NFS: fix the handling of NFS_INO_INVALID_DATA flag in nfs_revalidate_mapping
[deliverable/linux.git] / fs / nfs / dir.c
1 /*
2 * linux/fs/nfs/dir.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs directory handling functions
7 *
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
18 */
19
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
28 #include <linux/mm.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
40
41 #include "delegation.h"
42 #include "iostat.h"
43 #include "internal.h"
44 #include "fscache.h"
45
46 #include "nfstrace.h"
47
48 /* #define NFS_DEBUG_VERBOSE 1 */
49
50 static int nfs_opendir(struct inode *, struct file *);
51 static int nfs_closedir(struct inode *, struct file *);
52 static int nfs_readdir(struct file *, struct dir_context *);
53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55 static void nfs_readdir_clear_array(struct page*);
56
57 const struct file_operations nfs_dir_operations = {
58 .llseek = nfs_llseek_dir,
59 .read = generic_read_dir,
60 .iterate = nfs_readdir,
61 .open = nfs_opendir,
62 .release = nfs_closedir,
63 .fsync = nfs_fsync_dir,
64 };
65
66 const struct address_space_operations nfs_dir_aops = {
67 .freepage = nfs_readdir_clear_array,
68 };
69
70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
71 {
72 struct nfs_open_dir_context *ctx;
73 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
74 if (ctx != NULL) {
75 ctx->duped = 0;
76 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
77 ctx->dir_cookie = 0;
78 ctx->dup_cookie = 0;
79 ctx->cred = get_rpccred(cred);
80 return ctx;
81 }
82 return ERR_PTR(-ENOMEM);
83 }
84
85 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
86 {
87 put_rpccred(ctx->cred);
88 kfree(ctx);
89 }
90
91 /*
92 * Open file
93 */
94 static int
95 nfs_opendir(struct inode *inode, struct file *filp)
96 {
97 int res = 0;
98 struct nfs_open_dir_context *ctx;
99 struct rpc_cred *cred;
100
101 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
102
103 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
104
105 cred = rpc_lookup_cred();
106 if (IS_ERR(cred))
107 return PTR_ERR(cred);
108 ctx = alloc_nfs_open_dir_context(inode, cred);
109 if (IS_ERR(ctx)) {
110 res = PTR_ERR(ctx);
111 goto out;
112 }
113 filp->private_data = ctx;
114 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
115 /* This is a mountpoint, so d_revalidate will never
116 * have been called, so we need to refresh the
117 * inode (for close-open consistency) ourselves.
118 */
119 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
120 }
121 out:
122 put_rpccred(cred);
123 return res;
124 }
125
126 static int
127 nfs_closedir(struct inode *inode, struct file *filp)
128 {
129 put_nfs_open_dir_context(filp->private_data);
130 return 0;
131 }
132
133 struct nfs_cache_array_entry {
134 u64 cookie;
135 u64 ino;
136 struct qstr string;
137 unsigned char d_type;
138 };
139
140 struct nfs_cache_array {
141 int size;
142 int eof_index;
143 u64 last_cookie;
144 struct nfs_cache_array_entry array[0];
145 };
146
147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
148 typedef struct {
149 struct file *file;
150 struct page *page;
151 struct dir_context *ctx;
152 unsigned long page_index;
153 u64 *dir_cookie;
154 u64 last_cookie;
155 loff_t current_index;
156 decode_dirent_t decode;
157
158 unsigned long timestamp;
159 unsigned long gencount;
160 unsigned int cache_entry_index;
161 unsigned int plus:1;
162 unsigned int eof:1;
163 } nfs_readdir_descriptor_t;
164
165 /*
166 * The caller is responsible for calling nfs_readdir_release_array(page)
167 */
168 static
169 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
170 {
171 void *ptr;
172 if (page == NULL)
173 return ERR_PTR(-EIO);
174 ptr = kmap(page);
175 if (ptr == NULL)
176 return ERR_PTR(-ENOMEM);
177 return ptr;
178 }
179
180 static
181 void nfs_readdir_release_array(struct page *page)
182 {
183 kunmap(page);
184 }
185
186 /*
187 * we are freeing strings created by nfs_add_to_readdir_array()
188 */
189 static
190 void nfs_readdir_clear_array(struct page *page)
191 {
192 struct nfs_cache_array *array;
193 int i;
194
195 array = kmap_atomic(page);
196 for (i = 0; i < array->size; i++)
197 kfree(array->array[i].string.name);
198 kunmap_atomic(array);
199 }
200
201 /*
202 * the caller is responsible for freeing qstr.name
203 * when called by nfs_readdir_add_to_array, the strings will be freed in
204 * nfs_clear_readdir_array()
205 */
206 static
207 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
208 {
209 string->len = len;
210 string->name = kmemdup(name, len, GFP_KERNEL);
211 if (string->name == NULL)
212 return -ENOMEM;
213 /*
214 * Avoid a kmemleak false positive. The pointer to the name is stored
215 * in a page cache page which kmemleak does not scan.
216 */
217 kmemleak_not_leak(string->name);
218 string->hash = full_name_hash(name, len);
219 return 0;
220 }
221
222 static
223 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
224 {
225 struct nfs_cache_array *array = nfs_readdir_get_array(page);
226 struct nfs_cache_array_entry *cache_entry;
227 int ret;
228
229 if (IS_ERR(array))
230 return PTR_ERR(array);
231
232 cache_entry = &array->array[array->size];
233
234 /* Check that this entry lies within the page bounds */
235 ret = -ENOSPC;
236 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
237 goto out;
238
239 cache_entry->cookie = entry->prev_cookie;
240 cache_entry->ino = entry->ino;
241 cache_entry->d_type = entry->d_type;
242 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
243 if (ret)
244 goto out;
245 array->last_cookie = entry->cookie;
246 array->size++;
247 if (entry->eof != 0)
248 array->eof_index = array->size;
249 out:
250 nfs_readdir_release_array(page);
251 return ret;
252 }
253
254 static
255 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
256 {
257 loff_t diff = desc->ctx->pos - desc->current_index;
258 unsigned int index;
259
260 if (diff < 0)
261 goto out_eof;
262 if (diff >= array->size) {
263 if (array->eof_index >= 0)
264 goto out_eof;
265 return -EAGAIN;
266 }
267
268 index = (unsigned int)diff;
269 *desc->dir_cookie = array->array[index].cookie;
270 desc->cache_entry_index = index;
271 return 0;
272 out_eof:
273 desc->eof = 1;
274 return -EBADCOOKIE;
275 }
276
277 static
278 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
279 {
280 int i;
281 loff_t new_pos;
282 int status = -EAGAIN;
283
284 for (i = 0; i < array->size; i++) {
285 if (array->array[i].cookie == *desc->dir_cookie) {
286 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
287 struct nfs_open_dir_context *ctx = desc->file->private_data;
288
289 new_pos = desc->current_index + i;
290 if (ctx->attr_gencount != nfsi->attr_gencount
291 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
292 || test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) {
293 ctx->duped = 0;
294 ctx->attr_gencount = nfsi->attr_gencount;
295 } else if (new_pos < desc->ctx->pos) {
296 if (ctx->duped > 0
297 && ctx->dup_cookie == *desc->dir_cookie) {
298 if (printk_ratelimit()) {
299 pr_notice("NFS: directory %pD2 contains a readdir loop."
300 "Please contact your server vendor. "
301 "The file: %s has duplicate cookie %llu\n",
302 desc->file,
303 array->array[i].string.name,
304 *desc->dir_cookie);
305 }
306 status = -ELOOP;
307 goto out;
308 }
309 ctx->dup_cookie = *desc->dir_cookie;
310 ctx->duped = -1;
311 }
312 desc->ctx->pos = new_pos;
313 desc->cache_entry_index = i;
314 return 0;
315 }
316 }
317 if (array->eof_index >= 0) {
318 status = -EBADCOOKIE;
319 if (*desc->dir_cookie == array->last_cookie)
320 desc->eof = 1;
321 }
322 out:
323 return status;
324 }
325
326 static
327 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
328 {
329 struct nfs_cache_array *array;
330 int status;
331
332 array = nfs_readdir_get_array(desc->page);
333 if (IS_ERR(array)) {
334 status = PTR_ERR(array);
335 goto out;
336 }
337
338 if (*desc->dir_cookie == 0)
339 status = nfs_readdir_search_for_pos(array, desc);
340 else
341 status = nfs_readdir_search_for_cookie(array, desc);
342
343 if (status == -EAGAIN) {
344 desc->last_cookie = array->last_cookie;
345 desc->current_index += array->size;
346 desc->page_index++;
347 }
348 nfs_readdir_release_array(desc->page);
349 out:
350 return status;
351 }
352
353 /* Fill a page with xdr information before transferring to the cache page */
354 static
355 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
356 struct nfs_entry *entry, struct file *file, struct inode *inode)
357 {
358 struct nfs_open_dir_context *ctx = file->private_data;
359 struct rpc_cred *cred = ctx->cred;
360 unsigned long timestamp, gencount;
361 int error;
362
363 again:
364 timestamp = jiffies;
365 gencount = nfs_inc_attr_generation_counter();
366 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
367 NFS_SERVER(inode)->dtsize, desc->plus);
368 if (error < 0) {
369 /* We requested READDIRPLUS, but the server doesn't grok it */
370 if (error == -ENOTSUPP && desc->plus) {
371 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
372 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
373 desc->plus = 0;
374 goto again;
375 }
376 goto error;
377 }
378 desc->timestamp = timestamp;
379 desc->gencount = gencount;
380 error:
381 return error;
382 }
383
384 static int xdr_decode(nfs_readdir_descriptor_t *desc,
385 struct nfs_entry *entry, struct xdr_stream *xdr)
386 {
387 int error;
388
389 error = desc->decode(xdr, entry, desc->plus);
390 if (error)
391 return error;
392 entry->fattr->time_start = desc->timestamp;
393 entry->fattr->gencount = desc->gencount;
394 return 0;
395 }
396
397 static
398 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
399 {
400 if (dentry->d_inode == NULL)
401 goto different;
402 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
403 goto different;
404 return 1;
405 different:
406 return 0;
407 }
408
409 static
410 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
411 {
412 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
413 return false;
414 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
415 return true;
416 if (ctx->pos == 0)
417 return true;
418 return false;
419 }
420
421 /*
422 * This function is called by the lookup code to request the use of
423 * readdirplus to accelerate any future lookups in the same
424 * directory.
425 */
426 static
427 void nfs_advise_use_readdirplus(struct inode *dir)
428 {
429 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
430 }
431
432 static
433 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
434 {
435 struct qstr filename = QSTR_INIT(entry->name, entry->len);
436 struct dentry *dentry;
437 struct dentry *alias;
438 struct inode *dir = parent->d_inode;
439 struct inode *inode;
440 int status;
441
442 if (filename.name[0] == '.') {
443 if (filename.len == 1)
444 return;
445 if (filename.len == 2 && filename.name[1] == '.')
446 return;
447 }
448 filename.hash = full_name_hash(filename.name, filename.len);
449
450 dentry = d_lookup(parent, &filename);
451 if (dentry != NULL) {
452 if (nfs_same_file(dentry, entry)) {
453 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
454 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
455 if (!status)
456 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
457 goto out;
458 } else {
459 if (d_invalidate(dentry) != 0)
460 goto out;
461 dput(dentry);
462 }
463 }
464
465 dentry = d_alloc(parent, &filename);
466 if (dentry == NULL)
467 return;
468
469 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
470 if (IS_ERR(inode))
471 goto out;
472
473 alias = d_materialise_unique(dentry, inode);
474 if (IS_ERR(alias))
475 goto out;
476 else if (alias) {
477 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
478 dput(alias);
479 } else
480 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
481
482 out:
483 dput(dentry);
484 }
485
486 /* Perform conversion from xdr to cache array */
487 static
488 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
489 struct page **xdr_pages, struct page *page, unsigned int buflen)
490 {
491 struct xdr_stream stream;
492 struct xdr_buf buf;
493 struct page *scratch;
494 struct nfs_cache_array *array;
495 unsigned int count = 0;
496 int status;
497
498 scratch = alloc_page(GFP_KERNEL);
499 if (scratch == NULL)
500 return -ENOMEM;
501
502 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
503 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
504
505 do {
506 status = xdr_decode(desc, entry, &stream);
507 if (status != 0) {
508 if (status == -EAGAIN)
509 status = 0;
510 break;
511 }
512
513 count++;
514
515 if (desc->plus != 0)
516 nfs_prime_dcache(desc->file->f_path.dentry, entry);
517
518 status = nfs_readdir_add_to_array(entry, page);
519 if (status != 0)
520 break;
521 } while (!entry->eof);
522
523 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
524 array = nfs_readdir_get_array(page);
525 if (!IS_ERR(array)) {
526 array->eof_index = array->size;
527 status = 0;
528 nfs_readdir_release_array(page);
529 } else
530 status = PTR_ERR(array);
531 }
532
533 put_page(scratch);
534 return status;
535 }
536
537 static
538 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
539 {
540 unsigned int i;
541 for (i = 0; i < npages; i++)
542 put_page(pages[i]);
543 }
544
545 static
546 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
547 unsigned int npages)
548 {
549 nfs_readdir_free_pagearray(pages, npages);
550 }
551
552 /*
553 * nfs_readdir_large_page will allocate pages that must be freed with a call
554 * to nfs_readdir_free_large_page
555 */
556 static
557 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
558 {
559 unsigned int i;
560
561 for (i = 0; i < npages; i++) {
562 struct page *page = alloc_page(GFP_KERNEL);
563 if (page == NULL)
564 goto out_freepages;
565 pages[i] = page;
566 }
567 return 0;
568
569 out_freepages:
570 nfs_readdir_free_pagearray(pages, i);
571 return -ENOMEM;
572 }
573
574 static
575 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
576 {
577 struct page *pages[NFS_MAX_READDIR_PAGES];
578 void *pages_ptr = NULL;
579 struct nfs_entry entry;
580 struct file *file = desc->file;
581 struct nfs_cache_array *array;
582 int status = -ENOMEM;
583 unsigned int array_size = ARRAY_SIZE(pages);
584
585 entry.prev_cookie = 0;
586 entry.cookie = desc->last_cookie;
587 entry.eof = 0;
588 entry.fh = nfs_alloc_fhandle();
589 entry.fattr = nfs_alloc_fattr();
590 entry.server = NFS_SERVER(inode);
591 if (entry.fh == NULL || entry.fattr == NULL)
592 goto out;
593
594 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
595 if (IS_ERR(entry.label)) {
596 status = PTR_ERR(entry.label);
597 goto out;
598 }
599
600 array = nfs_readdir_get_array(page);
601 if (IS_ERR(array)) {
602 status = PTR_ERR(array);
603 goto out_label_free;
604 }
605 memset(array, 0, sizeof(struct nfs_cache_array));
606 array->eof_index = -1;
607
608 status = nfs_readdir_large_page(pages, array_size);
609 if (status < 0)
610 goto out_release_array;
611 do {
612 unsigned int pglen;
613 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
614
615 if (status < 0)
616 break;
617 pglen = status;
618 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
619 if (status < 0) {
620 if (status == -ENOSPC)
621 status = 0;
622 break;
623 }
624 } while (array->eof_index < 0);
625
626 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
627 out_release_array:
628 nfs_readdir_release_array(page);
629 out_label_free:
630 nfs4_label_free(entry.label);
631 out:
632 nfs_free_fattr(entry.fattr);
633 nfs_free_fhandle(entry.fh);
634 return status;
635 }
636
637 /*
638 * Now we cache directories properly, by converting xdr information
639 * to an array that can be used for lookups later. This results in
640 * fewer cache pages, since we can store more information on each page.
641 * We only need to convert from xdr once so future lookups are much simpler
642 */
643 static
644 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
645 {
646 struct inode *inode = file_inode(desc->file);
647 int ret;
648
649 ret = nfs_readdir_xdr_to_array(desc, page, inode);
650 if (ret < 0)
651 goto error;
652 SetPageUptodate(page);
653
654 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
655 /* Should never happen */
656 nfs_zap_mapping(inode, inode->i_mapping);
657 }
658 unlock_page(page);
659 return 0;
660 error:
661 unlock_page(page);
662 return ret;
663 }
664
665 static
666 void cache_page_release(nfs_readdir_descriptor_t *desc)
667 {
668 if (!desc->page->mapping)
669 nfs_readdir_clear_array(desc->page);
670 page_cache_release(desc->page);
671 desc->page = NULL;
672 }
673
674 static
675 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
676 {
677 return read_cache_page(file_inode(desc->file)->i_mapping,
678 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
679 }
680
681 /*
682 * Returns 0 if desc->dir_cookie was found on page desc->page_index
683 */
684 static
685 int find_cache_page(nfs_readdir_descriptor_t *desc)
686 {
687 int res;
688
689 desc->page = get_cache_page(desc);
690 if (IS_ERR(desc->page))
691 return PTR_ERR(desc->page);
692
693 res = nfs_readdir_search_array(desc);
694 if (res != 0)
695 cache_page_release(desc);
696 return res;
697 }
698
699 /* Search for desc->dir_cookie from the beginning of the page cache */
700 static inline
701 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
702 {
703 int res;
704
705 if (desc->page_index == 0) {
706 desc->current_index = 0;
707 desc->last_cookie = 0;
708 }
709 do {
710 res = find_cache_page(desc);
711 } while (res == -EAGAIN);
712 return res;
713 }
714
715 /*
716 * Once we've found the start of the dirent within a page: fill 'er up...
717 */
718 static
719 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
720 {
721 struct file *file = desc->file;
722 int i = 0;
723 int res = 0;
724 struct nfs_cache_array *array = NULL;
725 struct nfs_open_dir_context *ctx = file->private_data;
726
727 array = nfs_readdir_get_array(desc->page);
728 if (IS_ERR(array)) {
729 res = PTR_ERR(array);
730 goto out;
731 }
732
733 for (i = desc->cache_entry_index; i < array->size; i++) {
734 struct nfs_cache_array_entry *ent;
735
736 ent = &array->array[i];
737 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
738 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
739 desc->eof = 1;
740 break;
741 }
742 desc->ctx->pos++;
743 if (i < (array->size-1))
744 *desc->dir_cookie = array->array[i+1].cookie;
745 else
746 *desc->dir_cookie = array->last_cookie;
747 if (ctx->duped != 0)
748 ctx->duped = 1;
749 }
750 if (array->eof_index >= 0)
751 desc->eof = 1;
752
753 nfs_readdir_release_array(desc->page);
754 out:
755 cache_page_release(desc);
756 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
757 (unsigned long long)*desc->dir_cookie, res);
758 return res;
759 }
760
761 /*
762 * If we cannot find a cookie in our cache, we suspect that this is
763 * because it points to a deleted file, so we ask the server to return
764 * whatever it thinks is the next entry. We then feed this to filldir.
765 * If all goes well, we should then be able to find our way round the
766 * cache on the next call to readdir_search_pagecache();
767 *
768 * NOTE: we cannot add the anonymous page to the pagecache because
769 * the data it contains might not be page aligned. Besides,
770 * we should already have a complete representation of the
771 * directory in the page cache by the time we get here.
772 */
773 static inline
774 int uncached_readdir(nfs_readdir_descriptor_t *desc)
775 {
776 struct page *page = NULL;
777 int status;
778 struct inode *inode = file_inode(desc->file);
779 struct nfs_open_dir_context *ctx = desc->file->private_data;
780
781 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
782 (unsigned long long)*desc->dir_cookie);
783
784 page = alloc_page(GFP_HIGHUSER);
785 if (!page) {
786 status = -ENOMEM;
787 goto out;
788 }
789
790 desc->page_index = 0;
791 desc->last_cookie = *desc->dir_cookie;
792 desc->page = page;
793 ctx->duped = 0;
794
795 status = nfs_readdir_xdr_to_array(desc, page, inode);
796 if (status < 0)
797 goto out_release;
798
799 status = nfs_do_filldir(desc);
800
801 out:
802 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
803 __func__, status);
804 return status;
805 out_release:
806 cache_page_release(desc);
807 goto out;
808 }
809
810 /* The file offset position represents the dirent entry number. A
811 last cookie cache takes care of the common case of reading the
812 whole directory.
813 */
814 static int nfs_readdir(struct file *file, struct dir_context *ctx)
815 {
816 struct dentry *dentry = file->f_path.dentry;
817 struct inode *inode = dentry->d_inode;
818 nfs_readdir_descriptor_t my_desc,
819 *desc = &my_desc;
820 struct nfs_open_dir_context *dir_ctx = file->private_data;
821 int res = 0;
822
823 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
824 file, (long long)ctx->pos);
825 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
826
827 /*
828 * ctx->pos points to the dirent entry number.
829 * *desc->dir_cookie has the cookie for the next entry. We have
830 * to either find the entry with the appropriate number or
831 * revalidate the cookie.
832 */
833 memset(desc, 0, sizeof(*desc));
834
835 desc->file = file;
836 desc->ctx = ctx;
837 desc->dir_cookie = &dir_ctx->dir_cookie;
838 desc->decode = NFS_PROTO(inode)->decode_dirent;
839 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
840
841 nfs_block_sillyrename(dentry);
842 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
843 res = nfs_revalidate_mapping(inode, file->f_mapping);
844 if (res < 0)
845 goto out;
846
847 do {
848 res = readdir_search_pagecache(desc);
849
850 if (res == -EBADCOOKIE) {
851 res = 0;
852 /* This means either end of directory */
853 if (*desc->dir_cookie && desc->eof == 0) {
854 /* Or that the server has 'lost' a cookie */
855 res = uncached_readdir(desc);
856 if (res == 0)
857 continue;
858 }
859 break;
860 }
861 if (res == -ETOOSMALL && desc->plus) {
862 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
863 nfs_zap_caches(inode);
864 desc->page_index = 0;
865 desc->plus = 0;
866 desc->eof = 0;
867 continue;
868 }
869 if (res < 0)
870 break;
871
872 res = nfs_do_filldir(desc);
873 if (res < 0)
874 break;
875 } while (!desc->eof);
876 out:
877 nfs_unblock_sillyrename(dentry);
878 if (res > 0)
879 res = 0;
880 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
881 return res;
882 }
883
884 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
885 {
886 struct inode *inode = file_inode(filp);
887 struct nfs_open_dir_context *dir_ctx = filp->private_data;
888
889 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
890 filp, offset, whence);
891
892 mutex_lock(&inode->i_mutex);
893 switch (whence) {
894 case 1:
895 offset += filp->f_pos;
896 case 0:
897 if (offset >= 0)
898 break;
899 default:
900 offset = -EINVAL;
901 goto out;
902 }
903 if (offset != filp->f_pos) {
904 filp->f_pos = offset;
905 dir_ctx->dir_cookie = 0;
906 dir_ctx->duped = 0;
907 }
908 out:
909 mutex_unlock(&inode->i_mutex);
910 return offset;
911 }
912
913 /*
914 * All directory operations under NFS are synchronous, so fsync()
915 * is a dummy operation.
916 */
917 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
918 int datasync)
919 {
920 struct inode *inode = file_inode(filp);
921
922 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
923
924 mutex_lock(&inode->i_mutex);
925 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
926 mutex_unlock(&inode->i_mutex);
927 return 0;
928 }
929
930 /**
931 * nfs_force_lookup_revalidate - Mark the directory as having changed
932 * @dir - pointer to directory inode
933 *
934 * This forces the revalidation code in nfs_lookup_revalidate() to do a
935 * full lookup on all child dentries of 'dir' whenever a change occurs
936 * on the server that might have invalidated our dcache.
937 *
938 * The caller should be holding dir->i_lock
939 */
940 void nfs_force_lookup_revalidate(struct inode *dir)
941 {
942 NFS_I(dir)->cache_change_attribute++;
943 }
944 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
945
946 /*
947 * A check for whether or not the parent directory has changed.
948 * In the case it has, we assume that the dentries are untrustworthy
949 * and may need to be looked up again.
950 */
951 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
952 {
953 if (IS_ROOT(dentry))
954 return 1;
955 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
956 return 0;
957 if (!nfs_verify_change_attribute(dir, dentry->d_time))
958 return 0;
959 /* Revalidate nfsi->cache_change_attribute before we declare a match */
960 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
961 return 0;
962 if (!nfs_verify_change_attribute(dir, dentry->d_time))
963 return 0;
964 return 1;
965 }
966
967 /*
968 * Use intent information to check whether or not we're going to do
969 * an O_EXCL create using this path component.
970 */
971 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
972 {
973 if (NFS_PROTO(dir)->version == 2)
974 return 0;
975 return flags & LOOKUP_EXCL;
976 }
977
978 /*
979 * Inode and filehandle revalidation for lookups.
980 *
981 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
982 * or if the intent information indicates that we're about to open this
983 * particular file and the "nocto" mount flag is not set.
984 *
985 */
986 static
987 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
988 {
989 struct nfs_server *server = NFS_SERVER(inode);
990 int ret;
991
992 if (IS_AUTOMOUNT(inode))
993 return 0;
994 /* VFS wants an on-the-wire revalidation */
995 if (flags & LOOKUP_REVAL)
996 goto out_force;
997 /* This is an open(2) */
998 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
999 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1000 goto out_force;
1001 out:
1002 return (inode->i_nlink == 0) ? -ENOENT : 0;
1003 out_force:
1004 ret = __nfs_revalidate_inode(server, inode);
1005 if (ret != 0)
1006 return ret;
1007 goto out;
1008 }
1009
1010 /*
1011 * We judge how long we want to trust negative
1012 * dentries by looking at the parent inode mtime.
1013 *
1014 * If parent mtime has changed, we revalidate, else we wait for a
1015 * period corresponding to the parent's attribute cache timeout value.
1016 */
1017 static inline
1018 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1019 unsigned int flags)
1020 {
1021 /* Don't revalidate a negative dentry if we're creating a new file */
1022 if (flags & LOOKUP_CREATE)
1023 return 0;
1024 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1025 return 1;
1026 return !nfs_check_verifier(dir, dentry);
1027 }
1028
1029 /*
1030 * This is called every time the dcache has a lookup hit,
1031 * and we should check whether we can really trust that
1032 * lookup.
1033 *
1034 * NOTE! The hit can be a negative hit too, don't assume
1035 * we have an inode!
1036 *
1037 * If the parent directory is seen to have changed, we throw out the
1038 * cached dentry and do a new lookup.
1039 */
1040 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1041 {
1042 struct inode *dir;
1043 struct inode *inode;
1044 struct dentry *parent;
1045 struct nfs_fh *fhandle = NULL;
1046 struct nfs_fattr *fattr = NULL;
1047 struct nfs4_label *label = NULL;
1048 int error;
1049
1050 if (flags & LOOKUP_RCU)
1051 return -ECHILD;
1052
1053 parent = dget_parent(dentry);
1054 dir = parent->d_inode;
1055 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1056 inode = dentry->d_inode;
1057
1058 if (!inode) {
1059 if (nfs_neg_need_reval(dir, dentry, flags))
1060 goto out_bad;
1061 goto out_valid_noent;
1062 }
1063
1064 if (is_bad_inode(inode)) {
1065 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1066 __func__, dentry);
1067 goto out_bad;
1068 }
1069
1070 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1071 goto out_set_verifier;
1072
1073 /* Force a full look up iff the parent directory has changed */
1074 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1075 if (nfs_lookup_verify_inode(inode, flags))
1076 goto out_zap_parent;
1077 goto out_valid;
1078 }
1079
1080 if (NFS_STALE(inode))
1081 goto out_bad;
1082
1083 error = -ENOMEM;
1084 fhandle = nfs_alloc_fhandle();
1085 fattr = nfs_alloc_fattr();
1086 if (fhandle == NULL || fattr == NULL)
1087 goto out_error;
1088
1089 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1090 if (IS_ERR(label))
1091 goto out_error;
1092
1093 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1094 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1095 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1096 if (error)
1097 goto out_bad;
1098 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1099 goto out_bad;
1100 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1101 goto out_bad;
1102
1103 nfs_setsecurity(inode, fattr, label);
1104
1105 nfs_free_fattr(fattr);
1106 nfs_free_fhandle(fhandle);
1107 nfs4_label_free(label);
1108
1109 out_set_verifier:
1110 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1111 out_valid:
1112 /* Success: notify readdir to use READDIRPLUS */
1113 nfs_advise_use_readdirplus(dir);
1114 out_valid_noent:
1115 dput(parent);
1116 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1117 __func__, dentry);
1118 return 1;
1119 out_zap_parent:
1120 nfs_zap_caches(dir);
1121 out_bad:
1122 nfs_free_fattr(fattr);
1123 nfs_free_fhandle(fhandle);
1124 nfs4_label_free(label);
1125 nfs_mark_for_revalidate(dir);
1126 if (inode && S_ISDIR(inode->i_mode)) {
1127 /* Purge readdir caches. */
1128 nfs_zap_caches(inode);
1129 /*
1130 * We can't d_drop the root of a disconnected tree:
1131 * its d_hash is on the s_anon list and d_drop() would hide
1132 * it from shrink_dcache_for_unmount(), leading to busy
1133 * inodes on unmount and further oopses.
1134 */
1135 if (IS_ROOT(dentry))
1136 goto out_valid;
1137 }
1138 /* If we have submounts, don't unhash ! */
1139 if (check_submounts_and_drop(dentry) != 0)
1140 goto out_valid;
1141
1142 dput(parent);
1143 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1144 __func__, dentry);
1145 return 0;
1146 out_error:
1147 nfs_free_fattr(fattr);
1148 nfs_free_fhandle(fhandle);
1149 nfs4_label_free(label);
1150 dput(parent);
1151 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1152 __func__, dentry, error);
1153 return error;
1154 }
1155
1156 /*
1157 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1158 * when we don't really care about the dentry name. This is called when a
1159 * pathwalk ends on a dentry that was not found via a normal lookup in the
1160 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1161 *
1162 * In this situation, we just want to verify that the inode itself is OK
1163 * since the dentry might have changed on the server.
1164 */
1165 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1166 {
1167 int error;
1168 struct inode *inode = dentry->d_inode;
1169
1170 /*
1171 * I believe we can only get a negative dentry here in the case of a
1172 * procfs-style symlink. Just assume it's correct for now, but we may
1173 * eventually need to do something more here.
1174 */
1175 if (!inode) {
1176 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1177 __func__, dentry);
1178 return 1;
1179 }
1180
1181 if (is_bad_inode(inode)) {
1182 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1183 __func__, dentry);
1184 return 0;
1185 }
1186
1187 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1188 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1189 __func__, inode->i_ino, error ? "invalid" : "valid");
1190 return !error;
1191 }
1192
1193 /*
1194 * This is called from dput() when d_count is going to 0.
1195 */
1196 static int nfs_dentry_delete(const struct dentry *dentry)
1197 {
1198 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1199 dentry, dentry->d_flags);
1200
1201 /* Unhash any dentry with a stale inode */
1202 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1203 return 1;
1204
1205 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1206 /* Unhash it, so that ->d_iput() would be called */
1207 return 1;
1208 }
1209 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1210 /* Unhash it, so that ancestors of killed async unlink
1211 * files will be cleaned up during umount */
1212 return 1;
1213 }
1214 return 0;
1215
1216 }
1217
1218 /* Ensure that we revalidate inode->i_nlink */
1219 static void nfs_drop_nlink(struct inode *inode)
1220 {
1221 spin_lock(&inode->i_lock);
1222 /* drop the inode if we're reasonably sure this is the last link */
1223 if (inode->i_nlink == 1)
1224 clear_nlink(inode);
1225 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1226 spin_unlock(&inode->i_lock);
1227 }
1228
1229 /*
1230 * Called when the dentry loses inode.
1231 * We use it to clean up silly-renamed files.
1232 */
1233 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1234 {
1235 if (S_ISDIR(inode->i_mode))
1236 /* drop any readdir cache as it could easily be old */
1237 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1238
1239 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1240 nfs_complete_unlink(dentry, inode);
1241 nfs_drop_nlink(inode);
1242 }
1243 iput(inode);
1244 }
1245
1246 static void nfs_d_release(struct dentry *dentry)
1247 {
1248 /* free cached devname value, if it survived that far */
1249 if (unlikely(dentry->d_fsdata)) {
1250 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1251 WARN_ON(1);
1252 else
1253 kfree(dentry->d_fsdata);
1254 }
1255 }
1256
1257 const struct dentry_operations nfs_dentry_operations = {
1258 .d_revalidate = nfs_lookup_revalidate,
1259 .d_weak_revalidate = nfs_weak_revalidate,
1260 .d_delete = nfs_dentry_delete,
1261 .d_iput = nfs_dentry_iput,
1262 .d_automount = nfs_d_automount,
1263 .d_release = nfs_d_release,
1264 };
1265 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1266
1267 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1268 {
1269 struct dentry *res;
1270 struct dentry *parent;
1271 struct inode *inode = NULL;
1272 struct nfs_fh *fhandle = NULL;
1273 struct nfs_fattr *fattr = NULL;
1274 struct nfs4_label *label = NULL;
1275 int error;
1276
1277 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1278 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1279
1280 res = ERR_PTR(-ENAMETOOLONG);
1281 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1282 goto out;
1283
1284 /*
1285 * If we're doing an exclusive create, optimize away the lookup
1286 * but don't hash the dentry.
1287 */
1288 if (nfs_is_exclusive_create(dir, flags)) {
1289 d_instantiate(dentry, NULL);
1290 res = NULL;
1291 goto out;
1292 }
1293
1294 res = ERR_PTR(-ENOMEM);
1295 fhandle = nfs_alloc_fhandle();
1296 fattr = nfs_alloc_fattr();
1297 if (fhandle == NULL || fattr == NULL)
1298 goto out;
1299
1300 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1301 if (IS_ERR(label))
1302 goto out;
1303
1304 parent = dentry->d_parent;
1305 /* Protect against concurrent sillydeletes */
1306 trace_nfs_lookup_enter(dir, dentry, flags);
1307 nfs_block_sillyrename(parent);
1308 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1309 if (error == -ENOENT)
1310 goto no_entry;
1311 if (error < 0) {
1312 res = ERR_PTR(error);
1313 goto out_unblock_sillyrename;
1314 }
1315 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1316 res = ERR_CAST(inode);
1317 if (IS_ERR(res))
1318 goto out_unblock_sillyrename;
1319
1320 /* Success: notify readdir to use READDIRPLUS */
1321 nfs_advise_use_readdirplus(dir);
1322
1323 no_entry:
1324 res = d_materialise_unique(dentry, inode);
1325 if (res != NULL) {
1326 if (IS_ERR(res))
1327 goto out_unblock_sillyrename;
1328 dentry = res;
1329 }
1330 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1331 out_unblock_sillyrename:
1332 nfs_unblock_sillyrename(parent);
1333 trace_nfs_lookup_exit(dir, dentry, flags, error);
1334 nfs4_label_free(label);
1335 out:
1336 nfs_free_fattr(fattr);
1337 nfs_free_fhandle(fhandle);
1338 return res;
1339 }
1340 EXPORT_SYMBOL_GPL(nfs_lookup);
1341
1342 #if IS_ENABLED(CONFIG_NFS_V4)
1343 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1344
1345 const struct dentry_operations nfs4_dentry_operations = {
1346 .d_revalidate = nfs4_lookup_revalidate,
1347 .d_delete = nfs_dentry_delete,
1348 .d_iput = nfs_dentry_iput,
1349 .d_automount = nfs_d_automount,
1350 .d_release = nfs_d_release,
1351 };
1352 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1353
1354 static fmode_t flags_to_mode(int flags)
1355 {
1356 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1357 if ((flags & O_ACCMODE) != O_WRONLY)
1358 res |= FMODE_READ;
1359 if ((flags & O_ACCMODE) != O_RDONLY)
1360 res |= FMODE_WRITE;
1361 return res;
1362 }
1363
1364 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1365 {
1366 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1367 }
1368
1369 static int do_open(struct inode *inode, struct file *filp)
1370 {
1371 nfs_fscache_open_file(inode, filp);
1372 return 0;
1373 }
1374
1375 static int nfs_finish_open(struct nfs_open_context *ctx,
1376 struct dentry *dentry,
1377 struct file *file, unsigned open_flags,
1378 int *opened)
1379 {
1380 int err;
1381
1382 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1383 *opened |= FILE_CREATED;
1384
1385 err = finish_open(file, dentry, do_open, opened);
1386 if (err)
1387 goto out;
1388 nfs_file_set_open_context(file, ctx);
1389
1390 out:
1391 return err;
1392 }
1393
1394 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1395 struct file *file, unsigned open_flags,
1396 umode_t mode, int *opened)
1397 {
1398 struct nfs_open_context *ctx;
1399 struct dentry *res;
1400 struct iattr attr = { .ia_valid = ATTR_OPEN };
1401 struct inode *inode;
1402 unsigned int lookup_flags = 0;
1403 int err;
1404
1405 /* Expect a negative dentry */
1406 BUG_ON(dentry->d_inode);
1407
1408 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1409 dir->i_sb->s_id, dir->i_ino, dentry);
1410
1411 err = nfs_check_flags(open_flags);
1412 if (err)
1413 return err;
1414
1415 /* NFS only supports OPEN on regular files */
1416 if ((open_flags & O_DIRECTORY)) {
1417 if (!d_unhashed(dentry)) {
1418 /*
1419 * Hashed negative dentry with O_DIRECTORY: dentry was
1420 * revalidated and is fine, no need to perform lookup
1421 * again
1422 */
1423 return -ENOENT;
1424 }
1425 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1426 goto no_open;
1427 }
1428
1429 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1430 return -ENAMETOOLONG;
1431
1432 if (open_flags & O_CREAT) {
1433 attr.ia_valid |= ATTR_MODE;
1434 attr.ia_mode = mode & ~current_umask();
1435 }
1436 if (open_flags & O_TRUNC) {
1437 attr.ia_valid |= ATTR_SIZE;
1438 attr.ia_size = 0;
1439 }
1440
1441 ctx = create_nfs_open_context(dentry, open_flags);
1442 err = PTR_ERR(ctx);
1443 if (IS_ERR(ctx))
1444 goto out;
1445
1446 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1447 nfs_block_sillyrename(dentry->d_parent);
1448 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1449 nfs_unblock_sillyrename(dentry->d_parent);
1450 if (IS_ERR(inode)) {
1451 err = PTR_ERR(inode);
1452 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1453 put_nfs_open_context(ctx);
1454 switch (err) {
1455 case -ENOENT:
1456 d_drop(dentry);
1457 d_add(dentry, NULL);
1458 break;
1459 case -EISDIR:
1460 case -ENOTDIR:
1461 goto no_open;
1462 case -ELOOP:
1463 if (!(open_flags & O_NOFOLLOW))
1464 goto no_open;
1465 break;
1466 /* case -EINVAL: */
1467 default:
1468 break;
1469 }
1470 goto out;
1471 }
1472
1473 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1474 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1475 put_nfs_open_context(ctx);
1476 out:
1477 return err;
1478
1479 no_open:
1480 res = nfs_lookup(dir, dentry, lookup_flags);
1481 err = PTR_ERR(res);
1482 if (IS_ERR(res))
1483 goto out;
1484
1485 return finish_no_open(file, res);
1486 }
1487 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1488
1489 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1490 {
1491 struct dentry *parent = NULL;
1492 struct inode *inode;
1493 struct inode *dir;
1494 int ret = 0;
1495
1496 if (flags & LOOKUP_RCU)
1497 return -ECHILD;
1498
1499 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1500 goto no_open;
1501 if (d_mountpoint(dentry))
1502 goto no_open;
1503 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1504 goto no_open;
1505
1506 inode = dentry->d_inode;
1507 parent = dget_parent(dentry);
1508 dir = parent->d_inode;
1509
1510 /* We can't create new files in nfs_open_revalidate(), so we
1511 * optimize away revalidation of negative dentries.
1512 */
1513 if (inode == NULL) {
1514 if (!nfs_neg_need_reval(dir, dentry, flags))
1515 ret = 1;
1516 goto out;
1517 }
1518
1519 /* NFS only supports OPEN on regular files */
1520 if (!S_ISREG(inode->i_mode))
1521 goto no_open_dput;
1522 /* We cannot do exclusive creation on a positive dentry */
1523 if (flags & LOOKUP_EXCL)
1524 goto no_open_dput;
1525
1526 /* Let f_op->open() actually open (and revalidate) the file */
1527 ret = 1;
1528
1529 out:
1530 dput(parent);
1531 return ret;
1532
1533 no_open_dput:
1534 dput(parent);
1535 no_open:
1536 return nfs_lookup_revalidate(dentry, flags);
1537 }
1538
1539 #endif /* CONFIG_NFSV4 */
1540
1541 /*
1542 * Code common to create, mkdir, and mknod.
1543 */
1544 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1545 struct nfs_fattr *fattr,
1546 struct nfs4_label *label)
1547 {
1548 struct dentry *parent = dget_parent(dentry);
1549 struct inode *dir = parent->d_inode;
1550 struct inode *inode;
1551 int error = -EACCES;
1552
1553 d_drop(dentry);
1554
1555 /* We may have been initialized further down */
1556 if (dentry->d_inode)
1557 goto out;
1558 if (fhandle->size == 0) {
1559 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1560 if (error)
1561 goto out_error;
1562 }
1563 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1564 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1565 struct nfs_server *server = NFS_SB(dentry->d_sb);
1566 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1567 if (error < 0)
1568 goto out_error;
1569 }
1570 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1571 error = PTR_ERR(inode);
1572 if (IS_ERR(inode))
1573 goto out_error;
1574 d_add(dentry, inode);
1575 out:
1576 dput(parent);
1577 return 0;
1578 out_error:
1579 nfs_mark_for_revalidate(dir);
1580 dput(parent);
1581 return error;
1582 }
1583 EXPORT_SYMBOL_GPL(nfs_instantiate);
1584
1585 /*
1586 * Following a failed create operation, we drop the dentry rather
1587 * than retain a negative dentry. This avoids a problem in the event
1588 * that the operation succeeded on the server, but an error in the
1589 * reply path made it appear to have failed.
1590 */
1591 int nfs_create(struct inode *dir, struct dentry *dentry,
1592 umode_t mode, bool excl)
1593 {
1594 struct iattr attr;
1595 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1596 int error;
1597
1598 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1599 dir->i_sb->s_id, dir->i_ino, dentry);
1600
1601 attr.ia_mode = mode;
1602 attr.ia_valid = ATTR_MODE;
1603
1604 trace_nfs_create_enter(dir, dentry, open_flags);
1605 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1606 trace_nfs_create_exit(dir, dentry, open_flags, error);
1607 if (error != 0)
1608 goto out_err;
1609 return 0;
1610 out_err:
1611 d_drop(dentry);
1612 return error;
1613 }
1614 EXPORT_SYMBOL_GPL(nfs_create);
1615
1616 /*
1617 * See comments for nfs_proc_create regarding failed operations.
1618 */
1619 int
1620 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1621 {
1622 struct iattr attr;
1623 int status;
1624
1625 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1626 dir->i_sb->s_id, dir->i_ino, dentry);
1627
1628 if (!new_valid_dev(rdev))
1629 return -EINVAL;
1630
1631 attr.ia_mode = mode;
1632 attr.ia_valid = ATTR_MODE;
1633
1634 trace_nfs_mknod_enter(dir, dentry);
1635 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1636 trace_nfs_mknod_exit(dir, dentry, status);
1637 if (status != 0)
1638 goto out_err;
1639 return 0;
1640 out_err:
1641 d_drop(dentry);
1642 return status;
1643 }
1644 EXPORT_SYMBOL_GPL(nfs_mknod);
1645
1646 /*
1647 * See comments for nfs_proc_create regarding failed operations.
1648 */
1649 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1650 {
1651 struct iattr attr;
1652 int error;
1653
1654 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1655 dir->i_sb->s_id, dir->i_ino, dentry);
1656
1657 attr.ia_valid = ATTR_MODE;
1658 attr.ia_mode = mode | S_IFDIR;
1659
1660 trace_nfs_mkdir_enter(dir, dentry);
1661 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1662 trace_nfs_mkdir_exit(dir, dentry, error);
1663 if (error != 0)
1664 goto out_err;
1665 return 0;
1666 out_err:
1667 d_drop(dentry);
1668 return error;
1669 }
1670 EXPORT_SYMBOL_GPL(nfs_mkdir);
1671
1672 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1673 {
1674 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1675 d_delete(dentry);
1676 }
1677
1678 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1679 {
1680 int error;
1681
1682 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1683 dir->i_sb->s_id, dir->i_ino, dentry);
1684
1685 trace_nfs_rmdir_enter(dir, dentry);
1686 if (dentry->d_inode) {
1687 nfs_wait_on_sillyrename(dentry);
1688 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1689 /* Ensure the VFS deletes this inode */
1690 switch (error) {
1691 case 0:
1692 clear_nlink(dentry->d_inode);
1693 break;
1694 case -ENOENT:
1695 nfs_dentry_handle_enoent(dentry);
1696 }
1697 } else
1698 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1699 trace_nfs_rmdir_exit(dir, dentry, error);
1700
1701 return error;
1702 }
1703 EXPORT_SYMBOL_GPL(nfs_rmdir);
1704
1705 /*
1706 * Remove a file after making sure there are no pending writes,
1707 * and after checking that the file has only one user.
1708 *
1709 * We invalidate the attribute cache and free the inode prior to the operation
1710 * to avoid possible races if the server reuses the inode.
1711 */
1712 static int nfs_safe_remove(struct dentry *dentry)
1713 {
1714 struct inode *dir = dentry->d_parent->d_inode;
1715 struct inode *inode = dentry->d_inode;
1716 int error = -EBUSY;
1717
1718 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1719
1720 /* If the dentry was sillyrenamed, we simply call d_delete() */
1721 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1722 error = 0;
1723 goto out;
1724 }
1725
1726 trace_nfs_remove_enter(dir, dentry);
1727 if (inode != NULL) {
1728 NFS_PROTO(inode)->return_delegation(inode);
1729 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1730 if (error == 0)
1731 nfs_drop_nlink(inode);
1732 } else
1733 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1734 if (error == -ENOENT)
1735 nfs_dentry_handle_enoent(dentry);
1736 trace_nfs_remove_exit(dir, dentry, error);
1737 out:
1738 return error;
1739 }
1740
1741 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1742 * belongs to an active ".nfs..." file and we return -EBUSY.
1743 *
1744 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1745 */
1746 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1747 {
1748 int error;
1749 int need_rehash = 0;
1750
1751 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1752 dir->i_ino, dentry);
1753
1754 trace_nfs_unlink_enter(dir, dentry);
1755 spin_lock(&dentry->d_lock);
1756 if (d_count(dentry) > 1) {
1757 spin_unlock(&dentry->d_lock);
1758 /* Start asynchronous writeout of the inode */
1759 write_inode_now(dentry->d_inode, 0);
1760 error = nfs_sillyrename(dir, dentry);
1761 goto out;
1762 }
1763 if (!d_unhashed(dentry)) {
1764 __d_drop(dentry);
1765 need_rehash = 1;
1766 }
1767 spin_unlock(&dentry->d_lock);
1768 error = nfs_safe_remove(dentry);
1769 if (!error || error == -ENOENT) {
1770 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1771 } else if (need_rehash)
1772 d_rehash(dentry);
1773 out:
1774 trace_nfs_unlink_exit(dir, dentry, error);
1775 return error;
1776 }
1777 EXPORT_SYMBOL_GPL(nfs_unlink);
1778
1779 /*
1780 * To create a symbolic link, most file systems instantiate a new inode,
1781 * add a page to it containing the path, then write it out to the disk
1782 * using prepare_write/commit_write.
1783 *
1784 * Unfortunately the NFS client can't create the in-core inode first
1785 * because it needs a file handle to create an in-core inode (see
1786 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1787 * symlink request has completed on the server.
1788 *
1789 * So instead we allocate a raw page, copy the symname into it, then do
1790 * the SYMLINK request with the page as the buffer. If it succeeds, we
1791 * now have a new file handle and can instantiate an in-core NFS inode
1792 * and move the raw page into its mapping.
1793 */
1794 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1795 {
1796 struct page *page;
1797 char *kaddr;
1798 struct iattr attr;
1799 unsigned int pathlen = strlen(symname);
1800 int error;
1801
1802 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1803 dir->i_ino, dentry, symname);
1804
1805 if (pathlen > PAGE_SIZE)
1806 return -ENAMETOOLONG;
1807
1808 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1809 attr.ia_valid = ATTR_MODE;
1810
1811 page = alloc_page(GFP_HIGHUSER);
1812 if (!page)
1813 return -ENOMEM;
1814
1815 kaddr = kmap_atomic(page);
1816 memcpy(kaddr, symname, pathlen);
1817 if (pathlen < PAGE_SIZE)
1818 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1819 kunmap_atomic(kaddr);
1820
1821 trace_nfs_symlink_enter(dir, dentry);
1822 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1823 trace_nfs_symlink_exit(dir, dentry, error);
1824 if (error != 0) {
1825 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1826 dir->i_sb->s_id, dir->i_ino,
1827 dentry, symname, error);
1828 d_drop(dentry);
1829 __free_page(page);
1830 return error;
1831 }
1832
1833 /*
1834 * No big deal if we can't add this page to the page cache here.
1835 * READLINK will get the missing page from the server if needed.
1836 */
1837 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1838 GFP_KERNEL)) {
1839 SetPageUptodate(page);
1840 unlock_page(page);
1841 } else
1842 __free_page(page);
1843
1844 return 0;
1845 }
1846 EXPORT_SYMBOL_GPL(nfs_symlink);
1847
1848 int
1849 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1850 {
1851 struct inode *inode = old_dentry->d_inode;
1852 int error;
1853
1854 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1855 old_dentry, dentry);
1856
1857 trace_nfs_link_enter(inode, dir, dentry);
1858 NFS_PROTO(inode)->return_delegation(inode);
1859
1860 d_drop(dentry);
1861 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1862 if (error == 0) {
1863 ihold(inode);
1864 d_add(dentry, inode);
1865 }
1866 trace_nfs_link_exit(inode, dir, dentry, error);
1867 return error;
1868 }
1869 EXPORT_SYMBOL_GPL(nfs_link);
1870
1871 /*
1872 * RENAME
1873 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1874 * different file handle for the same inode after a rename (e.g. when
1875 * moving to a different directory). A fail-safe method to do so would
1876 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1877 * rename the old file using the sillyrename stuff. This way, the original
1878 * file in old_dir will go away when the last process iput()s the inode.
1879 *
1880 * FIXED.
1881 *
1882 * It actually works quite well. One needs to have the possibility for
1883 * at least one ".nfs..." file in each directory the file ever gets
1884 * moved or linked to which happens automagically with the new
1885 * implementation that only depends on the dcache stuff instead of
1886 * using the inode layer
1887 *
1888 * Unfortunately, things are a little more complicated than indicated
1889 * above. For a cross-directory move, we want to make sure we can get
1890 * rid of the old inode after the operation. This means there must be
1891 * no pending writes (if it's a file), and the use count must be 1.
1892 * If these conditions are met, we can drop the dentries before doing
1893 * the rename.
1894 */
1895 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1896 struct inode *new_dir, struct dentry *new_dentry)
1897 {
1898 struct inode *old_inode = old_dentry->d_inode;
1899 struct inode *new_inode = new_dentry->d_inode;
1900 struct dentry *dentry = NULL, *rehash = NULL;
1901 int error = -EBUSY;
1902
1903 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1904 old_dentry, new_dentry,
1905 d_count(new_dentry));
1906
1907 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1908 /*
1909 * For non-directories, check whether the target is busy and if so,
1910 * make a copy of the dentry and then do a silly-rename. If the
1911 * silly-rename succeeds, the copied dentry is hashed and becomes
1912 * the new target.
1913 */
1914 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1915 /*
1916 * To prevent any new references to the target during the
1917 * rename, we unhash the dentry in advance.
1918 */
1919 if (!d_unhashed(new_dentry)) {
1920 d_drop(new_dentry);
1921 rehash = new_dentry;
1922 }
1923
1924 if (d_count(new_dentry) > 2) {
1925 int err;
1926
1927 /* copy the target dentry's name */
1928 dentry = d_alloc(new_dentry->d_parent,
1929 &new_dentry->d_name);
1930 if (!dentry)
1931 goto out;
1932
1933 /* silly-rename the existing target ... */
1934 err = nfs_sillyrename(new_dir, new_dentry);
1935 if (err)
1936 goto out;
1937
1938 new_dentry = dentry;
1939 rehash = NULL;
1940 new_inode = NULL;
1941 }
1942 }
1943
1944 NFS_PROTO(old_inode)->return_delegation(old_inode);
1945 if (new_inode != NULL)
1946 NFS_PROTO(new_inode)->return_delegation(new_inode);
1947
1948 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1949 new_dir, &new_dentry->d_name);
1950 nfs_mark_for_revalidate(old_inode);
1951 out:
1952 if (rehash)
1953 d_rehash(rehash);
1954 trace_nfs_rename_exit(old_dir, old_dentry,
1955 new_dir, new_dentry, error);
1956 if (!error) {
1957 if (new_inode != NULL)
1958 nfs_drop_nlink(new_inode);
1959 d_move(old_dentry, new_dentry);
1960 nfs_set_verifier(new_dentry,
1961 nfs_save_change_attribute(new_dir));
1962 } else if (error == -ENOENT)
1963 nfs_dentry_handle_enoent(old_dentry);
1964
1965 /* new dentry created? */
1966 if (dentry)
1967 dput(dentry);
1968 return error;
1969 }
1970 EXPORT_SYMBOL_GPL(nfs_rename);
1971
1972 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1973 static LIST_HEAD(nfs_access_lru_list);
1974 static atomic_long_t nfs_access_nr_entries;
1975
1976 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1977 {
1978 put_rpccred(entry->cred);
1979 kfree(entry);
1980 smp_mb__before_atomic_dec();
1981 atomic_long_dec(&nfs_access_nr_entries);
1982 smp_mb__after_atomic_dec();
1983 }
1984
1985 static void nfs_access_free_list(struct list_head *head)
1986 {
1987 struct nfs_access_entry *cache;
1988
1989 while (!list_empty(head)) {
1990 cache = list_entry(head->next, struct nfs_access_entry, lru);
1991 list_del(&cache->lru);
1992 nfs_access_free_entry(cache);
1993 }
1994 }
1995
1996 unsigned long
1997 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
1998 {
1999 LIST_HEAD(head);
2000 struct nfs_inode *nfsi, *next;
2001 struct nfs_access_entry *cache;
2002 int nr_to_scan = sc->nr_to_scan;
2003 gfp_t gfp_mask = sc->gfp_mask;
2004 long freed = 0;
2005
2006 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2007 return SHRINK_STOP;
2008
2009 spin_lock(&nfs_access_lru_lock);
2010 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2011 struct inode *inode;
2012
2013 if (nr_to_scan-- == 0)
2014 break;
2015 inode = &nfsi->vfs_inode;
2016 spin_lock(&inode->i_lock);
2017 if (list_empty(&nfsi->access_cache_entry_lru))
2018 goto remove_lru_entry;
2019 cache = list_entry(nfsi->access_cache_entry_lru.next,
2020 struct nfs_access_entry, lru);
2021 list_move(&cache->lru, &head);
2022 rb_erase(&cache->rb_node, &nfsi->access_cache);
2023 freed++;
2024 if (!list_empty(&nfsi->access_cache_entry_lru))
2025 list_move_tail(&nfsi->access_cache_inode_lru,
2026 &nfs_access_lru_list);
2027 else {
2028 remove_lru_entry:
2029 list_del_init(&nfsi->access_cache_inode_lru);
2030 smp_mb__before_clear_bit();
2031 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2032 smp_mb__after_clear_bit();
2033 }
2034 spin_unlock(&inode->i_lock);
2035 }
2036 spin_unlock(&nfs_access_lru_lock);
2037 nfs_access_free_list(&head);
2038 return freed;
2039 }
2040
2041 unsigned long
2042 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2043 {
2044 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2045 }
2046
2047 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2048 {
2049 struct rb_root *root_node = &nfsi->access_cache;
2050 struct rb_node *n;
2051 struct nfs_access_entry *entry;
2052
2053 /* Unhook entries from the cache */
2054 while ((n = rb_first(root_node)) != NULL) {
2055 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2056 rb_erase(n, root_node);
2057 list_move(&entry->lru, head);
2058 }
2059 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2060 }
2061
2062 void nfs_access_zap_cache(struct inode *inode)
2063 {
2064 LIST_HEAD(head);
2065
2066 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2067 return;
2068 /* Remove from global LRU init */
2069 spin_lock(&nfs_access_lru_lock);
2070 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2071 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2072
2073 spin_lock(&inode->i_lock);
2074 __nfs_access_zap_cache(NFS_I(inode), &head);
2075 spin_unlock(&inode->i_lock);
2076 spin_unlock(&nfs_access_lru_lock);
2077 nfs_access_free_list(&head);
2078 }
2079 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2080
2081 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2082 {
2083 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2084 struct nfs_access_entry *entry;
2085
2086 while (n != NULL) {
2087 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2088
2089 if (cred < entry->cred)
2090 n = n->rb_left;
2091 else if (cred > entry->cred)
2092 n = n->rb_right;
2093 else
2094 return entry;
2095 }
2096 return NULL;
2097 }
2098
2099 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2100 {
2101 struct nfs_inode *nfsi = NFS_I(inode);
2102 struct nfs_access_entry *cache;
2103 int err = -ENOENT;
2104
2105 spin_lock(&inode->i_lock);
2106 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2107 goto out_zap;
2108 cache = nfs_access_search_rbtree(inode, cred);
2109 if (cache == NULL)
2110 goto out;
2111 if (!nfs_have_delegated_attributes(inode) &&
2112 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2113 goto out_stale;
2114 res->jiffies = cache->jiffies;
2115 res->cred = cache->cred;
2116 res->mask = cache->mask;
2117 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2118 err = 0;
2119 out:
2120 spin_unlock(&inode->i_lock);
2121 return err;
2122 out_stale:
2123 rb_erase(&cache->rb_node, &nfsi->access_cache);
2124 list_del(&cache->lru);
2125 spin_unlock(&inode->i_lock);
2126 nfs_access_free_entry(cache);
2127 return -ENOENT;
2128 out_zap:
2129 spin_unlock(&inode->i_lock);
2130 nfs_access_zap_cache(inode);
2131 return -ENOENT;
2132 }
2133
2134 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2135 {
2136 struct nfs_inode *nfsi = NFS_I(inode);
2137 struct rb_root *root_node = &nfsi->access_cache;
2138 struct rb_node **p = &root_node->rb_node;
2139 struct rb_node *parent = NULL;
2140 struct nfs_access_entry *entry;
2141
2142 spin_lock(&inode->i_lock);
2143 while (*p != NULL) {
2144 parent = *p;
2145 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2146
2147 if (set->cred < entry->cred)
2148 p = &parent->rb_left;
2149 else if (set->cred > entry->cred)
2150 p = &parent->rb_right;
2151 else
2152 goto found;
2153 }
2154 rb_link_node(&set->rb_node, parent, p);
2155 rb_insert_color(&set->rb_node, root_node);
2156 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2157 spin_unlock(&inode->i_lock);
2158 return;
2159 found:
2160 rb_replace_node(parent, &set->rb_node, root_node);
2161 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2162 list_del(&entry->lru);
2163 spin_unlock(&inode->i_lock);
2164 nfs_access_free_entry(entry);
2165 }
2166
2167 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2168 {
2169 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2170 if (cache == NULL)
2171 return;
2172 RB_CLEAR_NODE(&cache->rb_node);
2173 cache->jiffies = set->jiffies;
2174 cache->cred = get_rpccred(set->cred);
2175 cache->mask = set->mask;
2176
2177 nfs_access_add_rbtree(inode, cache);
2178
2179 /* Update accounting */
2180 smp_mb__before_atomic_inc();
2181 atomic_long_inc(&nfs_access_nr_entries);
2182 smp_mb__after_atomic_inc();
2183
2184 /* Add inode to global LRU list */
2185 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2186 spin_lock(&nfs_access_lru_lock);
2187 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2188 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2189 &nfs_access_lru_list);
2190 spin_unlock(&nfs_access_lru_lock);
2191 }
2192 }
2193 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2194
2195 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2196 {
2197 entry->mask = 0;
2198 if (access_result & NFS4_ACCESS_READ)
2199 entry->mask |= MAY_READ;
2200 if (access_result &
2201 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2202 entry->mask |= MAY_WRITE;
2203 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2204 entry->mask |= MAY_EXEC;
2205 }
2206 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2207
2208 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2209 {
2210 struct nfs_access_entry cache;
2211 int status;
2212
2213 trace_nfs_access_enter(inode);
2214
2215 status = nfs_access_get_cached(inode, cred, &cache);
2216 if (status == 0)
2217 goto out_cached;
2218
2219 /* Be clever: ask server to check for all possible rights */
2220 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2221 cache.cred = cred;
2222 cache.jiffies = jiffies;
2223 status = NFS_PROTO(inode)->access(inode, &cache);
2224 if (status != 0) {
2225 if (status == -ESTALE) {
2226 nfs_zap_caches(inode);
2227 if (!S_ISDIR(inode->i_mode))
2228 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2229 }
2230 goto out;
2231 }
2232 nfs_access_add_cache(inode, &cache);
2233 out_cached:
2234 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2235 status = -EACCES;
2236 out:
2237 trace_nfs_access_exit(inode, status);
2238 return status;
2239 }
2240
2241 static int nfs_open_permission_mask(int openflags)
2242 {
2243 int mask = 0;
2244
2245 if (openflags & __FMODE_EXEC) {
2246 /* ONLY check exec rights */
2247 mask = MAY_EXEC;
2248 } else {
2249 if ((openflags & O_ACCMODE) != O_WRONLY)
2250 mask |= MAY_READ;
2251 if ((openflags & O_ACCMODE) != O_RDONLY)
2252 mask |= MAY_WRITE;
2253 }
2254
2255 return mask;
2256 }
2257
2258 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2259 {
2260 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2261 }
2262 EXPORT_SYMBOL_GPL(nfs_may_open);
2263
2264 int nfs_permission(struct inode *inode, int mask)
2265 {
2266 struct rpc_cred *cred;
2267 int res = 0;
2268
2269 if (mask & MAY_NOT_BLOCK)
2270 return -ECHILD;
2271
2272 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2273
2274 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2275 goto out;
2276 /* Is this sys_access() ? */
2277 if (mask & (MAY_ACCESS | MAY_CHDIR))
2278 goto force_lookup;
2279
2280 switch (inode->i_mode & S_IFMT) {
2281 case S_IFLNK:
2282 goto out;
2283 case S_IFREG:
2284 break;
2285 case S_IFDIR:
2286 /*
2287 * Optimize away all write operations, since the server
2288 * will check permissions when we perform the op.
2289 */
2290 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2291 goto out;
2292 }
2293
2294 force_lookup:
2295 if (!NFS_PROTO(inode)->access)
2296 goto out_notsup;
2297
2298 cred = rpc_lookup_cred();
2299 if (!IS_ERR(cred)) {
2300 res = nfs_do_access(inode, cred, mask);
2301 put_rpccred(cred);
2302 } else
2303 res = PTR_ERR(cred);
2304 out:
2305 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2306 res = -EACCES;
2307
2308 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2309 inode->i_sb->s_id, inode->i_ino, mask, res);
2310 return res;
2311 out_notsup:
2312 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2313 if (res == 0)
2314 res = generic_permission(inode, mask);
2315 goto out;
2316 }
2317 EXPORT_SYMBOL_GPL(nfs_permission);
2318
2319 /*
2320 * Local variables:
2321 * version-control: t
2322 * kept-new-versions: 5
2323 * End:
2324 */
This page took 0.076861 seconds and 5 git commands to generate.