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  */
hash_32(const char * s,int len,u32 seed)98 static u32 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 	struct qstr *name = &dentry->d_name;
160 	struct page *page;
161 	struct logfs_disk_dentry *dd;
162 	u32 hash = 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 				(filler_t *)logfs_readpage, NULL);
178 		if (IS_ERR(page))
179 			return page;
180 		dd = kmap_atomic(page, KM_USER0);
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, KM_USER0);
186 			page_cache_release(page);
187 			continue;
188 		}
189 
190 		kunmap_atomic(dd, KM_USER0);
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 = dentry->d_inode;
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;
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 	page_cache_release(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 = dentry->d_inode;
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 */
284 #define IMPLICIT_NODES 2
__logfs_readdir(struct file * file,void * buf,filldir_t filldir)285 static int __logfs_readdir(struct file *file, void *buf, filldir_t filldir)
286 {
287 	struct inode *dir = file->f_dentry->d_inode;
288 	loff_t pos = file->f_pos - IMPLICIT_NODES;
289 	struct page *page;
290 	struct logfs_disk_dentry *dd;
291 	int full;
292 
293 	BUG_ON(pos < 0);
294 	for (;; pos++) {
295 		if (beyond_eof(dir, pos))
296 			break;
297 		if (!logfs_exist_block(dir, pos)) {
298 			/* deleted dentry */
299 			pos = dir_seek_data(dir, pos);
300 			continue;
301 		}
302 		page = read_cache_page(dir->i_mapping, pos,
303 				(filler_t *)logfs_readpage, NULL);
304 		if (IS_ERR(page))
305 			return PTR_ERR(page);
306 		dd = kmap(page);
307 		BUG_ON(dd->namelen == 0);
308 
309 		full = filldir(buf, (char *)dd->name, be16_to_cpu(dd->namelen),
310 				pos, be64_to_cpu(dd->ino), dd->type);
311 		kunmap(page);
312 		page_cache_release(page);
313 		if (full)
314 			break;
315 	}
316 
317 	file->f_pos = pos + IMPLICIT_NODES;
318 	return 0;
319 }
320 
logfs_readdir(struct file * file,void * buf,filldir_t filldir)321 static int logfs_readdir(struct file *file, void *buf, filldir_t filldir)
322 {
323 	struct inode *inode = file->f_dentry->d_inode;
324 	ino_t pino = parent_ino(file->f_dentry);
325 	int err;
326 
327 	if (file->f_pos < 0)
328 		return -EINVAL;
329 
330 	if (file->f_pos == 0) {
331 		if (filldir(buf, ".", 1, 1, inode->i_ino, DT_DIR) < 0)
332 			return 0;
333 		file->f_pos++;
334 	}
335 	if (file->f_pos == 1) {
336 		if (filldir(buf, "..", 2, 2, pino, DT_DIR) < 0)
337 			return 0;
338 		file->f_pos++;
339 	}
340 
341 	err = __logfs_readdir(file, buf, filldir);
342 	return err;
343 }
344 
logfs_set_name(struct logfs_disk_dentry * dd,struct qstr * name)345 static void logfs_set_name(struct logfs_disk_dentry *dd, struct qstr *name)
346 {
347 	dd->namelen = cpu_to_be16(name->len);
348 	memcpy(dd->name, name->name, name->len);
349 }
350 
logfs_lookup(struct inode * dir,struct dentry * dentry,struct nameidata * nd)351 static struct dentry *logfs_lookup(struct inode *dir, struct dentry *dentry,
352 		struct nameidata *nd)
353 {
354 	struct page *page;
355 	struct logfs_disk_dentry *dd;
356 	pgoff_t index;
357 	u64 ino = 0;
358 	struct inode *inode;
359 
360 	page = logfs_get_dd_page(dir, dentry);
361 	if (IS_ERR(page))
362 		return ERR_CAST(page);
363 	if (!page) {
364 		d_add(dentry, NULL);
365 		return NULL;
366 	}
367 	index = page->index;
368 	dd = kmap_atomic(page, KM_USER0);
369 	ino = be64_to_cpu(dd->ino);
370 	kunmap_atomic(dd, KM_USER0);
371 	page_cache_release(page);
372 
373 	inode = logfs_iget(dir->i_sb, ino);
374 	if (IS_ERR(inode))
375 		printk(KERN_ERR"LogFS: Cannot read inode #%llx for dentry (%lx, %lx)n",
376 				ino, dir->i_ino, index);
377 	return d_splice_alias(inode, dentry);
378 }
379 
grow_dir(struct inode * dir,loff_t index)380 static void grow_dir(struct inode *dir, loff_t index)
381 {
382 	index = (index + 1) << dir->i_sb->s_blocksize_bits;
383 	if (i_size_read(dir) < index)
384 		i_size_write(dir, index);
385 }
386 
logfs_write_dir(struct inode * dir,struct dentry * dentry,struct inode * inode)387 static int logfs_write_dir(struct inode *dir, struct dentry *dentry,
388 		struct inode *inode)
389 {
390 	struct page *page;
391 	struct logfs_disk_dentry *dd;
392 	u32 hash = hash_32(dentry->d_name.name, dentry->d_name.len, 0);
393 	pgoff_t index;
394 	int round, err;
395 
396 	for (round = 0; round < 20; round++) {
397 		index = hash_index(hash, round);
398 
399 		if (logfs_exist_block(dir, index))
400 			continue;
401 		page = find_or_create_page(dir->i_mapping, index, GFP_KERNEL);
402 		if (!page)
403 			return -ENOMEM;
404 
405 		dd = kmap_atomic(page, KM_USER0);
406 		memset(dd, 0, sizeof(*dd));
407 		dd->ino = cpu_to_be64(inode->i_ino);
408 		dd->type = logfs_type(inode);
409 		logfs_set_name(dd, &dentry->d_name);
410 		kunmap_atomic(dd, KM_USER0);
411 
412 		err = logfs_write_buf(dir, page, WF_LOCK);
413 		unlock_page(page);
414 		page_cache_release(page);
415 		if (!err)
416 			grow_dir(dir, index);
417 		return err;
418 	}
419 	/* FIXME: Is there a better return value?  In most cases neither
420 	 * the filesystem nor the directory are full.  But we have had
421 	 * too many collisions for this particular hash and no fallback.
422 	 */
423 	return -ENOSPC;
424 }
425 
__logfs_create(struct inode * dir,struct dentry * dentry,struct inode * inode,const char * dest,long destlen)426 static int __logfs_create(struct inode *dir, struct dentry *dentry,
427 		struct inode *inode, const char *dest, long destlen)
428 {
429 	struct logfs_super *super = logfs_super(dir->i_sb);
430 	struct logfs_inode *li = logfs_inode(inode);
431 	struct logfs_transaction *ta;
432 	int ret;
433 
434 	ta = kzalloc(sizeof(*ta), GFP_KERNEL);
435 	if (!ta) {
436 		drop_nlink(inode);
437 		iput(inode);
438 		return -ENOMEM;
439 	}
440 
441 	ta->state = CREATE_1;
442 	ta->ino = inode->i_ino;
443 	mutex_lock(&super->s_dirop_mutex);
444 	logfs_add_transaction(inode, ta);
445 
446 	if (dest) {
447 		/* symlink */
448 		ret = logfs_inode_write(inode, dest, destlen, 0, WF_LOCK, NULL);
449 		if (!ret)
450 			ret = write_inode(inode);
451 	} else {
452 		/* creat/mkdir/mknod */
453 		ret = write_inode(inode);
454 	}
455 	if (ret) {
456 		abort_transaction(inode, ta);
457 		li->li_flags |= LOGFS_IF_STILLBORN;
458 		/* FIXME: truncate symlink */
459 		drop_nlink(inode);
460 		iput(inode);
461 		goto out;
462 	}
463 
464 	ta->state = CREATE_2;
465 	logfs_add_transaction(dir, ta);
466 	ret = logfs_write_dir(dir, dentry, inode);
467 	/* sync directory */
468 	if (!ret)
469 		ret = write_inode(dir);
470 
471 	if (ret) {
472 		logfs_del_transaction(dir, ta);
473 		ta->state = CREATE_2;
474 		logfs_add_transaction(inode, ta);
475 		logfs_remove_inode(inode);
476 		iput(inode);
477 		goto out;
478 	}
479 	d_instantiate(dentry, inode);
480 out:
481 	mutex_unlock(&super->s_dirop_mutex);
482 	return ret;
483 }
484 
logfs_mkdir(struct inode * dir,struct dentry * dentry,umode_t mode)485 static int logfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
486 {
487 	struct inode *inode;
488 
489 	/*
490 	 * FIXME: why do we have to fill in S_IFDIR, while the mode is
491 	 * correct for mknod, creat, etc.?  Smells like the vfs *should*
492 	 * do it for us but for some reason fails to do so.
493 	 */
494 	inode = logfs_new_inode(dir, S_IFDIR | mode);
495 	if (IS_ERR(inode))
496 		return PTR_ERR(inode);
497 
498 	inode->i_op = &logfs_dir_iops;
499 	inode->i_fop = &logfs_dir_fops;
500 
501 	return __logfs_create(dir, dentry, inode, NULL, 0);
502 }
503 
logfs_create(struct inode * dir,struct dentry * dentry,umode_t mode,struct nameidata * nd)504 static int logfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
505 		struct nameidata *nd)
506 {
507 	struct inode *inode;
508 
509 	inode = logfs_new_inode(dir, mode);
510 	if (IS_ERR(inode))
511 		return PTR_ERR(inode);
512 
513 	inode->i_op = &logfs_reg_iops;
514 	inode->i_fop = &logfs_reg_fops;
515 	inode->i_mapping->a_ops = &logfs_reg_aops;
516 
517 	return __logfs_create(dir, dentry, inode, NULL, 0);
518 }
519 
logfs_mknod(struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)520 static int logfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
521 		dev_t rdev)
522 {
523 	struct inode *inode;
524 
525 	if (dentry->d_name.len > LOGFS_MAX_NAMELEN)
526 		return -ENAMETOOLONG;
527 
528 	inode = logfs_new_inode(dir, mode);
529 	if (IS_ERR(inode))
530 		return PTR_ERR(inode);
531 
532 	init_special_inode(inode, mode, rdev);
533 
534 	return __logfs_create(dir, dentry, inode, NULL, 0);
535 }
536 
logfs_symlink(struct inode * dir,struct dentry * dentry,const char * target)537 static int logfs_symlink(struct inode *dir, struct dentry *dentry,
538 		const char *target)
539 {
540 	struct inode *inode;
541 	size_t destlen = strlen(target) + 1;
542 
543 	if (destlen > dir->i_sb->s_blocksize)
544 		return -ENAMETOOLONG;
545 
546 	inode = logfs_new_inode(dir, S_IFLNK | 0777);
547 	if (IS_ERR(inode))
548 		return PTR_ERR(inode);
549 
550 	inode->i_op = &logfs_symlink_iops;
551 	inode->i_mapping->a_ops = &logfs_reg_aops;
552 
553 	return __logfs_create(dir, dentry, inode, target, destlen);
554 }
555 
logfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)556 static int logfs_link(struct dentry *old_dentry, struct inode *dir,
557 		struct dentry *dentry)
558 {
559 	struct inode *inode = old_dentry->d_inode;
560 
561 	if (inode->i_nlink >= LOGFS_LINK_MAX)
562 		return -EMLINK;
563 
564 	inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
565 	ihold(inode);
566 	inc_nlink(inode);
567 	mark_inode_dirty_sync(inode);
568 
569 	return __logfs_create(dir, dentry, inode, NULL, 0);
570 }
571 
logfs_get_dd(struct inode * dir,struct dentry * dentry,struct logfs_disk_dentry * dd,loff_t * pos)572 static int logfs_get_dd(struct inode *dir, struct dentry *dentry,
573 		struct logfs_disk_dentry *dd, loff_t *pos)
574 {
575 	struct page *page;
576 	void *map;
577 
578 	page = logfs_get_dd_page(dir, dentry);
579 	if (IS_ERR(page))
580 		return PTR_ERR(page);
581 	*pos = page->index;
582 	map = kmap_atomic(page, KM_USER0);
583 	memcpy(dd, map, sizeof(*dd));
584 	kunmap_atomic(map, KM_USER0);
585 	page_cache_release(page);
586 	return 0;
587 }
588 
logfs_delete_dd(struct inode * dir,loff_t pos)589 static int logfs_delete_dd(struct inode *dir, loff_t pos)
590 {
591 	/*
592 	 * Getting called with pos somewhere beyond eof is either a goofup
593 	 * within this file or means someone maliciously edited the
594 	 * (crc-protected) journal.
595 	 */
596 	BUG_ON(beyond_eof(dir, pos));
597 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
598 	log_dir(" Delete dentry (%lx, %llx)\n", dir->i_ino, pos);
599 	return logfs_delete(dir, pos, NULL);
600 }
601 
602 /*
603  * Cross-directory rename, target does not exist.  Just a little nasty.
604  * Create a new dentry in the target dir, then remove the old dentry,
605  * all the while taking care to remember our operation in the journal.
606  */
logfs_rename_cross(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)607 static int logfs_rename_cross(struct inode *old_dir, struct dentry *old_dentry,
608 			      struct inode *new_dir, struct dentry *new_dentry)
609 {
610 	struct logfs_super *super = logfs_super(old_dir->i_sb);
611 	struct logfs_disk_dentry dd;
612 	struct logfs_transaction *ta;
613 	loff_t pos;
614 	int err;
615 
616 	/* 1. locate source dd */
617 	err = logfs_get_dd(old_dir, old_dentry, &dd, &pos);
618 	if (err)
619 		return err;
620 
621 	ta = kzalloc(sizeof(*ta), GFP_KERNEL);
622 	if (!ta)
623 		return -ENOMEM;
624 
625 	ta->state = CROSS_RENAME_1;
626 	ta->dir = old_dir->i_ino;
627 	ta->pos = pos;
628 
629 	/* 2. write target dd */
630 	mutex_lock(&super->s_dirop_mutex);
631 	logfs_add_transaction(new_dir, ta);
632 	err = logfs_write_dir(new_dir, new_dentry, old_dentry->d_inode);
633 	if (!err)
634 		err = write_inode(new_dir);
635 
636 	if (err) {
637 		super->s_rename_dir = 0;
638 		super->s_rename_pos = 0;
639 		abort_transaction(new_dir, ta);
640 		goto out;
641 	}
642 
643 	/* 3. remove source dd */
644 	ta->state = CROSS_RENAME_2;
645 	logfs_add_transaction(old_dir, ta);
646 	err = logfs_delete_dd(old_dir, pos);
647 	if (!err)
648 		err = write_inode(old_dir);
649 	LOGFS_BUG_ON(err, old_dir->i_sb);
650 out:
651 	mutex_unlock(&super->s_dirop_mutex);
652 	return err;
653 }
654 
logfs_replace_inode(struct inode * dir,struct dentry * dentry,struct logfs_disk_dentry * dd,struct inode * inode)655 static int logfs_replace_inode(struct inode *dir, struct dentry *dentry,
656 		struct logfs_disk_dentry *dd, struct inode *inode)
657 {
658 	loff_t pos;
659 	int err;
660 
661 	err = logfs_get_dd(dir, dentry, dd, &pos);
662 	if (err)
663 		return err;
664 	dd->ino = cpu_to_be64(inode->i_ino);
665 	dd->type = logfs_type(inode);
666 
667 	err = write_dir(dir, dd, pos);
668 	if (err)
669 		return err;
670 	log_dir("Replace dentry (%lx, %llx) %s -> %llx\n", dir->i_ino, pos,
671 			dd->name, be64_to_cpu(dd->ino));
672 	return write_inode(dir);
673 }
674 
675 /* Target dentry exists - the worst case.  We need to attach the source
676  * inode to the target dentry, then remove the orphaned target inode and
677  * source dentry.
678  */
logfs_rename_target(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)679 static int logfs_rename_target(struct inode *old_dir, struct dentry *old_dentry,
680 			       struct inode *new_dir, struct dentry *new_dentry)
681 {
682 	struct logfs_super *super = logfs_super(old_dir->i_sb);
683 	struct inode *old_inode = old_dentry->d_inode;
684 	struct inode *new_inode = new_dentry->d_inode;
685 	int isdir = S_ISDIR(old_inode->i_mode);
686 	struct logfs_disk_dentry dd;
687 	struct logfs_transaction *ta;
688 	loff_t pos;
689 	int err;
690 
691 	BUG_ON(isdir != S_ISDIR(new_inode->i_mode));
692 	if (isdir) {
693 		if (!logfs_empty_dir(new_inode))
694 			return -ENOTEMPTY;
695 	}
696 
697 	/* 1. locate source dd */
698 	err = logfs_get_dd(old_dir, old_dentry, &dd, &pos);
699 	if (err)
700 		return err;
701 
702 	ta = kzalloc(sizeof(*ta), GFP_KERNEL);
703 	if (!ta)
704 		return -ENOMEM;
705 
706 	ta->state = TARGET_RENAME_1;
707 	ta->dir = old_dir->i_ino;
708 	ta->pos = pos;
709 	ta->ino = new_inode->i_ino;
710 
711 	/* 2. attach source inode to target dd */
712 	mutex_lock(&super->s_dirop_mutex);
713 	logfs_add_transaction(new_dir, ta);
714 	err = logfs_replace_inode(new_dir, new_dentry, &dd, old_inode);
715 	if (err) {
716 		super->s_rename_dir = 0;
717 		super->s_rename_pos = 0;
718 		super->s_victim_ino = 0;
719 		abort_transaction(new_dir, ta);
720 		goto out;
721 	}
722 
723 	/* 3. remove source dd */
724 	ta->state = TARGET_RENAME_2;
725 	logfs_add_transaction(old_dir, ta);
726 	err = logfs_delete_dd(old_dir, pos);
727 	if (!err)
728 		err = write_inode(old_dir);
729 	LOGFS_BUG_ON(err, old_dir->i_sb);
730 
731 	/* 4. remove target inode */
732 	ta->state = TARGET_RENAME_3;
733 	logfs_add_transaction(new_inode, ta);
734 	err = logfs_remove_inode(new_inode);
735 
736 out:
737 	mutex_unlock(&super->s_dirop_mutex);
738 	return err;
739 }
740 
logfs_rename(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)741 static int logfs_rename(struct inode *old_dir, struct dentry *old_dentry,
742 			struct inode *new_dir, struct dentry *new_dentry)
743 {
744 	if (new_dentry->d_inode)
745 		return logfs_rename_target(old_dir, old_dentry,
746 					   new_dir, new_dentry);
747 	return logfs_rename_cross(old_dir, old_dentry, new_dir, new_dentry);
748 }
749 
750 /* No locking done here, as this is called before .get_sb() returns. */
logfs_replay_journal(struct super_block * sb)751 int logfs_replay_journal(struct super_block *sb)
752 {
753 	struct logfs_super *super = logfs_super(sb);
754 	struct inode *inode;
755 	u64 ino, pos;
756 	int err;
757 
758 	if (super->s_victim_ino) {
759 		/* delete victim inode */
760 		ino = super->s_victim_ino;
761 		printk(KERN_INFO"LogFS: delete unmapped inode #%llx\n", ino);
762 		inode = logfs_iget(sb, ino);
763 		if (IS_ERR(inode))
764 			goto fail;
765 
766 		LOGFS_BUG_ON(i_size_read(inode) > 0, sb);
767 		super->s_victim_ino = 0;
768 		err = logfs_remove_inode(inode);
769 		iput(inode);
770 		if (err) {
771 			super->s_victim_ino = ino;
772 			goto fail;
773 		}
774 	}
775 	if (super->s_rename_dir) {
776 		/* delete old dd from rename */
777 		ino = super->s_rename_dir;
778 		pos = super->s_rename_pos;
779 		printk(KERN_INFO"LogFS: delete unbacked dentry (%llx, %llx)\n",
780 				ino, pos);
781 		inode = logfs_iget(sb, ino);
782 		if (IS_ERR(inode))
783 			goto fail;
784 
785 		super->s_rename_dir = 0;
786 		super->s_rename_pos = 0;
787 		err = logfs_delete_dd(inode, pos);
788 		iput(inode);
789 		if (err) {
790 			super->s_rename_dir = ino;
791 			super->s_rename_pos = pos;
792 			goto fail;
793 		}
794 	}
795 	return 0;
796 fail:
797 	LOGFS_BUG(sb);
798 	return -EIO;
799 }
800 
801 const struct inode_operations logfs_symlink_iops = {
802 	.readlink	= generic_readlink,
803 	.follow_link	= page_follow_link_light,
804 };
805 
806 const struct inode_operations logfs_dir_iops = {
807 	.create		= logfs_create,
808 	.link		= logfs_link,
809 	.lookup		= logfs_lookup,
810 	.mkdir		= logfs_mkdir,
811 	.mknod		= logfs_mknod,
812 	.rename		= logfs_rename,
813 	.rmdir		= logfs_rmdir,
814 	.symlink	= logfs_symlink,
815 	.unlink		= logfs_unlink,
816 };
817 const struct file_operations logfs_dir_fops = {
818 	.fsync		= logfs_fsync,
819 	.unlocked_ioctl	= logfs_ioctl,
820 	.readdir	= logfs_readdir,
821 	.read		= generic_read_dir,
822 	.llseek		= default_llseek,
823 };
824