blob: 2fc99877cb0d1b55c74329a40cd23f2eaae91e13 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* dir.c: AFS filesystem directory handling
2 *
3 * Copyright (C) 2002 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#include <linux/kernel.h>
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/sched.h>
16#include <linux/slab.h>
17#include <linux/fs.h>
18#include <linux/pagemap.h>
19#include <linux/smp_lock.h>
20#include "vnode.h"
21#include "volume.h"
22#include <rxrpc/call.h>
23#include "super.h"
24#include "internal.h"
25
26static struct dentry *afs_dir_lookup(struct inode *dir, struct dentry *dentry,
27 struct nameidata *nd);
28static int afs_dir_open(struct inode *inode, struct file *file);
29static int afs_dir_readdir(struct file *file, void *dirent, filldir_t filldir);
30static int afs_d_revalidate(struct dentry *dentry, struct nameidata *nd);
31static int afs_d_delete(struct dentry *dentry);
32static int afs_dir_lookup_filldir(void *_cookie, const char *name, int nlen,
33 loff_t fpos, ino_t ino, unsigned dtype);
34
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080035const struct file_operations afs_dir_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 .open = afs_dir_open,
37 .readdir = afs_dir_readdir,
38};
39
40struct inode_operations afs_dir_inode_operations = {
41 .lookup = afs_dir_lookup,
42 .getattr = afs_inode_getattr,
43#if 0 /* TODO */
44 .create = afs_dir_create,
45 .link = afs_dir_link,
46 .unlink = afs_dir_unlink,
47 .symlink = afs_dir_symlink,
48 .mkdir = afs_dir_mkdir,
49 .rmdir = afs_dir_rmdir,
50 .mknod = afs_dir_mknod,
51 .rename = afs_dir_rename,
52#endif
53};
54
55static struct dentry_operations afs_fs_dentry_operations = {
56 .d_revalidate = afs_d_revalidate,
57 .d_delete = afs_d_delete,
58};
59
60#define AFS_DIR_HASHTBL_SIZE 128
61#define AFS_DIR_DIRENT_SIZE 32
62#define AFS_DIRENT_PER_BLOCK 64
63
64union afs_dirent {
65 struct {
66 uint8_t valid;
67 uint8_t unused[1];
68 __be16 hash_next;
69 __be32 vnode;
70 __be32 unique;
71 uint8_t name[16];
72 uint8_t overflow[4]; /* if any char of the name (inc
73 * NUL) reaches here, consume
74 * the next dirent too */
75 } u;
76 uint8_t extended_name[32];
77};
78
79/* AFS directory page header (one at the beginning of every 2048-byte chunk) */
80struct afs_dir_pagehdr {
81 __be16 npages;
82 __be16 magic;
83#define AFS_DIR_MAGIC htons(1234)
84 uint8_t nentries;
85 uint8_t bitmap[8];
86 uint8_t pad[19];
87};
88
89/* directory block layout */
90union afs_dir_block {
91
92 struct afs_dir_pagehdr pagehdr;
93
94 struct {
95 struct afs_dir_pagehdr pagehdr;
96 uint8_t alloc_ctrs[128];
97 /* dir hash table */
98 uint16_t hashtable[AFS_DIR_HASHTBL_SIZE];
99 } hdr;
100
101 union afs_dirent dirents[AFS_DIRENT_PER_BLOCK];
102};
103
104/* layout on a linux VM page */
105struct afs_dir_page {
106 union afs_dir_block blocks[PAGE_SIZE / sizeof(union afs_dir_block)];
107};
108
109struct afs_dir_lookup_cookie {
110 struct afs_fid fid;
111 const char *name;
112 size_t nlen;
113 int found;
114};
115
116/*****************************************************************************/
117/*
118 * check that a directory page is valid
119 */
120static inline void afs_dir_check_page(struct inode *dir, struct page *page)
121{
122 struct afs_dir_page *dbuf;
123 loff_t latter;
124 int tmp, qty;
125
126#if 0
127 /* check the page count */
128 qty = desc.size / sizeof(dbuf->blocks[0]);
129 if (qty == 0)
130 goto error;
131
132 if (page->index==0 && qty!=ntohs(dbuf->blocks[0].pagehdr.npages)) {
133 printk("kAFS: %s(%lu): wrong number of dir blocks %d!=%hu\n",
134 __FUNCTION__,dir->i_ino,qty,ntohs(dbuf->blocks[0].pagehdr.npages));
135 goto error;
136 }
137#endif
138
139 /* determine how many magic numbers there should be in this page */
Andrew Morton54b21a72006-01-08 01:03:05 -0800140 latter = dir->i_size - page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 if (latter >= PAGE_SIZE)
142 qty = PAGE_SIZE;
143 else
144 qty = latter;
145 qty /= sizeof(union afs_dir_block);
146
147 /* check them */
148 dbuf = page_address(page);
149 for (tmp = 0; tmp < qty; tmp++) {
150 if (dbuf->blocks[tmp].pagehdr.magic != AFS_DIR_MAGIC) {
151 printk("kAFS: %s(%lu): bad magic %d/%d is %04hx\n",
152 __FUNCTION__, dir->i_ino, tmp, qty,
153 ntohs(dbuf->blocks[tmp].pagehdr.magic));
154 goto error;
155 }
156 }
157
158 SetPageChecked(page);
159 return;
160
161 error:
162 SetPageChecked(page);
163 SetPageError(page);
164
165} /* end afs_dir_check_page() */
166
167/*****************************************************************************/
168/*
169 * discard a page cached in the pagecache
170 */
171static inline void afs_dir_put_page(struct page *page)
172{
173 kunmap(page);
174 page_cache_release(page);
175
176} /* end afs_dir_put_page() */
177
178/*****************************************************************************/
179/*
180 * get a page into the pagecache
181 */
182static struct page *afs_dir_get_page(struct inode *dir, unsigned long index)
183{
184 struct page *page;
185
186 _enter("{%lu},%lu", dir->i_ino, index);
187
Pekka Enberg090d2b12006-06-23 02:05:08 -0700188 page = read_mapping_page(dir->i_mapping, index, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 if (!IS_ERR(page)) {
190 wait_on_page_locked(page);
191 kmap(page);
192 if (!PageUptodate(page))
193 goto fail;
194 if (!PageChecked(page))
195 afs_dir_check_page(dir, page);
196 if (PageError(page))
197 goto fail;
198 }
199 return page;
200
201 fail:
202 afs_dir_put_page(page);
203 return ERR_PTR(-EIO);
204} /* end afs_dir_get_page() */
205
206/*****************************************************************************/
207/*
208 * open an AFS directory file
209 */
210static int afs_dir_open(struct inode *inode, struct file *file)
211{
212 _enter("{%lu}", inode->i_ino);
213
214 BUG_ON(sizeof(union afs_dir_block) != 2048);
215 BUG_ON(sizeof(union afs_dirent) != 32);
216
217 if (AFS_FS_I(inode)->flags & AFS_VNODE_DELETED)
218 return -ENOENT;
219
220 _leave(" = 0");
221 return 0;
222
223} /* end afs_dir_open() */
224
225/*****************************************************************************/
226/*
227 * deal with one block in an AFS directory
228 */
229static int afs_dir_iterate_block(unsigned *fpos,
230 union afs_dir_block *block,
231 unsigned blkoff,
232 void *cookie,
233 filldir_t filldir)
234{
235 union afs_dirent *dire;
236 unsigned offset, next, curr;
237 size_t nlen;
238 int tmp, ret;
239
240 _enter("%u,%x,%p,,",*fpos,blkoff,block);
241
242 curr = (*fpos - blkoff) / sizeof(union afs_dirent);
243
244 /* walk through the block, an entry at a time */
245 for (offset = AFS_DIRENT_PER_BLOCK - block->pagehdr.nentries;
246 offset < AFS_DIRENT_PER_BLOCK;
247 offset = next
248 ) {
249 next = offset + 1;
250
251 /* skip entries marked unused in the bitmap */
252 if (!(block->pagehdr.bitmap[offset / 8] &
253 (1 << (offset % 8)))) {
254 _debug("ENT[%Zu.%u]: unused\n",
255 blkoff / sizeof(union afs_dir_block), offset);
256 if (offset >= curr)
257 *fpos = blkoff +
258 next * sizeof(union afs_dirent);
259 continue;
260 }
261
262 /* got a valid entry */
263 dire = &block->dirents[offset];
264 nlen = strnlen(dire->u.name,
265 sizeof(*block) -
266 offset * sizeof(union afs_dirent));
267
268 _debug("ENT[%Zu.%u]: %s %Zu \"%s\"\n",
269 blkoff / sizeof(union afs_dir_block), offset,
270 (offset < curr ? "skip" : "fill"),
271 nlen, dire->u.name);
272
273 /* work out where the next possible entry is */
274 for (tmp = nlen; tmp > 15; tmp -= sizeof(union afs_dirent)) {
275 if (next >= AFS_DIRENT_PER_BLOCK) {
276 _debug("ENT[%Zu.%u]:"
277 " %u travelled beyond end dir block"
278 " (len %u/%Zu)\n",
279 blkoff / sizeof(union afs_dir_block),
280 offset, next, tmp, nlen);
281 return -EIO;
282 }
283 if (!(block->pagehdr.bitmap[next / 8] &
284 (1 << (next % 8)))) {
285 _debug("ENT[%Zu.%u]:"
286 " %u unmarked extension (len %u/%Zu)\n",
287 blkoff / sizeof(union afs_dir_block),
288 offset, next, tmp, nlen);
289 return -EIO;
290 }
291
292 _debug("ENT[%Zu.%u]: ext %u/%Zu\n",
293 blkoff / sizeof(union afs_dir_block),
294 next, tmp, nlen);
295 next++;
296 }
297
298 /* skip if starts before the current position */
299 if (offset < curr)
300 continue;
301
302 /* found the next entry */
303 ret = filldir(cookie,
304 dire->u.name,
305 nlen,
306 blkoff + offset * sizeof(union afs_dirent),
307 ntohl(dire->u.vnode),
308 filldir == afs_dir_lookup_filldir ?
309 ntohl(dire->u.unique) : DT_UNKNOWN);
310 if (ret < 0) {
311 _leave(" = 0 [full]");
312 return 0;
313 }
314
315 *fpos = blkoff + next * sizeof(union afs_dirent);
316 }
317
318 _leave(" = 1 [more]");
319 return 1;
320} /* end afs_dir_iterate_block() */
321
322/*****************************************************************************/
323/*
324 * read an AFS directory
325 */
326static int afs_dir_iterate(struct inode *dir, unsigned *fpos, void *cookie,
327 filldir_t filldir)
328{
329 union afs_dir_block *dblock;
330 struct afs_dir_page *dbuf;
331 struct page *page;
332 unsigned blkoff, limit;
333 int ret;
334
335 _enter("{%lu},%u,,", dir->i_ino, *fpos);
336
337 if (AFS_FS_I(dir)->flags & AFS_VNODE_DELETED) {
338 _leave(" = -ESTALE");
339 return -ESTALE;
340 }
341
342 /* round the file position up to the next entry boundary */
343 *fpos += sizeof(union afs_dirent) - 1;
344 *fpos &= ~(sizeof(union afs_dirent) - 1);
345
346 /* walk through the blocks in sequence */
347 ret = 0;
348 while (*fpos < dir->i_size) {
349 blkoff = *fpos & ~(sizeof(union afs_dir_block) - 1);
350
351 /* fetch the appropriate page from the directory */
352 page = afs_dir_get_page(dir, blkoff / PAGE_SIZE);
353 if (IS_ERR(page)) {
354 ret = PTR_ERR(page);
355 break;
356 }
357
358 limit = blkoff & ~(PAGE_SIZE - 1);
359
360 dbuf = page_address(page);
361
362 /* deal with the individual blocks stashed on this page */
363 do {
364 dblock = &dbuf->blocks[(blkoff % PAGE_SIZE) /
365 sizeof(union afs_dir_block)];
366 ret = afs_dir_iterate_block(fpos, dblock, blkoff,
367 cookie, filldir);
368 if (ret != 1) {
369 afs_dir_put_page(page);
370 goto out;
371 }
372
373 blkoff += sizeof(union afs_dir_block);
374
375 } while (*fpos < dir->i_size && blkoff < limit);
376
377 afs_dir_put_page(page);
378 ret = 0;
379 }
380
381 out:
382 _leave(" = %d", ret);
383 return ret;
384} /* end afs_dir_iterate() */
385
386/*****************************************************************************/
387/*
388 * read an AFS directory
389 */
390static int afs_dir_readdir(struct file *file, void *cookie, filldir_t filldir)
391{
392 unsigned fpos;
393 int ret;
394
395 _enter("{%Ld,{%lu}}", file->f_pos, file->f_dentry->d_inode->i_ino);
396
397 fpos = file->f_pos;
398 ret = afs_dir_iterate(file->f_dentry->d_inode, &fpos, cookie, filldir);
399 file->f_pos = fpos;
400
401 _leave(" = %d", ret);
402 return ret;
403} /* end afs_dir_readdir() */
404
405/*****************************************************************************/
406/*
407 * search the directory for a name
408 * - if afs_dir_iterate_block() spots this function, it'll pass the FID
409 * uniquifier through dtype
410 */
411static int afs_dir_lookup_filldir(void *_cookie, const char *name, int nlen,
412 loff_t fpos, ino_t ino, unsigned dtype)
413{
414 struct afs_dir_lookup_cookie *cookie = _cookie;
415
416 _enter("{%s,%Zu},%s,%u,,%lu,%u",
417 cookie->name, cookie->nlen, name, nlen, ino, dtype);
418
419 if (cookie->nlen != nlen || memcmp(cookie->name, name, nlen) != 0) {
420 _leave(" = 0 [no]");
421 return 0;
422 }
423
424 cookie->fid.vnode = ino;
425 cookie->fid.unique = dtype;
426 cookie->found = 1;
427
428 _leave(" = -1 [found]");
429 return -1;
430} /* end afs_dir_lookup_filldir() */
431
432/*****************************************************************************/
433/*
434 * look up an entry in a directory
435 */
436static struct dentry *afs_dir_lookup(struct inode *dir, struct dentry *dentry,
437 struct nameidata *nd)
438{
439 struct afs_dir_lookup_cookie cookie;
440 struct afs_super_info *as;
441 struct afs_vnode *vnode;
442 struct inode *inode;
443 unsigned fpos;
444 int ret;
445
446 _enter("{%lu},%p{%s}", dir->i_ino, dentry, dentry->d_name.name);
447
448 /* insanity checks first */
449 BUG_ON(sizeof(union afs_dir_block) != 2048);
450 BUG_ON(sizeof(union afs_dirent) != 32);
451
452 if (dentry->d_name.len > 255) {
453 _leave(" = -ENAMETOOLONG");
454 return ERR_PTR(-ENAMETOOLONG);
455 }
456
457 vnode = AFS_FS_I(dir);
458 if (vnode->flags & AFS_VNODE_DELETED) {
459 _leave(" = -ESTALE");
460 return ERR_PTR(-ESTALE);
461 }
462
463 as = dir->i_sb->s_fs_info;
464
465 /* search the directory */
466 cookie.name = dentry->d_name.name;
467 cookie.nlen = dentry->d_name.len;
468 cookie.fid.vid = as->volume->vid;
469 cookie.found = 0;
470
471 fpos = 0;
472 ret = afs_dir_iterate(dir, &fpos, &cookie, afs_dir_lookup_filldir);
473 if (ret < 0) {
474 _leave(" = %d", ret);
475 return ERR_PTR(ret);
476 }
477
478 ret = -ENOENT;
479 if (!cookie.found) {
480 _leave(" = %d", ret);
481 return ERR_PTR(ret);
482 }
483
484 /* instantiate the dentry */
485 ret = afs_iget(dir->i_sb, &cookie.fid, &inode);
486 if (ret < 0) {
487 _leave(" = %d", ret);
488 return ERR_PTR(ret);
489 }
490
491 dentry->d_op = &afs_fs_dentry_operations;
492 dentry->d_fsdata = (void *) (unsigned long) vnode->status.version;
493
494 d_add(dentry, inode);
495 _leave(" = 0 { vn=%u u=%u } -> { ino=%lu v=%lu }",
496 cookie.fid.vnode,
497 cookie.fid.unique,
498 dentry->d_inode->i_ino,
499 dentry->d_inode->i_version);
500
501 return NULL;
502} /* end afs_dir_lookup() */
503
504/*****************************************************************************/
505/*
506 * check that a dentry lookup hit has found a valid entry
507 * - NOTE! the hit can be a negative hit too, so we can't assume we have an
508 * inode
509 * (derived from nfs_lookup_revalidate)
510 */
511static int afs_d_revalidate(struct dentry *dentry, struct nameidata *nd)
512{
513 struct afs_dir_lookup_cookie cookie;
514 struct dentry *parent;
515 struct inode *inode, *dir;
516 unsigned fpos;
517 int ret;
518
519 _enter("{sb=%p n=%s},", dentry->d_sb, dentry->d_name.name);
520
521 /* lock down the parent dentry so we can peer at it */
522 parent = dget_parent(dentry->d_parent);
523
524 dir = parent->d_inode;
525 inode = dentry->d_inode;
526
527 /* handle a negative dentry */
528 if (!inode)
529 goto out_bad;
530
531 /* handle a bad inode */
532 if (is_bad_inode(inode)) {
533 printk("kAFS: afs_d_revalidate: %s/%s has bad inode\n",
534 dentry->d_parent->d_name.name, dentry->d_name.name);
535 goto out_bad;
536 }
537
538 /* force a full look up if the parent directory changed since last the
539 * server was consulted
540 * - otherwise this inode must still exist, even if the inode details
541 * themselves have changed
542 */
543 if (AFS_FS_I(dir)->flags & AFS_VNODE_CHANGED)
544 afs_vnode_fetch_status(AFS_FS_I(dir));
545
546 if (AFS_FS_I(dir)->flags & AFS_VNODE_DELETED) {
547 _debug("%s: parent dir deleted", dentry->d_name.name);
548 goto out_bad;
549 }
550
551 if (AFS_FS_I(inode)->flags & AFS_VNODE_DELETED) {
552 _debug("%s: file already deleted", dentry->d_name.name);
553 goto out_bad;
554 }
555
556 if ((unsigned long) dentry->d_fsdata !=
557 (unsigned long) AFS_FS_I(dir)->status.version) {
558 _debug("%s: parent changed %lu -> %u",
559 dentry->d_name.name,
560 (unsigned long) dentry->d_fsdata,
561 (unsigned) AFS_FS_I(dir)->status.version);
562
563 /* search the directory for this vnode */
564 cookie.name = dentry->d_name.name;
565 cookie.nlen = dentry->d_name.len;
566 cookie.fid.vid = AFS_FS_I(inode)->volume->vid;
567 cookie.found = 0;
568
569 fpos = 0;
570 ret = afs_dir_iterate(dir, &fpos, &cookie,
571 afs_dir_lookup_filldir);
572 if (ret < 0) {
573 _debug("failed to iterate dir %s: %d",
574 parent->d_name.name, ret);
575 goto out_bad;
576 }
577
578 if (!cookie.found) {
579 _debug("%s: dirent not found", dentry->d_name.name);
580 goto not_found;
581 }
582
583 /* if the vnode ID has changed, then the dirent points to a
584 * different file */
585 if (cookie.fid.vnode != AFS_FS_I(inode)->fid.vnode) {
586 _debug("%s: dirent changed", dentry->d_name.name);
587 goto not_found;
588 }
589
590 /* if the vnode ID uniqifier has changed, then the file has
591 * been deleted */
592 if (cookie.fid.unique != AFS_FS_I(inode)->fid.unique) {
593 _debug("%s: file deleted (uq %u -> %u I:%lu)",
594 dentry->d_name.name,
595 cookie.fid.unique,
596 AFS_FS_I(inode)->fid.unique,
597 inode->i_version);
598 spin_lock(&AFS_FS_I(inode)->lock);
599 AFS_FS_I(inode)->flags |= AFS_VNODE_DELETED;
600 spin_unlock(&AFS_FS_I(inode)->lock);
601 invalidate_remote_inode(inode);
602 goto out_bad;
603 }
604
605 dentry->d_fsdata =
606 (void *) (unsigned long) AFS_FS_I(dir)->status.version;
607 }
608
609 out_valid:
610 dput(parent);
611 _leave(" = 1 [valid]");
612 return 1;
613
614 /* the dirent, if it exists, now points to a different vnode */
615 not_found:
616 spin_lock(&dentry->d_lock);
617 dentry->d_flags |= DCACHE_NFSFS_RENAMED;
618 spin_unlock(&dentry->d_lock);
619
620 out_bad:
621 if (inode) {
622 /* don't unhash if we have submounts */
623 if (have_submounts(dentry))
624 goto out_valid;
625 }
626
627 shrink_dcache_parent(dentry);
628
629 _debug("dropping dentry %s/%s",
630 dentry->d_parent->d_name.name, dentry->d_name.name);
631 d_drop(dentry);
632
633 dput(parent);
634
635 _leave(" = 0 [bad]");
636 return 0;
637} /* end afs_d_revalidate() */
638
639/*****************************************************************************/
640/*
641 * allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
642 * sleep)
643 * - called from dput() when d_count is going to 0.
644 * - return 1 to request dentry be unhashed, 0 otherwise
645 */
646static int afs_d_delete(struct dentry *dentry)
647{
648 _enter("%s", dentry->d_name.name);
649
650 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
651 goto zap;
652
653 if (dentry->d_inode) {
654 if (AFS_FS_I(dentry->d_inode)->flags & AFS_VNODE_DELETED)
655 goto zap;
656 }
657
658 _leave(" = 0 [keep]");
659 return 0;
660
661 zap:
662 _leave(" = 1 [zap]");
663 return 1;
664} /* end afs_d_delete() */