xref: /openbmc/linux/fs/ext4/inode.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  *  linux/fs/ext4/inode.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/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Goal-directed block allocation by Stephen Tweedie
16  *	(sct@redhat.com), 1993, 1998
17  *  Big-endian to little-endian byte-swapping/bitmaps by
18  *        David S. Miller (davem@caip.rutgers.edu), 1995
19  *  64-bit file support on 64-bit platforms by Jakub Jelinek
20  *	(jj@sunsite.ms.mff.cuni.cz)
21  *
22  *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23  */
24 
25 #include <linux/module.h>
26 #include <linux/fs.h>
27 #include <linux/time.h>
28 #include <linux/ext4_jbd2.h>
29 #include <linux/jbd2.h>
30 #include <linux/highuid.h>
31 #include <linux/pagemap.h>
32 #include <linux/quotaops.h>
33 #include <linux/string.h>
34 #include <linux/buffer_head.h>
35 #include <linux/writeback.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include "xattr.h"
40 #include "acl.h"
41 
42 /*
43  * Test whether an inode is a fast symlink.
44  */
45 static int ext4_inode_is_fast_symlink(struct inode *inode)
46 {
47 	int ea_blocks = EXT4_I(inode)->i_file_acl ?
48 		(inode->i_sb->s_blocksize >> 9) : 0;
49 
50 	return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
51 }
52 
53 /*
54  * The ext4 forget function must perform a revoke if we are freeing data
55  * which has been journaled.  Metadata (eg. indirect blocks) must be
56  * revoked in all cases.
57  *
58  * "bh" may be NULL: a metadata block may have been freed from memory
59  * but there may still be a record of it in the journal, and that record
60  * still needs to be revoked.
61  */
62 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
63 			struct buffer_head *bh, ext4_fsblk_t blocknr)
64 {
65 	int err;
66 
67 	might_sleep();
68 
69 	BUFFER_TRACE(bh, "enter");
70 
71 	jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
72 		  "data mode %lx\n",
73 		  bh, is_metadata, inode->i_mode,
74 		  test_opt(inode->i_sb, DATA_FLAGS));
75 
76 	/* Never use the revoke function if we are doing full data
77 	 * journaling: there is no need to, and a V1 superblock won't
78 	 * support it.  Otherwise, only skip the revoke on un-journaled
79 	 * data blocks. */
80 
81 	if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
82 	    (!is_metadata && !ext4_should_journal_data(inode))) {
83 		if (bh) {
84 			BUFFER_TRACE(bh, "call jbd2_journal_forget");
85 			return ext4_journal_forget(handle, bh);
86 		}
87 		return 0;
88 	}
89 
90 	/*
91 	 * data!=journal && (is_metadata || should_journal_data(inode))
92 	 */
93 	BUFFER_TRACE(bh, "call ext4_journal_revoke");
94 	err = ext4_journal_revoke(handle, blocknr, bh);
95 	if (err)
96 		ext4_abort(inode->i_sb, __FUNCTION__,
97 			   "error %d when attempting revoke", err);
98 	BUFFER_TRACE(bh, "exit");
99 	return err;
100 }
101 
102 /*
103  * Work out how many blocks we need to proceed with the next chunk of a
104  * truncate transaction.
105  */
106 static unsigned long blocks_for_truncate(struct inode *inode)
107 {
108 	unsigned long needed;
109 
110 	needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
111 
112 	/* Give ourselves just enough room to cope with inodes in which
113 	 * i_blocks is corrupt: we've seen disk corruptions in the past
114 	 * which resulted in random data in an inode which looked enough
115 	 * like a regular file for ext4 to try to delete it.  Things
116 	 * will go a bit crazy if that happens, but at least we should
117 	 * try not to panic the whole kernel. */
118 	if (needed < 2)
119 		needed = 2;
120 
121 	/* But we need to bound the transaction so we don't overflow the
122 	 * journal. */
123 	if (needed > EXT4_MAX_TRANS_DATA)
124 		needed = EXT4_MAX_TRANS_DATA;
125 
126 	return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
127 }
128 
129 /*
130  * Truncate transactions can be complex and absolutely huge.  So we need to
131  * be able to restart the transaction at a conventient checkpoint to make
132  * sure we don't overflow the journal.
133  *
134  * start_transaction gets us a new handle for a truncate transaction,
135  * and extend_transaction tries to extend the existing one a bit.  If
136  * extend fails, we need to propagate the failure up and restart the
137  * transaction in the top-level truncate loop. --sct
138  */
139 static handle_t *start_transaction(struct inode *inode)
140 {
141 	handle_t *result;
142 
143 	result = ext4_journal_start(inode, blocks_for_truncate(inode));
144 	if (!IS_ERR(result))
145 		return result;
146 
147 	ext4_std_error(inode->i_sb, PTR_ERR(result));
148 	return result;
149 }
150 
151 /*
152  * Try to extend this transaction for the purposes of truncation.
153  *
154  * Returns 0 if we managed to create more room.  If we can't create more
155  * room, and the transaction must be restarted we return 1.
156  */
157 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
158 {
159 	if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
160 		return 0;
161 	if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
162 		return 0;
163 	return 1;
164 }
165 
166 /*
167  * Restart the transaction associated with *handle.  This does a commit,
168  * so before we call here everything must be consistently dirtied against
169  * this transaction.
170  */
171 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
172 {
173 	jbd_debug(2, "restarting handle %p\n", handle);
174 	return ext4_journal_restart(handle, blocks_for_truncate(inode));
175 }
176 
177 /*
178  * Called at the last iput() if i_nlink is zero.
179  */
180 void ext4_delete_inode (struct inode * inode)
181 {
182 	handle_t *handle;
183 
184 	truncate_inode_pages(&inode->i_data, 0);
185 
186 	if (is_bad_inode(inode))
187 		goto no_delete;
188 
189 	handle = start_transaction(inode);
190 	if (IS_ERR(handle)) {
191 		/*
192 		 * If we're going to skip the normal cleanup, we still need to
193 		 * make sure that the in-core orphan linked list is properly
194 		 * cleaned up.
195 		 */
196 		ext4_orphan_del(NULL, inode);
197 		goto no_delete;
198 	}
199 
200 	if (IS_SYNC(inode))
201 		handle->h_sync = 1;
202 	inode->i_size = 0;
203 	if (inode->i_blocks)
204 		ext4_truncate(inode);
205 	/*
206 	 * Kill off the orphan record which ext4_truncate created.
207 	 * AKPM: I think this can be inside the above `if'.
208 	 * Note that ext4_orphan_del() has to be able to cope with the
209 	 * deletion of a non-existent orphan - this is because we don't
210 	 * know if ext4_truncate() actually created an orphan record.
211 	 * (Well, we could do this if we need to, but heck - it works)
212 	 */
213 	ext4_orphan_del(handle, inode);
214 	EXT4_I(inode)->i_dtime	= get_seconds();
215 
216 	/*
217 	 * One subtle ordering requirement: if anything has gone wrong
218 	 * (transaction abort, IO errors, whatever), then we can still
219 	 * do these next steps (the fs will already have been marked as
220 	 * having errors), but we can't free the inode if the mark_dirty
221 	 * fails.
222 	 */
223 	if (ext4_mark_inode_dirty(handle, inode))
224 		/* If that failed, just do the required in-core inode clear. */
225 		clear_inode(inode);
226 	else
227 		ext4_free_inode(handle, inode);
228 	ext4_journal_stop(handle);
229 	return;
230 no_delete:
231 	clear_inode(inode);	/* We must guarantee clearing of inode... */
232 }
233 
234 typedef struct {
235 	__le32	*p;
236 	__le32	key;
237 	struct buffer_head *bh;
238 } Indirect;
239 
240 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
241 {
242 	p->key = *(p->p = v);
243 	p->bh = bh;
244 }
245 
246 static int verify_chain(Indirect *from, Indirect *to)
247 {
248 	while (from <= to && from->key == *from->p)
249 		from++;
250 	return (from > to);
251 }
252 
253 /**
254  *	ext4_block_to_path - parse the block number into array of offsets
255  *	@inode: inode in question (we are only interested in its superblock)
256  *	@i_block: block number to be parsed
257  *	@offsets: array to store the offsets in
258  *      @boundary: set this non-zero if the referred-to block is likely to be
259  *             followed (on disk) by an indirect block.
260  *
261  *	To store the locations of file's data ext4 uses a data structure common
262  *	for UNIX filesystems - tree of pointers anchored in the inode, with
263  *	data blocks at leaves and indirect blocks in intermediate nodes.
264  *	This function translates the block number into path in that tree -
265  *	return value is the path length and @offsets[n] is the offset of
266  *	pointer to (n+1)th node in the nth one. If @block is out of range
267  *	(negative or too large) warning is printed and zero returned.
268  *
269  *	Note: function doesn't find node addresses, so no IO is needed. All
270  *	we need to know is the capacity of indirect blocks (taken from the
271  *	inode->i_sb).
272  */
273 
274 /*
275  * Portability note: the last comparison (check that we fit into triple
276  * indirect block) is spelled differently, because otherwise on an
277  * architecture with 32-bit longs and 8Kb pages we might get into trouble
278  * if our filesystem had 8Kb blocks. We might use long long, but that would
279  * kill us on x86. Oh, well, at least the sign propagation does not matter -
280  * i_block would have to be negative in the very beginning, so we would not
281  * get there at all.
282  */
283 
284 static int ext4_block_to_path(struct inode *inode,
285 			long i_block, int offsets[4], int *boundary)
286 {
287 	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
288 	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
289 	const long direct_blocks = EXT4_NDIR_BLOCKS,
290 		indirect_blocks = ptrs,
291 		double_blocks = (1 << (ptrs_bits * 2));
292 	int n = 0;
293 	int final = 0;
294 
295 	if (i_block < 0) {
296 		ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
297 	} else if (i_block < direct_blocks) {
298 		offsets[n++] = i_block;
299 		final = direct_blocks;
300 	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
301 		offsets[n++] = EXT4_IND_BLOCK;
302 		offsets[n++] = i_block;
303 		final = ptrs;
304 	} else if ((i_block -= indirect_blocks) < double_blocks) {
305 		offsets[n++] = EXT4_DIND_BLOCK;
306 		offsets[n++] = i_block >> ptrs_bits;
307 		offsets[n++] = i_block & (ptrs - 1);
308 		final = ptrs;
309 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
310 		offsets[n++] = EXT4_TIND_BLOCK;
311 		offsets[n++] = i_block >> (ptrs_bits * 2);
312 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
313 		offsets[n++] = i_block & (ptrs - 1);
314 		final = ptrs;
315 	} else {
316 		ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big");
317 	}
318 	if (boundary)
319 		*boundary = final - 1 - (i_block & (ptrs - 1));
320 	return n;
321 }
322 
323 /**
324  *	ext4_get_branch - read the chain of indirect blocks leading to data
325  *	@inode: inode in question
326  *	@depth: depth of the chain (1 - direct pointer, etc.)
327  *	@offsets: offsets of pointers in inode/indirect blocks
328  *	@chain: place to store the result
329  *	@err: here we store the error value
330  *
331  *	Function fills the array of triples <key, p, bh> and returns %NULL
332  *	if everything went OK or the pointer to the last filled triple
333  *	(incomplete one) otherwise. Upon the return chain[i].key contains
334  *	the number of (i+1)-th block in the chain (as it is stored in memory,
335  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
336  *	number (it points into struct inode for i==0 and into the bh->b_data
337  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
338  *	block for i>0 and NULL for i==0. In other words, it holds the block
339  *	numbers of the chain, addresses they were taken from (and where we can
340  *	verify that chain did not change) and buffer_heads hosting these
341  *	numbers.
342  *
343  *	Function stops when it stumbles upon zero pointer (absent block)
344  *		(pointer to last triple returned, *@err == 0)
345  *	or when it gets an IO error reading an indirect block
346  *		(ditto, *@err == -EIO)
347  *	or when it notices that chain had been changed while it was reading
348  *		(ditto, *@err == -EAGAIN)
349  *	or when it reads all @depth-1 indirect blocks successfully and finds
350  *	the whole chain, all way to the data (returns %NULL, *err == 0).
351  */
352 static Indirect *ext4_get_branch(struct inode *inode, int depth, int *offsets,
353 				 Indirect chain[4], int *err)
354 {
355 	struct super_block *sb = inode->i_sb;
356 	Indirect *p = chain;
357 	struct buffer_head *bh;
358 
359 	*err = 0;
360 	/* i_data is not going away, no lock needed */
361 	add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
362 	if (!p->key)
363 		goto no_block;
364 	while (--depth) {
365 		bh = sb_bread(sb, le32_to_cpu(p->key));
366 		if (!bh)
367 			goto failure;
368 		/* Reader: pointers */
369 		if (!verify_chain(chain, p))
370 			goto changed;
371 		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
372 		/* Reader: end */
373 		if (!p->key)
374 			goto no_block;
375 	}
376 	return NULL;
377 
378 changed:
379 	brelse(bh);
380 	*err = -EAGAIN;
381 	goto no_block;
382 failure:
383 	*err = -EIO;
384 no_block:
385 	return p;
386 }
387 
388 /**
389  *	ext4_find_near - find a place for allocation with sufficient locality
390  *	@inode: owner
391  *	@ind: descriptor of indirect block.
392  *
393  *	This function returns the prefered place for block allocation.
394  *	It is used when heuristic for sequential allocation fails.
395  *	Rules are:
396  *	  + if there is a block to the left of our position - allocate near it.
397  *	  + if pointer will live in indirect block - allocate near that block.
398  *	  + if pointer will live in inode - allocate in the same
399  *	    cylinder group.
400  *
401  * In the latter case we colour the starting block by the callers PID to
402  * prevent it from clashing with concurrent allocations for a different inode
403  * in the same block group.   The PID is used here so that functionally related
404  * files will be close-by on-disk.
405  *
406  *	Caller must make sure that @ind is valid and will stay that way.
407  */
408 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
409 {
410 	struct ext4_inode_info *ei = EXT4_I(inode);
411 	__le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
412 	__le32 *p;
413 	ext4_fsblk_t bg_start;
414 	ext4_grpblk_t colour;
415 
416 	/* Try to find previous block */
417 	for (p = ind->p - 1; p >= start; p--) {
418 		if (*p)
419 			return le32_to_cpu(*p);
420 	}
421 
422 	/* No such thing, so let's try location of indirect block */
423 	if (ind->bh)
424 		return ind->bh->b_blocknr;
425 
426 	/*
427 	 * It is going to be referred to from the inode itself? OK, just put it
428 	 * into the same cylinder group then.
429 	 */
430 	bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
431 	colour = (current->pid % 16) *
432 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
433 	return bg_start + colour;
434 }
435 
436 /**
437  *	ext4_find_goal - find a prefered place for allocation.
438  *	@inode: owner
439  *	@block:  block we want
440  *	@chain:  chain of indirect blocks
441  *	@partial: pointer to the last triple within a chain
442  *	@goal:	place to store the result.
443  *
444  *	Normally this function find the prefered place for block allocation,
445  *	stores it in *@goal and returns zero.
446  */
447 
448 static ext4_fsblk_t ext4_find_goal(struct inode *inode, long block,
449 		Indirect chain[4], Indirect *partial)
450 {
451 	struct ext4_block_alloc_info *block_i;
452 
453 	block_i =  EXT4_I(inode)->i_block_alloc_info;
454 
455 	/*
456 	 * try the heuristic for sequential allocation,
457 	 * failing that at least try to get decent locality.
458 	 */
459 	if (block_i && (block == block_i->last_alloc_logical_block + 1)
460 		&& (block_i->last_alloc_physical_block != 0)) {
461 		return block_i->last_alloc_physical_block + 1;
462 	}
463 
464 	return ext4_find_near(inode, partial);
465 }
466 
467 /**
468  *	ext4_blks_to_allocate: Look up the block map and count the number
469  *	of direct blocks need to be allocated for the given branch.
470  *
471  *	@branch: chain of indirect blocks
472  *	@k: number of blocks need for indirect blocks
473  *	@blks: number of data blocks to be mapped.
474  *	@blocks_to_boundary:  the offset in the indirect block
475  *
476  *	return the total number of blocks to be allocate, including the
477  *	direct and indirect blocks.
478  */
479 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
480 		int blocks_to_boundary)
481 {
482 	unsigned long count = 0;
483 
484 	/*
485 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
486 	 * then it's clear blocks on that path have not allocated
487 	 */
488 	if (k > 0) {
489 		/* right now we don't handle cross boundary allocation */
490 		if (blks < blocks_to_boundary + 1)
491 			count += blks;
492 		else
493 			count += blocks_to_boundary + 1;
494 		return count;
495 	}
496 
497 	count++;
498 	while (count < blks && count <= blocks_to_boundary &&
499 		le32_to_cpu(*(branch[0].p + count)) == 0) {
500 		count++;
501 	}
502 	return count;
503 }
504 
505 /**
506  *	ext4_alloc_blocks: multiple allocate blocks needed for a branch
507  *	@indirect_blks: the number of blocks need to allocate for indirect
508  *			blocks
509  *
510  *	@new_blocks: on return it will store the new block numbers for
511  *	the indirect blocks(if needed) and the first direct block,
512  *	@blks:	on return it will store the total number of allocated
513  *		direct blocks
514  */
515 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
516 			ext4_fsblk_t goal, int indirect_blks, int blks,
517 			ext4_fsblk_t new_blocks[4], int *err)
518 {
519 	int target, i;
520 	unsigned long count = 0;
521 	int index = 0;
522 	ext4_fsblk_t current_block = 0;
523 	int ret = 0;
524 
525 	/*
526 	 * Here we try to allocate the requested multiple blocks at once,
527 	 * on a best-effort basis.
528 	 * To build a branch, we should allocate blocks for
529 	 * the indirect blocks(if not allocated yet), and at least
530 	 * the first direct block of this branch.  That's the
531 	 * minimum number of blocks need to allocate(required)
532 	 */
533 	target = blks + indirect_blks;
534 
535 	while (1) {
536 		count = target;
537 		/* allocating blocks for indirect blocks and direct blocks */
538 		current_block = ext4_new_blocks(handle,inode,goal,&count,err);
539 		if (*err)
540 			goto failed_out;
541 
542 		target -= count;
543 		/* allocate blocks for indirect blocks */
544 		while (index < indirect_blks && count) {
545 			new_blocks[index++] = current_block++;
546 			count--;
547 		}
548 
549 		if (count > 0)
550 			break;
551 	}
552 
553 	/* save the new block number for the first direct block */
554 	new_blocks[index] = current_block;
555 
556 	/* total number of blocks allocated for direct blocks */
557 	ret = count;
558 	*err = 0;
559 	return ret;
560 failed_out:
561 	for (i = 0; i <index; i++)
562 		ext4_free_blocks(handle, inode, new_blocks[i], 1);
563 	return ret;
564 }
565 
566 /**
567  *	ext4_alloc_branch - allocate and set up a chain of blocks.
568  *	@inode: owner
569  *	@indirect_blks: number of allocated indirect blocks
570  *	@blks: number of allocated direct blocks
571  *	@offsets: offsets (in the blocks) to store the pointers to next.
572  *	@branch: place to store the chain in.
573  *
574  *	This function allocates blocks, zeroes out all but the last one,
575  *	links them into chain and (if we are synchronous) writes them to disk.
576  *	In other words, it prepares a branch that can be spliced onto the
577  *	inode. It stores the information about that chain in the branch[], in
578  *	the same format as ext4_get_branch() would do. We are calling it after
579  *	we had read the existing part of chain and partial points to the last
580  *	triple of that (one with zero ->key). Upon the exit we have the same
581  *	picture as after the successful ext4_get_block(), except that in one
582  *	place chain is disconnected - *branch->p is still zero (we did not
583  *	set the last link), but branch->key contains the number that should
584  *	be placed into *branch->p to fill that gap.
585  *
586  *	If allocation fails we free all blocks we've allocated (and forget
587  *	their buffer_heads) and return the error value the from failed
588  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
589  *	as described above and return 0.
590  */
591 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
592 			int indirect_blks, int *blks, ext4_fsblk_t goal,
593 			int *offsets, Indirect *branch)
594 {
595 	int blocksize = inode->i_sb->s_blocksize;
596 	int i, n = 0;
597 	int err = 0;
598 	struct buffer_head *bh;
599 	int num;
600 	ext4_fsblk_t new_blocks[4];
601 	ext4_fsblk_t current_block;
602 
603 	num = ext4_alloc_blocks(handle, inode, goal, indirect_blks,
604 				*blks, new_blocks, &err);
605 	if (err)
606 		return err;
607 
608 	branch[0].key = cpu_to_le32(new_blocks[0]);
609 	/*
610 	 * metadata blocks and data blocks are allocated.
611 	 */
612 	for (n = 1; n <= indirect_blks;  n++) {
613 		/*
614 		 * Get buffer_head for parent block, zero it out
615 		 * and set the pointer to new one, then send
616 		 * parent to disk.
617 		 */
618 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
619 		branch[n].bh = bh;
620 		lock_buffer(bh);
621 		BUFFER_TRACE(bh, "call get_create_access");
622 		err = ext4_journal_get_create_access(handle, bh);
623 		if (err) {
624 			unlock_buffer(bh);
625 			brelse(bh);
626 			goto failed;
627 		}
628 
629 		memset(bh->b_data, 0, blocksize);
630 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
631 		branch[n].key = cpu_to_le32(new_blocks[n]);
632 		*branch[n].p = branch[n].key;
633 		if ( n == indirect_blks) {
634 			current_block = new_blocks[n];
635 			/*
636 			 * End of chain, update the last new metablock of
637 			 * the chain to point to the new allocated
638 			 * data blocks numbers
639 			 */
640 			for (i=1; i < num; i++)
641 				*(branch[n].p + i) = cpu_to_le32(++current_block);
642 		}
643 		BUFFER_TRACE(bh, "marking uptodate");
644 		set_buffer_uptodate(bh);
645 		unlock_buffer(bh);
646 
647 		BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
648 		err = ext4_journal_dirty_metadata(handle, bh);
649 		if (err)
650 			goto failed;
651 	}
652 	*blks = num;
653 	return err;
654 failed:
655 	/* Allocation failed, free what we already allocated */
656 	for (i = 1; i <= n ; i++) {
657 		BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
658 		ext4_journal_forget(handle, branch[i].bh);
659 	}
660 	for (i = 0; i <indirect_blks; i++)
661 		ext4_free_blocks(handle, inode, new_blocks[i], 1);
662 
663 	ext4_free_blocks(handle, inode, new_blocks[i], num);
664 
665 	return err;
666 }
667 
668 /**
669  * ext4_splice_branch - splice the allocated branch onto inode.
670  * @inode: owner
671  * @block: (logical) number of block we are adding
672  * @chain: chain of indirect blocks (with a missing link - see
673  *	ext4_alloc_branch)
674  * @where: location of missing link
675  * @num:   number of indirect blocks we are adding
676  * @blks:  number of direct blocks we are adding
677  *
678  * This function fills the missing link and does all housekeeping needed in
679  * inode (->i_blocks, etc.). In case of success we end up with the full
680  * chain to new block and return 0.
681  */
682 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
683 			long block, Indirect *where, int num, int blks)
684 {
685 	int i;
686 	int err = 0;
687 	struct ext4_block_alloc_info *block_i;
688 	ext4_fsblk_t current_block;
689 
690 	block_i = EXT4_I(inode)->i_block_alloc_info;
691 	/*
692 	 * If we're splicing into a [td]indirect block (as opposed to the
693 	 * inode) then we need to get write access to the [td]indirect block
694 	 * before the splice.
695 	 */
696 	if (where->bh) {
697 		BUFFER_TRACE(where->bh, "get_write_access");
698 		err = ext4_journal_get_write_access(handle, where->bh);
699 		if (err)
700 			goto err_out;
701 	}
702 	/* That's it */
703 
704 	*where->p = where->key;
705 
706 	/*
707 	 * Update the host buffer_head or inode to point to more just allocated
708 	 * direct blocks blocks
709 	 */
710 	if (num == 0 && blks > 1) {
711 		current_block = le32_to_cpu(where->key) + 1;
712 		for (i = 1; i < blks; i++)
713 			*(where->p + i ) = cpu_to_le32(current_block++);
714 	}
715 
716 	/*
717 	 * update the most recently allocated logical & physical block
718 	 * in i_block_alloc_info, to assist find the proper goal block for next
719 	 * allocation
720 	 */
721 	if (block_i) {
722 		block_i->last_alloc_logical_block = block + blks - 1;
723 		block_i->last_alloc_physical_block =
724 				le32_to_cpu(where[num].key) + blks - 1;
725 	}
726 
727 	/* We are done with atomic stuff, now do the rest of housekeeping */
728 
729 	inode->i_ctime = CURRENT_TIME_SEC;
730 	ext4_mark_inode_dirty(handle, inode);
731 
732 	/* had we spliced it onto indirect block? */
733 	if (where->bh) {
734 		/*
735 		 * If we spliced it onto an indirect block, we haven't
736 		 * altered the inode.  Note however that if it is being spliced
737 		 * onto an indirect block at the very end of the file (the
738 		 * file is growing) then we *will* alter the inode to reflect
739 		 * the new i_size.  But that is not done here - it is done in
740 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
741 		 */
742 		jbd_debug(5, "splicing indirect only\n");
743 		BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
744 		err = ext4_journal_dirty_metadata(handle, where->bh);
745 		if (err)
746 			goto err_out;
747 	} else {
748 		/*
749 		 * OK, we spliced it into the inode itself on a direct block.
750 		 * Inode was dirtied above.
751 		 */
752 		jbd_debug(5, "splicing direct\n");
753 	}
754 	return err;
755 
756 err_out:
757 	for (i = 1; i <= num; i++) {
758 		BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
759 		ext4_journal_forget(handle, where[i].bh);
760 		ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
761 	}
762 	ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
763 
764 	return err;
765 }
766 
767 /*
768  * Allocation strategy is simple: if we have to allocate something, we will
769  * have to go the whole way to leaf. So let's do it before attaching anything
770  * to tree, set linkage between the newborn blocks, write them if sync is
771  * required, recheck the path, free and repeat if check fails, otherwise
772  * set the last missing link (that will protect us from any truncate-generated
773  * removals - all blocks on the path are immune now) and possibly force the
774  * write on the parent block.
775  * That has a nice additional property: no special recovery from the failed
776  * allocations is needed - we simply release blocks and do not touch anything
777  * reachable from inode.
778  *
779  * `handle' can be NULL if create == 0.
780  *
781  * The BKL may not be held on entry here.  Be sure to take it early.
782  * return > 0, # of blocks mapped or allocated.
783  * return = 0, if plain lookup failed.
784  * return < 0, error case.
785  */
786 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
787 		sector_t iblock, unsigned long maxblocks,
788 		struct buffer_head *bh_result,
789 		int create, int extend_disksize)
790 {
791 	int err = -EIO;
792 	int offsets[4];
793 	Indirect chain[4];
794 	Indirect *partial;
795 	ext4_fsblk_t goal;
796 	int indirect_blks;
797 	int blocks_to_boundary = 0;
798 	int depth;
799 	struct ext4_inode_info *ei = EXT4_I(inode);
800 	int count = 0;
801 	ext4_fsblk_t first_block = 0;
802 
803 
804 	J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
805 	J_ASSERT(handle != NULL || create == 0);
806 	depth = ext4_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
807 
808 	if (depth == 0)
809 		goto out;
810 
811 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
812 
813 	/* Simplest case - block found, no allocation needed */
814 	if (!partial) {
815 		first_block = le32_to_cpu(chain[depth - 1].key);
816 		clear_buffer_new(bh_result);
817 		count++;
818 		/*map more blocks*/
819 		while (count < maxblocks && count <= blocks_to_boundary) {
820 			ext4_fsblk_t blk;
821 
822 			if (!verify_chain(chain, partial)) {
823 				/*
824 				 * Indirect block might be removed by
825 				 * truncate while we were reading it.
826 				 * Handling of that case: forget what we've
827 				 * got now. Flag the err as EAGAIN, so it
828 				 * will reread.
829 				 */
830 				err = -EAGAIN;
831 				count = 0;
832 				break;
833 			}
834 			blk = le32_to_cpu(*(chain[depth-1].p + count));
835 
836 			if (blk == first_block + count)
837 				count++;
838 			else
839 				break;
840 		}
841 		if (err != -EAGAIN)
842 			goto got_it;
843 	}
844 
845 	/* Next simple case - plain lookup or failed read of indirect block */
846 	if (!create || err == -EIO)
847 		goto cleanup;
848 
849 	mutex_lock(&ei->truncate_mutex);
850 
851 	/*
852 	 * If the indirect block is missing while we are reading
853 	 * the chain(ext4_get_branch() returns -EAGAIN err), or
854 	 * if the chain has been changed after we grab the semaphore,
855 	 * (either because another process truncated this branch, or
856 	 * another get_block allocated this branch) re-grab the chain to see if
857 	 * the request block has been allocated or not.
858 	 *
859 	 * Since we already block the truncate/other get_block
860 	 * at this point, we will have the current copy of the chain when we
861 	 * splice the branch into the tree.
862 	 */
863 	if (err == -EAGAIN || !verify_chain(chain, partial)) {
864 		while (partial > chain) {
865 			brelse(partial->bh);
866 			partial--;
867 		}
868 		partial = ext4_get_branch(inode, depth, offsets, chain, &err);
869 		if (!partial) {
870 			count++;
871 			mutex_unlock(&ei->truncate_mutex);
872 			if (err)
873 				goto cleanup;
874 			clear_buffer_new(bh_result);
875 			goto got_it;
876 		}
877 	}
878 
879 	/*
880 	 * Okay, we need to do block allocation.  Lazily initialize the block
881 	 * allocation info here if necessary
882 	*/
883 	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
884 		ext4_init_block_alloc_info(inode);
885 
886 	goal = ext4_find_goal(inode, iblock, chain, partial);
887 
888 	/* the number of blocks need to allocate for [d,t]indirect blocks */
889 	indirect_blks = (chain + depth) - partial - 1;
890 
891 	/*
892 	 * Next look up the indirect map to count the totoal number of
893 	 * direct blocks to allocate for this branch.
894 	 */
895 	count = ext4_blks_to_allocate(partial, indirect_blks,
896 					maxblocks, blocks_to_boundary);
897 	/*
898 	 * Block out ext4_truncate while we alter the tree
899 	 */
900 	err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal,
901 				offsets + (partial - chain), partial);
902 
903 	/*
904 	 * The ext4_splice_branch call will free and forget any buffers
905 	 * on the new chain if there is a failure, but that risks using
906 	 * up transaction credits, especially for bitmaps where the
907 	 * credits cannot be returned.  Can we handle this somehow?  We
908 	 * may need to return -EAGAIN upwards in the worst case.  --sct
909 	 */
910 	if (!err)
911 		err = ext4_splice_branch(handle, inode, iblock,
912 					partial, indirect_blks, count);
913 	/*
914 	 * i_disksize growing is protected by truncate_mutex.  Don't forget to
915 	 * protect it if you're about to implement concurrent
916 	 * ext4_get_block() -bzzz
917 	*/
918 	if (!err && extend_disksize && inode->i_size > ei->i_disksize)
919 		ei->i_disksize = inode->i_size;
920 	mutex_unlock(&ei->truncate_mutex);
921 	if (err)
922 		goto cleanup;
923 
924 	set_buffer_new(bh_result);
925 got_it:
926 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
927 	if (count > blocks_to_boundary)
928 		set_buffer_boundary(bh_result);
929 	err = count;
930 	/* Clean up and exit */
931 	partial = chain + depth - 1;	/* the whole chain */
932 cleanup:
933 	while (partial > chain) {
934 		BUFFER_TRACE(partial->bh, "call brelse");
935 		brelse(partial->bh);
936 		partial--;
937 	}
938 	BUFFER_TRACE(bh_result, "returned");
939 out:
940 	return err;
941 }
942 
943 #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
944 
945 static int ext4_get_block(struct inode *inode, sector_t iblock,
946 			struct buffer_head *bh_result, int create)
947 {
948 	handle_t *handle = ext4_journal_current_handle();
949 	int ret = 0;
950 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
951 
952 	if (!create)
953 		goto get_block;		/* A read */
954 
955 	if (max_blocks == 1)
956 		goto get_block;		/* A single block get */
957 
958 	if (handle->h_transaction->t_state == T_LOCKED) {
959 		/*
960 		 * Huge direct-io writes can hold off commits for long
961 		 * periods of time.  Let this commit run.
962 		 */
963 		ext4_journal_stop(handle);
964 		handle = ext4_journal_start(inode, DIO_CREDITS);
965 		if (IS_ERR(handle))
966 			ret = PTR_ERR(handle);
967 		goto get_block;
968 	}
969 
970 	if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) {
971 		/*
972 		 * Getting low on buffer credits...
973 		 */
974 		ret = ext4_journal_extend(handle, DIO_CREDITS);
975 		if (ret > 0) {
976 			/*
977 			 * Couldn't extend the transaction.  Start a new one.
978 			 */
979 			ret = ext4_journal_restart(handle, DIO_CREDITS);
980 		}
981 	}
982 
983 get_block:
984 	if (ret == 0) {
985 		ret = ext4_get_blocks_wrap(handle, inode, iblock,
986 					max_blocks, bh_result, create, 0);
987 		if (ret > 0) {
988 			bh_result->b_size = (ret << inode->i_blkbits);
989 			ret = 0;
990 		}
991 	}
992 	return ret;
993 }
994 
995 /*
996  * `handle' can be NULL if create is zero
997  */
998 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
999 				long block, int create, int *errp)
1000 {
1001 	struct buffer_head dummy;
1002 	int fatal = 0, err;
1003 
1004 	J_ASSERT(handle != NULL || create == 0);
1005 
1006 	dummy.b_state = 0;
1007 	dummy.b_blocknr = -1000;
1008 	buffer_trace_init(&dummy.b_history);
1009 	err = ext4_get_blocks_wrap(handle, inode, block, 1,
1010 					&dummy, create, 1);
1011 	/*
1012 	 * ext4_get_blocks_handle() returns number of blocks
1013 	 * mapped. 0 in case of a HOLE.
1014 	 */
1015 	if (err > 0) {
1016 		if (err > 1)
1017 			WARN_ON(1);
1018 		err = 0;
1019 	}
1020 	*errp = err;
1021 	if (!err && buffer_mapped(&dummy)) {
1022 		struct buffer_head *bh;
1023 		bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1024 		if (!bh) {
1025 			*errp = -EIO;
1026 			goto err;
1027 		}
1028 		if (buffer_new(&dummy)) {
1029 			J_ASSERT(create != 0);
1030 			J_ASSERT(handle != 0);
1031 
1032 			/*
1033 			 * Now that we do not always journal data, we should
1034 			 * keep in mind whether this should always journal the
1035 			 * new buffer as metadata.  For now, regular file
1036 			 * writes use ext4_get_block instead, so it's not a
1037 			 * problem.
1038 			 */
1039 			lock_buffer(bh);
1040 			BUFFER_TRACE(bh, "call get_create_access");
1041 			fatal = ext4_journal_get_create_access(handle, bh);
1042 			if (!fatal && !buffer_uptodate(bh)) {
1043 				memset(bh->b_data,0,inode->i_sb->s_blocksize);
1044 				set_buffer_uptodate(bh);
1045 			}
1046 			unlock_buffer(bh);
1047 			BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1048 			err = ext4_journal_dirty_metadata(handle, bh);
1049 			if (!fatal)
1050 				fatal = err;
1051 		} else {
1052 			BUFFER_TRACE(bh, "not a new buffer");
1053 		}
1054 		if (fatal) {
1055 			*errp = fatal;
1056 			brelse(bh);
1057 			bh = NULL;
1058 		}
1059 		return bh;
1060 	}
1061 err:
1062 	return NULL;
1063 }
1064 
1065 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1066 			       int block, int create, int *err)
1067 {
1068 	struct buffer_head * bh;
1069 
1070 	bh = ext4_getblk(handle, inode, block, create, err);
1071 	if (!bh)
1072 		return bh;
1073 	if (buffer_uptodate(bh))
1074 		return bh;
1075 	ll_rw_block(READ_META, 1, &bh);
1076 	wait_on_buffer(bh);
1077 	if (buffer_uptodate(bh))
1078 		return bh;
1079 	put_bh(bh);
1080 	*err = -EIO;
1081 	return NULL;
1082 }
1083 
1084 static int walk_page_buffers(	handle_t *handle,
1085 				struct buffer_head *head,
1086 				unsigned from,
1087 				unsigned to,
1088 				int *partial,
1089 				int (*fn)(	handle_t *handle,
1090 						struct buffer_head *bh))
1091 {
1092 	struct buffer_head *bh;
1093 	unsigned block_start, block_end;
1094 	unsigned blocksize = head->b_size;
1095 	int err, ret = 0;
1096 	struct buffer_head *next;
1097 
1098 	for (	bh = head, block_start = 0;
1099 		ret == 0 && (bh != head || !block_start);
1100 		block_start = block_end, bh = next)
1101 	{
1102 		next = bh->b_this_page;
1103 		block_end = block_start + blocksize;
1104 		if (block_end <= from || block_start >= to) {
1105 			if (partial && !buffer_uptodate(bh))
1106 				*partial = 1;
1107 			continue;
1108 		}
1109 		err = (*fn)(handle, bh);
1110 		if (!ret)
1111 			ret = err;
1112 	}
1113 	return ret;
1114 }
1115 
1116 /*
1117  * To preserve ordering, it is essential that the hole instantiation and
1118  * the data write be encapsulated in a single transaction.  We cannot
1119  * close off a transaction and start a new one between the ext4_get_block()
1120  * and the commit_write().  So doing the jbd2_journal_start at the start of
1121  * prepare_write() is the right place.
1122  *
1123  * Also, this function can nest inside ext4_writepage() ->
1124  * block_write_full_page(). In that case, we *know* that ext4_writepage()
1125  * has generated enough buffer credits to do the whole page.  So we won't
1126  * block on the journal in that case, which is good, because the caller may
1127  * be PF_MEMALLOC.
1128  *
1129  * By accident, ext4 can be reentered when a transaction is open via
1130  * quota file writes.  If we were to commit the transaction while thus
1131  * reentered, there can be a deadlock - we would be holding a quota
1132  * lock, and the commit would never complete if another thread had a
1133  * transaction open and was blocking on the quota lock - a ranking
1134  * violation.
1135  *
1136  * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1137  * will _not_ run commit under these circumstances because handle->h_ref
1138  * is elevated.  We'll still have enough credits for the tiny quotafile
1139  * write.
1140  */
1141 static int do_journal_get_write_access(handle_t *handle,
1142 					struct buffer_head *bh)
1143 {
1144 	if (!buffer_mapped(bh) || buffer_freed(bh))
1145 		return 0;
1146 	return ext4_journal_get_write_access(handle, bh);
1147 }
1148 
1149 static int ext4_prepare_write(struct file *file, struct page *page,
1150 			      unsigned from, unsigned to)
1151 {
1152 	struct inode *inode = page->mapping->host;
1153 	int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1154 	handle_t *handle;
1155 	int retries = 0;
1156 
1157 retry:
1158 	handle = ext4_journal_start(inode, needed_blocks);
1159 	if (IS_ERR(handle)) {
1160 		ret = PTR_ERR(handle);
1161 		goto out;
1162 	}
1163 	if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1164 		ret = nobh_prepare_write(page, from, to, ext4_get_block);
1165 	else
1166 		ret = block_prepare_write(page, from, to, ext4_get_block);
1167 	if (ret)
1168 		goto prepare_write_failed;
1169 
1170 	if (ext4_should_journal_data(inode)) {
1171 		ret = walk_page_buffers(handle, page_buffers(page),
1172 				from, to, NULL, do_journal_get_write_access);
1173 	}
1174 prepare_write_failed:
1175 	if (ret)
1176 		ext4_journal_stop(handle);
1177 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1178 		goto retry;
1179 out:
1180 	return ret;
1181 }
1182 
1183 int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1184 {
1185 	int err = jbd2_journal_dirty_data(handle, bh);
1186 	if (err)
1187 		ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1188 						bh, handle,err);
1189 	return err;
1190 }
1191 
1192 /* For commit_write() in data=journal mode */
1193 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1194 {
1195 	if (!buffer_mapped(bh) || buffer_freed(bh))
1196 		return 0;
1197 	set_buffer_uptodate(bh);
1198 	return ext4_journal_dirty_metadata(handle, bh);
1199 }
1200 
1201 /*
1202  * We need to pick up the new inode size which generic_commit_write gave us
1203  * `file' can be NULL - eg, when called from page_symlink().
1204  *
1205  * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1206  * buffers are managed internally.
1207  */
1208 static int ext4_ordered_commit_write(struct file *file, struct page *page,
1209 			     unsigned from, unsigned to)
1210 {
1211 	handle_t *handle = ext4_journal_current_handle();
1212 	struct inode *inode = page->mapping->host;
1213 	int ret = 0, ret2;
1214 
1215 	ret = walk_page_buffers(handle, page_buffers(page),
1216 		from, to, NULL, ext4_journal_dirty_data);
1217 
1218 	if (ret == 0) {
1219 		/*
1220 		 * generic_commit_write() will run mark_inode_dirty() if i_size
1221 		 * changes.  So let's piggyback the i_disksize mark_inode_dirty
1222 		 * into that.
1223 		 */
1224 		loff_t new_i_size;
1225 
1226 		new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1227 		if (new_i_size > EXT4_I(inode)->i_disksize)
1228 			EXT4_I(inode)->i_disksize = new_i_size;
1229 		ret = generic_commit_write(file, page, from, to);
1230 	}
1231 	ret2 = ext4_journal_stop(handle);
1232 	if (!ret)
1233 		ret = ret2;
1234 	return ret;
1235 }
1236 
1237 static int ext4_writeback_commit_write(struct file *file, struct page *page,
1238 			     unsigned from, unsigned to)
1239 {
1240 	handle_t *handle = ext4_journal_current_handle();
1241 	struct inode *inode = page->mapping->host;
1242 	int ret = 0, ret2;
1243 	loff_t new_i_size;
1244 
1245 	new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1246 	if (new_i_size > EXT4_I(inode)->i_disksize)
1247 		EXT4_I(inode)->i_disksize = new_i_size;
1248 
1249 	if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1250 		ret = nobh_commit_write(file, page, from, to);
1251 	else
1252 		ret = generic_commit_write(file, page, from, to);
1253 
1254 	ret2 = ext4_journal_stop(handle);
1255 	if (!ret)
1256 		ret = ret2;
1257 	return ret;
1258 }
1259 
1260 static int ext4_journalled_commit_write(struct file *file,
1261 			struct page *page, unsigned from, unsigned to)
1262 {
1263 	handle_t *handle = ext4_journal_current_handle();
1264 	struct inode *inode = page->mapping->host;
1265 	int ret = 0, ret2;
1266 	int partial = 0;
1267 	loff_t pos;
1268 
1269 	/*
1270 	 * Here we duplicate the generic_commit_write() functionality
1271 	 */
1272 	pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1273 
1274 	ret = walk_page_buffers(handle, page_buffers(page), from,
1275 				to, &partial, commit_write_fn);
1276 	if (!partial)
1277 		SetPageUptodate(page);
1278 	if (pos > inode->i_size)
1279 		i_size_write(inode, pos);
1280 	EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1281 	if (inode->i_size > EXT4_I(inode)->i_disksize) {
1282 		EXT4_I(inode)->i_disksize = inode->i_size;
1283 		ret2 = ext4_mark_inode_dirty(handle, inode);
1284 		if (!ret)
1285 			ret = ret2;
1286 	}
1287 	ret2 = ext4_journal_stop(handle);
1288 	if (!ret)
1289 		ret = ret2;
1290 	return ret;
1291 }
1292 
1293 /*
1294  * bmap() is special.  It gets used by applications such as lilo and by
1295  * the swapper to find the on-disk block of a specific piece of data.
1296  *
1297  * Naturally, this is dangerous if the block concerned is still in the
1298  * journal.  If somebody makes a swapfile on an ext4 data-journaling
1299  * filesystem and enables swap, then they may get a nasty shock when the
1300  * data getting swapped to that swapfile suddenly gets overwritten by
1301  * the original zero's written out previously to the journal and
1302  * awaiting writeback in the kernel's buffer cache.
1303  *
1304  * So, if we see any bmap calls here on a modified, data-journaled file,
1305  * take extra steps to flush any blocks which might be in the cache.
1306  */
1307 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
1308 {
1309 	struct inode *inode = mapping->host;
1310 	journal_t *journal;
1311 	int err;
1312 
1313 	if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
1314 		/*
1315 		 * This is a REALLY heavyweight approach, but the use of
1316 		 * bmap on dirty files is expected to be extremely rare:
1317 		 * only if we run lilo or swapon on a freshly made file
1318 		 * do we expect this to happen.
1319 		 *
1320 		 * (bmap requires CAP_SYS_RAWIO so this does not
1321 		 * represent an unprivileged user DOS attack --- we'd be
1322 		 * in trouble if mortal users could trigger this path at
1323 		 * will.)
1324 		 *
1325 		 * NB. EXT4_STATE_JDATA is not set on files other than
1326 		 * regular files.  If somebody wants to bmap a directory
1327 		 * or symlink and gets confused because the buffer
1328 		 * hasn't yet been flushed to disk, they deserve
1329 		 * everything they get.
1330 		 */
1331 
1332 		EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
1333 		journal = EXT4_JOURNAL(inode);
1334 		jbd2_journal_lock_updates(journal);
1335 		err = jbd2_journal_flush(journal);
1336 		jbd2_journal_unlock_updates(journal);
1337 
1338 		if (err)
1339 			return 0;
1340 	}
1341 
1342 	return generic_block_bmap(mapping,block,ext4_get_block);
1343 }
1344 
1345 static int bget_one(handle_t *handle, struct buffer_head *bh)
1346 {
1347 	get_bh(bh);
1348 	return 0;
1349 }
1350 
1351 static int bput_one(handle_t *handle, struct buffer_head *bh)
1352 {
1353 	put_bh(bh);
1354 	return 0;
1355 }
1356 
1357 static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1358 {
1359 	if (buffer_mapped(bh))
1360 		return ext4_journal_dirty_data(handle, bh);
1361 	return 0;
1362 }
1363 
1364 /*
1365  * Note that we always start a transaction even if we're not journalling
1366  * data.  This is to preserve ordering: any hole instantiation within
1367  * __block_write_full_page -> ext4_get_block() should be journalled
1368  * along with the data so we don't crash and then get metadata which
1369  * refers to old data.
1370  *
1371  * In all journalling modes block_write_full_page() will start the I/O.
1372  *
1373  * Problem:
1374  *
1375  *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1376  *		ext4_writepage()
1377  *
1378  * Similar for:
1379  *
1380  *	ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1381  *
1382  * Same applies to ext4_get_block().  We will deadlock on various things like
1383  * lock_journal and i_truncate_mutex.
1384  *
1385  * Setting PF_MEMALLOC here doesn't work - too many internal memory
1386  * allocations fail.
1387  *
1388  * 16May01: If we're reentered then journal_current_handle() will be
1389  *	    non-zero. We simply *return*.
1390  *
1391  * 1 July 2001: @@@ FIXME:
1392  *   In journalled data mode, a data buffer may be metadata against the
1393  *   current transaction.  But the same file is part of a shared mapping
1394  *   and someone does a writepage() on it.
1395  *
1396  *   We will move the buffer onto the async_data list, but *after* it has
1397  *   been dirtied. So there's a small window where we have dirty data on
1398  *   BJ_Metadata.
1399  *
1400  *   Note that this only applies to the last partial page in the file.  The
1401  *   bit which block_write_full_page() uses prepare/commit for.  (That's
1402  *   broken code anyway: it's wrong for msync()).
1403  *
1404  *   It's a rare case: affects the final partial page, for journalled data
1405  *   where the file is subject to bith write() and writepage() in the same
1406  *   transction.  To fix it we'll need a custom block_write_full_page().
1407  *   We'll probably need that anyway for journalling writepage() output.
1408  *
1409  * We don't honour synchronous mounts for writepage().  That would be
1410  * disastrous.  Any write() or metadata operation will sync the fs for
1411  * us.
1412  *
1413  * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1414  * we don't need to open a transaction here.
1415  */
1416 static int ext4_ordered_writepage(struct page *page,
1417 				struct writeback_control *wbc)
1418 {
1419 	struct inode *inode = page->mapping->host;
1420 	struct buffer_head *page_bufs;
1421 	handle_t *handle = NULL;
1422 	int ret = 0;
1423 	int err;
1424 
1425 	J_ASSERT(PageLocked(page));
1426 
1427 	/*
1428 	 * We give up here if we're reentered, because it might be for a
1429 	 * different filesystem.
1430 	 */
1431 	if (ext4_journal_current_handle())
1432 		goto out_fail;
1433 
1434 	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1435 
1436 	if (IS_ERR(handle)) {
1437 		ret = PTR_ERR(handle);
1438 		goto out_fail;
1439 	}
1440 
1441 	if (!page_has_buffers(page)) {
1442 		create_empty_buffers(page, inode->i_sb->s_blocksize,
1443 				(1 << BH_Dirty)|(1 << BH_Uptodate));
1444 	}
1445 	page_bufs = page_buffers(page);
1446 	walk_page_buffers(handle, page_bufs, 0,
1447 			PAGE_CACHE_SIZE, NULL, bget_one);
1448 
1449 	ret = block_write_full_page(page, ext4_get_block, wbc);
1450 
1451 	/*
1452 	 * The page can become unlocked at any point now, and
1453 	 * truncate can then come in and change things.  So we
1454 	 * can't touch *page from now on.  But *page_bufs is
1455 	 * safe due to elevated refcount.
1456 	 */
1457 
1458 	/*
1459 	 * And attach them to the current transaction.  But only if
1460 	 * block_write_full_page() succeeded.  Otherwise they are unmapped,
1461 	 * and generally junk.
1462 	 */
1463 	if (ret == 0) {
1464 		err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1465 					NULL, jbd2_journal_dirty_data_fn);
1466 		if (!ret)
1467 			ret = err;
1468 	}
1469 	walk_page_buffers(handle, page_bufs, 0,
1470 			PAGE_CACHE_SIZE, NULL, bput_one);
1471 	err = ext4_journal_stop(handle);
1472 	if (!ret)
1473 		ret = err;
1474 	return ret;
1475 
1476 out_fail:
1477 	redirty_page_for_writepage(wbc, page);
1478 	unlock_page(page);
1479 	return ret;
1480 }
1481 
1482 static int ext4_writeback_writepage(struct page *page,
1483 				struct writeback_control *wbc)
1484 {
1485 	struct inode *inode = page->mapping->host;
1486 	handle_t *handle = NULL;
1487 	int ret = 0;
1488 	int err;
1489 
1490 	if (ext4_journal_current_handle())
1491 		goto out_fail;
1492 
1493 	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1494 	if (IS_ERR(handle)) {
1495 		ret = PTR_ERR(handle);
1496 		goto out_fail;
1497 	}
1498 
1499 	if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1500 		ret = nobh_writepage(page, ext4_get_block, wbc);
1501 	else
1502 		ret = block_write_full_page(page, ext4_get_block, wbc);
1503 
1504 	err = ext4_journal_stop(handle);
1505 	if (!ret)
1506 		ret = err;
1507 	return ret;
1508 
1509 out_fail:
1510 	redirty_page_for_writepage(wbc, page);
1511 	unlock_page(page);
1512 	return ret;
1513 }
1514 
1515 static int ext4_journalled_writepage(struct page *page,
1516 				struct writeback_control *wbc)
1517 {
1518 	struct inode *inode = page->mapping->host;
1519 	handle_t *handle = NULL;
1520 	int ret = 0;
1521 	int err;
1522 
1523 	if (ext4_journal_current_handle())
1524 		goto no_write;
1525 
1526 	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1527 	if (IS_ERR(handle)) {
1528 		ret = PTR_ERR(handle);
1529 		goto no_write;
1530 	}
1531 
1532 	if (!page_has_buffers(page) || PageChecked(page)) {
1533 		/*
1534 		 * It's mmapped pagecache.  Add buffers and journal it.  There
1535 		 * doesn't seem much point in redirtying the page here.
1536 		 */
1537 		ClearPageChecked(page);
1538 		ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1539 					ext4_get_block);
1540 		if (ret != 0) {
1541 			ext4_journal_stop(handle);
1542 			goto out_unlock;
1543 		}
1544 		ret = walk_page_buffers(handle, page_buffers(page), 0,
1545 			PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1546 
1547 		err = walk_page_buffers(handle, page_buffers(page), 0,
1548 				PAGE_CACHE_SIZE, NULL, commit_write_fn);
1549 		if (ret == 0)
1550 			ret = err;
1551 		EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1552 		unlock_page(page);
1553 	} else {
1554 		/*
1555 		 * It may be a page full of checkpoint-mode buffers.  We don't
1556 		 * really know unless we go poke around in the buffer_heads.
1557 		 * But block_write_full_page will do the right thing.
1558 		 */
1559 		ret = block_write_full_page(page, ext4_get_block, wbc);
1560 	}
1561 	err = ext4_journal_stop(handle);
1562 	if (!ret)
1563 		ret = err;
1564 out:
1565 	return ret;
1566 
1567 no_write:
1568 	redirty_page_for_writepage(wbc, page);
1569 out_unlock:
1570 	unlock_page(page);
1571 	goto out;
1572 }
1573 
1574 static int ext4_readpage(struct file *file, struct page *page)
1575 {
1576 	return mpage_readpage(page, ext4_get_block);
1577 }
1578 
1579 static int
1580 ext4_readpages(struct file *file, struct address_space *mapping,
1581 		struct list_head *pages, unsigned nr_pages)
1582 {
1583 	return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
1584 }
1585 
1586 static void ext4_invalidatepage(struct page *page, unsigned long offset)
1587 {
1588 	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1589 
1590 	/*
1591 	 * If it's a full truncate we just forget about the pending dirtying
1592 	 */
1593 	if (offset == 0)
1594 		ClearPageChecked(page);
1595 
1596 	jbd2_journal_invalidatepage(journal, page, offset);
1597 }
1598 
1599 static int ext4_releasepage(struct page *page, gfp_t wait)
1600 {
1601 	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1602 
1603 	WARN_ON(PageChecked(page));
1604 	if (!page_has_buffers(page))
1605 		return 0;
1606 	return jbd2_journal_try_to_free_buffers(journal, page, wait);
1607 }
1608 
1609 /*
1610  * If the O_DIRECT write will extend the file then add this inode to the
1611  * orphan list.  So recovery will truncate it back to the original size
1612  * if the machine crashes during the write.
1613  *
1614  * If the O_DIRECT write is intantiating holes inside i_size and the machine
1615  * crashes then stale disk data _may_ be exposed inside the file.
1616  */
1617 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
1618 			const struct iovec *iov, loff_t offset,
1619 			unsigned long nr_segs)
1620 {
1621 	struct file *file = iocb->ki_filp;
1622 	struct inode *inode = file->f_mapping->host;
1623 	struct ext4_inode_info *ei = EXT4_I(inode);
1624 	handle_t *handle = NULL;
1625 	ssize_t ret;
1626 	int orphan = 0;
1627 	size_t count = iov_length(iov, nr_segs);
1628 
1629 	if (rw == WRITE) {
1630 		loff_t final_size = offset + count;
1631 
1632 		handle = ext4_journal_start(inode, DIO_CREDITS);
1633 		if (IS_ERR(handle)) {
1634 			ret = PTR_ERR(handle);
1635 			goto out;
1636 		}
1637 		if (final_size > inode->i_size) {
1638 			ret = ext4_orphan_add(handle, inode);
1639 			if (ret)
1640 				goto out_stop;
1641 			orphan = 1;
1642 			ei->i_disksize = inode->i_size;
1643 		}
1644 	}
1645 
1646 	ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1647 				 offset, nr_segs,
1648 				 ext4_get_block, NULL);
1649 
1650 	/*
1651 	 * Reacquire the handle: ext4_get_block() can restart the transaction
1652 	 */
1653 	handle = ext4_journal_current_handle();
1654 
1655 out_stop:
1656 	if (handle) {
1657 		int err;
1658 
1659 		if (orphan && inode->i_nlink)
1660 			ext4_orphan_del(handle, inode);
1661 		if (orphan && ret > 0) {
1662 			loff_t end = offset + ret;
1663 			if (end > inode->i_size) {
1664 				ei->i_disksize = end;
1665 				i_size_write(inode, end);
1666 				/*
1667 				 * We're going to return a positive `ret'
1668 				 * here due to non-zero-length I/O, so there's
1669 				 * no way of reporting error returns from
1670 				 * ext4_mark_inode_dirty() to userspace.  So
1671 				 * ignore it.
1672 				 */
1673 				ext4_mark_inode_dirty(handle, inode);
1674 			}
1675 		}
1676 		err = ext4_journal_stop(handle);
1677 		if (ret == 0)
1678 			ret = err;
1679 	}
1680 out:
1681 	return ret;
1682 }
1683 
1684 /*
1685  * Pages can be marked dirty completely asynchronously from ext4's journalling
1686  * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
1687  * much here because ->set_page_dirty is called under VFS locks.  The page is
1688  * not necessarily locked.
1689  *
1690  * We cannot just dirty the page and leave attached buffers clean, because the
1691  * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
1692  * or jbddirty because all the journalling code will explode.
1693  *
1694  * So what we do is to mark the page "pending dirty" and next time writepage
1695  * is called, propagate that into the buffers appropriately.
1696  */
1697 static int ext4_journalled_set_page_dirty(struct page *page)
1698 {
1699 	SetPageChecked(page);
1700 	return __set_page_dirty_nobuffers(page);
1701 }
1702 
1703 static const struct address_space_operations ext4_ordered_aops = {
1704 	.readpage	= ext4_readpage,
1705 	.readpages	= ext4_readpages,
1706 	.writepage	= ext4_ordered_writepage,
1707 	.sync_page	= block_sync_page,
1708 	.prepare_write	= ext4_prepare_write,
1709 	.commit_write	= ext4_ordered_commit_write,
1710 	.bmap		= ext4_bmap,
1711 	.invalidatepage	= ext4_invalidatepage,
1712 	.releasepage	= ext4_releasepage,
1713 	.direct_IO	= ext4_direct_IO,
1714 	.migratepage	= buffer_migrate_page,
1715 };
1716 
1717 static const struct address_space_operations ext4_writeback_aops = {
1718 	.readpage	= ext4_readpage,
1719 	.readpages	= ext4_readpages,
1720 	.writepage	= ext4_writeback_writepage,
1721 	.sync_page	= block_sync_page,
1722 	.prepare_write	= ext4_prepare_write,
1723 	.commit_write	= ext4_writeback_commit_write,
1724 	.bmap		= ext4_bmap,
1725 	.invalidatepage	= ext4_invalidatepage,
1726 	.releasepage	= ext4_releasepage,
1727 	.direct_IO	= ext4_direct_IO,
1728 	.migratepage	= buffer_migrate_page,
1729 };
1730 
1731 static const struct address_space_operations ext4_journalled_aops = {
1732 	.readpage	= ext4_readpage,
1733 	.readpages	= ext4_readpages,
1734 	.writepage	= ext4_journalled_writepage,
1735 	.sync_page	= block_sync_page,
1736 	.prepare_write	= ext4_prepare_write,
1737 	.commit_write	= ext4_journalled_commit_write,
1738 	.set_page_dirty	= ext4_journalled_set_page_dirty,
1739 	.bmap		= ext4_bmap,
1740 	.invalidatepage	= ext4_invalidatepage,
1741 	.releasepage	= ext4_releasepage,
1742 };
1743 
1744 void ext4_set_aops(struct inode *inode)
1745 {
1746 	if (ext4_should_order_data(inode))
1747 		inode->i_mapping->a_ops = &ext4_ordered_aops;
1748 	else if (ext4_should_writeback_data(inode))
1749 		inode->i_mapping->a_ops = &ext4_writeback_aops;
1750 	else
1751 		inode->i_mapping->a_ops = &ext4_journalled_aops;
1752 }
1753 
1754 /*
1755  * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
1756  * up to the end of the block which corresponds to `from'.
1757  * This required during truncate. We need to physically zero the tail end
1758  * of that block so it doesn't yield old data if the file is later grown.
1759  */
1760 int ext4_block_truncate_page(handle_t *handle, struct page *page,
1761 		struct address_space *mapping, loff_t from)
1762 {
1763 	ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
1764 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
1765 	unsigned blocksize, iblock, length, pos;
1766 	struct inode *inode = mapping->host;
1767 	struct buffer_head *bh;
1768 	int err = 0;
1769 	void *kaddr;
1770 
1771 	blocksize = inode->i_sb->s_blocksize;
1772 	length = blocksize - (offset & (blocksize - 1));
1773 	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1774 
1775 	/*
1776 	 * For "nobh" option,  we can only work if we don't need to
1777 	 * read-in the page - otherwise we create buffers to do the IO.
1778 	 */
1779 	if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1780 	     ext4_should_writeback_data(inode) && PageUptodate(page)) {
1781 		kaddr = kmap_atomic(page, KM_USER0);
1782 		memset(kaddr + offset, 0, length);
1783 		flush_dcache_page(page);
1784 		kunmap_atomic(kaddr, KM_USER0);
1785 		set_page_dirty(page);
1786 		goto unlock;
1787 	}
1788 
1789 	if (!page_has_buffers(page))
1790 		create_empty_buffers(page, blocksize, 0);
1791 
1792 	/* Find the buffer that contains "offset" */
1793 	bh = page_buffers(page);
1794 	pos = blocksize;
1795 	while (offset >= pos) {
1796 		bh = bh->b_this_page;
1797 		iblock++;
1798 		pos += blocksize;
1799 	}
1800 
1801 	err = 0;
1802 	if (buffer_freed(bh)) {
1803 		BUFFER_TRACE(bh, "freed: skip");
1804 		goto unlock;
1805 	}
1806 
1807 	if (!buffer_mapped(bh)) {
1808 		BUFFER_TRACE(bh, "unmapped");
1809 		ext4_get_block(inode, iblock, bh, 0);
1810 		/* unmapped? It's a hole - nothing to do */
1811 		if (!buffer_mapped(bh)) {
1812 			BUFFER_TRACE(bh, "still unmapped");
1813 			goto unlock;
1814 		}
1815 	}
1816 
1817 	/* Ok, it's mapped. Make sure it's up-to-date */
1818 	if (PageUptodate(page))
1819 		set_buffer_uptodate(bh);
1820 
1821 	if (!buffer_uptodate(bh)) {
1822 		err = -EIO;
1823 		ll_rw_block(READ, 1, &bh);
1824 		wait_on_buffer(bh);
1825 		/* Uhhuh. Read error. Complain and punt. */
1826 		if (!buffer_uptodate(bh))
1827 			goto unlock;
1828 	}
1829 
1830 	if (ext4_should_journal_data(inode)) {
1831 		BUFFER_TRACE(bh, "get write access");
1832 		err = ext4_journal_get_write_access(handle, bh);
1833 		if (err)
1834 			goto unlock;
1835 	}
1836 
1837 	kaddr = kmap_atomic(page, KM_USER0);
1838 	memset(kaddr + offset, 0, length);
1839 	flush_dcache_page(page);
1840 	kunmap_atomic(kaddr, KM_USER0);
1841 
1842 	BUFFER_TRACE(bh, "zeroed end of block");
1843 
1844 	err = 0;
1845 	if (ext4_should_journal_data(inode)) {
1846 		err = ext4_journal_dirty_metadata(handle, bh);
1847 	} else {
1848 		if (ext4_should_order_data(inode))
1849 			err = ext4_journal_dirty_data(handle, bh);
1850 		mark_buffer_dirty(bh);
1851 	}
1852 
1853 unlock:
1854 	unlock_page(page);
1855 	page_cache_release(page);
1856 	return err;
1857 }
1858 
1859 /*
1860  * Probably it should be a library function... search for first non-zero word
1861  * or memcmp with zero_page, whatever is better for particular architecture.
1862  * Linus?
1863  */
1864 static inline int all_zeroes(__le32 *p, __le32 *q)
1865 {
1866 	while (p < q)
1867 		if (*p++)
1868 			return 0;
1869 	return 1;
1870 }
1871 
1872 /**
1873  *	ext4_find_shared - find the indirect blocks for partial truncation.
1874  *	@inode:	  inode in question
1875  *	@depth:	  depth of the affected branch
1876  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
1877  *	@chain:	  place to store the pointers to partial indirect blocks
1878  *	@top:	  place to the (detached) top of branch
1879  *
1880  *	This is a helper function used by ext4_truncate().
1881  *
1882  *	When we do truncate() we may have to clean the ends of several
1883  *	indirect blocks but leave the blocks themselves alive. Block is
1884  *	partially truncated if some data below the new i_size is refered
1885  *	from it (and it is on the path to the first completely truncated
1886  *	data block, indeed).  We have to free the top of that path along
1887  *	with everything to the right of the path. Since no allocation
1888  *	past the truncation point is possible until ext4_truncate()
1889  *	finishes, we may safely do the latter, but top of branch may
1890  *	require special attention - pageout below the truncation point
1891  *	might try to populate it.
1892  *
1893  *	We atomically detach the top of branch from the tree, store the
1894  *	block number of its root in *@top, pointers to buffer_heads of
1895  *	partially truncated blocks - in @chain[].bh and pointers to
1896  *	their last elements that should not be removed - in
1897  *	@chain[].p. Return value is the pointer to last filled element
1898  *	of @chain.
1899  *
1900  *	The work left to caller to do the actual freeing of subtrees:
1901  *		a) free the subtree starting from *@top
1902  *		b) free the subtrees whose roots are stored in
1903  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1904  *		c) free the subtrees growing from the inode past the @chain[0].
1905  *			(no partially truncated stuff there).  */
1906 
1907 static Indirect *ext4_find_shared(struct inode *inode, int depth,
1908 			int offsets[4], Indirect chain[4], __le32 *top)
1909 {
1910 	Indirect *partial, *p;
1911 	int k, err;
1912 
1913 	*top = 0;
1914 	/* Make k index the deepest non-null offest + 1 */
1915 	for (k = depth; k > 1 && !offsets[k-1]; k--)
1916 		;
1917 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
1918 	/* Writer: pointers */
1919 	if (!partial)
1920 		partial = chain + k-1;
1921 	/*
1922 	 * If the branch acquired continuation since we've looked at it -
1923 	 * fine, it should all survive and (new) top doesn't belong to us.
1924 	 */
1925 	if (!partial->key && *partial->p)
1926 		/* Writer: end */
1927 		goto no_top;
1928 	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1929 		;
1930 	/*
1931 	 * OK, we've found the last block that must survive. The rest of our
1932 	 * branch should be detached before unlocking. However, if that rest
1933 	 * of branch is all ours and does not grow immediately from the inode
1934 	 * it's easier to cheat and just decrement partial->p.
1935 	 */
1936 	if (p == chain + k - 1 && p > chain) {
1937 		p->p--;
1938 	} else {
1939 		*top = *p->p;
1940 		/* Nope, don't do this in ext4.  Must leave the tree intact */
1941 #if 0
1942 		*p->p = 0;
1943 #endif
1944 	}
1945 	/* Writer: end */
1946 
1947 	while(partial > p) {
1948 		brelse(partial->bh);
1949 		partial--;
1950 	}
1951 no_top:
1952 	return partial;
1953 }
1954 
1955 /*
1956  * Zero a number of block pointers in either an inode or an indirect block.
1957  * If we restart the transaction we must again get write access to the
1958  * indirect block for further modification.
1959  *
1960  * We release `count' blocks on disk, but (last - first) may be greater
1961  * than `count' because there can be holes in there.
1962  */
1963 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
1964 		struct buffer_head *bh, ext4_fsblk_t block_to_free,
1965 		unsigned long count, __le32 *first, __le32 *last)
1966 {
1967 	__le32 *p;
1968 	if (try_to_extend_transaction(handle, inode)) {
1969 		if (bh) {
1970 			BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1971 			ext4_journal_dirty_metadata(handle, bh);
1972 		}
1973 		ext4_mark_inode_dirty(handle, inode);
1974 		ext4_journal_test_restart(handle, inode);
1975 		if (bh) {
1976 			BUFFER_TRACE(bh, "retaking write access");
1977 			ext4_journal_get_write_access(handle, bh);
1978 		}
1979 	}
1980 
1981 	/*
1982 	 * Any buffers which are on the journal will be in memory. We find
1983 	 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
1984 	 * on them.  We've already detached each block from the file, so
1985 	 * bforget() in jbd2_journal_forget() should be safe.
1986 	 *
1987 	 * AKPM: turn on bforget in jbd2_journal_forget()!!!
1988 	 */
1989 	for (p = first; p < last; p++) {
1990 		u32 nr = le32_to_cpu(*p);
1991 		if (nr) {
1992 			struct buffer_head *bh;
1993 
1994 			*p = 0;
1995 			bh = sb_find_get_block(inode->i_sb, nr);
1996 			ext4_forget(handle, 0, inode, bh, nr);
1997 		}
1998 	}
1999 
2000 	ext4_free_blocks(handle, inode, block_to_free, count);
2001 }
2002 
2003 /**
2004  * ext4_free_data - free a list of data blocks
2005  * @handle:	handle for this transaction
2006  * @inode:	inode we are dealing with
2007  * @this_bh:	indirect buffer_head which contains *@first and *@last
2008  * @first:	array of block numbers
2009  * @last:	points immediately past the end of array
2010  *
2011  * We are freeing all blocks refered from that array (numbers are stored as
2012  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2013  *
2014  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
2015  * blocks are contiguous then releasing them at one time will only affect one
2016  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2017  * actually use a lot of journal space.
2018  *
2019  * @this_bh will be %NULL if @first and @last point into the inode's direct
2020  * block pointers.
2021  */
2022 static void ext4_free_data(handle_t *handle, struct inode *inode,
2023 			   struct buffer_head *this_bh,
2024 			   __le32 *first, __le32 *last)
2025 {
2026 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
2027 	unsigned long count = 0;	    /* Number of blocks in the run */
2028 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
2029 					       corresponding to
2030 					       block_to_free */
2031 	ext4_fsblk_t nr;		    /* Current block # */
2032 	__le32 *p;			    /* Pointer into inode/ind
2033 					       for current block */
2034 	int err;
2035 
2036 	if (this_bh) {				/* For indirect block */
2037 		BUFFER_TRACE(this_bh, "get_write_access");
2038 		err = ext4_journal_get_write_access(handle, this_bh);
2039 		/* Important: if we can't update the indirect pointers
2040 		 * to the blocks, we can't free them. */
2041 		if (err)
2042 			return;
2043 	}
2044 
2045 	for (p = first; p < last; p++) {
2046 		nr = le32_to_cpu(*p);
2047 		if (nr) {
2048 			/* accumulate blocks to free if they're contiguous */
2049 			if (count == 0) {
2050 				block_to_free = nr;
2051 				block_to_free_p = p;
2052 				count = 1;
2053 			} else if (nr == block_to_free + count) {
2054 				count++;
2055 			} else {
2056 				ext4_clear_blocks(handle, inode, this_bh,
2057 						  block_to_free,
2058 						  count, block_to_free_p, p);
2059 				block_to_free = nr;
2060 				block_to_free_p = p;
2061 				count = 1;
2062 			}
2063 		}
2064 	}
2065 
2066 	if (count > 0)
2067 		ext4_clear_blocks(handle, inode, this_bh, block_to_free,
2068 				  count, block_to_free_p, p);
2069 
2070 	if (this_bh) {
2071 		BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2072 		ext4_journal_dirty_metadata(handle, this_bh);
2073 	}
2074 }
2075 
2076 /**
2077  *	ext4_free_branches - free an array of branches
2078  *	@handle: JBD handle for this transaction
2079  *	@inode:	inode we are dealing with
2080  *	@parent_bh: the buffer_head which contains *@first and *@last
2081  *	@first:	array of block numbers
2082  *	@last:	pointer immediately past the end of array
2083  *	@depth:	depth of the branches to free
2084  *
2085  *	We are freeing all blocks refered from these branches (numbers are
2086  *	stored as little-endian 32-bit) and updating @inode->i_blocks
2087  *	appropriately.
2088  */
2089 static void ext4_free_branches(handle_t *handle, struct inode *inode,
2090 			       struct buffer_head *parent_bh,
2091 			       __le32 *first, __le32 *last, int depth)
2092 {
2093 	ext4_fsblk_t nr;
2094 	__le32 *p;
2095 
2096 	if (is_handle_aborted(handle))
2097 		return;
2098 
2099 	if (depth--) {
2100 		struct buffer_head *bh;
2101 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2102 		p = last;
2103 		while (--p >= first) {
2104 			nr = le32_to_cpu(*p);
2105 			if (!nr)
2106 				continue;		/* A hole */
2107 
2108 			/* Go read the buffer for the next level down */
2109 			bh = sb_bread(inode->i_sb, nr);
2110 
2111 			/*
2112 			 * A read failure? Report error and clear slot
2113 			 * (should be rare).
2114 			 */
2115 			if (!bh) {
2116 				ext4_error(inode->i_sb, "ext4_free_branches",
2117 					   "Read failure, inode=%lu, block=%llu",
2118 					   inode->i_ino, nr);
2119 				continue;
2120 			}
2121 
2122 			/* This zaps the entire block.  Bottom up. */
2123 			BUFFER_TRACE(bh, "free child branches");
2124 			ext4_free_branches(handle, inode, bh,
2125 					   (__le32*)bh->b_data,
2126 					   (__le32*)bh->b_data + addr_per_block,
2127 					   depth);
2128 
2129 			/*
2130 			 * We've probably journalled the indirect block several
2131 			 * times during the truncate.  But it's no longer
2132 			 * needed and we now drop it from the transaction via
2133 			 * jbd2_journal_revoke().
2134 			 *
2135 			 * That's easy if it's exclusively part of this
2136 			 * transaction.  But if it's part of the committing
2137 			 * transaction then jbd2_journal_forget() will simply
2138 			 * brelse() it.  That means that if the underlying
2139 			 * block is reallocated in ext4_get_block(),
2140 			 * unmap_underlying_metadata() will find this block
2141 			 * and will try to get rid of it.  damn, damn.
2142 			 *
2143 			 * If this block has already been committed to the
2144 			 * journal, a revoke record will be written.  And
2145 			 * revoke records must be emitted *before* clearing
2146 			 * this block's bit in the bitmaps.
2147 			 */
2148 			ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
2149 
2150 			/*
2151 			 * Everything below this this pointer has been
2152 			 * released.  Now let this top-of-subtree go.
2153 			 *
2154 			 * We want the freeing of this indirect block to be
2155 			 * atomic in the journal with the updating of the
2156 			 * bitmap block which owns it.  So make some room in
2157 			 * the journal.
2158 			 *
2159 			 * We zero the parent pointer *after* freeing its
2160 			 * pointee in the bitmaps, so if extend_transaction()
2161 			 * for some reason fails to put the bitmap changes and
2162 			 * the release into the same transaction, recovery
2163 			 * will merely complain about releasing a free block,
2164 			 * rather than leaking blocks.
2165 			 */
2166 			if (is_handle_aborted(handle))
2167 				return;
2168 			if (try_to_extend_transaction(handle, inode)) {
2169 				ext4_mark_inode_dirty(handle, inode);
2170 				ext4_journal_test_restart(handle, inode);
2171 			}
2172 
2173 			ext4_free_blocks(handle, inode, nr, 1);
2174 
2175 			if (parent_bh) {
2176 				/*
2177 				 * The block which we have just freed is
2178 				 * pointed to by an indirect block: journal it
2179 				 */
2180 				BUFFER_TRACE(parent_bh, "get_write_access");
2181 				if (!ext4_journal_get_write_access(handle,
2182 								   parent_bh)){
2183 					*p = 0;
2184 					BUFFER_TRACE(parent_bh,
2185 					"call ext4_journal_dirty_metadata");
2186 					ext4_journal_dirty_metadata(handle,
2187 								    parent_bh);
2188 				}
2189 			}
2190 		}
2191 	} else {
2192 		/* We have reached the bottom of the tree. */
2193 		BUFFER_TRACE(parent_bh, "free data blocks");
2194 		ext4_free_data(handle, inode, parent_bh, first, last);
2195 	}
2196 }
2197 
2198 /*
2199  * ext4_truncate()
2200  *
2201  * We block out ext4_get_block() block instantiations across the entire
2202  * transaction, and VFS/VM ensures that ext4_truncate() cannot run
2203  * simultaneously on behalf of the same inode.
2204  *
2205  * As we work through the truncate and commmit bits of it to the journal there
2206  * is one core, guiding principle: the file's tree must always be consistent on
2207  * disk.  We must be able to restart the truncate after a crash.
2208  *
2209  * The file's tree may be transiently inconsistent in memory (although it
2210  * probably isn't), but whenever we close off and commit a journal transaction,
2211  * the contents of (the filesystem + the journal) must be consistent and
2212  * restartable.  It's pretty simple, really: bottom up, right to left (although
2213  * left-to-right works OK too).
2214  *
2215  * Note that at recovery time, journal replay occurs *before* the restart of
2216  * truncate against the orphan inode list.
2217  *
2218  * The committed inode has the new, desired i_size (which is the same as
2219  * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
2220  * that this inode's truncate did not complete and it will again call
2221  * ext4_truncate() to have another go.  So there will be instantiated blocks
2222  * to the right of the truncation point in a crashed ext4 filesystem.  But
2223  * that's fine - as long as they are linked from the inode, the post-crash
2224  * ext4_truncate() run will find them and release them.
2225  */
2226 void ext4_truncate(struct inode *inode)
2227 {
2228 	handle_t *handle;
2229 	struct ext4_inode_info *ei = EXT4_I(inode);
2230 	__le32 *i_data = ei->i_data;
2231 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2232 	struct address_space *mapping = inode->i_mapping;
2233 	int offsets[4];
2234 	Indirect chain[4];
2235 	Indirect *partial;
2236 	__le32 nr = 0;
2237 	int n;
2238 	long last_block;
2239 	unsigned blocksize = inode->i_sb->s_blocksize;
2240 	struct page *page;
2241 
2242 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2243 	    S_ISLNK(inode->i_mode)))
2244 		return;
2245 	if (ext4_inode_is_fast_symlink(inode))
2246 		return;
2247 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2248 		return;
2249 
2250 	/*
2251 	 * We have to lock the EOF page here, because lock_page() nests
2252 	 * outside jbd2_journal_start().
2253 	 */
2254 	if ((inode->i_size & (blocksize - 1)) == 0) {
2255 		/* Block boundary? Nothing to do */
2256 		page = NULL;
2257 	} else {
2258 		page = grab_cache_page(mapping,
2259 				inode->i_size >> PAGE_CACHE_SHIFT);
2260 		if (!page)
2261 			return;
2262 	}
2263 
2264 	if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
2265 		return ext4_ext_truncate(inode, page);
2266 
2267 	handle = start_transaction(inode);
2268 	if (IS_ERR(handle)) {
2269 		if (page) {
2270 			clear_highpage(page);
2271 			flush_dcache_page(page);
2272 			unlock_page(page);
2273 			page_cache_release(page);
2274 		}
2275 		return;		/* AKPM: return what? */
2276 	}
2277 
2278 	last_block = (inode->i_size + blocksize-1)
2279 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
2280 
2281 	if (page)
2282 		ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2283 
2284 	n = ext4_block_to_path(inode, last_block, offsets, NULL);
2285 	if (n == 0)
2286 		goto out_stop;	/* error */
2287 
2288 	/*
2289 	 * OK.  This truncate is going to happen.  We add the inode to the
2290 	 * orphan list, so that if this truncate spans multiple transactions,
2291 	 * and we crash, we will resume the truncate when the filesystem
2292 	 * recovers.  It also marks the inode dirty, to catch the new size.
2293 	 *
2294 	 * Implication: the file must always be in a sane, consistent
2295 	 * truncatable state while each transaction commits.
2296 	 */
2297 	if (ext4_orphan_add(handle, inode))
2298 		goto out_stop;
2299 
2300 	/*
2301 	 * The orphan list entry will now protect us from any crash which
2302 	 * occurs before the truncate completes, so it is now safe to propagate
2303 	 * the new, shorter inode size (held for now in i_size) into the
2304 	 * on-disk inode. We do this via i_disksize, which is the value which
2305 	 * ext4 *really* writes onto the disk inode.
2306 	 */
2307 	ei->i_disksize = inode->i_size;
2308 
2309 	/*
2310 	 * From here we block out all ext4_get_block() callers who want to
2311 	 * modify the block allocation tree.
2312 	 */
2313 	mutex_lock(&ei->truncate_mutex);
2314 
2315 	if (n == 1) {		/* direct blocks */
2316 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2317 			       i_data + EXT4_NDIR_BLOCKS);
2318 		goto do_indirects;
2319 	}
2320 
2321 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
2322 	/* Kill the top of shared branch (not detached) */
2323 	if (nr) {
2324 		if (partial == chain) {
2325 			/* Shared branch grows from the inode */
2326 			ext4_free_branches(handle, inode, NULL,
2327 					   &nr, &nr+1, (chain+n-1) - partial);
2328 			*partial->p = 0;
2329 			/*
2330 			 * We mark the inode dirty prior to restart,
2331 			 * and prior to stop.  No need for it here.
2332 			 */
2333 		} else {
2334 			/* Shared branch grows from an indirect block */
2335 			BUFFER_TRACE(partial->bh, "get_write_access");
2336 			ext4_free_branches(handle, inode, partial->bh,
2337 					partial->p,
2338 					partial->p+1, (chain+n-1) - partial);
2339 		}
2340 	}
2341 	/* Clear the ends of indirect blocks on the shared branch */
2342 	while (partial > chain) {
2343 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
2344 				   (__le32*)partial->bh->b_data+addr_per_block,
2345 				   (chain+n-1) - partial);
2346 		BUFFER_TRACE(partial->bh, "call brelse");
2347 		brelse (partial->bh);
2348 		partial--;
2349 	}
2350 do_indirects:
2351 	/* Kill the remaining (whole) subtrees */
2352 	switch (offsets[0]) {
2353 	default:
2354 		nr = i_data[EXT4_IND_BLOCK];
2355 		if (nr) {
2356 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2357 			i_data[EXT4_IND_BLOCK] = 0;
2358 		}
2359 	case EXT4_IND_BLOCK:
2360 		nr = i_data[EXT4_DIND_BLOCK];
2361 		if (nr) {
2362 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2363 			i_data[EXT4_DIND_BLOCK] = 0;
2364 		}
2365 	case EXT4_DIND_BLOCK:
2366 		nr = i_data[EXT4_TIND_BLOCK];
2367 		if (nr) {
2368 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2369 			i_data[EXT4_TIND_BLOCK] = 0;
2370 		}
2371 	case EXT4_TIND_BLOCK:
2372 		;
2373 	}
2374 
2375 	ext4_discard_reservation(inode);
2376 
2377 	mutex_unlock(&ei->truncate_mutex);
2378 	inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2379 	ext4_mark_inode_dirty(handle, inode);
2380 
2381 	/*
2382 	 * In a multi-transaction truncate, we only make the final transaction
2383 	 * synchronous
2384 	 */
2385 	if (IS_SYNC(inode))
2386 		handle->h_sync = 1;
2387 out_stop:
2388 	/*
2389 	 * If this was a simple ftruncate(), and the file will remain alive
2390 	 * then we need to clear up the orphan record which we created above.
2391 	 * However, if this was a real unlink then we were called by
2392 	 * ext4_delete_inode(), and we allow that function to clean up the
2393 	 * orphan info for us.
2394 	 */
2395 	if (inode->i_nlink)
2396 		ext4_orphan_del(handle, inode);
2397 
2398 	ext4_journal_stop(handle);
2399 }
2400 
2401 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
2402 		unsigned long ino, struct ext4_iloc *iloc)
2403 {
2404 	unsigned long desc, group_desc, block_group;
2405 	unsigned long offset;
2406 	ext4_fsblk_t block;
2407 	struct buffer_head *bh;
2408 	struct ext4_group_desc * gdp;
2409 
2410 	if (!ext4_valid_inum(sb, ino)) {
2411 		/*
2412 		 * This error is already checked for in namei.c unless we are
2413 		 * looking at an NFS filehandle, in which case no error
2414 		 * report is needed
2415 		 */
2416 		return 0;
2417 	}
2418 
2419 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
2420 	if (block_group >= EXT4_SB(sb)->s_groups_count) {
2421 		ext4_error(sb,"ext4_get_inode_block","group >= groups count");
2422 		return 0;
2423 	}
2424 	smp_rmb();
2425 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
2426 	desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2427 	bh = EXT4_SB(sb)->s_group_desc[group_desc];
2428 	if (!bh) {
2429 		ext4_error (sb, "ext4_get_inode_block",
2430 			    "Descriptor not loaded");
2431 		return 0;
2432 	}
2433 
2434 	gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
2435 		desc * EXT4_DESC_SIZE(sb));
2436 	/*
2437 	 * Figure out the offset within the block group inode table
2438 	 */
2439 	offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
2440 		EXT4_INODE_SIZE(sb);
2441 	block = ext4_inode_table(sb, gdp) +
2442 		(offset >> EXT4_BLOCK_SIZE_BITS(sb));
2443 
2444 	iloc->block_group = block_group;
2445 	iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
2446 	return block;
2447 }
2448 
2449 /*
2450  * ext4_get_inode_loc returns with an extra refcount against the inode's
2451  * underlying buffer_head on success. If 'in_mem' is true, we have all
2452  * data in memory that is needed to recreate the on-disk version of this
2453  * inode.
2454  */
2455 static int __ext4_get_inode_loc(struct inode *inode,
2456 				struct ext4_iloc *iloc, int in_mem)
2457 {
2458 	ext4_fsblk_t block;
2459 	struct buffer_head *bh;
2460 
2461 	block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2462 	if (!block)
2463 		return -EIO;
2464 
2465 	bh = sb_getblk(inode->i_sb, block);
2466 	if (!bh) {
2467 		ext4_error (inode->i_sb, "ext4_get_inode_loc",
2468 				"unable to read inode block - "
2469 				"inode=%lu, block=%llu",
2470 				 inode->i_ino, block);
2471 		return -EIO;
2472 	}
2473 	if (!buffer_uptodate(bh)) {
2474 		lock_buffer(bh);
2475 		if (buffer_uptodate(bh)) {
2476 			/* someone brought it uptodate while we waited */
2477 			unlock_buffer(bh);
2478 			goto has_buffer;
2479 		}
2480 
2481 		/*
2482 		 * If we have all information of the inode in memory and this
2483 		 * is the only valid inode in the block, we need not read the
2484 		 * block.
2485 		 */
2486 		if (in_mem) {
2487 			struct buffer_head *bitmap_bh;
2488 			struct ext4_group_desc *desc;
2489 			int inodes_per_buffer;
2490 			int inode_offset, i;
2491 			int block_group;
2492 			int start;
2493 
2494 			block_group = (inode->i_ino - 1) /
2495 					EXT4_INODES_PER_GROUP(inode->i_sb);
2496 			inodes_per_buffer = bh->b_size /
2497 				EXT4_INODE_SIZE(inode->i_sb);
2498 			inode_offset = ((inode->i_ino - 1) %
2499 					EXT4_INODES_PER_GROUP(inode->i_sb));
2500 			start = inode_offset & ~(inodes_per_buffer - 1);
2501 
2502 			/* Is the inode bitmap in cache? */
2503 			desc = ext4_get_group_desc(inode->i_sb,
2504 						block_group, NULL);
2505 			if (!desc)
2506 				goto make_io;
2507 
2508 			bitmap_bh = sb_getblk(inode->i_sb,
2509 				ext4_inode_bitmap(inode->i_sb, desc));
2510 			if (!bitmap_bh)
2511 				goto make_io;
2512 
2513 			/*
2514 			 * If the inode bitmap isn't in cache then the
2515 			 * optimisation may end up performing two reads instead
2516 			 * of one, so skip it.
2517 			 */
2518 			if (!buffer_uptodate(bitmap_bh)) {
2519 				brelse(bitmap_bh);
2520 				goto make_io;
2521 			}
2522 			for (i = start; i < start + inodes_per_buffer; i++) {
2523 				if (i == inode_offset)
2524 					continue;
2525 				if (ext4_test_bit(i, bitmap_bh->b_data))
2526 					break;
2527 			}
2528 			brelse(bitmap_bh);
2529 			if (i == start + inodes_per_buffer) {
2530 				/* all other inodes are free, so skip I/O */
2531 				memset(bh->b_data, 0, bh->b_size);
2532 				set_buffer_uptodate(bh);
2533 				unlock_buffer(bh);
2534 				goto has_buffer;
2535 			}
2536 		}
2537 
2538 make_io:
2539 		/*
2540 		 * There are other valid inodes in the buffer, this inode
2541 		 * has in-inode xattrs, or we don't have this inode in memory.
2542 		 * Read the block from disk.
2543 		 */
2544 		get_bh(bh);
2545 		bh->b_end_io = end_buffer_read_sync;
2546 		submit_bh(READ_META, bh);
2547 		wait_on_buffer(bh);
2548 		if (!buffer_uptodate(bh)) {
2549 			ext4_error(inode->i_sb, "ext4_get_inode_loc",
2550 					"unable to read inode block - "
2551 					"inode=%lu, block=%llu",
2552 					inode->i_ino, block);
2553 			brelse(bh);
2554 			return -EIO;
2555 		}
2556 	}
2557 has_buffer:
2558 	iloc->bh = bh;
2559 	return 0;
2560 }
2561 
2562 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
2563 {
2564 	/* We have all inode data except xattrs in memory here. */
2565 	return __ext4_get_inode_loc(inode, iloc,
2566 		!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
2567 }
2568 
2569 void ext4_set_inode_flags(struct inode *inode)
2570 {
2571 	unsigned int flags = EXT4_I(inode)->i_flags;
2572 
2573 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2574 	if (flags & EXT4_SYNC_FL)
2575 		inode->i_flags |= S_SYNC;
2576 	if (flags & EXT4_APPEND_FL)
2577 		inode->i_flags |= S_APPEND;
2578 	if (flags & EXT4_IMMUTABLE_FL)
2579 		inode->i_flags |= S_IMMUTABLE;
2580 	if (flags & EXT4_NOATIME_FL)
2581 		inode->i_flags |= S_NOATIME;
2582 	if (flags & EXT4_DIRSYNC_FL)
2583 		inode->i_flags |= S_DIRSYNC;
2584 }
2585 
2586 void ext4_read_inode(struct inode * inode)
2587 {
2588 	struct ext4_iloc iloc;
2589 	struct ext4_inode *raw_inode;
2590 	struct ext4_inode_info *ei = EXT4_I(inode);
2591 	struct buffer_head *bh;
2592 	int block;
2593 
2594 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2595 	ei->i_acl = EXT4_ACL_NOT_CACHED;
2596 	ei->i_default_acl = EXT4_ACL_NOT_CACHED;
2597 #endif
2598 	ei->i_block_alloc_info = NULL;
2599 
2600 	if (__ext4_get_inode_loc(inode, &iloc, 0))
2601 		goto bad_inode;
2602 	bh = iloc.bh;
2603 	raw_inode = ext4_raw_inode(&iloc);
2604 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2605 	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2606 	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2607 	if(!(test_opt (inode->i_sb, NO_UID32))) {
2608 		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2609 		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2610 	}
2611 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2612 	inode->i_size = le32_to_cpu(raw_inode->i_size);
2613 	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2614 	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2615 	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
2616 	inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2617 
2618 	ei->i_state = 0;
2619 	ei->i_dir_start_lookup = 0;
2620 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2621 	/* We now have enough fields to check if the inode was active or not.
2622 	 * This is needed because nfsd might try to access dead inodes
2623 	 * the test is that same one that e2fsck uses
2624 	 * NeilBrown 1999oct15
2625 	 */
2626 	if (inode->i_nlink == 0) {
2627 		if (inode->i_mode == 0 ||
2628 		    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
2629 			/* this inode is deleted */
2630 			brelse (bh);
2631 			goto bad_inode;
2632 		}
2633 		/* The only unlinked inodes we let through here have
2634 		 * valid i_mode and are being read by the orphan
2635 		 * recovery code: that's fine, we're about to complete
2636 		 * the process of deleting those. */
2637 	}
2638 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2639 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2640 #ifdef EXT4_FRAGMENTS
2641 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2642 	ei->i_frag_no = raw_inode->i_frag;
2643 	ei->i_frag_size = raw_inode->i_fsize;
2644 #endif
2645 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2646 	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2647 	    cpu_to_le32(EXT4_OS_HURD))
2648 		ei->i_file_acl |=
2649 			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
2650 	if (!S_ISREG(inode->i_mode)) {
2651 		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2652 	} else {
2653 		inode->i_size |=
2654 			((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2655 	}
2656 	ei->i_disksize = inode->i_size;
2657 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2658 	ei->i_block_group = iloc.block_group;
2659 	/*
2660 	 * NOTE! The in-memory inode i_data array is in little-endian order
2661 	 * even on big-endian machines: we do NOT byteswap the block numbers!
2662 	 */
2663 	for (block = 0; block < EXT4_N_BLOCKS; block++)
2664 		ei->i_data[block] = raw_inode->i_block[block];
2665 	INIT_LIST_HEAD(&ei->i_orphan);
2666 
2667 	if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
2668 	    EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
2669 		/*
2670 		 * When mke2fs creates big inodes it does not zero out
2671 		 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
2672 		 * so ignore those first few inodes.
2673 		 */
2674 		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2675 		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2676 		    EXT4_INODE_SIZE(inode->i_sb))
2677 			goto bad_inode;
2678 		if (ei->i_extra_isize == 0) {
2679 			/* The extra space is currently unused. Use it. */
2680 			ei->i_extra_isize = sizeof(struct ext4_inode) -
2681 					    EXT4_GOOD_OLD_INODE_SIZE;
2682 		} else {
2683 			__le32 *magic = (void *)raw_inode +
2684 					EXT4_GOOD_OLD_INODE_SIZE +
2685 					ei->i_extra_isize;
2686 			if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2687 				 ei->i_state |= EXT4_STATE_XATTR;
2688 		}
2689 	} else
2690 		ei->i_extra_isize = 0;
2691 
2692 	if (S_ISREG(inode->i_mode)) {
2693 		inode->i_op = &ext4_file_inode_operations;
2694 		inode->i_fop = &ext4_file_operations;
2695 		ext4_set_aops(inode);
2696 	} else if (S_ISDIR(inode->i_mode)) {
2697 		inode->i_op = &ext4_dir_inode_operations;
2698 		inode->i_fop = &ext4_dir_operations;
2699 	} else if (S_ISLNK(inode->i_mode)) {
2700 		if (ext4_inode_is_fast_symlink(inode))
2701 			inode->i_op = &ext4_fast_symlink_inode_operations;
2702 		else {
2703 			inode->i_op = &ext4_symlink_inode_operations;
2704 			ext4_set_aops(inode);
2705 		}
2706 	} else {
2707 		inode->i_op = &ext4_special_inode_operations;
2708 		if (raw_inode->i_block[0])
2709 			init_special_inode(inode, inode->i_mode,
2710 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2711 		else
2712 			init_special_inode(inode, inode->i_mode,
2713 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2714 	}
2715 	brelse (iloc.bh);
2716 	ext4_set_inode_flags(inode);
2717 	return;
2718 
2719 bad_inode:
2720 	make_bad_inode(inode);
2721 	return;
2722 }
2723 
2724 /*
2725  * Post the struct inode info into an on-disk inode location in the
2726  * buffer-cache.  This gobbles the caller's reference to the
2727  * buffer_head in the inode location struct.
2728  *
2729  * The caller must have write access to iloc->bh.
2730  */
2731 static int ext4_do_update_inode(handle_t *handle,
2732 				struct inode *inode,
2733 				struct ext4_iloc *iloc)
2734 {
2735 	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
2736 	struct ext4_inode_info *ei = EXT4_I(inode);
2737 	struct buffer_head *bh = iloc->bh;
2738 	int err = 0, rc, block;
2739 
2740 	/* For fields not not tracking in the in-memory inode,
2741 	 * initialise them to zero for new inodes. */
2742 	if (ei->i_state & EXT4_STATE_NEW)
2743 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2744 
2745 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2746 	if(!(test_opt(inode->i_sb, NO_UID32))) {
2747 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2748 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2749 /*
2750  * Fix up interoperability with old kernels. Otherwise, old inodes get
2751  * re-used with the upper 16 bits of the uid/gid intact
2752  */
2753 		if(!ei->i_dtime) {
2754 			raw_inode->i_uid_high =
2755 				cpu_to_le16(high_16_bits(inode->i_uid));
2756 			raw_inode->i_gid_high =
2757 				cpu_to_le16(high_16_bits(inode->i_gid));
2758 		} else {
2759 			raw_inode->i_uid_high = 0;
2760 			raw_inode->i_gid_high = 0;
2761 		}
2762 	} else {
2763 		raw_inode->i_uid_low =
2764 			cpu_to_le16(fs_high2lowuid(inode->i_uid));
2765 		raw_inode->i_gid_low =
2766 			cpu_to_le16(fs_high2lowgid(inode->i_gid));
2767 		raw_inode->i_uid_high = 0;
2768 		raw_inode->i_gid_high = 0;
2769 	}
2770 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2771 	raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2772 	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2773 	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2774 	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2775 	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2776 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2777 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2778 #ifdef EXT4_FRAGMENTS
2779 	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2780 	raw_inode->i_frag = ei->i_frag_no;
2781 	raw_inode->i_fsize = ei->i_frag_size;
2782 #endif
2783 	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2784 	    cpu_to_le32(EXT4_OS_HURD))
2785 		raw_inode->i_file_acl_high =
2786 			cpu_to_le16(ei->i_file_acl >> 32);
2787 	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2788 	if (!S_ISREG(inode->i_mode)) {
2789 		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2790 	} else {
2791 		raw_inode->i_size_high =
2792 			cpu_to_le32(ei->i_disksize >> 32);
2793 		if (ei->i_disksize > 0x7fffffffULL) {
2794 			struct super_block *sb = inode->i_sb;
2795 			if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
2796 					EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
2797 			    EXT4_SB(sb)->s_es->s_rev_level ==
2798 					cpu_to_le32(EXT4_GOOD_OLD_REV)) {
2799 			       /* If this is the first large file
2800 				* created, add a flag to the superblock.
2801 				*/
2802 				err = ext4_journal_get_write_access(handle,
2803 						EXT4_SB(sb)->s_sbh);
2804 				if (err)
2805 					goto out_brelse;
2806 				ext4_update_dynamic_rev(sb);
2807 				EXT4_SET_RO_COMPAT_FEATURE(sb,
2808 					EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
2809 				sb->s_dirt = 1;
2810 				handle->h_sync = 1;
2811 				err = ext4_journal_dirty_metadata(handle,
2812 						EXT4_SB(sb)->s_sbh);
2813 			}
2814 		}
2815 	}
2816 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2817 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2818 		if (old_valid_dev(inode->i_rdev)) {
2819 			raw_inode->i_block[0] =
2820 				cpu_to_le32(old_encode_dev(inode->i_rdev));
2821 			raw_inode->i_block[1] = 0;
2822 		} else {
2823 			raw_inode->i_block[0] = 0;
2824 			raw_inode->i_block[1] =
2825 				cpu_to_le32(new_encode_dev(inode->i_rdev));
2826 			raw_inode->i_block[2] = 0;
2827 		}
2828 	} else for (block = 0; block < EXT4_N_BLOCKS; block++)
2829 		raw_inode->i_block[block] = ei->i_data[block];
2830 
2831 	if (ei->i_extra_isize)
2832 		raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2833 
2834 	BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2835 	rc = ext4_journal_dirty_metadata(handle, bh);
2836 	if (!err)
2837 		err = rc;
2838 	ei->i_state &= ~EXT4_STATE_NEW;
2839 
2840 out_brelse:
2841 	brelse (bh);
2842 	ext4_std_error(inode->i_sb, err);
2843 	return err;
2844 }
2845 
2846 /*
2847  * ext4_write_inode()
2848  *
2849  * We are called from a few places:
2850  *
2851  * - Within generic_file_write() for O_SYNC files.
2852  *   Here, there will be no transaction running. We wait for any running
2853  *   trasnaction to commit.
2854  *
2855  * - Within sys_sync(), kupdate and such.
2856  *   We wait on commit, if tol to.
2857  *
2858  * - Within prune_icache() (PF_MEMALLOC == true)
2859  *   Here we simply return.  We can't afford to block kswapd on the
2860  *   journal commit.
2861  *
2862  * In all cases it is actually safe for us to return without doing anything,
2863  * because the inode has been copied into a raw inode buffer in
2864  * ext4_mark_inode_dirty().  This is a correctness thing for O_SYNC and for
2865  * knfsd.
2866  *
2867  * Note that we are absolutely dependent upon all inode dirtiers doing the
2868  * right thing: they *must* call mark_inode_dirty() after dirtying info in
2869  * which we are interested.
2870  *
2871  * It would be a bug for them to not do this.  The code:
2872  *
2873  *	mark_inode_dirty(inode)
2874  *	stuff();
2875  *	inode->i_size = expr;
2876  *
2877  * is in error because a kswapd-driven write_inode() could occur while
2878  * `stuff()' is running, and the new i_size will be lost.  Plus the inode
2879  * will no longer be on the superblock's dirty inode list.
2880  */
2881 int ext4_write_inode(struct inode *inode, int wait)
2882 {
2883 	if (current->flags & PF_MEMALLOC)
2884 		return 0;
2885 
2886 	if (ext4_journal_current_handle()) {
2887 		jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2888 		dump_stack();
2889 		return -EIO;
2890 	}
2891 
2892 	if (!wait)
2893 		return 0;
2894 
2895 	return ext4_force_commit(inode->i_sb);
2896 }
2897 
2898 /*
2899  * ext4_setattr()
2900  *
2901  * Called from notify_change.
2902  *
2903  * We want to trap VFS attempts to truncate the file as soon as
2904  * possible.  In particular, we want to make sure that when the VFS
2905  * shrinks i_size, we put the inode on the orphan list and modify
2906  * i_disksize immediately, so that during the subsequent flushing of
2907  * dirty pages and freeing of disk blocks, we can guarantee that any
2908  * commit will leave the blocks being flushed in an unused state on
2909  * disk.  (On recovery, the inode will get truncated and the blocks will
2910  * be freed, so we have a strong guarantee that no future commit will
2911  * leave these blocks visible to the user.)
2912  *
2913  * Called with inode->sem down.
2914  */
2915 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
2916 {
2917 	struct inode *inode = dentry->d_inode;
2918 	int error, rc = 0;
2919 	const unsigned int ia_valid = attr->ia_valid;
2920 
2921 	error = inode_change_ok(inode, attr);
2922 	if (error)
2923 		return error;
2924 
2925 	if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2926 		(ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2927 		handle_t *handle;
2928 
2929 		/* (user+group)*(old+new) structure, inode write (sb,
2930 		 * inode block, ? - but truncate inode update has it) */
2931 		handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
2932 					EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
2933 		if (IS_ERR(handle)) {
2934 			error = PTR_ERR(handle);
2935 			goto err_out;
2936 		}
2937 		error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2938 		if (error) {
2939 			ext4_journal_stop(handle);
2940 			return error;
2941 		}
2942 		/* Update corresponding info in inode so that everything is in
2943 		 * one transaction */
2944 		if (attr->ia_valid & ATTR_UID)
2945 			inode->i_uid = attr->ia_uid;
2946 		if (attr->ia_valid & ATTR_GID)
2947 			inode->i_gid = attr->ia_gid;
2948 		error = ext4_mark_inode_dirty(handle, inode);
2949 		ext4_journal_stop(handle);
2950 	}
2951 
2952 	if (S_ISREG(inode->i_mode) &&
2953 	    attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2954 		handle_t *handle;
2955 
2956 		handle = ext4_journal_start(inode, 3);
2957 		if (IS_ERR(handle)) {
2958 			error = PTR_ERR(handle);
2959 			goto err_out;
2960 		}
2961 
2962 		error = ext4_orphan_add(handle, inode);
2963 		EXT4_I(inode)->i_disksize = attr->ia_size;
2964 		rc = ext4_mark_inode_dirty(handle, inode);
2965 		if (!error)
2966 			error = rc;
2967 		ext4_journal_stop(handle);
2968 	}
2969 
2970 	rc = inode_setattr(inode, attr);
2971 
2972 	/* If inode_setattr's call to ext4_truncate failed to get a
2973 	 * transaction handle at all, we need to clean up the in-core
2974 	 * orphan list manually. */
2975 	if (inode->i_nlink)
2976 		ext4_orphan_del(NULL, inode);
2977 
2978 	if (!rc && (ia_valid & ATTR_MODE))
2979 		rc = ext4_acl_chmod(inode);
2980 
2981 err_out:
2982 	ext4_std_error(inode->i_sb, error);
2983 	if (!error)
2984 		error = rc;
2985 	return error;
2986 }
2987 
2988 
2989 /*
2990  * How many blocks doth make a writepage()?
2991  *
2992  * With N blocks per page, it may be:
2993  * N data blocks
2994  * 2 indirect block
2995  * 2 dindirect
2996  * 1 tindirect
2997  * N+5 bitmap blocks (from the above)
2998  * N+5 group descriptor summary blocks
2999  * 1 inode block
3000  * 1 superblock.
3001  * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
3002  *
3003  * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
3004  *
3005  * With ordered or writeback data it's the same, less the N data blocks.
3006  *
3007  * If the inode's direct blocks can hold an integral number of pages then a
3008  * page cannot straddle two indirect blocks, and we can only touch one indirect
3009  * and dindirect block, and the "5" above becomes "3".
3010  *
3011  * This still overestimates under most circumstances.  If we were to pass the
3012  * start and end offsets in here as well we could do block_to_path() on each
3013  * block and work out the exact number of indirects which are touched.  Pah.
3014  */
3015 
3016 int ext4_writepage_trans_blocks(struct inode *inode)
3017 {
3018 	int bpp = ext4_journal_blocks_per_page(inode);
3019 	int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
3020 	int ret;
3021 
3022 	if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3023 		return ext4_ext_writepage_trans_blocks(inode, bpp);
3024 
3025 	if (ext4_should_journal_data(inode))
3026 		ret = 3 * (bpp + indirects) + 2;
3027 	else
3028 		ret = 2 * (bpp + indirects) + 2;
3029 
3030 #ifdef CONFIG_QUOTA
3031 	/* We know that structure was already allocated during DQUOT_INIT so
3032 	 * we will be updating only the data blocks + inodes */
3033 	ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
3034 #endif
3035 
3036 	return ret;
3037 }
3038 
3039 /*
3040  * The caller must have previously called ext4_reserve_inode_write().
3041  * Give this, we know that the caller already has write access to iloc->bh.
3042  */
3043 int ext4_mark_iloc_dirty(handle_t *handle,
3044 		struct inode *inode, struct ext4_iloc *iloc)
3045 {
3046 	int err = 0;
3047 
3048 	/* the do_update_inode consumes one bh->b_count */
3049 	get_bh(iloc->bh);
3050 
3051 	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
3052 	err = ext4_do_update_inode(handle, inode, iloc);
3053 	put_bh(iloc->bh);
3054 	return err;
3055 }
3056 
3057 /*
3058  * On success, We end up with an outstanding reference count against
3059  * iloc->bh.  This _must_ be cleaned up later.
3060  */
3061 
3062 int
3063 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
3064 			 struct ext4_iloc *iloc)
3065 {
3066 	int err = 0;
3067 	if (handle) {
3068 		err = ext4_get_inode_loc(inode, iloc);
3069 		if (!err) {
3070 			BUFFER_TRACE(iloc->bh, "get_write_access");
3071 			err = ext4_journal_get_write_access(handle, iloc->bh);
3072 			if (err) {
3073 				brelse(iloc->bh);
3074 				iloc->bh = NULL;
3075 			}
3076 		}
3077 	}
3078 	ext4_std_error(inode->i_sb, err);
3079 	return err;
3080 }
3081 
3082 /*
3083  * What we do here is to mark the in-core inode as clean with respect to inode
3084  * dirtiness (it may still be data-dirty).
3085  * This means that the in-core inode may be reaped by prune_icache
3086  * without having to perform any I/O.  This is a very good thing,
3087  * because *any* task may call prune_icache - even ones which
3088  * have a transaction open against a different journal.
3089  *
3090  * Is this cheating?  Not really.  Sure, we haven't written the
3091  * inode out, but prune_icache isn't a user-visible syncing function.
3092  * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3093  * we start and wait on commits.
3094  *
3095  * Is this efficient/effective?  Well, we're being nice to the system
3096  * by cleaning up our inodes proactively so they can be reaped
3097  * without I/O.  But we are potentially leaving up to five seconds'
3098  * worth of inodes floating about which prune_icache wants us to
3099  * write out.  One way to fix that would be to get prune_icache()
3100  * to do a write_super() to free up some memory.  It has the desired
3101  * effect.
3102  */
3103 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
3104 {
3105 	struct ext4_iloc iloc;
3106 	int err;
3107 
3108 	might_sleep();
3109 	err = ext4_reserve_inode_write(handle, inode, &iloc);
3110 	if (!err)
3111 		err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3112 	return err;
3113 }
3114 
3115 /*
3116  * ext4_dirty_inode() is called from __mark_inode_dirty()
3117  *
3118  * We're really interested in the case where a file is being extended.
3119  * i_size has been changed by generic_commit_write() and we thus need
3120  * to include the updated inode in the current transaction.
3121  *
3122  * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3123  * are allocated to the file.
3124  *
3125  * If the inode is marked synchronous, we don't honour that here - doing
3126  * so would cause a commit on atime updates, which we don't bother doing.
3127  * We handle synchronous inodes at the highest possible level.
3128  */
3129 void ext4_dirty_inode(struct inode *inode)
3130 {
3131 	handle_t *current_handle = ext4_journal_current_handle();
3132 	handle_t *handle;
3133 
3134 	handle = ext4_journal_start(inode, 2);
3135 	if (IS_ERR(handle))
3136 		goto out;
3137 	if (current_handle &&
3138 		current_handle->h_transaction != handle->h_transaction) {
3139 		/* This task has a transaction open against a different fs */
3140 		printk(KERN_EMERG "%s: transactions do not match!\n",
3141 		       __FUNCTION__);
3142 	} else {
3143 		jbd_debug(5, "marking dirty.  outer handle=%p\n",
3144 				current_handle);
3145 		ext4_mark_inode_dirty(handle, inode);
3146 	}
3147 	ext4_journal_stop(handle);
3148 out:
3149 	return;
3150 }
3151 
3152 #if 0
3153 /*
3154  * Bind an inode's backing buffer_head into this transaction, to prevent
3155  * it from being flushed to disk early.  Unlike
3156  * ext4_reserve_inode_write, this leaves behind no bh reference and
3157  * returns no iloc structure, so the caller needs to repeat the iloc
3158  * lookup to mark the inode dirty later.
3159  */
3160 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
3161 {
3162 	struct ext4_iloc iloc;
3163 
3164 	int err = 0;
3165 	if (handle) {
3166 		err = ext4_get_inode_loc(inode, &iloc);
3167 		if (!err) {
3168 			BUFFER_TRACE(iloc.bh, "get_write_access");
3169 			err = jbd2_journal_get_write_access(handle, iloc.bh);
3170 			if (!err)
3171 				err = ext4_journal_dirty_metadata(handle,
3172 								  iloc.bh);
3173 			brelse(iloc.bh);
3174 		}
3175 	}
3176 	ext4_std_error(inode->i_sb, err);
3177 	return err;
3178 }
3179 #endif
3180 
3181 int ext4_change_inode_journal_flag(struct inode *inode, int val)
3182 {
3183 	journal_t *journal;
3184 	handle_t *handle;
3185 	int err;
3186 
3187 	/*
3188 	 * We have to be very careful here: changing a data block's
3189 	 * journaling status dynamically is dangerous.  If we write a
3190 	 * data block to the journal, change the status and then delete
3191 	 * that block, we risk forgetting to revoke the old log record
3192 	 * from the journal and so a subsequent replay can corrupt data.
3193 	 * So, first we make sure that the journal is empty and that
3194 	 * nobody is changing anything.
3195 	 */
3196 
3197 	journal = EXT4_JOURNAL(inode);
3198 	if (is_journal_aborted(journal) || IS_RDONLY(inode))
3199 		return -EROFS;
3200 
3201 	jbd2_journal_lock_updates(journal);
3202 	jbd2_journal_flush(journal);
3203 
3204 	/*
3205 	 * OK, there are no updates running now, and all cached data is
3206 	 * synced to disk.  We are now in a completely consistent state
3207 	 * which doesn't have anything in the journal, and we know that
3208 	 * no filesystem updates are running, so it is safe to modify
3209 	 * the inode's in-core data-journaling state flag now.
3210 	 */
3211 
3212 	if (val)
3213 		EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
3214 	else
3215 		EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
3216 	ext4_set_aops(inode);
3217 
3218 	jbd2_journal_unlock_updates(journal);
3219 
3220 	/* Finally we can mark the inode as dirty. */
3221 
3222 	handle = ext4_journal_start(inode, 1);
3223 	if (IS_ERR(handle))
3224 		return PTR_ERR(handle);
3225 
3226 	err = ext4_mark_inode_dirty(handle, inode);
3227 	handle->h_sync = 1;
3228 	ext4_journal_stop(handle);
3229 	ext4_std_error(inode->i_sb, err);
3230 
3231 	return err;
3232 }
3233