xref: /linux/fs/ocfs2/dir.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dir.c
5  *
6  * Creates, reads, walks and deletes directory-nodes
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38 
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 
44 #define MLOG_MASK_PREFIX ML_NAMEI
45 #include <cluster/masklog.h>
46 
47 #include "ocfs2.h"
48 
49 #include "alloc.h"
50 #include "dir.h"
51 #include "dlmglue.h"
52 #include "extent_map.h"
53 #include "file.h"
54 #include "inode.h"
55 #include "journal.h"
56 #include "namei.h"
57 #include "suballoc.h"
58 #include "uptodate.h"
59 
60 #include "buffer_head_io.h"
61 
62 static unsigned char ocfs2_filetype_table[] = {
63 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
64 };
65 
66 static int ocfs2_extend_dir(struct ocfs2_super *osb,
67 			    struct inode *dir,
68 			    struct buffer_head *parent_fe_bh,
69 			    struct buffer_head **new_de_bh);
70 /*
71  * ocfs2_readdir()
72  *
73  */
74 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
75 {
76 	int error = 0;
77 	unsigned long offset, blk;
78 	int i, num, stored;
79 	struct buffer_head * bh, * tmp;
80 	struct ocfs2_dir_entry * de;
81 	int err;
82 	struct inode *inode = filp->f_dentry->d_inode;
83 	struct super_block * sb = inode->i_sb;
84 	int have_disk_lock = 0;
85 
86 	mlog_entry("dirino=%llu\n",
87 		   (unsigned long long)OCFS2_I(inode)->ip_blkno);
88 
89 	stored = 0;
90 	bh = NULL;
91 
92 	error = ocfs2_meta_lock(inode, NULL, NULL, 0);
93 	if (error < 0) {
94 		if (error != -ENOENT)
95 			mlog_errno(error);
96 		/* we haven't got any yet, so propagate the error. */
97 		stored = error;
98 		goto bail;
99 	}
100 	have_disk_lock = 1;
101 
102 	offset = filp->f_pos & (sb->s_blocksize - 1);
103 
104 	while (!error && !stored && filp->f_pos < i_size_read(inode)) {
105 		blk = (filp->f_pos) >> sb->s_blocksize_bits;
106 		bh = ocfs2_bread(inode, blk, &err, 0);
107 		if (!bh) {
108 			mlog(ML_ERROR,
109 			     "directory #%llu contains a hole at offset %lld\n",
110 			     (unsigned long long)OCFS2_I(inode)->ip_blkno,
111 			     filp->f_pos);
112 			filp->f_pos += sb->s_blocksize - offset;
113 			continue;
114 		}
115 
116 		/*
117 		 * Do the readahead (8k)
118 		 */
119 		if (!offset) {
120 			for (i = 16 >> (sb->s_blocksize_bits - 9), num = 0;
121 			     i > 0; i--) {
122 				tmp = ocfs2_bread(inode, ++blk, &err, 1);
123 				if (tmp)
124 					brelse(tmp);
125 			}
126 		}
127 
128 revalidate:
129 		/* If the dir block has changed since the last call to
130 		 * readdir(2), then we might be pointing to an invalid
131 		 * dirent right now.  Scan from the start of the block
132 		 * to make sure. */
133 		if (filp->f_version != inode->i_version) {
134 			for (i = 0; i < sb->s_blocksize && i < offset; ) {
135 				de = (struct ocfs2_dir_entry *) (bh->b_data + i);
136 				/* It's too expensive to do a full
137 				 * dirent test each time round this
138 				 * loop, but we do have to test at
139 				 * least that it is non-zero.  A
140 				 * failure will be detected in the
141 				 * dirent test below. */
142 				if (le16_to_cpu(de->rec_len) <
143 				    OCFS2_DIR_REC_LEN(1))
144 					break;
145 				i += le16_to_cpu(de->rec_len);
146 			}
147 			offset = i;
148 			filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
149 				| offset;
150 			filp->f_version = inode->i_version;
151 		}
152 
153 		while (!error && filp->f_pos < i_size_read(inode)
154 		       && offset < sb->s_blocksize) {
155 			de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
156 			if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
157 				/* On error, skip the f_pos to the
158 				   next block. */
159 				filp->f_pos = (filp->f_pos |
160 					       (sb->s_blocksize - 1)) + 1;
161 				brelse(bh);
162 				goto bail;
163 			}
164 			offset += le16_to_cpu(de->rec_len);
165 			if (le64_to_cpu(de->inode)) {
166 				/* We might block in the next section
167 				 * if the data destination is
168 				 * currently swapped out.  So, use a
169 				 * version stamp to detect whether or
170 				 * not the directory has been modified
171 				 * during the copy operation.
172 				 */
173 				unsigned long version = filp->f_version;
174 				unsigned char d_type = DT_UNKNOWN;
175 
176 				if (de->file_type < OCFS2_FT_MAX)
177 					d_type = ocfs2_filetype_table[de->file_type];
178 				error = filldir(dirent, de->name,
179 						de->name_len,
180 						filp->f_pos,
181 						ino_from_blkno(sb, le64_to_cpu(de->inode)),
182 						d_type);
183 				if (error)
184 					break;
185 				if (version != filp->f_version)
186 					goto revalidate;
187 				stored ++;
188 			}
189 			filp->f_pos += le16_to_cpu(de->rec_len);
190 		}
191 		offset = 0;
192 		brelse(bh);
193 	}
194 
195 	stored = 0;
196 bail:
197 	if (have_disk_lock)
198 		ocfs2_meta_unlock(inode, 0);
199 
200 	mlog_exit(stored);
201 
202 	return stored;
203 }
204 
205 /*
206  * NOTE: this should always be called with parent dir i_mutex taken.
207  */
208 int ocfs2_find_files_on_disk(const char *name,
209 			     int namelen,
210 			     u64 *blkno,
211 			     struct inode *inode,
212 			     struct buffer_head **dirent_bh,
213 			     struct ocfs2_dir_entry **dirent)
214 {
215 	int status = -ENOENT;
216 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
217 
218 	mlog_entry("(osb=%p, parent=%llu, name='%.*s', blkno=%p, inode=%p)\n",
219 		   osb, (unsigned long long)OCFS2_I(inode)->ip_blkno,
220 		   namelen, name, blkno, inode);
221 
222 	*dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
223 	if (!*dirent_bh || !*dirent) {
224 		status = -ENOENT;
225 		goto leave;
226 	}
227 
228 	*blkno = le64_to_cpu((*dirent)->inode);
229 
230 	status = 0;
231 leave:
232 	if (status < 0) {
233 		*dirent = NULL;
234 		if (*dirent_bh) {
235 			brelse(*dirent_bh);
236 			*dirent_bh = NULL;
237 		}
238 	}
239 
240 	mlog_exit(status);
241 	return status;
242 }
243 
244 /* Check for a name within a directory.
245  *
246  * Return 0 if the name does not exist
247  * Return -EEXIST if the directory contains the name
248  *
249  * Callers should have i_mutex + a cluster lock on dir
250  */
251 int ocfs2_check_dir_for_entry(struct inode *dir,
252 			      const char *name,
253 			      int namelen)
254 {
255 	int ret;
256 	struct buffer_head *dirent_bh = NULL;
257 	struct ocfs2_dir_entry *dirent = NULL;
258 
259 	mlog_entry("dir %llu, name '%.*s'\n",
260 		   (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
261 
262 	ret = -EEXIST;
263 	dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
264 	if (dirent_bh)
265 		goto bail;
266 
267 	ret = 0;
268 bail:
269 	if (dirent_bh)
270 		brelse(dirent_bh);
271 
272 	mlog_exit(ret);
273 	return ret;
274 }
275 
276 /*
277  * routine to check that the specified directory is empty (for rmdir)
278  */
279 int ocfs2_empty_dir(struct inode *inode)
280 {
281 	unsigned long offset;
282 	struct buffer_head * bh;
283 	struct ocfs2_dir_entry * de, * de1;
284 	struct super_block * sb;
285 	int err;
286 
287 	sb = inode->i_sb;
288 	if ((i_size_read(inode) <
289 	     (OCFS2_DIR_REC_LEN(1) + OCFS2_DIR_REC_LEN(2))) ||
290 	    !(bh = ocfs2_bread(inode, 0, &err, 0))) {
291 	    	mlog(ML_ERROR, "bad directory (dir #%llu) - no data block\n",
292 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
293 		return 1;
294 	}
295 
296 	de = (struct ocfs2_dir_entry *) bh->b_data;
297 	de1 = (struct ocfs2_dir_entry *)
298 			((char *)de + le16_to_cpu(de->rec_len));
299 	if ((le64_to_cpu(de->inode) != OCFS2_I(inode)->ip_blkno) ||
300 			!le64_to_cpu(de1->inode) ||
301 			strcmp(".", de->name) ||
302 			strcmp("..", de1->name)) {
303 	    	mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
304 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
305 		brelse(bh);
306 		return 1;
307 	}
308 	offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
309 	de = (struct ocfs2_dir_entry *)((char *)de1 + le16_to_cpu(de1->rec_len));
310 	while (offset < i_size_read(inode) ) {
311 		if (!bh || (void *)de >= (void *)(bh->b_data + sb->s_blocksize)) {
312 			brelse(bh);
313 			bh = ocfs2_bread(inode,
314 					 offset >> sb->s_blocksize_bits, &err, 0);
315 			if (!bh) {
316 				mlog(ML_ERROR, "dir %llu has a hole at %lu\n",
317 				     (unsigned long long)OCFS2_I(inode)->ip_blkno, offset);
318 				offset += sb->s_blocksize;
319 				continue;
320 			}
321 			de = (struct ocfs2_dir_entry *) bh->b_data;
322 		}
323 		if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
324 			brelse(bh);
325 			return 1;
326 		}
327 		if (le64_to_cpu(de->inode)) {
328 			brelse(bh);
329 			return 0;
330 		}
331 		offset += le16_to_cpu(de->rec_len);
332 		de = (struct ocfs2_dir_entry *)
333 			((char *)de + le16_to_cpu(de->rec_len));
334 	}
335 	brelse(bh);
336 	return 1;
337 }
338 
339 /* returns a bh of the 1st new block in the allocation. */
340 int ocfs2_do_extend_dir(struct super_block *sb,
341 			struct ocfs2_journal_handle *handle,
342 			struct inode *dir,
343 			struct buffer_head *parent_fe_bh,
344 			struct ocfs2_alloc_context *data_ac,
345 			struct ocfs2_alloc_context *meta_ac,
346 			struct buffer_head **new_bh)
347 {
348 	int status;
349 	int extend;
350 	u64 p_blkno;
351 
352 	spin_lock(&OCFS2_I(dir)->ip_lock);
353 	extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
354 	spin_unlock(&OCFS2_I(dir)->ip_lock);
355 
356 	if (extend) {
357 		status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, 1,
358 						    parent_fe_bh, handle,
359 						    data_ac, meta_ac, NULL);
360 		BUG_ON(status == -EAGAIN);
361 		if (status < 0) {
362 			mlog_errno(status);
363 			goto bail;
364 		}
365 	}
366 
367 	status = ocfs2_extent_map_get_blocks(dir, (dir->i_blocks >>
368 						   (sb->s_blocksize_bits - 9)),
369 					     1, &p_blkno, NULL);
370 	if (status < 0) {
371 		mlog_errno(status);
372 		goto bail;
373 	}
374 
375 	*new_bh = sb_getblk(sb, p_blkno);
376 	if (!*new_bh) {
377 		status = -EIO;
378 		mlog_errno(status);
379 		goto bail;
380 	}
381 	status = 0;
382 bail:
383 	mlog_exit(status);
384 	return status;
385 }
386 
387 /* assumes you already have a cluster lock on the directory. */
388 static int ocfs2_extend_dir(struct ocfs2_super *osb,
389 			    struct inode *dir,
390 			    struct buffer_head *parent_fe_bh,
391 			    struct buffer_head **new_de_bh)
392 {
393 	int status = 0;
394 	int credits, num_free_extents;
395 	loff_t dir_i_size;
396 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
397 	struct ocfs2_alloc_context *data_ac = NULL;
398 	struct ocfs2_alloc_context *meta_ac = NULL;
399 	struct ocfs2_journal_handle *handle = NULL;
400 	struct buffer_head *new_bh = NULL;
401 	struct ocfs2_dir_entry * de;
402 	struct super_block *sb = osb->sb;
403 
404 	mlog_entry_void();
405 
406 	dir_i_size = i_size_read(dir);
407 	mlog(0, "extending dir %llu (i_size = %lld)\n",
408 	     (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
409 
410 	handle = ocfs2_alloc_handle(osb);
411 	if (handle == NULL) {
412 		status = -ENOMEM;
413 		mlog_errno(status);
414 		goto bail;
415 	}
416 
417 	/* dir->i_size is always block aligned. */
418 	spin_lock(&OCFS2_I(dir)->ip_lock);
419 	if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
420 		spin_unlock(&OCFS2_I(dir)->ip_lock);
421 		num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
422 		if (num_free_extents < 0) {
423 			status = num_free_extents;
424 			mlog_errno(status);
425 			goto bail;
426 		}
427 
428 		if (!num_free_extents) {
429 			status = ocfs2_reserve_new_metadata(osb, handle,
430 							    fe, &meta_ac);
431 			if (status < 0) {
432 				if (status != -ENOSPC)
433 					mlog_errno(status);
434 				goto bail;
435 			}
436 		}
437 
438 		status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
439 		if (status < 0) {
440 			if (status != -ENOSPC)
441 				mlog_errno(status);
442 			goto bail;
443 		}
444 
445 		credits = ocfs2_calc_extend_credits(sb, fe, 1);
446 	} else {
447 		spin_unlock(&OCFS2_I(dir)->ip_lock);
448 		credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
449 	}
450 
451 	handle = ocfs2_start_trans(osb, handle, credits);
452 	if (IS_ERR(handle)) {
453 		status = PTR_ERR(handle);
454 		handle = NULL;
455 		mlog_errno(status);
456 		goto bail;
457 	}
458 
459 	status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
460 				     data_ac, meta_ac, &new_bh);
461 	if (status < 0) {
462 		mlog_errno(status);
463 		goto bail;
464 	}
465 
466 	ocfs2_set_new_buffer_uptodate(dir, new_bh);
467 
468 	status = ocfs2_journal_access(handle, dir, new_bh,
469 				      OCFS2_JOURNAL_ACCESS_CREATE);
470 	if (status < 0) {
471 		mlog_errno(status);
472 		goto bail;
473 	}
474 	memset(new_bh->b_data, 0, sb->s_blocksize);
475 	de = (struct ocfs2_dir_entry *) new_bh->b_data;
476 	de->inode = 0;
477 	de->rec_len = cpu_to_le16(sb->s_blocksize);
478 	status = ocfs2_journal_dirty(handle, new_bh);
479 	if (status < 0) {
480 		mlog_errno(status);
481 		goto bail;
482 	}
483 
484 	dir_i_size += dir->i_sb->s_blocksize;
485 	i_size_write(dir, dir_i_size);
486 	dir->i_blocks = ocfs2_align_bytes_to_sectors(dir_i_size);
487 	status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
488 	if (status < 0) {
489 		mlog_errno(status);
490 		goto bail;
491 	}
492 
493 	*new_de_bh = new_bh;
494 	get_bh(*new_de_bh);
495 bail:
496 	if (handle)
497 		ocfs2_commit_trans(handle);
498 
499 	if (data_ac)
500 		ocfs2_free_alloc_context(data_ac);
501 	if (meta_ac)
502 		ocfs2_free_alloc_context(meta_ac);
503 
504 	if (new_bh)
505 		brelse(new_bh);
506 
507 	mlog_exit(status);
508 	return status;
509 }
510 
511 /*
512  * Search the dir for a good spot, extending it if necessary. The
513  * block containing an appropriate record is returned in ret_de_bh.
514  */
515 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
516 				 struct inode *dir,
517 				 struct buffer_head *parent_fe_bh,
518 				 const char *name,
519 				 int namelen,
520 				 struct buffer_head **ret_de_bh)
521 {
522 	unsigned long offset;
523 	struct buffer_head * bh = NULL;
524 	unsigned short rec_len;
525 	struct ocfs2_dinode *fe;
526 	struct ocfs2_dir_entry *de;
527 	struct super_block *sb;
528 	int status;
529 
530 	mlog_entry_void();
531 
532 	mlog(0, "getting ready to insert namelen %d into dir %llu\n",
533 	     namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
534 
535 	BUG_ON(!S_ISDIR(dir->i_mode));
536 	fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
537 	BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
538 
539 	sb = dir->i_sb;
540 
541 	if (!namelen) {
542 		status = -EINVAL;
543 		mlog_errno(status);
544 		goto bail;
545 	}
546 
547 	bh = ocfs2_bread(dir, 0, &status, 0);
548 	if (!bh) {
549 		mlog_errno(status);
550 		goto bail;
551 	}
552 
553 	rec_len = OCFS2_DIR_REC_LEN(namelen);
554 	offset = 0;
555 	de = (struct ocfs2_dir_entry *) bh->b_data;
556 	while (1) {
557 		if ((char *)de >= sb->s_blocksize + bh->b_data) {
558 			brelse(bh);
559 			bh = NULL;
560 
561 			if (i_size_read(dir) <= offset) {
562 				status = ocfs2_extend_dir(osb,
563 							  dir,
564 							  parent_fe_bh,
565 							  &bh);
566 				if (status < 0) {
567 					mlog_errno(status);
568 					goto bail;
569 				}
570 				BUG_ON(!bh);
571 				*ret_de_bh = bh;
572 				get_bh(*ret_de_bh);
573 				goto bail;
574 			}
575 			bh = ocfs2_bread(dir,
576 					 offset >> sb->s_blocksize_bits,
577 					 &status,
578 					 0);
579 			if (!bh) {
580 				mlog_errno(status);
581 				goto bail;
582 			}
583 			/* move to next block */
584 			de = (struct ocfs2_dir_entry *) bh->b_data;
585 		}
586 		if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
587 			status = -ENOENT;
588 			goto bail;
589 		}
590 		if (ocfs2_match(namelen, name, de)) {
591 			status = -EEXIST;
592 			goto bail;
593 		}
594 		if (((le64_to_cpu(de->inode) == 0) &&
595 		     (le16_to_cpu(de->rec_len) >= rec_len)) ||
596 		    (le16_to_cpu(de->rec_len) >=
597 		     (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
598 			/* Ok, we found a spot. Return this bh and let
599 			 * the caller actually fill it in. */
600 			*ret_de_bh = bh;
601 			get_bh(*ret_de_bh);
602 			status = 0;
603 			goto bail;
604 		}
605 		offset += le16_to_cpu(de->rec_len);
606 		de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
607 	}
608 
609 	status = 0;
610 bail:
611 	if (bh)
612 		brelse(bh);
613 
614 	mlog_exit(status);
615 	return status;
616 }
617