• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * fs/logfs/dir.c	- directory-related code
3  *
4  * As should be obvious for Linux kernel code, license is GPLv2
5  *
6  * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
7  */
8 #include "logfs.h"
9 #include <linux/slab.h>
10 
11 /*
12  * Atomic dir operations
13  *
14  * Directory operations are by default not atomic.  Dentries and Inodes are
15  * created/removed/altered in separate operations.  Therefore we need to do
16  * a small amount of journaling.
17  *
18  * Create, link, mkdir, mknod and symlink all share the same function to do
19  * the work: __logfs_create.  This function works in two atomic steps:
20  * 1. allocate inode (remember in journal)
21  * 2. allocate dentry (clear journal)
22  *
23  * As we can only get interrupted between the two, when the inode we just
24  * created is simply stored in the anchor.  On next mount, if we were
25  * interrupted, we delete the inode.  From a users point of view the
26  * operation never happened.
27  *
28  * Unlink and rmdir also share the same function: unlink.  Again, this
29  * function works in two atomic steps
30  * 1. remove dentry (remember inode in journal)
31  * 2. unlink inode (clear journal)
32  *
33  * And again, on the next mount, if we were interrupted, we delete the inode.
34  * From a users point of view the operation succeeded.
35  *
36  * Rename is the real pain to deal with, harder than all the other methods
37  * combined.  Depending on the circumstances we can run into three cases.
38  * A "target rename" where the target dentry already existed, a "local
39  * rename" where both parent directories are identical or a "cross-directory
40  * rename" in the remaining case.
41  *
42  * Local rename is atomic, as the old dentry is simply rewritten with a new
43  * name.
44  *
45  * Cross-directory rename works in two steps, similar to __logfs_create and
46  * logfs_unlink:
47  * 1. Write new dentry (remember old dentry in journal)
48  * 2. Remove old dentry (clear journal)
49  *
50  * Here we remember a dentry instead of an inode.  On next mount, if we were
51  * interrupted, we delete the dentry.  From a users point of view, the
52  * operation succeeded.
53  *
54  * Target rename works in three atomic steps:
55  * 1. Attach old inode to new dentry (remember old dentry and new inode)
56  * 2. Remove old dentry (still remember the new inode)
57  * 3. Remove victim inode
58  *
59  * Here we remember both an inode an a dentry.  If we get interrupted
60  * between steps 1 and 2, we delete both the dentry and the inode.  If
61  * we get interrupted between steps 2 and 3, we delete just the inode.
62  * In either case, the remaining objects are deleted on next mount.  From
63  * a users point of view, the operation succeeded.
64  */
65 
write_dir(struct inode * dir,struct logfs_disk_dentry * dd,loff_t pos)66 static int write_dir(struct inode *dir, struct logfs_disk_dentry *dd,
67 		loff_t pos)
68 {
69 	return logfs_inode_write(dir, dd, sizeof(*dd), pos, WF_LOCK, NULL);
70 }
71 
write_inode(struct inode * inode)72 static int write_inode(struct inode *inode)
73 {
74 	return __logfs_write_inode(inode, NULL, WF_LOCK);
75 }
76 
dir_seek_data(struct inode * inode,s64 pos)77 static s64 dir_seek_data(struct inode *inode, s64 pos)
78 {
79 	s64 new_pos = logfs_seek_data(inode, pos);
80 
81 	return max(pos, new_pos - 1);
82 }
83 
beyond_eof(struct inode * inode,loff_t bix)84 static int beyond_eof(struct inode *inode, loff_t bix)
85 {
86 	loff_t pos = bix << inode->i_sb->s_blocksize_bits;
87 	return pos >= i_size_read(inode);
88 }
89 
90 /*
91  * Prime value was chosen to be roughly 256 + 26.  r5 hash uses 11,
92  * so short names (len <= 9) don't even occupy the complete 32bit name
93  * space.  A prime >256 ensures short names quickly spread the 32bit
94  * name space.  Add about 26 for the estimated amount of information
95  * of each character and pick a prime nearby, preferably a bit-sparse
96  * one.
97  */
logfs_hash_32(const char * s,int len,u32 seed)98 static u32 logfs_hash_32(const char *s, int len, u32 seed)
99 {
100 	u32 hash = seed;
101 	int i;
102 
103 	for (i = 0; i < len; i++)
104 		hash = hash * 293 + s[i];
105 	return hash;
106 }
107 
108 /*
109  * We have to satisfy several conflicting requirements here.  Small
110  * directories should stay fairly compact and not require too many
111  * indirect blocks.  The number of possible locations for a given hash
112  * should be small to make lookup() fast.  And we should try hard not
113  * to overflow the 32bit name space or nfs and 32bit host systems will
114  * be unhappy.
115  *
116  * So we use the following scheme.  First we reduce the hash to 0..15
117  * and try a direct block.  If that is occupied we reduce the hash to
118  * 16..255 and try an indirect block.  Same for 2x and 3x indirect
119  * blocks.  Lastly we reduce the hash to 0x800_0000 .. 0xffff_ffff,
120  * but use buckets containing eight entries instead of a single one.
121  *
122  * Using 16 entries should allow for a reasonable amount of hash
123  * collisions, so the 32bit name space can be packed fairly tight
124  * before overflowing.  Oh and currently we don't overflow but return
125  * and error.
126  *
127  * How likely are collisions?  Doing the appropriate math is beyond me
128  * and the Bronstein textbook.  But running a test program to brute
129  * force collisions for a couple of days showed that on average the
130  * first collision occurs after 598M entries, with 290M being the
131  * smallest result.  Obviously 21 entries could already cause a
132  * collision if all entries are carefully chosen.
133  */
hash_index(u32 hash,int round)134 static pgoff_t hash_index(u32 hash, int round)
135 {
136 	u32 i0_blocks = I0_BLOCKS;
137 	u32 i1_blocks = I1_BLOCKS;
138 	u32 i2_blocks = I2_BLOCKS;
139 	u32 i3_blocks = I3_BLOCKS;
140 
141 	switch (round) {
142 	case 0:
143 		return hash % i0_blocks;
144 	case 1:
145 		return i0_blocks + hash % (i1_blocks - i0_blocks);
146 	case 2:
147 		return i1_blocks + hash % (i2_blocks - i1_blocks);
148 	case 3:
149 		return i2_blocks + hash % (i3_blocks - i2_blocks);
150 	case 4 ... 19:
151 		return i3_blocks + 16 * (hash % (((1<<31) - i3_blocks) / 16))
152 			+ round - 4;
153 	}
154 	BUG();
155 }
156 
logfs_get_dd_page(struct inode * dir,struct dentry * dentry)157 static struct page *logfs_get_dd_page(struct inode *dir, struct dentry *dentry)
158 {
159 	const struct qstr *name = &dentry->d_name;
160 	struct page *page;
161 	struct logfs_disk_dentry *dd;
162 	u32 hash = logfs_hash_32(name->name, name->len, 0);
163 	pgoff_t index;
164 	int round;
165 
166 	if (name->len > LOGFS_MAX_NAMELEN)
167 		return ERR_PTR(-ENAMETOOLONG);
168 
169 	for (round = 0; round < 20; round++) {
170 		index = hash_index(hash, round);
171 
172 		if (beyond_eof(dir, index))
173 			return NULL;
174 		if (!logfs_exist_block(dir, index))
175 			continue;
176 		page = read_cache_page(dir->i_mapping, index,
177 				logfs_readpage, NULL);
178 		if (IS_ERR(page))
179 			return page;
180 		dd = kmap_atomic(page);
181 		BUG_ON(dd->namelen == 0);
182 
183 		if (name->len != be16_to_cpu(dd->namelen) ||
184 				memcmp(name->name, dd->name, name->len)) {
185 			kunmap_atomic(dd);
186 			put_page(page);
187 			continue;
188 		}
189 
190 		kunmap_atomic(dd);
191 		return page;
192 	}
193 	return NULL;
194 }
195 
logfs_remove_inode(struct inode * inode)196 static int logfs_remove_inode(struct inode *inode)
197 {
198 	int ret;
199 
200 	drop_nlink(inode);
201 	ret = write_inode(inode);
202 	LOGFS_BUG_ON(ret, inode->i_sb);
203 	return ret;
204 }
205 
abort_transaction(struct inode * inode,struct logfs_transaction * ta)206 static void abort_transaction(struct inode *inode, struct logfs_transaction *ta)
207 {
208 	if (logfs_inode(inode)->li_block)
209 		logfs_inode(inode)->li_block->ta = NULL;
210 	kfree(ta);
211 }
212 
logfs_unlink(struct inode * dir,struct dentry * dentry)213 static int logfs_unlink(struct inode *dir, struct dentry *dentry)
214 {
215 	struct logfs_super *super = logfs_super(dir->i_sb);
216 	struct inode *inode = d_inode(dentry);
217 	struct logfs_transaction *ta;
218 	struct page *page;
219 	pgoff_t index;
220 	int ret;
221 
222 	ta = kzalloc(sizeof(*ta), GFP_KERNEL);
223 	if (!ta)
224 		return -ENOMEM;
225 
226 	ta->state = UNLINK_1;
227 	ta->ino = inode->i_ino;
228 
229 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
230 
231 	page = logfs_get_dd_page(dir, dentry);
232 	if (!page) {
233 		kfree(ta);
234 		return -ENOENT;
235 	}
236 	if (IS_ERR(page)) {
237 		kfree(ta);
238 		return PTR_ERR(page);
239 	}
240 	index = page->index;
241 	put_page(page);
242 
243 	mutex_lock(&super->s_dirop_mutex);
244 	logfs_add_transaction(dir, ta);
245 
246 	ret = logfs_delete(dir, index, NULL);
247 	if (!ret)
248 		ret = write_inode(dir);
249 
250 	if (ret) {
251 		abort_transaction(dir, ta);
252 		printk(KERN_ERR"LOGFS: unable to delete inode\n");
253 		goto out;
254 	}
255 
256 	ta->state = UNLINK_2;
257 	logfs_add_transaction(inode, ta);
258 	ret = logfs_remove_inode(inode);
259 out:
260 	mutex_unlock(&super->s_dirop_mutex);
261 	return ret;
262 }
263 
logfs_empty_dir(struct inode * dir)264 static inline int logfs_empty_dir(struct inode *dir)
265 {
266 	u64 data;
267 
268 	data = logfs_seek_data(dir, 0) << dir->i_sb->s_blocksize_bits;
269 	return data >= i_size_read(dir);
270 }
271 
logfs_rmdir(struct inode * dir,struct dentry * dentry)272 static int logfs_rmdir(struct inode *dir, struct dentry *dentry)
273 {
274 	struct inode *inode = d_inode(dentry);
275 
276 	if (!logfs_empty_dir(inode))
277 		return -ENOTEMPTY;
278 
279 	return logfs_unlink(dir, dentry);
280 }
281 
282 /* FIXME: readdir currently has it's own dir_walk code.  I don't see a good
283  * way to combine the two copies */
logfs_readdir(struct file * file,struct dir_context * ctx)284 static int logfs_readdir(struct file *file, struct dir_context *ctx)
285 {
286 	struct inode *dir = file_inode(file);
287 	loff_t pos;
288 	struct page *page;
289 	struct logfs_disk_dentry *dd;
290 
291 	if (ctx->pos < 0)
292 		return -EINVAL;
293 
294 	if (!dir_emit_dots(file, ctx))
295 		return 0;
296 
297 	pos = ctx->pos - 2;
298 	BUG_ON(pos < 0);
299 	for (;; pos++, ctx->pos++) {
300 		bool full;
301 		if (beyond_eof(dir, pos))
302 			break;
303 		if (!logfs_exist_block(dir, pos)) {
304 			/* deleted dentry */
305 			pos = dir_seek_data(dir, pos);
306 			continue;
307 		}
308 		page = read_cache_page(dir->i_mapping, pos,
309 				logfs_readpage, NULL);
310 		if (IS_ERR(page))
311 			return PTR_ERR(page);
312 		dd = kmap(page);
313 		BUG_ON(dd->namelen == 0);
314 
315 		full = !dir_emit(ctx, (char *)dd->name,
316 				be16_to_cpu(dd->namelen),
317 				be64_to_cpu(dd->ino), dd->type);
318 		kunmap(page);
319 		put_page(page);
320 		if (full)
321 			break;
322 	}
323 	return 0;
324 }
325 
logfs_set_name(struct logfs_disk_dentry * dd,const struct qstr * name)326 static void logfs_set_name(struct logfs_disk_dentry *dd, const struct qstr *name)
327 {
328 	dd->namelen = cpu_to_be16(name->len);
329 	memcpy(dd->name, name->name, name->len);
330 }
331 
logfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)332 static struct dentry *logfs_lookup(struct inode *dir, struct dentry *dentry,
333 		unsigned int flags)
334 {
335 	struct page *page;
336 	struct logfs_disk_dentry *dd;
337 	pgoff_t index;
338 	u64 ino = 0;
339 	struct inode *inode;
340 
341 	page = logfs_get_dd_page(dir, dentry);
342 	if (IS_ERR(page))
343 		return ERR_CAST(page);
344 	if (!page) {
345 		d_add(dentry, NULL);
346 		return NULL;
347 	}
348 	index = page->index;
349 	dd = kmap_atomic(page);
350 	ino = be64_to_cpu(dd->ino);
351 	kunmap_atomic(dd);
352 	put_page(page);
353 
354 	inode = logfs_iget(dir->i_sb, ino);
355 	if (IS_ERR(inode))
356 		printk(KERN_ERR"LogFS: Cannot read inode #%llx for dentry (%lx, %lx)n",
357 				ino, dir->i_ino, index);
358 	return d_splice_alias(inode, dentry);
359 }
360 
grow_dir(struct inode * dir,loff_t index)361 static void grow_dir(struct inode *dir, loff_t index)
362 {
363 	index = (index + 1) << dir->i_sb->s_blocksize_bits;
364 	if (i_size_read(dir) < index)
365 		i_size_write(dir, index);
366 }
367 
logfs_write_dir(struct inode * dir,struct dentry * dentry,struct inode * inode)368 static int logfs_write_dir(struct inode *dir, struct dentry *dentry,
369 		struct inode *inode)
370 {
371 	struct page *page;
372 	struct logfs_disk_dentry *dd;
373 	u32 hash = logfs_hash_32(dentry->d_name.name, dentry->d_name.len, 0);
374 	pgoff_t index;
375 	int round, err;
376 
377 	for (round = 0; round < 20; round++) {
378 		index = hash_index(hash, round);
379 
380 		if (logfs_exist_block(dir, index))
381 			continue;
382 		page = find_or_create_page(dir->i_mapping, index, GFP_KERNEL);
383 		if (!page)
384 			return -ENOMEM;
385 
386 		dd = kmap_atomic(page);
387 		memset(dd, 0, sizeof(*dd));
388 		dd->ino = cpu_to_be64(inode->i_ino);
389 		dd->type = logfs_type(inode);
390 		logfs_set_name(dd, &dentry->d_name);
391 		kunmap_atomic(dd);
392 
393 		err = logfs_write_buf(dir, page, WF_LOCK);
394 		unlock_page(page);
395 		put_page(page);
396 		if (!err)
397 			grow_dir(dir, index);
398 		return err;
399 	}
400 	/* FIXME: Is there a better return value?  In most cases neither
401 	 * the filesystem nor the directory are full.  But we have had
402 	 * too many collisions for this particular hash and no fallback.
403 	 */
404 	return -ENOSPC;
405 }
406 
__logfs_create(struct inode * dir,struct dentry * dentry,struct inode * inode,const char * dest,long destlen)407 static int __logfs_create(struct inode *dir, struct dentry *dentry,
408 		struct inode *inode, const char *dest, long destlen)
409 {
410 	struct logfs_super *super = logfs_super(dir->i_sb);
411 	struct logfs_inode *li = logfs_inode(inode);
412 	struct logfs_transaction *ta;
413 	int ret;
414 
415 	ta = kzalloc(sizeof(*ta), GFP_KERNEL);
416 	if (!ta) {
417 		drop_nlink(inode);
418 		iput(inode);
419 		return -ENOMEM;
420 	}
421 
422 	ta->state = CREATE_1;
423 	ta->ino = inode->i_ino;
424 	mutex_lock(&super->s_dirop_mutex);
425 	logfs_add_transaction(inode, ta);
426 
427 	if (dest) {
428 		/* symlink */
429 		ret = logfs_inode_write(inode, dest, destlen, 0, WF_LOCK, NULL);
430 		if (!ret)
431 			ret = write_inode(inode);
432 	} else {
433 		/* creat/mkdir/mknod */
434 		ret = write_inode(inode);
435 	}
436 	if (ret) {
437 		abort_transaction(inode, ta);
438 		li->li_flags |= LOGFS_IF_STILLBORN;
439 		/* FIXME: truncate symlink */
440 		drop_nlink(inode);
441 		iput(inode);
442 		goto out;
443 	}
444 
445 	ta->state = CREATE_2;
446 	logfs_add_transaction(dir, ta);
447 	ret = logfs_write_dir(dir, dentry, inode);
448 	/* sync directory */
449 	if (!ret)
450 		ret = write_inode(dir);
451 
452 	if (ret) {
453 		logfs_del_transaction(dir, ta);
454 		ta->state = CREATE_2;
455 		logfs_add_transaction(inode, ta);
456 		logfs_remove_inode(inode);
457 		iput(inode);
458 		goto out;
459 	}
460 	d_instantiate(dentry, inode);
461 out:
462 	mutex_unlock(&super->s_dirop_mutex);
463 	return ret;
464 }
465 
logfs_mkdir(struct inode * dir,struct dentry * dentry,umode_t mode)466 static int logfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
467 {
468 	struct inode *inode;
469 
470 	/*
471 	 * FIXME: why do we have to fill in S_IFDIR, while the mode is
472 	 * correct for mknod, creat, etc.?  Smells like the vfs *should*
473 	 * do it for us but for some reason fails to do so.
474 	 */
475 	inode = logfs_new_inode(dir, S_IFDIR | mode);
476 	if (IS_ERR(inode))
477 		return PTR_ERR(inode);
478 
479 	inode->i_op = &logfs_dir_iops;
480 	inode->i_fop = &logfs_dir_fops;
481 
482 	return __logfs_create(dir, dentry, inode, NULL, 0);
483 }
484 
logfs_create(struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)485 static int logfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
486 		bool excl)
487 {
488 	struct inode *inode;
489 
490 	inode = logfs_new_inode(dir, mode);
491 	if (IS_ERR(inode))
492 		return PTR_ERR(inode);
493 
494 	inode->i_op = &logfs_reg_iops;
495 	inode->i_fop = &logfs_reg_fops;
496 	inode->i_mapping->a_ops = &logfs_reg_aops;
497 
498 	return __logfs_create(dir, dentry, inode, NULL, 0);
499 }
500 
logfs_mknod(struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)501 static int logfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
502 		dev_t rdev)
503 {
504 	struct inode *inode;
505 
506 	if (dentry->d_name.len > LOGFS_MAX_NAMELEN)
507 		return -ENAMETOOLONG;
508 
509 	inode = logfs_new_inode(dir, mode);
510 	if (IS_ERR(inode))
511 		return PTR_ERR(inode);
512 
513 	init_special_inode(inode, mode, rdev);
514 
515 	return __logfs_create(dir, dentry, inode, NULL, 0);
516 }
517 
logfs_symlink(struct inode * dir,struct dentry * dentry,const char * target)518 static int logfs_symlink(struct inode *dir, struct dentry *dentry,
519 		const char *target)
520 {
521 	struct inode *inode;
522 	size_t destlen = strlen(target) + 1;
523 
524 	if (destlen > dir->i_sb->s_blocksize)
525 		return -ENAMETOOLONG;
526 
527 	inode = logfs_new_inode(dir, S_IFLNK | 0777);
528 	if (IS_ERR(inode))
529 		return PTR_ERR(inode);
530 
531 	inode->i_op = &page_symlink_inode_operations;
532 	inode_nohighmem(inode);
533 	inode->i_mapping->a_ops = &logfs_reg_aops;
534 
535 	return __logfs_create(dir, dentry, inode, target, destlen);
536 }
537 
logfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)538 static int logfs_link(struct dentry *old_dentry, struct inode *dir,
539 		struct dentry *dentry)
540 {
541 	struct inode *inode = d_inode(old_dentry);
542 
543 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
544 	ihold(inode);
545 	inc_nlink(inode);
546 	mark_inode_dirty_sync(inode);
547 
548 	return __logfs_create(dir, dentry, inode, NULL, 0);
549 }
550 
logfs_get_dd(struct inode * dir,struct dentry * dentry,struct logfs_disk_dentry * dd,loff_t * pos)551 static int logfs_get_dd(struct inode *dir, struct dentry *dentry,
552 		struct logfs_disk_dentry *dd, loff_t *pos)
553 {
554 	struct page *page;
555 	void *map;
556 
557 	page = logfs_get_dd_page(dir, dentry);
558 	if (IS_ERR(page))
559 		return PTR_ERR(page);
560 	*pos = page->index;
561 	map = kmap_atomic(page);
562 	memcpy(dd, map, sizeof(*dd));
563 	kunmap_atomic(map);
564 	put_page(page);
565 	return 0;
566 }
567 
logfs_delete_dd(struct inode * dir,loff_t pos)568 static int logfs_delete_dd(struct inode *dir, loff_t pos)
569 {
570 	/*
571 	 * Getting called with pos somewhere beyond eof is either a goofup
572 	 * within this file or means someone maliciously edited the
573 	 * (crc-protected) journal.
574 	 */
575 	BUG_ON(beyond_eof(dir, pos));
576 	dir->i_ctime = dir->i_mtime = current_time(dir);
577 	log_dir(" Delete dentry (%lx, %llx)\n", dir->i_ino, pos);
578 	return logfs_delete(dir, pos, NULL);
579 }
580 
581 /*
582  * Cross-directory rename, target does not exist.  Just a little nasty.
583  * Create a new dentry in the target dir, then remove the old dentry,
584  * all the while taking care to remember our operation in the journal.
585  */
logfs_rename_cross(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)586 static int logfs_rename_cross(struct inode *old_dir, struct dentry *old_dentry,
587 			      struct inode *new_dir, struct dentry *new_dentry)
588 {
589 	struct logfs_super *super = logfs_super(old_dir->i_sb);
590 	struct logfs_disk_dentry dd;
591 	struct logfs_transaction *ta;
592 	loff_t pos;
593 	int err;
594 
595 	/* 1. locate source dd */
596 	err = logfs_get_dd(old_dir, old_dentry, &dd, &pos);
597 	if (err)
598 		return err;
599 
600 	ta = kzalloc(sizeof(*ta), GFP_KERNEL);
601 	if (!ta)
602 		return -ENOMEM;
603 
604 	ta->state = CROSS_RENAME_1;
605 	ta->dir = old_dir->i_ino;
606 	ta->pos = pos;
607 
608 	/* 2. write target dd */
609 	mutex_lock(&super->s_dirop_mutex);
610 	logfs_add_transaction(new_dir, ta);
611 	err = logfs_write_dir(new_dir, new_dentry, d_inode(old_dentry));
612 	if (!err)
613 		err = write_inode(new_dir);
614 
615 	if (err) {
616 		super->s_rename_dir = 0;
617 		super->s_rename_pos = 0;
618 		abort_transaction(new_dir, ta);
619 		goto out;
620 	}
621 
622 	/* 3. remove source dd */
623 	ta->state = CROSS_RENAME_2;
624 	logfs_add_transaction(old_dir, ta);
625 	err = logfs_delete_dd(old_dir, pos);
626 	if (!err)
627 		err = write_inode(old_dir);
628 	LOGFS_BUG_ON(err, old_dir->i_sb);
629 out:
630 	mutex_unlock(&super->s_dirop_mutex);
631 	return err;
632 }
633 
logfs_replace_inode(struct inode * dir,struct dentry * dentry,struct logfs_disk_dentry * dd,struct inode * inode)634 static int logfs_replace_inode(struct inode *dir, struct dentry *dentry,
635 		struct logfs_disk_dentry *dd, struct inode *inode)
636 {
637 	loff_t pos;
638 	int err;
639 
640 	err = logfs_get_dd(dir, dentry, dd, &pos);
641 	if (err)
642 		return err;
643 	dd->ino = cpu_to_be64(inode->i_ino);
644 	dd->type = logfs_type(inode);
645 
646 	err = write_dir(dir, dd, pos);
647 	if (err)
648 		return err;
649 	log_dir("Replace dentry (%lx, %llx) %s -> %llx\n", dir->i_ino, pos,
650 			dd->name, be64_to_cpu(dd->ino));
651 	return write_inode(dir);
652 }
653 
654 /* Target dentry exists - the worst case.  We need to attach the source
655  * inode to the target dentry, then remove the orphaned target inode and
656  * source dentry.
657  */
logfs_rename_target(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)658 static int logfs_rename_target(struct inode *old_dir, struct dentry *old_dentry,
659 			       struct inode *new_dir, struct dentry *new_dentry)
660 {
661 	struct logfs_super *super = logfs_super(old_dir->i_sb);
662 	struct inode *old_inode = d_inode(old_dentry);
663 	struct inode *new_inode = d_inode(new_dentry);
664 	int isdir = S_ISDIR(old_inode->i_mode);
665 	struct logfs_disk_dentry dd;
666 	struct logfs_transaction *ta;
667 	loff_t pos;
668 	int err;
669 
670 	BUG_ON(isdir != S_ISDIR(new_inode->i_mode));
671 	if (isdir) {
672 		if (!logfs_empty_dir(new_inode))
673 			return -ENOTEMPTY;
674 	}
675 
676 	/* 1. locate source dd */
677 	err = logfs_get_dd(old_dir, old_dentry, &dd, &pos);
678 	if (err)
679 		return err;
680 
681 	ta = kzalloc(sizeof(*ta), GFP_KERNEL);
682 	if (!ta)
683 		return -ENOMEM;
684 
685 	ta->state = TARGET_RENAME_1;
686 	ta->dir = old_dir->i_ino;
687 	ta->pos = pos;
688 	ta->ino = new_inode->i_ino;
689 
690 	/* 2. attach source inode to target dd */
691 	mutex_lock(&super->s_dirop_mutex);
692 	logfs_add_transaction(new_dir, ta);
693 	err = logfs_replace_inode(new_dir, new_dentry, &dd, old_inode);
694 	if (err) {
695 		super->s_rename_dir = 0;
696 		super->s_rename_pos = 0;
697 		super->s_victim_ino = 0;
698 		abort_transaction(new_dir, ta);
699 		goto out;
700 	}
701 
702 	/* 3. remove source dd */
703 	ta->state = TARGET_RENAME_2;
704 	logfs_add_transaction(old_dir, ta);
705 	err = logfs_delete_dd(old_dir, pos);
706 	if (!err)
707 		err = write_inode(old_dir);
708 	LOGFS_BUG_ON(err, old_dir->i_sb);
709 
710 	/* 4. remove target inode */
711 	ta->state = TARGET_RENAME_3;
712 	logfs_add_transaction(new_inode, ta);
713 	err = logfs_remove_inode(new_inode);
714 
715 out:
716 	mutex_unlock(&super->s_dirop_mutex);
717 	return err;
718 }
719 
logfs_rename(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)720 static int logfs_rename(struct inode *old_dir, struct dentry *old_dentry,
721 			struct inode *new_dir, struct dentry *new_dentry,
722 			unsigned int flags)
723 {
724 	if (flags & ~RENAME_NOREPLACE)
725 		return -EINVAL;
726 
727 	if (d_really_is_positive(new_dentry))
728 		return logfs_rename_target(old_dir, old_dentry,
729 					   new_dir, new_dentry);
730 	return logfs_rename_cross(old_dir, old_dentry, new_dir, new_dentry);
731 }
732 
733 /* No locking done here, as this is called before .get_sb() returns. */
logfs_replay_journal(struct super_block * sb)734 int logfs_replay_journal(struct super_block *sb)
735 {
736 	struct logfs_super *super = logfs_super(sb);
737 	struct inode *inode;
738 	u64 ino, pos;
739 	int err;
740 
741 	if (super->s_victim_ino) {
742 		/* delete victim inode */
743 		ino = super->s_victim_ino;
744 		printk(KERN_INFO"LogFS: delete unmapped inode #%llx\n", ino);
745 		inode = logfs_iget(sb, ino);
746 		if (IS_ERR(inode))
747 			goto fail;
748 
749 		LOGFS_BUG_ON(i_size_read(inode) > 0, sb);
750 		super->s_victim_ino = 0;
751 		err = logfs_remove_inode(inode);
752 		iput(inode);
753 		if (err) {
754 			super->s_victim_ino = ino;
755 			goto fail;
756 		}
757 	}
758 	if (super->s_rename_dir) {
759 		/* delete old dd from rename */
760 		ino = super->s_rename_dir;
761 		pos = super->s_rename_pos;
762 		printk(KERN_INFO"LogFS: delete unbacked dentry (%llx, %llx)\n",
763 				ino, pos);
764 		inode = logfs_iget(sb, ino);
765 		if (IS_ERR(inode))
766 			goto fail;
767 
768 		super->s_rename_dir = 0;
769 		super->s_rename_pos = 0;
770 		err = logfs_delete_dd(inode, pos);
771 		iput(inode);
772 		if (err) {
773 			super->s_rename_dir = ino;
774 			super->s_rename_pos = pos;
775 			goto fail;
776 		}
777 	}
778 	return 0;
779 fail:
780 	LOGFS_BUG(sb);
781 	return -EIO;
782 }
783 
784 const struct inode_operations logfs_dir_iops = {
785 	.create		= logfs_create,
786 	.link		= logfs_link,
787 	.lookup		= logfs_lookup,
788 	.mkdir		= logfs_mkdir,
789 	.mknod		= logfs_mknod,
790 	.rename		= logfs_rename,
791 	.rmdir		= logfs_rmdir,
792 	.symlink	= logfs_symlink,
793 	.unlink		= logfs_unlink,
794 };
795 const struct file_operations logfs_dir_fops = {
796 	.fsync		= logfs_fsync,
797 	.unlocked_ioctl	= logfs_ioctl,
798 	.iterate_shared	= logfs_readdir,
799 	.read		= generic_read_dir,
800 	.llseek		= generic_file_llseek,
801 };
802