xref: /openbmc/linux/fs/ext4/inode.c (revision fd589a8f)
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/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 
41 #include "ext4_jbd2.h"
42 #include "xattr.h"
43 #include "acl.h"
44 #include "ext4_extents.h"
45 
46 #include <trace/events/ext4.h>
47 
48 #define MPAGE_DA_EXTENT_TAIL 0x01
49 
50 static inline int ext4_begin_ordered_truncate(struct inode *inode,
51 					      loff_t new_size)
52 {
53 	return jbd2_journal_begin_ordered_truncate(
54 					EXT4_SB(inode->i_sb)->s_journal,
55 					&EXT4_I(inode)->jinode,
56 					new_size);
57 }
58 
59 static void ext4_invalidatepage(struct page *page, unsigned long offset);
60 
61 /*
62  * Test whether an inode is a fast symlink.
63  */
64 static int ext4_inode_is_fast_symlink(struct inode *inode)
65 {
66 	int ea_blocks = EXT4_I(inode)->i_file_acl ?
67 		(inode->i_sb->s_blocksize >> 9) : 0;
68 
69 	return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
70 }
71 
72 /*
73  * The ext4 forget function must perform a revoke if we are freeing data
74  * which has been journaled.  Metadata (eg. indirect blocks) must be
75  * revoked in all cases.
76  *
77  * "bh" may be NULL: a metadata block may have been freed from memory
78  * but there may still be a record of it in the journal, and that record
79  * still needs to be revoked.
80  *
81  * If the handle isn't valid we're not journaling, but we still need to
82  * call into ext4_journal_revoke() to put the buffer head.
83  */
84 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
85 		struct buffer_head *bh, ext4_fsblk_t blocknr)
86 {
87 	int err;
88 
89 	might_sleep();
90 
91 	BUFFER_TRACE(bh, "enter");
92 
93 	jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
94 		  "data mode %x\n",
95 		  bh, is_metadata, inode->i_mode,
96 		  test_opt(inode->i_sb, DATA_FLAGS));
97 
98 	/* Never use the revoke function if we are doing full data
99 	 * journaling: there is no need to, and a V1 superblock won't
100 	 * support it.  Otherwise, only skip the revoke on un-journaled
101 	 * data blocks. */
102 
103 	if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
104 	    (!is_metadata && !ext4_should_journal_data(inode))) {
105 		if (bh) {
106 			BUFFER_TRACE(bh, "call jbd2_journal_forget");
107 			return ext4_journal_forget(handle, bh);
108 		}
109 		return 0;
110 	}
111 
112 	/*
113 	 * data!=journal && (is_metadata || should_journal_data(inode))
114 	 */
115 	BUFFER_TRACE(bh, "call ext4_journal_revoke");
116 	err = ext4_journal_revoke(handle, blocknr, bh);
117 	if (err)
118 		ext4_abort(inode->i_sb, __func__,
119 			   "error %d when attempting revoke", err);
120 	BUFFER_TRACE(bh, "exit");
121 	return err;
122 }
123 
124 /*
125  * Work out how many blocks we need to proceed with the next chunk of a
126  * truncate transaction.
127  */
128 static unsigned long blocks_for_truncate(struct inode *inode)
129 {
130 	ext4_lblk_t needed;
131 
132 	needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
133 
134 	/* Give ourselves just enough room to cope with inodes in which
135 	 * i_blocks is corrupt: we've seen disk corruptions in the past
136 	 * which resulted in random data in an inode which looked enough
137 	 * like a regular file for ext4 to try to delete it.  Things
138 	 * will go a bit crazy if that happens, but at least we should
139 	 * try not to panic the whole kernel. */
140 	if (needed < 2)
141 		needed = 2;
142 
143 	/* But we need to bound the transaction so we don't overflow the
144 	 * journal. */
145 	if (needed > EXT4_MAX_TRANS_DATA)
146 		needed = EXT4_MAX_TRANS_DATA;
147 
148 	return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
149 }
150 
151 /*
152  * Truncate transactions can be complex and absolutely huge.  So we need to
153  * be able to restart the transaction at a conventient checkpoint to make
154  * sure we don't overflow the journal.
155  *
156  * start_transaction gets us a new handle for a truncate transaction,
157  * and extend_transaction tries to extend the existing one a bit.  If
158  * extend fails, we need to propagate the failure up and restart the
159  * transaction in the top-level truncate loop. --sct
160  */
161 static handle_t *start_transaction(struct inode *inode)
162 {
163 	handle_t *result;
164 
165 	result = ext4_journal_start(inode, blocks_for_truncate(inode));
166 	if (!IS_ERR(result))
167 		return result;
168 
169 	ext4_std_error(inode->i_sb, PTR_ERR(result));
170 	return result;
171 }
172 
173 /*
174  * Try to extend this transaction for the purposes of truncation.
175  *
176  * Returns 0 if we managed to create more room.  If we can't create more
177  * room, and the transaction must be restarted we return 1.
178  */
179 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
180 {
181 	if (!ext4_handle_valid(handle))
182 		return 0;
183 	if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
184 		return 0;
185 	if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
186 		return 0;
187 	return 1;
188 }
189 
190 /*
191  * Restart the transaction associated with *handle.  This does a commit,
192  * so before we call here everything must be consistently dirtied against
193  * this transaction.
194  */
195  int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
196 				 int nblocks)
197 {
198 	int ret;
199 
200 	/*
201 	 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
202 	 * moment, get_block can be called only for blocks inside i_size since
203 	 * page cache has been already dropped and writes are blocked by
204 	 * i_mutex. So we can safely drop the i_data_sem here.
205 	 */
206 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
207 	jbd_debug(2, "restarting handle %p\n", handle);
208 	up_write(&EXT4_I(inode)->i_data_sem);
209 	ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
210 	down_write(&EXT4_I(inode)->i_data_sem);
211 
212 	return ret;
213 }
214 
215 /*
216  * Called at the last iput() if i_nlink is zero.
217  */
218 void ext4_delete_inode(struct inode *inode)
219 {
220 	handle_t *handle;
221 	int err;
222 
223 	if (ext4_should_order_data(inode))
224 		ext4_begin_ordered_truncate(inode, 0);
225 	truncate_inode_pages(&inode->i_data, 0);
226 
227 	if (is_bad_inode(inode))
228 		goto no_delete;
229 
230 	handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
231 	if (IS_ERR(handle)) {
232 		ext4_std_error(inode->i_sb, PTR_ERR(handle));
233 		/*
234 		 * If we're going to skip the normal cleanup, we still need to
235 		 * make sure that the in-core orphan linked list is properly
236 		 * cleaned up.
237 		 */
238 		ext4_orphan_del(NULL, inode);
239 		goto no_delete;
240 	}
241 
242 	if (IS_SYNC(inode))
243 		ext4_handle_sync(handle);
244 	inode->i_size = 0;
245 	err = ext4_mark_inode_dirty(handle, inode);
246 	if (err) {
247 		ext4_warning(inode->i_sb, __func__,
248 			     "couldn't mark inode dirty (err %d)", err);
249 		goto stop_handle;
250 	}
251 	if (inode->i_blocks)
252 		ext4_truncate(inode);
253 
254 	/*
255 	 * ext4_ext_truncate() doesn't reserve any slop when it
256 	 * restarts journal transactions; therefore there may not be
257 	 * enough credits left in the handle to remove the inode from
258 	 * the orphan list and set the dtime field.
259 	 */
260 	if (!ext4_handle_has_enough_credits(handle, 3)) {
261 		err = ext4_journal_extend(handle, 3);
262 		if (err > 0)
263 			err = ext4_journal_restart(handle, 3);
264 		if (err != 0) {
265 			ext4_warning(inode->i_sb, __func__,
266 				     "couldn't extend journal (err %d)", err);
267 		stop_handle:
268 			ext4_journal_stop(handle);
269 			goto no_delete;
270 		}
271 	}
272 
273 	/*
274 	 * Kill off the orphan record which ext4_truncate created.
275 	 * AKPM: I think this can be inside the above `if'.
276 	 * Note that ext4_orphan_del() has to be able to cope with the
277 	 * deletion of a non-existent orphan - this is because we don't
278 	 * know if ext4_truncate() actually created an orphan record.
279 	 * (Well, we could do this if we need to, but heck - it works)
280 	 */
281 	ext4_orphan_del(handle, inode);
282 	EXT4_I(inode)->i_dtime	= get_seconds();
283 
284 	/*
285 	 * One subtle ordering requirement: if anything has gone wrong
286 	 * (transaction abort, IO errors, whatever), then we can still
287 	 * do these next steps (the fs will already have been marked as
288 	 * having errors), but we can't free the inode if the mark_dirty
289 	 * fails.
290 	 */
291 	if (ext4_mark_inode_dirty(handle, inode))
292 		/* If that failed, just do the required in-core inode clear. */
293 		clear_inode(inode);
294 	else
295 		ext4_free_inode(handle, inode);
296 	ext4_journal_stop(handle);
297 	return;
298 no_delete:
299 	clear_inode(inode);	/* We must guarantee clearing of inode... */
300 }
301 
302 typedef struct {
303 	__le32	*p;
304 	__le32	key;
305 	struct buffer_head *bh;
306 } Indirect;
307 
308 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
309 {
310 	p->key = *(p->p = v);
311 	p->bh = bh;
312 }
313 
314 /**
315  *	ext4_block_to_path - parse the block number into array of offsets
316  *	@inode: inode in question (we are only interested in its superblock)
317  *	@i_block: block number to be parsed
318  *	@offsets: array to store the offsets in
319  *	@boundary: set this non-zero if the referred-to block is likely to be
320  *	       followed (on disk) by an indirect block.
321  *
322  *	To store the locations of file's data ext4 uses a data structure common
323  *	for UNIX filesystems - tree of pointers anchored in the inode, with
324  *	data blocks at leaves and indirect blocks in intermediate nodes.
325  *	This function translates the block number into path in that tree -
326  *	return value is the path length and @offsets[n] is the offset of
327  *	pointer to (n+1)th node in the nth one. If @block is out of range
328  *	(negative or too large) warning is printed and zero returned.
329  *
330  *	Note: function doesn't find node addresses, so no IO is needed. All
331  *	we need to know is the capacity of indirect blocks (taken from the
332  *	inode->i_sb).
333  */
334 
335 /*
336  * Portability note: the last comparison (check that we fit into triple
337  * indirect block) is spelled differently, because otherwise on an
338  * architecture with 32-bit longs and 8Kb pages we might get into trouble
339  * if our filesystem had 8Kb blocks. We might use long long, but that would
340  * kill us on x86. Oh, well, at least the sign propagation does not matter -
341  * i_block would have to be negative in the very beginning, so we would not
342  * get there at all.
343  */
344 
345 static int ext4_block_to_path(struct inode *inode,
346 			      ext4_lblk_t i_block,
347 			      ext4_lblk_t offsets[4], int *boundary)
348 {
349 	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
350 	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
351 	const long direct_blocks = EXT4_NDIR_BLOCKS,
352 		indirect_blocks = ptrs,
353 		double_blocks = (1 << (ptrs_bits * 2));
354 	int n = 0;
355 	int final = 0;
356 
357 	if (i_block < direct_blocks) {
358 		offsets[n++] = i_block;
359 		final = direct_blocks;
360 	} else if ((i_block -= direct_blocks) < indirect_blocks) {
361 		offsets[n++] = EXT4_IND_BLOCK;
362 		offsets[n++] = i_block;
363 		final = ptrs;
364 	} else if ((i_block -= indirect_blocks) < double_blocks) {
365 		offsets[n++] = EXT4_DIND_BLOCK;
366 		offsets[n++] = i_block >> ptrs_bits;
367 		offsets[n++] = i_block & (ptrs - 1);
368 		final = ptrs;
369 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
370 		offsets[n++] = EXT4_TIND_BLOCK;
371 		offsets[n++] = i_block >> (ptrs_bits * 2);
372 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
373 		offsets[n++] = i_block & (ptrs - 1);
374 		final = ptrs;
375 	} else {
376 		ext4_warning(inode->i_sb, "ext4_block_to_path",
377 			     "block %lu > max in inode %lu",
378 			     i_block + direct_blocks +
379 			     indirect_blocks + double_blocks, inode->i_ino);
380 	}
381 	if (boundary)
382 		*boundary = final - 1 - (i_block & (ptrs - 1));
383 	return n;
384 }
385 
386 static int __ext4_check_blockref(const char *function, struct inode *inode,
387 				 __le32 *p, unsigned int max)
388 {
389 	__le32 *bref = p;
390 	unsigned int blk;
391 
392 	while (bref < p+max) {
393 		blk = le32_to_cpu(*bref++);
394 		if (blk &&
395 		    unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
396 						    blk, 1))) {
397 			ext4_error(inode->i_sb, function,
398 				   "invalid block reference %u "
399 				   "in inode #%lu", blk, inode->i_ino);
400 			return -EIO;
401 		}
402 	}
403 	return 0;
404 }
405 
406 
407 #define ext4_check_indirect_blockref(inode, bh)                         \
408 	__ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data,  \
409 			      EXT4_ADDR_PER_BLOCK((inode)->i_sb))
410 
411 #define ext4_check_inode_blockref(inode)                                \
412 	__ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data,   \
413 			      EXT4_NDIR_BLOCKS)
414 
415 /**
416  *	ext4_get_branch - read the chain of indirect blocks leading to data
417  *	@inode: inode in question
418  *	@depth: depth of the chain (1 - direct pointer, etc.)
419  *	@offsets: offsets of pointers in inode/indirect blocks
420  *	@chain: place to store the result
421  *	@err: here we store the error value
422  *
423  *	Function fills the array of triples <key, p, bh> and returns %NULL
424  *	if everything went OK or the pointer to the last filled triple
425  *	(incomplete one) otherwise. Upon the return chain[i].key contains
426  *	the number of (i+1)-th block in the chain (as it is stored in memory,
427  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
428  *	number (it points into struct inode for i==0 and into the bh->b_data
429  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
430  *	block for i>0 and NULL for i==0. In other words, it holds the block
431  *	numbers of the chain, addresses they were taken from (and where we can
432  *	verify that chain did not change) and buffer_heads hosting these
433  *	numbers.
434  *
435  *	Function stops when it stumbles upon zero pointer (absent block)
436  *		(pointer to last triple returned, *@err == 0)
437  *	or when it gets an IO error reading an indirect block
438  *		(ditto, *@err == -EIO)
439  *	or when it reads all @depth-1 indirect blocks successfully and finds
440  *	the whole chain, all way to the data (returns %NULL, *err == 0).
441  *
442  *      Need to be called with
443  *      down_read(&EXT4_I(inode)->i_data_sem)
444  */
445 static Indirect *ext4_get_branch(struct inode *inode, int depth,
446 				 ext4_lblk_t  *offsets,
447 				 Indirect chain[4], int *err)
448 {
449 	struct super_block *sb = inode->i_sb;
450 	Indirect *p = chain;
451 	struct buffer_head *bh;
452 
453 	*err = 0;
454 	/* i_data is not going away, no lock needed */
455 	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
456 	if (!p->key)
457 		goto no_block;
458 	while (--depth) {
459 		bh = sb_getblk(sb, le32_to_cpu(p->key));
460 		if (unlikely(!bh))
461 			goto failure;
462 
463 		if (!bh_uptodate_or_lock(bh)) {
464 			if (bh_submit_read(bh) < 0) {
465 				put_bh(bh);
466 				goto failure;
467 			}
468 			/* validate block references */
469 			if (ext4_check_indirect_blockref(inode, bh)) {
470 				put_bh(bh);
471 				goto failure;
472 			}
473 		}
474 
475 		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
476 		/* Reader: end */
477 		if (!p->key)
478 			goto no_block;
479 	}
480 	return NULL;
481 
482 failure:
483 	*err = -EIO;
484 no_block:
485 	return p;
486 }
487 
488 /**
489  *	ext4_find_near - find a place for allocation with sufficient locality
490  *	@inode: owner
491  *	@ind: descriptor of indirect block.
492  *
493  *	This function returns the preferred place for block allocation.
494  *	It is used when heuristic for sequential allocation fails.
495  *	Rules are:
496  *	  + if there is a block to the left of our position - allocate near it.
497  *	  + if pointer will live in indirect block - allocate near that block.
498  *	  + if pointer will live in inode - allocate in the same
499  *	    cylinder group.
500  *
501  * In the latter case we colour the starting block by the callers PID to
502  * prevent it from clashing with concurrent allocations for a different inode
503  * in the same block group.   The PID is used here so that functionally related
504  * files will be close-by on-disk.
505  *
506  *	Caller must make sure that @ind is valid and will stay that way.
507  */
508 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
509 {
510 	struct ext4_inode_info *ei = EXT4_I(inode);
511 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
512 	__le32 *p;
513 	ext4_fsblk_t bg_start;
514 	ext4_fsblk_t last_block;
515 	ext4_grpblk_t colour;
516 	ext4_group_t block_group;
517 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
518 
519 	/* Try to find previous block */
520 	for (p = ind->p - 1; p >= start; p--) {
521 		if (*p)
522 			return le32_to_cpu(*p);
523 	}
524 
525 	/* No such thing, so let's try location of indirect block */
526 	if (ind->bh)
527 		return ind->bh->b_blocknr;
528 
529 	/*
530 	 * It is going to be referred to from the inode itself? OK, just put it
531 	 * into the same cylinder group then.
532 	 */
533 	block_group = ei->i_block_group;
534 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
535 		block_group &= ~(flex_size-1);
536 		if (S_ISREG(inode->i_mode))
537 			block_group++;
538 	}
539 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
540 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
541 
542 	/*
543 	 * If we are doing delayed allocation, we don't need take
544 	 * colour into account.
545 	 */
546 	if (test_opt(inode->i_sb, DELALLOC))
547 		return bg_start;
548 
549 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
550 		colour = (current->pid % 16) *
551 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
552 	else
553 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
554 	return bg_start + colour;
555 }
556 
557 /**
558  *	ext4_find_goal - find a preferred place for allocation.
559  *	@inode: owner
560  *	@block:  block we want
561  *	@partial: pointer to the last triple within a chain
562  *
563  *	Normally this function find the preferred place for block allocation,
564  *	returns it.
565  *	Because this is only used for non-extent files, we limit the block nr
566  *	to 32 bits.
567  */
568 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
569 				   Indirect *partial)
570 {
571 	ext4_fsblk_t goal;
572 
573 	/*
574 	 * XXX need to get goal block from mballoc's data structures
575 	 */
576 
577 	goal = ext4_find_near(inode, partial);
578 	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
579 	return goal;
580 }
581 
582 /**
583  *	ext4_blks_to_allocate: Look up the block map and count the number
584  *	of direct blocks need to be allocated for the given branch.
585  *
586  *	@branch: chain of indirect blocks
587  *	@k: number of blocks need for indirect blocks
588  *	@blks: number of data blocks to be mapped.
589  *	@blocks_to_boundary:  the offset in the indirect block
590  *
591  *	return the total number of blocks to be allocate, including the
592  *	direct and indirect blocks.
593  */
594 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
595 				 int blocks_to_boundary)
596 {
597 	unsigned int count = 0;
598 
599 	/*
600 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
601 	 * then it's clear blocks on that path have not allocated
602 	 */
603 	if (k > 0) {
604 		/* right now we don't handle cross boundary allocation */
605 		if (blks < blocks_to_boundary + 1)
606 			count += blks;
607 		else
608 			count += blocks_to_boundary + 1;
609 		return count;
610 	}
611 
612 	count++;
613 	while (count < blks && count <= blocks_to_boundary &&
614 		le32_to_cpu(*(branch[0].p + count)) == 0) {
615 		count++;
616 	}
617 	return count;
618 }
619 
620 /**
621  *	ext4_alloc_blocks: multiple allocate blocks needed for a branch
622  *	@indirect_blks: the number of blocks need to allocate for indirect
623  *			blocks
624  *
625  *	@new_blocks: on return it will store the new block numbers for
626  *	the indirect blocks(if needed) and the first direct block,
627  *	@blks:	on return it will store the total number of allocated
628  *		direct blocks
629  */
630 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
631 			     ext4_lblk_t iblock, ext4_fsblk_t goal,
632 			     int indirect_blks, int blks,
633 			     ext4_fsblk_t new_blocks[4], int *err)
634 {
635 	struct ext4_allocation_request ar;
636 	int target, i;
637 	unsigned long count = 0, blk_allocated = 0;
638 	int index = 0;
639 	ext4_fsblk_t current_block = 0;
640 	int ret = 0;
641 
642 	/*
643 	 * Here we try to allocate the requested multiple blocks at once,
644 	 * on a best-effort basis.
645 	 * To build a branch, we should allocate blocks for
646 	 * the indirect blocks(if not allocated yet), and at least
647 	 * the first direct block of this branch.  That's the
648 	 * minimum number of blocks need to allocate(required)
649 	 */
650 	/* first we try to allocate the indirect blocks */
651 	target = indirect_blks;
652 	while (target > 0) {
653 		count = target;
654 		/* allocating blocks for indirect blocks and direct blocks */
655 		current_block = ext4_new_meta_blocks(handle, inode,
656 							goal, &count, err);
657 		if (*err)
658 			goto failed_out;
659 
660 		BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS);
661 
662 		target -= count;
663 		/* allocate blocks for indirect blocks */
664 		while (index < indirect_blks && count) {
665 			new_blocks[index++] = current_block++;
666 			count--;
667 		}
668 		if (count > 0) {
669 			/*
670 			 * save the new block number
671 			 * for the first direct block
672 			 */
673 			new_blocks[index] = current_block;
674 			printk(KERN_INFO "%s returned more blocks than "
675 						"requested\n", __func__);
676 			WARN_ON(1);
677 			break;
678 		}
679 	}
680 
681 	target = blks - count ;
682 	blk_allocated = count;
683 	if (!target)
684 		goto allocated;
685 	/* Now allocate data blocks */
686 	memset(&ar, 0, sizeof(ar));
687 	ar.inode = inode;
688 	ar.goal = goal;
689 	ar.len = target;
690 	ar.logical = iblock;
691 	if (S_ISREG(inode->i_mode))
692 		/* enable in-core preallocation only for regular files */
693 		ar.flags = EXT4_MB_HINT_DATA;
694 
695 	current_block = ext4_mb_new_blocks(handle, &ar, err);
696 	BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS);
697 
698 	if (*err && (target == blks)) {
699 		/*
700 		 * if the allocation failed and we didn't allocate
701 		 * any blocks before
702 		 */
703 		goto failed_out;
704 	}
705 	if (!*err) {
706 		if (target == blks) {
707 			/*
708 			 * save the new block number
709 			 * for the first direct block
710 			 */
711 			new_blocks[index] = current_block;
712 		}
713 		blk_allocated += ar.len;
714 	}
715 allocated:
716 	/* total number of blocks allocated for direct blocks */
717 	ret = blk_allocated;
718 	*err = 0;
719 	return ret;
720 failed_out:
721 	for (i = 0; i < index; i++)
722 		ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
723 	return ret;
724 }
725 
726 /**
727  *	ext4_alloc_branch - allocate and set up a chain of blocks.
728  *	@inode: owner
729  *	@indirect_blks: number of allocated indirect blocks
730  *	@blks: number of allocated direct blocks
731  *	@offsets: offsets (in the blocks) to store the pointers to next.
732  *	@branch: place to store the chain in.
733  *
734  *	This function allocates blocks, zeroes out all but the last one,
735  *	links them into chain and (if we are synchronous) writes them to disk.
736  *	In other words, it prepares a branch that can be spliced onto the
737  *	inode. It stores the information about that chain in the branch[], in
738  *	the same format as ext4_get_branch() would do. We are calling it after
739  *	we had read the existing part of chain and partial points to the last
740  *	triple of that (one with zero ->key). Upon the exit we have the same
741  *	picture as after the successful ext4_get_block(), except that in one
742  *	place chain is disconnected - *branch->p is still zero (we did not
743  *	set the last link), but branch->key contains the number that should
744  *	be placed into *branch->p to fill that gap.
745  *
746  *	If allocation fails we free all blocks we've allocated (and forget
747  *	their buffer_heads) and return the error value the from failed
748  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
749  *	as described above and return 0.
750  */
751 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
752 			     ext4_lblk_t iblock, int indirect_blks,
753 			     int *blks, ext4_fsblk_t goal,
754 			     ext4_lblk_t *offsets, Indirect *branch)
755 {
756 	int blocksize = inode->i_sb->s_blocksize;
757 	int i, n = 0;
758 	int err = 0;
759 	struct buffer_head *bh;
760 	int num;
761 	ext4_fsblk_t new_blocks[4];
762 	ext4_fsblk_t current_block;
763 
764 	num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
765 				*blks, new_blocks, &err);
766 	if (err)
767 		return err;
768 
769 	branch[0].key = cpu_to_le32(new_blocks[0]);
770 	/*
771 	 * metadata blocks and data blocks are allocated.
772 	 */
773 	for (n = 1; n <= indirect_blks;  n++) {
774 		/*
775 		 * Get buffer_head for parent block, zero it out
776 		 * and set the pointer to new one, then send
777 		 * parent to disk.
778 		 */
779 		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
780 		branch[n].bh = bh;
781 		lock_buffer(bh);
782 		BUFFER_TRACE(bh, "call get_create_access");
783 		err = ext4_journal_get_create_access(handle, bh);
784 		if (err) {
785 			/* Don't brelse(bh) here; it's done in
786 			 * ext4_journal_forget() below */
787 			unlock_buffer(bh);
788 			goto failed;
789 		}
790 
791 		memset(bh->b_data, 0, blocksize);
792 		branch[n].p = (__le32 *) bh->b_data + offsets[n];
793 		branch[n].key = cpu_to_le32(new_blocks[n]);
794 		*branch[n].p = branch[n].key;
795 		if (n == indirect_blks) {
796 			current_block = new_blocks[n];
797 			/*
798 			 * End of chain, update the last new metablock of
799 			 * the chain to point to the new allocated
800 			 * data blocks numbers
801 			 */
802 			for (i = 1; i < num; i++)
803 				*(branch[n].p + i) = cpu_to_le32(++current_block);
804 		}
805 		BUFFER_TRACE(bh, "marking uptodate");
806 		set_buffer_uptodate(bh);
807 		unlock_buffer(bh);
808 
809 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
810 		err = ext4_handle_dirty_metadata(handle, inode, bh);
811 		if (err)
812 			goto failed;
813 	}
814 	*blks = num;
815 	return err;
816 failed:
817 	/* Allocation failed, free what we already allocated */
818 	for (i = 1; i <= n ; i++) {
819 		BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
820 		ext4_journal_forget(handle, branch[i].bh);
821 	}
822 	for (i = 0; i < indirect_blks; i++)
823 		ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
824 
825 	ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
826 
827 	return err;
828 }
829 
830 /**
831  * ext4_splice_branch - splice the allocated branch onto inode.
832  * @inode: owner
833  * @block: (logical) number of block we are adding
834  * @chain: chain of indirect blocks (with a missing link - see
835  *	ext4_alloc_branch)
836  * @where: location of missing link
837  * @num:   number of indirect blocks we are adding
838  * @blks:  number of direct blocks we are adding
839  *
840  * This function fills the missing link and does all housekeeping needed in
841  * inode (->i_blocks, etc.). In case of success we end up with the full
842  * chain to new block and return 0.
843  */
844 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
845 			      ext4_lblk_t block, Indirect *where, int num,
846 			      int blks)
847 {
848 	int i;
849 	int err = 0;
850 	ext4_fsblk_t current_block;
851 
852 	/*
853 	 * If we're splicing into a [td]indirect block (as opposed to the
854 	 * inode) then we need to get write access to the [td]indirect block
855 	 * before the splice.
856 	 */
857 	if (where->bh) {
858 		BUFFER_TRACE(where->bh, "get_write_access");
859 		err = ext4_journal_get_write_access(handle, where->bh);
860 		if (err)
861 			goto err_out;
862 	}
863 	/* That's it */
864 
865 	*where->p = where->key;
866 
867 	/*
868 	 * Update the host buffer_head or inode to point to more just allocated
869 	 * direct blocks blocks
870 	 */
871 	if (num == 0 && blks > 1) {
872 		current_block = le32_to_cpu(where->key) + 1;
873 		for (i = 1; i < blks; i++)
874 			*(where->p + i) = cpu_to_le32(current_block++);
875 	}
876 
877 	/* We are done with atomic stuff, now do the rest of housekeeping */
878 	/* had we spliced it onto indirect block? */
879 	if (where->bh) {
880 		/*
881 		 * If we spliced it onto an indirect block, we haven't
882 		 * altered the inode.  Note however that if it is being spliced
883 		 * onto an indirect block at the very end of the file (the
884 		 * file is growing) then we *will* alter the inode to reflect
885 		 * the new i_size.  But that is not done here - it is done in
886 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
887 		 */
888 		jbd_debug(5, "splicing indirect only\n");
889 		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
890 		err = ext4_handle_dirty_metadata(handle, inode, where->bh);
891 		if (err)
892 			goto err_out;
893 	} else {
894 		/*
895 		 * OK, we spliced it into the inode itself on a direct block.
896 		 */
897 		ext4_mark_inode_dirty(handle, inode);
898 		jbd_debug(5, "splicing direct\n");
899 	}
900 	return err;
901 
902 err_out:
903 	for (i = 1; i <= num; i++) {
904 		BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
905 		ext4_journal_forget(handle, where[i].bh);
906 		ext4_free_blocks(handle, inode,
907 					le32_to_cpu(where[i-1].key), 1, 0);
908 	}
909 	ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
910 
911 	return err;
912 }
913 
914 /*
915  * The ext4_ind_get_blocks() function handles non-extents inodes
916  * (i.e., using the traditional indirect/double-indirect i_blocks
917  * scheme) for ext4_get_blocks().
918  *
919  * Allocation strategy is simple: if we have to allocate something, we will
920  * have to go the whole way to leaf. So let's do it before attaching anything
921  * to tree, set linkage between the newborn blocks, write them if sync is
922  * required, recheck the path, free and repeat if check fails, otherwise
923  * set the last missing link (that will protect us from any truncate-generated
924  * removals - all blocks on the path are immune now) and possibly force the
925  * write on the parent block.
926  * That has a nice additional property: no special recovery from the failed
927  * allocations is needed - we simply release blocks and do not touch anything
928  * reachable from inode.
929  *
930  * `handle' can be NULL if create == 0.
931  *
932  * return > 0, # of blocks mapped or allocated.
933  * return = 0, if plain lookup failed.
934  * return < 0, error case.
935  *
936  * The ext4_ind_get_blocks() function should be called with
937  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
938  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
939  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
940  * blocks.
941  */
942 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
943 			       ext4_lblk_t iblock, unsigned int maxblocks,
944 			       struct buffer_head *bh_result,
945 			       int flags)
946 {
947 	int err = -EIO;
948 	ext4_lblk_t offsets[4];
949 	Indirect chain[4];
950 	Indirect *partial;
951 	ext4_fsblk_t goal;
952 	int indirect_blks;
953 	int blocks_to_boundary = 0;
954 	int depth;
955 	int count = 0;
956 	ext4_fsblk_t first_block = 0;
957 
958 	J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
959 	J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
960 	depth = ext4_block_to_path(inode, iblock, offsets,
961 				   &blocks_to_boundary);
962 
963 	if (depth == 0)
964 		goto out;
965 
966 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
967 
968 	/* Simplest case - block found, no allocation needed */
969 	if (!partial) {
970 		first_block = le32_to_cpu(chain[depth - 1].key);
971 		clear_buffer_new(bh_result);
972 		count++;
973 		/*map more blocks*/
974 		while (count < maxblocks && count <= blocks_to_boundary) {
975 			ext4_fsblk_t blk;
976 
977 			blk = le32_to_cpu(*(chain[depth-1].p + count));
978 
979 			if (blk == first_block + count)
980 				count++;
981 			else
982 				break;
983 		}
984 		goto got_it;
985 	}
986 
987 	/* Next simple case - plain lookup or failed read of indirect block */
988 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
989 		goto cleanup;
990 
991 	/*
992 	 * Okay, we need to do block allocation.
993 	*/
994 	goal = ext4_find_goal(inode, iblock, partial);
995 
996 	/* the number of blocks need to allocate for [d,t]indirect blocks */
997 	indirect_blks = (chain + depth) - partial - 1;
998 
999 	/*
1000 	 * Next look up the indirect map to count the totoal number of
1001 	 * direct blocks to allocate for this branch.
1002 	 */
1003 	count = ext4_blks_to_allocate(partial, indirect_blks,
1004 					maxblocks, blocks_to_boundary);
1005 	/*
1006 	 * Block out ext4_truncate while we alter the tree
1007 	 */
1008 	err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
1009 				&count, goal,
1010 				offsets + (partial - chain), partial);
1011 
1012 	/*
1013 	 * The ext4_splice_branch call will free and forget any buffers
1014 	 * on the new chain if there is a failure, but that risks using
1015 	 * up transaction credits, especially for bitmaps where the
1016 	 * credits cannot be returned.  Can we handle this somehow?  We
1017 	 * may need to return -EAGAIN upwards in the worst case.  --sct
1018 	 */
1019 	if (!err)
1020 		err = ext4_splice_branch(handle, inode, iblock,
1021 					 partial, indirect_blks, count);
1022 	else
1023 		goto cleanup;
1024 
1025 	set_buffer_new(bh_result);
1026 got_it:
1027 	map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1028 	if (count > blocks_to_boundary)
1029 		set_buffer_boundary(bh_result);
1030 	err = count;
1031 	/* Clean up and exit */
1032 	partial = chain + depth - 1;	/* the whole chain */
1033 cleanup:
1034 	while (partial > chain) {
1035 		BUFFER_TRACE(partial->bh, "call brelse");
1036 		brelse(partial->bh);
1037 		partial--;
1038 	}
1039 	BUFFER_TRACE(bh_result, "returned");
1040 out:
1041 	return err;
1042 }
1043 
1044 qsize_t ext4_get_reserved_space(struct inode *inode)
1045 {
1046 	unsigned long long total;
1047 
1048 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1049 	total = EXT4_I(inode)->i_reserved_data_blocks +
1050 		EXT4_I(inode)->i_reserved_meta_blocks;
1051 	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1052 
1053 	return total;
1054 }
1055 /*
1056  * Calculate the number of metadata blocks need to reserve
1057  * to allocate @blocks for non extent file based file
1058  */
1059 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1060 {
1061 	int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1062 	int ind_blks, dind_blks, tind_blks;
1063 
1064 	/* number of new indirect blocks needed */
1065 	ind_blks = (blocks + icap - 1) / icap;
1066 
1067 	dind_blks = (ind_blks + icap - 1) / icap;
1068 
1069 	tind_blks = 1;
1070 
1071 	return ind_blks + dind_blks + tind_blks;
1072 }
1073 
1074 /*
1075  * Calculate the number of metadata blocks need to reserve
1076  * to allocate given number of blocks
1077  */
1078 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1079 {
1080 	if (!blocks)
1081 		return 0;
1082 
1083 	if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1084 		return ext4_ext_calc_metadata_amount(inode, blocks);
1085 
1086 	return ext4_indirect_calc_metadata_amount(inode, blocks);
1087 }
1088 
1089 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1090 {
1091 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1092 	int total, mdb, mdb_free;
1093 
1094 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1095 	/* recalculate the number of metablocks still need to be reserved */
1096 	total = EXT4_I(inode)->i_reserved_data_blocks - used;
1097 	mdb = ext4_calc_metadata_amount(inode, total);
1098 
1099 	/* figure out how many metablocks to release */
1100 	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1101 	mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1102 
1103 	if (mdb_free) {
1104 		/* Account for allocated meta_blocks */
1105 		mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1106 
1107 		/* update fs dirty blocks counter */
1108 		percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1109 		EXT4_I(inode)->i_allocated_meta_blocks = 0;
1110 		EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1111 	}
1112 
1113 	/* update per-inode reservations */
1114 	BUG_ON(used  > EXT4_I(inode)->i_reserved_data_blocks);
1115 	EXT4_I(inode)->i_reserved_data_blocks -= used;
1116 	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1117 
1118 	/*
1119 	 * free those over-booking quota for metadata blocks
1120 	 */
1121 	if (mdb_free)
1122 		vfs_dq_release_reservation_block(inode, mdb_free);
1123 
1124 	/*
1125 	 * If we have done all the pending block allocations and if
1126 	 * there aren't any writers on the inode, we can discard the
1127 	 * inode's preallocations.
1128 	 */
1129 	if (!total && (atomic_read(&inode->i_writecount) == 0))
1130 		ext4_discard_preallocations(inode);
1131 }
1132 
1133 static int check_block_validity(struct inode *inode, const char *msg,
1134 				sector_t logical, sector_t phys, int len)
1135 {
1136 	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
1137 		ext4_error(inode->i_sb, msg,
1138 			   "inode #%lu logical block %llu mapped to %llu "
1139 			   "(size %d)", inode->i_ino,
1140 			   (unsigned long long) logical,
1141 			   (unsigned long long) phys, len);
1142 		return -EIO;
1143 	}
1144 	return 0;
1145 }
1146 
1147 /*
1148  * The ext4_get_blocks() function tries to look up the requested blocks,
1149  * and returns if the blocks are already mapped.
1150  *
1151  * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1152  * and store the allocated blocks in the result buffer head and mark it
1153  * mapped.
1154  *
1155  * If file type is extents based, it will call ext4_ext_get_blocks(),
1156  * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1157  * based files
1158  *
1159  * On success, it returns the number of blocks being mapped or allocate.
1160  * if create==0 and the blocks are pre-allocated and uninitialized block,
1161  * the result buffer head is unmapped. If the create ==1, it will make sure
1162  * the buffer head is mapped.
1163  *
1164  * It returns 0 if plain look up failed (blocks have not been allocated), in
1165  * that casem, buffer head is unmapped
1166  *
1167  * It returns the error in case of allocation failure.
1168  */
1169 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1170 		    unsigned int max_blocks, struct buffer_head *bh,
1171 		    int flags)
1172 {
1173 	int retval;
1174 
1175 	clear_buffer_mapped(bh);
1176 	clear_buffer_unwritten(bh);
1177 
1178 	/*
1179 	 * Try to see if we can get the block without requesting a new
1180 	 * file system block.
1181 	 */
1182 	down_read((&EXT4_I(inode)->i_data_sem));
1183 	if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1184 		retval =  ext4_ext_get_blocks(handle, inode, block, max_blocks,
1185 				bh, 0);
1186 	} else {
1187 		retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1188 					     bh, 0);
1189 	}
1190 	up_read((&EXT4_I(inode)->i_data_sem));
1191 
1192 	if (retval > 0 && buffer_mapped(bh)) {
1193 		int ret = check_block_validity(inode, "file system corruption",
1194 					       block, bh->b_blocknr, retval);
1195 		if (ret != 0)
1196 			return ret;
1197 	}
1198 
1199 	/* If it is only a block(s) look up */
1200 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1201 		return retval;
1202 
1203 	/*
1204 	 * Returns if the blocks have already allocated
1205 	 *
1206 	 * Note that if blocks have been preallocated
1207 	 * ext4_ext_get_block() returns th create = 0
1208 	 * with buffer head unmapped.
1209 	 */
1210 	if (retval > 0 && buffer_mapped(bh))
1211 		return retval;
1212 
1213 	/*
1214 	 * When we call get_blocks without the create flag, the
1215 	 * BH_Unwritten flag could have gotten set if the blocks
1216 	 * requested were part of a uninitialized extent.  We need to
1217 	 * clear this flag now that we are committed to convert all or
1218 	 * part of the uninitialized extent to be an initialized
1219 	 * extent.  This is because we need to avoid the combination
1220 	 * of BH_Unwritten and BH_Mapped flags being simultaneously
1221 	 * set on the buffer_head.
1222 	 */
1223 	clear_buffer_unwritten(bh);
1224 
1225 	/*
1226 	 * New blocks allocate and/or writing to uninitialized extent
1227 	 * will possibly result in updating i_data, so we take
1228 	 * the write lock of i_data_sem, and call get_blocks()
1229 	 * with create == 1 flag.
1230 	 */
1231 	down_write((&EXT4_I(inode)->i_data_sem));
1232 
1233 	/*
1234 	 * if the caller is from delayed allocation writeout path
1235 	 * we have already reserved fs blocks for allocation
1236 	 * let the underlying get_block() function know to
1237 	 * avoid double accounting
1238 	 */
1239 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1240 		EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1241 	/*
1242 	 * We need to check for EXT4 here because migrate
1243 	 * could have changed the inode type in between
1244 	 */
1245 	if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1246 		retval =  ext4_ext_get_blocks(handle, inode, block, max_blocks,
1247 					      bh, flags);
1248 	} else {
1249 		retval = ext4_ind_get_blocks(handle, inode, block,
1250 					     max_blocks, bh, flags);
1251 
1252 		if (retval > 0 && buffer_new(bh)) {
1253 			/*
1254 			 * We allocated new blocks which will result in
1255 			 * i_data's format changing.  Force the migrate
1256 			 * to fail by clearing migrate flags
1257 			 */
1258 			EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE;
1259 		}
1260 	}
1261 
1262 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1263 		EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1264 
1265 	/*
1266 	 * Update reserved blocks/metadata blocks after successful
1267 	 * block allocation which had been deferred till now.
1268 	 */
1269 	if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1270 		ext4_da_update_reserve_space(inode, retval);
1271 
1272 	up_write((&EXT4_I(inode)->i_data_sem));
1273 	if (retval > 0 && buffer_mapped(bh)) {
1274 		int ret = check_block_validity(inode, "file system "
1275 					       "corruption after allocation",
1276 					       block, bh->b_blocknr, retval);
1277 		if (ret != 0)
1278 			return ret;
1279 	}
1280 	return retval;
1281 }
1282 
1283 /* Maximum number of blocks we map for direct IO at once. */
1284 #define DIO_MAX_BLOCKS 4096
1285 
1286 int ext4_get_block(struct inode *inode, sector_t iblock,
1287 		   struct buffer_head *bh_result, int create)
1288 {
1289 	handle_t *handle = ext4_journal_current_handle();
1290 	int ret = 0, started = 0;
1291 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1292 	int dio_credits;
1293 
1294 	if (create && !handle) {
1295 		/* Direct IO write... */
1296 		if (max_blocks > DIO_MAX_BLOCKS)
1297 			max_blocks = DIO_MAX_BLOCKS;
1298 		dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1299 		handle = ext4_journal_start(inode, dio_credits);
1300 		if (IS_ERR(handle)) {
1301 			ret = PTR_ERR(handle);
1302 			goto out;
1303 		}
1304 		started = 1;
1305 	}
1306 
1307 	ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1308 			      create ? EXT4_GET_BLOCKS_CREATE : 0);
1309 	if (ret > 0) {
1310 		bh_result->b_size = (ret << inode->i_blkbits);
1311 		ret = 0;
1312 	}
1313 	if (started)
1314 		ext4_journal_stop(handle);
1315 out:
1316 	return ret;
1317 }
1318 
1319 /*
1320  * `handle' can be NULL if create is zero
1321  */
1322 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1323 				ext4_lblk_t block, int create, int *errp)
1324 {
1325 	struct buffer_head dummy;
1326 	int fatal = 0, err;
1327 	int flags = 0;
1328 
1329 	J_ASSERT(handle != NULL || create == 0);
1330 
1331 	dummy.b_state = 0;
1332 	dummy.b_blocknr = -1000;
1333 	buffer_trace_init(&dummy.b_history);
1334 	if (create)
1335 		flags |= EXT4_GET_BLOCKS_CREATE;
1336 	err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1337 	/*
1338 	 * ext4_get_blocks() returns number of blocks mapped. 0 in
1339 	 * case of a HOLE.
1340 	 */
1341 	if (err > 0) {
1342 		if (err > 1)
1343 			WARN_ON(1);
1344 		err = 0;
1345 	}
1346 	*errp = err;
1347 	if (!err && buffer_mapped(&dummy)) {
1348 		struct buffer_head *bh;
1349 		bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1350 		if (!bh) {
1351 			*errp = -EIO;
1352 			goto err;
1353 		}
1354 		if (buffer_new(&dummy)) {
1355 			J_ASSERT(create != 0);
1356 			J_ASSERT(handle != NULL);
1357 
1358 			/*
1359 			 * Now that we do not always journal data, we should
1360 			 * keep in mind whether this should always journal the
1361 			 * new buffer as metadata.  For now, regular file
1362 			 * writes use ext4_get_block instead, so it's not a
1363 			 * problem.
1364 			 */
1365 			lock_buffer(bh);
1366 			BUFFER_TRACE(bh, "call get_create_access");
1367 			fatal = ext4_journal_get_create_access(handle, bh);
1368 			if (!fatal && !buffer_uptodate(bh)) {
1369 				memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1370 				set_buffer_uptodate(bh);
1371 			}
1372 			unlock_buffer(bh);
1373 			BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1374 			err = ext4_handle_dirty_metadata(handle, inode, bh);
1375 			if (!fatal)
1376 				fatal = err;
1377 		} else {
1378 			BUFFER_TRACE(bh, "not a new buffer");
1379 		}
1380 		if (fatal) {
1381 			*errp = fatal;
1382 			brelse(bh);
1383 			bh = NULL;
1384 		}
1385 		return bh;
1386 	}
1387 err:
1388 	return NULL;
1389 }
1390 
1391 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1392 			       ext4_lblk_t block, int create, int *err)
1393 {
1394 	struct buffer_head *bh;
1395 
1396 	bh = ext4_getblk(handle, inode, block, create, err);
1397 	if (!bh)
1398 		return bh;
1399 	if (buffer_uptodate(bh))
1400 		return bh;
1401 	ll_rw_block(READ_META, 1, &bh);
1402 	wait_on_buffer(bh);
1403 	if (buffer_uptodate(bh))
1404 		return bh;
1405 	put_bh(bh);
1406 	*err = -EIO;
1407 	return NULL;
1408 }
1409 
1410 static int walk_page_buffers(handle_t *handle,
1411 			     struct buffer_head *head,
1412 			     unsigned from,
1413 			     unsigned to,
1414 			     int *partial,
1415 			     int (*fn)(handle_t *handle,
1416 				       struct buffer_head *bh))
1417 {
1418 	struct buffer_head *bh;
1419 	unsigned block_start, block_end;
1420 	unsigned blocksize = head->b_size;
1421 	int err, ret = 0;
1422 	struct buffer_head *next;
1423 
1424 	for (bh = head, block_start = 0;
1425 	     ret == 0 && (bh != head || !block_start);
1426 	     block_start = block_end, bh = next) {
1427 		next = bh->b_this_page;
1428 		block_end = block_start + blocksize;
1429 		if (block_end <= from || block_start >= to) {
1430 			if (partial && !buffer_uptodate(bh))
1431 				*partial = 1;
1432 			continue;
1433 		}
1434 		err = (*fn)(handle, bh);
1435 		if (!ret)
1436 			ret = err;
1437 	}
1438 	return ret;
1439 }
1440 
1441 /*
1442  * To preserve ordering, it is essential that the hole instantiation and
1443  * the data write be encapsulated in a single transaction.  We cannot
1444  * close off a transaction and start a new one between the ext4_get_block()
1445  * and the commit_write().  So doing the jbd2_journal_start at the start of
1446  * prepare_write() is the right place.
1447  *
1448  * Also, this function can nest inside ext4_writepage() ->
1449  * block_write_full_page(). In that case, we *know* that ext4_writepage()
1450  * has generated enough buffer credits to do the whole page.  So we won't
1451  * block on the journal in that case, which is good, because the caller may
1452  * be PF_MEMALLOC.
1453  *
1454  * By accident, ext4 can be reentered when a transaction is open via
1455  * quota file writes.  If we were to commit the transaction while thus
1456  * reentered, there can be a deadlock - we would be holding a quota
1457  * lock, and the commit would never complete if another thread had a
1458  * transaction open and was blocking on the quota lock - a ranking
1459  * violation.
1460  *
1461  * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1462  * will _not_ run commit under these circumstances because handle->h_ref
1463  * is elevated.  We'll still have enough credits for the tiny quotafile
1464  * write.
1465  */
1466 static int do_journal_get_write_access(handle_t *handle,
1467 				       struct buffer_head *bh)
1468 {
1469 	if (!buffer_mapped(bh) || buffer_freed(bh))
1470 		return 0;
1471 	return ext4_journal_get_write_access(handle, bh);
1472 }
1473 
1474 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1475 			    loff_t pos, unsigned len, unsigned flags,
1476 			    struct page **pagep, void **fsdata)
1477 {
1478 	struct inode *inode = mapping->host;
1479 	int ret, needed_blocks;
1480 	handle_t *handle;
1481 	int retries = 0;
1482 	struct page *page;
1483 	pgoff_t index;
1484 	unsigned from, to;
1485 
1486 	trace_ext4_write_begin(inode, pos, len, flags);
1487 	/*
1488 	 * Reserve one block more for addition to orphan list in case
1489 	 * we allocate blocks but write fails for some reason
1490 	 */
1491 	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1492 	index = pos >> PAGE_CACHE_SHIFT;
1493 	from = pos & (PAGE_CACHE_SIZE - 1);
1494 	to = from + len;
1495 
1496 retry:
1497 	handle = ext4_journal_start(inode, needed_blocks);
1498 	if (IS_ERR(handle)) {
1499 		ret = PTR_ERR(handle);
1500 		goto out;
1501 	}
1502 
1503 	/* We cannot recurse into the filesystem as the transaction is already
1504 	 * started */
1505 	flags |= AOP_FLAG_NOFS;
1506 
1507 	page = grab_cache_page_write_begin(mapping, index, flags);
1508 	if (!page) {
1509 		ext4_journal_stop(handle);
1510 		ret = -ENOMEM;
1511 		goto out;
1512 	}
1513 	*pagep = page;
1514 
1515 	ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1516 				ext4_get_block);
1517 
1518 	if (!ret && ext4_should_journal_data(inode)) {
1519 		ret = walk_page_buffers(handle, page_buffers(page),
1520 				from, to, NULL, do_journal_get_write_access);
1521 	}
1522 
1523 	if (ret) {
1524 		unlock_page(page);
1525 		page_cache_release(page);
1526 		/*
1527 		 * block_write_begin may have instantiated a few blocks
1528 		 * outside i_size.  Trim these off again. Don't need
1529 		 * i_size_read because we hold i_mutex.
1530 		 *
1531 		 * Add inode to orphan list in case we crash before
1532 		 * truncate finishes
1533 		 */
1534 		if (pos + len > inode->i_size && ext4_can_truncate(inode))
1535 			ext4_orphan_add(handle, inode);
1536 
1537 		ext4_journal_stop(handle);
1538 		if (pos + len > inode->i_size) {
1539 			ext4_truncate(inode);
1540 			/*
1541 			 * If truncate failed early the inode might
1542 			 * still be on the orphan list; we need to
1543 			 * make sure the inode is removed from the
1544 			 * orphan list in that case.
1545 			 */
1546 			if (inode->i_nlink)
1547 				ext4_orphan_del(NULL, inode);
1548 		}
1549 	}
1550 
1551 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1552 		goto retry;
1553 out:
1554 	return ret;
1555 }
1556 
1557 /* For write_end() in data=journal mode */
1558 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1559 {
1560 	if (!buffer_mapped(bh) || buffer_freed(bh))
1561 		return 0;
1562 	set_buffer_uptodate(bh);
1563 	return ext4_handle_dirty_metadata(handle, NULL, bh);
1564 }
1565 
1566 static int ext4_generic_write_end(struct file *file,
1567 				  struct address_space *mapping,
1568 				  loff_t pos, unsigned len, unsigned copied,
1569 				  struct page *page, void *fsdata)
1570 {
1571 	int i_size_changed = 0;
1572 	struct inode *inode = mapping->host;
1573 	handle_t *handle = ext4_journal_current_handle();
1574 
1575 	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1576 
1577 	/*
1578 	 * No need to use i_size_read() here, the i_size
1579 	 * cannot change under us because we hold i_mutex.
1580 	 *
1581 	 * But it's important to update i_size while still holding page lock:
1582 	 * page writeout could otherwise come in and zero beyond i_size.
1583 	 */
1584 	if (pos + copied > inode->i_size) {
1585 		i_size_write(inode, pos + copied);
1586 		i_size_changed = 1;
1587 	}
1588 
1589 	if (pos + copied >  EXT4_I(inode)->i_disksize) {
1590 		/* We need to mark inode dirty even if
1591 		 * new_i_size is less that inode->i_size
1592 		 * bu greater than i_disksize.(hint delalloc)
1593 		 */
1594 		ext4_update_i_disksize(inode, (pos + copied));
1595 		i_size_changed = 1;
1596 	}
1597 	unlock_page(page);
1598 	page_cache_release(page);
1599 
1600 	/*
1601 	 * Don't mark the inode dirty under page lock. First, it unnecessarily
1602 	 * makes the holding time of page lock longer. Second, it forces lock
1603 	 * ordering of page lock and transaction start for journaling
1604 	 * filesystems.
1605 	 */
1606 	if (i_size_changed)
1607 		ext4_mark_inode_dirty(handle, inode);
1608 
1609 	return copied;
1610 }
1611 
1612 /*
1613  * We need to pick up the new inode size which generic_commit_write gave us
1614  * `file' can be NULL - eg, when called from page_symlink().
1615  *
1616  * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1617  * buffers are managed internally.
1618  */
1619 static int ext4_ordered_write_end(struct file *file,
1620 				  struct address_space *mapping,
1621 				  loff_t pos, unsigned len, unsigned copied,
1622 				  struct page *page, void *fsdata)
1623 {
1624 	handle_t *handle = ext4_journal_current_handle();
1625 	struct inode *inode = mapping->host;
1626 	int ret = 0, ret2;
1627 
1628 	trace_ext4_ordered_write_end(inode, pos, len, copied);
1629 	ret = ext4_jbd2_file_inode(handle, inode);
1630 
1631 	if (ret == 0) {
1632 		ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1633 							page, fsdata);
1634 		copied = ret2;
1635 		if (pos + len > inode->i_size && ext4_can_truncate(inode))
1636 			/* if we have allocated more blocks and copied
1637 			 * less. We will have blocks allocated outside
1638 			 * inode->i_size. So truncate them
1639 			 */
1640 			ext4_orphan_add(handle, inode);
1641 		if (ret2 < 0)
1642 			ret = ret2;
1643 	}
1644 	ret2 = ext4_journal_stop(handle);
1645 	if (!ret)
1646 		ret = ret2;
1647 
1648 	if (pos + len > inode->i_size) {
1649 		ext4_truncate(inode);
1650 		/*
1651 		 * If truncate failed early the inode might still be
1652 		 * on the orphan list; we need to make sure the inode
1653 		 * is removed from the orphan list in that case.
1654 		 */
1655 		if (inode->i_nlink)
1656 			ext4_orphan_del(NULL, inode);
1657 	}
1658 
1659 
1660 	return ret ? ret : copied;
1661 }
1662 
1663 static int ext4_writeback_write_end(struct file *file,
1664 				    struct address_space *mapping,
1665 				    loff_t pos, unsigned len, unsigned copied,
1666 				    struct page *page, void *fsdata)
1667 {
1668 	handle_t *handle = ext4_journal_current_handle();
1669 	struct inode *inode = mapping->host;
1670 	int ret = 0, ret2;
1671 
1672 	trace_ext4_writeback_write_end(inode, pos, len, copied);
1673 	ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1674 							page, fsdata);
1675 	copied = ret2;
1676 	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1677 		/* if we have allocated more blocks and copied
1678 		 * less. We will have blocks allocated outside
1679 		 * inode->i_size. So truncate them
1680 		 */
1681 		ext4_orphan_add(handle, inode);
1682 
1683 	if (ret2 < 0)
1684 		ret = ret2;
1685 
1686 	ret2 = ext4_journal_stop(handle);
1687 	if (!ret)
1688 		ret = ret2;
1689 
1690 	if (pos + len > inode->i_size) {
1691 		ext4_truncate(inode);
1692 		/*
1693 		 * If truncate failed early the inode might still be
1694 		 * on the orphan list; we need to make sure the inode
1695 		 * is removed from the orphan list in that case.
1696 		 */
1697 		if (inode->i_nlink)
1698 			ext4_orphan_del(NULL, inode);
1699 	}
1700 
1701 	return ret ? ret : copied;
1702 }
1703 
1704 static int ext4_journalled_write_end(struct file *file,
1705 				     struct address_space *mapping,
1706 				     loff_t pos, unsigned len, unsigned copied,
1707 				     struct page *page, void *fsdata)
1708 {
1709 	handle_t *handle = ext4_journal_current_handle();
1710 	struct inode *inode = mapping->host;
1711 	int ret = 0, ret2;
1712 	int partial = 0;
1713 	unsigned from, to;
1714 	loff_t new_i_size;
1715 
1716 	trace_ext4_journalled_write_end(inode, pos, len, copied);
1717 	from = pos & (PAGE_CACHE_SIZE - 1);
1718 	to = from + len;
1719 
1720 	if (copied < len) {
1721 		if (!PageUptodate(page))
1722 			copied = 0;
1723 		page_zero_new_buffers(page, from+copied, to);
1724 	}
1725 
1726 	ret = walk_page_buffers(handle, page_buffers(page), from,
1727 				to, &partial, write_end_fn);
1728 	if (!partial)
1729 		SetPageUptodate(page);
1730 	new_i_size = pos + copied;
1731 	if (new_i_size > inode->i_size)
1732 		i_size_write(inode, pos+copied);
1733 	EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1734 	if (new_i_size > EXT4_I(inode)->i_disksize) {
1735 		ext4_update_i_disksize(inode, new_i_size);
1736 		ret2 = ext4_mark_inode_dirty(handle, inode);
1737 		if (!ret)
1738 			ret = ret2;
1739 	}
1740 
1741 	unlock_page(page);
1742 	page_cache_release(page);
1743 	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1744 		/* if we have allocated more blocks and copied
1745 		 * less. We will have blocks allocated outside
1746 		 * inode->i_size. So truncate them
1747 		 */
1748 		ext4_orphan_add(handle, inode);
1749 
1750 	ret2 = ext4_journal_stop(handle);
1751 	if (!ret)
1752 		ret = ret2;
1753 	if (pos + len > inode->i_size) {
1754 		ext4_truncate(inode);
1755 		/*
1756 		 * If truncate failed early the inode might still be
1757 		 * on the orphan list; we need to make sure the inode
1758 		 * is removed from the orphan list in that case.
1759 		 */
1760 		if (inode->i_nlink)
1761 			ext4_orphan_del(NULL, inode);
1762 	}
1763 
1764 	return ret ? ret : copied;
1765 }
1766 
1767 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1768 {
1769 	int retries = 0;
1770 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1771 	unsigned long md_needed, mdblocks, total = 0;
1772 
1773 	/*
1774 	 * recalculate the amount of metadata blocks to reserve
1775 	 * in order to allocate nrblocks
1776 	 * worse case is one extent per block
1777 	 */
1778 repeat:
1779 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1780 	total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1781 	mdblocks = ext4_calc_metadata_amount(inode, total);
1782 	BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1783 
1784 	md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1785 	total = md_needed + nrblocks;
1786 
1787 	/*
1788 	 * Make quota reservation here to prevent quota overflow
1789 	 * later. Real quota accounting is done at pages writeout
1790 	 * time.
1791 	 */
1792 	if (vfs_dq_reserve_block(inode, total)) {
1793 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1794 		return -EDQUOT;
1795 	}
1796 
1797 	if (ext4_claim_free_blocks(sbi, total)) {
1798 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1799 		if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1800 			yield();
1801 			goto repeat;
1802 		}
1803 		vfs_dq_release_reservation_block(inode, total);
1804 		return -ENOSPC;
1805 	}
1806 	EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1807 	EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1808 
1809 	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1810 	return 0;       /* success */
1811 }
1812 
1813 static void ext4_da_release_space(struct inode *inode, int to_free)
1814 {
1815 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1816 	int total, mdb, mdb_free, release;
1817 
1818 	if (!to_free)
1819 		return;		/* Nothing to release, exit */
1820 
1821 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1822 
1823 	if (!EXT4_I(inode)->i_reserved_data_blocks) {
1824 		/*
1825 		 * if there is no reserved blocks, but we try to free some
1826 		 * then the counter is messed up somewhere.
1827 		 * but since this function is called from invalidate
1828 		 * page, it's harmless to return without any action
1829 		 */
1830 		printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1831 			    "blocks for inode %lu, but there is no reserved "
1832 			    "data blocks\n", to_free, inode->i_ino);
1833 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1834 		return;
1835 	}
1836 
1837 	/* recalculate the number of metablocks still need to be reserved */
1838 	total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1839 	mdb = ext4_calc_metadata_amount(inode, total);
1840 
1841 	/* figure out how many metablocks to release */
1842 	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1843 	mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1844 
1845 	release = to_free + mdb_free;
1846 
1847 	/* update fs dirty blocks counter for truncate case */
1848 	percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1849 
1850 	/* update per-inode reservations */
1851 	BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1852 	EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1853 
1854 	BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1855 	EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1856 	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1857 
1858 	vfs_dq_release_reservation_block(inode, release);
1859 }
1860 
1861 static void ext4_da_page_release_reservation(struct page *page,
1862 					     unsigned long offset)
1863 {
1864 	int to_release = 0;
1865 	struct buffer_head *head, *bh;
1866 	unsigned int curr_off = 0;
1867 
1868 	head = page_buffers(page);
1869 	bh = head;
1870 	do {
1871 		unsigned int next_off = curr_off + bh->b_size;
1872 
1873 		if ((offset <= curr_off) && (buffer_delay(bh))) {
1874 			to_release++;
1875 			clear_buffer_delay(bh);
1876 		}
1877 		curr_off = next_off;
1878 	} while ((bh = bh->b_this_page) != head);
1879 	ext4_da_release_space(page->mapping->host, to_release);
1880 }
1881 
1882 /*
1883  * Delayed allocation stuff
1884  */
1885 
1886 /*
1887  * mpage_da_submit_io - walks through extent of pages and try to write
1888  * them with writepage() call back
1889  *
1890  * @mpd->inode: inode
1891  * @mpd->first_page: first page of the extent
1892  * @mpd->next_page: page after the last page of the extent
1893  *
1894  * By the time mpage_da_submit_io() is called we expect all blocks
1895  * to be allocated. this may be wrong if allocation failed.
1896  *
1897  * As pages are already locked by write_cache_pages(), we can't use it
1898  */
1899 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1900 {
1901 	long pages_skipped;
1902 	struct pagevec pvec;
1903 	unsigned long index, end;
1904 	int ret = 0, err, nr_pages, i;
1905 	struct inode *inode = mpd->inode;
1906 	struct address_space *mapping = inode->i_mapping;
1907 
1908 	BUG_ON(mpd->next_page <= mpd->first_page);
1909 	/*
1910 	 * We need to start from the first_page to the next_page - 1
1911 	 * to make sure we also write the mapped dirty buffer_heads.
1912 	 * If we look at mpd->b_blocknr we would only be looking
1913 	 * at the currently mapped buffer_heads.
1914 	 */
1915 	index = mpd->first_page;
1916 	end = mpd->next_page - 1;
1917 
1918 	pagevec_init(&pvec, 0);
1919 	while (index <= end) {
1920 		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1921 		if (nr_pages == 0)
1922 			break;
1923 		for (i = 0; i < nr_pages; i++) {
1924 			struct page *page = pvec.pages[i];
1925 
1926 			index = page->index;
1927 			if (index > end)
1928 				break;
1929 			index++;
1930 
1931 			BUG_ON(!PageLocked(page));
1932 			BUG_ON(PageWriteback(page));
1933 
1934 			pages_skipped = mpd->wbc->pages_skipped;
1935 			err = mapping->a_ops->writepage(page, mpd->wbc);
1936 			if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1937 				/*
1938 				 * have successfully written the page
1939 				 * without skipping the same
1940 				 */
1941 				mpd->pages_written++;
1942 			/*
1943 			 * In error case, we have to continue because
1944 			 * remaining pages are still locked
1945 			 * XXX: unlock and re-dirty them?
1946 			 */
1947 			if (ret == 0)
1948 				ret = err;
1949 		}
1950 		pagevec_release(&pvec);
1951 	}
1952 	return ret;
1953 }
1954 
1955 /*
1956  * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1957  *
1958  * @mpd->inode - inode to walk through
1959  * @exbh->b_blocknr - first block on a disk
1960  * @exbh->b_size - amount of space in bytes
1961  * @logical - first logical block to start assignment with
1962  *
1963  * the function goes through all passed space and put actual disk
1964  * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1965  */
1966 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1967 				 struct buffer_head *exbh)
1968 {
1969 	struct inode *inode = mpd->inode;
1970 	struct address_space *mapping = inode->i_mapping;
1971 	int blocks = exbh->b_size >> inode->i_blkbits;
1972 	sector_t pblock = exbh->b_blocknr, cur_logical;
1973 	struct buffer_head *head, *bh;
1974 	pgoff_t index, end;
1975 	struct pagevec pvec;
1976 	int nr_pages, i;
1977 
1978 	index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1979 	end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1980 	cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1981 
1982 	pagevec_init(&pvec, 0);
1983 
1984 	while (index <= end) {
1985 		/* XXX: optimize tail */
1986 		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1987 		if (nr_pages == 0)
1988 			break;
1989 		for (i = 0; i < nr_pages; i++) {
1990 			struct page *page = pvec.pages[i];
1991 
1992 			index = page->index;
1993 			if (index > end)
1994 				break;
1995 			index++;
1996 
1997 			BUG_ON(!PageLocked(page));
1998 			BUG_ON(PageWriteback(page));
1999 			BUG_ON(!page_has_buffers(page));
2000 
2001 			bh = page_buffers(page);
2002 			head = bh;
2003 
2004 			/* skip blocks out of the range */
2005 			do {
2006 				if (cur_logical >= logical)
2007 					break;
2008 				cur_logical++;
2009 			} while ((bh = bh->b_this_page) != head);
2010 
2011 			do {
2012 				if (cur_logical >= logical + blocks)
2013 					break;
2014 
2015 				if (buffer_delay(bh) ||
2016 						buffer_unwritten(bh)) {
2017 
2018 					BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2019 
2020 					if (buffer_delay(bh)) {
2021 						clear_buffer_delay(bh);
2022 						bh->b_blocknr = pblock;
2023 					} else {
2024 						/*
2025 						 * unwritten already should have
2026 						 * blocknr assigned. Verify that
2027 						 */
2028 						clear_buffer_unwritten(bh);
2029 						BUG_ON(bh->b_blocknr != pblock);
2030 					}
2031 
2032 				} else if (buffer_mapped(bh))
2033 					BUG_ON(bh->b_blocknr != pblock);
2034 
2035 				cur_logical++;
2036 				pblock++;
2037 			} while ((bh = bh->b_this_page) != head);
2038 		}
2039 		pagevec_release(&pvec);
2040 	}
2041 }
2042 
2043 
2044 /*
2045  * __unmap_underlying_blocks - just a helper function to unmap
2046  * set of blocks described by @bh
2047  */
2048 static inline void __unmap_underlying_blocks(struct inode *inode,
2049 					     struct buffer_head *bh)
2050 {
2051 	struct block_device *bdev = inode->i_sb->s_bdev;
2052 	int blocks, i;
2053 
2054 	blocks = bh->b_size >> inode->i_blkbits;
2055 	for (i = 0; i < blocks; i++)
2056 		unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2057 }
2058 
2059 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2060 					sector_t logical, long blk_cnt)
2061 {
2062 	int nr_pages, i;
2063 	pgoff_t index, end;
2064 	struct pagevec pvec;
2065 	struct inode *inode = mpd->inode;
2066 	struct address_space *mapping = inode->i_mapping;
2067 
2068 	index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2069 	end   = (logical + blk_cnt - 1) >>
2070 				(PAGE_CACHE_SHIFT - inode->i_blkbits);
2071 	while (index <= end) {
2072 		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2073 		if (nr_pages == 0)
2074 			break;
2075 		for (i = 0; i < nr_pages; i++) {
2076 			struct page *page = pvec.pages[i];
2077 			index = page->index;
2078 			if (index > end)
2079 				break;
2080 			index++;
2081 
2082 			BUG_ON(!PageLocked(page));
2083 			BUG_ON(PageWriteback(page));
2084 			block_invalidatepage(page, 0);
2085 			ClearPageUptodate(page);
2086 			unlock_page(page);
2087 		}
2088 	}
2089 	return;
2090 }
2091 
2092 static void ext4_print_free_blocks(struct inode *inode)
2093 {
2094 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2095 	printk(KERN_EMERG "Total free blocks count %lld\n",
2096 			ext4_count_free_blocks(inode->i_sb));
2097 	printk(KERN_EMERG "Free/Dirty block details\n");
2098 	printk(KERN_EMERG "free_blocks=%lld\n",
2099 			(long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
2100 	printk(KERN_EMERG "dirty_blocks=%lld\n",
2101 			(long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2102 	printk(KERN_EMERG "Block reservation details\n");
2103 	printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
2104 			EXT4_I(inode)->i_reserved_data_blocks);
2105 	printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
2106 			EXT4_I(inode)->i_reserved_meta_blocks);
2107 	return;
2108 }
2109 
2110 /*
2111  * mpage_da_map_blocks - go through given space
2112  *
2113  * @mpd - bh describing space
2114  *
2115  * The function skips space we know is already mapped to disk blocks.
2116  *
2117  */
2118 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2119 {
2120 	int err, blks, get_blocks_flags;
2121 	struct buffer_head new;
2122 	sector_t next = mpd->b_blocknr;
2123 	unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2124 	loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2125 	handle_t *handle = NULL;
2126 
2127 	/*
2128 	 * We consider only non-mapped and non-allocated blocks
2129 	 */
2130 	if ((mpd->b_state  & (1 << BH_Mapped)) &&
2131 		!(mpd->b_state & (1 << BH_Delay)) &&
2132 		!(mpd->b_state & (1 << BH_Unwritten)))
2133 		return 0;
2134 
2135 	/*
2136 	 * If we didn't accumulate anything to write simply return
2137 	 */
2138 	if (!mpd->b_size)
2139 		return 0;
2140 
2141 	handle = ext4_journal_current_handle();
2142 	BUG_ON(!handle);
2143 
2144 	/*
2145 	 * Call ext4_get_blocks() to allocate any delayed allocation
2146 	 * blocks, or to convert an uninitialized extent to be
2147 	 * initialized (in the case where we have written into
2148 	 * one or more preallocated blocks).
2149 	 *
2150 	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2151 	 * indicate that we are on the delayed allocation path.  This
2152 	 * affects functions in many different parts of the allocation
2153 	 * call path.  This flag exists primarily because we don't
2154 	 * want to change *many* call functions, so ext4_get_blocks()
2155 	 * will set the magic i_delalloc_reserved_flag once the
2156 	 * inode's allocation semaphore is taken.
2157 	 *
2158 	 * If the blocks in questions were delalloc blocks, set
2159 	 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2160 	 * variables are updated after the blocks have been allocated.
2161 	 */
2162 	new.b_state = 0;
2163 	get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2164 			    EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2165 	if (mpd->b_state & (1 << BH_Delay))
2166 		get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
2167 	blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2168 			       &new, get_blocks_flags);
2169 	if (blks < 0) {
2170 		err = blks;
2171 		/*
2172 		 * If get block returns with error we simply
2173 		 * return. Later writepage will redirty the page and
2174 		 * writepages will find the dirty page again
2175 		 */
2176 		if (err == -EAGAIN)
2177 			return 0;
2178 
2179 		if (err == -ENOSPC &&
2180 		    ext4_count_free_blocks(mpd->inode->i_sb)) {
2181 			mpd->retval = err;
2182 			return 0;
2183 		}
2184 
2185 		/*
2186 		 * get block failure will cause us to loop in
2187 		 * writepages, because a_ops->writepage won't be able
2188 		 * to make progress. The page will be redirtied by
2189 		 * writepage and writepages will again try to write
2190 		 * the same.
2191 		 */
2192 		printk(KERN_EMERG "%s block allocation failed for inode %lu "
2193 				  "at logical offset %llu with max blocks "
2194 				  "%zd with error %d\n",
2195 				  __func__, mpd->inode->i_ino,
2196 				  (unsigned long long)next,
2197 				  mpd->b_size >> mpd->inode->i_blkbits, err);
2198 		printk(KERN_EMERG "This should not happen.!! "
2199 					"Data will be lost\n");
2200 		if (err == -ENOSPC) {
2201 			ext4_print_free_blocks(mpd->inode);
2202 		}
2203 		/* invalidate all the pages */
2204 		ext4_da_block_invalidatepages(mpd, next,
2205 				mpd->b_size >> mpd->inode->i_blkbits);
2206 		return err;
2207 	}
2208 	BUG_ON(blks == 0);
2209 
2210 	new.b_size = (blks << mpd->inode->i_blkbits);
2211 
2212 	if (buffer_new(&new))
2213 		__unmap_underlying_blocks(mpd->inode, &new);
2214 
2215 	/*
2216 	 * If blocks are delayed marked, we need to
2217 	 * put actual blocknr and drop delayed bit
2218 	 */
2219 	if ((mpd->b_state & (1 << BH_Delay)) ||
2220 	    (mpd->b_state & (1 << BH_Unwritten)))
2221 		mpage_put_bnr_to_bhs(mpd, next, &new);
2222 
2223 	if (ext4_should_order_data(mpd->inode)) {
2224 		err = ext4_jbd2_file_inode(handle, mpd->inode);
2225 		if (err)
2226 			return err;
2227 	}
2228 
2229 	/*
2230 	 * Update on-disk size along with block allocation.
2231 	 */
2232 	disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2233 	if (disksize > i_size_read(mpd->inode))
2234 		disksize = i_size_read(mpd->inode);
2235 	if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2236 		ext4_update_i_disksize(mpd->inode, disksize);
2237 		return ext4_mark_inode_dirty(handle, mpd->inode);
2238 	}
2239 
2240 	return 0;
2241 }
2242 
2243 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2244 		(1 << BH_Delay) | (1 << BH_Unwritten))
2245 
2246 /*
2247  * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2248  *
2249  * @mpd->lbh - extent of blocks
2250  * @logical - logical number of the block in the file
2251  * @bh - bh of the block (used to access block's state)
2252  *
2253  * the function is used to collect contig. blocks in same state
2254  */
2255 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2256 				   sector_t logical, size_t b_size,
2257 				   unsigned long b_state)
2258 {
2259 	sector_t next;
2260 	int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2261 
2262 	/* check if thereserved journal credits might overflow */
2263 	if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2264 		if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2265 			/*
2266 			 * With non-extent format we are limited by the journal
2267 			 * credit available.  Total credit needed to insert
2268 			 * nrblocks contiguous blocks is dependent on the
2269 			 * nrblocks.  So limit nrblocks.
2270 			 */
2271 			goto flush_it;
2272 		} else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2273 				EXT4_MAX_TRANS_DATA) {
2274 			/*
2275 			 * Adding the new buffer_head would make it cross the
2276 			 * allowed limit for which we have journal credit
2277 			 * reserved. So limit the new bh->b_size
2278 			 */
2279 			b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2280 						mpd->inode->i_blkbits;
2281 			/* we will do mpage_da_submit_io in the next loop */
2282 		}
2283 	}
2284 	/*
2285 	 * First block in the extent
2286 	 */
2287 	if (mpd->b_size == 0) {
2288 		mpd->b_blocknr = logical;
2289 		mpd->b_size = b_size;
2290 		mpd->b_state = b_state & BH_FLAGS;
2291 		return;
2292 	}
2293 
2294 	next = mpd->b_blocknr + nrblocks;
2295 	/*
2296 	 * Can we merge the block to our big extent?
2297 	 */
2298 	if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2299 		mpd->b_size += b_size;
2300 		return;
2301 	}
2302 
2303 flush_it:
2304 	/*
2305 	 * We couldn't merge the block to our extent, so we
2306 	 * need to flush current  extent and start new one
2307 	 */
2308 	if (mpage_da_map_blocks(mpd) == 0)
2309 		mpage_da_submit_io(mpd);
2310 	mpd->io_done = 1;
2311 	return;
2312 }
2313 
2314 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2315 {
2316 	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2317 }
2318 
2319 /*
2320  * __mpage_da_writepage - finds extent of pages and blocks
2321  *
2322  * @page: page to consider
2323  * @wbc: not used, we just follow rules
2324  * @data: context
2325  *
2326  * The function finds extents of pages and scan them for all blocks.
2327  */
2328 static int __mpage_da_writepage(struct page *page,
2329 				struct writeback_control *wbc, void *data)
2330 {
2331 	struct mpage_da_data *mpd = data;
2332 	struct inode *inode = mpd->inode;
2333 	struct buffer_head *bh, *head;
2334 	sector_t logical;
2335 
2336 	if (mpd->io_done) {
2337 		/*
2338 		 * Rest of the page in the page_vec
2339 		 * redirty then and skip then. We will
2340 		 * try to write them again after
2341 		 * starting a new transaction
2342 		 */
2343 		redirty_page_for_writepage(wbc, page);
2344 		unlock_page(page);
2345 		return MPAGE_DA_EXTENT_TAIL;
2346 	}
2347 	/*
2348 	 * Can we merge this page to current extent?
2349 	 */
2350 	if (mpd->next_page != page->index) {
2351 		/*
2352 		 * Nope, we can't. So, we map non-allocated blocks
2353 		 * and start IO on them using writepage()
2354 		 */
2355 		if (mpd->next_page != mpd->first_page) {
2356 			if (mpage_da_map_blocks(mpd) == 0)
2357 				mpage_da_submit_io(mpd);
2358 			/*
2359 			 * skip rest of the page in the page_vec
2360 			 */
2361 			mpd->io_done = 1;
2362 			redirty_page_for_writepage(wbc, page);
2363 			unlock_page(page);
2364 			return MPAGE_DA_EXTENT_TAIL;
2365 		}
2366 
2367 		/*
2368 		 * Start next extent of pages ...
2369 		 */
2370 		mpd->first_page = page->index;
2371 
2372 		/*
2373 		 * ... and blocks
2374 		 */
2375 		mpd->b_size = 0;
2376 		mpd->b_state = 0;
2377 		mpd->b_blocknr = 0;
2378 	}
2379 
2380 	mpd->next_page = page->index + 1;
2381 	logical = (sector_t) page->index <<
2382 		  (PAGE_CACHE_SHIFT - inode->i_blkbits);
2383 
2384 	if (!page_has_buffers(page)) {
2385 		mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2386 				       (1 << BH_Dirty) | (1 << BH_Uptodate));
2387 		if (mpd->io_done)
2388 			return MPAGE_DA_EXTENT_TAIL;
2389 	} else {
2390 		/*
2391 		 * Page with regular buffer heads, just add all dirty ones
2392 		 */
2393 		head = page_buffers(page);
2394 		bh = head;
2395 		do {
2396 			BUG_ON(buffer_locked(bh));
2397 			/*
2398 			 * We need to try to allocate
2399 			 * unmapped blocks in the same page.
2400 			 * Otherwise we won't make progress
2401 			 * with the page in ext4_writepage
2402 			 */
2403 			if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2404 				mpage_add_bh_to_extent(mpd, logical,
2405 						       bh->b_size,
2406 						       bh->b_state);
2407 				if (mpd->io_done)
2408 					return MPAGE_DA_EXTENT_TAIL;
2409 			} else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2410 				/*
2411 				 * mapped dirty buffer. We need to update
2412 				 * the b_state because we look at
2413 				 * b_state in mpage_da_map_blocks. We don't
2414 				 * update b_size because if we find an
2415 				 * unmapped buffer_head later we need to
2416 				 * use the b_state flag of that buffer_head.
2417 				 */
2418 				if (mpd->b_size == 0)
2419 					mpd->b_state = bh->b_state & BH_FLAGS;
2420 			}
2421 			logical++;
2422 		} while ((bh = bh->b_this_page) != head);
2423 	}
2424 
2425 	return 0;
2426 }
2427 
2428 /*
2429  * This is a special get_blocks_t callback which is used by
2430  * ext4_da_write_begin().  It will either return mapped block or
2431  * reserve space for a single block.
2432  *
2433  * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2434  * We also have b_blocknr = -1 and b_bdev initialized properly
2435  *
2436  * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2437  * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2438  * initialized properly.
2439  */
2440 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2441 				  struct buffer_head *bh_result, int create)
2442 {
2443 	int ret = 0;
2444 	sector_t invalid_block = ~((sector_t) 0xffff);
2445 
2446 	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2447 		invalid_block = ~0;
2448 
2449 	BUG_ON(create == 0);
2450 	BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2451 
2452 	/*
2453 	 * first, we need to know whether the block is allocated already
2454 	 * preallocated blocks are unmapped but should treated
2455 	 * the same as allocated blocks.
2456 	 */
2457 	ret = ext4_get_blocks(NULL, inode, iblock, 1,  bh_result, 0);
2458 	if ((ret == 0) && !buffer_delay(bh_result)) {
2459 		/* the block isn't (pre)allocated yet, let's reserve space */
2460 		/*
2461 		 * XXX: __block_prepare_write() unmaps passed block,
2462 		 * is it OK?
2463 		 */
2464 		ret = ext4_da_reserve_space(inode, 1);
2465 		if (ret)
2466 			/* not enough space to reserve */
2467 			return ret;
2468 
2469 		map_bh(bh_result, inode->i_sb, invalid_block);
2470 		set_buffer_new(bh_result);
2471 		set_buffer_delay(bh_result);
2472 	} else if (ret > 0) {
2473 		bh_result->b_size = (ret << inode->i_blkbits);
2474 		if (buffer_unwritten(bh_result)) {
2475 			/* A delayed write to unwritten bh should
2476 			 * be marked new and mapped.  Mapped ensures
2477 			 * that we don't do get_block multiple times
2478 			 * when we write to the same offset and new
2479 			 * ensures that we do proper zero out for
2480 			 * partial write.
2481 			 */
2482 			set_buffer_new(bh_result);
2483 			set_buffer_mapped(bh_result);
2484 		}
2485 		ret = 0;
2486 	}
2487 
2488 	return ret;
2489 }
2490 
2491 /*
2492  * This function is used as a standard get_block_t calback function
2493  * when there is no desire to allocate any blocks.  It is used as a
2494  * callback function for block_prepare_write(), nobh_writepage(), and
2495  * block_write_full_page().  These functions should only try to map a
2496  * single block at a time.
2497  *
2498  * Since this function doesn't do block allocations even if the caller
2499  * requests it by passing in create=1, it is critically important that
2500  * any caller checks to make sure that any buffer heads are returned
2501  * by this function are either all already mapped or marked for
2502  * delayed allocation before calling nobh_writepage() or
2503  * block_write_full_page().  Otherwise, b_blocknr could be left
2504  * unitialized, and the page write functions will be taken by
2505  * surprise.
2506  */
2507 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2508 				   struct buffer_head *bh_result, int create)
2509 {
2510 	int ret = 0;
2511 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2512 
2513 	BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2514 
2515 	/*
2516 	 * we don't want to do block allocation in writepage
2517 	 * so call get_block_wrap with create = 0
2518 	 */
2519 	ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2520 	if (ret > 0) {
2521 		bh_result->b_size = (ret << inode->i_blkbits);
2522 		ret = 0;
2523 	}
2524 	return ret;
2525 }
2526 
2527 static int bget_one(handle_t *handle, struct buffer_head *bh)
2528 {
2529 	get_bh(bh);
2530 	return 0;
2531 }
2532 
2533 static int bput_one(handle_t *handle, struct buffer_head *bh)
2534 {
2535 	put_bh(bh);
2536 	return 0;
2537 }
2538 
2539 static int __ext4_journalled_writepage(struct page *page,
2540 				       struct writeback_control *wbc,
2541 				       unsigned int len)
2542 {
2543 	struct address_space *mapping = page->mapping;
2544 	struct inode *inode = mapping->host;
2545 	struct buffer_head *page_bufs;
2546 	handle_t *handle = NULL;
2547 	int ret = 0;
2548 	int err;
2549 
2550 	page_bufs = page_buffers(page);
2551 	BUG_ON(!page_bufs);
2552 	walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2553 	/* As soon as we unlock the page, it can go away, but we have
2554 	 * references to buffers so we are safe */
2555 	unlock_page(page);
2556 
2557 	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2558 	if (IS_ERR(handle)) {
2559 		ret = PTR_ERR(handle);
2560 		goto out;
2561 	}
2562 
2563 	ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2564 				do_journal_get_write_access);
2565 
2566 	err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2567 				write_end_fn);
2568 	if (ret == 0)
2569 		ret = err;
2570 	err = ext4_journal_stop(handle);
2571 	if (!ret)
2572 		ret = err;
2573 
2574 	walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2575 	EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2576 out:
2577 	return ret;
2578 }
2579 
2580 /*
2581  * Note that we don't need to start a transaction unless we're journaling data
2582  * because we should have holes filled from ext4_page_mkwrite(). We even don't
2583  * need to file the inode to the transaction's list in ordered mode because if
2584  * we are writing back data added by write(), the inode is already there and if
2585  * we are writing back data modified via mmap(), noone guarantees in which
2586  * transaction the data will hit the disk. In case we are journaling data, we
2587  * cannot start transaction directly because transaction start ranks above page
2588  * lock so we have to do some magic.
2589  *
2590  * This function can get called via...
2591  *   - ext4_da_writepages after taking page lock (have journal handle)
2592  *   - journal_submit_inode_data_buffers (no journal handle)
2593  *   - shrink_page_list via pdflush (no journal handle)
2594  *   - grab_page_cache when doing write_begin (have journal handle)
2595  *
2596  * We don't do any block allocation in this function. If we have page with
2597  * multiple blocks we need to write those buffer_heads that are mapped. This
2598  * is important for mmaped based write. So if we do with blocksize 1K
2599  * truncate(f, 1024);
2600  * a = mmap(f, 0, 4096);
2601  * a[0] = 'a';
2602  * truncate(f, 4096);
2603  * we have in the page first buffer_head mapped via page_mkwrite call back
2604  * but other bufer_heads would be unmapped but dirty(dirty done via the
2605  * do_wp_page). So writepage should write the first block. If we modify
2606  * the mmap area beyond 1024 we will again get a page_fault and the
2607  * page_mkwrite callback will do the block allocation and mark the
2608  * buffer_heads mapped.
2609  *
2610  * We redirty the page if we have any buffer_heads that is either delay or
2611  * unwritten in the page.
2612  *
2613  * We can get recursively called as show below.
2614  *
2615  *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2616  *		ext4_writepage()
2617  *
2618  * But since we don't do any block allocation we should not deadlock.
2619  * Page also have the dirty flag cleared so we don't get recurive page_lock.
2620  */
2621 static int ext4_writepage(struct page *page,
2622 			  struct writeback_control *wbc)
2623 {
2624 	int ret = 0;
2625 	loff_t size;
2626 	unsigned int len;
2627 	struct buffer_head *page_bufs;
2628 	struct inode *inode = page->mapping->host;
2629 
2630 	trace_ext4_writepage(inode, page);
2631 	size = i_size_read(inode);
2632 	if (page->index == size >> PAGE_CACHE_SHIFT)
2633 		len = size & ~PAGE_CACHE_MASK;
2634 	else
2635 		len = PAGE_CACHE_SIZE;
2636 
2637 	if (page_has_buffers(page)) {
2638 		page_bufs = page_buffers(page);
2639 		if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2640 					ext4_bh_delay_or_unwritten)) {
2641 			/*
2642 			 * We don't want to do  block allocation
2643 			 * So redirty the page and return
2644 			 * We may reach here when we do a journal commit
2645 			 * via journal_submit_inode_data_buffers.
2646 			 * If we don't have mapping block we just ignore
2647 			 * them. We can also reach here via shrink_page_list
2648 			 */
2649 			redirty_page_for_writepage(wbc, page);
2650 			unlock_page(page);
2651 			return 0;
2652 		}
2653 	} else {
2654 		/*
2655 		 * The test for page_has_buffers() is subtle:
2656 		 * We know the page is dirty but it lost buffers. That means
2657 		 * that at some moment in time after write_begin()/write_end()
2658 		 * has been called all buffers have been clean and thus they
2659 		 * must have been written at least once. So they are all
2660 		 * mapped and we can happily proceed with mapping them
2661 		 * and writing the page.
2662 		 *
2663 		 * Try to initialize the buffer_heads and check whether
2664 		 * all are mapped and non delay. We don't want to
2665 		 * do block allocation here.
2666 		 */
2667 		ret = block_prepare_write(page, 0, len,
2668 					  noalloc_get_block_write);
2669 		if (!ret) {
2670 			page_bufs = page_buffers(page);
2671 			/* check whether all are mapped and non delay */
2672 			if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2673 						ext4_bh_delay_or_unwritten)) {
2674 				redirty_page_for_writepage(wbc, page);
2675 				unlock_page(page);
2676 				return 0;
2677 			}
2678 		} else {
2679 			/*
2680 			 * We can't do block allocation here
2681 			 * so just redity the page and unlock
2682 			 * and return
2683 			 */
2684 			redirty_page_for_writepage(wbc, page);
2685 			unlock_page(page);
2686 			return 0;
2687 		}
2688 		/* now mark the buffer_heads as dirty and uptodate */
2689 		block_commit_write(page, 0, len);
2690 	}
2691 
2692 	if (PageChecked(page) && ext4_should_journal_data(inode)) {
2693 		/*
2694 		 * It's mmapped pagecache.  Add buffers and journal it.  There
2695 		 * doesn't seem much point in redirtying the page here.
2696 		 */
2697 		ClearPageChecked(page);
2698 		return __ext4_journalled_writepage(page, wbc, len);
2699 	}
2700 
2701 	if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2702 		ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2703 	else
2704 		ret = block_write_full_page(page, noalloc_get_block_write,
2705 					    wbc);
2706 
2707 	return ret;
2708 }
2709 
2710 /*
2711  * This is called via ext4_da_writepages() to
2712  * calulate the total number of credits to reserve to fit
2713  * a single extent allocation into a single transaction,
2714  * ext4_da_writpeages() will loop calling this before
2715  * the block allocation.
2716  */
2717 
2718 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2719 {
2720 	int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2721 
2722 	/*
2723 	 * With non-extent format the journal credit needed to
2724 	 * insert nrblocks contiguous block is dependent on
2725 	 * number of contiguous block. So we will limit
2726 	 * number of contiguous block to a sane value
2727 	 */
2728 	if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2729 	    (max_blocks > EXT4_MAX_TRANS_DATA))
2730 		max_blocks = EXT4_MAX_TRANS_DATA;
2731 
2732 	return ext4_chunk_trans_blocks(inode, max_blocks);
2733 }
2734 
2735 static int ext4_da_writepages(struct address_space *mapping,
2736 			      struct writeback_control *wbc)
2737 {
2738 	pgoff_t	index;
2739 	int range_whole = 0;
2740 	handle_t *handle = NULL;
2741 	struct mpage_da_data mpd;
2742 	struct inode *inode = mapping->host;
2743 	int no_nrwrite_index_update;
2744 	int pages_written = 0;
2745 	long pages_skipped;
2746 	int range_cyclic, cycled = 1, io_done = 0;
2747 	int needed_blocks, ret = 0, nr_to_writebump = 0;
2748 	loff_t range_start = wbc->range_start;
2749 	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2750 
2751 	trace_ext4_da_writepages(inode, wbc);
2752 
2753 	/*
2754 	 * No pages to write? This is mainly a kludge to avoid starting
2755 	 * a transaction for special inodes like journal inode on last iput()
2756 	 * because that could violate lock ordering on umount
2757 	 */
2758 	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2759 		return 0;
2760 
2761 	/*
2762 	 * If the filesystem has aborted, it is read-only, so return
2763 	 * right away instead of dumping stack traces later on that
2764 	 * will obscure the real source of the problem.  We test
2765 	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2766 	 * the latter could be true if the filesystem is mounted
2767 	 * read-only, and in that case, ext4_da_writepages should
2768 	 * *never* be called, so if that ever happens, we would want
2769 	 * the stack trace.
2770 	 */
2771 	if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2772 		return -EROFS;
2773 
2774 	/*
2775 	 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2776 	 * This make sure small files blocks are allocated in
2777 	 * single attempt. This ensure that small files
2778 	 * get less fragmented.
2779 	 */
2780 	if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2781 		nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2782 		wbc->nr_to_write = sbi->s_mb_stream_request;
2783 	}
2784 	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2785 		range_whole = 1;
2786 
2787 	range_cyclic = wbc->range_cyclic;
2788 	if (wbc->range_cyclic) {
2789 		index = mapping->writeback_index;
2790 		if (index)
2791 			cycled = 0;
2792 		wbc->range_start = index << PAGE_CACHE_SHIFT;
2793 		wbc->range_end  = LLONG_MAX;
2794 		wbc->range_cyclic = 0;
2795 	} else
2796 		index = wbc->range_start >> PAGE_CACHE_SHIFT;
2797 
2798 	mpd.wbc = wbc;
2799 	mpd.inode = mapping->host;
2800 
2801 	/*
2802 	 * we don't want write_cache_pages to update
2803 	 * nr_to_write and writeback_index
2804 	 */
2805 	no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2806 	wbc->no_nrwrite_index_update = 1;
2807 	pages_skipped = wbc->pages_skipped;
2808 
2809 retry:
2810 	while (!ret && wbc->nr_to_write > 0) {
2811 
2812 		/*
2813 		 * we  insert one extent at a time. So we need
2814 		 * credit needed for single extent allocation.
2815 		 * journalled mode is currently not supported
2816 		 * by delalloc
2817 		 */
2818 		BUG_ON(ext4_should_journal_data(inode));
2819 		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2820 
2821 		/* start a new transaction*/
2822 		handle = ext4_journal_start(inode, needed_blocks);
2823 		if (IS_ERR(handle)) {
2824 			ret = PTR_ERR(handle);
2825 			printk(KERN_CRIT "%s: jbd2_start: "
2826 			       "%ld pages, ino %lu; err %d\n", __func__,
2827 				wbc->nr_to_write, inode->i_ino, ret);
2828 			dump_stack();
2829 			goto out_writepages;
2830 		}
2831 
2832 		/*
2833 		 * Now call __mpage_da_writepage to find the next
2834 		 * contiguous region of logical blocks that need
2835 		 * blocks to be allocated by ext4.  We don't actually
2836 		 * submit the blocks for I/O here, even though
2837 		 * write_cache_pages thinks it will, and will set the
2838 		 * pages as clean for write before calling
2839 		 * __mpage_da_writepage().
2840 		 */
2841 		mpd.b_size = 0;
2842 		mpd.b_state = 0;
2843 		mpd.b_blocknr = 0;
2844 		mpd.first_page = 0;
2845 		mpd.next_page = 0;
2846 		mpd.io_done = 0;
2847 		mpd.pages_written = 0;
2848 		mpd.retval = 0;
2849 		ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2850 					&mpd);
2851 		/*
2852 		 * If we have a contigous extent of pages and we
2853 		 * haven't done the I/O yet, map the blocks and submit
2854 		 * them for I/O.
2855 		 */
2856 		if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2857 			if (mpage_da_map_blocks(&mpd) == 0)
2858 				mpage_da_submit_io(&mpd);
2859 			mpd.io_done = 1;
2860 			ret = MPAGE_DA_EXTENT_TAIL;
2861 		}
2862 		trace_ext4_da_write_pages(inode, &mpd);
2863 		wbc->nr_to_write -= mpd.pages_written;
2864 
2865 		ext4_journal_stop(handle);
2866 
2867 		if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2868 			/* commit the transaction which would
2869 			 * free blocks released in the transaction
2870 			 * and try again
2871 			 */
2872 			jbd2_journal_force_commit_nested(sbi->s_journal);
2873 			wbc->pages_skipped = pages_skipped;
2874 			ret = 0;
2875 		} else if (ret == MPAGE_DA_EXTENT_TAIL) {
2876 			/*
2877 			 * got one extent now try with
2878 			 * rest of the pages
2879 			 */
2880 			pages_written += mpd.pages_written;
2881 			wbc->pages_skipped = pages_skipped;
2882 			ret = 0;
2883 			io_done = 1;
2884 		} else if (wbc->nr_to_write)
2885 			/*
2886 			 * There is no more writeout needed
2887 			 * or we requested for a noblocking writeout
2888 			 * and we found the device congested
2889 			 */
2890 			break;
2891 	}
2892 	if (!io_done && !cycled) {
2893 		cycled = 1;
2894 		index = 0;
2895 		wbc->range_start = index << PAGE_CACHE_SHIFT;
2896 		wbc->range_end  = mapping->writeback_index - 1;
2897 		goto retry;
2898 	}
2899 	if (pages_skipped != wbc->pages_skipped)
2900 		printk(KERN_EMERG "This should not happen leaving %s "
2901 				"with nr_to_write = %ld ret = %d\n",
2902 				__func__, wbc->nr_to_write, ret);
2903 
2904 	/* Update index */
2905 	index += pages_written;
2906 	wbc->range_cyclic = range_cyclic;
2907 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2908 		/*
2909 		 * set the writeback_index so that range_cyclic
2910 		 * mode will write it back later
2911 		 */
2912 		mapping->writeback_index = index;
2913 
2914 out_writepages:
2915 	if (!no_nrwrite_index_update)
2916 		wbc->no_nrwrite_index_update = 0;
2917 	wbc->nr_to_write -= nr_to_writebump;
2918 	wbc->range_start = range_start;
2919 	trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2920 	return ret;
2921 }
2922 
2923 #define FALL_BACK_TO_NONDELALLOC 1
2924 static int ext4_nonda_switch(struct super_block *sb)
2925 {
2926 	s64 free_blocks, dirty_blocks;
2927 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2928 
2929 	/*
2930 	 * switch to non delalloc mode if we are running low
2931 	 * on free block. The free block accounting via percpu
2932 	 * counters can get slightly wrong with percpu_counter_batch getting
2933 	 * accumulated on each CPU without updating global counters
2934 	 * Delalloc need an accurate free block accounting. So switch
2935 	 * to non delalloc when we are near to error range.
2936 	 */
2937 	free_blocks  = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2938 	dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2939 	if (2 * free_blocks < 3 * dirty_blocks ||
2940 		free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2941 		/*
2942 		 * free block count is less that 150% of dirty blocks
2943 		 * or free blocks is less that watermark
2944 		 */
2945 		return 1;
2946 	}
2947 	return 0;
2948 }
2949 
2950 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2951 			       loff_t pos, unsigned len, unsigned flags,
2952 			       struct page **pagep, void **fsdata)
2953 {
2954 	int ret, retries = 0;
2955 	struct page *page;
2956 	pgoff_t index;
2957 	unsigned from, to;
2958 	struct inode *inode = mapping->host;
2959 	handle_t *handle;
2960 
2961 	index = pos >> PAGE_CACHE_SHIFT;
2962 	from = pos & (PAGE_CACHE_SIZE - 1);
2963 	to = from + len;
2964 
2965 	if (ext4_nonda_switch(inode->i_sb)) {
2966 		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2967 		return ext4_write_begin(file, mapping, pos,
2968 					len, flags, pagep, fsdata);
2969 	}
2970 	*fsdata = (void *)0;
2971 	trace_ext4_da_write_begin(inode, pos, len, flags);
2972 retry:
2973 	/*
2974 	 * With delayed allocation, we don't log the i_disksize update
2975 	 * if there is delayed block allocation. But we still need
2976 	 * to journalling the i_disksize update if writes to the end
2977 	 * of file which has an already mapped buffer.
2978 	 */
2979 	handle = ext4_journal_start(inode, 1);
2980 	if (IS_ERR(handle)) {
2981 		ret = PTR_ERR(handle);
2982 		goto out;
2983 	}
2984 	/* We cannot recurse into the filesystem as the transaction is already
2985 	 * started */
2986 	flags |= AOP_FLAG_NOFS;
2987 
2988 	page = grab_cache_page_write_begin(mapping, index, flags);
2989 	if (!page) {
2990 		ext4_journal_stop(handle);
2991 		ret = -ENOMEM;
2992 		goto out;
2993 	}
2994 	*pagep = page;
2995 
2996 	ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2997 				ext4_da_get_block_prep);
2998 	if (ret < 0) {
2999 		unlock_page(page);
3000 		ext4_journal_stop(handle);
3001 		page_cache_release(page);
3002 		/*
3003 		 * block_write_begin may have instantiated a few blocks
3004 		 * outside i_size.  Trim these off again. Don't need
3005 		 * i_size_read because we hold i_mutex.
3006 		 */
3007 		if (pos + len > inode->i_size)
3008 			ext4_truncate(inode);
3009 	}
3010 
3011 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3012 		goto retry;
3013 out:
3014 	return ret;
3015 }
3016 
3017 /*
3018  * Check if we should update i_disksize
3019  * when write to the end of file but not require block allocation
3020  */
3021 static int ext4_da_should_update_i_disksize(struct page *page,
3022 					    unsigned long offset)
3023 {
3024 	struct buffer_head *bh;
3025 	struct inode *inode = page->mapping->host;
3026 	unsigned int idx;
3027 	int i;
3028 
3029 	bh = page_buffers(page);
3030 	idx = offset >> inode->i_blkbits;
3031 
3032 	for (i = 0; i < idx; i++)
3033 		bh = bh->b_this_page;
3034 
3035 	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3036 		return 0;
3037 	return 1;
3038 }
3039 
3040 static int ext4_da_write_end(struct file *file,
3041 			     struct address_space *mapping,
3042 			     loff_t pos, unsigned len, unsigned copied,
3043 			     struct page *page, void *fsdata)
3044 {
3045 	struct inode *inode = mapping->host;
3046 	int ret = 0, ret2;
3047 	handle_t *handle = ext4_journal_current_handle();
3048 	loff_t new_i_size;
3049 	unsigned long start, end;
3050 	int write_mode = (int)(unsigned long)fsdata;
3051 
3052 	if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3053 		if (ext4_should_order_data(inode)) {
3054 			return ext4_ordered_write_end(file, mapping, pos,
3055 					len, copied, page, fsdata);
3056 		} else if (ext4_should_writeback_data(inode)) {
3057 			return ext4_writeback_write_end(file, mapping, pos,
3058 					len, copied, page, fsdata);
3059 		} else {
3060 			BUG();
3061 		}
3062 	}
3063 
3064 	trace_ext4_da_write_end(inode, pos, len, copied);
3065 	start = pos & (PAGE_CACHE_SIZE - 1);
3066 	end = start + copied - 1;
3067 
3068 	/*
3069 	 * generic_write_end() will run mark_inode_dirty() if i_size
3070 	 * changes.  So let's piggyback the i_disksize mark_inode_dirty
3071 	 * into that.
3072 	 */
3073 
3074 	new_i_size = pos + copied;
3075 	if (new_i_size > EXT4_I(inode)->i_disksize) {
3076 		if (ext4_da_should_update_i_disksize(page, end)) {
3077 			down_write(&EXT4_I(inode)->i_data_sem);
3078 			if (new_i_size > EXT4_I(inode)->i_disksize) {
3079 				/*
3080 				 * Updating i_disksize when extending file
3081 				 * without needing block allocation
3082 				 */
3083 				if (ext4_should_order_data(inode))
3084 					ret = ext4_jbd2_file_inode(handle,
3085 								   inode);
3086 
3087 				EXT4_I(inode)->i_disksize = new_i_size;
3088 			}
3089 			up_write(&EXT4_I(inode)->i_data_sem);
3090 			/* We need to mark inode dirty even if
3091 			 * new_i_size is less that inode->i_size
3092 			 * bu greater than i_disksize.(hint delalloc)
3093 			 */
3094 			ext4_mark_inode_dirty(handle, inode);
3095 		}
3096 	}
3097 	ret2 = generic_write_end(file, mapping, pos, len, copied,
3098 							page, fsdata);
3099 	copied = ret2;
3100 	if (ret2 < 0)
3101 		ret = ret2;
3102 	ret2 = ext4_journal_stop(handle);
3103 	if (!ret)
3104 		ret = ret2;
3105 
3106 	return ret ? ret : copied;
3107 }
3108 
3109 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3110 {
3111 	/*
3112 	 * Drop reserved blocks
3113 	 */
3114 	BUG_ON(!PageLocked(page));
3115 	if (!page_has_buffers(page))
3116 		goto out;
3117 
3118 	ext4_da_page_release_reservation(page, offset);
3119 
3120 out:
3121 	ext4_invalidatepage(page, offset);
3122 
3123 	return;
3124 }
3125 
3126 /*
3127  * Force all delayed allocation blocks to be allocated for a given inode.
3128  */
3129 int ext4_alloc_da_blocks(struct inode *inode)
3130 {
3131 	trace_ext4_alloc_da_blocks(inode);
3132 
3133 	if (!EXT4_I(inode)->i_reserved_data_blocks &&
3134 	    !EXT4_I(inode)->i_reserved_meta_blocks)
3135 		return 0;
3136 
3137 	/*
3138 	 * We do something simple for now.  The filemap_flush() will
3139 	 * also start triggering a write of the data blocks, which is
3140 	 * not strictly speaking necessary (and for users of
3141 	 * laptop_mode, not even desirable).  However, to do otherwise
3142 	 * would require replicating code paths in:
3143 	 *
3144 	 * ext4_da_writepages() ->
3145 	 *    write_cache_pages() ---> (via passed in callback function)
3146 	 *        __mpage_da_writepage() -->
3147 	 *           mpage_add_bh_to_extent()
3148 	 *           mpage_da_map_blocks()
3149 	 *
3150 	 * The problem is that write_cache_pages(), located in
3151 	 * mm/page-writeback.c, marks pages clean in preparation for
3152 	 * doing I/O, which is not desirable if we're not planning on
3153 	 * doing I/O at all.
3154 	 *
3155 	 * We could call write_cache_pages(), and then redirty all of
3156 	 * the pages by calling redirty_page_for_writeback() but that
3157 	 * would be ugly in the extreme.  So instead we would need to
3158 	 * replicate parts of the code in the above functions,
3159 	 * simplifying them becuase we wouldn't actually intend to
3160 	 * write out the pages, but rather only collect contiguous
3161 	 * logical block extents, call the multi-block allocator, and
3162 	 * then update the buffer heads with the block allocations.
3163 	 *
3164 	 * For now, though, we'll cheat by calling filemap_flush(),
3165 	 * which will map the blocks, and start the I/O, but not
3166 	 * actually wait for the I/O to complete.
3167 	 */
3168 	return filemap_flush(inode->i_mapping);
3169 }
3170 
3171 /*
3172  * bmap() is special.  It gets used by applications such as lilo and by
3173  * the swapper to find the on-disk block of a specific piece of data.
3174  *
3175  * Naturally, this is dangerous if the block concerned is still in the
3176  * journal.  If somebody makes a swapfile on an ext4 data-journaling
3177  * filesystem and enables swap, then they may get a nasty shock when the
3178  * data getting swapped to that swapfile suddenly gets overwritten by
3179  * the original zero's written out previously to the journal and
3180  * awaiting writeback in the kernel's buffer cache.
3181  *
3182  * So, if we see any bmap calls here on a modified, data-journaled file,
3183  * take extra steps to flush any blocks which might be in the cache.
3184  */
3185 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3186 {
3187 	struct inode *inode = mapping->host;
3188 	journal_t *journal;
3189 	int err;
3190 
3191 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3192 			test_opt(inode->i_sb, DELALLOC)) {
3193 		/*
3194 		 * With delalloc we want to sync the file
3195 		 * so that we can make sure we allocate
3196 		 * blocks for file
3197 		 */
3198 		filemap_write_and_wait(mapping);
3199 	}
3200 
3201 	if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
3202 		/*
3203 		 * This is a REALLY heavyweight approach, but the use of
3204 		 * bmap on dirty files is expected to be extremely rare:
3205 		 * only if we run lilo or swapon on a freshly made file
3206 		 * do we expect this to happen.
3207 		 *
3208 		 * (bmap requires CAP_SYS_RAWIO so this does not
3209 		 * represent an unprivileged user DOS attack --- we'd be
3210 		 * in trouble if mortal users could trigger this path at
3211 		 * will.)
3212 		 *
3213 		 * NB. EXT4_STATE_JDATA is not set on files other than
3214 		 * regular files.  If somebody wants to bmap a directory
3215 		 * or symlink and gets confused because the buffer
3216 		 * hasn't yet been flushed to disk, they deserve
3217 		 * everything they get.
3218 		 */
3219 
3220 		EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3221 		journal = EXT4_JOURNAL(inode);
3222 		jbd2_journal_lock_updates(journal);
3223 		err = jbd2_journal_flush(journal);
3224 		jbd2_journal_unlock_updates(journal);
3225 
3226 		if (err)
3227 			return 0;
3228 	}
3229 
3230 	return generic_block_bmap(mapping, block, ext4_get_block);
3231 }
3232 
3233 static int ext4_readpage(struct file *file, struct page *page)
3234 {
3235 	return mpage_readpage(page, ext4_get_block);
3236 }
3237 
3238 static int
3239 ext4_readpages(struct file *file, struct address_space *mapping,
3240 		struct list_head *pages, unsigned nr_pages)
3241 {
3242 	return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3243 }
3244 
3245 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3246 {
3247 	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3248 
3249 	/*
3250 	 * If it's a full truncate we just forget about the pending dirtying
3251 	 */
3252 	if (offset == 0)
3253 		ClearPageChecked(page);
3254 
3255 	if (journal)
3256 		jbd2_journal_invalidatepage(journal, page, offset);
3257 	else
3258 		block_invalidatepage(page, offset);
3259 }
3260 
3261 static int ext4_releasepage(struct page *page, gfp_t wait)
3262 {
3263 	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3264 
3265 	WARN_ON(PageChecked(page));
3266 	if (!page_has_buffers(page))
3267 		return 0;
3268 	if (journal)
3269 		return jbd2_journal_try_to_free_buffers(journal, page, wait);
3270 	else
3271 		return try_to_free_buffers(page);
3272 }
3273 
3274 /*
3275  * If the O_DIRECT write will extend the file then add this inode to the
3276  * orphan list.  So recovery will truncate it back to the original size
3277  * if the machine crashes during the write.
3278  *
3279  * If the O_DIRECT write is intantiating holes inside i_size and the machine
3280  * crashes then stale disk data _may_ be exposed inside the file. But current
3281  * VFS code falls back into buffered path in that case so we are safe.
3282  */
3283 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3284 			      const struct iovec *iov, loff_t offset,
3285 			      unsigned long nr_segs)
3286 {
3287 	struct file *file = iocb->ki_filp;
3288 	struct inode *inode = file->f_mapping->host;
3289 	struct ext4_inode_info *ei = EXT4_I(inode);
3290 	handle_t *handle;
3291 	ssize_t ret;
3292 	int orphan = 0;
3293 	size_t count = iov_length(iov, nr_segs);
3294 
3295 	if (rw == WRITE) {
3296 		loff_t final_size = offset + count;
3297 
3298 		if (final_size > inode->i_size) {
3299 			/* Credits for sb + inode write */
3300 			handle = ext4_journal_start(inode, 2);
3301 			if (IS_ERR(handle)) {
3302 				ret = PTR_ERR(handle);
3303 				goto out;
3304 			}
3305 			ret = ext4_orphan_add(handle, inode);
3306 			if (ret) {
3307 				ext4_journal_stop(handle);
3308 				goto out;
3309 			}
3310 			orphan = 1;
3311 			ei->i_disksize = inode->i_size;
3312 			ext4_journal_stop(handle);
3313 		}
3314 	}
3315 
3316 	ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3317 				 offset, nr_segs,
3318 				 ext4_get_block, NULL);
3319 
3320 	if (orphan) {
3321 		int err;
3322 
3323 		/* Credits for sb + inode write */
3324 		handle = ext4_journal_start(inode, 2);
3325 		if (IS_ERR(handle)) {
3326 			/* This is really bad luck. We've written the data
3327 			 * but cannot extend i_size. Bail out and pretend
3328 			 * the write failed... */
3329 			ret = PTR_ERR(handle);
3330 			goto out;
3331 		}
3332 		if (inode->i_nlink)
3333 			ext4_orphan_del(handle, inode);
3334 		if (ret > 0) {
3335 			loff_t end = offset + ret;
3336 			if (end > inode->i_size) {
3337 				ei->i_disksize = end;
3338 				i_size_write(inode, end);
3339 				/*
3340 				 * We're going to return a positive `ret'
3341 				 * here due to non-zero-length I/O, so there's
3342 				 * no way of reporting error returns from
3343 				 * ext4_mark_inode_dirty() to userspace.  So
3344 				 * ignore it.
3345 				 */
3346 				ext4_mark_inode_dirty(handle, inode);
3347 			}
3348 		}
3349 		err = ext4_journal_stop(handle);
3350 		if (ret == 0)
3351 			ret = err;
3352 	}
3353 out:
3354 	return ret;
3355 }
3356 
3357 /*
3358  * Pages can be marked dirty completely asynchronously from ext4's journalling
3359  * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3360  * much here because ->set_page_dirty is called under VFS locks.  The page is
3361  * not necessarily locked.
3362  *
3363  * We cannot just dirty the page and leave attached buffers clean, because the
3364  * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3365  * or jbddirty because all the journalling code will explode.
3366  *
3367  * So what we do is to mark the page "pending dirty" and next time writepage
3368  * is called, propagate that into the buffers appropriately.
3369  */
3370 static int ext4_journalled_set_page_dirty(struct page *page)
3371 {
3372 	SetPageChecked(page);
3373 	return __set_page_dirty_nobuffers(page);
3374 }
3375 
3376 static const struct address_space_operations ext4_ordered_aops = {
3377 	.readpage		= ext4_readpage,
3378 	.readpages		= ext4_readpages,
3379 	.writepage		= ext4_writepage,
3380 	.sync_page		= block_sync_page,
3381 	.write_begin		= ext4_write_begin,
3382 	.write_end		= ext4_ordered_write_end,
3383 	.bmap			= ext4_bmap,
3384 	.invalidatepage		= ext4_invalidatepage,
3385 	.releasepage		= ext4_releasepage,
3386 	.direct_IO		= ext4_direct_IO,
3387 	.migratepage		= buffer_migrate_page,
3388 	.is_partially_uptodate  = block_is_partially_uptodate,
3389 };
3390 
3391 static const struct address_space_operations ext4_writeback_aops = {
3392 	.readpage		= ext4_readpage,
3393 	.readpages		= ext4_readpages,
3394 	.writepage		= ext4_writepage,
3395 	.sync_page		= block_sync_page,
3396 	.write_begin		= ext4_write_begin,
3397 	.write_end		= ext4_writeback_write_end,
3398 	.bmap			= ext4_bmap,
3399 	.invalidatepage		= ext4_invalidatepage,
3400 	.releasepage		= ext4_releasepage,
3401 	.direct_IO		= ext4_direct_IO,
3402 	.migratepage		= buffer_migrate_page,
3403 	.is_partially_uptodate  = block_is_partially_uptodate,
3404 };
3405 
3406 static const struct address_space_operations ext4_journalled_aops = {
3407 	.readpage		= ext4_readpage,
3408 	.readpages		= ext4_readpages,
3409 	.writepage		= ext4_writepage,
3410 	.sync_page		= block_sync_page,
3411 	.write_begin		= ext4_write_begin,
3412 	.write_end		= ext4_journalled_write_end,
3413 	.set_page_dirty		= ext4_journalled_set_page_dirty,
3414 	.bmap			= ext4_bmap,
3415 	.invalidatepage		= ext4_invalidatepage,
3416 	.releasepage		= ext4_releasepage,
3417 	.is_partially_uptodate  = block_is_partially_uptodate,
3418 };
3419 
3420 static const struct address_space_operations ext4_da_aops = {
3421 	.readpage		= ext4_readpage,
3422 	.readpages		= ext4_readpages,
3423 	.writepage		= ext4_writepage,
3424 	.writepages		= ext4_da_writepages,
3425 	.sync_page		= block_sync_page,
3426 	.write_begin		= ext4_da_write_begin,
3427 	.write_end		= ext4_da_write_end,
3428 	.bmap			= ext4_bmap,
3429 	.invalidatepage		= ext4_da_invalidatepage,
3430 	.releasepage		= ext4_releasepage,
3431 	.direct_IO		= ext4_direct_IO,
3432 	.migratepage		= buffer_migrate_page,
3433 	.is_partially_uptodate  = block_is_partially_uptodate,
3434 };
3435 
3436 void ext4_set_aops(struct inode *inode)
3437 {
3438 	if (ext4_should_order_data(inode) &&
3439 		test_opt(inode->i_sb, DELALLOC))
3440 		inode->i_mapping->a_ops = &ext4_da_aops;
3441 	else if (ext4_should_order_data(inode))
3442 		inode->i_mapping->a_ops = &ext4_ordered_aops;
3443 	else if (ext4_should_writeback_data(inode) &&
3444 		 test_opt(inode->i_sb, DELALLOC))
3445 		inode->i_mapping->a_ops = &ext4_da_aops;
3446 	else if (ext4_should_writeback_data(inode))
3447 		inode->i_mapping->a_ops = &ext4_writeback_aops;
3448 	else
3449 		inode->i_mapping->a_ops = &ext4_journalled_aops;
3450 }
3451 
3452 /*
3453  * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3454  * up to the end of the block which corresponds to `from'.
3455  * This required during truncate. We need to physically zero the tail end
3456  * of that block so it doesn't yield old data if the file is later grown.
3457  */
3458 int ext4_block_truncate_page(handle_t *handle,
3459 		struct address_space *mapping, loff_t from)
3460 {
3461 	ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3462 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
3463 	unsigned blocksize, length, pos;
3464 	ext4_lblk_t iblock;
3465 	struct inode *inode = mapping->host;
3466 	struct buffer_head *bh;
3467 	struct page *page;
3468 	int err = 0;
3469 
3470 	page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3471 				   mapping_gfp_mask(mapping) & ~__GFP_FS);
3472 	if (!page)
3473 		return -EINVAL;
3474 
3475 	blocksize = inode->i_sb->s_blocksize;
3476 	length = blocksize - (offset & (blocksize - 1));
3477 	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3478 
3479 	/*
3480 	 * For "nobh" option,  we can only work if we don't need to
3481 	 * read-in the page - otherwise we create buffers to do the IO.
3482 	 */
3483 	if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3484 	     ext4_should_writeback_data(inode) && PageUptodate(page)) {
3485 		zero_user(page, offset, length);
3486 		set_page_dirty(page);
3487 		goto unlock;
3488 	}
3489 
3490 	if (!page_has_buffers(page))
3491 		create_empty_buffers(page, blocksize, 0);
3492 
3493 	/* Find the buffer that contains "offset" */
3494 	bh = page_buffers(page);
3495 	pos = blocksize;
3496 	while (offset >= pos) {
3497 		bh = bh->b_this_page;
3498 		iblock++;
3499 		pos += blocksize;
3500 	}
3501 
3502 	err = 0;
3503 	if (buffer_freed(bh)) {
3504 		BUFFER_TRACE(bh, "freed: skip");
3505 		goto unlock;
3506 	}
3507 
3508 	if (!buffer_mapped(bh)) {
3509 		BUFFER_TRACE(bh, "unmapped");
3510 		ext4_get_block(inode, iblock, bh, 0);
3511 		/* unmapped? It's a hole - nothing to do */
3512 		if (!buffer_mapped(bh)) {
3513 			BUFFER_TRACE(bh, "still unmapped");
3514 			goto unlock;
3515 		}
3516 	}
3517 
3518 	/* Ok, it's mapped. Make sure it's up-to-date */
3519 	if (PageUptodate(page))
3520 		set_buffer_uptodate(bh);
3521 
3522 	if (!buffer_uptodate(bh)) {
3523 		err = -EIO;
3524 		ll_rw_block(READ, 1, &bh);
3525 		wait_on_buffer(bh);
3526 		/* Uhhuh. Read error. Complain and punt. */
3527 		if (!buffer_uptodate(bh))
3528 			goto unlock;
3529 	}
3530 
3531 	if (ext4_should_journal_data(inode)) {
3532 		BUFFER_TRACE(bh, "get write access");
3533 		err = ext4_journal_get_write_access(handle, bh);
3534 		if (err)
3535 			goto unlock;
3536 	}
3537 
3538 	zero_user(page, offset, length);
3539 
3540 	BUFFER_TRACE(bh, "zeroed end of block");
3541 
3542 	err = 0;
3543 	if (ext4_should_journal_data(inode)) {
3544 		err = ext4_handle_dirty_metadata(handle, inode, bh);
3545 	} else {
3546 		if (ext4_should_order_data(inode))
3547 			err = ext4_jbd2_file_inode(handle, inode);
3548 		mark_buffer_dirty(bh);
3549 	}
3550 
3551 unlock:
3552 	unlock_page(page);
3553 	page_cache_release(page);
3554 	return err;
3555 }
3556 
3557 /*
3558  * Probably it should be a library function... search for first non-zero word
3559  * or memcmp with zero_page, whatever is better for particular architecture.
3560  * Linus?
3561  */
3562 static inline int all_zeroes(__le32 *p, __le32 *q)
3563 {
3564 	while (p < q)
3565 		if (*p++)
3566 			return 0;
3567 	return 1;
3568 }
3569 
3570 /**
3571  *	ext4_find_shared - find the indirect blocks for partial truncation.
3572  *	@inode:	  inode in question
3573  *	@depth:	  depth of the affected branch
3574  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
3575  *	@chain:	  place to store the pointers to partial indirect blocks
3576  *	@top:	  place to the (detached) top of branch
3577  *
3578  *	This is a helper function used by ext4_truncate().
3579  *
3580  *	When we do truncate() we may have to clean the ends of several
3581  *	indirect blocks but leave the blocks themselves alive. Block is
3582  *	partially truncated if some data below the new i_size is refered
3583  *	from it (and it is on the path to the first completely truncated
3584  *	data block, indeed).  We have to free the top of that path along
3585  *	with everything to the right of the path. Since no allocation
3586  *	past the truncation point is possible until ext4_truncate()
3587  *	finishes, we may safely do the latter, but top of branch may
3588  *	require special attention - pageout below the truncation point
3589  *	might try to populate it.
3590  *
3591  *	We atomically detach the top of branch from the tree, store the
3592  *	block number of its root in *@top, pointers to buffer_heads of
3593  *	partially truncated blocks - in @chain[].bh and pointers to
3594  *	their last elements that should not be removed - in
3595  *	@chain[].p. Return value is the pointer to last filled element
3596  *	of @chain.
3597  *
3598  *	The work left to caller to do the actual freeing of subtrees:
3599  *		a) free the subtree starting from *@top
3600  *		b) free the subtrees whose roots are stored in
3601  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
3602  *		c) free the subtrees growing from the inode past the @chain[0].
3603  *			(no partially truncated stuff there).  */
3604 
3605 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3606 				  ext4_lblk_t offsets[4], Indirect chain[4],
3607 				  __le32 *top)
3608 {
3609 	Indirect *partial, *p;
3610 	int k, err;
3611 
3612 	*top = 0;
3613 	/* Make k index the deepest non-null offest + 1 */
3614 	for (k = depth; k > 1 && !offsets[k-1]; k--)
3615 		;
3616 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
3617 	/* Writer: pointers */
3618 	if (!partial)
3619 		partial = chain + k-1;
3620 	/*
3621 	 * If the branch acquired continuation since we've looked at it -
3622 	 * fine, it should all survive and (new) top doesn't belong to us.
3623 	 */
3624 	if (!partial->key && *partial->p)
3625 		/* Writer: end */
3626 		goto no_top;
3627 	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3628 		;
3629 	/*
3630 	 * OK, we've found the last block that must survive. The rest of our
3631 	 * branch should be detached before unlocking. However, if that rest
3632 	 * of branch is all ours and does not grow immediately from the inode
3633 	 * it's easier to cheat and just decrement partial->p.
3634 	 */
3635 	if (p == chain + k - 1 && p > chain) {
3636 		p->p--;
3637 	} else {
3638 		*top = *p->p;
3639 		/* Nope, don't do this in ext4.  Must leave the tree intact */
3640 #if 0
3641 		*p->p = 0;
3642 #endif
3643 	}
3644 	/* Writer: end */
3645 
3646 	while (partial > p) {
3647 		brelse(partial->bh);
3648 		partial--;
3649 	}
3650 no_top:
3651 	return partial;
3652 }
3653 
3654 /*
3655  * Zero a number of block pointers in either an inode or an indirect block.
3656  * If we restart the transaction we must again get write access to the
3657  * indirect block for further modification.
3658  *
3659  * We release `count' blocks on disk, but (last - first) may be greater
3660  * than `count' because there can be holes in there.
3661  */
3662 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3663 			      struct buffer_head *bh,
3664 			      ext4_fsblk_t block_to_free,
3665 			      unsigned long count, __le32 *first,
3666 			      __le32 *last)
3667 {
3668 	__le32 *p;
3669 	if (try_to_extend_transaction(handle, inode)) {
3670 		if (bh) {
3671 			BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3672 			ext4_handle_dirty_metadata(handle, inode, bh);
3673 		}
3674 		ext4_mark_inode_dirty(handle, inode);
3675 		ext4_truncate_restart_trans(handle, inode,
3676 					    blocks_for_truncate(inode));
3677 		if (bh) {
3678 			BUFFER_TRACE(bh, "retaking write access");
3679 			ext4_journal_get_write_access(handle, bh);
3680 		}
3681 	}
3682 
3683 	/*
3684 	 * Any buffers which are on the journal will be in memory. We
3685 	 * find them on the hash table so jbd2_journal_revoke() will
3686 	 * run jbd2_journal_forget() on them.  We've already detached
3687 	 * each block from the file, so bforget() in
3688 	 * jbd2_journal_forget() should be safe.
3689 	 *
3690 	 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3691 	 */
3692 	for (p = first; p < last; p++) {
3693 		u32 nr = le32_to_cpu(*p);
3694 		if (nr) {
3695 			struct buffer_head *tbh;
3696 
3697 			*p = 0;
3698 			tbh = sb_find_get_block(inode->i_sb, nr);
3699 			ext4_forget(handle, 0, inode, tbh, nr);
3700 		}
3701 	}
3702 
3703 	ext4_free_blocks(handle, inode, block_to_free, count, 0);
3704 }
3705 
3706 /**
3707  * ext4_free_data - free a list of data blocks
3708  * @handle:	handle for this transaction
3709  * @inode:	inode we are dealing with
3710  * @this_bh:	indirect buffer_head which contains *@first and *@last
3711  * @first:	array of block numbers
3712  * @last:	points immediately past the end of array
3713  *
3714  * We are freeing all blocks refered from that array (numbers are stored as
3715  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3716  *
3717  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
3718  * blocks are contiguous then releasing them at one time will only affect one
3719  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3720  * actually use a lot of journal space.
3721  *
3722  * @this_bh will be %NULL if @first and @last point into the inode's direct
3723  * block pointers.
3724  */
3725 static void ext4_free_data(handle_t *handle, struct inode *inode,
3726 			   struct buffer_head *this_bh,
3727 			   __le32 *first, __le32 *last)
3728 {
3729 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
3730 	unsigned long count = 0;	    /* Number of blocks in the run */
3731 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
3732 					       corresponding to
3733 					       block_to_free */
3734 	ext4_fsblk_t nr;		    /* Current block # */
3735 	__le32 *p;			    /* Pointer into inode/ind
3736 					       for current block */
3737 	int err;
3738 
3739 	if (this_bh) {				/* For indirect block */
3740 		BUFFER_TRACE(this_bh, "get_write_access");
3741 		err = ext4_journal_get_write_access(handle, this_bh);
3742 		/* Important: if we can't update the indirect pointers
3743 		 * to the blocks, we can't free them. */
3744 		if (err)
3745 			return;
3746 	}
3747 
3748 	for (p = first; p < last; p++) {
3749 		nr = le32_to_cpu(*p);
3750 		if (nr) {
3751 			/* accumulate blocks to free if they're contiguous */
3752 			if (count == 0) {
3753 				block_to_free = nr;
3754 				block_to_free_p = p;
3755 				count = 1;
3756 			} else if (nr == block_to_free + count) {
3757 				count++;
3758 			} else {
3759 				ext4_clear_blocks(handle, inode, this_bh,
3760 						  block_to_free,
3761 						  count, block_to_free_p, p);
3762 				block_to_free = nr;
3763 				block_to_free_p = p;
3764 				count = 1;
3765 			}
3766 		}
3767 	}
3768 
3769 	if (count > 0)
3770 		ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3771 				  count, block_to_free_p, p);
3772 
3773 	if (this_bh) {
3774 		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3775 
3776 		/*
3777 		 * The buffer head should have an attached journal head at this
3778 		 * point. However, if the data is corrupted and an indirect
3779 		 * block pointed to itself, it would have been detached when
3780 		 * the block was cleared. Check for this instead of OOPSing.
3781 		 */
3782 		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3783 			ext4_handle_dirty_metadata(handle, inode, this_bh);
3784 		else
3785 			ext4_error(inode->i_sb, __func__,
3786 				   "circular indirect block detected, "
3787 				   "inode=%lu, block=%llu",
3788 				   inode->i_ino,
3789 				   (unsigned long long) this_bh->b_blocknr);
3790 	}
3791 }
3792 
3793 /**
3794  *	ext4_free_branches - free an array of branches
3795  *	@handle: JBD handle for this transaction
3796  *	@inode:	inode we are dealing with
3797  *	@parent_bh: the buffer_head which contains *@first and *@last
3798  *	@first:	array of block numbers
3799  *	@last:	pointer immediately past the end of array
3800  *	@depth:	depth of the branches to free
3801  *
3802  *	We are freeing all blocks refered from these branches (numbers are
3803  *	stored as little-endian 32-bit) and updating @inode->i_blocks
3804  *	appropriately.
3805  */
3806 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3807 			       struct buffer_head *parent_bh,
3808 			       __le32 *first, __le32 *last, int depth)
3809 {
3810 	ext4_fsblk_t nr;
3811 	__le32 *p;
3812 
3813 	if (ext4_handle_is_aborted(handle))
3814 		return;
3815 
3816 	if (depth--) {
3817 		struct buffer_head *bh;
3818 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3819 		p = last;
3820 		while (--p >= first) {
3821 			nr = le32_to_cpu(*p);
3822 			if (!nr)
3823 				continue;		/* A hole */
3824 
3825 			/* Go read the buffer for the next level down */
3826 			bh = sb_bread(inode->i_sb, nr);
3827 
3828 			/*
3829 			 * A read failure? Report error and clear slot
3830 			 * (should be rare).
3831 			 */
3832 			if (!bh) {
3833 				ext4_error(inode->i_sb, "ext4_free_branches",
3834 					   "Read failure, inode=%lu, block=%llu",
3835 					   inode->i_ino, nr);
3836 				continue;
3837 			}
3838 
3839 			/* This zaps the entire block.  Bottom up. */
3840 			BUFFER_TRACE(bh, "free child branches");
3841 			ext4_free_branches(handle, inode, bh,
3842 					(__le32 *) bh->b_data,
3843 					(__le32 *) bh->b_data + addr_per_block,
3844 					depth);
3845 
3846 			/*
3847 			 * We've probably journalled the indirect block several
3848 			 * times during the truncate.  But it's no longer
3849 			 * needed and we now drop it from the transaction via
3850 			 * jbd2_journal_revoke().
3851 			 *
3852 			 * That's easy if it's exclusively part of this
3853 			 * transaction.  But if it's part of the committing
3854 			 * transaction then jbd2_journal_forget() will simply
3855 			 * brelse() it.  That means that if the underlying
3856 			 * block is reallocated in ext4_get_block(),
3857 			 * unmap_underlying_metadata() will find this block
3858 			 * and will try to get rid of it.  damn, damn.
3859 			 *
3860 			 * If this block has already been committed to the
3861 			 * journal, a revoke record will be written.  And
3862 			 * revoke records must be emitted *before* clearing
3863 			 * this block's bit in the bitmaps.
3864 			 */
3865 			ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3866 
3867 			/*
3868 			 * Everything below this this pointer has been
3869 			 * released.  Now let this top-of-subtree go.
3870 			 *
3871 			 * We want the freeing of this indirect block to be
3872 			 * atomic in the journal with the updating of the
3873 			 * bitmap block which owns it.  So make some room in
3874 			 * the journal.
3875 			 *
3876 			 * We zero the parent pointer *after* freeing its
3877 			 * pointee in the bitmaps, so if extend_transaction()
3878 			 * for some reason fails to put the bitmap changes and
3879 			 * the release into the same transaction, recovery
3880 			 * will merely complain about releasing a free block,
3881 			 * rather than leaking blocks.
3882 			 */
3883 			if (ext4_handle_is_aborted(handle))
3884 				return;
3885 			if (try_to_extend_transaction(handle, inode)) {
3886 				ext4_mark_inode_dirty(handle, inode);
3887 				ext4_truncate_restart_trans(handle, inode,
3888 					    blocks_for_truncate(inode));
3889 			}
3890 
3891 			ext4_free_blocks(handle, inode, nr, 1, 1);
3892 
3893 			if (parent_bh) {
3894 				/*
3895 				 * The block which we have just freed is
3896 				 * pointed to by an indirect block: journal it
3897 				 */
3898 				BUFFER_TRACE(parent_bh, "get_write_access");
3899 				if (!ext4_journal_get_write_access(handle,
3900 								   parent_bh)){
3901 					*p = 0;
3902 					BUFFER_TRACE(parent_bh,
3903 					"call ext4_handle_dirty_metadata");
3904 					ext4_handle_dirty_metadata(handle,
3905 								   inode,
3906 								   parent_bh);
3907 				}
3908 			}
3909 		}
3910 	} else {
3911 		/* We have reached the bottom of the tree. */
3912 		BUFFER_TRACE(parent_bh, "free data blocks");
3913 		ext4_free_data(handle, inode, parent_bh, first, last);
3914 	}
3915 }
3916 
3917 int ext4_can_truncate(struct inode *inode)
3918 {
3919 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3920 		return 0;
3921 	if (S_ISREG(inode->i_mode))
3922 		return 1;
3923 	if (S_ISDIR(inode->i_mode))
3924 		return 1;
3925 	if (S_ISLNK(inode->i_mode))
3926 		return !ext4_inode_is_fast_symlink(inode);
3927 	return 0;
3928 }
3929 
3930 /*
3931  * ext4_truncate()
3932  *
3933  * We block out ext4_get_block() block instantiations across the entire
3934  * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3935  * simultaneously on behalf of the same inode.
3936  *
3937  * As we work through the truncate and commmit bits of it to the journal there
3938  * is one core, guiding principle: the file's tree must always be consistent on
3939  * disk.  We must be able to restart the truncate after a crash.
3940  *
3941  * The file's tree may be transiently inconsistent in memory (although it
3942  * probably isn't), but whenever we close off and commit a journal transaction,
3943  * the contents of (the filesystem + the journal) must be consistent and
3944  * restartable.  It's pretty simple, really: bottom up, right to left (although
3945  * left-to-right works OK too).
3946  *
3947  * Note that at recovery time, journal replay occurs *before* the restart of
3948  * truncate against the orphan inode list.
3949  *
3950  * The committed inode has the new, desired i_size (which is the same as
3951  * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
3952  * that this inode's truncate did not complete and it will again call
3953  * ext4_truncate() to have another go.  So there will be instantiated blocks
3954  * to the right of the truncation point in a crashed ext4 filesystem.  But
3955  * that's fine - as long as they are linked from the inode, the post-crash
3956  * ext4_truncate() run will find them and release them.
3957  */
3958 void ext4_truncate(struct inode *inode)
3959 {
3960 	handle_t *handle;
3961 	struct ext4_inode_info *ei = EXT4_I(inode);
3962 	__le32 *i_data = ei->i_data;
3963 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3964 	struct address_space *mapping = inode->i_mapping;
3965 	ext4_lblk_t offsets[4];
3966 	Indirect chain[4];
3967 	Indirect *partial;
3968 	__le32 nr = 0;
3969 	int n;
3970 	ext4_lblk_t last_block;
3971 	unsigned blocksize = inode->i_sb->s_blocksize;
3972 
3973 	if (!ext4_can_truncate(inode))
3974 		return;
3975 
3976 	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3977 		ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
3978 
3979 	if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3980 		ext4_ext_truncate(inode);
3981 		return;
3982 	}
3983 
3984 	handle = start_transaction(inode);
3985 	if (IS_ERR(handle))
3986 		return;		/* AKPM: return what? */
3987 
3988 	last_block = (inode->i_size + blocksize-1)
3989 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3990 
3991 	if (inode->i_size & (blocksize - 1))
3992 		if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3993 			goto out_stop;
3994 
3995 	n = ext4_block_to_path(inode, last_block, offsets, NULL);
3996 	if (n == 0)
3997 		goto out_stop;	/* error */
3998 
3999 	/*
4000 	 * OK.  This truncate is going to happen.  We add the inode to the
4001 	 * orphan list, so that if this truncate spans multiple transactions,
4002 	 * and we crash, we will resume the truncate when the filesystem
4003 	 * recovers.  It also marks the inode dirty, to catch the new size.
4004 	 *
4005 	 * Implication: the file must always be in a sane, consistent
4006 	 * truncatable state while each transaction commits.
4007 	 */
4008 	if (ext4_orphan_add(handle, inode))
4009 		goto out_stop;
4010 
4011 	/*
4012 	 * From here we block out all ext4_get_block() callers who want to
4013 	 * modify the block allocation tree.
4014 	 */
4015 	down_write(&ei->i_data_sem);
4016 
4017 	ext4_discard_preallocations(inode);
4018 
4019 	/*
4020 	 * The orphan list entry will now protect us from any crash which
4021 	 * occurs before the truncate completes, so it is now safe to propagate
4022 	 * the new, shorter inode size (held for now in i_size) into the
4023 	 * on-disk inode. We do this via i_disksize, which is the value which
4024 	 * ext4 *really* writes onto the disk inode.
4025 	 */
4026 	ei->i_disksize = inode->i_size;
4027 
4028 	if (n == 1) {		/* direct blocks */
4029 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4030 			       i_data + EXT4_NDIR_BLOCKS);
4031 		goto do_indirects;
4032 	}
4033 
4034 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4035 	/* Kill the top of shared branch (not detached) */
4036 	if (nr) {
4037 		if (partial == chain) {
4038 			/* Shared branch grows from the inode */
4039 			ext4_free_branches(handle, inode, NULL,
4040 					   &nr, &nr+1, (chain+n-1) - partial);
4041 			*partial->p = 0;
4042 			/*
4043 			 * We mark the inode dirty prior to restart,
4044 			 * and prior to stop.  No need for it here.
4045 			 */
4046 		} else {
4047 			/* Shared branch grows from an indirect block */
4048 			BUFFER_TRACE(partial->bh, "get_write_access");
4049 			ext4_free_branches(handle, inode, partial->bh,
4050 					partial->p,
4051 					partial->p+1, (chain+n-1) - partial);
4052 		}
4053 	}
4054 	/* Clear the ends of indirect blocks on the shared branch */
4055 	while (partial > chain) {
4056 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4057 				   (__le32*)partial->bh->b_data+addr_per_block,
4058 				   (chain+n-1) - partial);
4059 		BUFFER_TRACE(partial->bh, "call brelse");
4060 		brelse(partial->bh);
4061 		partial--;
4062 	}
4063 do_indirects:
4064 	/* Kill the remaining (whole) subtrees */
4065 	switch (offsets[0]) {
4066 	default:
4067 		nr = i_data[EXT4_IND_BLOCK];
4068 		if (nr) {
4069 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4070 			i_data[EXT4_IND_BLOCK] = 0;
4071 		}
4072 	case EXT4_IND_BLOCK:
4073 		nr = i_data[EXT4_DIND_BLOCK];
4074 		if (nr) {
4075 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4076 			i_data[EXT4_DIND_BLOCK] = 0;
4077 		}
4078 	case EXT4_DIND_BLOCK:
4079 		nr = i_data[EXT4_TIND_BLOCK];
4080 		if (nr) {
4081 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4082 			i_data[EXT4_TIND_BLOCK] = 0;
4083 		}
4084 	case EXT4_TIND_BLOCK:
4085 		;
4086 	}
4087 
4088 	up_write(&ei->i_data_sem);
4089 	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4090 	ext4_mark_inode_dirty(handle, inode);
4091 
4092 	/*
4093 	 * In a multi-transaction truncate, we only make the final transaction
4094 	 * synchronous
4095 	 */
4096 	if (IS_SYNC(inode))
4097 		ext4_handle_sync(handle);
4098 out_stop:
4099 	/*
4100 	 * If this was a simple ftruncate(), and the file will remain alive
4101 	 * then we need to clear up the orphan record which we created above.
4102 	 * However, if this was a real unlink then we were called by
4103 	 * ext4_delete_inode(), and we allow that function to clean up the
4104 	 * orphan info for us.
4105 	 */
4106 	if (inode->i_nlink)
4107 		ext4_orphan_del(handle, inode);
4108 
4109 	ext4_journal_stop(handle);
4110 }
4111 
4112 /*
4113  * ext4_get_inode_loc returns with an extra refcount against the inode's
4114  * underlying buffer_head on success. If 'in_mem' is true, we have all
4115  * data in memory that is needed to recreate the on-disk version of this
4116  * inode.
4117  */
4118 static int __ext4_get_inode_loc(struct inode *inode,
4119 				struct ext4_iloc *iloc, int in_mem)
4120 {
4121 	struct ext4_group_desc	*gdp;
4122 	struct buffer_head	*bh;
4123 	struct super_block	*sb = inode->i_sb;
4124 	ext4_fsblk_t		block;
4125 	int			inodes_per_block, inode_offset;
4126 
4127 	iloc->bh = NULL;
4128 	if (!ext4_valid_inum(sb, inode->i_ino))
4129 		return -EIO;
4130 
4131 	iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4132 	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4133 	if (!gdp)
4134 		return -EIO;
4135 
4136 	/*
4137 	 * Figure out the offset within the block group inode table
4138 	 */
4139 	inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4140 	inode_offset = ((inode->i_ino - 1) %
4141 			EXT4_INODES_PER_GROUP(sb));
4142 	block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4143 	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4144 
4145 	bh = sb_getblk(sb, block);
4146 	if (!bh) {
4147 		ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4148 			   "inode block - inode=%lu, block=%llu",
4149 			   inode->i_ino, block);
4150 		return -EIO;
4151 	}
4152 	if (!buffer_uptodate(bh)) {
4153 		lock_buffer(bh);
4154 
4155 		/*
4156 		 * If the buffer has the write error flag, we have failed
4157 		 * to write out another inode in the same block.  In this
4158 		 * case, we don't have to read the block because we may
4159 		 * read the old inode data successfully.
4160 		 */
4161 		if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4162 			set_buffer_uptodate(bh);
4163 
4164 		if (buffer_uptodate(bh)) {
4165 			/* someone brought it uptodate while we waited */
4166 			unlock_buffer(bh);
4167 			goto has_buffer;
4168 		}
4169 
4170 		/*
4171 		 * If we have all information of the inode in memory and this
4172 		 * is the only valid inode in the block, we need not read the
4173 		 * block.
4174 		 */
4175 		if (in_mem) {
4176 			struct buffer_head *bitmap_bh;
4177 			int i, start;
4178 
4179 			start = inode_offset & ~(inodes_per_block - 1);
4180 
4181 			/* Is the inode bitmap in cache? */
4182 			bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4183 			if (!bitmap_bh)
4184 				goto make_io;
4185 
4186 			/*
4187 			 * If the inode bitmap isn't in cache then the
4188 			 * optimisation may end up performing two reads instead
4189 			 * of one, so skip it.
4190 			 */
4191 			if (!buffer_uptodate(bitmap_bh)) {
4192 				brelse(bitmap_bh);
4193 				goto make_io;
4194 			}
4195 			for (i = start; i < start + inodes_per_block; i++) {
4196 				if (i == inode_offset)
4197 					continue;
4198 				if (ext4_test_bit(i, bitmap_bh->b_data))
4199 					break;
4200 			}
4201 			brelse(bitmap_bh);
4202 			if (i == start + inodes_per_block) {
4203 				/* all other inodes are free, so skip I/O */
4204 				memset(bh->b_data, 0, bh->b_size);
4205 				set_buffer_uptodate(bh);
4206 				unlock_buffer(bh);
4207 				goto has_buffer;
4208 			}
4209 		}
4210 
4211 make_io:
4212 		/*
4213 		 * If we need to do any I/O, try to pre-readahead extra
4214 		 * blocks from the inode table.
4215 		 */
4216 		if (EXT4_SB(sb)->s_inode_readahead_blks) {
4217 			ext4_fsblk_t b, end, table;
4218 			unsigned num;
4219 
4220 			table = ext4_inode_table(sb, gdp);
4221 			/* s_inode_readahead_blks is always a power of 2 */
4222 			b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4223 			if (table > b)
4224 				b = table;
4225 			end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4226 			num = EXT4_INODES_PER_GROUP(sb);
4227 			if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4228 				       EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4229 				num -= ext4_itable_unused_count(sb, gdp);
4230 			table += num / inodes_per_block;
4231 			if (end > table)
4232 				end = table;
4233 			while (b <= end)
4234 				sb_breadahead(sb, b++);
4235 		}
4236 
4237 		/*
4238 		 * There are other valid inodes in the buffer, this inode
4239 		 * has in-inode xattrs, or we don't have this inode in memory.
4240 		 * Read the block from disk.
4241 		 */
4242 		get_bh(bh);
4243 		bh->b_end_io = end_buffer_read_sync;
4244 		submit_bh(READ_META, bh);
4245 		wait_on_buffer(bh);
4246 		if (!buffer_uptodate(bh)) {
4247 			ext4_error(sb, __func__,
4248 				   "unable to read inode block - inode=%lu, "
4249 				   "block=%llu", inode->i_ino, block);
4250 			brelse(bh);
4251 			return -EIO;
4252 		}
4253 	}
4254 has_buffer:
4255 	iloc->bh = bh;
4256 	return 0;
4257 }
4258 
4259 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4260 {
4261 	/* We have all inode data except xattrs in memory here. */
4262 	return __ext4_get_inode_loc(inode, iloc,
4263 		!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4264 }
4265 
4266 void ext4_set_inode_flags(struct inode *inode)
4267 {
4268 	unsigned int flags = EXT4_I(inode)->i_flags;
4269 
4270 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4271 	if (flags & EXT4_SYNC_FL)
4272 		inode->i_flags |= S_SYNC;
4273 	if (flags & EXT4_APPEND_FL)
4274 		inode->i_flags |= S_APPEND;
4275 	if (flags & EXT4_IMMUTABLE_FL)
4276 		inode->i_flags |= S_IMMUTABLE;
4277 	if (flags & EXT4_NOATIME_FL)
4278 		inode->i_flags |= S_NOATIME;
4279 	if (flags & EXT4_DIRSYNC_FL)
4280 		inode->i_flags |= S_DIRSYNC;
4281 }
4282 
4283 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4284 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4285 {
4286 	unsigned int flags = ei->vfs_inode.i_flags;
4287 
4288 	ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4289 			EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4290 	if (flags & S_SYNC)
4291 		ei->i_flags |= EXT4_SYNC_FL;
4292 	if (flags & S_APPEND)
4293 		ei->i_flags |= EXT4_APPEND_FL;
4294 	if (flags & S_IMMUTABLE)
4295 		ei->i_flags |= EXT4_IMMUTABLE_FL;
4296 	if (flags & S_NOATIME)
4297 		ei->i_flags |= EXT4_NOATIME_FL;
4298 	if (flags & S_DIRSYNC)
4299 		ei->i_flags |= EXT4_DIRSYNC_FL;
4300 }
4301 
4302 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4303 				  struct ext4_inode_info *ei)
4304 {
4305 	blkcnt_t i_blocks ;
4306 	struct inode *inode = &(ei->vfs_inode);
4307 	struct super_block *sb = inode->i_sb;
4308 
4309 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4310 				EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4311 		/* we are using combined 48 bit field */
4312 		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4313 					le32_to_cpu(raw_inode->i_blocks_lo);
4314 		if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4315 			/* i_blocks represent file system block size */
4316 			return i_blocks  << (inode->i_blkbits - 9);
4317 		} else {
4318 			return i_blocks;
4319 		}
4320 	} else {
4321 		return le32_to_cpu(raw_inode->i_blocks_lo);
4322 	}
4323 }
4324 
4325 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4326 {
4327 	struct ext4_iloc iloc;
4328 	struct ext4_inode *raw_inode;
4329 	struct ext4_inode_info *ei;
4330 	struct buffer_head *bh;
4331 	struct inode *inode;
4332 	long ret;
4333 	int block;
4334 
4335 	inode = iget_locked(sb, ino);
4336 	if (!inode)
4337 		return ERR_PTR(-ENOMEM);
4338 	if (!(inode->i_state & I_NEW))
4339 		return inode;
4340 
4341 	ei = EXT4_I(inode);
4342 
4343 	ret = __ext4_get_inode_loc(inode, &iloc, 0);
4344 	if (ret < 0)
4345 		goto bad_inode;
4346 	bh = iloc.bh;
4347 	raw_inode = ext4_raw_inode(&iloc);
4348 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4349 	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4350 	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4351 	if (!(test_opt(inode->i_sb, NO_UID32))) {
4352 		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4353 		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4354 	}
4355 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4356 
4357 	ei->i_state = 0;
4358 	ei->i_dir_start_lookup = 0;
4359 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4360 	/* We now have enough fields to check if the inode was active or not.
4361 	 * This is needed because nfsd might try to access dead inodes
4362 	 * the test is that same one that e2fsck uses
4363 	 * NeilBrown 1999oct15
4364 	 */
4365 	if (inode->i_nlink == 0) {
4366 		if (inode->i_mode == 0 ||
4367 		    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4368 			/* this inode is deleted */
4369 			brelse(bh);
4370 			ret = -ESTALE;
4371 			goto bad_inode;
4372 		}
4373 		/* The only unlinked inodes we let through here have
4374 		 * valid i_mode and are being read by the orphan
4375 		 * recovery code: that's fine, we're about to complete
4376 		 * the process of deleting those. */
4377 	}
4378 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4379 	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4380 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4381 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4382 		ei->i_file_acl |=
4383 			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4384 	inode->i_size = ext4_isize(raw_inode);
4385 	ei->i_disksize = inode->i_size;
4386 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4387 	ei->i_block_group = iloc.block_group;
4388 	ei->i_last_alloc_group = ~0;
4389 	/*
4390 	 * NOTE! The in-memory inode i_data array is in little-endian order
4391 	 * even on big-endian machines: we do NOT byteswap the block numbers!
4392 	 */
4393 	for (block = 0; block < EXT4_N_BLOCKS; block++)
4394 		ei->i_data[block] = raw_inode->i_block[block];
4395 	INIT_LIST_HEAD(&ei->i_orphan);
4396 
4397 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4398 		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4399 		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4400 		    EXT4_INODE_SIZE(inode->i_sb)) {
4401 			brelse(bh);
4402 			ret = -EIO;
4403 			goto bad_inode;
4404 		}
4405 		if (ei->i_extra_isize == 0) {
4406 			/* The extra space is currently unused. Use it. */
4407 			ei->i_extra_isize = sizeof(struct ext4_inode) -
4408 					    EXT4_GOOD_OLD_INODE_SIZE;
4409 		} else {
4410 			__le32 *magic = (void *)raw_inode +
4411 					EXT4_GOOD_OLD_INODE_SIZE +
4412 					ei->i_extra_isize;
4413 			if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4414 				ei->i_state |= EXT4_STATE_XATTR;
4415 		}
4416 	} else
4417 		ei->i_extra_isize = 0;
4418 
4419 	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4420 	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4421 	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4422 	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4423 
4424 	inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4425 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4426 		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4427 			inode->i_version |=
4428 			(__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4429 	}
4430 
4431 	ret = 0;
4432 	if (ei->i_file_acl &&
4433 	    ((ei->i_file_acl <
4434 	      (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4435 	       EXT4_SB(sb)->s_gdb_count)) ||
4436 	     (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4437 		ext4_error(sb, __func__,
4438 			   "bad extended attribute block %llu in inode #%lu",
4439 			   ei->i_file_acl, inode->i_ino);
4440 		ret = -EIO;
4441 		goto bad_inode;
4442 	} else if (ei->i_flags & EXT4_EXTENTS_FL) {
4443 		if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4444 		    (S_ISLNK(inode->i_mode) &&
4445 		     !ext4_inode_is_fast_symlink(inode)))
4446 			/* Validate extent which is part of inode */
4447 			ret = ext4_ext_check_inode(inode);
4448 	} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4449 		   (S_ISLNK(inode->i_mode) &&
4450 		    !ext4_inode_is_fast_symlink(inode))) {
4451 		/* Validate block references which are part of inode */
4452 		ret = ext4_check_inode_blockref(inode);
4453 	}
4454 	if (ret) {
4455 		brelse(bh);
4456 		goto bad_inode;
4457 	}
4458 
4459 	if (S_ISREG(inode->i_mode)) {
4460 		inode->i_op = &ext4_file_inode_operations;
4461 		inode->i_fop = &ext4_file_operations;
4462 		ext4_set_aops(inode);
4463 	} else if (S_ISDIR(inode->i_mode)) {
4464 		inode->i_op = &ext4_dir_inode_operations;
4465 		inode->i_fop = &ext4_dir_operations;
4466 	} else if (S_ISLNK(inode->i_mode)) {
4467 		if (ext4_inode_is_fast_symlink(inode)) {
4468 			inode->i_op = &ext4_fast_symlink_inode_operations;
4469 			nd_terminate_link(ei->i_data, inode->i_size,
4470 				sizeof(ei->i_data) - 1);
4471 		} else {
4472 			inode->i_op = &ext4_symlink_inode_operations;
4473 			ext4_set_aops(inode);
4474 		}
4475 	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4476 	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4477 		inode->i_op = &ext4_special_inode_operations;
4478 		if (raw_inode->i_block[0])
4479 			init_special_inode(inode, inode->i_mode,
4480 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4481 		else
4482 			init_special_inode(inode, inode->i_mode,
4483 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4484 	} else {
4485 		brelse(bh);
4486 		ret = -EIO;
4487 		ext4_error(inode->i_sb, __func__,
4488 			   "bogus i_mode (%o) for inode=%lu",
4489 			   inode->i_mode, inode->i_ino);
4490 		goto bad_inode;
4491 	}
4492 	brelse(iloc.bh);
4493 	ext4_set_inode_flags(inode);
4494 	unlock_new_inode(inode);
4495 	return inode;
4496 
4497 bad_inode:
4498 	iget_failed(inode);
4499 	return ERR_PTR(ret);
4500 }
4501 
4502 static int ext4_inode_blocks_set(handle_t *handle,
4503 				struct ext4_inode *raw_inode,
4504 				struct ext4_inode_info *ei)
4505 {
4506 	struct inode *inode = &(ei->vfs_inode);
4507 	u64 i_blocks = inode->i_blocks;
4508 	struct super_block *sb = inode->i_sb;
4509 
4510 	if (i_blocks <= ~0U) {
4511 		/*
4512 		 * i_blocks can be represnted in a 32 bit variable
4513 		 * as multiple of 512 bytes
4514 		 */
4515 		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4516 		raw_inode->i_blocks_high = 0;
4517 		ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4518 		return 0;
4519 	}
4520 	if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4521 		return -EFBIG;
4522 
4523 	if (i_blocks <= 0xffffffffffffULL) {
4524 		/*
4525 		 * i_blocks can be represented in a 48 bit variable
4526 		 * as multiple of 512 bytes
4527 		 */
4528 		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4529 		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4530 		ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4531 	} else {
4532 		ei->i_flags |= EXT4_HUGE_FILE_FL;
4533 		/* i_block is stored in file system block size */
4534 		i_blocks = i_blocks >> (inode->i_blkbits - 9);
4535 		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4536 		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4537 	}
4538 	return 0;
4539 }
4540 
4541 /*
4542  * Post the struct inode info into an on-disk inode location in the
4543  * buffer-cache.  This gobbles the caller's reference to the
4544  * buffer_head in the inode location struct.
4545  *
4546  * The caller must have write access to iloc->bh.
4547  */
4548 static int ext4_do_update_inode(handle_t *handle,
4549 				struct inode *inode,
4550 				struct ext4_iloc *iloc,
4551 				int do_sync)
4552 {
4553 	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4554 	struct ext4_inode_info *ei = EXT4_I(inode);
4555 	struct buffer_head *bh = iloc->bh;
4556 	int err = 0, rc, block;
4557 
4558 	/* For fields not not tracking in the in-memory inode,
4559 	 * initialise them to zero for new inodes. */
4560 	if (ei->i_state & EXT4_STATE_NEW)
4561 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4562 
4563 	ext4_get_inode_flags(ei);
4564 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4565 	if (!(test_opt(inode->i_sb, NO_UID32))) {
4566 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4567 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4568 /*
4569  * Fix up interoperability with old kernels. Otherwise, old inodes get
4570  * re-used with the upper 16 bits of the uid/gid intact
4571  */
4572 		if (!ei->i_dtime) {
4573 			raw_inode->i_uid_high =
4574 				cpu_to_le16(high_16_bits(inode->i_uid));
4575 			raw_inode->i_gid_high =
4576 				cpu_to_le16(high_16_bits(inode->i_gid));
4577 		} else {
4578 			raw_inode->i_uid_high = 0;
4579 			raw_inode->i_gid_high = 0;
4580 		}
4581 	} else {
4582 		raw_inode->i_uid_low =
4583 			cpu_to_le16(fs_high2lowuid(inode->i_uid));
4584 		raw_inode->i_gid_low =
4585 			cpu_to_le16(fs_high2lowgid(inode->i_gid));
4586 		raw_inode->i_uid_high = 0;
4587 		raw_inode->i_gid_high = 0;
4588 	}
4589 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4590 
4591 	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4592 	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4593 	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4594 	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4595 
4596 	if (ext4_inode_blocks_set(handle, raw_inode, ei))
4597 		goto out_brelse;
4598 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4599 	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
4600 	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4601 	    cpu_to_le32(EXT4_OS_HURD))
4602 		raw_inode->i_file_acl_high =
4603 			cpu_to_le16(ei->i_file_acl >> 32);
4604 	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4605 	ext4_isize_set(raw_inode, ei->i_disksize);
4606 	if (ei->i_disksize > 0x7fffffffULL) {
4607 		struct super_block *sb = inode->i_sb;
4608 		if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4609 				EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4610 				EXT4_SB(sb)->s_es->s_rev_level ==
4611 				cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4612 			/* If this is the first large file
4613 			 * created, add a flag to the superblock.
4614 			 */
4615 			err = ext4_journal_get_write_access(handle,
4616 					EXT4_SB(sb)->s_sbh);
4617 			if (err)
4618 				goto out_brelse;
4619 			ext4_update_dynamic_rev(sb);
4620 			EXT4_SET_RO_COMPAT_FEATURE(sb,
4621 					EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4622 			sb->s_dirt = 1;
4623 			ext4_handle_sync(handle);
4624 			err = ext4_handle_dirty_metadata(handle, inode,
4625 					EXT4_SB(sb)->s_sbh);
4626 		}
4627 	}
4628 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4629 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4630 		if (old_valid_dev(inode->i_rdev)) {
4631 			raw_inode->i_block[0] =
4632 				cpu_to_le32(old_encode_dev(inode->i_rdev));
4633 			raw_inode->i_block[1] = 0;
4634 		} else {
4635 			raw_inode->i_block[0] = 0;
4636 			raw_inode->i_block[1] =
4637 				cpu_to_le32(new_encode_dev(inode->i_rdev));
4638 			raw_inode->i_block[2] = 0;
4639 		}
4640 	} else
4641 		for (block = 0; block < EXT4_N_BLOCKS; block++)
4642 			raw_inode->i_block[block] = ei->i_data[block];
4643 
4644 	raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4645 	if (ei->i_extra_isize) {
4646 		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4647 			raw_inode->i_version_hi =
4648 			cpu_to_le32(inode->i_version >> 32);
4649 		raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4650 	}
4651 
4652 	/*
4653 	 * If we're not using a journal and we were called from
4654 	 * ext4_write_inode() to sync the inode (making do_sync true),
4655 	 * we can just use sync_dirty_buffer() directly to do our dirty
4656 	 * work.  Testing s_journal here is a bit redundant but it's
4657 	 * worth it to avoid potential future trouble.
4658 	 */
4659 	if (EXT4_SB(inode->i_sb)->s_journal == NULL && do_sync) {
4660 		BUFFER_TRACE(bh, "call sync_dirty_buffer");
4661 		sync_dirty_buffer(bh);
4662 	} else {
4663 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4664 		rc = ext4_handle_dirty_metadata(handle, inode, bh);
4665 		if (!err)
4666 			err = rc;
4667 	}
4668 	ei->i_state &= ~EXT4_STATE_NEW;
4669 
4670 out_brelse:
4671 	brelse(bh);
4672 	ext4_std_error(inode->i_sb, err);
4673 	return err;
4674 }
4675 
4676 /*
4677  * ext4_write_inode()
4678  *
4679  * We are called from a few places:
4680  *
4681  * - Within generic_file_write() for O_SYNC files.
4682  *   Here, there will be no transaction running. We wait for any running
4683  *   trasnaction to commit.
4684  *
4685  * - Within sys_sync(), kupdate and such.
4686  *   We wait on commit, if tol to.
4687  *
4688  * - Within prune_icache() (PF_MEMALLOC == true)
4689  *   Here we simply return.  We can't afford to block kswapd on the
4690  *   journal commit.
4691  *
4692  * In all cases it is actually safe for us to return without doing anything,
4693  * because the inode has been copied into a raw inode buffer in
4694  * ext4_mark_inode_dirty().  This is a correctness thing for O_SYNC and for
4695  * knfsd.
4696  *
4697  * Note that we are absolutely dependent upon all inode dirtiers doing the
4698  * right thing: they *must* call mark_inode_dirty() after dirtying info in
4699  * which we are interested.
4700  *
4701  * It would be a bug for them to not do this.  The code:
4702  *
4703  *	mark_inode_dirty(inode)
4704  *	stuff();
4705  *	inode->i_size = expr;
4706  *
4707  * is in error because a kswapd-driven write_inode() could occur while
4708  * `stuff()' is running, and the new i_size will be lost.  Plus the inode
4709  * will no longer be on the superblock's dirty inode list.
4710  */
4711 int ext4_write_inode(struct inode *inode, int wait)
4712 {
4713 	int err;
4714 
4715 	if (current->flags & PF_MEMALLOC)
4716 		return 0;
4717 
4718 	if (EXT4_SB(inode->i_sb)->s_journal) {
4719 		if (ext4_journal_current_handle()) {
4720 			jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4721 			dump_stack();
4722 			return -EIO;
4723 		}
4724 
4725 		if (!wait)
4726 			return 0;
4727 
4728 		err = ext4_force_commit(inode->i_sb);
4729 	} else {
4730 		struct ext4_iloc iloc;
4731 
4732 		err = ext4_get_inode_loc(inode, &iloc);
4733 		if (err)
4734 			return err;
4735 		err = ext4_do_update_inode(EXT4_NOJOURNAL_HANDLE,
4736 					   inode, &iloc, wait);
4737 	}
4738 	return err;
4739 }
4740 
4741 /*
4742  * ext4_setattr()
4743  *
4744  * Called from notify_change.
4745  *
4746  * We want to trap VFS attempts to truncate the file as soon as
4747  * possible.  In particular, we want to make sure that when the VFS
4748  * shrinks i_size, we put the inode on the orphan list and modify
4749  * i_disksize immediately, so that during the subsequent flushing of
4750  * dirty pages and freeing of disk blocks, we can guarantee that any
4751  * commit will leave the blocks being flushed in an unused state on
4752  * disk.  (On recovery, the inode will get truncated and the blocks will
4753  * be freed, so we have a strong guarantee that no future commit will
4754  * leave these blocks visible to the user.)
4755  *
4756  * Another thing we have to assure is that if we are in ordered mode
4757  * and inode is still attached to the committing transaction, we must
4758  * we start writeout of all the dirty pages which are being truncated.
4759  * This way we are sure that all the data written in the previous
4760  * transaction are already on disk (truncate waits for pages under
4761  * writeback).
4762  *
4763  * Called with inode->i_mutex down.
4764  */
4765 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4766 {
4767 	struct inode *inode = dentry->d_inode;
4768 	int error, rc = 0;
4769 	const unsigned int ia_valid = attr->ia_valid;
4770 
4771 	error = inode_change_ok(inode, attr);
4772 	if (error)
4773 		return error;
4774 
4775 	if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4776 		(ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4777 		handle_t *handle;
4778 
4779 		/* (user+group)*(old+new) structure, inode write (sb,
4780 		 * inode block, ? - but truncate inode update has it) */
4781 		handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4782 					EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4783 		if (IS_ERR(handle)) {
4784 			error = PTR_ERR(handle);
4785 			goto err_out;
4786 		}
4787 		error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
4788 		if (error) {
4789 			ext4_journal_stop(handle);
4790 			return error;
4791 		}
4792 		/* Update corresponding info in inode so that everything is in
4793 		 * one transaction */
4794 		if (attr->ia_valid & ATTR_UID)
4795 			inode->i_uid = attr->ia_uid;
4796 		if (attr->ia_valid & ATTR_GID)
4797 			inode->i_gid = attr->ia_gid;
4798 		error = ext4_mark_inode_dirty(handle, inode);
4799 		ext4_journal_stop(handle);
4800 	}
4801 
4802 	if (attr->ia_valid & ATTR_SIZE) {
4803 		if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4804 			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4805 
4806 			if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4807 				error = -EFBIG;
4808 				goto err_out;
4809 			}
4810 		}
4811 	}
4812 
4813 	if (S_ISREG(inode->i_mode) &&
4814 	    attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4815 		handle_t *handle;
4816 
4817 		handle = ext4_journal_start(inode, 3);
4818 		if (IS_ERR(handle)) {
4819 			error = PTR_ERR(handle);
4820 			goto err_out;
4821 		}
4822 
4823 		error = ext4_orphan_add(handle, inode);
4824 		EXT4_I(inode)->i_disksize = attr->ia_size;
4825 		rc = ext4_mark_inode_dirty(handle, inode);
4826 		if (!error)
4827 			error = rc;
4828 		ext4_journal_stop(handle);
4829 
4830 		if (ext4_should_order_data(inode)) {
4831 			error = ext4_begin_ordered_truncate(inode,
4832 							    attr->ia_size);
4833 			if (error) {
4834 				/* Do as much error cleanup as possible */
4835 				handle = ext4_journal_start(inode, 3);
4836 				if (IS_ERR(handle)) {
4837 					ext4_orphan_del(NULL, inode);
4838 					goto err_out;
4839 				}
4840 				ext4_orphan_del(handle, inode);
4841 				ext4_journal_stop(handle);
4842 				goto err_out;
4843 			}
4844 		}
4845 	}
4846 
4847 	rc = inode_setattr(inode, attr);
4848 
4849 	/* If inode_setattr's call to ext4_truncate failed to get a
4850 	 * transaction handle at all, we need to clean up the in-core
4851 	 * orphan list manually. */
4852 	if (inode->i_nlink)
4853 		ext4_orphan_del(NULL, inode);
4854 
4855 	if (!rc && (ia_valid & ATTR_MODE))
4856 		rc = ext4_acl_chmod(inode);
4857 
4858 err_out:
4859 	ext4_std_error(inode->i_sb, error);
4860 	if (!error)
4861 		error = rc;
4862 	return error;
4863 }
4864 
4865 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4866 		 struct kstat *stat)
4867 {
4868 	struct inode *inode;
4869 	unsigned long delalloc_blocks;
4870 
4871 	inode = dentry->d_inode;
4872 	generic_fillattr(inode, stat);
4873 
4874 	/*
4875 	 * We can't update i_blocks if the block allocation is delayed
4876 	 * otherwise in the case of system crash before the real block
4877 	 * allocation is done, we will have i_blocks inconsistent with
4878 	 * on-disk file blocks.
4879 	 * We always keep i_blocks updated together with real
4880 	 * allocation. But to not confuse with user, stat
4881 	 * will return the blocks that include the delayed allocation
4882 	 * blocks for this file.
4883 	 */
4884 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4885 	delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4886 	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4887 
4888 	stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4889 	return 0;
4890 }
4891 
4892 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4893 				      int chunk)
4894 {
4895 	int indirects;
4896 
4897 	/* if nrblocks are contiguous */
4898 	if (chunk) {
4899 		/*
4900 		 * With N contiguous data blocks, it need at most
4901 		 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4902 		 * 2 dindirect blocks
4903 		 * 1 tindirect block
4904 		 */
4905 		indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4906 		return indirects + 3;
4907 	}
4908 	/*
4909 	 * if nrblocks are not contiguous, worse case, each block touch
4910 	 * a indirect block, and each indirect block touch a double indirect
4911 	 * block, plus a triple indirect block
4912 	 */
4913 	indirects = nrblocks * 2 + 1;
4914 	return indirects;
4915 }
4916 
4917 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4918 {
4919 	if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4920 		return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4921 	return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4922 }
4923 
4924 /*
4925  * Account for index blocks, block groups bitmaps and block group
4926  * descriptor blocks if modify datablocks and index blocks
4927  * worse case, the indexs blocks spread over different block groups
4928  *
4929  * If datablocks are discontiguous, they are possible to spread over
4930  * different block groups too. If they are contiugous, with flexbg,
4931  * they could still across block group boundary.
4932  *
4933  * Also account for superblock, inode, quota and xattr blocks
4934  */
4935 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4936 {
4937 	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4938 	int gdpblocks;
4939 	int idxblocks;
4940 	int ret = 0;
4941 
4942 	/*
4943 	 * How many index blocks need to touch to modify nrblocks?
4944 	 * The "Chunk" flag indicating whether the nrblocks is
4945 	 * physically contiguous on disk
4946 	 *
4947 	 * For Direct IO and fallocate, they calls get_block to allocate
4948 	 * one single extent at a time, so they could set the "Chunk" flag
4949 	 */
4950 	idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4951 
4952 	ret = idxblocks;
4953 
4954 	/*
4955 	 * Now let's see how many group bitmaps and group descriptors need
4956 	 * to account
4957 	 */
4958 	groups = idxblocks;
4959 	if (chunk)
4960 		groups += 1;
4961 	else
4962 		groups += nrblocks;
4963 
4964 	gdpblocks = groups;
4965 	if (groups > ngroups)
4966 		groups = ngroups;
4967 	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4968 		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4969 
4970 	/* bitmaps and block group descriptor blocks */
4971 	ret += groups + gdpblocks;
4972 
4973 	/* Blocks for super block, inode, quota and xattr blocks */
4974 	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4975 
4976 	return ret;
4977 }
4978 
4979 /*
4980  * Calulate the total number of credits to reserve to fit
4981  * the modification of a single pages into a single transaction,
4982  * which may include multiple chunks of block allocations.
4983  *
4984  * This could be called via ext4_write_begin()
4985  *
4986  * We need to consider the worse case, when
4987  * one new block per extent.
4988  */
4989 int ext4_writepage_trans_blocks(struct inode *inode)
4990 {
4991 	int bpp = ext4_journal_blocks_per_page(inode);
4992 	int ret;
4993 
4994 	ret = ext4_meta_trans_blocks(inode, bpp, 0);
4995 
4996 	/* Account for data blocks for journalled mode */
4997 	if (ext4_should_journal_data(inode))
4998 		ret += bpp;
4999 	return ret;
5000 }
5001 
5002 /*
5003  * Calculate the journal credits for a chunk of data modification.
5004  *
5005  * This is called from DIO, fallocate or whoever calling
5006  * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5007  *
5008  * journal buffers for data blocks are not included here, as DIO
5009  * and fallocate do no need to journal data buffers.
5010  */
5011 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5012 {
5013 	return ext4_meta_trans_blocks(inode, nrblocks, 1);
5014 }
5015 
5016 /*
5017  * The caller must have previously called ext4_reserve_inode_write().
5018  * Give this, we know that the caller already has write access to iloc->bh.
5019  */
5020 int ext4_mark_iloc_dirty(handle_t *handle,
5021 			 struct inode *inode, struct ext4_iloc *iloc)
5022 {
5023 	int err = 0;
5024 
5025 	if (test_opt(inode->i_sb, I_VERSION))
5026 		inode_inc_iversion(inode);
5027 
5028 	/* the do_update_inode consumes one bh->b_count */
5029 	get_bh(iloc->bh);
5030 
5031 	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5032 	err = ext4_do_update_inode(handle, inode, iloc, 0);
5033 	put_bh(iloc->bh);
5034 	return err;
5035 }
5036 
5037 /*
5038  * On success, We end up with an outstanding reference count against
5039  * iloc->bh.  This _must_ be cleaned up later.
5040  */
5041 
5042 int
5043 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5044 			 struct ext4_iloc *iloc)
5045 {
5046 	int err;
5047 
5048 	err = ext4_get_inode_loc(inode, iloc);
5049 	if (!err) {
5050 		BUFFER_TRACE(iloc->bh, "get_write_access");
5051 		err = ext4_journal_get_write_access(handle, iloc->bh);
5052 		if (err) {
5053 			brelse(iloc->bh);
5054 			iloc->bh = NULL;
5055 		}
5056 	}
5057 	ext4_std_error(inode->i_sb, err);
5058 	return err;
5059 }
5060 
5061 /*
5062  * Expand an inode by new_extra_isize bytes.
5063  * Returns 0 on success or negative error number on failure.
5064  */
5065 static int ext4_expand_extra_isize(struct inode *inode,
5066 				   unsigned int new_extra_isize,
5067 				   struct ext4_iloc iloc,
5068 				   handle_t *handle)
5069 {
5070 	struct ext4_inode *raw_inode;
5071 	struct ext4_xattr_ibody_header *header;
5072 	struct ext4_xattr_entry *entry;
5073 
5074 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5075 		return 0;
5076 
5077 	raw_inode = ext4_raw_inode(&iloc);
5078 
5079 	header = IHDR(inode, raw_inode);
5080 	entry = IFIRST(header);
5081 
5082 	/* No extended attributes present */
5083 	if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5084 		header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5085 		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5086 			new_extra_isize);
5087 		EXT4_I(inode)->i_extra_isize = new_extra_isize;
5088 		return 0;
5089 	}
5090 
5091 	/* try to expand with EAs present */
5092 	return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5093 					  raw_inode, handle);
5094 }
5095 
5096 /*
5097  * What we do here is to mark the in-core inode as clean with respect to inode
5098  * dirtiness (it may still be data-dirty).
5099  * This means that the in-core inode may be reaped by prune_icache
5100  * without having to perform any I/O.  This is a very good thing,
5101  * because *any* task may call prune_icache - even ones which
5102  * have a transaction open against a different journal.
5103  *
5104  * Is this cheating?  Not really.  Sure, we haven't written the
5105  * inode out, but prune_icache isn't a user-visible syncing function.
5106  * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5107  * we start and wait on commits.
5108  *
5109  * Is this efficient/effective?  Well, we're being nice to the system
5110  * by cleaning up our inodes proactively so they can be reaped
5111  * without I/O.  But we are potentially leaving up to five seconds'
5112  * worth of inodes floating about which prune_icache wants us to
5113  * write out.  One way to fix that would be to get prune_icache()
5114  * to do a write_super() to free up some memory.  It has the desired
5115  * effect.
5116  */
5117 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5118 {
5119 	struct ext4_iloc iloc;
5120 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5121 	static unsigned int mnt_count;
5122 	int err, ret;
5123 
5124 	might_sleep();
5125 	err = ext4_reserve_inode_write(handle, inode, &iloc);
5126 	if (ext4_handle_valid(handle) &&
5127 	    EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5128 	    !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5129 		/*
5130 		 * We need extra buffer credits since we may write into EA block
5131 		 * with this same handle. If journal_extend fails, then it will
5132 		 * only result in a minor loss of functionality for that inode.
5133 		 * If this is felt to be critical, then e2fsck should be run to
5134 		 * force a large enough s_min_extra_isize.
5135 		 */
5136 		if ((jbd2_journal_extend(handle,
5137 			     EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5138 			ret = ext4_expand_extra_isize(inode,
5139 						      sbi->s_want_extra_isize,
5140 						      iloc, handle);
5141 			if (ret) {
5142 				EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5143 				if (mnt_count !=
5144 					le16_to_cpu(sbi->s_es->s_mnt_count)) {
5145 					ext4_warning(inode->i_sb, __func__,
5146 					"Unable to expand inode %lu. Delete"
5147 					" some EAs or run e2fsck.",
5148 					inode->i_ino);
5149 					mnt_count =
5150 					  le16_to_cpu(sbi->s_es->s_mnt_count);
5151 				}
5152 			}
5153 		}
5154 	}
5155 	if (!err)
5156 		err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5157 	return err;
5158 }
5159 
5160 /*
5161  * ext4_dirty_inode() is called from __mark_inode_dirty()
5162  *
5163  * We're really interested in the case where a file is being extended.
5164  * i_size has been changed by generic_commit_write() and we thus need
5165  * to include the updated inode in the current transaction.
5166  *
5167  * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5168  * are allocated to the file.
5169  *
5170  * If the inode is marked synchronous, we don't honour that here - doing
5171  * so would cause a commit on atime updates, which we don't bother doing.
5172  * We handle synchronous inodes at the highest possible level.
5173  */
5174 void ext4_dirty_inode(struct inode *inode)
5175 {
5176 	handle_t *current_handle = ext4_journal_current_handle();
5177 	handle_t *handle;
5178 
5179 	if (!ext4_handle_valid(current_handle)) {
5180 		ext4_mark_inode_dirty(current_handle, inode);
5181 		return;
5182 	}
5183 
5184 	handle = ext4_journal_start(inode, 2);
5185 	if (IS_ERR(handle))
5186 		goto out;
5187 	if (current_handle &&
5188 		current_handle->h_transaction != handle->h_transaction) {
5189 		/* This task has a transaction open against a different fs */
5190 		printk(KERN_EMERG "%s: transactions do not match!\n",
5191 		       __func__);
5192 	} else {
5193 		jbd_debug(5, "marking dirty.  outer handle=%p\n",
5194 				current_handle);
5195 		ext4_mark_inode_dirty(handle, inode);
5196 	}
5197 	ext4_journal_stop(handle);
5198 out:
5199 	return;
5200 }
5201 
5202 #if 0
5203 /*
5204  * Bind an inode's backing buffer_head into this transaction, to prevent
5205  * it from being flushed to disk early.  Unlike
5206  * ext4_reserve_inode_write, this leaves behind no bh reference and
5207  * returns no iloc structure, so the caller needs to repeat the iloc
5208  * lookup to mark the inode dirty later.
5209  */
5210 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5211 {
5212 	struct ext4_iloc iloc;
5213 
5214 	int err = 0;
5215 	if (handle) {
5216 		err = ext4_get_inode_loc(inode, &iloc);
5217 		if (!err) {
5218 			BUFFER_TRACE(iloc.bh, "get_write_access");
5219 			err = jbd2_journal_get_write_access(handle, iloc.bh);
5220 			if (!err)
5221 				err = ext4_handle_dirty_metadata(handle,
5222 								 inode,
5223 								 iloc.bh);
5224 			brelse(iloc.bh);
5225 		}
5226 	}
5227 	ext4_std_error(inode->i_sb, err);
5228 	return err;
5229 }
5230 #endif
5231 
5232 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5233 {
5234 	journal_t *journal;
5235 	handle_t *handle;
5236 	int err;
5237 
5238 	/*
5239 	 * We have to be very careful here: changing a data block's
5240 	 * journaling status dynamically is dangerous.  If we write a
5241 	 * data block to the journal, change the status and then delete
5242 	 * that block, we risk forgetting to revoke the old log record
5243 	 * from the journal and so a subsequent replay can corrupt data.
5244 	 * So, first we make sure that the journal is empty and that
5245 	 * nobody is changing anything.
5246 	 */
5247 
5248 	journal = EXT4_JOURNAL(inode);
5249 	if (!journal)
5250 		return 0;
5251 	if (is_journal_aborted(journal))
5252 		return -EROFS;
5253 
5254 	jbd2_journal_lock_updates(journal);
5255 	jbd2_journal_flush(journal);
5256 
5257 	/*
5258 	 * OK, there are no updates running now, and all cached data is
5259 	 * synced to disk.  We are now in a completely consistent state
5260 	 * which doesn't have anything in the journal, and we know that
5261 	 * no filesystem updates are running, so it is safe to modify
5262 	 * the inode's in-core data-journaling state flag now.
5263 	 */
5264 
5265 	if (val)
5266 		EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5267 	else
5268 		EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5269 	ext4_set_aops(inode);
5270 
5271 	jbd2_journal_unlock_updates(journal);
5272 
5273 	/* Finally we can mark the inode as dirty. */
5274 
5275 	handle = ext4_journal_start(inode, 1);
5276 	if (IS_ERR(handle))
5277 		return PTR_ERR(handle);
5278 
5279 	err = ext4_mark_inode_dirty(handle, inode);
5280 	ext4_handle_sync(handle);
5281 	ext4_journal_stop(handle);
5282 	ext4_std_error(inode->i_sb, err);
5283 
5284 	return err;
5285 }
5286 
5287 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5288 {
5289 	return !buffer_mapped(bh);
5290 }
5291 
5292 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5293 {
5294 	struct page *page = vmf->page;
5295 	loff_t size;
5296 	unsigned long len;
5297 	int ret = -EINVAL;
5298 	void *fsdata;
5299 	struct file *file = vma->vm_file;
5300 	struct inode *inode = file->f_path.dentry->d_inode;
5301 	struct address_space *mapping = inode->i_mapping;
5302 
5303 	/*
5304 	 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5305 	 * get i_mutex because we are already holding mmap_sem.
5306 	 */
5307 	down_read(&inode->i_alloc_sem);
5308 	size = i_size_read(inode);
5309 	if (page->mapping != mapping || size <= page_offset(page)
5310 	    || !PageUptodate(page)) {
5311 		/* page got truncated from under us? */
5312 		goto out_unlock;
5313 	}
5314 	ret = 0;
5315 	if (PageMappedToDisk(page))
5316 		goto out_unlock;
5317 
5318 	if (page->index == size >> PAGE_CACHE_SHIFT)
5319 		len = size & ~PAGE_CACHE_MASK;
5320 	else
5321 		len = PAGE_CACHE_SIZE;
5322 
5323 	lock_page(page);
5324 	/*
5325 	 * return if we have all the buffers mapped. This avoid
5326 	 * the need to call write_begin/write_end which does a
5327 	 * journal_start/journal_stop which can block and take
5328 	 * long time
5329 	 */
5330 	if (page_has_buffers(page)) {
5331 		if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5332 					ext4_bh_unmapped)) {
5333 			unlock_page(page);
5334 			goto out_unlock;
5335 		}
5336 	}
5337 	unlock_page(page);
5338 	/*
5339 	 * OK, we need to fill the hole... Do write_begin write_end
5340 	 * to do block allocation/reservation.We are not holding
5341 	 * inode.i__mutex here. That allow * parallel write_begin,
5342 	 * write_end call. lock_page prevent this from happening
5343 	 * on the same page though
5344 	 */
5345 	ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5346 			len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5347 	if (ret < 0)
5348 		goto out_unlock;
5349 	ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5350 			len, len, page, fsdata);
5351 	if (ret < 0)
5352 		goto out_unlock;
5353 	ret = 0;
5354 out_unlock:
5355 	if (ret)
5356 		ret = VM_FAULT_SIGBUS;
5357 	up_read(&inode->i_alloc_sem);
5358 	return ret;
5359 }
5360