blob: 5853f4440af41949081cfd2bf3e2b8f1b9363c18 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext3/dir.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/dir.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * ext3 directory handling functions
16 *
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 *
20 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
21 *
22 */
23
24#include <linux/fs.h>
25#include <linux/jbd.h>
26#include <linux/ext3_fs.h>
27#include <linux/buffer_head.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/slab.h>
29#include <linux/rbtree.h>
30
31static unsigned char ext3_filetype_table[] = {
32 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
33};
34
35static int ext3_readdir(struct file *, void *, filldir_t);
36static int ext3_dx_readdir(struct file * filp,
37 void * dirent, filldir_t filldir);
38static int ext3_release_dir (struct inode * inode,
39 struct file * filp);
40
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080041const struct file_operations ext3_dir_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 .llseek = generic_file_llseek,
43 .read = generic_read_dir,
44 .readdir = ext3_readdir, /* we take BKL. needed?*/
45 .ioctl = ext3_ioctl, /* BKL held */
David Howells52a700c2006-08-29 19:06:23 +010046#ifdef CONFIG_COMPAT
47 .compat_ioctl = ext3_compat_ioctl,
48#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 .fsync = ext3_sync_file, /* BKL held */
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 .release = ext3_release_dir,
Linus Torvalds1da177e2005-04-16 15:20:36 -070051};
52
53
54static unsigned char get_dtype(struct super_block *sb, int filetype)
55{
56 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
57 (filetype >= EXT3_FT_MAX))
58 return DT_UNKNOWN;
59
60 return (ext3_filetype_table[filetype]);
61}
Mingming Caoae6ddcc2006-09-27 01:49:27 -070062
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64int ext3_check_dir_entry (const char * function, struct inode * dir,
65 struct ext3_dir_entry_2 * de,
66 struct buffer_head * bh,
67 unsigned long offset)
68{
69 const char * error_msg = NULL;
Jan Kara7c06a8d2007-11-14 17:00:19 -080070 const int rlen = ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
72 if (rlen < EXT3_DIR_REC_LEN(1))
73 error_msg = "rec_len is smaller than minimal";
74 else if (rlen % 4 != 0)
75 error_msg = "rec_len % 4 != 0";
76 else if (rlen < EXT3_DIR_REC_LEN(de->name_len))
77 error_msg = "rec_len is too small for name_len";
78 else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
79 error_msg = "directory entry across blocks";
80 else if (le32_to_cpu(de->inode) >
81 le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
82 error_msg = "inode out of bounds";
83
84 if (error_msg != NULL)
85 ext3_error (dir->i_sb, function,
86 "bad entry in directory #%lu: %s - "
87 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
88 dir->i_ino, error_msg, offset,
89 (unsigned long) le32_to_cpu(de->inode),
90 rlen, de->name_len);
91 return error_msg == NULL ? 1 : 0;
92}
93
94static int ext3_readdir(struct file * filp,
95 void * dirent, filldir_t filldir)
96{
97 int error = 0;
Andrew Mortond8733c22006-03-23 03:00:11 -080098 unsigned long offset;
99 int i, stored;
100 struct ext3_dir_entry_2 *de;
101 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 int err;
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800103 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 int ret = 0;
Eric Sandeencdbf6db2008-10-18 20:28:00 -0700105 int dir_has_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107 sb = inode->i_sb;
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
110 EXT3_FEATURE_COMPAT_DIR_INDEX) &&
111 ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
112 ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
113 err = ext3_dx_readdir(filp, dirent, filldir);
114 if (err != ERR_BAD_DX_DIR) {
115 ret = err;
116 goto out;
117 }
118 /*
119 * We don't set the inode dirty flag since it's not
120 * critical that it get flushed back to the disk.
121 */
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800122 EXT3_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 stored = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 offset = filp->f_pos & (sb->s_blocksize - 1);
126
127 while (!error && !stored && filp->f_pos < inode->i_size) {
Andrew Mortond8733c22006-03-23 03:00:11 -0800128 unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
129 struct buffer_head map_bh;
130 struct buffer_head *bh = NULL;
131
132 map_bh.b_state = 0;
Mingming Cao89747d32006-03-26 01:37:55 -0800133 err = ext3_get_blocks_handle(NULL, inode, blk, 1,
134 &map_bh, 0, 0);
135 if (err > 0) {
Fengguang Wudc7868f2007-07-19 01:48:04 -0700136 pgoff_t index = map_bh.b_blocknr >>
137 (PAGE_CACHE_SHIFT - inode->i_blkbits);
138 if (!ra_has_index(&filp->f_ra, index))
Rusty Russellcf914a72007-07-19 01:48:08 -0700139 page_cache_sync_readahead(
Fengguang Wudc7868f2007-07-19 01:48:04 -0700140 sb->s_bdev->bd_inode->i_mapping,
141 &filp->f_ra, filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700142 index, 1);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700143 filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Andrew Mortond8733c22006-03-23 03:00:11 -0800144 bh = ext3_bread(NULL, inode, blk, 0, &err);
145 }
146
147 /*
148 * We ignore I/O errors on directories so users have a chance
149 * of recovering data when there's a bad sector
150 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 if (!bh) {
Eric Sandeencdbf6db2008-10-18 20:28:00 -0700152 if (!dir_has_error) {
153 ext3_error(sb, __func__, "directory #%lu "
154 "contains a hole at offset %lld",
155 inode->i_ino, filp->f_pos);
156 dir_has_error = 1;
157 }
Eric Sandeen40b85132006-12-06 20:36:26 -0800158 /* corrupt size? Maybe no more blocks to read */
159 if (filp->f_pos > inode->i_blocks << 9)
160 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 filp->f_pos += sb->s_blocksize - offset;
162 continue;
163 }
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165revalidate:
166 /* If the dir block has changed since the last call to
167 * readdir(2), then we might be pointing to an invalid
168 * dirent right now. Scan from the start of the block
169 * to make sure. */
170 if (filp->f_version != inode->i_version) {
171 for (i = 0; i < sb->s_blocksize && i < offset; ) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700172 de = (struct ext3_dir_entry_2 *)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 (bh->b_data + i);
174 /* It's too expensive to do a full
175 * dirent test each time round this
176 * loop, but we do have to test at
177 * least that it is non-zero. A
178 * failure will be detected in the
179 * dirent test below. */
Jan Kara7c06a8d2007-11-14 17:00:19 -0800180 if (ext3_rec_len_from_disk(de->rec_len) <
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 EXT3_DIR_REC_LEN(1))
182 break;
Jan Kara7c06a8d2007-11-14 17:00:19 -0800183 i += ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 }
185 offset = i;
186 filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
187 | offset;
188 filp->f_version = inode->i_version;
189 }
190
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700191 while (!error && filp->f_pos < inode->i_size
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 && offset < sb->s_blocksize) {
193 de = (struct ext3_dir_entry_2 *) (bh->b_data + offset);
194 if (!ext3_check_dir_entry ("ext3_readdir", inode, de,
195 bh, offset)) {
196 /* On error, skip the f_pos to the
197 next block. */
198 filp->f_pos = (filp->f_pos |
199 (sb->s_blocksize - 1)) + 1;
200 brelse (bh);
201 ret = stored;
202 goto out;
203 }
Jan Kara7c06a8d2007-11-14 17:00:19 -0800204 offset += ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 if (le32_to_cpu(de->inode)) {
206 /* We might block in the next section
207 * if the data destination is
208 * currently swapped out. So, use a
209 * version stamp to detect whether or
210 * not the directory has been modified
211 * during the copy operation.
212 */
Mathieu Desnoyers2b47c362007-10-16 23:27:21 -0700213 u64 version = filp->f_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
215 error = filldir(dirent, de->name,
216 de->name_len,
217 filp->f_pos,
218 le32_to_cpu(de->inode),
219 get_dtype(sb, de->file_type));
220 if (error)
221 break;
222 if (version != filp->f_version)
223 goto revalidate;
224 stored ++;
225 }
Jan Kara7c06a8d2007-11-14 17:00:19 -0800226 filp->f_pos += ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 }
228 offset = 0;
229 brelse (bh);
230 }
231out:
232 return ret;
233}
234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235/*
236 * These functions convert from the major/minor hash to an f_pos
237 * value.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700238 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 * Currently we only use major hash numer. This is unfortunate, but
240 * on 32-bit machines, the same VFS interface is used for lseek and
241 * llseek, so if we use the 64 bit offset, then the 32-bit versions of
242 * lseek/telldir/seekdir will blow out spectacularly, and from within
243 * the ext2 low-level routine, we don't know if we're being called by
244 * a 64-bit version of the system call or the 32-bit version of the
245 * system call. Worse yet, NFSv2 only allows for a 32-bit readdir
246 * cookie. Sigh.
247 */
248#define hash2pos(major, minor) (major >> 1)
249#define pos2maj_hash(pos) ((pos << 1) & 0xffffffff)
250#define pos2min_hash(pos) (0)
251
252/*
253 * This structure holds the nodes of the red-black tree used to store
254 * the directory entry in hash order.
255 */
256struct fname {
257 __u32 hash;
258 __u32 minor_hash;
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700259 struct rb_node rb_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 struct fname *next;
261 __u32 inode;
262 __u8 name_len;
263 __u8 file_type;
264 char name[0];
265};
266
267/*
268 * This functoin implements a non-recursive way of freeing all of the
269 * nodes in the red-black tree.
270 */
271static void free_rb_tree_fname(struct rb_root *root)
272{
273 struct rb_node *n = root->rb_node;
274 struct rb_node *parent;
275 struct fname *fname;
276
277 while (n) {
278 /* Do the node's children first */
Shen Feng9ebfbe92008-07-25 01:46:21 -0700279 if (n->rb_left) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 n = n->rb_left;
281 continue;
282 }
283 if (n->rb_right) {
284 n = n->rb_right;
285 continue;
286 }
287 /*
288 * The node has no children; free it, and then zero
289 * out parent's link to it. Finally go to the
290 * beginning of the loop and try to free the parent
291 * node.
292 */
David Woodhouse52b51082006-04-21 13:15:57 +0100293 parent = rb_parent(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 fname = rb_entry(n, struct fname, rb_hash);
295 while (fname) {
296 struct fname * old = fname;
297 fname = fname->next;
298 kfree (old);
299 }
300 if (!parent)
301 root->rb_node = NULL;
302 else if (parent->rb_left == n)
303 parent->rb_left = NULL;
304 else if (parent->rb_right == n)
305 parent->rb_right = NULL;
306 n = parent;
307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308}
309
310
Shen Feng9ebfbe92008-07-25 01:46:21 -0700311static struct dir_private_info *ext3_htree_create_dir_info(loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
313 struct dir_private_info *p;
314
Shen Feng9ebfbe92008-07-25 01:46:21 -0700315 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 if (!p)
317 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 p->curr_hash = pos2maj_hash(pos);
319 p->curr_minor_hash = pos2min_hash(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 return p;
321}
322
323void ext3_htree_free_dir_info(struct dir_private_info *p)
324{
325 free_rb_tree_fname(&p->root);
326 kfree(p);
327}
328
329/*
330 * Given a directory entry, enter it into the fname rb tree.
331 */
332int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
333 __u32 minor_hash,
334 struct ext3_dir_entry_2 *dirent)
335{
336 struct rb_node **p, *parent = NULL;
337 struct fname * fname, *new_fn;
338 struct dir_private_info *info;
339 int len;
340
341 info = (struct dir_private_info *) dir_file->private_data;
342 p = &info->root.rb_node;
343
344 /* Create and allocate the fname structure */
345 len = sizeof(struct fname) + dirent->name_len + 1;
Panagiotis Issarisf8314dc2006-09-27 01:49:37 -0700346 new_fn = kzalloc(len, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 if (!new_fn)
348 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 new_fn->hash = hash;
350 new_fn->minor_hash = minor_hash;
351 new_fn->inode = le32_to_cpu(dirent->inode);
352 new_fn->name_len = dirent->name_len;
353 new_fn->file_type = dirent->file_type;
354 memcpy(new_fn->name, dirent->name, dirent->name_len);
355 new_fn->name[dirent->name_len] = 0;
356
357 while (*p) {
358 parent = *p;
359 fname = rb_entry(parent, struct fname, rb_hash);
360
361 /*
362 * If the hash and minor hash match up, then we put
363 * them on a linked list. This rarely happens...
364 */
365 if ((new_fn->hash == fname->hash) &&
366 (new_fn->minor_hash == fname->minor_hash)) {
367 new_fn->next = fname->next;
368 fname->next = new_fn;
369 return 0;
370 }
371
372 if (new_fn->hash < fname->hash)
373 p = &(*p)->rb_left;
374 else if (new_fn->hash > fname->hash)
375 p = &(*p)->rb_right;
376 else if (new_fn->minor_hash < fname->minor_hash)
377 p = &(*p)->rb_left;
378 else /* if (new_fn->minor_hash > fname->minor_hash) */
379 p = &(*p)->rb_right;
380 }
381
382 rb_link_node(&new_fn->rb_hash, parent, p);
383 rb_insert_color(&new_fn->rb_hash, &info->root);
384 return 0;
385}
386
387
388
389/*
390 * This is a helper function for ext3_dx_readdir. It calls filldir
391 * for all entres on the fname linked list. (Normally there is only
392 * one entry on the linked list, unless there are 62 bit hash collisions.)
393 */
394static int call_filldir(struct file * filp, void * dirent,
395 filldir_t filldir, struct fname *fname)
396{
397 struct dir_private_info *info = filp->private_data;
398 loff_t curr_pos;
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800399 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 struct super_block * sb;
401 int error;
402
403 sb = inode->i_sb;
404
405 if (!fname) {
406 printk("call_filldir: called with null fname?!?\n");
407 return 0;
408 }
409 curr_pos = hash2pos(fname->hash, fname->minor_hash);
410 while (fname) {
411 error = filldir(dirent, fname->name,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700412 fname->name_len, curr_pos,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 fname->inode,
414 get_dtype(sb, fname->file_type));
415 if (error) {
416 filp->f_pos = curr_pos;
Eugene Dashevsky6a897cf2008-10-18 20:27:59 -0700417 info->extra_fname = fname;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return error;
419 }
420 fname = fname->next;
421 }
422 return 0;
423}
424
425static int ext3_dx_readdir(struct file * filp,
426 void * dirent, filldir_t filldir)
427{
428 struct dir_private_info *info = filp->private_data;
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800429 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 struct fname *fname;
431 int ret;
432
433 if (!info) {
Shen Feng9ebfbe92008-07-25 01:46:21 -0700434 info = ext3_htree_create_dir_info(filp->f_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 if (!info)
436 return -ENOMEM;
437 filp->private_data = info;
438 }
439
440 if (filp->f_pos == EXT3_HTREE_EOF)
441 return 0; /* EOF */
442
443 /* Some one has messed with f_pos; reset the world */
444 if (info->last_pos != filp->f_pos) {
445 free_rb_tree_fname(&info->root);
446 info->curr_node = NULL;
447 info->extra_fname = NULL;
448 info->curr_hash = pos2maj_hash(filp->f_pos);
449 info->curr_minor_hash = pos2min_hash(filp->f_pos);
450 }
451
452 /*
453 * If there are any leftover names on the hash collision
454 * chain, return them first.
455 */
Eugene Dashevsky6a897cf2008-10-18 20:27:59 -0700456 if (info->extra_fname) {
457 if (call_filldir(filp, dirent, filldir, info->extra_fname))
458 goto finished;
Eugene Dashevsky6a897cf2008-10-18 20:27:59 -0700459 info->extra_fname = NULL;
Theodore Ts'o8c9fa932008-10-25 11:38:37 -0400460 goto next_node;
Eugene Dashevsky6a897cf2008-10-18 20:27:59 -0700461 } else if (!info->curr_node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 info->curr_node = rb_first(&info->root);
463
464 while (1) {
465 /*
466 * Fill the rbtree if we have no more entries,
467 * or the inode has changed since we last read in the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700468 * cached entries.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 */
470 if ((!info->curr_node) ||
471 (filp->f_version != inode->i_version)) {
472 info->curr_node = NULL;
473 free_rb_tree_fname(&info->root);
474 filp->f_version = inode->i_version;
475 ret = ext3_htree_fill_tree(filp, info->curr_hash,
476 info->curr_minor_hash,
477 &info->next_hash);
478 if (ret < 0)
479 return ret;
480 if (ret == 0) {
481 filp->f_pos = EXT3_HTREE_EOF;
482 break;
483 }
484 info->curr_node = rb_first(&info->root);
485 }
486
487 fname = rb_entry(info->curr_node, struct fname, rb_hash);
488 info->curr_hash = fname->hash;
489 info->curr_minor_hash = fname->minor_hash;
490 if (call_filldir(filp, dirent, filldir, fname))
491 break;
Theodore Ts'o8c9fa932008-10-25 11:38:37 -0400492 next_node:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 info->curr_node = rb_next(info->curr_node);
Theodore Ts'o8c9fa932008-10-25 11:38:37 -0400494 if (info->curr_node) {
495 fname = rb_entry(info->curr_node, struct fname,
496 rb_hash);
497 info->curr_hash = fname->hash;
498 info->curr_minor_hash = fname->minor_hash;
499 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (info->next_hash == ~0) {
501 filp->f_pos = EXT3_HTREE_EOF;
502 break;
503 }
504 info->curr_hash = info->next_hash;
505 info->curr_minor_hash = 0;
506 }
507 }
508finished:
509 info->last_pos = filp->f_pos;
510 return 0;
511}
512
513static int ext3_release_dir (struct inode * inode, struct file * filp)
514{
515 if (filp->private_data)
516 ext3_htree_free_dir_info(filp->private_data);
517
518 return 0;
519}