2468 return error;
2469}
2470
2471/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2472 * belongs to an active ".nfs..." file and we return -EBUSY.
2473 *
2474 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2475 */
2476int nfs_unlink(struct inode *dir, struct dentry *dentry)
2477{
2478 int error;
2479
2480 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2481 dir->i_ino, dentry);
2482
2483 trace_nfs_unlink_enter(dir, dentry);
2484 spin_lock(&dentry->d_lock);
2485 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2486 &NFS_I(d_inode(dentry))->flags)) {
2487 spin_unlock(&dentry->d_lock);
2488 /* Start asynchronous writeout of the inode */
2489 write_inode_now(d_inode(dentry), 0);
2490 error = nfs_sillyrename(dir, dentry);
2491 goto out;
2492 }
2493 /* We must prevent any concurrent open until the unlink
2494 * completes. ->d_revalidate will wait for ->d_fsdata
2495 * to clear. We set it here to ensure no lookup succeeds until
2496 * the unlink is complete on the server.
2497 */
2498 error = -ETXTBSY;
2499 if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2500 WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2501 spin_unlock(&dentry->d_lock);
2502 goto out;
2503 }
2504 /* old devname */
2505 kfree(dentry->d_fsdata);
2506 dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2507
2508 spin_unlock(&dentry->d_lock);
2509 error = nfs_safe_remove(dentry);
2510 nfs_dentry_remove_handle_error(dir, dentry, error);
2511 dentry->d_fsdata = NULL;
2512 wake_up_var(&dentry->d_fsdata);
2513out:
2514 trace_nfs_unlink_exit(dir, dentry, error);
2515 return error;
2516}
2517EXPORT_SYMBOL_GPL(nfs_unlink);
2518
2519/*
2520 * To create a symbolic link, most file systems instantiate a new inode,
2521 * add a page to it containing the path, then write it out to the disk
2522 * using prepare_write/commit_write.
2523 *
2524 * Unfortunately the NFS client can't create the in-core inode first
2525 * because it needs a file handle to create an in-core inode (see
2526 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2527 * symlink request has completed on the server.
2528 *
2529 * So instead we allocate a raw page, copy the symname into it, then do
2530 * the SYMLINK request with the page as the buffer. If it succeeds, we
2531 * now have a new file handle and can instantiate an in-core NFS inode
2532 * and move the raw page into its mapping.
2533 */
2534int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2535 struct dentry *dentry, const char *symname)
2536{
2537 struct folio *folio;
2538 char *kaddr;
2539 struct iattr attr;
2540 unsigned int pathlen = strlen(symname);
2541 int error;
2542
2543 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2544 dir->i_ino, dentry, symname);
2545
2546 if (pathlen > PAGE_SIZE)
2547 return -ENAMETOOLONG;
2548
2549 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2550 attr.ia_valid = ATTR_MODE;
2551
2552 folio = folio_alloc(GFP_USER, 0);
2553 if (!folio)
2554 return -ENOMEM;
2555
2556 kaddr = folio_address(folio);
2557 memcpy(kaddr, symname, pathlen);
2558 if (pathlen < PAGE_SIZE)
2559 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2560
2561 trace_nfs_symlink_enter(dir, dentry);
2562 error = NFS_PROTO(dir)->symlink(dir, dentry, folio, pathlen, &attr);
2563 trace_nfs_symlink_exit(dir, dentry, error);
2564 if (error != 0) {
2565 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2566 dir->i_sb->s_id, dir->i_ino,
2567 dentry, symname, error);
2568 d_drop(dentry);
2569 folio_put(folio);
2570 return error;
2571 }
2572
2573 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2574
2575 /*
2576 * No big deal if we can't add this page to the page cache here.
2577 * READLINK will get the missing page from the server if needed.
2578 */
2579 if (filemap_add_folio(d_inode(dentry)->i_mapping, folio, 0,
2580 GFP_KERNEL) == 0) {
2581 folio_mark_uptodate(folio);
2582 folio_unlock(folio);
2583 }
2584
2585 folio_put(folio);
2586 return 0;
2587}
2588EXPORT_SYMBOL_GPL(nfs_symlink);
2589
2590int
2591nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2592{
2593 struct inode *inode = d_inode(old_dentry);
2594 int error;
2595
2596 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2597 old_dentry, dentry);
2598
2599 trace_nfs_link_enter(inode, dir, dentry);
2600 d_drop(dentry);
2601 if (S_ISREG(inode->i_mode))
2602 nfs_sync_inode(inode);
2603 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2604 if (error == 0) {
2605 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2606 ihold(inode);
2607 d_add(dentry, inode);
2608 }
2609 trace_nfs_link_exit(inode, dir, dentry, error);
2610 return error;
2611}
2612EXPORT_SYMBOL_GPL(nfs_link);
2613
2614static void
2615nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2616{
2617 struct dentry *new_dentry = data->new_dentry;
2618
2619 new_dentry->d_fsdata = NULL;
2620 wake_up_var(&new_dentry->d_fsdata);
2621}
2622
2623/*
2624 * RENAME
2625 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2626 * different file handle for the same inode after a rename (e.g. when
2627 * moving to a different directory). A fail-safe method to do so would
2628 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2629 * rename the old file using the sillyrename stuff. This way, the original
2630 * file in old_dir will go away when the last process iput()s the inode.
2631 *
2632 * FIXED.
2633 *
2634 * It actually works quite well. One needs to have the possibility for
2635 * at least one ".nfs..." file in each directory the file ever gets
2636 * moved or linked to which happens automagically with the new
2637 * implementation that only depends on the dcache stuff instead of
2638 * using the inode layer
2639 *
2640 * Unfortunately, things are a little more complicated than indicated
2641 * above. For a cross-directory move, we want to make sure we can get
2642 * rid of the old inode after the operation. This means there must be
2643 * no pending writes (if it's a file), and the use count must be 1.
2644 * If these conditions are met, we can drop the dentries before doing
2645 * the rename.
2646 */
2647int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2648 struct dentry *old_dentry, struct inode *new_dir,
2649 struct dentry *new_dentry, unsigned int flags)
2650{
2651 struct inode *old_inode = d_inode(old_dentry);
2652 struct inode *new_inode = d_inode(new_dentry);
2653 struct dentry *dentry = NULL;
2654 struct rpc_task *task;
2655 bool must_unblock = false;
2656 int error = -EBUSY;
2657
2658 if (flags)
2659 return -EINVAL;
2660
2661 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2662 old_dentry, new_dentry,
2663 d_count(new_dentry));
2664
2665 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2666 /*
2667 * For non-directories, check whether the target is busy and if so,
2668 * make a copy of the dentry and then do a silly-rename. If the
2669 * silly-rename succeeds, the copied dentry is hashed and becomes
2670 * the new target.
2671 */
2672 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2673 /* We must prevent any concurrent open until the unlink
2674 * completes. ->d_revalidate will wait for ->d_fsdata
2675 * to clear. We set it here to ensure no lookup succeeds until
2676 * the unlink is complete on the server.
2677 */
2678 error = -ETXTBSY;
2679 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2680 WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2681 goto out;
2682 if (new_dentry->d_fsdata) {
2683 /* old devname */
2684 kfree(new_dentry->d_fsdata);
2685 new_dentry->d_fsdata = NULL;
2686 }
2687
2688 spin_lock(&new_dentry->d_lock);
2689 if (d_count(new_dentry) > 2) {
2690 int err;
2691
2692 spin_unlock(&new_dentry->d_lock);
2693
2694 /* copy the target dentry's name */
2695 dentry = d_alloc(new_dentry->d_parent,
2696 &new_dentry->d_name);
2697 if (!dentry)
2698 goto out;
2699
2700 /* silly-rename the existing target ... */
2701 err = nfs_sillyrename(new_dir, new_dentry);
2702 if (err)
2703 goto out;
2704
2705 new_dentry = dentry;
2706 new_inode = NULL;
2707 } else {
2708 new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2709 must_unblock = true;
2710 spin_unlock(&new_dentry->d_lock);
2711 }
2712
2713 }
2714
2715 if (S_ISREG(old_inode->i_mode))
2716 nfs_sync_inode(old_inode);
2717 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2718 must_unblock ? nfs_unblock_rename : NULL);
2719 if (IS_ERR(task)) {
2720 error = PTR_ERR(task);
2721 goto out;
2722 }
2723
2724 error = rpc_wait_for_completion_task(task);
2725 if (error != 0) {
2726 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2727 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2728 smp_wmb();
2729 } else
2730 error = task->tk_status;
2731 rpc_put_task(task);
2732 /* Ensure the inode attributes are revalidated */
2733 if (error == 0) {
2734 spin_lock(&old_inode->i_lock);
2735 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2736 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2737 NFS_INO_INVALID_CTIME |
2738 NFS_INO_REVAL_FORCED);
2739 spin_unlock(&old_inode->i_lock);
2740 }
2741out:
2742 trace_nfs_rename_exit(old_dir, old_dentry,
2743 new_dir, new_dentry, error);
2744 if (!error) {
2745 if (new_inode != NULL)
2746 nfs_drop_nlink(new_inode);
2747 /*
2748 * The d_move() should be here instead of in an async RPC completion
2749 * handler because we need the proper locks to move the dentry. If
2750 * we're interrupted by a signal, the async RPC completion handler
2751 * should mark the directories for revalidation.
2752 */
2753 d_move(old_dentry, new_dentry);
2754 nfs_set_verifier(old_dentry,
2755 nfs_save_change_attribute(new_dir));
2756 } else if (error == -ENOENT)
2757 nfs_dentry_handle_enoent(old_dentry);
2758
2759 /* new dentry created? */
2760 if (dentry)
2761 dput(dentry);
2762 return error;
2763}
2764EXPORT_SYMBOL_GPL(nfs_rename);
2765
2766static DEFINE_SPINLOCK(nfs_access_lru_lock);
2767static LIST_HEAD(nfs_access_lru_list);
2768static atomic_long_t nfs_access_nr_entries;
2769
2770static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2771module_param(nfs_access_max_cachesize, ulong, 0644);
2772MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2773
2774static void nfs_access_free_entry(struct nfs_access_entry *entry)
2775{
2776 put_group_info(entry->group_info);
2777 kfree_rcu(entry, rcu_head);
2778 smp_mb__before_atomic();
2779 atomic_long_dec(&nfs_access_nr_entries);
2780 smp_mb__after_atomic();
2781}
2782
2783static void nfs_access_free_list(struct list_head *head)
2784{
2785 struct nfs_access_entry *cache;
2786
2787 while (!list_empty(head)) {
2788 cache = list_entry(head->next, struct nfs_access_entry, lru);
2789 list_del(&cache->lru);
2790 nfs_access_free_entry(cache);
2791 }
2792}
2793
2794static unsigned long
2795nfs_do_access_cache_scan(unsigned int nr_to_scan)
2796{
2797 LIST_HEAD(head);
2798 struct nfs_inode *nfsi, *next;
2799 struct nfs_access_entry *cache;
2800 long freed = 0;
2801
2802 spin_lock(&nfs_access_lru_lock);
2803 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2804 struct inode *inode;
2805
2806 if (nr_to_scan-- == 0)
2807 break;
2808 inode = &nfsi->vfs_inode;
2809 spin_lock(&inode->i_lock);
2810 if (list_empty(&nfsi->access_cache_entry_lru))
2811 goto remove_lru_entry;
2812 cache = list_entry(nfsi->access_cache_entry_lru.next,
2813 struct nfs_access_entry, lru);
2814 list_move(&cache->lru, &head);
2815 rb_erase(&cache->rb_node, &nfsi->access_cache);
2816 freed++;
2817 if (!list_empty(&nfsi->access_cache_entry_lru))
2818 list_move_tail(&nfsi->access_cache_inode_lru,
2819 &nfs_access_lru_list);
2820 else {
2821remove_lru_entry:
2822 list_del_init(&nfsi->access_cache_inode_lru);
2823 smp_mb__before_atomic();
2824 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2825 smp_mb__after_atomic();
2826 }
2827 spin_unlock(&inode->i_lock);
2828 }
2829 spin_unlock(&nfs_access_lru_lock);
2830 nfs_access_free_list(&head);
2831 return freed;
2832}
2833
2834unsigned long
2835nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2836{
2837 int nr_to_scan = sc->nr_to_scan;
2838 gfp_t gfp_mask = sc->gfp_mask;
2839
2840 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2841 return SHRINK_STOP;
2842 return nfs_do_access_cache_scan(nr_to_scan);
2843}
2844
2845
2846unsigned long
2847nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2848{
2849 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2850}
2851
2852static void
2853nfs_access_cache_enforce_limit(void)
2854{
2855 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2856 unsigned long diff;
2857 unsigned int nr_to_scan;
2858
2859 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2860 return;
2861 nr_to_scan = 100;
2862 diff = nr_entries - nfs_access_max_cachesize;
2863 if (diff < nr_to_scan)
2864 nr_to_scan = diff;
2865 nfs_do_access_cache_scan(nr_to_scan);
2866}
2867
2868static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2869{
2870 struct rb_root *root_node = &nfsi->access_cache;
2871 struct rb_node *n;
2872 struct nfs_access_entry *entry;
2873
2874 /* Unhook entries from the cache */
2875 while ((n = rb_first(root_node)) != NULL) {
2876 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2877 rb_erase(n, root_node);
2878 list_move(&entry->lru, head);
2879 }
2880 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2881}
2882
2883void nfs_access_zap_cache(struct inode *inode)
2884{
2885 LIST_HEAD(head);
2886
2887 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2888 return;
2889 /* Remove from global LRU init */
2890 spin_lock(&nfs_access_lru_lock);
2891 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2892 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2893
2894 spin_lock(&inode->i_lock);
2895 __nfs_access_zap_cache(NFS_I(inode), &head);
2896 spin_unlock(&inode->i_lock);
2897 spin_unlock(&nfs_access_lru_lock);
2898 nfs_access_free_list(&head);
2899}
2900EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2901
2902static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2903{
2904 struct group_info *ga, *gb;
2905 int g;
2906
2907 if (uid_lt(a->fsuid, b->fsuid))
2908 return -1;
2909 if (uid_gt(a->fsuid, b->fsuid))
2910 return 1;
2911
2912 if (gid_lt(a->fsgid, b->fsgid))
2913 return -1;
2914 if (gid_gt(a->fsgid, b->fsgid))
2915 return 1;
2916
2917 ga = a->group_info;
2918 gb = b->group_info;
2919 if (ga == gb)
2920 return 0;
2921 if (ga == NULL)
2922 return -1;
2923 if (gb == NULL)
2924 return 1;
2925 if (ga->ngroups < gb->ngroups)
2926 return -1;
2927 if (ga->ngroups > gb->ngroups)
2928 return 1;
2929
2930 for (g = 0; g < ga->ngroups; g++) {
2931 if (gid_lt(ga->gid[g], gb->gid[g]))
2932 return -1;
2933 if (gid_gt(ga->gid[g], gb->gid[g]))
2934 return 1;
2935 }
2936 return 0;
2937}
2938
2939static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2940{
2941 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2942
2943 while (n != NULL) {
2944 struct nfs_access_entry *entry =
2945 rb_entry(n, struct nfs_access_entry, rb_node);
2946 int cmp = access_cmp(cred, entry);
2947
2948 if (cmp < 0)
2949 n = n->rb_left;
2950 else if (cmp > 0)
2951 n = n->rb_right;
2952 else
2953 return entry;
2954 }
2955 return NULL;
2956}
2957
2958static u64 nfs_access_login_time(const struct task_struct *task,
2959 const struct cred *cred)
2960{
2961 const struct task_struct *parent;
2962 const struct cred *pcred;
2963 u64 ret;
2964
2965 rcu_read_lock();
2966 for (;;) {
2967 parent = rcu_dereference(task->real_parent);
2968 pcred = __task_cred(parent);
2969 if (parent == task || cred_fscmp(pcred, cred) != 0)
2970 break;
2971 task = parent;
2972 }
2973 ret = task->start_time;
2974 rcu_read_unlock();
2975 return ret;
2976}
2977
2978static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2979{
2980 struct nfs_inode *nfsi = NFS_I(inode);
2981 u64 login_time = nfs_access_login_time(current, cred);
2982 struct nfs_access_entry *cache;
2983 bool retry = true;
2984 int err;
2985
2986 spin_lock(&inode->i_lock);
2987 for(;;) {
2988 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2989 goto out_zap;
2990 cache = nfs_access_search_rbtree(inode, cred);
2991 err = -ENOENT;
2992 if (cache == NULL)
2993 goto out;
2994 /* Found an entry, is our attribute cache valid? */
2995 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2996 break;
2997 if (!retry)
2998 break;
2999 err = -ECHILD;
3000 if (!may_block)
3001 goto out;
3002 spin_unlock(&inode->i_lock);
3003 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3004 if (err)
3005 return err;
3006 spin_lock(&inode->i_lock);
3007 retry = false;
3008 }
3009 err = -ENOENT;
3010 if ((s64)(login_time - cache->timestamp) > 0)
3011 goto out;
3012 *mask = cache->mask;
3013 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3014 err = 0;
3015out:
3016 spin_unlock(&inode->i_lock);
3017 return err;
3018out_zap:
3019 spin_unlock(&inode->i_lock);
3020 nfs_access_zap_cache(inode);
3021 return -ENOENT;
3022}
3023
3024static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3025{
3026 /* Only check the most recently returned cache entry,
3027 * but do it without locking.
3028 */
3029 struct nfs_inode *nfsi = NFS_I(inode);
3030 u64 login_time = nfs_access_login_time(current, cred);
3031 struct nfs_access_entry *cache;
3032 int err = -ECHILD;
3033 struct list_head *lh;
3034
3035 rcu_read_lock();
3036 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3037 goto out;
3038 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3039 cache = list_entry(lh, struct nfs_access_entry, lru);
3040 if (lh == &nfsi->access_cache_entry_lru ||
3041 access_cmp(cred, cache) != 0)
3042 cache = NULL;
3043 if (cache == NULL)
3044 goto out;
3045 if ((s64)(login_time - cache->timestamp) > 0)
3046 goto out;
3047 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3048 goto out;
3049 *mask = cache->mask;
3050 err = 0;
3051out:
3052 rcu_read_unlock();
3053 return err;
3054}
3055
3056int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3057 u32 *mask, bool may_block)
3058{
3059 int status;
3060
3061 status = nfs_access_get_cached_rcu(inode, cred, mask);
3062 if (status != 0)
3063 status = nfs_access_get_cached_locked(inode, cred, mask,
3064 may_block);
3065
3066 return status;
3067}
3068EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3069
3070static void nfs_access_add_rbtree(struct inode *inode,
3071 struct nfs_access_entry *set,
3072 const struct cred *cred)
3073{
3074 struct nfs_inode *nfsi = NFS_I(inode);
3075 struct rb_root *root_node = &nfsi->access_cache;
3076 struct rb_node **p = &root_node->rb_node;
3077 struct rb_node *parent = NULL;
3078 struct nfs_access_entry *entry;
3079 int cmp;
3080
3081 spin_lock(&inode->i_lock);
3082 while (*p != NULL) {
3083 parent = *p;
3084 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3085 cmp = access_cmp(cred, entry);
3086
3087 if (cmp < 0)
3088 p = &parent->rb_left;
3089 else if (cmp > 0)
3090 p = &parent->rb_right;
3091 else
3092 goto found;
3093 }
3094 rb_link_node(&set->rb_node, parent, p);
3095 rb_insert_color(&set->rb_node, root_node);
3096 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3097 spin_unlock(&inode->i_lock);
3098 return;
3099found:
3100 rb_replace_node(parent, &set->rb_node, root_node);
3101 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3102 list_del(&entry->lru);
3103 spin_unlock(&inode->i_lock);
3104 nfs_access_free_entry(entry);
3105}
3106
3107void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3108 const struct cred *cred)
3109{
3110 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3111 if (cache == NULL)
3112 return;
3113 RB_CLEAR_NODE(&cache->rb_node);
3114 cache->fsuid = cred->fsuid;
3115 cache->fsgid = cred->fsgid;
3116 cache->group_info = get_group_info(cred->group_info);
3117 cache->mask = set->mask;
3118 cache->timestamp = ktime_get_ns();
3119
3120 /* The above field assignments must be visible
3121 * before this item appears on the lru. We cannot easily
3122 * use rcu_assign_pointer, so just force the memory barrier.
3123 */
3124 smp_wmb();
3125 nfs_access_add_rbtree(inode, cache, cred);
3126
3127 /* Update accounting */
3128 smp_mb__before_atomic();
3129 atomic_long_inc(&nfs_access_nr_entries);
3130 smp_mb__after_atomic();
3131
3132 /* Add inode to global LRU list */
3133 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3134 spin_lock(&nfs_access_lru_lock);
3135 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3136 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3137 &nfs_access_lru_list);
3138 spin_unlock(&nfs_access_lru_lock);
3139 }
3140 nfs_access_cache_enforce_limit();
3141}
3142EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3143
3144#define NFS_MAY_READ (NFS_ACCESS_READ)
3145#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3146 NFS_ACCESS_EXTEND | \
3147 NFS_ACCESS_DELETE)
3148#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3149 NFS_ACCESS_EXTEND)
3150#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3151#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3152#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3153static int
3154nfs_access_calc_mask(u32 access_result, umode_t umode)
3155{
3156 int mask = 0;
3157
3158 if (access_result & NFS_MAY_READ)
3159 mask |= MAY_READ;
3160 if (S_ISDIR(umode)) {
3161 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3162 mask |= MAY_WRITE;
3163 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3164 mask |= MAY_EXEC;
3165 } else if (S_ISREG(umode)) {
3166 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3167 mask |= MAY_WRITE;
3168 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3169 mask |= MAY_EXEC;
3170 } else if (access_result & NFS_MAY_WRITE)
3171 mask |= MAY_WRITE;
3172 return mask;
3173}
3174
3175void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3176{
3177 entry->mask = access_result;
3178}
3179EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3180
3181static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3182{
3183 struct nfs_access_entry cache;
3184 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3185 int cache_mask = -1;
3186 int status;
3187
3188 trace_nfs_access_enter(inode);
3189
3190 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3191 if (status == 0)
3192 goto out_cached;
3193
3194 status = -ECHILD;
3195 if (!may_block)
3196 goto out;
3197
3198 /*
3199 * Determine which access bits we want to ask for...
3200 */
3201 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3202 nfs_access_xattr_mask(NFS_SERVER(inode));
3203 if (S_ISDIR(inode->i_mode))
3204 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3205 else
3206 cache.mask |= NFS_ACCESS_EXECUTE;
3207 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3208 if (status != 0) {
3209 if (status == -ESTALE) {
3210 if (!S_ISDIR(inode->i_mode))
3211 nfs_set_inode_stale(inode);
3212 else
3213 nfs_zap_caches(inode);
3214 }
3215 goto out;
3216 }
3217 nfs_access_add_cache(inode, &cache, cred);
3218out_cached:
3219 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3220 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3221 status = -EACCES;
3222out:
3223 trace_nfs_access_exit(inode, mask, cache_mask, status);
3224 return status;
3225}
3226
3227static int nfs_open_permission_mask(int openflags)
3228{
3229 int mask = 0;
3230
3231 if (openflags & __FMODE_EXEC) {
3232 /* ONLY check exec rights */
3233 mask = MAY_EXEC;
3234 } else {
3235 if ((openflags & O_ACCMODE) != O_WRONLY)
3236 mask |= MAY_READ;
3237 if ((openflags & O_ACCMODE) != O_RDONLY)
3238 mask |= MAY_WRITE;
3239 }
3240
3241 return mask;
3242}
3243
3244int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3245{
3246 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3247}
3248EXPORT_SYMBOL_GPL(nfs_may_open);
3249
3250static int nfs_execute_ok(struct inode *inode, int mask)
3251{
3252 struct nfs_server *server = NFS_SERVER(inode);
3253 int ret = 0;
3254
3255 if (S_ISDIR(inode->i_mode))
3256 return 0;
3257 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3258 if (mask & MAY_NOT_BLOCK)
3259 return -ECHILD;
3260 ret = __nfs_revalidate_inode(server, inode);
3261 }
3262 if (ret == 0 && !execute_ok(inode))
3263 ret = -EACCES;
3264 return ret;
3265}
3266
3267int nfs_permission(struct mnt_idmap *idmap,
3268 struct inode *inode,
3269 int mask)
3270{
3271 const struct cred *cred = current_cred();
3272 int res = 0;
3273
3274 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3275
3276 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3277 goto out;
3278 /* Is this sys_access() ? */
3279 if (mask & (MAY_ACCESS | MAY_CHDIR))
3280 goto force_lookup;
3281
3282 switch (inode->i_mode & S_IFMT) {
3283 case S_IFLNK:
3284 goto out;
3285 case S_IFREG:
3286 if ((mask & MAY_OPEN) &&
3287 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3288 return 0;
3289 break;
3290 case S_IFDIR:
3291 /*
3292 * Optimize away all write operations, since the server
3293 * will check permissions when we perform the op.
3294 */
3295 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3296 goto out;
3297 }
3298
3299force_lookup:
3300 if (!NFS_PROTO(inode)->access)
3301 goto out_notsup;
3302
3303 res = nfs_do_access(inode, cred, mask);
3304out:
3305 if (!res && (mask & MAY_EXEC))
3306 res = nfs_execute_ok(inode, mask);
3307
3308 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3309 inode->i_sb->s_id, inode->i_ino, mask, res);
3310 return res;
3311out_notsup:
3312 if (mask & MAY_NOT_BLOCK)
3313 return -ECHILD;
3314
3315 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3316 NFS_INO_INVALID_OTHER);
3317 if (res == 0)
3318 res = generic_permission(&nop_mnt_idmap, inode, mask);
3319 goto out;
3320}
3321EXPORT_SYMBOL_GPL(nfs_permission);