]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/nfs/dir.c
8a45e6d1f6a477cfdb18fe29698c59c8d3f70a3b
[mirror_ubuntu-zesty-kernel.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/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
27 #include <linux/mm.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/kmemleak.h>
37 #include <linux/xattr.h>
38
39 #include "delegation.h"
40 #include "iostat.h"
41 #include "internal.h"
42 #include "fscache.h"
43
44 /* #define NFS_DEBUG_VERBOSE 1 */
45
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_closedir(struct inode *, struct file *);
48 static int nfs_readdir(struct file *, void *, filldir_t);
49 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, int);
52 static int nfs_rmdir(struct inode *, struct dentry *);
53 static int nfs_unlink(struct inode *, struct dentry *);
54 static int nfs_symlink(struct inode *, struct dentry *, const char *);
55 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, int);
60 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
61 static void nfs_readdir_clear_array(struct page*);
62
63 const struct file_operations nfs_dir_operations = {
64 .llseek = nfs_llseek_dir,
65 .read = generic_read_dir,
66 .readdir = nfs_readdir,
67 .open = nfs_opendir,
68 .release = nfs_closedir,
69 .fsync = nfs_fsync_dir,
70 };
71
72 const struct inode_operations nfs_dir_inode_operations = {
73 .create = nfs_create,
74 .lookup = nfs_lookup,
75 .link = nfs_link,
76 .unlink = nfs_unlink,
77 .symlink = nfs_symlink,
78 .mkdir = nfs_mkdir,
79 .rmdir = nfs_rmdir,
80 .mknod = nfs_mknod,
81 .rename = nfs_rename,
82 .permission = nfs_permission,
83 .getattr = nfs_getattr,
84 .setattr = nfs_setattr,
85 };
86
87 const struct address_space_operations nfs_dir_aops = {
88 .freepage = nfs_readdir_clear_array,
89 };
90
91 #ifdef CONFIG_NFS_V3
92 const struct inode_operations nfs3_dir_inode_operations = {
93 .create = nfs_create,
94 .lookup = nfs_lookup,
95 .link = nfs_link,
96 .unlink = nfs_unlink,
97 .symlink = nfs_symlink,
98 .mkdir = nfs_mkdir,
99 .rmdir = nfs_rmdir,
100 .mknod = nfs_mknod,
101 .rename = nfs_rename,
102 .permission = nfs_permission,
103 .getattr = nfs_getattr,
104 .setattr = nfs_setattr,
105 .listxattr = nfs3_listxattr,
106 .getxattr = nfs3_getxattr,
107 .setxattr = nfs3_setxattr,
108 .removexattr = nfs3_removexattr,
109 };
110 #endif /* CONFIG_NFS_V3 */
111
112 #ifdef CONFIG_NFS_V4
113
114 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
115 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
116 const struct inode_operations nfs4_dir_inode_operations = {
117 .create = nfs_open_create,
118 .lookup = nfs_atomic_lookup,
119 .link = nfs_link,
120 .unlink = nfs_unlink,
121 .symlink = nfs_symlink,
122 .mkdir = nfs_mkdir,
123 .rmdir = nfs_rmdir,
124 .mknod = nfs_mknod,
125 .rename = nfs_rename,
126 .permission = nfs_permission,
127 .getattr = nfs_getattr,
128 .setattr = nfs_setattr,
129 .getxattr = generic_getxattr,
130 .setxattr = generic_setxattr,
131 .listxattr = generic_listxattr,
132 .removexattr = generic_removexattr,
133 };
134
135 #endif /* CONFIG_NFS_V4 */
136
137 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct rpc_cred *cred)
138 {
139 struct nfs_open_dir_context *ctx;
140 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
141 if (ctx != NULL) {
142 ctx->duped = 0;
143 ctx->dir_cookie = 0;
144 ctx->dup_cookie = 0;
145 ctx->cred = get_rpccred(cred);
146 } else
147 ctx = ERR_PTR(-ENOMEM);
148 return ctx;
149 }
150
151 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
152 {
153 put_rpccred(ctx->cred);
154 kfree(ctx);
155 }
156
157 /*
158 * Open file
159 */
160 static int
161 nfs_opendir(struct inode *inode, struct file *filp)
162 {
163 int res = 0;
164 struct nfs_open_dir_context *ctx;
165 struct rpc_cred *cred;
166
167 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
168 filp->f_path.dentry->d_parent->d_name.name,
169 filp->f_path.dentry->d_name.name);
170
171 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
172
173 cred = rpc_lookup_cred();
174 if (IS_ERR(cred))
175 return PTR_ERR(cred);
176 ctx = alloc_nfs_open_dir_context(cred);
177 if (IS_ERR(ctx)) {
178 res = PTR_ERR(ctx);
179 goto out;
180 }
181 filp->private_data = ctx;
182 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
183 /* This is a mountpoint, so d_revalidate will never
184 * have been called, so we need to refresh the
185 * inode (for close-open consistency) ourselves.
186 */
187 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
188 }
189 out:
190 put_rpccred(cred);
191 return res;
192 }
193
194 static int
195 nfs_closedir(struct inode *inode, struct file *filp)
196 {
197 put_nfs_open_dir_context(filp->private_data);
198 return 0;
199 }
200
201 struct nfs_cache_array_entry {
202 u64 cookie;
203 u64 ino;
204 struct qstr string;
205 unsigned char d_type;
206 };
207
208 struct nfs_cache_array {
209 unsigned int size;
210 int eof_index;
211 u64 last_cookie;
212 struct nfs_cache_array_entry array[0];
213 };
214
215 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
216 typedef struct {
217 struct file *file;
218 struct page *page;
219 unsigned long page_index;
220 u64 *dir_cookie;
221 u64 last_cookie;
222 loff_t current_index;
223 decode_dirent_t decode;
224
225 unsigned long timestamp;
226 unsigned long gencount;
227 unsigned int cache_entry_index;
228 unsigned int plus:1;
229 unsigned int eof:1;
230 } nfs_readdir_descriptor_t;
231
232 /*
233 * The caller is responsible for calling nfs_readdir_release_array(page)
234 */
235 static
236 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
237 {
238 void *ptr;
239 if (page == NULL)
240 return ERR_PTR(-EIO);
241 ptr = kmap(page);
242 if (ptr == NULL)
243 return ERR_PTR(-ENOMEM);
244 return ptr;
245 }
246
247 static
248 void nfs_readdir_release_array(struct page *page)
249 {
250 kunmap(page);
251 }
252
253 /*
254 * we are freeing strings created by nfs_add_to_readdir_array()
255 */
256 static
257 void nfs_readdir_clear_array(struct page *page)
258 {
259 struct nfs_cache_array *array;
260 int i;
261
262 array = kmap_atomic(page, KM_USER0);
263 for (i = 0; i < array->size; i++)
264 kfree(array->array[i].string.name);
265 kunmap_atomic(array, KM_USER0);
266 }
267
268 /*
269 * the caller is responsible for freeing qstr.name
270 * when called by nfs_readdir_add_to_array, the strings will be freed in
271 * nfs_clear_readdir_array()
272 */
273 static
274 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
275 {
276 string->len = len;
277 string->name = kmemdup(name, len, GFP_KERNEL);
278 if (string->name == NULL)
279 return -ENOMEM;
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 kmemleak_not_leak(string->name);
285 string->hash = full_name_hash(name, len);
286 return 0;
287 }
288
289 static
290 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
291 {
292 struct nfs_cache_array *array = nfs_readdir_get_array(page);
293 struct nfs_cache_array_entry *cache_entry;
294 int ret;
295
296 if (IS_ERR(array))
297 return PTR_ERR(array);
298
299 cache_entry = &array->array[array->size];
300
301 /* Check that this entry lies within the page bounds */
302 ret = -ENOSPC;
303 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
304 goto out;
305
306 cache_entry->cookie = entry->prev_cookie;
307 cache_entry->ino = entry->ino;
308 cache_entry->d_type = entry->d_type;
309 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
310 if (ret)
311 goto out;
312 array->last_cookie = entry->cookie;
313 array->size++;
314 if (entry->eof != 0)
315 array->eof_index = array->size;
316 out:
317 nfs_readdir_release_array(page);
318 return ret;
319 }
320
321 static
322 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
323 {
324 loff_t diff = desc->file->f_pos - desc->current_index;
325 unsigned int index;
326 struct nfs_open_dir_context *ctx = desc->file->private_data;
327
328 if (diff < 0)
329 goto out_eof;
330 if (diff >= array->size) {
331 if (array->eof_index >= 0)
332 goto out_eof;
333 return -EAGAIN;
334 }
335
336 index = (unsigned int)diff;
337 *desc->dir_cookie = array->array[index].cookie;
338 desc->cache_entry_index = index;
339 ctx->duped = 0;
340 return 0;
341 out_eof:
342 desc->eof = 1;
343 return -EBADCOOKIE;
344 }
345
346 static
347 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
348 {
349 int i;
350 loff_t new_pos;
351 int status = -EAGAIN;
352 struct nfs_open_dir_context *ctx = desc->file->private_data;
353
354 for (i = 0; i < array->size; i++) {
355 if (array->array[i].cookie == *desc->dir_cookie) {
356 new_pos = desc->current_index + i;
357 if (new_pos < desc->file->f_pos) {
358 ctx->dup_cookie = *desc->dir_cookie;
359 ctx->duped = 1;
360 }
361 desc->file->f_pos = new_pos;
362 desc->cache_entry_index = i;
363 return 0;
364 }
365 }
366 if (array->eof_index >= 0) {
367 status = -EBADCOOKIE;
368 if (*desc->dir_cookie == array->last_cookie)
369 desc->eof = 1;
370 }
371 return status;
372 }
373
374 static
375 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
376 {
377 struct nfs_cache_array *array;
378 int status;
379
380 array = nfs_readdir_get_array(desc->page);
381 if (IS_ERR(array)) {
382 status = PTR_ERR(array);
383 goto out;
384 }
385
386 if (*desc->dir_cookie == 0)
387 status = nfs_readdir_search_for_pos(array, desc);
388 else
389 status = nfs_readdir_search_for_cookie(array, desc);
390
391 if (status == -EAGAIN) {
392 desc->last_cookie = array->last_cookie;
393 desc->current_index += array->size;
394 desc->page_index++;
395 }
396 nfs_readdir_release_array(desc->page);
397 out:
398 return status;
399 }
400
401 /* Fill a page with xdr information before transferring to the cache page */
402 static
403 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
404 struct nfs_entry *entry, struct file *file, struct inode *inode)
405 {
406 struct nfs_open_dir_context *ctx = file->private_data;
407 struct rpc_cred *cred = ctx->cred;
408 unsigned long timestamp, gencount;
409 int error;
410
411 again:
412 timestamp = jiffies;
413 gencount = nfs_inc_attr_generation_counter();
414 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
415 NFS_SERVER(inode)->dtsize, desc->plus);
416 if (error < 0) {
417 /* We requested READDIRPLUS, but the server doesn't grok it */
418 if (error == -ENOTSUPP && desc->plus) {
419 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
420 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
421 desc->plus = 0;
422 goto again;
423 }
424 goto error;
425 }
426 desc->timestamp = timestamp;
427 desc->gencount = gencount;
428 error:
429 return error;
430 }
431
432 static int xdr_decode(nfs_readdir_descriptor_t *desc,
433 struct nfs_entry *entry, struct xdr_stream *xdr)
434 {
435 int error;
436
437 error = desc->decode(xdr, entry, desc->plus);
438 if (error)
439 return error;
440 entry->fattr->time_start = desc->timestamp;
441 entry->fattr->gencount = desc->gencount;
442 return 0;
443 }
444
445 static
446 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
447 {
448 if (dentry->d_inode == NULL)
449 goto different;
450 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
451 goto different;
452 return 1;
453 different:
454 return 0;
455 }
456
457 static
458 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
459 {
460 struct qstr filename = {
461 .len = entry->len,
462 .name = entry->name,
463 };
464 struct dentry *dentry;
465 struct dentry *alias;
466 struct inode *dir = parent->d_inode;
467 struct inode *inode;
468
469 if (filename.name[0] == '.') {
470 if (filename.len == 1)
471 return;
472 if (filename.len == 2 && filename.name[1] == '.')
473 return;
474 }
475 filename.hash = full_name_hash(filename.name, filename.len);
476
477 dentry = d_lookup(parent, &filename);
478 if (dentry != NULL) {
479 if (nfs_same_file(dentry, entry)) {
480 nfs_refresh_inode(dentry->d_inode, entry->fattr);
481 goto out;
482 } else {
483 d_drop(dentry);
484 dput(dentry);
485 }
486 }
487
488 dentry = d_alloc(parent, &filename);
489 if (dentry == NULL)
490 return;
491
492 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
493 if (IS_ERR(inode))
494 goto out;
495
496 alias = d_materialise_unique(dentry, inode);
497 if (IS_ERR(alias))
498 goto out;
499 else if (alias) {
500 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
501 dput(alias);
502 } else
503 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
504
505 out:
506 dput(dentry);
507 }
508
509 /* Perform conversion from xdr to cache array */
510 static
511 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
512 struct page **xdr_pages, struct page *page, unsigned int buflen)
513 {
514 struct xdr_stream stream;
515 struct xdr_buf buf;
516 struct page *scratch;
517 struct nfs_cache_array *array;
518 unsigned int count = 0;
519 int status;
520
521 scratch = alloc_page(GFP_KERNEL);
522 if (scratch == NULL)
523 return -ENOMEM;
524
525 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
526 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
527
528 do {
529 status = xdr_decode(desc, entry, &stream);
530 if (status != 0) {
531 if (status == -EAGAIN)
532 status = 0;
533 break;
534 }
535
536 count++;
537
538 if (desc->plus != 0)
539 nfs_prime_dcache(desc->file->f_path.dentry, entry);
540
541 status = nfs_readdir_add_to_array(entry, page);
542 if (status != 0)
543 break;
544 } while (!entry->eof);
545
546 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
547 array = nfs_readdir_get_array(page);
548 if (!IS_ERR(array)) {
549 array->eof_index = array->size;
550 status = 0;
551 nfs_readdir_release_array(page);
552 } else
553 status = PTR_ERR(array);
554 }
555
556 put_page(scratch);
557 return status;
558 }
559
560 static
561 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
562 {
563 unsigned int i;
564 for (i = 0; i < npages; i++)
565 put_page(pages[i]);
566 }
567
568 static
569 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
570 unsigned int npages)
571 {
572 nfs_readdir_free_pagearray(pages, npages);
573 }
574
575 /*
576 * nfs_readdir_large_page will allocate pages that must be freed with a call
577 * to nfs_readdir_free_large_page
578 */
579 static
580 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
581 {
582 unsigned int i;
583
584 for (i = 0; i < npages; i++) {
585 struct page *page = alloc_page(GFP_KERNEL);
586 if (page == NULL)
587 goto out_freepages;
588 pages[i] = page;
589 }
590 return 0;
591
592 out_freepages:
593 nfs_readdir_free_pagearray(pages, i);
594 return -ENOMEM;
595 }
596
597 static
598 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
599 {
600 struct page *pages[NFS_MAX_READDIR_PAGES];
601 void *pages_ptr = NULL;
602 struct nfs_entry entry;
603 struct file *file = desc->file;
604 struct nfs_cache_array *array;
605 int status = -ENOMEM;
606 unsigned int array_size = ARRAY_SIZE(pages);
607
608 entry.prev_cookie = 0;
609 entry.cookie = desc->last_cookie;
610 entry.eof = 0;
611 entry.fh = nfs_alloc_fhandle();
612 entry.fattr = nfs_alloc_fattr();
613 entry.server = NFS_SERVER(inode);
614 if (entry.fh == NULL || entry.fattr == NULL)
615 goto out;
616
617 array = nfs_readdir_get_array(page);
618 if (IS_ERR(array)) {
619 status = PTR_ERR(array);
620 goto out;
621 }
622 memset(array, 0, sizeof(struct nfs_cache_array));
623 array->eof_index = -1;
624
625 status = nfs_readdir_large_page(pages, array_size);
626 if (status < 0)
627 goto out_release_array;
628 do {
629 unsigned int pglen;
630 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
631
632 if (status < 0)
633 break;
634 pglen = status;
635 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
636 if (status < 0) {
637 if (status == -ENOSPC)
638 status = 0;
639 break;
640 }
641 } while (array->eof_index < 0);
642
643 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
644 out_release_array:
645 nfs_readdir_release_array(page);
646 out:
647 nfs_free_fattr(entry.fattr);
648 nfs_free_fhandle(entry.fh);
649 return status;
650 }
651
652 /*
653 * Now we cache directories properly, by converting xdr information
654 * to an array that can be used for lookups later. This results in
655 * fewer cache pages, since we can store more information on each page.
656 * We only need to convert from xdr once so future lookups are much simpler
657 */
658 static
659 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
660 {
661 struct inode *inode = desc->file->f_path.dentry->d_inode;
662 int ret;
663
664 ret = nfs_readdir_xdr_to_array(desc, page, inode);
665 if (ret < 0)
666 goto error;
667 SetPageUptodate(page);
668
669 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
670 /* Should never happen */
671 nfs_zap_mapping(inode, inode->i_mapping);
672 }
673 unlock_page(page);
674 return 0;
675 error:
676 unlock_page(page);
677 return ret;
678 }
679
680 static
681 void cache_page_release(nfs_readdir_descriptor_t *desc)
682 {
683 if (!desc->page->mapping)
684 nfs_readdir_clear_array(desc->page);
685 page_cache_release(desc->page);
686 desc->page = NULL;
687 }
688
689 static
690 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
691 {
692 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
693 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
694 }
695
696 /*
697 * Returns 0 if desc->dir_cookie was found on page desc->page_index
698 */
699 static
700 int find_cache_page(nfs_readdir_descriptor_t *desc)
701 {
702 int res;
703
704 desc->page = get_cache_page(desc);
705 if (IS_ERR(desc->page))
706 return PTR_ERR(desc->page);
707
708 res = nfs_readdir_search_array(desc);
709 if (res != 0)
710 cache_page_release(desc);
711 return res;
712 }
713
714 /* Search for desc->dir_cookie from the beginning of the page cache */
715 static inline
716 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
717 {
718 int res;
719
720 if (desc->page_index == 0) {
721 desc->current_index = 0;
722 desc->last_cookie = 0;
723 }
724 do {
725 res = find_cache_page(desc);
726 } while (res == -EAGAIN);
727 return res;
728 }
729
730 /*
731 * Once we've found the start of the dirent within a page: fill 'er up...
732 */
733 static
734 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
735 filldir_t filldir)
736 {
737 struct file *file = desc->file;
738 int i = 0;
739 int res = 0;
740 struct nfs_cache_array *array = NULL;
741 struct nfs_open_dir_context *ctx = file->private_data;
742
743 if (ctx->duped != 0 && ctx->dup_cookie == *desc->dir_cookie) {
744 if (printk_ratelimit()) {
745 pr_notice("NFS: directory %s/%s contains a readdir loop. "
746 "Please contact your server vendor. "
747 "Offending cookie: %llu\n",
748 file->f_dentry->d_parent->d_name.name,
749 file->f_dentry->d_name.name,
750 *desc->dir_cookie);
751 }
752 res = -ELOOP;
753 goto out;
754 }
755
756 array = nfs_readdir_get_array(desc->page);
757 if (IS_ERR(array)) {
758 res = PTR_ERR(array);
759 goto out;
760 }
761
762 for (i = desc->cache_entry_index; i < array->size; i++) {
763 struct nfs_cache_array_entry *ent;
764
765 ent = &array->array[i];
766 if (filldir(dirent, ent->string.name, ent->string.len,
767 file->f_pos, nfs_compat_user_ino64(ent->ino),
768 ent->d_type) < 0) {
769 desc->eof = 1;
770 break;
771 }
772 file->f_pos++;
773 if (i < (array->size-1))
774 *desc->dir_cookie = array->array[i+1].cookie;
775 else
776 *desc->dir_cookie = array->last_cookie;
777 }
778 if (array->eof_index >= 0)
779 desc->eof = 1;
780
781 nfs_readdir_release_array(desc->page);
782 out:
783 cache_page_release(desc);
784 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
785 (unsigned long long)*desc->dir_cookie, res);
786 return res;
787 }
788
789 /*
790 * If we cannot find a cookie in our cache, we suspect that this is
791 * because it points to a deleted file, so we ask the server to return
792 * whatever it thinks is the next entry. We then feed this to filldir.
793 * If all goes well, we should then be able to find our way round the
794 * cache on the next call to readdir_search_pagecache();
795 *
796 * NOTE: we cannot add the anonymous page to the pagecache because
797 * the data it contains might not be page aligned. Besides,
798 * we should already have a complete representation of the
799 * directory in the page cache by the time we get here.
800 */
801 static inline
802 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
803 filldir_t filldir)
804 {
805 struct page *page = NULL;
806 int status;
807 struct inode *inode = desc->file->f_path.dentry->d_inode;
808
809 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
810 (unsigned long long)*desc->dir_cookie);
811
812 page = alloc_page(GFP_HIGHUSER);
813 if (!page) {
814 status = -ENOMEM;
815 goto out;
816 }
817
818 desc->page_index = 0;
819 desc->last_cookie = *desc->dir_cookie;
820 desc->page = page;
821
822 status = nfs_readdir_xdr_to_array(desc, page, inode);
823 if (status < 0)
824 goto out_release;
825
826 status = nfs_do_filldir(desc, dirent, filldir);
827
828 out:
829 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
830 __func__, status);
831 return status;
832 out_release:
833 cache_page_release(desc);
834 goto out;
835 }
836
837 /* The file offset position represents the dirent entry number. A
838 last cookie cache takes care of the common case of reading the
839 whole directory.
840 */
841 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
842 {
843 struct dentry *dentry = filp->f_path.dentry;
844 struct inode *inode = dentry->d_inode;
845 nfs_readdir_descriptor_t my_desc,
846 *desc = &my_desc;
847 struct nfs_open_dir_context *dir_ctx = filp->private_data;
848 int res;
849
850 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
851 dentry->d_parent->d_name.name, dentry->d_name.name,
852 (long long)filp->f_pos);
853 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
854
855 /*
856 * filp->f_pos points to the dirent entry number.
857 * *desc->dir_cookie has the cookie for the next entry. We have
858 * to either find the entry with the appropriate number or
859 * revalidate the cookie.
860 */
861 memset(desc, 0, sizeof(*desc));
862
863 desc->file = filp;
864 desc->dir_cookie = &dir_ctx->dir_cookie;
865 desc->decode = NFS_PROTO(inode)->decode_dirent;
866 desc->plus = NFS_USE_READDIRPLUS(inode);
867
868 nfs_block_sillyrename(dentry);
869 res = nfs_revalidate_mapping(inode, filp->f_mapping);
870 if (res < 0)
871 goto out;
872
873 do {
874 res = readdir_search_pagecache(desc);
875
876 if (res == -EBADCOOKIE) {
877 res = 0;
878 /* This means either end of directory */
879 if (*desc->dir_cookie && desc->eof == 0) {
880 /* Or that the server has 'lost' a cookie */
881 res = uncached_readdir(desc, dirent, filldir);
882 if (res == 0)
883 continue;
884 }
885 break;
886 }
887 if (res == -ETOOSMALL && desc->plus) {
888 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
889 nfs_zap_caches(inode);
890 desc->page_index = 0;
891 desc->plus = 0;
892 desc->eof = 0;
893 continue;
894 }
895 if (res < 0)
896 break;
897
898 res = nfs_do_filldir(desc, dirent, filldir);
899 if (res < 0)
900 break;
901 } while (!desc->eof);
902 out:
903 nfs_unblock_sillyrename(dentry);
904 if (res > 0)
905 res = 0;
906 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
907 dentry->d_parent->d_name.name, dentry->d_name.name,
908 res);
909 return res;
910 }
911
912 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
913 {
914 struct dentry *dentry = filp->f_path.dentry;
915 struct inode *inode = dentry->d_inode;
916 struct nfs_open_dir_context *dir_ctx = filp->private_data;
917
918 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
919 dentry->d_parent->d_name.name,
920 dentry->d_name.name,
921 offset, origin);
922
923 mutex_lock(&inode->i_mutex);
924 switch (origin) {
925 case 1:
926 offset += filp->f_pos;
927 case 0:
928 if (offset >= 0)
929 break;
930 default:
931 offset = -EINVAL;
932 goto out;
933 }
934 if (offset != filp->f_pos) {
935 filp->f_pos = offset;
936 dir_ctx->dir_cookie = 0;
937 dir_ctx->duped = 0;
938 }
939 out:
940 mutex_unlock(&inode->i_mutex);
941 return offset;
942 }
943
944 /*
945 * All directory operations under NFS are synchronous, so fsync()
946 * is a dummy operation.
947 */
948 static int nfs_fsync_dir(struct file *filp, int datasync)
949 {
950 struct dentry *dentry = filp->f_path.dentry;
951
952 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
953 dentry->d_parent->d_name.name, dentry->d_name.name,
954 datasync);
955
956 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
957 return 0;
958 }
959
960 /**
961 * nfs_force_lookup_revalidate - Mark the directory as having changed
962 * @dir - pointer to directory inode
963 *
964 * This forces the revalidation code in nfs_lookup_revalidate() to do a
965 * full lookup on all child dentries of 'dir' whenever a change occurs
966 * on the server that might have invalidated our dcache.
967 *
968 * The caller should be holding dir->i_lock
969 */
970 void nfs_force_lookup_revalidate(struct inode *dir)
971 {
972 NFS_I(dir)->cache_change_attribute++;
973 }
974
975 /*
976 * A check for whether or not the parent directory has changed.
977 * In the case it has, we assume that the dentries are untrustworthy
978 * and may need to be looked up again.
979 */
980 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
981 {
982 if (IS_ROOT(dentry))
983 return 1;
984 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
985 return 0;
986 if (!nfs_verify_change_attribute(dir, dentry->d_time))
987 return 0;
988 /* Revalidate nfsi->cache_change_attribute before we declare a match */
989 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
990 return 0;
991 if (!nfs_verify_change_attribute(dir, dentry->d_time))
992 return 0;
993 return 1;
994 }
995
996 /*
997 * Return the intent data that applies to this particular path component
998 *
999 * Note that the current set of intents only apply to the very last
1000 * component of the path and none of them is set before that last
1001 * component.
1002 */
1003 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1004 unsigned int mask)
1005 {
1006 return nd->flags & mask;
1007 }
1008
1009 /*
1010 * Use intent information to check whether or not we're going to do
1011 * an O_EXCL create using this path component.
1012 */
1013 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1014 {
1015 if (NFS_PROTO(dir)->version == 2)
1016 return 0;
1017 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1018 }
1019
1020 /*
1021 * Inode and filehandle revalidation for lookups.
1022 *
1023 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1024 * or if the intent information indicates that we're about to open this
1025 * particular file and the "nocto" mount flag is not set.
1026 *
1027 */
1028 static inline
1029 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1030 {
1031 struct nfs_server *server = NFS_SERVER(inode);
1032
1033 if (IS_AUTOMOUNT(inode))
1034 return 0;
1035 if (nd != NULL) {
1036 /* VFS wants an on-the-wire revalidation */
1037 if (nd->flags & LOOKUP_REVAL)
1038 goto out_force;
1039 /* This is an open(2) */
1040 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1041 !(server->flags & NFS_MOUNT_NOCTO) &&
1042 (S_ISREG(inode->i_mode) ||
1043 S_ISDIR(inode->i_mode)))
1044 goto out_force;
1045 return 0;
1046 }
1047 return nfs_revalidate_inode(server, inode);
1048 out_force:
1049 return __nfs_revalidate_inode(server, inode);
1050 }
1051
1052 /*
1053 * We judge how long we want to trust negative
1054 * dentries by looking at the parent inode mtime.
1055 *
1056 * If parent mtime has changed, we revalidate, else we wait for a
1057 * period corresponding to the parent's attribute cache timeout value.
1058 */
1059 static inline
1060 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1061 struct nameidata *nd)
1062 {
1063 /* Don't revalidate a negative dentry if we're creating a new file */
1064 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1065 return 0;
1066 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1067 return 1;
1068 return !nfs_check_verifier(dir, dentry);
1069 }
1070
1071 /*
1072 * This is called every time the dcache has a lookup hit,
1073 * and we should check whether we can really trust that
1074 * lookup.
1075 *
1076 * NOTE! The hit can be a negative hit too, don't assume
1077 * we have an inode!
1078 *
1079 * If the parent directory is seen to have changed, we throw out the
1080 * cached dentry and do a new lookup.
1081 */
1082 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1083 {
1084 struct inode *dir;
1085 struct inode *inode;
1086 struct dentry *parent;
1087 struct nfs_fh *fhandle = NULL;
1088 struct nfs_fattr *fattr = NULL;
1089 int error;
1090
1091 if (nd->flags & LOOKUP_RCU)
1092 return -ECHILD;
1093
1094 parent = dget_parent(dentry);
1095 dir = parent->d_inode;
1096 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1097 inode = dentry->d_inode;
1098
1099 if (!inode) {
1100 if (nfs_neg_need_reval(dir, dentry, nd))
1101 goto out_bad;
1102 goto out_valid;
1103 }
1104
1105 if (is_bad_inode(inode)) {
1106 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1107 __func__, dentry->d_parent->d_name.name,
1108 dentry->d_name.name);
1109 goto out_bad;
1110 }
1111
1112 if (nfs_have_delegation(inode, FMODE_READ))
1113 goto out_set_verifier;
1114
1115 /* Force a full look up iff the parent directory has changed */
1116 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1117 if (nfs_lookup_verify_inode(inode, nd))
1118 goto out_zap_parent;
1119 goto out_valid;
1120 }
1121
1122 if (NFS_STALE(inode))
1123 goto out_bad;
1124
1125 error = -ENOMEM;
1126 fhandle = nfs_alloc_fhandle();
1127 fattr = nfs_alloc_fattr();
1128 if (fhandle == NULL || fattr == NULL)
1129 goto out_error;
1130
1131 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1132 if (error)
1133 goto out_bad;
1134 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1135 goto out_bad;
1136 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1137 goto out_bad;
1138
1139 nfs_free_fattr(fattr);
1140 nfs_free_fhandle(fhandle);
1141 out_set_verifier:
1142 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1143 out_valid:
1144 dput(parent);
1145 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1146 __func__, dentry->d_parent->d_name.name,
1147 dentry->d_name.name);
1148 return 1;
1149 out_zap_parent:
1150 nfs_zap_caches(dir);
1151 out_bad:
1152 nfs_mark_for_revalidate(dir);
1153 if (inode && S_ISDIR(inode->i_mode)) {
1154 /* Purge readdir caches. */
1155 nfs_zap_caches(inode);
1156 /* If we have submounts, don't unhash ! */
1157 if (have_submounts(dentry))
1158 goto out_valid;
1159 if (dentry->d_flags & DCACHE_DISCONNECTED)
1160 goto out_valid;
1161 shrink_dcache_parent(dentry);
1162 }
1163 d_drop(dentry);
1164 nfs_free_fattr(fattr);
1165 nfs_free_fhandle(fhandle);
1166 dput(parent);
1167 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1168 __func__, dentry->d_parent->d_name.name,
1169 dentry->d_name.name);
1170 return 0;
1171 out_error:
1172 nfs_free_fattr(fattr);
1173 nfs_free_fhandle(fhandle);
1174 dput(parent);
1175 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1176 __func__, dentry->d_parent->d_name.name,
1177 dentry->d_name.name, error);
1178 return error;
1179 }
1180
1181 /*
1182 * This is called from dput() when d_count is going to 0.
1183 */
1184 static int nfs_dentry_delete(const struct dentry *dentry)
1185 {
1186 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1187 dentry->d_parent->d_name.name, dentry->d_name.name,
1188 dentry->d_flags);
1189
1190 /* Unhash any dentry with a stale inode */
1191 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1192 return 1;
1193
1194 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1195 /* Unhash it, so that ->d_iput() would be called */
1196 return 1;
1197 }
1198 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1199 /* Unhash it, so that ancestors of killed async unlink
1200 * files will be cleaned up during umount */
1201 return 1;
1202 }
1203 return 0;
1204
1205 }
1206
1207 static void nfs_drop_nlink(struct inode *inode)
1208 {
1209 spin_lock(&inode->i_lock);
1210 if (inode->i_nlink > 0)
1211 drop_nlink(inode);
1212 spin_unlock(&inode->i_lock);
1213 }
1214
1215 /*
1216 * Called when the dentry loses inode.
1217 * We use it to clean up silly-renamed files.
1218 */
1219 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1220 {
1221 if (S_ISDIR(inode->i_mode))
1222 /* drop any readdir cache as it could easily be old */
1223 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1224
1225 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1226 drop_nlink(inode);
1227 nfs_complete_unlink(dentry, inode);
1228 }
1229 iput(inode);
1230 }
1231
1232 static void nfs_d_release(struct dentry *dentry)
1233 {
1234 /* free cached devname value, if it survived that far */
1235 if (unlikely(dentry->d_fsdata)) {
1236 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1237 WARN_ON(1);
1238 else
1239 kfree(dentry->d_fsdata);
1240 }
1241 }
1242
1243 const struct dentry_operations nfs_dentry_operations = {
1244 .d_revalidate = nfs_lookup_revalidate,
1245 .d_delete = nfs_dentry_delete,
1246 .d_iput = nfs_dentry_iput,
1247 .d_automount = nfs_d_automount,
1248 .d_release = nfs_d_release,
1249 };
1250
1251 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1252 {
1253 struct dentry *res;
1254 struct dentry *parent;
1255 struct inode *inode = NULL;
1256 struct nfs_fh *fhandle = NULL;
1257 struct nfs_fattr *fattr = NULL;
1258 int error;
1259
1260 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1261 dentry->d_parent->d_name.name, dentry->d_name.name);
1262 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1263
1264 res = ERR_PTR(-ENAMETOOLONG);
1265 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1266 goto out;
1267
1268 /*
1269 * If we're doing an exclusive create, optimize away the lookup
1270 * but don't hash the dentry.
1271 */
1272 if (nfs_is_exclusive_create(dir, nd)) {
1273 d_instantiate(dentry, NULL);
1274 res = NULL;
1275 goto out;
1276 }
1277
1278 res = ERR_PTR(-ENOMEM);
1279 fhandle = nfs_alloc_fhandle();
1280 fattr = nfs_alloc_fattr();
1281 if (fhandle == NULL || fattr == NULL)
1282 goto out;
1283
1284 parent = dentry->d_parent;
1285 /* Protect against concurrent sillydeletes */
1286 nfs_block_sillyrename(parent);
1287 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1288 if (error == -ENOENT)
1289 goto no_entry;
1290 if (error < 0) {
1291 res = ERR_PTR(error);
1292 goto out_unblock_sillyrename;
1293 }
1294 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1295 res = ERR_CAST(inode);
1296 if (IS_ERR(res))
1297 goto out_unblock_sillyrename;
1298
1299 no_entry:
1300 res = d_materialise_unique(dentry, inode);
1301 if (res != NULL) {
1302 if (IS_ERR(res))
1303 goto out_unblock_sillyrename;
1304 dentry = res;
1305 }
1306 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1307 out_unblock_sillyrename:
1308 nfs_unblock_sillyrename(parent);
1309 out:
1310 nfs_free_fattr(fattr);
1311 nfs_free_fhandle(fhandle);
1312 return res;
1313 }
1314
1315 #ifdef CONFIG_NFS_V4
1316 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1317
1318 const struct dentry_operations nfs4_dentry_operations = {
1319 .d_revalidate = nfs_open_revalidate,
1320 .d_delete = nfs_dentry_delete,
1321 .d_iput = nfs_dentry_iput,
1322 .d_automount = nfs_d_automount,
1323 .d_release = nfs_d_release,
1324 };
1325
1326 /*
1327 * Use intent information to determine whether we need to substitute
1328 * the NFSv4-style stateful OPEN for the LOOKUP call
1329 */
1330 static int is_atomic_open(struct nameidata *nd)
1331 {
1332 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1333 return 0;
1334 /* NFS does not (yet) have a stateful open for directories */
1335 if (nd->flags & LOOKUP_DIRECTORY)
1336 return 0;
1337 /* Are we trying to write to a read only partition? */
1338 if (__mnt_is_readonly(nd->path.mnt) &&
1339 (nd->intent.open.flags & (O_CREAT|O_TRUNC|O_ACCMODE)))
1340 return 0;
1341 return 1;
1342 }
1343
1344 static fmode_t flags_to_mode(int flags)
1345 {
1346 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1347 if ((flags & O_ACCMODE) != O_WRONLY)
1348 res |= FMODE_READ;
1349 if ((flags & O_ACCMODE) != O_RDONLY)
1350 res |= FMODE_WRITE;
1351 return res;
1352 }
1353
1354 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1355 {
1356 struct nfs_open_context *ctx;
1357 struct rpc_cred *cred;
1358 fmode_t fmode = flags_to_mode(open_flags);
1359
1360 cred = rpc_lookup_cred();
1361 if (IS_ERR(cred))
1362 return ERR_CAST(cred);
1363 ctx = alloc_nfs_open_context(dentry, cred, fmode);
1364 put_rpccred(cred);
1365 if (ctx == NULL)
1366 return ERR_PTR(-ENOMEM);
1367 return ctx;
1368 }
1369
1370 static int do_open(struct inode *inode, struct file *filp)
1371 {
1372 nfs_fscache_set_inode_cookie(inode, filp);
1373 return 0;
1374 }
1375
1376 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1377 {
1378 struct file *filp;
1379 int ret = 0;
1380
1381 /* If the open_intent is for execute, we have an extra check to make */
1382 if (ctx->mode & FMODE_EXEC) {
1383 ret = nfs_may_open(ctx->dentry->d_inode,
1384 ctx->cred,
1385 nd->intent.open.flags);
1386 if (ret < 0)
1387 goto out;
1388 }
1389 filp = lookup_instantiate_filp(nd, ctx->dentry, do_open);
1390 if (IS_ERR(filp))
1391 ret = PTR_ERR(filp);
1392 else
1393 nfs_file_set_open_context(filp, ctx);
1394 out:
1395 put_nfs_open_context(ctx);
1396 return ret;
1397 }
1398
1399 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1400 {
1401 struct nfs_open_context *ctx;
1402 struct iattr attr;
1403 struct dentry *res = NULL;
1404 struct inode *inode;
1405 int open_flags;
1406 int err;
1407
1408 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1409 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1410
1411 /* Check that we are indeed trying to open this file */
1412 if (!is_atomic_open(nd))
1413 goto no_open;
1414
1415 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1416 res = ERR_PTR(-ENAMETOOLONG);
1417 goto out;
1418 }
1419
1420 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1421 * the dentry. */
1422 if (nd->flags & LOOKUP_EXCL) {
1423 d_instantiate(dentry, NULL);
1424 goto out;
1425 }
1426
1427 open_flags = nd->intent.open.flags;
1428
1429 ctx = create_nfs_open_context(dentry, open_flags);
1430 res = ERR_CAST(ctx);
1431 if (IS_ERR(ctx))
1432 goto out;
1433
1434 if (nd->flags & LOOKUP_CREATE) {
1435 attr.ia_mode = nd->intent.open.create_mode;
1436 attr.ia_valid = ATTR_MODE;
1437 attr.ia_mode &= ~current_umask();
1438 } else {
1439 open_flags &= ~(O_EXCL | O_CREAT);
1440 attr.ia_valid = 0;
1441 }
1442
1443 /* Open the file on the server */
1444 nfs_block_sillyrename(dentry->d_parent);
1445 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1446 if (IS_ERR(inode)) {
1447 nfs_unblock_sillyrename(dentry->d_parent);
1448 put_nfs_open_context(ctx);
1449 switch (PTR_ERR(inode)) {
1450 /* Make a negative dentry */
1451 case -ENOENT:
1452 d_add(dentry, NULL);
1453 res = NULL;
1454 goto out;
1455 /* This turned out not to be a regular file */
1456 case -ENOTDIR:
1457 goto no_open;
1458 case -ELOOP:
1459 if (!(nd->intent.open.flags & O_NOFOLLOW))
1460 goto no_open;
1461 /* case -EISDIR: */
1462 /* case -EINVAL: */
1463 default:
1464 res = ERR_CAST(inode);
1465 goto out;
1466 }
1467 }
1468 res = d_add_unique(dentry, inode);
1469 nfs_unblock_sillyrename(dentry->d_parent);
1470 if (res != NULL) {
1471 dput(ctx->dentry);
1472 ctx->dentry = dget(res);
1473 dentry = res;
1474 }
1475 err = nfs_intent_set_file(nd, ctx);
1476 if (err < 0) {
1477 if (res != NULL)
1478 dput(res);
1479 return ERR_PTR(err);
1480 }
1481 out:
1482 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1483 return res;
1484 no_open:
1485 return nfs_lookup(dir, dentry, nd);
1486 }
1487
1488 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1489 {
1490 struct dentry *parent = NULL;
1491 struct inode *inode;
1492 struct inode *dir;
1493 struct nfs_open_context *ctx;
1494 int openflags, ret = 0;
1495
1496 if (nd->flags & LOOKUP_RCU)
1497 return -ECHILD;
1498
1499 inode = dentry->d_inode;
1500 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1501 goto no_open;
1502
1503 parent = dget_parent(dentry);
1504 dir = parent->d_inode;
1505
1506 /* We can't create new files in nfs_open_revalidate(), so we
1507 * optimize away revalidation of negative dentries.
1508 */
1509 if (inode == NULL) {
1510 if (!nfs_neg_need_reval(dir, dentry, nd))
1511 ret = 1;
1512 goto out;
1513 }
1514
1515 /* NFS only supports OPEN on regular files */
1516 if (!S_ISREG(inode->i_mode))
1517 goto no_open_dput;
1518 openflags = nd->intent.open.flags;
1519 /* We cannot do exclusive creation on a positive dentry */
1520 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1521 goto no_open_dput;
1522 /* We can't create new files, or truncate existing ones here */
1523 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1524
1525 ctx = create_nfs_open_context(dentry, openflags);
1526 ret = PTR_ERR(ctx);
1527 if (IS_ERR(ctx))
1528 goto out;
1529 /*
1530 * Note: we're not holding inode->i_mutex and so may be racing with
1531 * operations that change the directory. We therefore save the
1532 * change attribute *before* we do the RPC call.
1533 */
1534 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1535 if (IS_ERR(inode)) {
1536 ret = PTR_ERR(inode);
1537 switch (ret) {
1538 case -EPERM:
1539 case -EACCES:
1540 case -EDQUOT:
1541 case -ENOSPC:
1542 case -EROFS:
1543 goto out_put_ctx;
1544 default:
1545 goto out_drop;
1546 }
1547 }
1548 iput(inode);
1549 if (inode != dentry->d_inode)
1550 goto out_drop;
1551
1552 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1553 ret = nfs_intent_set_file(nd, ctx);
1554 if (ret >= 0)
1555 ret = 1;
1556 out:
1557 dput(parent);
1558 return ret;
1559 out_drop:
1560 d_drop(dentry);
1561 ret = 0;
1562 out_put_ctx:
1563 put_nfs_open_context(ctx);
1564 goto out;
1565
1566 no_open_dput:
1567 dput(parent);
1568 no_open:
1569 return nfs_lookup_revalidate(dentry, nd);
1570 }
1571
1572 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1573 struct nameidata *nd)
1574 {
1575 struct nfs_open_context *ctx = NULL;
1576 struct iattr attr;
1577 int error;
1578 int open_flags = O_CREAT|O_EXCL;
1579
1580 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1581 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1582
1583 attr.ia_mode = mode;
1584 attr.ia_valid = ATTR_MODE;
1585
1586 if (nd)
1587 open_flags = nd->intent.open.flags;
1588
1589 ctx = create_nfs_open_context(dentry, open_flags);
1590 error = PTR_ERR(ctx);
1591 if (IS_ERR(ctx))
1592 goto out_err_drop;
1593
1594 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1595 if (error != 0)
1596 goto out_put_ctx;
1597 if (nd) {
1598 error = nfs_intent_set_file(nd, ctx);
1599 if (error < 0)
1600 goto out_err;
1601 } else {
1602 put_nfs_open_context(ctx);
1603 }
1604 return 0;
1605 out_put_ctx:
1606 put_nfs_open_context(ctx);
1607 out_err_drop:
1608 d_drop(dentry);
1609 out_err:
1610 return error;
1611 }
1612
1613 #endif /* CONFIG_NFSV4 */
1614
1615 /*
1616 * Code common to create, mkdir, and mknod.
1617 */
1618 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1619 struct nfs_fattr *fattr)
1620 {
1621 struct dentry *parent = dget_parent(dentry);
1622 struct inode *dir = parent->d_inode;
1623 struct inode *inode;
1624 int error = -EACCES;
1625
1626 d_drop(dentry);
1627
1628 /* We may have been initialized further down */
1629 if (dentry->d_inode)
1630 goto out;
1631 if (fhandle->size == 0) {
1632 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1633 if (error)
1634 goto out_error;
1635 }
1636 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1637 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1638 struct nfs_server *server = NFS_SB(dentry->d_sb);
1639 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1640 if (error < 0)
1641 goto out_error;
1642 }
1643 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1644 error = PTR_ERR(inode);
1645 if (IS_ERR(inode))
1646 goto out_error;
1647 d_add(dentry, inode);
1648 out:
1649 dput(parent);
1650 return 0;
1651 out_error:
1652 nfs_mark_for_revalidate(dir);
1653 dput(parent);
1654 return error;
1655 }
1656
1657 /*
1658 * Following a failed create operation, we drop the dentry rather
1659 * than retain a negative dentry. This avoids a problem in the event
1660 * that the operation succeeded on the server, but an error in the
1661 * reply path made it appear to have failed.
1662 */
1663 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1664 struct nameidata *nd)
1665 {
1666 struct iattr attr;
1667 int error;
1668 int open_flags = O_CREAT|O_EXCL;
1669
1670 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1671 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1672
1673 attr.ia_mode = mode;
1674 attr.ia_valid = ATTR_MODE;
1675
1676 if (nd)
1677 open_flags = nd->intent.open.flags;
1678
1679 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1680 if (error != 0)
1681 goto out_err;
1682 return 0;
1683 out_err:
1684 d_drop(dentry);
1685 return error;
1686 }
1687
1688 /*
1689 * See comments for nfs_proc_create regarding failed operations.
1690 */
1691 static int
1692 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1693 {
1694 struct iattr attr;
1695 int status;
1696
1697 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1698 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1699
1700 if (!new_valid_dev(rdev))
1701 return -EINVAL;
1702
1703 attr.ia_mode = mode;
1704 attr.ia_valid = ATTR_MODE;
1705
1706 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1707 if (status != 0)
1708 goto out_err;
1709 return 0;
1710 out_err:
1711 d_drop(dentry);
1712 return status;
1713 }
1714
1715 /*
1716 * See comments for nfs_proc_create regarding failed operations.
1717 */
1718 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1719 {
1720 struct iattr attr;
1721 int error;
1722
1723 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1724 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1725
1726 attr.ia_valid = ATTR_MODE;
1727 attr.ia_mode = mode | S_IFDIR;
1728
1729 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1730 if (error != 0)
1731 goto out_err;
1732 return 0;
1733 out_err:
1734 d_drop(dentry);
1735 return error;
1736 }
1737
1738 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1739 {
1740 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1741 d_delete(dentry);
1742 }
1743
1744 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1745 {
1746 int error;
1747
1748 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1749 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1750
1751 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1752 /* Ensure the VFS deletes this inode */
1753 if (error == 0 && dentry->d_inode != NULL)
1754 clear_nlink(dentry->d_inode);
1755 else if (error == -ENOENT)
1756 nfs_dentry_handle_enoent(dentry);
1757
1758 return error;
1759 }
1760
1761 /*
1762 * Remove a file after making sure there are no pending writes,
1763 * and after checking that the file has only one user.
1764 *
1765 * We invalidate the attribute cache and free the inode prior to the operation
1766 * to avoid possible races if the server reuses the inode.
1767 */
1768 static int nfs_safe_remove(struct dentry *dentry)
1769 {
1770 struct inode *dir = dentry->d_parent->d_inode;
1771 struct inode *inode = dentry->d_inode;
1772 int error = -EBUSY;
1773
1774 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1775 dentry->d_parent->d_name.name, dentry->d_name.name);
1776
1777 /* If the dentry was sillyrenamed, we simply call d_delete() */
1778 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1779 error = 0;
1780 goto out;
1781 }
1782
1783 if (inode != NULL) {
1784 nfs_inode_return_delegation(inode);
1785 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1786 /* The VFS may want to delete this inode */
1787 if (error == 0)
1788 nfs_drop_nlink(inode);
1789 nfs_mark_for_revalidate(inode);
1790 } else
1791 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1792 if (error == -ENOENT)
1793 nfs_dentry_handle_enoent(dentry);
1794 out:
1795 return error;
1796 }
1797
1798 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1799 * belongs to an active ".nfs..." file and we return -EBUSY.
1800 *
1801 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1802 */
1803 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1804 {
1805 int error;
1806 int need_rehash = 0;
1807
1808 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1809 dir->i_ino, dentry->d_name.name);
1810
1811 spin_lock(&dentry->d_lock);
1812 if (dentry->d_count > 1) {
1813 spin_unlock(&dentry->d_lock);
1814 /* Start asynchronous writeout of the inode */
1815 write_inode_now(dentry->d_inode, 0);
1816 error = nfs_sillyrename(dir, dentry);
1817 return error;
1818 }
1819 if (!d_unhashed(dentry)) {
1820 __d_drop(dentry);
1821 need_rehash = 1;
1822 }
1823 spin_unlock(&dentry->d_lock);
1824 error = nfs_safe_remove(dentry);
1825 if (!error || error == -ENOENT) {
1826 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1827 } else if (need_rehash)
1828 d_rehash(dentry);
1829 return error;
1830 }
1831
1832 /*
1833 * To create a symbolic link, most file systems instantiate a new inode,
1834 * add a page to it containing the path, then write it out to the disk
1835 * using prepare_write/commit_write.
1836 *
1837 * Unfortunately the NFS client can't create the in-core inode first
1838 * because it needs a file handle to create an in-core inode (see
1839 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1840 * symlink request has completed on the server.
1841 *
1842 * So instead we allocate a raw page, copy the symname into it, then do
1843 * the SYMLINK request with the page as the buffer. If it succeeds, we
1844 * now have a new file handle and can instantiate an in-core NFS inode
1845 * and move the raw page into its mapping.
1846 */
1847 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1848 {
1849 struct pagevec lru_pvec;
1850 struct page *page;
1851 char *kaddr;
1852 struct iattr attr;
1853 unsigned int pathlen = strlen(symname);
1854 int error;
1855
1856 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1857 dir->i_ino, dentry->d_name.name, symname);
1858
1859 if (pathlen > PAGE_SIZE)
1860 return -ENAMETOOLONG;
1861
1862 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1863 attr.ia_valid = ATTR_MODE;
1864
1865 page = alloc_page(GFP_HIGHUSER);
1866 if (!page)
1867 return -ENOMEM;
1868
1869 kaddr = kmap_atomic(page, KM_USER0);
1870 memcpy(kaddr, symname, pathlen);
1871 if (pathlen < PAGE_SIZE)
1872 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1873 kunmap_atomic(kaddr, KM_USER0);
1874
1875 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1876 if (error != 0) {
1877 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1878 dir->i_sb->s_id, dir->i_ino,
1879 dentry->d_name.name, symname, error);
1880 d_drop(dentry);
1881 __free_page(page);
1882 return error;
1883 }
1884
1885 /*
1886 * No big deal if we can't add this page to the page cache here.
1887 * READLINK will get the missing page from the server if needed.
1888 */
1889 pagevec_init(&lru_pvec, 0);
1890 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1891 GFP_KERNEL)) {
1892 pagevec_add(&lru_pvec, page);
1893 pagevec_lru_add_file(&lru_pvec);
1894 SetPageUptodate(page);
1895 unlock_page(page);
1896 } else
1897 __free_page(page);
1898
1899 return 0;
1900 }
1901
1902 static int
1903 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1904 {
1905 struct inode *inode = old_dentry->d_inode;
1906 int error;
1907
1908 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1909 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1910 dentry->d_parent->d_name.name, dentry->d_name.name);
1911
1912 nfs_inode_return_delegation(inode);
1913
1914 d_drop(dentry);
1915 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1916 if (error == 0) {
1917 ihold(inode);
1918 d_add(dentry, inode);
1919 }
1920 return error;
1921 }
1922
1923 /*
1924 * RENAME
1925 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1926 * different file handle for the same inode after a rename (e.g. when
1927 * moving to a different directory). A fail-safe method to do so would
1928 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1929 * rename the old file using the sillyrename stuff. This way, the original
1930 * file in old_dir will go away when the last process iput()s the inode.
1931 *
1932 * FIXED.
1933 *
1934 * It actually works quite well. One needs to have the possibility for
1935 * at least one ".nfs..." file in each directory the file ever gets
1936 * moved or linked to which happens automagically with the new
1937 * implementation that only depends on the dcache stuff instead of
1938 * using the inode layer
1939 *
1940 * Unfortunately, things are a little more complicated than indicated
1941 * above. For a cross-directory move, we want to make sure we can get
1942 * rid of the old inode after the operation. This means there must be
1943 * no pending writes (if it's a file), and the use count must be 1.
1944 * If these conditions are met, we can drop the dentries before doing
1945 * the rename.
1946 */
1947 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1948 struct inode *new_dir, struct dentry *new_dentry)
1949 {
1950 struct inode *old_inode = old_dentry->d_inode;
1951 struct inode *new_inode = new_dentry->d_inode;
1952 struct dentry *dentry = NULL, *rehash = NULL;
1953 int error = -EBUSY;
1954
1955 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1956 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1957 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1958 new_dentry->d_count);
1959
1960 /*
1961 * For non-directories, check whether the target is busy and if so,
1962 * make a copy of the dentry and then do a silly-rename. If the
1963 * silly-rename succeeds, the copied dentry is hashed and becomes
1964 * the new target.
1965 */
1966 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1967 /*
1968 * To prevent any new references to the target during the
1969 * rename, we unhash the dentry in advance.
1970 */
1971 if (!d_unhashed(new_dentry)) {
1972 d_drop(new_dentry);
1973 rehash = new_dentry;
1974 }
1975
1976 if (new_dentry->d_count > 2) {
1977 int err;
1978
1979 /* copy the target dentry's name */
1980 dentry = d_alloc(new_dentry->d_parent,
1981 &new_dentry->d_name);
1982 if (!dentry)
1983 goto out;
1984
1985 /* silly-rename the existing target ... */
1986 err = nfs_sillyrename(new_dir, new_dentry);
1987 if (err)
1988 goto out;
1989
1990 new_dentry = dentry;
1991 rehash = NULL;
1992 new_inode = NULL;
1993 }
1994 }
1995
1996 nfs_inode_return_delegation(old_inode);
1997 if (new_inode != NULL)
1998 nfs_inode_return_delegation(new_inode);
1999
2000 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
2001 new_dir, &new_dentry->d_name);
2002 nfs_mark_for_revalidate(old_inode);
2003 out:
2004 if (rehash)
2005 d_rehash(rehash);
2006 if (!error) {
2007 if (new_inode != NULL)
2008 nfs_drop_nlink(new_inode);
2009 d_move(old_dentry, new_dentry);
2010 nfs_set_verifier(new_dentry,
2011 nfs_save_change_attribute(new_dir));
2012 } else if (error == -ENOENT)
2013 nfs_dentry_handle_enoent(old_dentry);
2014
2015 /* new dentry created? */
2016 if (dentry)
2017 dput(dentry);
2018 return error;
2019 }
2020
2021 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2022 static LIST_HEAD(nfs_access_lru_list);
2023 static atomic_long_t nfs_access_nr_entries;
2024
2025 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2026 {
2027 put_rpccred(entry->cred);
2028 kfree(entry);
2029 smp_mb__before_atomic_dec();
2030 atomic_long_dec(&nfs_access_nr_entries);
2031 smp_mb__after_atomic_dec();
2032 }
2033
2034 static void nfs_access_free_list(struct list_head *head)
2035 {
2036 struct nfs_access_entry *cache;
2037
2038 while (!list_empty(head)) {
2039 cache = list_entry(head->next, struct nfs_access_entry, lru);
2040 list_del(&cache->lru);
2041 nfs_access_free_entry(cache);
2042 }
2043 }
2044
2045 int nfs_access_cache_shrinker(struct shrinker *shrink,
2046 struct shrink_control *sc)
2047 {
2048 LIST_HEAD(head);
2049 struct nfs_inode *nfsi, *next;
2050 struct nfs_access_entry *cache;
2051 int nr_to_scan = sc->nr_to_scan;
2052 gfp_t gfp_mask = sc->gfp_mask;
2053
2054 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2055 return (nr_to_scan == 0) ? 0 : -1;
2056
2057 spin_lock(&nfs_access_lru_lock);
2058 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2059 struct inode *inode;
2060
2061 if (nr_to_scan-- == 0)
2062 break;
2063 inode = &nfsi->vfs_inode;
2064 spin_lock(&inode->i_lock);
2065 if (list_empty(&nfsi->access_cache_entry_lru))
2066 goto remove_lru_entry;
2067 cache = list_entry(nfsi->access_cache_entry_lru.next,
2068 struct nfs_access_entry, lru);
2069 list_move(&cache->lru, &head);
2070 rb_erase(&cache->rb_node, &nfsi->access_cache);
2071 if (!list_empty(&nfsi->access_cache_entry_lru))
2072 list_move_tail(&nfsi->access_cache_inode_lru,
2073 &nfs_access_lru_list);
2074 else {
2075 remove_lru_entry:
2076 list_del_init(&nfsi->access_cache_inode_lru);
2077 smp_mb__before_clear_bit();
2078 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2079 smp_mb__after_clear_bit();
2080 }
2081 spin_unlock(&inode->i_lock);
2082 }
2083 spin_unlock(&nfs_access_lru_lock);
2084 nfs_access_free_list(&head);
2085 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2086 }
2087
2088 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2089 {
2090 struct rb_root *root_node = &nfsi->access_cache;
2091 struct rb_node *n;
2092 struct nfs_access_entry *entry;
2093
2094 /* Unhook entries from the cache */
2095 while ((n = rb_first(root_node)) != NULL) {
2096 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2097 rb_erase(n, root_node);
2098 list_move(&entry->lru, head);
2099 }
2100 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2101 }
2102
2103 void nfs_access_zap_cache(struct inode *inode)
2104 {
2105 LIST_HEAD(head);
2106
2107 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2108 return;
2109 /* Remove from global LRU init */
2110 spin_lock(&nfs_access_lru_lock);
2111 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2112 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2113
2114 spin_lock(&inode->i_lock);
2115 __nfs_access_zap_cache(NFS_I(inode), &head);
2116 spin_unlock(&inode->i_lock);
2117 spin_unlock(&nfs_access_lru_lock);
2118 nfs_access_free_list(&head);
2119 }
2120
2121 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2122 {
2123 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2124 struct nfs_access_entry *entry;
2125
2126 while (n != NULL) {
2127 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2128
2129 if (cred < entry->cred)
2130 n = n->rb_left;
2131 else if (cred > entry->cred)
2132 n = n->rb_right;
2133 else
2134 return entry;
2135 }
2136 return NULL;
2137 }
2138
2139 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2140 {
2141 struct nfs_inode *nfsi = NFS_I(inode);
2142 struct nfs_access_entry *cache;
2143 int err = -ENOENT;
2144
2145 spin_lock(&inode->i_lock);
2146 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2147 goto out_zap;
2148 cache = nfs_access_search_rbtree(inode, cred);
2149 if (cache == NULL)
2150 goto out;
2151 if (!nfs_have_delegated_attributes(inode) &&
2152 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2153 goto out_stale;
2154 res->jiffies = cache->jiffies;
2155 res->cred = cache->cred;
2156 res->mask = cache->mask;
2157 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2158 err = 0;
2159 out:
2160 spin_unlock(&inode->i_lock);
2161 return err;
2162 out_stale:
2163 rb_erase(&cache->rb_node, &nfsi->access_cache);
2164 list_del(&cache->lru);
2165 spin_unlock(&inode->i_lock);
2166 nfs_access_free_entry(cache);
2167 return -ENOENT;
2168 out_zap:
2169 spin_unlock(&inode->i_lock);
2170 nfs_access_zap_cache(inode);
2171 return -ENOENT;
2172 }
2173
2174 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2175 {
2176 struct nfs_inode *nfsi = NFS_I(inode);
2177 struct rb_root *root_node = &nfsi->access_cache;
2178 struct rb_node **p = &root_node->rb_node;
2179 struct rb_node *parent = NULL;
2180 struct nfs_access_entry *entry;
2181
2182 spin_lock(&inode->i_lock);
2183 while (*p != NULL) {
2184 parent = *p;
2185 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2186
2187 if (set->cred < entry->cred)
2188 p = &parent->rb_left;
2189 else if (set->cred > entry->cred)
2190 p = &parent->rb_right;
2191 else
2192 goto found;
2193 }
2194 rb_link_node(&set->rb_node, parent, p);
2195 rb_insert_color(&set->rb_node, root_node);
2196 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2197 spin_unlock(&inode->i_lock);
2198 return;
2199 found:
2200 rb_replace_node(parent, &set->rb_node, root_node);
2201 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2202 list_del(&entry->lru);
2203 spin_unlock(&inode->i_lock);
2204 nfs_access_free_entry(entry);
2205 }
2206
2207 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2208 {
2209 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2210 if (cache == NULL)
2211 return;
2212 RB_CLEAR_NODE(&cache->rb_node);
2213 cache->jiffies = set->jiffies;
2214 cache->cred = get_rpccred(set->cred);
2215 cache->mask = set->mask;
2216
2217 nfs_access_add_rbtree(inode, cache);
2218
2219 /* Update accounting */
2220 smp_mb__before_atomic_inc();
2221 atomic_long_inc(&nfs_access_nr_entries);
2222 smp_mb__after_atomic_inc();
2223
2224 /* Add inode to global LRU list */
2225 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2226 spin_lock(&nfs_access_lru_lock);
2227 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2228 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2229 &nfs_access_lru_list);
2230 spin_unlock(&nfs_access_lru_lock);
2231 }
2232 }
2233
2234 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2235 {
2236 struct nfs_access_entry cache;
2237 int status;
2238
2239 status = nfs_access_get_cached(inode, cred, &cache);
2240 if (status == 0)
2241 goto out;
2242
2243 /* Be clever: ask server to check for all possible rights */
2244 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2245 cache.cred = cred;
2246 cache.jiffies = jiffies;
2247 status = NFS_PROTO(inode)->access(inode, &cache);
2248 if (status != 0) {
2249 if (status == -ESTALE) {
2250 nfs_zap_caches(inode);
2251 if (!S_ISDIR(inode->i_mode))
2252 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2253 }
2254 return status;
2255 }
2256 nfs_access_add_cache(inode, &cache);
2257 out:
2258 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2259 return 0;
2260 return -EACCES;
2261 }
2262
2263 static int nfs_open_permission_mask(int openflags)
2264 {
2265 int mask = 0;
2266
2267 if ((openflags & O_ACCMODE) != O_WRONLY)
2268 mask |= MAY_READ;
2269 if ((openflags & O_ACCMODE) != O_RDONLY)
2270 mask |= MAY_WRITE;
2271 if (openflags & __FMODE_EXEC)
2272 mask |= MAY_EXEC;
2273 return mask;
2274 }
2275
2276 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2277 {
2278 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2279 }
2280
2281 int nfs_permission(struct inode *inode, int mask)
2282 {
2283 struct rpc_cred *cred;
2284 int res = 0;
2285
2286 if (mask & MAY_NOT_BLOCK)
2287 return -ECHILD;
2288
2289 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2290
2291 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2292 goto out;
2293 /* Is this sys_access() ? */
2294 if (mask & (MAY_ACCESS | MAY_CHDIR))
2295 goto force_lookup;
2296
2297 switch (inode->i_mode & S_IFMT) {
2298 case S_IFLNK:
2299 goto out;
2300 case S_IFREG:
2301 /* NFSv4 has atomic_open... */
2302 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2303 && (mask & MAY_OPEN)
2304 && !(mask & MAY_EXEC))
2305 goto out;
2306 break;
2307 case S_IFDIR:
2308 /*
2309 * Optimize away all write operations, since the server
2310 * will check permissions when we perform the op.
2311 */
2312 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2313 goto out;
2314 }
2315
2316 force_lookup:
2317 if (!NFS_PROTO(inode)->access)
2318 goto out_notsup;
2319
2320 cred = rpc_lookup_cred();
2321 if (!IS_ERR(cred)) {
2322 res = nfs_do_access(inode, cred, mask);
2323 put_rpccred(cred);
2324 } else
2325 res = PTR_ERR(cred);
2326 out:
2327 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2328 res = -EACCES;
2329
2330 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2331 inode->i_sb->s_id, inode->i_ino, mask, res);
2332 return res;
2333 out_notsup:
2334 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2335 if (res == 0)
2336 res = generic_permission(inode, mask);
2337 goto out;
2338 }
2339
2340 /*
2341 * Local variables:
2342 * version-control: t
2343 * kept-new-versions: 5
2344 * End:
2345 */