xref: /openbmc/linux/fs/ext4/inode.c (revision 2f828fb2)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/inode.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/inode.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  64-bit file support on 64-bit platforms by Jakub Jelinek
17  *	(jj@sunsite.ms.mff.cuni.cz)
18  *
19  *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
20  */
21 
22 #include <linux/fs.h>
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 
43 #include "ext4_jbd2.h"
44 #include "xattr.h"
45 #include "acl.h"
46 #include "truncate.h"
47 
48 #include <trace/events/ext4.h>
49 
50 #define MPAGE_DA_EXTENT_TAIL 0x01
51 
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 			      struct ext4_inode_info *ei)
54 {
55 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 	__u32 csum;
57 	__u16 dummy_csum = 0;
58 	int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 	unsigned int csum_size = sizeof(dummy_csum);
60 
61 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 	offset += csum_size;
64 	csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 			   EXT4_GOOD_OLD_INODE_SIZE - offset);
66 
67 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
68 		offset = offsetof(struct ext4_inode, i_checksum_hi);
69 		csum = ext4_chksum(sbi, csum, (__u8 *)raw +
70 				   EXT4_GOOD_OLD_INODE_SIZE,
71 				   offset - EXT4_GOOD_OLD_INODE_SIZE);
72 		if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
73 			csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
74 					   csum_size);
75 			offset += csum_size;
76 		}
77 		csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 				   EXT4_INODE_SIZE(inode->i_sb) - offset);
79 	}
80 
81 	return csum;
82 }
83 
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 				  struct ext4_inode_info *ei)
86 {
87 	__u32 provided, calculated;
88 
89 	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 	    cpu_to_le32(EXT4_OS_LINUX) ||
91 	    !ext4_has_metadata_csum(inode->i_sb))
92 		return 1;
93 
94 	provided = le16_to_cpu(raw->i_checksum_lo);
95 	calculated = ext4_inode_csum(inode, raw, ei);
96 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 		provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
99 	else
100 		calculated &= 0xFFFF;
101 
102 	return provided == calculated;
103 }
104 
105 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 				struct ext4_inode_info *ei)
107 {
108 	__u32 csum;
109 
110 	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 	    cpu_to_le32(EXT4_OS_LINUX) ||
112 	    !ext4_has_metadata_csum(inode->i_sb))
113 		return;
114 
115 	csum = ext4_inode_csum(inode, raw, ei);
116 	raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 		raw->i_checksum_hi = cpu_to_le16(csum >> 16);
120 }
121 
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
123 					      loff_t new_size)
124 {
125 	trace_ext4_begin_ordered_truncate(inode, new_size);
126 	/*
127 	 * If jinode is zero, then we never opened the file for
128 	 * writing, so there's no need to call
129 	 * jbd2_journal_begin_ordered_truncate() since there's no
130 	 * outstanding writes we need to flush.
131 	 */
132 	if (!EXT4_I(inode)->jinode)
133 		return 0;
134 	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 						   EXT4_I(inode)->jinode,
136 						   new_size);
137 }
138 
139 static void ext4_invalidatepage(struct page *page, unsigned int offset,
140 				unsigned int length);
141 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
143 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
144 				  int pextents);
145 
146 /*
147  * Test whether an inode is a fast symlink.
148  * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
149  */
150 int ext4_inode_is_fast_symlink(struct inode *inode)
151 {
152 	return S_ISLNK(inode->i_mode) && inode->i_size &&
153 	       (inode->i_size < EXT4_N_BLOCKS * 4);
154 }
155 
156 /*
157  * Restart the transaction associated with *handle.  This does a commit,
158  * so before we call here everything must be consistently dirtied against
159  * this transaction.
160  */
161 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
162 				 int nblocks)
163 {
164 	int ret;
165 
166 	/*
167 	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
168 	 * moment, get_block can be called only for blocks inside i_size since
169 	 * page cache has been already dropped and writes are blocked by
170 	 * i_mutex. So we can safely drop the i_data_sem here.
171 	 */
172 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
173 	jbd_debug(2, "restarting handle %p\n", handle);
174 	up_write(&EXT4_I(inode)->i_data_sem);
175 	ret = ext4_journal_restart(handle, nblocks);
176 	down_write(&EXT4_I(inode)->i_data_sem);
177 	ext4_discard_preallocations(inode);
178 
179 	return ret;
180 }
181 
182 /*
183  * Called at the last iput() if i_nlink is zero.
184  */
185 void ext4_evict_inode(struct inode *inode)
186 {
187 	handle_t *handle;
188 	int err;
189 	int extra_credits = 3;
190 	struct ext4_xattr_inode_array *ea_inode_array = NULL;
191 
192 	trace_ext4_evict_inode(inode);
193 
194 	if (inode->i_nlink) {
195 		/*
196 		 * When journalling data dirty buffers are tracked only in the
197 		 * journal. So although mm thinks everything is clean and
198 		 * ready for reaping the inode might still have some pages to
199 		 * write in the running transaction or waiting to be
200 		 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 		 * (via truncate_inode_pages()) to discard these buffers can
202 		 * cause data loss. Also even if we did not discard these
203 		 * buffers, we would have no way to find them after the inode
204 		 * is reaped and thus user could see stale data if he tries to
205 		 * read them before the transaction is checkpointed. So be
206 		 * careful and force everything to disk here... We use
207 		 * ei->i_datasync_tid to store the newest transaction
208 		 * containing inode's data.
209 		 *
210 		 * Note that directories do not have this problem because they
211 		 * don't use page cache.
212 		 */
213 		if (inode->i_ino != EXT4_JOURNAL_INO &&
214 		    ext4_should_journal_data(inode) &&
215 		    (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
216 		    inode->i_data.nrpages) {
217 			journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
218 			tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
219 
220 			jbd2_complete_transaction(journal, commit_tid);
221 			filemap_write_and_wait(&inode->i_data);
222 		}
223 		truncate_inode_pages_final(&inode->i_data);
224 
225 		goto no_delete;
226 	}
227 
228 	if (is_bad_inode(inode))
229 		goto no_delete;
230 	dquot_initialize(inode);
231 
232 	if (ext4_should_order_data(inode))
233 		ext4_begin_ordered_truncate(inode, 0);
234 	truncate_inode_pages_final(&inode->i_data);
235 
236 	/*
237 	 * Protect us against freezing - iput() caller didn't have to have any
238 	 * protection against it
239 	 */
240 	sb_start_intwrite(inode->i_sb);
241 
242 	if (!IS_NOQUOTA(inode))
243 		extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
244 
245 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
246 				 ext4_blocks_for_truncate(inode)+extra_credits);
247 	if (IS_ERR(handle)) {
248 		ext4_std_error(inode->i_sb, PTR_ERR(handle));
249 		/*
250 		 * If we're going to skip the normal cleanup, we still need to
251 		 * make sure that the in-core orphan linked list is properly
252 		 * cleaned up.
253 		 */
254 		ext4_orphan_del(NULL, inode);
255 		sb_end_intwrite(inode->i_sb);
256 		goto no_delete;
257 	}
258 
259 	if (IS_SYNC(inode))
260 		ext4_handle_sync(handle);
261 
262 	/*
263 	 * Set inode->i_size to 0 before calling ext4_truncate(). We need
264 	 * special handling of symlinks here because i_size is used to
265 	 * determine whether ext4_inode_info->i_data contains symlink data or
266 	 * block mappings. Setting i_size to 0 will remove its fast symlink
267 	 * status. Erase i_data so that it becomes a valid empty block map.
268 	 */
269 	if (ext4_inode_is_fast_symlink(inode))
270 		memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
271 	inode->i_size = 0;
272 	err = ext4_mark_inode_dirty(handle, inode);
273 	if (err) {
274 		ext4_warning(inode->i_sb,
275 			     "couldn't mark inode dirty (err %d)", err);
276 		goto stop_handle;
277 	}
278 	if (inode->i_blocks) {
279 		err = ext4_truncate(inode);
280 		if (err) {
281 			ext4_error(inode->i_sb,
282 				   "couldn't truncate inode %lu (err %d)",
283 				   inode->i_ino, err);
284 			goto stop_handle;
285 		}
286 	}
287 
288 	/* Remove xattr references. */
289 	err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
290 				      extra_credits);
291 	if (err) {
292 		ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
293 stop_handle:
294 		ext4_journal_stop(handle);
295 		ext4_orphan_del(NULL, inode);
296 		sb_end_intwrite(inode->i_sb);
297 		ext4_xattr_inode_array_free(ea_inode_array);
298 		goto no_delete;
299 	}
300 
301 	/*
302 	 * Kill off the orphan record which ext4_truncate created.
303 	 * AKPM: I think this can be inside the above `if'.
304 	 * Note that ext4_orphan_del() has to be able to cope with the
305 	 * deletion of a non-existent orphan - this is because we don't
306 	 * know if ext4_truncate() actually created an orphan record.
307 	 * (Well, we could do this if we need to, but heck - it works)
308 	 */
309 	ext4_orphan_del(handle, inode);
310 	EXT4_I(inode)->i_dtime	= get_seconds();
311 
312 	/*
313 	 * One subtle ordering requirement: if anything has gone wrong
314 	 * (transaction abort, IO errors, whatever), then we can still
315 	 * do these next steps (the fs will already have been marked as
316 	 * having errors), but we can't free the inode if the mark_dirty
317 	 * fails.
318 	 */
319 	if (ext4_mark_inode_dirty(handle, inode))
320 		/* If that failed, just do the required in-core inode clear. */
321 		ext4_clear_inode(inode);
322 	else
323 		ext4_free_inode(handle, inode);
324 	ext4_journal_stop(handle);
325 	sb_end_intwrite(inode->i_sb);
326 	ext4_xattr_inode_array_free(ea_inode_array);
327 	return;
328 no_delete:
329 	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
330 }
331 
332 #ifdef CONFIG_QUOTA
333 qsize_t *ext4_get_reserved_space(struct inode *inode)
334 {
335 	return &EXT4_I(inode)->i_reserved_quota;
336 }
337 #endif
338 
339 /*
340  * Called with i_data_sem down, which is important since we can call
341  * ext4_discard_preallocations() from here.
342  */
343 void ext4_da_update_reserve_space(struct inode *inode,
344 					int used, int quota_claim)
345 {
346 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
347 	struct ext4_inode_info *ei = EXT4_I(inode);
348 
349 	spin_lock(&ei->i_block_reservation_lock);
350 	trace_ext4_da_update_reserve_space(inode, used, quota_claim);
351 	if (unlikely(used > ei->i_reserved_data_blocks)) {
352 		ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
353 			 "with only %d reserved data blocks",
354 			 __func__, inode->i_ino, used,
355 			 ei->i_reserved_data_blocks);
356 		WARN_ON(1);
357 		used = ei->i_reserved_data_blocks;
358 	}
359 
360 	/* Update per-inode reservations */
361 	ei->i_reserved_data_blocks -= used;
362 	percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
363 
364 	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
365 
366 	/* Update quota subsystem for data blocks */
367 	if (quota_claim)
368 		dquot_claim_block(inode, EXT4_C2B(sbi, used));
369 	else {
370 		/*
371 		 * We did fallocate with an offset that is already delayed
372 		 * allocated. So on delayed allocated writeback we should
373 		 * not re-claim the quota for fallocated blocks.
374 		 */
375 		dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
376 	}
377 
378 	/*
379 	 * If we have done all the pending block allocations and if
380 	 * there aren't any writers on the inode, we can discard the
381 	 * inode's preallocations.
382 	 */
383 	if ((ei->i_reserved_data_blocks == 0) &&
384 	    (atomic_read(&inode->i_writecount) == 0))
385 		ext4_discard_preallocations(inode);
386 }
387 
388 static int __check_block_validity(struct inode *inode, const char *func,
389 				unsigned int line,
390 				struct ext4_map_blocks *map)
391 {
392 	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
393 				   map->m_len)) {
394 		ext4_error_inode(inode, func, line, map->m_pblk,
395 				 "lblock %lu mapped to illegal pblock "
396 				 "(length %d)", (unsigned long) map->m_lblk,
397 				 map->m_len);
398 		return -EFSCORRUPTED;
399 	}
400 	return 0;
401 }
402 
403 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
404 		       ext4_lblk_t len)
405 {
406 	int ret;
407 
408 	if (ext4_encrypted_inode(inode))
409 		return fscrypt_zeroout_range(inode, lblk, pblk, len);
410 
411 	ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
412 	if (ret > 0)
413 		ret = 0;
414 
415 	return ret;
416 }
417 
418 #define check_block_validity(inode, map)	\
419 	__check_block_validity((inode), __func__, __LINE__, (map))
420 
421 #ifdef ES_AGGRESSIVE_TEST
422 static void ext4_map_blocks_es_recheck(handle_t *handle,
423 				       struct inode *inode,
424 				       struct ext4_map_blocks *es_map,
425 				       struct ext4_map_blocks *map,
426 				       int flags)
427 {
428 	int retval;
429 
430 	map->m_flags = 0;
431 	/*
432 	 * There is a race window that the result is not the same.
433 	 * e.g. xfstests #223 when dioread_nolock enables.  The reason
434 	 * is that we lookup a block mapping in extent status tree with
435 	 * out taking i_data_sem.  So at the time the unwritten extent
436 	 * could be converted.
437 	 */
438 	down_read(&EXT4_I(inode)->i_data_sem);
439 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
440 		retval = ext4_ext_map_blocks(handle, inode, map, flags &
441 					     EXT4_GET_BLOCKS_KEEP_SIZE);
442 	} else {
443 		retval = ext4_ind_map_blocks(handle, inode, map, flags &
444 					     EXT4_GET_BLOCKS_KEEP_SIZE);
445 	}
446 	up_read((&EXT4_I(inode)->i_data_sem));
447 
448 	/*
449 	 * We don't check m_len because extent will be collpased in status
450 	 * tree.  So the m_len might not equal.
451 	 */
452 	if (es_map->m_lblk != map->m_lblk ||
453 	    es_map->m_flags != map->m_flags ||
454 	    es_map->m_pblk != map->m_pblk) {
455 		printk("ES cache assertion failed for inode: %lu "
456 		       "es_cached ex [%d/%d/%llu/%x] != "
457 		       "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
458 		       inode->i_ino, es_map->m_lblk, es_map->m_len,
459 		       es_map->m_pblk, es_map->m_flags, map->m_lblk,
460 		       map->m_len, map->m_pblk, map->m_flags,
461 		       retval, flags);
462 	}
463 }
464 #endif /* ES_AGGRESSIVE_TEST */
465 
466 /*
467  * The ext4_map_blocks() function tries to look up the requested blocks,
468  * and returns if the blocks are already mapped.
469  *
470  * Otherwise it takes the write lock of the i_data_sem and allocate blocks
471  * and store the allocated blocks in the result buffer head and mark it
472  * mapped.
473  *
474  * If file type is extents based, it will call ext4_ext_map_blocks(),
475  * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
476  * based files
477  *
478  * On success, it returns the number of blocks being mapped or allocated.  if
479  * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
480  * is marked as unwritten. If the create == 1, it will mark @map as mapped.
481  *
482  * It returns 0 if plain look up failed (blocks have not been allocated), in
483  * that case, @map is returned as unmapped but we still do fill map->m_len to
484  * indicate the length of a hole starting at map->m_lblk.
485  *
486  * It returns the error in case of allocation failure.
487  */
488 int ext4_map_blocks(handle_t *handle, struct inode *inode,
489 		    struct ext4_map_blocks *map, int flags)
490 {
491 	struct extent_status es;
492 	int retval;
493 	int ret = 0;
494 #ifdef ES_AGGRESSIVE_TEST
495 	struct ext4_map_blocks orig_map;
496 
497 	memcpy(&orig_map, map, sizeof(*map));
498 #endif
499 
500 	map->m_flags = 0;
501 	ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
502 		  "logical block %lu\n", inode->i_ino, flags, map->m_len,
503 		  (unsigned long) map->m_lblk);
504 
505 	/*
506 	 * ext4_map_blocks returns an int, and m_len is an unsigned int
507 	 */
508 	if (unlikely(map->m_len > INT_MAX))
509 		map->m_len = INT_MAX;
510 
511 	/* We can handle the block number less than EXT_MAX_BLOCKS */
512 	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
513 		return -EFSCORRUPTED;
514 
515 	/* Lookup extent status tree firstly */
516 	if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
517 		if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
518 			map->m_pblk = ext4_es_pblock(&es) +
519 					map->m_lblk - es.es_lblk;
520 			map->m_flags |= ext4_es_is_written(&es) ?
521 					EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
522 			retval = es.es_len - (map->m_lblk - es.es_lblk);
523 			if (retval > map->m_len)
524 				retval = map->m_len;
525 			map->m_len = retval;
526 		} else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
527 			map->m_pblk = 0;
528 			retval = es.es_len - (map->m_lblk - es.es_lblk);
529 			if (retval > map->m_len)
530 				retval = map->m_len;
531 			map->m_len = retval;
532 			retval = 0;
533 		} else {
534 			BUG_ON(1);
535 		}
536 #ifdef ES_AGGRESSIVE_TEST
537 		ext4_map_blocks_es_recheck(handle, inode, map,
538 					   &orig_map, flags);
539 #endif
540 		goto found;
541 	}
542 
543 	/*
544 	 * Try to see if we can get the block without requesting a new
545 	 * file system block.
546 	 */
547 	down_read(&EXT4_I(inode)->i_data_sem);
548 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
549 		retval = ext4_ext_map_blocks(handle, inode, map, flags &
550 					     EXT4_GET_BLOCKS_KEEP_SIZE);
551 	} else {
552 		retval = ext4_ind_map_blocks(handle, inode, map, flags &
553 					     EXT4_GET_BLOCKS_KEEP_SIZE);
554 	}
555 	if (retval > 0) {
556 		unsigned int status;
557 
558 		if (unlikely(retval != map->m_len)) {
559 			ext4_warning(inode->i_sb,
560 				     "ES len assertion failed for inode "
561 				     "%lu: retval %d != map->m_len %d",
562 				     inode->i_ino, retval, map->m_len);
563 			WARN_ON(1);
564 		}
565 
566 		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
567 				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
568 		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
569 		    !(status & EXTENT_STATUS_WRITTEN) &&
570 		    ext4_find_delalloc_range(inode, map->m_lblk,
571 					     map->m_lblk + map->m_len - 1))
572 			status |= EXTENT_STATUS_DELAYED;
573 		ret = ext4_es_insert_extent(inode, map->m_lblk,
574 					    map->m_len, map->m_pblk, status);
575 		if (ret < 0)
576 			retval = ret;
577 	}
578 	up_read((&EXT4_I(inode)->i_data_sem));
579 
580 found:
581 	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
582 		ret = check_block_validity(inode, map);
583 		if (ret != 0)
584 			return ret;
585 	}
586 
587 	/* If it is only a block(s) look up */
588 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
589 		return retval;
590 
591 	/*
592 	 * Returns if the blocks have already allocated
593 	 *
594 	 * Note that if blocks have been preallocated
595 	 * ext4_ext_get_block() returns the create = 0
596 	 * with buffer head unmapped.
597 	 */
598 	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
599 		/*
600 		 * If we need to convert extent to unwritten
601 		 * we continue and do the actual work in
602 		 * ext4_ext_map_blocks()
603 		 */
604 		if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
605 			return retval;
606 
607 	/*
608 	 * Here we clear m_flags because after allocating an new extent,
609 	 * it will be set again.
610 	 */
611 	map->m_flags &= ~EXT4_MAP_FLAGS;
612 
613 	/*
614 	 * New blocks allocate and/or writing to unwritten extent
615 	 * will possibly result in updating i_data, so we take
616 	 * the write lock of i_data_sem, and call get_block()
617 	 * with create == 1 flag.
618 	 */
619 	down_write(&EXT4_I(inode)->i_data_sem);
620 
621 	/*
622 	 * We need to check for EXT4 here because migrate
623 	 * could have changed the inode type in between
624 	 */
625 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
626 		retval = ext4_ext_map_blocks(handle, inode, map, flags);
627 	} else {
628 		retval = ext4_ind_map_blocks(handle, inode, map, flags);
629 
630 		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
631 			/*
632 			 * We allocated new blocks which will result in
633 			 * i_data's format changing.  Force the migrate
634 			 * to fail by clearing migrate flags
635 			 */
636 			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
637 		}
638 
639 		/*
640 		 * Update reserved blocks/metadata blocks after successful
641 		 * block allocation which had been deferred till now. We don't
642 		 * support fallocate for non extent files. So we can update
643 		 * reserve space here.
644 		 */
645 		if ((retval > 0) &&
646 			(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
647 			ext4_da_update_reserve_space(inode, retval, 1);
648 	}
649 
650 	if (retval > 0) {
651 		unsigned int status;
652 
653 		if (unlikely(retval != map->m_len)) {
654 			ext4_warning(inode->i_sb,
655 				     "ES len assertion failed for inode "
656 				     "%lu: retval %d != map->m_len %d",
657 				     inode->i_ino, retval, map->m_len);
658 			WARN_ON(1);
659 		}
660 
661 		/*
662 		 * We have to zeroout blocks before inserting them into extent
663 		 * status tree. Otherwise someone could look them up there and
664 		 * use them before they are really zeroed. We also have to
665 		 * unmap metadata before zeroing as otherwise writeback can
666 		 * overwrite zeros with stale data from block device.
667 		 */
668 		if (flags & EXT4_GET_BLOCKS_ZERO &&
669 		    map->m_flags & EXT4_MAP_MAPPED &&
670 		    map->m_flags & EXT4_MAP_NEW) {
671 			clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
672 					   map->m_len);
673 			ret = ext4_issue_zeroout(inode, map->m_lblk,
674 						 map->m_pblk, map->m_len);
675 			if (ret) {
676 				retval = ret;
677 				goto out_sem;
678 			}
679 		}
680 
681 		/*
682 		 * If the extent has been zeroed out, we don't need to update
683 		 * extent status tree.
684 		 */
685 		if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
686 		    ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
687 			if (ext4_es_is_written(&es))
688 				goto out_sem;
689 		}
690 		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
691 				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
692 		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
693 		    !(status & EXTENT_STATUS_WRITTEN) &&
694 		    ext4_find_delalloc_range(inode, map->m_lblk,
695 					     map->m_lblk + map->m_len - 1))
696 			status |= EXTENT_STATUS_DELAYED;
697 		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
698 					    map->m_pblk, status);
699 		if (ret < 0) {
700 			retval = ret;
701 			goto out_sem;
702 		}
703 	}
704 
705 out_sem:
706 	up_write((&EXT4_I(inode)->i_data_sem));
707 	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
708 		ret = check_block_validity(inode, map);
709 		if (ret != 0)
710 			return ret;
711 
712 		/*
713 		 * Inodes with freshly allocated blocks where contents will be
714 		 * visible after transaction commit must be on transaction's
715 		 * ordered data list.
716 		 */
717 		if (map->m_flags & EXT4_MAP_NEW &&
718 		    !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
719 		    !(flags & EXT4_GET_BLOCKS_ZERO) &&
720 		    !ext4_is_quota_file(inode) &&
721 		    ext4_should_order_data(inode)) {
722 			if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
723 				ret = ext4_jbd2_inode_add_wait(handle, inode);
724 			else
725 				ret = ext4_jbd2_inode_add_write(handle, inode);
726 			if (ret)
727 				return ret;
728 		}
729 	}
730 	return retval;
731 }
732 
733 /*
734  * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
735  * we have to be careful as someone else may be manipulating b_state as well.
736  */
737 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
738 {
739 	unsigned long old_state;
740 	unsigned long new_state;
741 
742 	flags &= EXT4_MAP_FLAGS;
743 
744 	/* Dummy buffer_head? Set non-atomically. */
745 	if (!bh->b_page) {
746 		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
747 		return;
748 	}
749 	/*
750 	 * Someone else may be modifying b_state. Be careful! This is ugly but
751 	 * once we get rid of using bh as a container for mapping information
752 	 * to pass to / from get_block functions, this can go away.
753 	 */
754 	do {
755 		old_state = READ_ONCE(bh->b_state);
756 		new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
757 	} while (unlikely(
758 		 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
759 }
760 
761 static int _ext4_get_block(struct inode *inode, sector_t iblock,
762 			   struct buffer_head *bh, int flags)
763 {
764 	struct ext4_map_blocks map;
765 	int ret = 0;
766 
767 	if (ext4_has_inline_data(inode))
768 		return -ERANGE;
769 
770 	map.m_lblk = iblock;
771 	map.m_len = bh->b_size >> inode->i_blkbits;
772 
773 	ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
774 			      flags);
775 	if (ret > 0) {
776 		map_bh(bh, inode->i_sb, map.m_pblk);
777 		ext4_update_bh_state(bh, map.m_flags);
778 		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
779 		ret = 0;
780 	} else if (ret == 0) {
781 		/* hole case, need to fill in bh->b_size */
782 		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
783 	}
784 	return ret;
785 }
786 
787 int ext4_get_block(struct inode *inode, sector_t iblock,
788 		   struct buffer_head *bh, int create)
789 {
790 	return _ext4_get_block(inode, iblock, bh,
791 			       create ? EXT4_GET_BLOCKS_CREATE : 0);
792 }
793 
794 /*
795  * Get block function used when preparing for buffered write if we require
796  * creating an unwritten extent if blocks haven't been allocated.  The extent
797  * will be converted to written after the IO is complete.
798  */
799 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
800 			     struct buffer_head *bh_result, int create)
801 {
802 	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
803 		   inode->i_ino, create);
804 	return _ext4_get_block(inode, iblock, bh_result,
805 			       EXT4_GET_BLOCKS_IO_CREATE_EXT);
806 }
807 
808 /* Maximum number of blocks we map for direct IO at once. */
809 #define DIO_MAX_BLOCKS 4096
810 
811 /*
812  * Get blocks function for the cases that need to start a transaction -
813  * generally difference cases of direct IO and DAX IO. It also handles retries
814  * in case of ENOSPC.
815  */
816 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
817 				struct buffer_head *bh_result, int flags)
818 {
819 	int dio_credits;
820 	handle_t *handle;
821 	int retries = 0;
822 	int ret;
823 
824 	/* Trim mapping request to maximum we can map at once for DIO */
825 	if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
826 		bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
827 	dio_credits = ext4_chunk_trans_blocks(inode,
828 				      bh_result->b_size >> inode->i_blkbits);
829 retry:
830 	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
831 	if (IS_ERR(handle))
832 		return PTR_ERR(handle);
833 
834 	ret = _ext4_get_block(inode, iblock, bh_result, flags);
835 	ext4_journal_stop(handle);
836 
837 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
838 		goto retry;
839 	return ret;
840 }
841 
842 /* Get block function for DIO reads and writes to inodes without extents */
843 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
844 		       struct buffer_head *bh, int create)
845 {
846 	/* We don't expect handle for direct IO */
847 	WARN_ON_ONCE(ext4_journal_current_handle());
848 
849 	if (!create)
850 		return _ext4_get_block(inode, iblock, bh, 0);
851 	return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
852 }
853 
854 /*
855  * Get block function for AIO DIO writes when we create unwritten extent if
856  * blocks are not allocated yet. The extent will be converted to written
857  * after IO is complete.
858  */
859 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
860 		sector_t iblock, struct buffer_head *bh_result,	int create)
861 {
862 	int ret;
863 
864 	/* We don't expect handle for direct IO */
865 	WARN_ON_ONCE(ext4_journal_current_handle());
866 
867 	ret = ext4_get_block_trans(inode, iblock, bh_result,
868 				   EXT4_GET_BLOCKS_IO_CREATE_EXT);
869 
870 	/*
871 	 * When doing DIO using unwritten extents, we need io_end to convert
872 	 * unwritten extents to written on IO completion. We allocate io_end
873 	 * once we spot unwritten extent and store it in b_private. Generic
874 	 * DIO code keeps b_private set and furthermore passes the value to
875 	 * our completion callback in 'private' argument.
876 	 */
877 	if (!ret && buffer_unwritten(bh_result)) {
878 		if (!bh_result->b_private) {
879 			ext4_io_end_t *io_end;
880 
881 			io_end = ext4_init_io_end(inode, GFP_KERNEL);
882 			if (!io_end)
883 				return -ENOMEM;
884 			bh_result->b_private = io_end;
885 			ext4_set_io_unwritten_flag(inode, io_end);
886 		}
887 		set_buffer_defer_completion(bh_result);
888 	}
889 
890 	return ret;
891 }
892 
893 /*
894  * Get block function for non-AIO DIO writes when we create unwritten extent if
895  * blocks are not allocated yet. The extent will be converted to written
896  * after IO is complete by ext4_direct_IO_write().
897  */
898 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
899 		sector_t iblock, struct buffer_head *bh_result,	int create)
900 {
901 	int ret;
902 
903 	/* We don't expect handle for direct IO */
904 	WARN_ON_ONCE(ext4_journal_current_handle());
905 
906 	ret = ext4_get_block_trans(inode, iblock, bh_result,
907 				   EXT4_GET_BLOCKS_IO_CREATE_EXT);
908 
909 	/*
910 	 * Mark inode as having pending DIO writes to unwritten extents.
911 	 * ext4_direct_IO_write() checks this flag and converts extents to
912 	 * written.
913 	 */
914 	if (!ret && buffer_unwritten(bh_result))
915 		ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
916 
917 	return ret;
918 }
919 
920 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
921 		   struct buffer_head *bh_result, int create)
922 {
923 	int ret;
924 
925 	ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
926 		   inode->i_ino, create);
927 	/* We don't expect handle for direct IO */
928 	WARN_ON_ONCE(ext4_journal_current_handle());
929 
930 	ret = _ext4_get_block(inode, iblock, bh_result, 0);
931 	/*
932 	 * Blocks should have been preallocated! ext4_file_write_iter() checks
933 	 * that.
934 	 */
935 	WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
936 
937 	return ret;
938 }
939 
940 
941 /*
942  * `handle' can be NULL if create is zero
943  */
944 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
945 				ext4_lblk_t block, int map_flags)
946 {
947 	struct ext4_map_blocks map;
948 	struct buffer_head *bh;
949 	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
950 	int err;
951 
952 	J_ASSERT(handle != NULL || create == 0);
953 
954 	map.m_lblk = block;
955 	map.m_len = 1;
956 	err = ext4_map_blocks(handle, inode, &map, map_flags);
957 
958 	if (err == 0)
959 		return create ? ERR_PTR(-ENOSPC) : NULL;
960 	if (err < 0)
961 		return ERR_PTR(err);
962 
963 	bh = sb_getblk(inode->i_sb, map.m_pblk);
964 	if (unlikely(!bh))
965 		return ERR_PTR(-ENOMEM);
966 	if (map.m_flags & EXT4_MAP_NEW) {
967 		J_ASSERT(create != 0);
968 		J_ASSERT(handle != NULL);
969 
970 		/*
971 		 * Now that we do not always journal data, we should
972 		 * keep in mind whether this should always journal the
973 		 * new buffer as metadata.  For now, regular file
974 		 * writes use ext4_get_block instead, so it's not a
975 		 * problem.
976 		 */
977 		lock_buffer(bh);
978 		BUFFER_TRACE(bh, "call get_create_access");
979 		err = ext4_journal_get_create_access(handle, bh);
980 		if (unlikely(err)) {
981 			unlock_buffer(bh);
982 			goto errout;
983 		}
984 		if (!buffer_uptodate(bh)) {
985 			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
986 			set_buffer_uptodate(bh);
987 		}
988 		unlock_buffer(bh);
989 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
990 		err = ext4_handle_dirty_metadata(handle, inode, bh);
991 		if (unlikely(err))
992 			goto errout;
993 	} else
994 		BUFFER_TRACE(bh, "not a new buffer");
995 	return bh;
996 errout:
997 	brelse(bh);
998 	return ERR_PTR(err);
999 }
1000 
1001 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1002 			       ext4_lblk_t block, int map_flags)
1003 {
1004 	struct buffer_head *bh;
1005 
1006 	bh = ext4_getblk(handle, inode, block, map_flags);
1007 	if (IS_ERR(bh))
1008 		return bh;
1009 	if (!bh || buffer_uptodate(bh))
1010 		return bh;
1011 	ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1012 	wait_on_buffer(bh);
1013 	if (buffer_uptodate(bh))
1014 		return bh;
1015 	put_bh(bh);
1016 	return ERR_PTR(-EIO);
1017 }
1018 
1019 /* Read a contiguous batch of blocks. */
1020 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1021 		     bool wait, struct buffer_head **bhs)
1022 {
1023 	int i, err;
1024 
1025 	for (i = 0; i < bh_count; i++) {
1026 		bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1027 		if (IS_ERR(bhs[i])) {
1028 			err = PTR_ERR(bhs[i]);
1029 			bh_count = i;
1030 			goto out_brelse;
1031 		}
1032 	}
1033 
1034 	for (i = 0; i < bh_count; i++)
1035 		/* Note that NULL bhs[i] is valid because of holes. */
1036 		if (bhs[i] && !buffer_uptodate(bhs[i]))
1037 			ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1038 				    &bhs[i]);
1039 
1040 	if (!wait)
1041 		return 0;
1042 
1043 	for (i = 0; i < bh_count; i++)
1044 		if (bhs[i])
1045 			wait_on_buffer(bhs[i]);
1046 
1047 	for (i = 0; i < bh_count; i++) {
1048 		if (bhs[i] && !buffer_uptodate(bhs[i])) {
1049 			err = -EIO;
1050 			goto out_brelse;
1051 		}
1052 	}
1053 	return 0;
1054 
1055 out_brelse:
1056 	for (i = 0; i < bh_count; i++) {
1057 		brelse(bhs[i]);
1058 		bhs[i] = NULL;
1059 	}
1060 	return err;
1061 }
1062 
1063 int ext4_walk_page_buffers(handle_t *handle,
1064 			   struct buffer_head *head,
1065 			   unsigned from,
1066 			   unsigned to,
1067 			   int *partial,
1068 			   int (*fn)(handle_t *handle,
1069 				     struct buffer_head *bh))
1070 {
1071 	struct buffer_head *bh;
1072 	unsigned block_start, block_end;
1073 	unsigned blocksize = head->b_size;
1074 	int err, ret = 0;
1075 	struct buffer_head *next;
1076 
1077 	for (bh = head, block_start = 0;
1078 	     ret == 0 && (bh != head || !block_start);
1079 	     block_start = block_end, bh = next) {
1080 		next = bh->b_this_page;
1081 		block_end = block_start + blocksize;
1082 		if (block_end <= from || block_start >= to) {
1083 			if (partial && !buffer_uptodate(bh))
1084 				*partial = 1;
1085 			continue;
1086 		}
1087 		err = (*fn)(handle, bh);
1088 		if (!ret)
1089 			ret = err;
1090 	}
1091 	return ret;
1092 }
1093 
1094 /*
1095  * To preserve ordering, it is essential that the hole instantiation and
1096  * the data write be encapsulated in a single transaction.  We cannot
1097  * close off a transaction and start a new one between the ext4_get_block()
1098  * and the commit_write().  So doing the jbd2_journal_start at the start of
1099  * prepare_write() is the right place.
1100  *
1101  * Also, this function can nest inside ext4_writepage().  In that case, we
1102  * *know* that ext4_writepage() has generated enough buffer credits to do the
1103  * whole page.  So we won't block on the journal in that case, which is good,
1104  * because the caller may be PF_MEMALLOC.
1105  *
1106  * By accident, ext4 can be reentered when a transaction is open via
1107  * quota file writes.  If we were to commit the transaction while thus
1108  * reentered, there can be a deadlock - we would be holding a quota
1109  * lock, and the commit would never complete if another thread had a
1110  * transaction open and was blocking on the quota lock - a ranking
1111  * violation.
1112  *
1113  * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1114  * will _not_ run commit under these circumstances because handle->h_ref
1115  * is elevated.  We'll still have enough credits for the tiny quotafile
1116  * write.
1117  */
1118 int do_journal_get_write_access(handle_t *handle,
1119 				struct buffer_head *bh)
1120 {
1121 	int dirty = buffer_dirty(bh);
1122 	int ret;
1123 
1124 	if (!buffer_mapped(bh) || buffer_freed(bh))
1125 		return 0;
1126 	/*
1127 	 * __block_write_begin() could have dirtied some buffers. Clean
1128 	 * the dirty bit as jbd2_journal_get_write_access() could complain
1129 	 * otherwise about fs integrity issues. Setting of the dirty bit
1130 	 * by __block_write_begin() isn't a real problem here as we clear
1131 	 * the bit before releasing a page lock and thus writeback cannot
1132 	 * ever write the buffer.
1133 	 */
1134 	if (dirty)
1135 		clear_buffer_dirty(bh);
1136 	BUFFER_TRACE(bh, "get write access");
1137 	ret = ext4_journal_get_write_access(handle, bh);
1138 	if (!ret && dirty)
1139 		ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1140 	return ret;
1141 }
1142 
1143 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1144 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1145 				  get_block_t *get_block)
1146 {
1147 	unsigned from = pos & (PAGE_SIZE - 1);
1148 	unsigned to = from + len;
1149 	struct inode *inode = page->mapping->host;
1150 	unsigned block_start, block_end;
1151 	sector_t block;
1152 	int err = 0;
1153 	unsigned blocksize = inode->i_sb->s_blocksize;
1154 	unsigned bbits;
1155 	struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1156 	bool decrypt = false;
1157 
1158 	BUG_ON(!PageLocked(page));
1159 	BUG_ON(from > PAGE_SIZE);
1160 	BUG_ON(to > PAGE_SIZE);
1161 	BUG_ON(from > to);
1162 
1163 	if (!page_has_buffers(page))
1164 		create_empty_buffers(page, blocksize, 0);
1165 	head = page_buffers(page);
1166 	bbits = ilog2(blocksize);
1167 	block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1168 
1169 	for (bh = head, block_start = 0; bh != head || !block_start;
1170 	    block++, block_start = block_end, bh = bh->b_this_page) {
1171 		block_end = block_start + blocksize;
1172 		if (block_end <= from || block_start >= to) {
1173 			if (PageUptodate(page)) {
1174 				if (!buffer_uptodate(bh))
1175 					set_buffer_uptodate(bh);
1176 			}
1177 			continue;
1178 		}
1179 		if (buffer_new(bh))
1180 			clear_buffer_new(bh);
1181 		if (!buffer_mapped(bh)) {
1182 			WARN_ON(bh->b_size != blocksize);
1183 			err = get_block(inode, block, bh, 1);
1184 			if (err)
1185 				break;
1186 			if (buffer_new(bh)) {
1187 				clean_bdev_bh_alias(bh);
1188 				if (PageUptodate(page)) {
1189 					clear_buffer_new(bh);
1190 					set_buffer_uptodate(bh);
1191 					mark_buffer_dirty(bh);
1192 					continue;
1193 				}
1194 				if (block_end > to || block_start < from)
1195 					zero_user_segments(page, to, block_end,
1196 							   block_start, from);
1197 				continue;
1198 			}
1199 		}
1200 		if (PageUptodate(page)) {
1201 			if (!buffer_uptodate(bh))
1202 				set_buffer_uptodate(bh);
1203 			continue;
1204 		}
1205 		if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1206 		    !buffer_unwritten(bh) &&
1207 		    (block_start < from || block_end > to)) {
1208 			ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1209 			*wait_bh++ = bh;
1210 			decrypt = ext4_encrypted_inode(inode) &&
1211 				S_ISREG(inode->i_mode);
1212 		}
1213 	}
1214 	/*
1215 	 * If we issued read requests, let them complete.
1216 	 */
1217 	while (wait_bh > wait) {
1218 		wait_on_buffer(*--wait_bh);
1219 		if (!buffer_uptodate(*wait_bh))
1220 			err = -EIO;
1221 	}
1222 	if (unlikely(err))
1223 		page_zero_new_buffers(page, from, to);
1224 	else if (decrypt)
1225 		err = fscrypt_decrypt_page(page->mapping->host, page,
1226 				PAGE_SIZE, 0, page->index);
1227 	return err;
1228 }
1229 #endif
1230 
1231 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1232 			    loff_t pos, unsigned len, unsigned flags,
1233 			    struct page **pagep, void **fsdata)
1234 {
1235 	struct inode *inode = mapping->host;
1236 	int ret, needed_blocks;
1237 	handle_t *handle;
1238 	int retries = 0;
1239 	struct page *page;
1240 	pgoff_t index;
1241 	unsigned from, to;
1242 
1243 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1244 		return -EIO;
1245 
1246 	trace_ext4_write_begin(inode, pos, len, flags);
1247 	/*
1248 	 * Reserve one block more for addition to orphan list in case
1249 	 * we allocate blocks but write fails for some reason
1250 	 */
1251 	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1252 	index = pos >> PAGE_SHIFT;
1253 	from = pos & (PAGE_SIZE - 1);
1254 	to = from + len;
1255 
1256 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1257 		ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1258 						    flags, pagep);
1259 		if (ret < 0)
1260 			return ret;
1261 		if (ret == 1)
1262 			return 0;
1263 	}
1264 
1265 	/*
1266 	 * grab_cache_page_write_begin() can take a long time if the
1267 	 * system is thrashing due to memory pressure, or if the page
1268 	 * is being written back.  So grab it first before we start
1269 	 * the transaction handle.  This also allows us to allocate
1270 	 * the page (if needed) without using GFP_NOFS.
1271 	 */
1272 retry_grab:
1273 	page = grab_cache_page_write_begin(mapping, index, flags);
1274 	if (!page)
1275 		return -ENOMEM;
1276 	unlock_page(page);
1277 
1278 retry_journal:
1279 	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1280 	if (IS_ERR(handle)) {
1281 		put_page(page);
1282 		return PTR_ERR(handle);
1283 	}
1284 
1285 	lock_page(page);
1286 	if (page->mapping != mapping) {
1287 		/* The page got truncated from under us */
1288 		unlock_page(page);
1289 		put_page(page);
1290 		ext4_journal_stop(handle);
1291 		goto retry_grab;
1292 	}
1293 	/* In case writeback began while the page was unlocked */
1294 	wait_for_stable_page(page);
1295 
1296 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1297 	if (ext4_should_dioread_nolock(inode))
1298 		ret = ext4_block_write_begin(page, pos, len,
1299 					     ext4_get_block_unwritten);
1300 	else
1301 		ret = ext4_block_write_begin(page, pos, len,
1302 					     ext4_get_block);
1303 #else
1304 	if (ext4_should_dioread_nolock(inode))
1305 		ret = __block_write_begin(page, pos, len,
1306 					  ext4_get_block_unwritten);
1307 	else
1308 		ret = __block_write_begin(page, pos, len, ext4_get_block);
1309 #endif
1310 	if (!ret && ext4_should_journal_data(inode)) {
1311 		ret = ext4_walk_page_buffers(handle, page_buffers(page),
1312 					     from, to, NULL,
1313 					     do_journal_get_write_access);
1314 	}
1315 
1316 	if (ret) {
1317 		unlock_page(page);
1318 		/*
1319 		 * __block_write_begin may have instantiated a few blocks
1320 		 * outside i_size.  Trim these off again. Don't need
1321 		 * i_size_read because we hold i_mutex.
1322 		 *
1323 		 * Add inode to orphan list in case we crash before
1324 		 * truncate finishes
1325 		 */
1326 		if (pos + len > inode->i_size && ext4_can_truncate(inode))
1327 			ext4_orphan_add(handle, inode);
1328 
1329 		ext4_journal_stop(handle);
1330 		if (pos + len > inode->i_size) {
1331 			ext4_truncate_failed_write(inode);
1332 			/*
1333 			 * If truncate failed early the inode might
1334 			 * still be on the orphan list; we need to
1335 			 * make sure the inode is removed from the
1336 			 * orphan list in that case.
1337 			 */
1338 			if (inode->i_nlink)
1339 				ext4_orphan_del(NULL, inode);
1340 		}
1341 
1342 		if (ret == -ENOSPC &&
1343 		    ext4_should_retry_alloc(inode->i_sb, &retries))
1344 			goto retry_journal;
1345 		put_page(page);
1346 		return ret;
1347 	}
1348 	*pagep = page;
1349 	return ret;
1350 }
1351 
1352 /* For write_end() in data=journal mode */
1353 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1354 {
1355 	int ret;
1356 	if (!buffer_mapped(bh) || buffer_freed(bh))
1357 		return 0;
1358 	set_buffer_uptodate(bh);
1359 	ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1360 	clear_buffer_meta(bh);
1361 	clear_buffer_prio(bh);
1362 	return ret;
1363 }
1364 
1365 /*
1366  * We need to pick up the new inode size which generic_commit_write gave us
1367  * `file' can be NULL - eg, when called from page_symlink().
1368  *
1369  * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1370  * buffers are managed internally.
1371  */
1372 static int ext4_write_end(struct file *file,
1373 			  struct address_space *mapping,
1374 			  loff_t pos, unsigned len, unsigned copied,
1375 			  struct page *page, void *fsdata)
1376 {
1377 	handle_t *handle = ext4_journal_current_handle();
1378 	struct inode *inode = mapping->host;
1379 	loff_t old_size = inode->i_size;
1380 	int ret = 0, ret2;
1381 	int i_size_changed = 0;
1382 
1383 	trace_ext4_write_end(inode, pos, len, copied);
1384 	if (ext4_has_inline_data(inode)) {
1385 		ret = ext4_write_inline_data_end(inode, pos, len,
1386 						 copied, page);
1387 		if (ret < 0) {
1388 			unlock_page(page);
1389 			put_page(page);
1390 			goto errout;
1391 		}
1392 		copied = ret;
1393 	} else
1394 		copied = block_write_end(file, mapping, pos,
1395 					 len, copied, page, fsdata);
1396 	/*
1397 	 * it's important to update i_size while still holding page lock:
1398 	 * page writeout could otherwise come in and zero beyond i_size.
1399 	 */
1400 	i_size_changed = ext4_update_inode_size(inode, pos + copied);
1401 	unlock_page(page);
1402 	put_page(page);
1403 
1404 	if (old_size < pos)
1405 		pagecache_isize_extended(inode, old_size, pos);
1406 	/*
1407 	 * Don't mark the inode dirty under page lock. First, it unnecessarily
1408 	 * makes the holding time of page lock longer. Second, it forces lock
1409 	 * ordering of page lock and transaction start for journaling
1410 	 * filesystems.
1411 	 */
1412 	if (i_size_changed)
1413 		ext4_mark_inode_dirty(handle, inode);
1414 
1415 	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1416 		/* if we have allocated more blocks and copied
1417 		 * less. We will have blocks allocated outside
1418 		 * inode->i_size. So truncate them
1419 		 */
1420 		ext4_orphan_add(handle, inode);
1421 errout:
1422 	ret2 = ext4_journal_stop(handle);
1423 	if (!ret)
1424 		ret = ret2;
1425 
1426 	if (pos + len > inode->i_size) {
1427 		ext4_truncate_failed_write(inode);
1428 		/*
1429 		 * If truncate failed early the inode might still be
1430 		 * on the orphan list; we need to make sure the inode
1431 		 * is removed from the orphan list in that case.
1432 		 */
1433 		if (inode->i_nlink)
1434 			ext4_orphan_del(NULL, inode);
1435 	}
1436 
1437 	return ret ? ret : copied;
1438 }
1439 
1440 /*
1441  * This is a private version of page_zero_new_buffers() which doesn't
1442  * set the buffer to be dirty, since in data=journalled mode we need
1443  * to call ext4_handle_dirty_metadata() instead.
1444  */
1445 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1446 					    struct page *page,
1447 					    unsigned from, unsigned to)
1448 {
1449 	unsigned int block_start = 0, block_end;
1450 	struct buffer_head *head, *bh;
1451 
1452 	bh = head = page_buffers(page);
1453 	do {
1454 		block_end = block_start + bh->b_size;
1455 		if (buffer_new(bh)) {
1456 			if (block_end > from && block_start < to) {
1457 				if (!PageUptodate(page)) {
1458 					unsigned start, size;
1459 
1460 					start = max(from, block_start);
1461 					size = min(to, block_end) - start;
1462 
1463 					zero_user(page, start, size);
1464 					write_end_fn(handle, bh);
1465 				}
1466 				clear_buffer_new(bh);
1467 			}
1468 		}
1469 		block_start = block_end;
1470 		bh = bh->b_this_page;
1471 	} while (bh != head);
1472 }
1473 
1474 static int ext4_journalled_write_end(struct file *file,
1475 				     struct address_space *mapping,
1476 				     loff_t pos, unsigned len, unsigned copied,
1477 				     struct page *page, void *fsdata)
1478 {
1479 	handle_t *handle = ext4_journal_current_handle();
1480 	struct inode *inode = mapping->host;
1481 	loff_t old_size = inode->i_size;
1482 	int ret = 0, ret2;
1483 	int partial = 0;
1484 	unsigned from, to;
1485 	int size_changed = 0;
1486 
1487 	trace_ext4_journalled_write_end(inode, pos, len, copied);
1488 	from = pos & (PAGE_SIZE - 1);
1489 	to = from + len;
1490 
1491 	BUG_ON(!ext4_handle_valid(handle));
1492 
1493 	if (ext4_has_inline_data(inode)) {
1494 		ret = ext4_write_inline_data_end(inode, pos, len,
1495 						 copied, page);
1496 		if (ret < 0) {
1497 			unlock_page(page);
1498 			put_page(page);
1499 			goto errout;
1500 		}
1501 		copied = ret;
1502 	} else if (unlikely(copied < len) && !PageUptodate(page)) {
1503 		copied = 0;
1504 		ext4_journalled_zero_new_buffers(handle, page, from, to);
1505 	} else {
1506 		if (unlikely(copied < len))
1507 			ext4_journalled_zero_new_buffers(handle, page,
1508 							 from + copied, to);
1509 		ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1510 					     from + copied, &partial,
1511 					     write_end_fn);
1512 		if (!partial)
1513 			SetPageUptodate(page);
1514 	}
1515 	size_changed = ext4_update_inode_size(inode, pos + copied);
1516 	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1517 	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1518 	unlock_page(page);
1519 	put_page(page);
1520 
1521 	if (old_size < pos)
1522 		pagecache_isize_extended(inode, old_size, pos);
1523 
1524 	if (size_changed) {
1525 		ret2 = ext4_mark_inode_dirty(handle, inode);
1526 		if (!ret)
1527 			ret = ret2;
1528 	}
1529 
1530 	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1531 		/* if we have allocated more blocks and copied
1532 		 * less. We will have blocks allocated outside
1533 		 * inode->i_size. So truncate them
1534 		 */
1535 		ext4_orphan_add(handle, inode);
1536 
1537 errout:
1538 	ret2 = ext4_journal_stop(handle);
1539 	if (!ret)
1540 		ret = ret2;
1541 	if (pos + len > inode->i_size) {
1542 		ext4_truncate_failed_write(inode);
1543 		/*
1544 		 * If truncate failed early the inode might still be
1545 		 * on the orphan list; we need to make sure the inode
1546 		 * is removed from the orphan list in that case.
1547 		 */
1548 		if (inode->i_nlink)
1549 			ext4_orphan_del(NULL, inode);
1550 	}
1551 
1552 	return ret ? ret : copied;
1553 }
1554 
1555 /*
1556  * Reserve space for a single cluster
1557  */
1558 static int ext4_da_reserve_space(struct inode *inode)
1559 {
1560 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1561 	struct ext4_inode_info *ei = EXT4_I(inode);
1562 	int ret;
1563 
1564 	/*
1565 	 * We will charge metadata quota at writeout time; this saves
1566 	 * us from metadata over-estimation, though we may go over by
1567 	 * a small amount in the end.  Here we just reserve for data.
1568 	 */
1569 	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1570 	if (ret)
1571 		return ret;
1572 
1573 	spin_lock(&ei->i_block_reservation_lock);
1574 	if (ext4_claim_free_clusters(sbi, 1, 0)) {
1575 		spin_unlock(&ei->i_block_reservation_lock);
1576 		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1577 		return -ENOSPC;
1578 	}
1579 	ei->i_reserved_data_blocks++;
1580 	trace_ext4_da_reserve_space(inode);
1581 	spin_unlock(&ei->i_block_reservation_lock);
1582 
1583 	return 0;       /* success */
1584 }
1585 
1586 static void ext4_da_release_space(struct inode *inode, int to_free)
1587 {
1588 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1589 	struct ext4_inode_info *ei = EXT4_I(inode);
1590 
1591 	if (!to_free)
1592 		return;		/* Nothing to release, exit */
1593 
1594 	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1595 
1596 	trace_ext4_da_release_space(inode, to_free);
1597 	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1598 		/*
1599 		 * if there aren't enough reserved blocks, then the
1600 		 * counter is messed up somewhere.  Since this
1601 		 * function is called from invalidate page, it's
1602 		 * harmless to return without any action.
1603 		 */
1604 		ext4_warning(inode->i_sb, "ext4_da_release_space: "
1605 			 "ino %lu, to_free %d with only %d reserved "
1606 			 "data blocks", inode->i_ino, to_free,
1607 			 ei->i_reserved_data_blocks);
1608 		WARN_ON(1);
1609 		to_free = ei->i_reserved_data_blocks;
1610 	}
1611 	ei->i_reserved_data_blocks -= to_free;
1612 
1613 	/* update fs dirty data blocks counter */
1614 	percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1615 
1616 	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1617 
1618 	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1619 }
1620 
1621 static void ext4_da_page_release_reservation(struct page *page,
1622 					     unsigned int offset,
1623 					     unsigned int length)
1624 {
1625 	int to_release = 0, contiguous_blks = 0;
1626 	struct buffer_head *head, *bh;
1627 	unsigned int curr_off = 0;
1628 	struct inode *inode = page->mapping->host;
1629 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1630 	unsigned int stop = offset + length;
1631 	int num_clusters;
1632 	ext4_fsblk_t lblk;
1633 
1634 	BUG_ON(stop > PAGE_SIZE || stop < length);
1635 
1636 	head = page_buffers(page);
1637 	bh = head;
1638 	do {
1639 		unsigned int next_off = curr_off + bh->b_size;
1640 
1641 		if (next_off > stop)
1642 			break;
1643 
1644 		if ((offset <= curr_off) && (buffer_delay(bh))) {
1645 			to_release++;
1646 			contiguous_blks++;
1647 			clear_buffer_delay(bh);
1648 		} else if (contiguous_blks) {
1649 			lblk = page->index <<
1650 			       (PAGE_SHIFT - inode->i_blkbits);
1651 			lblk += (curr_off >> inode->i_blkbits) -
1652 				contiguous_blks;
1653 			ext4_es_remove_extent(inode, lblk, contiguous_blks);
1654 			contiguous_blks = 0;
1655 		}
1656 		curr_off = next_off;
1657 	} while ((bh = bh->b_this_page) != head);
1658 
1659 	if (contiguous_blks) {
1660 		lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1661 		lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1662 		ext4_es_remove_extent(inode, lblk, contiguous_blks);
1663 	}
1664 
1665 	/* If we have released all the blocks belonging to a cluster, then we
1666 	 * need to release the reserved space for that cluster. */
1667 	num_clusters = EXT4_NUM_B2C(sbi, to_release);
1668 	while (num_clusters > 0) {
1669 		lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1670 			((num_clusters - 1) << sbi->s_cluster_bits);
1671 		if (sbi->s_cluster_ratio == 1 ||
1672 		    !ext4_find_delalloc_cluster(inode, lblk))
1673 			ext4_da_release_space(inode, 1);
1674 
1675 		num_clusters--;
1676 	}
1677 }
1678 
1679 /*
1680  * Delayed allocation stuff
1681  */
1682 
1683 struct mpage_da_data {
1684 	struct inode *inode;
1685 	struct writeback_control *wbc;
1686 
1687 	pgoff_t first_page;	/* The first page to write */
1688 	pgoff_t next_page;	/* Current page to examine */
1689 	pgoff_t last_page;	/* Last page to examine */
1690 	/*
1691 	 * Extent to map - this can be after first_page because that can be
1692 	 * fully mapped. We somewhat abuse m_flags to store whether the extent
1693 	 * is delalloc or unwritten.
1694 	 */
1695 	struct ext4_map_blocks map;
1696 	struct ext4_io_submit io_submit;	/* IO submission data */
1697 	unsigned int do_map:1;
1698 };
1699 
1700 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1701 				       bool invalidate)
1702 {
1703 	int nr_pages, i;
1704 	pgoff_t index, end;
1705 	struct pagevec pvec;
1706 	struct inode *inode = mpd->inode;
1707 	struct address_space *mapping = inode->i_mapping;
1708 
1709 	/* This is necessary when next_page == 0. */
1710 	if (mpd->first_page >= mpd->next_page)
1711 		return;
1712 
1713 	index = mpd->first_page;
1714 	end   = mpd->next_page - 1;
1715 	if (invalidate) {
1716 		ext4_lblk_t start, last;
1717 		start = index << (PAGE_SHIFT - inode->i_blkbits);
1718 		last = end << (PAGE_SHIFT - inode->i_blkbits);
1719 		ext4_es_remove_extent(inode, start, last - start + 1);
1720 	}
1721 
1722 	pagevec_init(&pvec);
1723 	while (index <= end) {
1724 		nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1725 		if (nr_pages == 0)
1726 			break;
1727 		for (i = 0; i < nr_pages; i++) {
1728 			struct page *page = pvec.pages[i];
1729 
1730 			BUG_ON(!PageLocked(page));
1731 			BUG_ON(PageWriteback(page));
1732 			if (invalidate) {
1733 				if (page_mapped(page))
1734 					clear_page_dirty_for_io(page);
1735 				block_invalidatepage(page, 0, PAGE_SIZE);
1736 				ClearPageUptodate(page);
1737 			}
1738 			unlock_page(page);
1739 		}
1740 		pagevec_release(&pvec);
1741 	}
1742 }
1743 
1744 static void ext4_print_free_blocks(struct inode *inode)
1745 {
1746 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1747 	struct super_block *sb = inode->i_sb;
1748 	struct ext4_inode_info *ei = EXT4_I(inode);
1749 
1750 	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1751 	       EXT4_C2B(EXT4_SB(inode->i_sb),
1752 			ext4_count_free_clusters(sb)));
1753 	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1754 	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1755 	       (long long) EXT4_C2B(EXT4_SB(sb),
1756 		percpu_counter_sum(&sbi->s_freeclusters_counter)));
1757 	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1758 	       (long long) EXT4_C2B(EXT4_SB(sb),
1759 		percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1760 	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1761 	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1762 		 ei->i_reserved_data_blocks);
1763 	return;
1764 }
1765 
1766 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1767 {
1768 	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1769 }
1770 
1771 /*
1772  * This function is grabs code from the very beginning of
1773  * ext4_map_blocks, but assumes that the caller is from delayed write
1774  * time. This function looks up the requested blocks and sets the
1775  * buffer delay bit under the protection of i_data_sem.
1776  */
1777 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1778 			      struct ext4_map_blocks *map,
1779 			      struct buffer_head *bh)
1780 {
1781 	struct extent_status es;
1782 	int retval;
1783 	sector_t invalid_block = ~((sector_t) 0xffff);
1784 #ifdef ES_AGGRESSIVE_TEST
1785 	struct ext4_map_blocks orig_map;
1786 
1787 	memcpy(&orig_map, map, sizeof(*map));
1788 #endif
1789 
1790 	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1791 		invalid_block = ~0;
1792 
1793 	map->m_flags = 0;
1794 	ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1795 		  "logical block %lu\n", inode->i_ino, map->m_len,
1796 		  (unsigned long) map->m_lblk);
1797 
1798 	/* Lookup extent status tree firstly */
1799 	if (ext4_es_lookup_extent(inode, iblock, &es)) {
1800 		if (ext4_es_is_hole(&es)) {
1801 			retval = 0;
1802 			down_read(&EXT4_I(inode)->i_data_sem);
1803 			goto add_delayed;
1804 		}
1805 
1806 		/*
1807 		 * Delayed extent could be allocated by fallocate.
1808 		 * So we need to check it.
1809 		 */
1810 		if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1811 			map_bh(bh, inode->i_sb, invalid_block);
1812 			set_buffer_new(bh);
1813 			set_buffer_delay(bh);
1814 			return 0;
1815 		}
1816 
1817 		map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1818 		retval = es.es_len - (iblock - es.es_lblk);
1819 		if (retval > map->m_len)
1820 			retval = map->m_len;
1821 		map->m_len = retval;
1822 		if (ext4_es_is_written(&es))
1823 			map->m_flags |= EXT4_MAP_MAPPED;
1824 		else if (ext4_es_is_unwritten(&es))
1825 			map->m_flags |= EXT4_MAP_UNWRITTEN;
1826 		else
1827 			BUG_ON(1);
1828 
1829 #ifdef ES_AGGRESSIVE_TEST
1830 		ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1831 #endif
1832 		return retval;
1833 	}
1834 
1835 	/*
1836 	 * Try to see if we can get the block without requesting a new
1837 	 * file system block.
1838 	 */
1839 	down_read(&EXT4_I(inode)->i_data_sem);
1840 	if (ext4_has_inline_data(inode))
1841 		retval = 0;
1842 	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1843 		retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1844 	else
1845 		retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1846 
1847 add_delayed:
1848 	if (retval == 0) {
1849 		int ret;
1850 		/*
1851 		 * XXX: __block_prepare_write() unmaps passed block,
1852 		 * is it OK?
1853 		 */
1854 		/*
1855 		 * If the block was allocated from previously allocated cluster,
1856 		 * then we don't need to reserve it again. However we still need
1857 		 * to reserve metadata for every block we're going to write.
1858 		 */
1859 		if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1860 		    !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1861 			ret = ext4_da_reserve_space(inode);
1862 			if (ret) {
1863 				/* not enough space to reserve */
1864 				retval = ret;
1865 				goto out_unlock;
1866 			}
1867 		}
1868 
1869 		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1870 					    ~0, EXTENT_STATUS_DELAYED);
1871 		if (ret) {
1872 			retval = ret;
1873 			goto out_unlock;
1874 		}
1875 
1876 		map_bh(bh, inode->i_sb, invalid_block);
1877 		set_buffer_new(bh);
1878 		set_buffer_delay(bh);
1879 	} else if (retval > 0) {
1880 		int ret;
1881 		unsigned int status;
1882 
1883 		if (unlikely(retval != map->m_len)) {
1884 			ext4_warning(inode->i_sb,
1885 				     "ES len assertion failed for inode "
1886 				     "%lu: retval %d != map->m_len %d",
1887 				     inode->i_ino, retval, map->m_len);
1888 			WARN_ON(1);
1889 		}
1890 
1891 		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1892 				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1893 		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1894 					    map->m_pblk, status);
1895 		if (ret != 0)
1896 			retval = ret;
1897 	}
1898 
1899 out_unlock:
1900 	up_read((&EXT4_I(inode)->i_data_sem));
1901 
1902 	return retval;
1903 }
1904 
1905 /*
1906  * This is a special get_block_t callback which is used by
1907  * ext4_da_write_begin().  It will either return mapped block or
1908  * reserve space for a single block.
1909  *
1910  * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1911  * We also have b_blocknr = -1 and b_bdev initialized properly
1912  *
1913  * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1914  * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1915  * initialized properly.
1916  */
1917 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1918 			   struct buffer_head *bh, int create)
1919 {
1920 	struct ext4_map_blocks map;
1921 	int ret = 0;
1922 
1923 	BUG_ON(create == 0);
1924 	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1925 
1926 	map.m_lblk = iblock;
1927 	map.m_len = 1;
1928 
1929 	/*
1930 	 * first, we need to know whether the block is allocated already
1931 	 * preallocated blocks are unmapped but should treated
1932 	 * the same as allocated blocks.
1933 	 */
1934 	ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1935 	if (ret <= 0)
1936 		return ret;
1937 
1938 	map_bh(bh, inode->i_sb, map.m_pblk);
1939 	ext4_update_bh_state(bh, map.m_flags);
1940 
1941 	if (buffer_unwritten(bh)) {
1942 		/* A delayed write to unwritten bh should be marked
1943 		 * new and mapped.  Mapped ensures that we don't do
1944 		 * get_block multiple times when we write to the same
1945 		 * offset and new ensures that we do proper zero out
1946 		 * for partial write.
1947 		 */
1948 		set_buffer_new(bh);
1949 		set_buffer_mapped(bh);
1950 	}
1951 	return 0;
1952 }
1953 
1954 static int bget_one(handle_t *handle, struct buffer_head *bh)
1955 {
1956 	get_bh(bh);
1957 	return 0;
1958 }
1959 
1960 static int bput_one(handle_t *handle, struct buffer_head *bh)
1961 {
1962 	put_bh(bh);
1963 	return 0;
1964 }
1965 
1966 static int __ext4_journalled_writepage(struct page *page,
1967 				       unsigned int len)
1968 {
1969 	struct address_space *mapping = page->mapping;
1970 	struct inode *inode = mapping->host;
1971 	struct buffer_head *page_bufs = NULL;
1972 	handle_t *handle = NULL;
1973 	int ret = 0, err = 0;
1974 	int inline_data = ext4_has_inline_data(inode);
1975 	struct buffer_head *inode_bh = NULL;
1976 
1977 	ClearPageChecked(page);
1978 
1979 	if (inline_data) {
1980 		BUG_ON(page->index != 0);
1981 		BUG_ON(len > ext4_get_max_inline_size(inode));
1982 		inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1983 		if (inode_bh == NULL)
1984 			goto out;
1985 	} else {
1986 		page_bufs = page_buffers(page);
1987 		if (!page_bufs) {
1988 			BUG();
1989 			goto out;
1990 		}
1991 		ext4_walk_page_buffers(handle, page_bufs, 0, len,
1992 				       NULL, bget_one);
1993 	}
1994 	/*
1995 	 * We need to release the page lock before we start the
1996 	 * journal, so grab a reference so the page won't disappear
1997 	 * out from under us.
1998 	 */
1999 	get_page(page);
2000 	unlock_page(page);
2001 
2002 	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2003 				    ext4_writepage_trans_blocks(inode));
2004 	if (IS_ERR(handle)) {
2005 		ret = PTR_ERR(handle);
2006 		put_page(page);
2007 		goto out_no_pagelock;
2008 	}
2009 	BUG_ON(!ext4_handle_valid(handle));
2010 
2011 	lock_page(page);
2012 	put_page(page);
2013 	if (page->mapping != mapping) {
2014 		/* The page got truncated from under us */
2015 		ext4_journal_stop(handle);
2016 		ret = 0;
2017 		goto out;
2018 	}
2019 
2020 	if (inline_data) {
2021 		BUFFER_TRACE(inode_bh, "get write access");
2022 		ret = ext4_journal_get_write_access(handle, inode_bh);
2023 
2024 		err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2025 
2026 	} else {
2027 		ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2028 					     do_journal_get_write_access);
2029 
2030 		err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2031 					     write_end_fn);
2032 	}
2033 	if (ret == 0)
2034 		ret = err;
2035 	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2036 	err = ext4_journal_stop(handle);
2037 	if (!ret)
2038 		ret = err;
2039 
2040 	if (!ext4_has_inline_data(inode))
2041 		ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2042 				       NULL, bput_one);
2043 	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2044 out:
2045 	unlock_page(page);
2046 out_no_pagelock:
2047 	brelse(inode_bh);
2048 	return ret;
2049 }
2050 
2051 /*
2052  * Note that we don't need to start a transaction unless we're journaling data
2053  * because we should have holes filled from ext4_page_mkwrite(). We even don't
2054  * need to file the inode to the transaction's list in ordered mode because if
2055  * we are writing back data added by write(), the inode is already there and if
2056  * we are writing back data modified via mmap(), no one guarantees in which
2057  * transaction the data will hit the disk. In case we are journaling data, we
2058  * cannot start transaction directly because transaction start ranks above page
2059  * lock so we have to do some magic.
2060  *
2061  * This function can get called via...
2062  *   - ext4_writepages after taking page lock (have journal handle)
2063  *   - journal_submit_inode_data_buffers (no journal handle)
2064  *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2065  *   - grab_page_cache when doing write_begin (have journal handle)
2066  *
2067  * We don't do any block allocation in this function. If we have page with
2068  * multiple blocks we need to write those buffer_heads that are mapped. This
2069  * is important for mmaped based write. So if we do with blocksize 1K
2070  * truncate(f, 1024);
2071  * a = mmap(f, 0, 4096);
2072  * a[0] = 'a';
2073  * truncate(f, 4096);
2074  * we have in the page first buffer_head mapped via page_mkwrite call back
2075  * but other buffer_heads would be unmapped but dirty (dirty done via the
2076  * do_wp_page). So writepage should write the first block. If we modify
2077  * the mmap area beyond 1024 we will again get a page_fault and the
2078  * page_mkwrite callback will do the block allocation and mark the
2079  * buffer_heads mapped.
2080  *
2081  * We redirty the page if we have any buffer_heads that is either delay or
2082  * unwritten in the page.
2083  *
2084  * We can get recursively called as show below.
2085  *
2086  *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2087  *		ext4_writepage()
2088  *
2089  * But since we don't do any block allocation we should not deadlock.
2090  * Page also have the dirty flag cleared so we don't get recurive page_lock.
2091  */
2092 static int ext4_writepage(struct page *page,
2093 			  struct writeback_control *wbc)
2094 {
2095 	int ret = 0;
2096 	loff_t size;
2097 	unsigned int len;
2098 	struct buffer_head *page_bufs = NULL;
2099 	struct inode *inode = page->mapping->host;
2100 	struct ext4_io_submit io_submit;
2101 	bool keep_towrite = false;
2102 
2103 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2104 		ext4_invalidatepage(page, 0, PAGE_SIZE);
2105 		unlock_page(page);
2106 		return -EIO;
2107 	}
2108 
2109 	trace_ext4_writepage(page);
2110 	size = i_size_read(inode);
2111 	if (page->index == size >> PAGE_SHIFT)
2112 		len = size & ~PAGE_MASK;
2113 	else
2114 		len = PAGE_SIZE;
2115 
2116 	page_bufs = page_buffers(page);
2117 	/*
2118 	 * We cannot do block allocation or other extent handling in this
2119 	 * function. If there are buffers needing that, we have to redirty
2120 	 * the page. But we may reach here when we do a journal commit via
2121 	 * journal_submit_inode_data_buffers() and in that case we must write
2122 	 * allocated buffers to achieve data=ordered mode guarantees.
2123 	 *
2124 	 * Also, if there is only one buffer per page (the fs block
2125 	 * size == the page size), if one buffer needs block
2126 	 * allocation or needs to modify the extent tree to clear the
2127 	 * unwritten flag, we know that the page can't be written at
2128 	 * all, so we might as well refuse the write immediately.
2129 	 * Unfortunately if the block size != page size, we can't as
2130 	 * easily detect this case using ext4_walk_page_buffers(), but
2131 	 * for the extremely common case, this is an optimization that
2132 	 * skips a useless round trip through ext4_bio_write_page().
2133 	 */
2134 	if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2135 				   ext4_bh_delay_or_unwritten)) {
2136 		redirty_page_for_writepage(wbc, page);
2137 		if ((current->flags & PF_MEMALLOC) ||
2138 		    (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2139 			/*
2140 			 * For memory cleaning there's no point in writing only
2141 			 * some buffers. So just bail out. Warn if we came here
2142 			 * from direct reclaim.
2143 			 */
2144 			WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2145 							== PF_MEMALLOC);
2146 			unlock_page(page);
2147 			return 0;
2148 		}
2149 		keep_towrite = true;
2150 	}
2151 
2152 	if (PageChecked(page) && ext4_should_journal_data(inode))
2153 		/*
2154 		 * It's mmapped pagecache.  Add buffers and journal it.  There
2155 		 * doesn't seem much point in redirtying the page here.
2156 		 */
2157 		return __ext4_journalled_writepage(page, len);
2158 
2159 	ext4_io_submit_init(&io_submit, wbc);
2160 	io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2161 	if (!io_submit.io_end) {
2162 		redirty_page_for_writepage(wbc, page);
2163 		unlock_page(page);
2164 		return -ENOMEM;
2165 	}
2166 	ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2167 	ext4_io_submit(&io_submit);
2168 	/* Drop io_end reference we got from init */
2169 	ext4_put_io_end_defer(io_submit.io_end);
2170 	return ret;
2171 }
2172 
2173 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2174 {
2175 	int len;
2176 	loff_t size;
2177 	int err;
2178 
2179 	BUG_ON(page->index != mpd->first_page);
2180 	clear_page_dirty_for_io(page);
2181 	/*
2182 	 * We have to be very careful here!  Nothing protects writeback path
2183 	 * against i_size changes and the page can be writeably mapped into
2184 	 * page tables. So an application can be growing i_size and writing
2185 	 * data through mmap while writeback runs. clear_page_dirty_for_io()
2186 	 * write-protects our page in page tables and the page cannot get
2187 	 * written to again until we release page lock. So only after
2188 	 * clear_page_dirty_for_io() we are safe to sample i_size for
2189 	 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2190 	 * on the barrier provided by TestClearPageDirty in
2191 	 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2192 	 * after page tables are updated.
2193 	 */
2194 	size = i_size_read(mpd->inode);
2195 	if (page->index == size >> PAGE_SHIFT)
2196 		len = size & ~PAGE_MASK;
2197 	else
2198 		len = PAGE_SIZE;
2199 	err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2200 	if (!err)
2201 		mpd->wbc->nr_to_write--;
2202 	mpd->first_page++;
2203 
2204 	return err;
2205 }
2206 
2207 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2208 
2209 /*
2210  * mballoc gives us at most this number of blocks...
2211  * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2212  * The rest of mballoc seems to handle chunks up to full group size.
2213  */
2214 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2215 
2216 /*
2217  * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2218  *
2219  * @mpd - extent of blocks
2220  * @lblk - logical number of the block in the file
2221  * @bh - buffer head we want to add to the extent
2222  *
2223  * The function is used to collect contig. blocks in the same state. If the
2224  * buffer doesn't require mapping for writeback and we haven't started the
2225  * extent of buffers to map yet, the function returns 'true' immediately - the
2226  * caller can write the buffer right away. Otherwise the function returns true
2227  * if the block has been added to the extent, false if the block couldn't be
2228  * added.
2229  */
2230 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2231 				   struct buffer_head *bh)
2232 {
2233 	struct ext4_map_blocks *map = &mpd->map;
2234 
2235 	/* Buffer that doesn't need mapping for writeback? */
2236 	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2237 	    (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2238 		/* So far no extent to map => we write the buffer right away */
2239 		if (map->m_len == 0)
2240 			return true;
2241 		return false;
2242 	}
2243 
2244 	/* First block in the extent? */
2245 	if (map->m_len == 0) {
2246 		/* We cannot map unless handle is started... */
2247 		if (!mpd->do_map)
2248 			return false;
2249 		map->m_lblk = lblk;
2250 		map->m_len = 1;
2251 		map->m_flags = bh->b_state & BH_FLAGS;
2252 		return true;
2253 	}
2254 
2255 	/* Don't go larger than mballoc is willing to allocate */
2256 	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2257 		return false;
2258 
2259 	/* Can we merge the block to our big extent? */
2260 	if (lblk == map->m_lblk + map->m_len &&
2261 	    (bh->b_state & BH_FLAGS) == map->m_flags) {
2262 		map->m_len++;
2263 		return true;
2264 	}
2265 	return false;
2266 }
2267 
2268 /*
2269  * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2270  *
2271  * @mpd - extent of blocks for mapping
2272  * @head - the first buffer in the page
2273  * @bh - buffer we should start processing from
2274  * @lblk - logical number of the block in the file corresponding to @bh
2275  *
2276  * Walk through page buffers from @bh upto @head (exclusive) and either submit
2277  * the page for IO if all buffers in this page were mapped and there's no
2278  * accumulated extent of buffers to map or add buffers in the page to the
2279  * extent of buffers to map. The function returns 1 if the caller can continue
2280  * by processing the next page, 0 if it should stop adding buffers to the
2281  * extent to map because we cannot extend it anymore. It can also return value
2282  * < 0 in case of error during IO submission.
2283  */
2284 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2285 				   struct buffer_head *head,
2286 				   struct buffer_head *bh,
2287 				   ext4_lblk_t lblk)
2288 {
2289 	struct inode *inode = mpd->inode;
2290 	int err;
2291 	ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2292 							>> inode->i_blkbits;
2293 
2294 	do {
2295 		BUG_ON(buffer_locked(bh));
2296 
2297 		if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2298 			/* Found extent to map? */
2299 			if (mpd->map.m_len)
2300 				return 0;
2301 			/* Buffer needs mapping and handle is not started? */
2302 			if (!mpd->do_map)
2303 				return 0;
2304 			/* Everything mapped so far and we hit EOF */
2305 			break;
2306 		}
2307 	} while (lblk++, (bh = bh->b_this_page) != head);
2308 	/* So far everything mapped? Submit the page for IO. */
2309 	if (mpd->map.m_len == 0) {
2310 		err = mpage_submit_page(mpd, head->b_page);
2311 		if (err < 0)
2312 			return err;
2313 	}
2314 	return lblk < blocks;
2315 }
2316 
2317 /*
2318  * mpage_map_buffers - update buffers corresponding to changed extent and
2319  *		       submit fully mapped pages for IO
2320  *
2321  * @mpd - description of extent to map, on return next extent to map
2322  *
2323  * Scan buffers corresponding to changed extent (we expect corresponding pages
2324  * to be already locked) and update buffer state according to new extent state.
2325  * We map delalloc buffers to their physical location, clear unwritten bits,
2326  * and mark buffers as uninit when we perform writes to unwritten extents
2327  * and do extent conversion after IO is finished. If the last page is not fully
2328  * mapped, we update @map to the next extent in the last page that needs
2329  * mapping. Otherwise we submit the page for IO.
2330  */
2331 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2332 {
2333 	struct pagevec pvec;
2334 	int nr_pages, i;
2335 	struct inode *inode = mpd->inode;
2336 	struct buffer_head *head, *bh;
2337 	int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2338 	pgoff_t start, end;
2339 	ext4_lblk_t lblk;
2340 	sector_t pblock;
2341 	int err;
2342 
2343 	start = mpd->map.m_lblk >> bpp_bits;
2344 	end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2345 	lblk = start << bpp_bits;
2346 	pblock = mpd->map.m_pblk;
2347 
2348 	pagevec_init(&pvec);
2349 	while (start <= end) {
2350 		nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2351 						&start, end);
2352 		if (nr_pages == 0)
2353 			break;
2354 		for (i = 0; i < nr_pages; i++) {
2355 			struct page *page = pvec.pages[i];
2356 
2357 			bh = head = page_buffers(page);
2358 			do {
2359 				if (lblk < mpd->map.m_lblk)
2360 					continue;
2361 				if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2362 					/*
2363 					 * Buffer after end of mapped extent.
2364 					 * Find next buffer in the page to map.
2365 					 */
2366 					mpd->map.m_len = 0;
2367 					mpd->map.m_flags = 0;
2368 					/*
2369 					 * FIXME: If dioread_nolock supports
2370 					 * blocksize < pagesize, we need to make
2371 					 * sure we add size mapped so far to
2372 					 * io_end->size as the following call
2373 					 * can submit the page for IO.
2374 					 */
2375 					err = mpage_process_page_bufs(mpd, head,
2376 								      bh, lblk);
2377 					pagevec_release(&pvec);
2378 					if (err > 0)
2379 						err = 0;
2380 					return err;
2381 				}
2382 				if (buffer_delay(bh)) {
2383 					clear_buffer_delay(bh);
2384 					bh->b_blocknr = pblock++;
2385 				}
2386 				clear_buffer_unwritten(bh);
2387 			} while (lblk++, (bh = bh->b_this_page) != head);
2388 
2389 			/*
2390 			 * FIXME: This is going to break if dioread_nolock
2391 			 * supports blocksize < pagesize as we will try to
2392 			 * convert potentially unmapped parts of inode.
2393 			 */
2394 			mpd->io_submit.io_end->size += PAGE_SIZE;
2395 			/* Page fully mapped - let IO run! */
2396 			err = mpage_submit_page(mpd, page);
2397 			if (err < 0) {
2398 				pagevec_release(&pvec);
2399 				return err;
2400 			}
2401 		}
2402 		pagevec_release(&pvec);
2403 	}
2404 	/* Extent fully mapped and matches with page boundary. We are done. */
2405 	mpd->map.m_len = 0;
2406 	mpd->map.m_flags = 0;
2407 	return 0;
2408 }
2409 
2410 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2411 {
2412 	struct inode *inode = mpd->inode;
2413 	struct ext4_map_blocks *map = &mpd->map;
2414 	int get_blocks_flags;
2415 	int err, dioread_nolock;
2416 
2417 	trace_ext4_da_write_pages_extent(inode, map);
2418 	/*
2419 	 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2420 	 * to convert an unwritten extent to be initialized (in the case
2421 	 * where we have written into one or more preallocated blocks).  It is
2422 	 * possible that we're going to need more metadata blocks than
2423 	 * previously reserved. However we must not fail because we're in
2424 	 * writeback and there is nothing we can do about it so it might result
2425 	 * in data loss.  So use reserved blocks to allocate metadata if
2426 	 * possible.
2427 	 *
2428 	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2429 	 * the blocks in question are delalloc blocks.  This indicates
2430 	 * that the blocks and quotas has already been checked when
2431 	 * the data was copied into the page cache.
2432 	 */
2433 	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2434 			   EXT4_GET_BLOCKS_METADATA_NOFAIL |
2435 			   EXT4_GET_BLOCKS_IO_SUBMIT;
2436 	dioread_nolock = ext4_should_dioread_nolock(inode);
2437 	if (dioread_nolock)
2438 		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2439 	if (map->m_flags & (1 << BH_Delay))
2440 		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2441 
2442 	err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2443 	if (err < 0)
2444 		return err;
2445 	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2446 		if (!mpd->io_submit.io_end->handle &&
2447 		    ext4_handle_valid(handle)) {
2448 			mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2449 			handle->h_rsv_handle = NULL;
2450 		}
2451 		ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2452 	}
2453 
2454 	BUG_ON(map->m_len == 0);
2455 	if (map->m_flags & EXT4_MAP_NEW) {
2456 		clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2457 				   map->m_len);
2458 	}
2459 	return 0;
2460 }
2461 
2462 /*
2463  * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2464  *				 mpd->len and submit pages underlying it for IO
2465  *
2466  * @handle - handle for journal operations
2467  * @mpd - extent to map
2468  * @give_up_on_write - we set this to true iff there is a fatal error and there
2469  *                     is no hope of writing the data. The caller should discard
2470  *                     dirty pages to avoid infinite loops.
2471  *
2472  * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2473  * delayed, blocks are allocated, if it is unwritten, we may need to convert
2474  * them to initialized or split the described range from larger unwritten
2475  * extent. Note that we need not map all the described range since allocation
2476  * can return less blocks or the range is covered by more unwritten extents. We
2477  * cannot map more because we are limited by reserved transaction credits. On
2478  * the other hand we always make sure that the last touched page is fully
2479  * mapped so that it can be written out (and thus forward progress is
2480  * guaranteed). After mapping we submit all mapped pages for IO.
2481  */
2482 static int mpage_map_and_submit_extent(handle_t *handle,
2483 				       struct mpage_da_data *mpd,
2484 				       bool *give_up_on_write)
2485 {
2486 	struct inode *inode = mpd->inode;
2487 	struct ext4_map_blocks *map = &mpd->map;
2488 	int err;
2489 	loff_t disksize;
2490 	int progress = 0;
2491 
2492 	mpd->io_submit.io_end->offset =
2493 				((loff_t)map->m_lblk) << inode->i_blkbits;
2494 	do {
2495 		err = mpage_map_one_extent(handle, mpd);
2496 		if (err < 0) {
2497 			struct super_block *sb = inode->i_sb;
2498 
2499 			if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2500 			    EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2501 				goto invalidate_dirty_pages;
2502 			/*
2503 			 * Let the uper layers retry transient errors.
2504 			 * In the case of ENOSPC, if ext4_count_free_blocks()
2505 			 * is non-zero, a commit should free up blocks.
2506 			 */
2507 			if ((err == -ENOMEM) ||
2508 			    (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2509 				if (progress)
2510 					goto update_disksize;
2511 				return err;
2512 			}
2513 			ext4_msg(sb, KERN_CRIT,
2514 				 "Delayed block allocation failed for "
2515 				 "inode %lu at logical offset %llu with"
2516 				 " max blocks %u with error %d",
2517 				 inode->i_ino,
2518 				 (unsigned long long)map->m_lblk,
2519 				 (unsigned)map->m_len, -err);
2520 			ext4_msg(sb, KERN_CRIT,
2521 				 "This should not happen!! Data will "
2522 				 "be lost\n");
2523 			if (err == -ENOSPC)
2524 				ext4_print_free_blocks(inode);
2525 		invalidate_dirty_pages:
2526 			*give_up_on_write = true;
2527 			return err;
2528 		}
2529 		progress = 1;
2530 		/*
2531 		 * Update buffer state, submit mapped pages, and get us new
2532 		 * extent to map
2533 		 */
2534 		err = mpage_map_and_submit_buffers(mpd);
2535 		if (err < 0)
2536 			goto update_disksize;
2537 	} while (map->m_len);
2538 
2539 update_disksize:
2540 	/*
2541 	 * Update on-disk size after IO is submitted.  Races with
2542 	 * truncate are avoided by checking i_size under i_data_sem.
2543 	 */
2544 	disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2545 	if (disksize > EXT4_I(inode)->i_disksize) {
2546 		int err2;
2547 		loff_t i_size;
2548 
2549 		down_write(&EXT4_I(inode)->i_data_sem);
2550 		i_size = i_size_read(inode);
2551 		if (disksize > i_size)
2552 			disksize = i_size;
2553 		if (disksize > EXT4_I(inode)->i_disksize)
2554 			EXT4_I(inode)->i_disksize = disksize;
2555 		up_write(&EXT4_I(inode)->i_data_sem);
2556 		err2 = ext4_mark_inode_dirty(handle, inode);
2557 		if (err2)
2558 			ext4_error(inode->i_sb,
2559 				   "Failed to mark inode %lu dirty",
2560 				   inode->i_ino);
2561 		if (!err)
2562 			err = err2;
2563 	}
2564 	return err;
2565 }
2566 
2567 /*
2568  * Calculate the total number of credits to reserve for one writepages
2569  * iteration. This is called from ext4_writepages(). We map an extent of
2570  * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2571  * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2572  * bpp - 1 blocks in bpp different extents.
2573  */
2574 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2575 {
2576 	int bpp = ext4_journal_blocks_per_page(inode);
2577 
2578 	return ext4_meta_trans_blocks(inode,
2579 				MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2580 }
2581 
2582 /*
2583  * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2584  * 				 and underlying extent to map
2585  *
2586  * @mpd - where to look for pages
2587  *
2588  * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2589  * IO immediately. When we find a page which isn't mapped we start accumulating
2590  * extent of buffers underlying these pages that needs mapping (formed by
2591  * either delayed or unwritten buffers). We also lock the pages containing
2592  * these buffers. The extent found is returned in @mpd structure (starting at
2593  * mpd->lblk with length mpd->len blocks).
2594  *
2595  * Note that this function can attach bios to one io_end structure which are
2596  * neither logically nor physically contiguous. Although it may seem as an
2597  * unnecessary complication, it is actually inevitable in blocksize < pagesize
2598  * case as we need to track IO to all buffers underlying a page in one io_end.
2599  */
2600 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2601 {
2602 	struct address_space *mapping = mpd->inode->i_mapping;
2603 	struct pagevec pvec;
2604 	unsigned int nr_pages;
2605 	long left = mpd->wbc->nr_to_write;
2606 	pgoff_t index = mpd->first_page;
2607 	pgoff_t end = mpd->last_page;
2608 	int tag;
2609 	int i, err = 0;
2610 	int blkbits = mpd->inode->i_blkbits;
2611 	ext4_lblk_t lblk;
2612 	struct buffer_head *head;
2613 
2614 	if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2615 		tag = PAGECACHE_TAG_TOWRITE;
2616 	else
2617 		tag = PAGECACHE_TAG_DIRTY;
2618 
2619 	pagevec_init(&pvec);
2620 	mpd->map.m_len = 0;
2621 	mpd->next_page = index;
2622 	while (index <= end) {
2623 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2624 				tag);
2625 		if (nr_pages == 0)
2626 			goto out;
2627 
2628 		for (i = 0; i < nr_pages; i++) {
2629 			struct page *page = pvec.pages[i];
2630 
2631 			/*
2632 			 * Accumulated enough dirty pages? This doesn't apply
2633 			 * to WB_SYNC_ALL mode. For integrity sync we have to
2634 			 * keep going because someone may be concurrently
2635 			 * dirtying pages, and we might have synced a lot of
2636 			 * newly appeared dirty pages, but have not synced all
2637 			 * of the old dirty pages.
2638 			 */
2639 			if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2640 				goto out;
2641 
2642 			/* If we can't merge this page, we are done. */
2643 			if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2644 				goto out;
2645 
2646 			lock_page(page);
2647 			/*
2648 			 * If the page is no longer dirty, or its mapping no
2649 			 * longer corresponds to inode we are writing (which
2650 			 * means it has been truncated or invalidated), or the
2651 			 * page is already under writeback and we are not doing
2652 			 * a data integrity writeback, skip the page
2653 			 */
2654 			if (!PageDirty(page) ||
2655 			    (PageWriteback(page) &&
2656 			     (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2657 			    unlikely(page->mapping != mapping)) {
2658 				unlock_page(page);
2659 				continue;
2660 			}
2661 
2662 			wait_on_page_writeback(page);
2663 			BUG_ON(PageWriteback(page));
2664 
2665 			if (mpd->map.m_len == 0)
2666 				mpd->first_page = page->index;
2667 			mpd->next_page = page->index + 1;
2668 			/* Add all dirty buffers to mpd */
2669 			lblk = ((ext4_lblk_t)page->index) <<
2670 				(PAGE_SHIFT - blkbits);
2671 			head = page_buffers(page);
2672 			err = mpage_process_page_bufs(mpd, head, head, lblk);
2673 			if (err <= 0)
2674 				goto out;
2675 			err = 0;
2676 			left--;
2677 		}
2678 		pagevec_release(&pvec);
2679 		cond_resched();
2680 	}
2681 	return 0;
2682 out:
2683 	pagevec_release(&pvec);
2684 	return err;
2685 }
2686 
2687 static int __writepage(struct page *page, struct writeback_control *wbc,
2688 		       void *data)
2689 {
2690 	struct address_space *mapping = data;
2691 	int ret = ext4_writepage(page, wbc);
2692 	mapping_set_error(mapping, ret);
2693 	return ret;
2694 }
2695 
2696 static int ext4_writepages(struct address_space *mapping,
2697 			   struct writeback_control *wbc)
2698 {
2699 	pgoff_t	writeback_index = 0;
2700 	long nr_to_write = wbc->nr_to_write;
2701 	int range_whole = 0;
2702 	int cycled = 1;
2703 	handle_t *handle = NULL;
2704 	struct mpage_da_data mpd;
2705 	struct inode *inode = mapping->host;
2706 	int needed_blocks, rsv_blocks = 0, ret = 0;
2707 	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2708 	bool done;
2709 	struct blk_plug plug;
2710 	bool give_up_on_write = false;
2711 
2712 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2713 		return -EIO;
2714 
2715 	percpu_down_read(&sbi->s_journal_flag_rwsem);
2716 	trace_ext4_writepages(inode, wbc);
2717 
2718 	if (dax_mapping(mapping)) {
2719 		ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2720 						  wbc);
2721 		goto out_writepages;
2722 	}
2723 
2724 	/*
2725 	 * No pages to write? This is mainly a kludge to avoid starting
2726 	 * a transaction for special inodes like journal inode on last iput()
2727 	 * because that could violate lock ordering on umount
2728 	 */
2729 	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2730 		goto out_writepages;
2731 
2732 	if (ext4_should_journal_data(inode)) {
2733 		struct blk_plug plug;
2734 
2735 		blk_start_plug(&plug);
2736 		ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2737 		blk_finish_plug(&plug);
2738 		goto out_writepages;
2739 	}
2740 
2741 	/*
2742 	 * If the filesystem has aborted, it is read-only, so return
2743 	 * right away instead of dumping stack traces later on that
2744 	 * will obscure the real source of the problem.  We test
2745 	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2746 	 * the latter could be true if the filesystem is mounted
2747 	 * read-only, and in that case, ext4_writepages should
2748 	 * *never* be called, so if that ever happens, we would want
2749 	 * the stack trace.
2750 	 */
2751 	if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2752 		     sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2753 		ret = -EROFS;
2754 		goto out_writepages;
2755 	}
2756 
2757 	if (ext4_should_dioread_nolock(inode)) {
2758 		/*
2759 		 * We may need to convert up to one extent per block in
2760 		 * the page and we may dirty the inode.
2761 		 */
2762 		rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2763 	}
2764 
2765 	/*
2766 	 * If we have inline data and arrive here, it means that
2767 	 * we will soon create the block for the 1st page, so
2768 	 * we'd better clear the inline data here.
2769 	 */
2770 	if (ext4_has_inline_data(inode)) {
2771 		/* Just inode will be modified... */
2772 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2773 		if (IS_ERR(handle)) {
2774 			ret = PTR_ERR(handle);
2775 			goto out_writepages;
2776 		}
2777 		BUG_ON(ext4_test_inode_state(inode,
2778 				EXT4_STATE_MAY_INLINE_DATA));
2779 		ext4_destroy_inline_data(handle, inode);
2780 		ext4_journal_stop(handle);
2781 	}
2782 
2783 	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2784 		range_whole = 1;
2785 
2786 	if (wbc->range_cyclic) {
2787 		writeback_index = mapping->writeback_index;
2788 		if (writeback_index)
2789 			cycled = 0;
2790 		mpd.first_page = writeback_index;
2791 		mpd.last_page = -1;
2792 	} else {
2793 		mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2794 		mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2795 	}
2796 
2797 	mpd.inode = inode;
2798 	mpd.wbc = wbc;
2799 	ext4_io_submit_init(&mpd.io_submit, wbc);
2800 retry:
2801 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2802 		tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2803 	done = false;
2804 	blk_start_plug(&plug);
2805 
2806 	/*
2807 	 * First writeback pages that don't need mapping - we can avoid
2808 	 * starting a transaction unnecessarily and also avoid being blocked
2809 	 * in the block layer on device congestion while having transaction
2810 	 * started.
2811 	 */
2812 	mpd.do_map = 0;
2813 	mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2814 	if (!mpd.io_submit.io_end) {
2815 		ret = -ENOMEM;
2816 		goto unplug;
2817 	}
2818 	ret = mpage_prepare_extent_to_map(&mpd);
2819 	/* Submit prepared bio */
2820 	ext4_io_submit(&mpd.io_submit);
2821 	ext4_put_io_end_defer(mpd.io_submit.io_end);
2822 	mpd.io_submit.io_end = NULL;
2823 	/* Unlock pages we didn't use */
2824 	mpage_release_unused_pages(&mpd, false);
2825 	if (ret < 0)
2826 		goto unplug;
2827 
2828 	while (!done && mpd.first_page <= mpd.last_page) {
2829 		/* For each extent of pages we use new io_end */
2830 		mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2831 		if (!mpd.io_submit.io_end) {
2832 			ret = -ENOMEM;
2833 			break;
2834 		}
2835 
2836 		/*
2837 		 * We have two constraints: We find one extent to map and we
2838 		 * must always write out whole page (makes a difference when
2839 		 * blocksize < pagesize) so that we don't block on IO when we
2840 		 * try to write out the rest of the page. Journalled mode is
2841 		 * not supported by delalloc.
2842 		 */
2843 		BUG_ON(ext4_should_journal_data(inode));
2844 		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2845 
2846 		/* start a new transaction */
2847 		handle = ext4_journal_start_with_reserve(inode,
2848 				EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2849 		if (IS_ERR(handle)) {
2850 			ret = PTR_ERR(handle);
2851 			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2852 			       "%ld pages, ino %lu; err %d", __func__,
2853 				wbc->nr_to_write, inode->i_ino, ret);
2854 			/* Release allocated io_end */
2855 			ext4_put_io_end(mpd.io_submit.io_end);
2856 			mpd.io_submit.io_end = NULL;
2857 			break;
2858 		}
2859 		mpd.do_map = 1;
2860 
2861 		trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2862 		ret = mpage_prepare_extent_to_map(&mpd);
2863 		if (!ret) {
2864 			if (mpd.map.m_len)
2865 				ret = mpage_map_and_submit_extent(handle, &mpd,
2866 					&give_up_on_write);
2867 			else {
2868 				/*
2869 				 * We scanned the whole range (or exhausted
2870 				 * nr_to_write), submitted what was mapped and
2871 				 * didn't find anything needing mapping. We are
2872 				 * done.
2873 				 */
2874 				done = true;
2875 			}
2876 		}
2877 		/*
2878 		 * Caution: If the handle is synchronous,
2879 		 * ext4_journal_stop() can wait for transaction commit
2880 		 * to finish which may depend on writeback of pages to
2881 		 * complete or on page lock to be released.  In that
2882 		 * case, we have to wait until after after we have
2883 		 * submitted all the IO, released page locks we hold,
2884 		 * and dropped io_end reference (for extent conversion
2885 		 * to be able to complete) before stopping the handle.
2886 		 */
2887 		if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2888 			ext4_journal_stop(handle);
2889 			handle = NULL;
2890 			mpd.do_map = 0;
2891 		}
2892 		/* Submit prepared bio */
2893 		ext4_io_submit(&mpd.io_submit);
2894 		/* Unlock pages we didn't use */
2895 		mpage_release_unused_pages(&mpd, give_up_on_write);
2896 		/*
2897 		 * Drop our io_end reference we got from init. We have
2898 		 * to be careful and use deferred io_end finishing if
2899 		 * we are still holding the transaction as we can
2900 		 * release the last reference to io_end which may end
2901 		 * up doing unwritten extent conversion.
2902 		 */
2903 		if (handle) {
2904 			ext4_put_io_end_defer(mpd.io_submit.io_end);
2905 			ext4_journal_stop(handle);
2906 		} else
2907 			ext4_put_io_end(mpd.io_submit.io_end);
2908 		mpd.io_submit.io_end = NULL;
2909 
2910 		if (ret == -ENOSPC && sbi->s_journal) {
2911 			/*
2912 			 * Commit the transaction which would
2913 			 * free blocks released in the transaction
2914 			 * and try again
2915 			 */
2916 			jbd2_journal_force_commit_nested(sbi->s_journal);
2917 			ret = 0;
2918 			continue;
2919 		}
2920 		/* Fatal error - ENOMEM, EIO... */
2921 		if (ret)
2922 			break;
2923 	}
2924 unplug:
2925 	blk_finish_plug(&plug);
2926 	if (!ret && !cycled && wbc->nr_to_write > 0) {
2927 		cycled = 1;
2928 		mpd.last_page = writeback_index - 1;
2929 		mpd.first_page = 0;
2930 		goto retry;
2931 	}
2932 
2933 	/* Update index */
2934 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2935 		/*
2936 		 * Set the writeback_index so that range_cyclic
2937 		 * mode will write it back later
2938 		 */
2939 		mapping->writeback_index = mpd.first_page;
2940 
2941 out_writepages:
2942 	trace_ext4_writepages_result(inode, wbc, ret,
2943 				     nr_to_write - wbc->nr_to_write);
2944 	percpu_up_read(&sbi->s_journal_flag_rwsem);
2945 	return ret;
2946 }
2947 
2948 static int ext4_nonda_switch(struct super_block *sb)
2949 {
2950 	s64 free_clusters, dirty_clusters;
2951 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2952 
2953 	/*
2954 	 * switch to non delalloc mode if we are running low
2955 	 * on free block. The free block accounting via percpu
2956 	 * counters can get slightly wrong with percpu_counter_batch getting
2957 	 * accumulated on each CPU without updating global counters
2958 	 * Delalloc need an accurate free block accounting. So switch
2959 	 * to non delalloc when we are near to error range.
2960 	 */
2961 	free_clusters =
2962 		percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2963 	dirty_clusters =
2964 		percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2965 	/*
2966 	 * Start pushing delalloc when 1/2 of free blocks are dirty.
2967 	 */
2968 	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2969 		try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2970 
2971 	if (2 * free_clusters < 3 * dirty_clusters ||
2972 	    free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2973 		/*
2974 		 * free block count is less than 150% of dirty blocks
2975 		 * or free blocks is less than watermark
2976 		 */
2977 		return 1;
2978 	}
2979 	return 0;
2980 }
2981 
2982 /* We always reserve for an inode update; the superblock could be there too */
2983 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2984 {
2985 	if (likely(ext4_has_feature_large_file(inode->i_sb)))
2986 		return 1;
2987 
2988 	if (pos + len <= 0x7fffffffULL)
2989 		return 1;
2990 
2991 	/* We might need to update the superblock to set LARGE_FILE */
2992 	return 2;
2993 }
2994 
2995 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2996 			       loff_t pos, unsigned len, unsigned flags,
2997 			       struct page **pagep, void **fsdata)
2998 {
2999 	int ret, retries = 0;
3000 	struct page *page;
3001 	pgoff_t index;
3002 	struct inode *inode = mapping->host;
3003 	handle_t *handle;
3004 
3005 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3006 		return -EIO;
3007 
3008 	index = pos >> PAGE_SHIFT;
3009 
3010 	if (ext4_nonda_switch(inode->i_sb) ||
3011 	    S_ISLNK(inode->i_mode)) {
3012 		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3013 		return ext4_write_begin(file, mapping, pos,
3014 					len, flags, pagep, fsdata);
3015 	}
3016 	*fsdata = (void *)0;
3017 	trace_ext4_da_write_begin(inode, pos, len, flags);
3018 
3019 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3020 		ret = ext4_da_write_inline_data_begin(mapping, inode,
3021 						      pos, len, flags,
3022 						      pagep, fsdata);
3023 		if (ret < 0)
3024 			return ret;
3025 		if (ret == 1)
3026 			return 0;
3027 	}
3028 
3029 	/*
3030 	 * grab_cache_page_write_begin() can take a long time if the
3031 	 * system is thrashing due to memory pressure, or if the page
3032 	 * is being written back.  So grab it first before we start
3033 	 * the transaction handle.  This also allows us to allocate
3034 	 * the page (if needed) without using GFP_NOFS.
3035 	 */
3036 retry_grab:
3037 	page = grab_cache_page_write_begin(mapping, index, flags);
3038 	if (!page)
3039 		return -ENOMEM;
3040 	unlock_page(page);
3041 
3042 	/*
3043 	 * With delayed allocation, we don't log the i_disksize update
3044 	 * if there is delayed block allocation. But we still need
3045 	 * to journalling the i_disksize update if writes to the end
3046 	 * of file which has an already mapped buffer.
3047 	 */
3048 retry_journal:
3049 	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3050 				ext4_da_write_credits(inode, pos, len));
3051 	if (IS_ERR(handle)) {
3052 		put_page(page);
3053 		return PTR_ERR(handle);
3054 	}
3055 
3056 	lock_page(page);
3057 	if (page->mapping != mapping) {
3058 		/* The page got truncated from under us */
3059 		unlock_page(page);
3060 		put_page(page);
3061 		ext4_journal_stop(handle);
3062 		goto retry_grab;
3063 	}
3064 	/* In case writeback began while the page was unlocked */
3065 	wait_for_stable_page(page);
3066 
3067 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3068 	ret = ext4_block_write_begin(page, pos, len,
3069 				     ext4_da_get_block_prep);
3070 #else
3071 	ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3072 #endif
3073 	if (ret < 0) {
3074 		unlock_page(page);
3075 		ext4_journal_stop(handle);
3076 		/*
3077 		 * block_write_begin may have instantiated a few blocks
3078 		 * outside i_size.  Trim these off again. Don't need
3079 		 * i_size_read because we hold i_mutex.
3080 		 */
3081 		if (pos + len > inode->i_size)
3082 			ext4_truncate_failed_write(inode);
3083 
3084 		if (ret == -ENOSPC &&
3085 		    ext4_should_retry_alloc(inode->i_sb, &retries))
3086 			goto retry_journal;
3087 
3088 		put_page(page);
3089 		return ret;
3090 	}
3091 
3092 	*pagep = page;
3093 	return ret;
3094 }
3095 
3096 /*
3097  * Check if we should update i_disksize
3098  * when write to the end of file but not require block allocation
3099  */
3100 static int ext4_da_should_update_i_disksize(struct page *page,
3101 					    unsigned long offset)
3102 {
3103 	struct buffer_head *bh;
3104 	struct inode *inode = page->mapping->host;
3105 	unsigned int idx;
3106 	int i;
3107 
3108 	bh = page_buffers(page);
3109 	idx = offset >> inode->i_blkbits;
3110 
3111 	for (i = 0; i < idx; i++)
3112 		bh = bh->b_this_page;
3113 
3114 	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3115 		return 0;
3116 	return 1;
3117 }
3118 
3119 static int ext4_da_write_end(struct file *file,
3120 			     struct address_space *mapping,
3121 			     loff_t pos, unsigned len, unsigned copied,
3122 			     struct page *page, void *fsdata)
3123 {
3124 	struct inode *inode = mapping->host;
3125 	int ret = 0, ret2;
3126 	handle_t *handle = ext4_journal_current_handle();
3127 	loff_t new_i_size;
3128 	unsigned long start, end;
3129 	int write_mode = (int)(unsigned long)fsdata;
3130 
3131 	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3132 		return ext4_write_end(file, mapping, pos,
3133 				      len, copied, page, fsdata);
3134 
3135 	trace_ext4_da_write_end(inode, pos, len, copied);
3136 	start = pos & (PAGE_SIZE - 1);
3137 	end = start + copied - 1;
3138 
3139 	/*
3140 	 * generic_write_end() will run mark_inode_dirty() if i_size
3141 	 * changes.  So let's piggyback the i_disksize mark_inode_dirty
3142 	 * into that.
3143 	 */
3144 	new_i_size = pos + copied;
3145 	if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3146 		if (ext4_has_inline_data(inode) ||
3147 		    ext4_da_should_update_i_disksize(page, end)) {
3148 			ext4_update_i_disksize(inode, new_i_size);
3149 			/* We need to mark inode dirty even if
3150 			 * new_i_size is less that inode->i_size
3151 			 * bu greater than i_disksize.(hint delalloc)
3152 			 */
3153 			ext4_mark_inode_dirty(handle, inode);
3154 		}
3155 	}
3156 
3157 	if (write_mode != CONVERT_INLINE_DATA &&
3158 	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3159 	    ext4_has_inline_data(inode))
3160 		ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3161 						     page);
3162 	else
3163 		ret2 = generic_write_end(file, mapping, pos, len, copied,
3164 							page, fsdata);
3165 
3166 	copied = ret2;
3167 	if (ret2 < 0)
3168 		ret = ret2;
3169 	ret2 = ext4_journal_stop(handle);
3170 	if (!ret)
3171 		ret = ret2;
3172 
3173 	return ret ? ret : copied;
3174 }
3175 
3176 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3177 				   unsigned int length)
3178 {
3179 	/*
3180 	 * Drop reserved blocks
3181 	 */
3182 	BUG_ON(!PageLocked(page));
3183 	if (!page_has_buffers(page))
3184 		goto out;
3185 
3186 	ext4_da_page_release_reservation(page, offset, length);
3187 
3188 out:
3189 	ext4_invalidatepage(page, offset, length);
3190 
3191 	return;
3192 }
3193 
3194 /*
3195  * Force all delayed allocation blocks to be allocated for a given inode.
3196  */
3197 int ext4_alloc_da_blocks(struct inode *inode)
3198 {
3199 	trace_ext4_alloc_da_blocks(inode);
3200 
3201 	if (!EXT4_I(inode)->i_reserved_data_blocks)
3202 		return 0;
3203 
3204 	/*
3205 	 * We do something simple for now.  The filemap_flush() will
3206 	 * also start triggering a write of the data blocks, which is
3207 	 * not strictly speaking necessary (and for users of
3208 	 * laptop_mode, not even desirable).  However, to do otherwise
3209 	 * would require replicating code paths in:
3210 	 *
3211 	 * ext4_writepages() ->
3212 	 *    write_cache_pages() ---> (via passed in callback function)
3213 	 *        __mpage_da_writepage() -->
3214 	 *           mpage_add_bh_to_extent()
3215 	 *           mpage_da_map_blocks()
3216 	 *
3217 	 * The problem is that write_cache_pages(), located in
3218 	 * mm/page-writeback.c, marks pages clean in preparation for
3219 	 * doing I/O, which is not desirable if we're not planning on
3220 	 * doing I/O at all.
3221 	 *
3222 	 * We could call write_cache_pages(), and then redirty all of
3223 	 * the pages by calling redirty_page_for_writepage() but that
3224 	 * would be ugly in the extreme.  So instead we would need to
3225 	 * replicate parts of the code in the above functions,
3226 	 * simplifying them because we wouldn't actually intend to
3227 	 * write out the pages, but rather only collect contiguous
3228 	 * logical block extents, call the multi-block allocator, and
3229 	 * then update the buffer heads with the block allocations.
3230 	 *
3231 	 * For now, though, we'll cheat by calling filemap_flush(),
3232 	 * which will map the blocks, and start the I/O, but not
3233 	 * actually wait for the I/O to complete.
3234 	 */
3235 	return filemap_flush(inode->i_mapping);
3236 }
3237 
3238 /*
3239  * bmap() is special.  It gets used by applications such as lilo and by
3240  * the swapper to find the on-disk block of a specific piece of data.
3241  *
3242  * Naturally, this is dangerous if the block concerned is still in the
3243  * journal.  If somebody makes a swapfile on an ext4 data-journaling
3244  * filesystem and enables swap, then they may get a nasty shock when the
3245  * data getting swapped to that swapfile suddenly gets overwritten by
3246  * the original zero's written out previously to the journal and
3247  * awaiting writeback in the kernel's buffer cache.
3248  *
3249  * So, if we see any bmap calls here on a modified, data-journaled file,
3250  * take extra steps to flush any blocks which might be in the cache.
3251  */
3252 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3253 {
3254 	struct inode *inode = mapping->host;
3255 	journal_t *journal;
3256 	int err;
3257 
3258 	/*
3259 	 * We can get here for an inline file via the FIBMAP ioctl
3260 	 */
3261 	if (ext4_has_inline_data(inode))
3262 		return 0;
3263 
3264 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3265 			test_opt(inode->i_sb, DELALLOC)) {
3266 		/*
3267 		 * With delalloc we want to sync the file
3268 		 * so that we can make sure we allocate
3269 		 * blocks for file
3270 		 */
3271 		filemap_write_and_wait(mapping);
3272 	}
3273 
3274 	if (EXT4_JOURNAL(inode) &&
3275 	    ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3276 		/*
3277 		 * This is a REALLY heavyweight approach, but the use of
3278 		 * bmap on dirty files is expected to be extremely rare:
3279 		 * only if we run lilo or swapon on a freshly made file
3280 		 * do we expect this to happen.
3281 		 *
3282 		 * (bmap requires CAP_SYS_RAWIO so this does not
3283 		 * represent an unprivileged user DOS attack --- we'd be
3284 		 * in trouble if mortal users could trigger this path at
3285 		 * will.)
3286 		 *
3287 		 * NB. EXT4_STATE_JDATA is not set on files other than
3288 		 * regular files.  If somebody wants to bmap a directory
3289 		 * or symlink and gets confused because the buffer
3290 		 * hasn't yet been flushed to disk, they deserve
3291 		 * everything they get.
3292 		 */
3293 
3294 		ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3295 		journal = EXT4_JOURNAL(inode);
3296 		jbd2_journal_lock_updates(journal);
3297 		err = jbd2_journal_flush(journal);
3298 		jbd2_journal_unlock_updates(journal);
3299 
3300 		if (err)
3301 			return 0;
3302 	}
3303 
3304 	return generic_block_bmap(mapping, block, ext4_get_block);
3305 }
3306 
3307 static int ext4_readpage(struct file *file, struct page *page)
3308 {
3309 	int ret = -EAGAIN;
3310 	struct inode *inode = page->mapping->host;
3311 
3312 	trace_ext4_readpage(page);
3313 
3314 	if (ext4_has_inline_data(inode))
3315 		ret = ext4_readpage_inline(inode, page);
3316 
3317 	if (ret == -EAGAIN)
3318 		return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3319 
3320 	return ret;
3321 }
3322 
3323 static int
3324 ext4_readpages(struct file *file, struct address_space *mapping,
3325 		struct list_head *pages, unsigned nr_pages)
3326 {
3327 	struct inode *inode = mapping->host;
3328 
3329 	/* If the file has inline data, no need to do readpages. */
3330 	if (ext4_has_inline_data(inode))
3331 		return 0;
3332 
3333 	return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3334 }
3335 
3336 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3337 				unsigned int length)
3338 {
3339 	trace_ext4_invalidatepage(page, offset, length);
3340 
3341 	/* No journalling happens on data buffers when this function is used */
3342 	WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3343 
3344 	block_invalidatepage(page, offset, length);
3345 }
3346 
3347 static int __ext4_journalled_invalidatepage(struct page *page,
3348 					    unsigned int offset,
3349 					    unsigned int length)
3350 {
3351 	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3352 
3353 	trace_ext4_journalled_invalidatepage(page, offset, length);
3354 
3355 	/*
3356 	 * If it's a full truncate we just forget about the pending dirtying
3357 	 */
3358 	if (offset == 0 && length == PAGE_SIZE)
3359 		ClearPageChecked(page);
3360 
3361 	return jbd2_journal_invalidatepage(journal, page, offset, length);
3362 }
3363 
3364 /* Wrapper for aops... */
3365 static void ext4_journalled_invalidatepage(struct page *page,
3366 					   unsigned int offset,
3367 					   unsigned int length)
3368 {
3369 	WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3370 }
3371 
3372 static int ext4_releasepage(struct page *page, gfp_t wait)
3373 {
3374 	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3375 
3376 	trace_ext4_releasepage(page);
3377 
3378 	/* Page has dirty journalled data -> cannot release */
3379 	if (PageChecked(page))
3380 		return 0;
3381 	if (journal)
3382 		return jbd2_journal_try_to_free_buffers(journal, page, wait);
3383 	else
3384 		return try_to_free_buffers(page);
3385 }
3386 
3387 static bool ext4_inode_datasync_dirty(struct inode *inode)
3388 {
3389 	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3390 
3391 	if (journal)
3392 		return !jbd2_transaction_committed(journal,
3393 					EXT4_I(inode)->i_datasync_tid);
3394 	/* Any metadata buffers to write? */
3395 	if (!list_empty(&inode->i_mapping->private_list))
3396 		return true;
3397 	return inode->i_state & I_DIRTY_DATASYNC;
3398 }
3399 
3400 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3401 			    unsigned flags, struct iomap *iomap)
3402 {
3403 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3404 	unsigned int blkbits = inode->i_blkbits;
3405 	unsigned long first_block = offset >> blkbits;
3406 	unsigned long last_block = (offset + length - 1) >> blkbits;
3407 	struct ext4_map_blocks map;
3408 	bool delalloc = false;
3409 	int ret;
3410 
3411 
3412 	if (flags & IOMAP_REPORT) {
3413 		if (ext4_has_inline_data(inode)) {
3414 			ret = ext4_inline_data_iomap(inode, iomap);
3415 			if (ret != -EAGAIN) {
3416 				if (ret == 0 && offset >= iomap->length)
3417 					ret = -ENOENT;
3418 				return ret;
3419 			}
3420 		}
3421 	} else {
3422 		if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3423 			return -ERANGE;
3424 	}
3425 
3426 	map.m_lblk = first_block;
3427 	map.m_len = last_block - first_block + 1;
3428 
3429 	if (flags & IOMAP_REPORT) {
3430 		ret = ext4_map_blocks(NULL, inode, &map, 0);
3431 		if (ret < 0)
3432 			return ret;
3433 
3434 		if (ret == 0) {
3435 			ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3436 			struct extent_status es;
3437 
3438 			ext4_es_find_delayed_extent_range(inode, map.m_lblk, end, &es);
3439 
3440 			if (!es.es_len || es.es_lblk > end) {
3441 				/* entire range is a hole */
3442 			} else if (es.es_lblk > map.m_lblk) {
3443 				/* range starts with a hole */
3444 				map.m_len = es.es_lblk - map.m_lblk;
3445 			} else {
3446 				ext4_lblk_t offs = 0;
3447 
3448 				if (es.es_lblk < map.m_lblk)
3449 					offs = map.m_lblk - es.es_lblk;
3450 				map.m_lblk = es.es_lblk + offs;
3451 				map.m_len = es.es_len - offs;
3452 				delalloc = true;
3453 			}
3454 		}
3455 	} else if (flags & IOMAP_WRITE) {
3456 		int dio_credits;
3457 		handle_t *handle;
3458 		int retries = 0;
3459 
3460 		/* Trim mapping request to maximum we can map at once for DIO */
3461 		if (map.m_len > DIO_MAX_BLOCKS)
3462 			map.m_len = DIO_MAX_BLOCKS;
3463 		dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3464 retry:
3465 		/*
3466 		 * Either we allocate blocks and then we don't get unwritten
3467 		 * extent so we have reserved enough credits, or the blocks
3468 		 * are already allocated and unwritten and in that case
3469 		 * extent conversion fits in the credits as well.
3470 		 */
3471 		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3472 					    dio_credits);
3473 		if (IS_ERR(handle))
3474 			return PTR_ERR(handle);
3475 
3476 		ret = ext4_map_blocks(handle, inode, &map,
3477 				      EXT4_GET_BLOCKS_CREATE_ZERO);
3478 		if (ret < 0) {
3479 			ext4_journal_stop(handle);
3480 			if (ret == -ENOSPC &&
3481 			    ext4_should_retry_alloc(inode->i_sb, &retries))
3482 				goto retry;
3483 			return ret;
3484 		}
3485 
3486 		/*
3487 		 * If we added blocks beyond i_size, we need to make sure they
3488 		 * will get truncated if we crash before updating i_size in
3489 		 * ext4_iomap_end(). For faults we don't need to do that (and
3490 		 * even cannot because for orphan list operations inode_lock is
3491 		 * required) - if we happen to instantiate block beyond i_size,
3492 		 * it is because we race with truncate which has already added
3493 		 * the inode to the orphan list.
3494 		 */
3495 		if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3496 		    (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3497 			int err;
3498 
3499 			err = ext4_orphan_add(handle, inode);
3500 			if (err < 0) {
3501 				ext4_journal_stop(handle);
3502 				return err;
3503 			}
3504 		}
3505 		ext4_journal_stop(handle);
3506 	} else {
3507 		ret = ext4_map_blocks(NULL, inode, &map, 0);
3508 		if (ret < 0)
3509 			return ret;
3510 	}
3511 
3512 	iomap->flags = 0;
3513 	if (ext4_inode_datasync_dirty(inode))
3514 		iomap->flags |= IOMAP_F_DIRTY;
3515 	iomap->bdev = inode->i_sb->s_bdev;
3516 	iomap->dax_dev = sbi->s_daxdev;
3517 	iomap->offset = first_block << blkbits;
3518 	iomap->length = (u64)map.m_len << blkbits;
3519 
3520 	if (ret == 0) {
3521 		iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3522 		iomap->addr = IOMAP_NULL_ADDR;
3523 	} else {
3524 		if (map.m_flags & EXT4_MAP_MAPPED) {
3525 			iomap->type = IOMAP_MAPPED;
3526 		} else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3527 			iomap->type = IOMAP_UNWRITTEN;
3528 		} else {
3529 			WARN_ON_ONCE(1);
3530 			return -EIO;
3531 		}
3532 		iomap->addr = (u64)map.m_pblk << blkbits;
3533 	}
3534 
3535 	if (map.m_flags & EXT4_MAP_NEW)
3536 		iomap->flags |= IOMAP_F_NEW;
3537 
3538 	return 0;
3539 }
3540 
3541 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3542 			  ssize_t written, unsigned flags, struct iomap *iomap)
3543 {
3544 	int ret = 0;
3545 	handle_t *handle;
3546 	int blkbits = inode->i_blkbits;
3547 	bool truncate = false;
3548 
3549 	if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3550 		return 0;
3551 
3552 	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3553 	if (IS_ERR(handle)) {
3554 		ret = PTR_ERR(handle);
3555 		goto orphan_del;
3556 	}
3557 	if (ext4_update_inode_size(inode, offset + written))
3558 		ext4_mark_inode_dirty(handle, inode);
3559 	/*
3560 	 * We may need to truncate allocated but not written blocks beyond EOF.
3561 	 */
3562 	if (iomap->offset + iomap->length >
3563 	    ALIGN(inode->i_size, 1 << blkbits)) {
3564 		ext4_lblk_t written_blk, end_blk;
3565 
3566 		written_blk = (offset + written) >> blkbits;
3567 		end_blk = (offset + length) >> blkbits;
3568 		if (written_blk < end_blk && ext4_can_truncate(inode))
3569 			truncate = true;
3570 	}
3571 	/*
3572 	 * Remove inode from orphan list if we were extending a inode and
3573 	 * everything went fine.
3574 	 */
3575 	if (!truncate && inode->i_nlink &&
3576 	    !list_empty(&EXT4_I(inode)->i_orphan))
3577 		ext4_orphan_del(handle, inode);
3578 	ext4_journal_stop(handle);
3579 	if (truncate) {
3580 		ext4_truncate_failed_write(inode);
3581 orphan_del:
3582 		/*
3583 		 * If truncate failed early the inode might still be on the
3584 		 * orphan list; we need to make sure the inode is removed from
3585 		 * the orphan list in that case.
3586 		 */
3587 		if (inode->i_nlink)
3588 			ext4_orphan_del(NULL, inode);
3589 	}
3590 	return ret;
3591 }
3592 
3593 const struct iomap_ops ext4_iomap_ops = {
3594 	.iomap_begin		= ext4_iomap_begin,
3595 	.iomap_end		= ext4_iomap_end,
3596 };
3597 
3598 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3599 			    ssize_t size, void *private)
3600 {
3601         ext4_io_end_t *io_end = private;
3602 
3603 	/* if not async direct IO just return */
3604 	if (!io_end)
3605 		return 0;
3606 
3607 	ext_debug("ext4_end_io_dio(): io_end 0x%p "
3608 		  "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3609 		  io_end, io_end->inode->i_ino, iocb, offset, size);
3610 
3611 	/*
3612 	 * Error during AIO DIO. We cannot convert unwritten extents as the
3613 	 * data was not written. Just clear the unwritten flag and drop io_end.
3614 	 */
3615 	if (size <= 0) {
3616 		ext4_clear_io_unwritten_flag(io_end);
3617 		size = 0;
3618 	}
3619 	io_end->offset = offset;
3620 	io_end->size = size;
3621 	ext4_put_io_end(io_end);
3622 
3623 	return 0;
3624 }
3625 
3626 /*
3627  * Handling of direct IO writes.
3628  *
3629  * For ext4 extent files, ext4 will do direct-io write even to holes,
3630  * preallocated extents, and those write extend the file, no need to
3631  * fall back to buffered IO.
3632  *
3633  * For holes, we fallocate those blocks, mark them as unwritten
3634  * If those blocks were preallocated, we mark sure they are split, but
3635  * still keep the range to write as unwritten.
3636  *
3637  * The unwritten extents will be converted to written when DIO is completed.
3638  * For async direct IO, since the IO may still pending when return, we
3639  * set up an end_io call back function, which will do the conversion
3640  * when async direct IO completed.
3641  *
3642  * If the O_DIRECT write will extend the file then add this inode to the
3643  * orphan list.  So recovery will truncate it back to the original size
3644  * if the machine crashes during the write.
3645  *
3646  */
3647 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3648 {
3649 	struct file *file = iocb->ki_filp;
3650 	struct inode *inode = file->f_mapping->host;
3651 	struct ext4_inode_info *ei = EXT4_I(inode);
3652 	ssize_t ret;
3653 	loff_t offset = iocb->ki_pos;
3654 	size_t count = iov_iter_count(iter);
3655 	int overwrite = 0;
3656 	get_block_t *get_block_func = NULL;
3657 	int dio_flags = 0;
3658 	loff_t final_size = offset + count;
3659 	int orphan = 0;
3660 	handle_t *handle;
3661 
3662 	if (final_size > inode->i_size) {
3663 		/* Credits for sb + inode write */
3664 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3665 		if (IS_ERR(handle)) {
3666 			ret = PTR_ERR(handle);
3667 			goto out;
3668 		}
3669 		ret = ext4_orphan_add(handle, inode);
3670 		if (ret) {
3671 			ext4_journal_stop(handle);
3672 			goto out;
3673 		}
3674 		orphan = 1;
3675 		ei->i_disksize = inode->i_size;
3676 		ext4_journal_stop(handle);
3677 	}
3678 
3679 	BUG_ON(iocb->private == NULL);
3680 
3681 	/*
3682 	 * Make all waiters for direct IO properly wait also for extent
3683 	 * conversion. This also disallows race between truncate() and
3684 	 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3685 	 */
3686 	inode_dio_begin(inode);
3687 
3688 	/* If we do a overwrite dio, i_mutex locking can be released */
3689 	overwrite = *((int *)iocb->private);
3690 
3691 	if (overwrite)
3692 		inode_unlock(inode);
3693 
3694 	/*
3695 	 * For extent mapped files we could direct write to holes and fallocate.
3696 	 *
3697 	 * Allocated blocks to fill the hole are marked as unwritten to prevent
3698 	 * parallel buffered read to expose the stale data before DIO complete
3699 	 * the data IO.
3700 	 *
3701 	 * As to previously fallocated extents, ext4 get_block will just simply
3702 	 * mark the buffer mapped but still keep the extents unwritten.
3703 	 *
3704 	 * For non AIO case, we will convert those unwritten extents to written
3705 	 * after return back from blockdev_direct_IO. That way we save us from
3706 	 * allocating io_end structure and also the overhead of offloading
3707 	 * the extent convertion to a workqueue.
3708 	 *
3709 	 * For async DIO, the conversion needs to be deferred when the
3710 	 * IO is completed. The ext4 end_io callback function will be
3711 	 * called to take care of the conversion work.  Here for async
3712 	 * case, we allocate an io_end structure to hook to the iocb.
3713 	 */
3714 	iocb->private = NULL;
3715 	if (overwrite)
3716 		get_block_func = ext4_dio_get_block_overwrite;
3717 	else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3718 		   round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3719 		get_block_func = ext4_dio_get_block;
3720 		dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3721 	} else if (is_sync_kiocb(iocb)) {
3722 		get_block_func = ext4_dio_get_block_unwritten_sync;
3723 		dio_flags = DIO_LOCKING;
3724 	} else {
3725 		get_block_func = ext4_dio_get_block_unwritten_async;
3726 		dio_flags = DIO_LOCKING;
3727 	}
3728 	ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3729 				   get_block_func, ext4_end_io_dio, NULL,
3730 				   dio_flags);
3731 
3732 	if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3733 						EXT4_STATE_DIO_UNWRITTEN)) {
3734 		int err;
3735 		/*
3736 		 * for non AIO case, since the IO is already
3737 		 * completed, we could do the conversion right here
3738 		 */
3739 		err = ext4_convert_unwritten_extents(NULL, inode,
3740 						     offset, ret);
3741 		if (err < 0)
3742 			ret = err;
3743 		ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3744 	}
3745 
3746 	inode_dio_end(inode);
3747 	/* take i_mutex locking again if we do a ovewrite dio */
3748 	if (overwrite)
3749 		inode_lock(inode);
3750 
3751 	if (ret < 0 && final_size > inode->i_size)
3752 		ext4_truncate_failed_write(inode);
3753 
3754 	/* Handle extending of i_size after direct IO write */
3755 	if (orphan) {
3756 		int err;
3757 
3758 		/* Credits for sb + inode write */
3759 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3760 		if (IS_ERR(handle)) {
3761 			/* This is really bad luck. We've written the data
3762 			 * but cannot extend i_size. Bail out and pretend
3763 			 * the write failed... */
3764 			ret = PTR_ERR(handle);
3765 			if (inode->i_nlink)
3766 				ext4_orphan_del(NULL, inode);
3767 
3768 			goto out;
3769 		}
3770 		if (inode->i_nlink)
3771 			ext4_orphan_del(handle, inode);
3772 		if (ret > 0) {
3773 			loff_t end = offset + ret;
3774 			if (end > inode->i_size) {
3775 				ei->i_disksize = end;
3776 				i_size_write(inode, end);
3777 				/*
3778 				 * We're going to return a positive `ret'
3779 				 * here due to non-zero-length I/O, so there's
3780 				 * no way of reporting error returns from
3781 				 * ext4_mark_inode_dirty() to userspace.  So
3782 				 * ignore it.
3783 				 */
3784 				ext4_mark_inode_dirty(handle, inode);
3785 			}
3786 		}
3787 		err = ext4_journal_stop(handle);
3788 		if (ret == 0)
3789 			ret = err;
3790 	}
3791 out:
3792 	return ret;
3793 }
3794 
3795 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3796 {
3797 	struct address_space *mapping = iocb->ki_filp->f_mapping;
3798 	struct inode *inode = mapping->host;
3799 	size_t count = iov_iter_count(iter);
3800 	ssize_t ret;
3801 
3802 	/*
3803 	 * Shared inode_lock is enough for us - it protects against concurrent
3804 	 * writes & truncates and since we take care of writing back page cache,
3805 	 * we are protected against page writeback as well.
3806 	 */
3807 	inode_lock_shared(inode);
3808 	ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3809 					   iocb->ki_pos + count - 1);
3810 	if (ret)
3811 		goto out_unlock;
3812 	ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3813 				   iter, ext4_dio_get_block, NULL, NULL, 0);
3814 out_unlock:
3815 	inode_unlock_shared(inode);
3816 	return ret;
3817 }
3818 
3819 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3820 {
3821 	struct file *file = iocb->ki_filp;
3822 	struct inode *inode = file->f_mapping->host;
3823 	size_t count = iov_iter_count(iter);
3824 	loff_t offset = iocb->ki_pos;
3825 	ssize_t ret;
3826 
3827 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3828 	if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3829 		return 0;
3830 #endif
3831 
3832 	/*
3833 	 * If we are doing data journalling we don't support O_DIRECT
3834 	 */
3835 	if (ext4_should_journal_data(inode))
3836 		return 0;
3837 
3838 	/* Let buffer I/O handle the inline data case. */
3839 	if (ext4_has_inline_data(inode))
3840 		return 0;
3841 
3842 	/* DAX uses iomap path now */
3843 	if (WARN_ON_ONCE(IS_DAX(inode)))
3844 		return 0;
3845 
3846 	trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3847 	if (iov_iter_rw(iter) == READ)
3848 		ret = ext4_direct_IO_read(iocb, iter);
3849 	else
3850 		ret = ext4_direct_IO_write(iocb, iter);
3851 	trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3852 	return ret;
3853 }
3854 
3855 /*
3856  * Pages can be marked dirty completely asynchronously from ext4's journalling
3857  * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3858  * much here because ->set_page_dirty is called under VFS locks.  The page is
3859  * not necessarily locked.
3860  *
3861  * We cannot just dirty the page and leave attached buffers clean, because the
3862  * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3863  * or jbddirty because all the journalling code will explode.
3864  *
3865  * So what we do is to mark the page "pending dirty" and next time writepage
3866  * is called, propagate that into the buffers appropriately.
3867  */
3868 static int ext4_journalled_set_page_dirty(struct page *page)
3869 {
3870 	SetPageChecked(page);
3871 	return __set_page_dirty_nobuffers(page);
3872 }
3873 
3874 static int ext4_set_page_dirty(struct page *page)
3875 {
3876 	WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3877 	WARN_ON_ONCE(!page_has_buffers(page));
3878 	return __set_page_dirty_buffers(page);
3879 }
3880 
3881 static const struct address_space_operations ext4_aops = {
3882 	.readpage		= ext4_readpage,
3883 	.readpages		= ext4_readpages,
3884 	.writepage		= ext4_writepage,
3885 	.writepages		= ext4_writepages,
3886 	.write_begin		= ext4_write_begin,
3887 	.write_end		= ext4_write_end,
3888 	.set_page_dirty		= ext4_set_page_dirty,
3889 	.bmap			= ext4_bmap,
3890 	.invalidatepage		= ext4_invalidatepage,
3891 	.releasepage		= ext4_releasepage,
3892 	.direct_IO		= ext4_direct_IO,
3893 	.migratepage		= buffer_migrate_page,
3894 	.is_partially_uptodate  = block_is_partially_uptodate,
3895 	.error_remove_page	= generic_error_remove_page,
3896 };
3897 
3898 static const struct address_space_operations ext4_journalled_aops = {
3899 	.readpage		= ext4_readpage,
3900 	.readpages		= ext4_readpages,
3901 	.writepage		= ext4_writepage,
3902 	.writepages		= ext4_writepages,
3903 	.write_begin		= ext4_write_begin,
3904 	.write_end		= ext4_journalled_write_end,
3905 	.set_page_dirty		= ext4_journalled_set_page_dirty,
3906 	.bmap			= ext4_bmap,
3907 	.invalidatepage		= ext4_journalled_invalidatepage,
3908 	.releasepage		= ext4_releasepage,
3909 	.direct_IO		= ext4_direct_IO,
3910 	.is_partially_uptodate  = block_is_partially_uptodate,
3911 	.error_remove_page	= generic_error_remove_page,
3912 };
3913 
3914 static const struct address_space_operations ext4_da_aops = {
3915 	.readpage		= ext4_readpage,
3916 	.readpages		= ext4_readpages,
3917 	.writepage		= ext4_writepage,
3918 	.writepages		= ext4_writepages,
3919 	.write_begin		= ext4_da_write_begin,
3920 	.write_end		= ext4_da_write_end,
3921 	.set_page_dirty		= ext4_set_page_dirty,
3922 	.bmap			= ext4_bmap,
3923 	.invalidatepage		= ext4_da_invalidatepage,
3924 	.releasepage		= ext4_releasepage,
3925 	.direct_IO		= ext4_direct_IO,
3926 	.migratepage		= buffer_migrate_page,
3927 	.is_partially_uptodate  = block_is_partially_uptodate,
3928 	.error_remove_page	= generic_error_remove_page,
3929 };
3930 
3931 void ext4_set_aops(struct inode *inode)
3932 {
3933 	switch (ext4_inode_journal_mode(inode)) {
3934 	case EXT4_INODE_ORDERED_DATA_MODE:
3935 	case EXT4_INODE_WRITEBACK_DATA_MODE:
3936 		break;
3937 	case EXT4_INODE_JOURNAL_DATA_MODE:
3938 		inode->i_mapping->a_ops = &ext4_journalled_aops;
3939 		return;
3940 	default:
3941 		BUG();
3942 	}
3943 	if (test_opt(inode->i_sb, DELALLOC))
3944 		inode->i_mapping->a_ops = &ext4_da_aops;
3945 	else
3946 		inode->i_mapping->a_ops = &ext4_aops;
3947 }
3948 
3949 static int __ext4_block_zero_page_range(handle_t *handle,
3950 		struct address_space *mapping, loff_t from, loff_t length)
3951 {
3952 	ext4_fsblk_t index = from >> PAGE_SHIFT;
3953 	unsigned offset = from & (PAGE_SIZE-1);
3954 	unsigned blocksize, pos;
3955 	ext4_lblk_t iblock;
3956 	struct inode *inode = mapping->host;
3957 	struct buffer_head *bh;
3958 	struct page *page;
3959 	int err = 0;
3960 
3961 	page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3962 				   mapping_gfp_constraint(mapping, ~__GFP_FS));
3963 	if (!page)
3964 		return -ENOMEM;
3965 
3966 	blocksize = inode->i_sb->s_blocksize;
3967 
3968 	iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3969 
3970 	if (!page_has_buffers(page))
3971 		create_empty_buffers(page, blocksize, 0);
3972 
3973 	/* Find the buffer that contains "offset" */
3974 	bh = page_buffers(page);
3975 	pos = blocksize;
3976 	while (offset >= pos) {
3977 		bh = bh->b_this_page;
3978 		iblock++;
3979 		pos += blocksize;
3980 	}
3981 	if (buffer_freed(bh)) {
3982 		BUFFER_TRACE(bh, "freed: skip");
3983 		goto unlock;
3984 	}
3985 	if (!buffer_mapped(bh)) {
3986 		BUFFER_TRACE(bh, "unmapped");
3987 		ext4_get_block(inode, iblock, bh, 0);
3988 		/* unmapped? It's a hole - nothing to do */
3989 		if (!buffer_mapped(bh)) {
3990 			BUFFER_TRACE(bh, "still unmapped");
3991 			goto unlock;
3992 		}
3993 	}
3994 
3995 	/* Ok, it's mapped. Make sure it's up-to-date */
3996 	if (PageUptodate(page))
3997 		set_buffer_uptodate(bh);
3998 
3999 	if (!buffer_uptodate(bh)) {
4000 		err = -EIO;
4001 		ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4002 		wait_on_buffer(bh);
4003 		/* Uhhuh. Read error. Complain and punt. */
4004 		if (!buffer_uptodate(bh))
4005 			goto unlock;
4006 		if (S_ISREG(inode->i_mode) &&
4007 		    ext4_encrypted_inode(inode)) {
4008 			/* We expect the key to be set. */
4009 			BUG_ON(!fscrypt_has_encryption_key(inode));
4010 			BUG_ON(blocksize != PAGE_SIZE);
4011 			WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4012 						page, PAGE_SIZE, 0, page->index));
4013 		}
4014 	}
4015 	if (ext4_should_journal_data(inode)) {
4016 		BUFFER_TRACE(bh, "get write access");
4017 		err = ext4_journal_get_write_access(handle, bh);
4018 		if (err)
4019 			goto unlock;
4020 	}
4021 	zero_user(page, offset, length);
4022 	BUFFER_TRACE(bh, "zeroed end of block");
4023 
4024 	if (ext4_should_journal_data(inode)) {
4025 		err = ext4_handle_dirty_metadata(handle, inode, bh);
4026 	} else {
4027 		err = 0;
4028 		mark_buffer_dirty(bh);
4029 		if (ext4_should_order_data(inode))
4030 			err = ext4_jbd2_inode_add_write(handle, inode);
4031 	}
4032 
4033 unlock:
4034 	unlock_page(page);
4035 	put_page(page);
4036 	return err;
4037 }
4038 
4039 /*
4040  * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4041  * starting from file offset 'from'.  The range to be zero'd must
4042  * be contained with in one block.  If the specified range exceeds
4043  * the end of the block it will be shortened to end of the block
4044  * that cooresponds to 'from'
4045  */
4046 static int ext4_block_zero_page_range(handle_t *handle,
4047 		struct address_space *mapping, loff_t from, loff_t length)
4048 {
4049 	struct inode *inode = mapping->host;
4050 	unsigned offset = from & (PAGE_SIZE-1);
4051 	unsigned blocksize = inode->i_sb->s_blocksize;
4052 	unsigned max = blocksize - (offset & (blocksize - 1));
4053 
4054 	/*
4055 	 * correct length if it does not fall between
4056 	 * 'from' and the end of the block
4057 	 */
4058 	if (length > max || length < 0)
4059 		length = max;
4060 
4061 	if (IS_DAX(inode)) {
4062 		return iomap_zero_range(inode, from, length, NULL,
4063 					&ext4_iomap_ops);
4064 	}
4065 	return __ext4_block_zero_page_range(handle, mapping, from, length);
4066 }
4067 
4068 /*
4069  * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4070  * up to the end of the block which corresponds to `from'.
4071  * This required during truncate. We need to physically zero the tail end
4072  * of that block so it doesn't yield old data if the file is later grown.
4073  */
4074 static int ext4_block_truncate_page(handle_t *handle,
4075 		struct address_space *mapping, loff_t from)
4076 {
4077 	unsigned offset = from & (PAGE_SIZE-1);
4078 	unsigned length;
4079 	unsigned blocksize;
4080 	struct inode *inode = mapping->host;
4081 
4082 	/* If we are processing an encrypted inode during orphan list handling */
4083 	if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4084 		return 0;
4085 
4086 	blocksize = inode->i_sb->s_blocksize;
4087 	length = blocksize - (offset & (blocksize - 1));
4088 
4089 	return ext4_block_zero_page_range(handle, mapping, from, length);
4090 }
4091 
4092 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4093 			     loff_t lstart, loff_t length)
4094 {
4095 	struct super_block *sb = inode->i_sb;
4096 	struct address_space *mapping = inode->i_mapping;
4097 	unsigned partial_start, partial_end;
4098 	ext4_fsblk_t start, end;
4099 	loff_t byte_end = (lstart + length - 1);
4100 	int err = 0;
4101 
4102 	partial_start = lstart & (sb->s_blocksize - 1);
4103 	partial_end = byte_end & (sb->s_blocksize - 1);
4104 
4105 	start = lstart >> sb->s_blocksize_bits;
4106 	end = byte_end >> sb->s_blocksize_bits;
4107 
4108 	/* Handle partial zero within the single block */
4109 	if (start == end &&
4110 	    (partial_start || (partial_end != sb->s_blocksize - 1))) {
4111 		err = ext4_block_zero_page_range(handle, mapping,
4112 						 lstart, length);
4113 		return err;
4114 	}
4115 	/* Handle partial zero out on the start of the range */
4116 	if (partial_start) {
4117 		err = ext4_block_zero_page_range(handle, mapping,
4118 						 lstart, sb->s_blocksize);
4119 		if (err)
4120 			return err;
4121 	}
4122 	/* Handle partial zero out on the end of the range */
4123 	if (partial_end != sb->s_blocksize - 1)
4124 		err = ext4_block_zero_page_range(handle, mapping,
4125 						 byte_end - partial_end,
4126 						 partial_end + 1);
4127 	return err;
4128 }
4129 
4130 int ext4_can_truncate(struct inode *inode)
4131 {
4132 	if (S_ISREG(inode->i_mode))
4133 		return 1;
4134 	if (S_ISDIR(inode->i_mode))
4135 		return 1;
4136 	if (S_ISLNK(inode->i_mode))
4137 		return !ext4_inode_is_fast_symlink(inode);
4138 	return 0;
4139 }
4140 
4141 /*
4142  * We have to make sure i_disksize gets properly updated before we truncate
4143  * page cache due to hole punching or zero range. Otherwise i_disksize update
4144  * can get lost as it may have been postponed to submission of writeback but
4145  * that will never happen after we truncate page cache.
4146  */
4147 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4148 				      loff_t len)
4149 {
4150 	handle_t *handle;
4151 	loff_t size = i_size_read(inode);
4152 
4153 	WARN_ON(!inode_is_locked(inode));
4154 	if (offset > size || offset + len < size)
4155 		return 0;
4156 
4157 	if (EXT4_I(inode)->i_disksize >= size)
4158 		return 0;
4159 
4160 	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4161 	if (IS_ERR(handle))
4162 		return PTR_ERR(handle);
4163 	ext4_update_i_disksize(inode, size);
4164 	ext4_mark_inode_dirty(handle, inode);
4165 	ext4_journal_stop(handle);
4166 
4167 	return 0;
4168 }
4169 
4170 /*
4171  * ext4_punch_hole: punches a hole in a file by releasing the blocks
4172  * associated with the given offset and length
4173  *
4174  * @inode:  File inode
4175  * @offset: The offset where the hole will begin
4176  * @len:    The length of the hole
4177  *
4178  * Returns: 0 on success or negative on failure
4179  */
4180 
4181 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4182 {
4183 	struct super_block *sb = inode->i_sb;
4184 	ext4_lblk_t first_block, stop_block;
4185 	struct address_space *mapping = inode->i_mapping;
4186 	loff_t first_block_offset, last_block_offset;
4187 	handle_t *handle;
4188 	unsigned int credits;
4189 	int ret = 0;
4190 
4191 	if (!S_ISREG(inode->i_mode))
4192 		return -EOPNOTSUPP;
4193 
4194 	trace_ext4_punch_hole(inode, offset, length, 0);
4195 
4196 	/*
4197 	 * Write out all dirty pages to avoid race conditions
4198 	 * Then release them.
4199 	 */
4200 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4201 		ret = filemap_write_and_wait_range(mapping, offset,
4202 						   offset + length - 1);
4203 		if (ret)
4204 			return ret;
4205 	}
4206 
4207 	inode_lock(inode);
4208 
4209 	/* No need to punch hole beyond i_size */
4210 	if (offset >= inode->i_size)
4211 		goto out_mutex;
4212 
4213 	/*
4214 	 * If the hole extends beyond i_size, set the hole
4215 	 * to end after the page that contains i_size
4216 	 */
4217 	if (offset + length > inode->i_size) {
4218 		length = inode->i_size +
4219 		   PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4220 		   offset;
4221 	}
4222 
4223 	if (offset & (sb->s_blocksize - 1) ||
4224 	    (offset + length) & (sb->s_blocksize - 1)) {
4225 		/*
4226 		 * Attach jinode to inode for jbd2 if we do any zeroing of
4227 		 * partial block
4228 		 */
4229 		ret = ext4_inode_attach_jinode(inode);
4230 		if (ret < 0)
4231 			goto out_mutex;
4232 
4233 	}
4234 
4235 	/* Wait all existing dio workers, newcomers will block on i_mutex */
4236 	ext4_inode_block_unlocked_dio(inode);
4237 	inode_dio_wait(inode);
4238 
4239 	/*
4240 	 * Prevent page faults from reinstantiating pages we have released from
4241 	 * page cache.
4242 	 */
4243 	down_write(&EXT4_I(inode)->i_mmap_sem);
4244 	first_block_offset = round_up(offset, sb->s_blocksize);
4245 	last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4246 
4247 	/* Now release the pages and zero block aligned part of pages*/
4248 	if (last_block_offset > first_block_offset) {
4249 		ret = ext4_update_disksize_before_punch(inode, offset, length);
4250 		if (ret)
4251 			goto out_dio;
4252 		truncate_pagecache_range(inode, first_block_offset,
4253 					 last_block_offset);
4254 	}
4255 
4256 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4257 		credits = ext4_writepage_trans_blocks(inode);
4258 	else
4259 		credits = ext4_blocks_for_truncate(inode);
4260 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4261 	if (IS_ERR(handle)) {
4262 		ret = PTR_ERR(handle);
4263 		ext4_std_error(sb, ret);
4264 		goto out_dio;
4265 	}
4266 
4267 	ret = ext4_zero_partial_blocks(handle, inode, offset,
4268 				       length);
4269 	if (ret)
4270 		goto out_stop;
4271 
4272 	first_block = (offset + sb->s_blocksize - 1) >>
4273 		EXT4_BLOCK_SIZE_BITS(sb);
4274 	stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4275 
4276 	/* If there are no blocks to remove, return now */
4277 	if (first_block >= stop_block)
4278 		goto out_stop;
4279 
4280 	down_write(&EXT4_I(inode)->i_data_sem);
4281 	ext4_discard_preallocations(inode);
4282 
4283 	ret = ext4_es_remove_extent(inode, first_block,
4284 				    stop_block - first_block);
4285 	if (ret) {
4286 		up_write(&EXT4_I(inode)->i_data_sem);
4287 		goto out_stop;
4288 	}
4289 
4290 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4291 		ret = ext4_ext_remove_space(inode, first_block,
4292 					    stop_block - 1);
4293 	else
4294 		ret = ext4_ind_remove_space(handle, inode, first_block,
4295 					    stop_block);
4296 
4297 	up_write(&EXT4_I(inode)->i_data_sem);
4298 	if (IS_SYNC(inode))
4299 		ext4_handle_sync(handle);
4300 
4301 	inode->i_mtime = inode->i_ctime = current_time(inode);
4302 	ext4_mark_inode_dirty(handle, inode);
4303 	if (ret >= 0)
4304 		ext4_update_inode_fsync_trans(handle, inode, 1);
4305 out_stop:
4306 	ext4_journal_stop(handle);
4307 out_dio:
4308 	up_write(&EXT4_I(inode)->i_mmap_sem);
4309 	ext4_inode_resume_unlocked_dio(inode);
4310 out_mutex:
4311 	inode_unlock(inode);
4312 	return ret;
4313 }
4314 
4315 int ext4_inode_attach_jinode(struct inode *inode)
4316 {
4317 	struct ext4_inode_info *ei = EXT4_I(inode);
4318 	struct jbd2_inode *jinode;
4319 
4320 	if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4321 		return 0;
4322 
4323 	jinode = jbd2_alloc_inode(GFP_KERNEL);
4324 	spin_lock(&inode->i_lock);
4325 	if (!ei->jinode) {
4326 		if (!jinode) {
4327 			spin_unlock(&inode->i_lock);
4328 			return -ENOMEM;
4329 		}
4330 		ei->jinode = jinode;
4331 		jbd2_journal_init_jbd_inode(ei->jinode, inode);
4332 		jinode = NULL;
4333 	}
4334 	spin_unlock(&inode->i_lock);
4335 	if (unlikely(jinode != NULL))
4336 		jbd2_free_inode(jinode);
4337 	return 0;
4338 }
4339 
4340 /*
4341  * ext4_truncate()
4342  *
4343  * We block out ext4_get_block() block instantiations across the entire
4344  * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4345  * simultaneously on behalf of the same inode.
4346  *
4347  * As we work through the truncate and commit bits of it to the journal there
4348  * is one core, guiding principle: the file's tree must always be consistent on
4349  * disk.  We must be able to restart the truncate after a crash.
4350  *
4351  * The file's tree may be transiently inconsistent in memory (although it
4352  * probably isn't), but whenever we close off and commit a journal transaction,
4353  * the contents of (the filesystem + the journal) must be consistent and
4354  * restartable.  It's pretty simple, really: bottom up, right to left (although
4355  * left-to-right works OK too).
4356  *
4357  * Note that at recovery time, journal replay occurs *before* the restart of
4358  * truncate against the orphan inode list.
4359  *
4360  * The committed inode has the new, desired i_size (which is the same as
4361  * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4362  * that this inode's truncate did not complete and it will again call
4363  * ext4_truncate() to have another go.  So there will be instantiated blocks
4364  * to the right of the truncation point in a crashed ext4 filesystem.  But
4365  * that's fine - as long as they are linked from the inode, the post-crash
4366  * ext4_truncate() run will find them and release them.
4367  */
4368 int ext4_truncate(struct inode *inode)
4369 {
4370 	struct ext4_inode_info *ei = EXT4_I(inode);
4371 	unsigned int credits;
4372 	int err = 0;
4373 	handle_t *handle;
4374 	struct address_space *mapping = inode->i_mapping;
4375 
4376 	/*
4377 	 * There is a possibility that we're either freeing the inode
4378 	 * or it's a completely new inode. In those cases we might not
4379 	 * have i_mutex locked because it's not necessary.
4380 	 */
4381 	if (!(inode->i_state & (I_NEW|I_FREEING)))
4382 		WARN_ON(!inode_is_locked(inode));
4383 	trace_ext4_truncate_enter(inode);
4384 
4385 	if (!ext4_can_truncate(inode))
4386 		return 0;
4387 
4388 	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4389 
4390 	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4391 		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4392 
4393 	if (ext4_has_inline_data(inode)) {
4394 		int has_inline = 1;
4395 
4396 		err = ext4_inline_data_truncate(inode, &has_inline);
4397 		if (err)
4398 			return err;
4399 		if (has_inline)
4400 			return 0;
4401 	}
4402 
4403 	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4404 	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4405 		if (ext4_inode_attach_jinode(inode) < 0)
4406 			return 0;
4407 	}
4408 
4409 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4410 		credits = ext4_writepage_trans_blocks(inode);
4411 	else
4412 		credits = ext4_blocks_for_truncate(inode);
4413 
4414 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4415 	if (IS_ERR(handle))
4416 		return PTR_ERR(handle);
4417 
4418 	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4419 		ext4_block_truncate_page(handle, mapping, inode->i_size);
4420 
4421 	/*
4422 	 * We add the inode to the orphan list, so that if this
4423 	 * truncate spans multiple transactions, and we crash, we will
4424 	 * resume the truncate when the filesystem recovers.  It also
4425 	 * marks the inode dirty, to catch the new size.
4426 	 *
4427 	 * Implication: the file must always be in a sane, consistent
4428 	 * truncatable state while each transaction commits.
4429 	 */
4430 	err = ext4_orphan_add(handle, inode);
4431 	if (err)
4432 		goto out_stop;
4433 
4434 	down_write(&EXT4_I(inode)->i_data_sem);
4435 
4436 	ext4_discard_preallocations(inode);
4437 
4438 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4439 		err = ext4_ext_truncate(handle, inode);
4440 	else
4441 		ext4_ind_truncate(handle, inode);
4442 
4443 	up_write(&ei->i_data_sem);
4444 	if (err)
4445 		goto out_stop;
4446 
4447 	if (IS_SYNC(inode))
4448 		ext4_handle_sync(handle);
4449 
4450 out_stop:
4451 	/*
4452 	 * If this was a simple ftruncate() and the file will remain alive,
4453 	 * then we need to clear up the orphan record which we created above.
4454 	 * However, if this was a real unlink then we were called by
4455 	 * ext4_evict_inode(), and we allow that function to clean up the
4456 	 * orphan info for us.
4457 	 */
4458 	if (inode->i_nlink)
4459 		ext4_orphan_del(handle, inode);
4460 
4461 	inode->i_mtime = inode->i_ctime = current_time(inode);
4462 	ext4_mark_inode_dirty(handle, inode);
4463 	ext4_journal_stop(handle);
4464 
4465 	trace_ext4_truncate_exit(inode);
4466 	return err;
4467 }
4468 
4469 /*
4470  * ext4_get_inode_loc returns with an extra refcount against the inode's
4471  * underlying buffer_head on success. If 'in_mem' is true, we have all
4472  * data in memory that is needed to recreate the on-disk version of this
4473  * inode.
4474  */
4475 static int __ext4_get_inode_loc(struct inode *inode,
4476 				struct ext4_iloc *iloc, int in_mem)
4477 {
4478 	struct ext4_group_desc	*gdp;
4479 	struct buffer_head	*bh;
4480 	struct super_block	*sb = inode->i_sb;
4481 	ext4_fsblk_t		block;
4482 	int			inodes_per_block, inode_offset;
4483 
4484 	iloc->bh = NULL;
4485 	if (!ext4_valid_inum(sb, inode->i_ino))
4486 		return -EFSCORRUPTED;
4487 
4488 	iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4489 	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4490 	if (!gdp)
4491 		return -EIO;
4492 
4493 	/*
4494 	 * Figure out the offset within the block group inode table
4495 	 */
4496 	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4497 	inode_offset = ((inode->i_ino - 1) %
4498 			EXT4_INODES_PER_GROUP(sb));
4499 	block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4500 	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4501 
4502 	bh = sb_getblk(sb, block);
4503 	if (unlikely(!bh))
4504 		return -ENOMEM;
4505 	if (!buffer_uptodate(bh)) {
4506 		lock_buffer(bh);
4507 
4508 		/*
4509 		 * If the buffer has the write error flag, we have failed
4510 		 * to write out another inode in the same block.  In this
4511 		 * case, we don't have to read the block because we may
4512 		 * read the old inode data successfully.
4513 		 */
4514 		if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4515 			set_buffer_uptodate(bh);
4516 
4517 		if (buffer_uptodate(bh)) {
4518 			/* someone brought it uptodate while we waited */
4519 			unlock_buffer(bh);
4520 			goto has_buffer;
4521 		}
4522 
4523 		/*
4524 		 * If we have all information of the inode in memory and this
4525 		 * is the only valid inode in the block, we need not read the
4526 		 * block.
4527 		 */
4528 		if (in_mem) {
4529 			struct buffer_head *bitmap_bh;
4530 			int i, start;
4531 
4532 			start = inode_offset & ~(inodes_per_block - 1);
4533 
4534 			/* Is the inode bitmap in cache? */
4535 			bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4536 			if (unlikely(!bitmap_bh))
4537 				goto make_io;
4538 
4539 			/*
4540 			 * If the inode bitmap isn't in cache then the
4541 			 * optimisation may end up performing two reads instead
4542 			 * of one, so skip it.
4543 			 */
4544 			if (!buffer_uptodate(bitmap_bh)) {
4545 				brelse(bitmap_bh);
4546 				goto make_io;
4547 			}
4548 			for (i = start; i < start + inodes_per_block; i++) {
4549 				if (i == inode_offset)
4550 					continue;
4551 				if (ext4_test_bit(i, bitmap_bh->b_data))
4552 					break;
4553 			}
4554 			brelse(bitmap_bh);
4555 			if (i == start + inodes_per_block) {
4556 				/* all other inodes are free, so skip I/O */
4557 				memset(bh->b_data, 0, bh->b_size);
4558 				set_buffer_uptodate(bh);
4559 				unlock_buffer(bh);
4560 				goto has_buffer;
4561 			}
4562 		}
4563 
4564 make_io:
4565 		/*
4566 		 * If we need to do any I/O, try to pre-readahead extra
4567 		 * blocks from the inode table.
4568 		 */
4569 		if (EXT4_SB(sb)->s_inode_readahead_blks) {
4570 			ext4_fsblk_t b, end, table;
4571 			unsigned num;
4572 			__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4573 
4574 			table = ext4_inode_table(sb, gdp);
4575 			/* s_inode_readahead_blks is always a power of 2 */
4576 			b = block & ~((ext4_fsblk_t) ra_blks - 1);
4577 			if (table > b)
4578 				b = table;
4579 			end = b + ra_blks;
4580 			num = EXT4_INODES_PER_GROUP(sb);
4581 			if (ext4_has_group_desc_csum(sb))
4582 				num -= ext4_itable_unused_count(sb, gdp);
4583 			table += num / inodes_per_block;
4584 			if (end > table)
4585 				end = table;
4586 			while (b <= end)
4587 				sb_breadahead(sb, b++);
4588 		}
4589 
4590 		/*
4591 		 * There are other valid inodes in the buffer, this inode
4592 		 * has in-inode xattrs, or we don't have this inode in memory.
4593 		 * Read the block from disk.
4594 		 */
4595 		trace_ext4_load_inode(inode);
4596 		get_bh(bh);
4597 		bh->b_end_io = end_buffer_read_sync;
4598 		submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4599 		wait_on_buffer(bh);
4600 		if (!buffer_uptodate(bh)) {
4601 			EXT4_ERROR_INODE_BLOCK(inode, block,
4602 					       "unable to read itable block");
4603 			brelse(bh);
4604 			return -EIO;
4605 		}
4606 	}
4607 has_buffer:
4608 	iloc->bh = bh;
4609 	return 0;
4610 }
4611 
4612 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4613 {
4614 	/* We have all inode data except xattrs in memory here. */
4615 	return __ext4_get_inode_loc(inode, iloc,
4616 		!ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4617 }
4618 
4619 static bool ext4_should_use_dax(struct inode *inode)
4620 {
4621 	if (!test_opt(inode->i_sb, DAX))
4622 		return false;
4623 	if (!S_ISREG(inode->i_mode))
4624 		return false;
4625 	if (ext4_should_journal_data(inode))
4626 		return false;
4627 	if (ext4_has_inline_data(inode))
4628 		return false;
4629 	if (ext4_encrypted_inode(inode))
4630 		return false;
4631 	return true;
4632 }
4633 
4634 void ext4_set_inode_flags(struct inode *inode)
4635 {
4636 	unsigned int flags = EXT4_I(inode)->i_flags;
4637 	unsigned int new_fl = 0;
4638 
4639 	if (flags & EXT4_SYNC_FL)
4640 		new_fl |= S_SYNC;
4641 	if (flags & EXT4_APPEND_FL)
4642 		new_fl |= S_APPEND;
4643 	if (flags & EXT4_IMMUTABLE_FL)
4644 		new_fl |= S_IMMUTABLE;
4645 	if (flags & EXT4_NOATIME_FL)
4646 		new_fl |= S_NOATIME;
4647 	if (flags & EXT4_DIRSYNC_FL)
4648 		new_fl |= S_DIRSYNC;
4649 	if (ext4_should_use_dax(inode))
4650 		new_fl |= S_DAX;
4651 	if (flags & EXT4_ENCRYPT_FL)
4652 		new_fl |= S_ENCRYPTED;
4653 	inode_set_flags(inode, new_fl,
4654 			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4655 			S_ENCRYPTED);
4656 }
4657 
4658 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4659 				  struct ext4_inode_info *ei)
4660 {
4661 	blkcnt_t i_blocks ;
4662 	struct inode *inode = &(ei->vfs_inode);
4663 	struct super_block *sb = inode->i_sb;
4664 
4665 	if (ext4_has_feature_huge_file(sb)) {
4666 		/* we are using combined 48 bit field */
4667 		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4668 					le32_to_cpu(raw_inode->i_blocks_lo);
4669 		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4670 			/* i_blocks represent file system block size */
4671 			return i_blocks  << (inode->i_blkbits - 9);
4672 		} else {
4673 			return i_blocks;
4674 		}
4675 	} else {
4676 		return le32_to_cpu(raw_inode->i_blocks_lo);
4677 	}
4678 }
4679 
4680 static inline void ext4_iget_extra_inode(struct inode *inode,
4681 					 struct ext4_inode *raw_inode,
4682 					 struct ext4_inode_info *ei)
4683 {
4684 	__le32 *magic = (void *)raw_inode +
4685 			EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4686 	if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4687 	    EXT4_INODE_SIZE(inode->i_sb) &&
4688 	    *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4689 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4690 		ext4_find_inline_data_nolock(inode);
4691 	} else
4692 		EXT4_I(inode)->i_inline_off = 0;
4693 }
4694 
4695 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4696 {
4697 	if (!ext4_has_feature_project(inode->i_sb))
4698 		return -EOPNOTSUPP;
4699 	*projid = EXT4_I(inode)->i_projid;
4700 	return 0;
4701 }
4702 
4703 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4704 {
4705 	struct ext4_iloc iloc;
4706 	struct ext4_inode *raw_inode;
4707 	struct ext4_inode_info *ei;
4708 	struct inode *inode;
4709 	journal_t *journal = EXT4_SB(sb)->s_journal;
4710 	long ret;
4711 	loff_t size;
4712 	int block;
4713 	uid_t i_uid;
4714 	gid_t i_gid;
4715 	projid_t i_projid;
4716 
4717 	inode = iget_locked(sb, ino);
4718 	if (!inode)
4719 		return ERR_PTR(-ENOMEM);
4720 	if (!(inode->i_state & I_NEW))
4721 		return inode;
4722 
4723 	ei = EXT4_I(inode);
4724 	iloc.bh = NULL;
4725 
4726 	ret = __ext4_get_inode_loc(inode, &iloc, 0);
4727 	if (ret < 0)
4728 		goto bad_inode;
4729 	raw_inode = ext4_raw_inode(&iloc);
4730 
4731 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4732 		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4733 		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4734 			EXT4_INODE_SIZE(inode->i_sb) ||
4735 		    (ei->i_extra_isize & 3)) {
4736 			EXT4_ERROR_INODE(inode,
4737 					 "bad extra_isize %u (inode size %u)",
4738 					 ei->i_extra_isize,
4739 					 EXT4_INODE_SIZE(inode->i_sb));
4740 			ret = -EFSCORRUPTED;
4741 			goto bad_inode;
4742 		}
4743 	} else
4744 		ei->i_extra_isize = 0;
4745 
4746 	/* Precompute checksum seed for inode metadata */
4747 	if (ext4_has_metadata_csum(sb)) {
4748 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4749 		__u32 csum;
4750 		__le32 inum = cpu_to_le32(inode->i_ino);
4751 		__le32 gen = raw_inode->i_generation;
4752 		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4753 				   sizeof(inum));
4754 		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4755 					      sizeof(gen));
4756 	}
4757 
4758 	if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4759 		EXT4_ERROR_INODE(inode, "checksum invalid");
4760 		ret = -EFSBADCRC;
4761 		goto bad_inode;
4762 	}
4763 
4764 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4765 	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4766 	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4767 	if (ext4_has_feature_project(sb) &&
4768 	    EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4769 	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4770 		i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4771 	else
4772 		i_projid = EXT4_DEF_PROJID;
4773 
4774 	if (!(test_opt(inode->i_sb, NO_UID32))) {
4775 		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4776 		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4777 	}
4778 	i_uid_write(inode, i_uid);
4779 	i_gid_write(inode, i_gid);
4780 	ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4781 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4782 
4783 	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
4784 	ei->i_inline_off = 0;
4785 	ei->i_dir_start_lookup = 0;
4786 	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4787 	/* We now have enough fields to check if the inode was active or not.
4788 	 * This is needed because nfsd might try to access dead inodes
4789 	 * the test is that same one that e2fsck uses
4790 	 * NeilBrown 1999oct15
4791 	 */
4792 	if (inode->i_nlink == 0) {
4793 		if ((inode->i_mode == 0 ||
4794 		     !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4795 		    ino != EXT4_BOOT_LOADER_INO) {
4796 			/* this inode is deleted */
4797 			ret = -ESTALE;
4798 			goto bad_inode;
4799 		}
4800 		/* The only unlinked inodes we let through here have
4801 		 * valid i_mode and are being read by the orphan
4802 		 * recovery code: that's fine, we're about to complete
4803 		 * the process of deleting those.
4804 		 * OR it is the EXT4_BOOT_LOADER_INO which is
4805 		 * not initialized on a new filesystem. */
4806 	}
4807 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4808 	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4809 	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4810 	if (ext4_has_feature_64bit(sb))
4811 		ei->i_file_acl |=
4812 			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4813 	inode->i_size = ext4_isize(sb, raw_inode);
4814 	if ((size = i_size_read(inode)) < 0) {
4815 		EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4816 		ret = -EFSCORRUPTED;
4817 		goto bad_inode;
4818 	}
4819 	ei->i_disksize = inode->i_size;
4820 #ifdef CONFIG_QUOTA
4821 	ei->i_reserved_quota = 0;
4822 #endif
4823 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4824 	ei->i_block_group = iloc.block_group;
4825 	ei->i_last_alloc_group = ~0;
4826 	/*
4827 	 * NOTE! The in-memory inode i_data array is in little-endian order
4828 	 * even on big-endian machines: we do NOT byteswap the block numbers!
4829 	 */
4830 	for (block = 0; block < EXT4_N_BLOCKS; block++)
4831 		ei->i_data[block] = raw_inode->i_block[block];
4832 	INIT_LIST_HEAD(&ei->i_orphan);
4833 
4834 	/*
4835 	 * Set transaction id's of transactions that have to be committed
4836 	 * to finish f[data]sync. We set them to currently running transaction
4837 	 * as we cannot be sure that the inode or some of its metadata isn't
4838 	 * part of the transaction - the inode could have been reclaimed and
4839 	 * now it is reread from disk.
4840 	 */
4841 	if (journal) {
4842 		transaction_t *transaction;
4843 		tid_t tid;
4844 
4845 		read_lock(&journal->j_state_lock);
4846 		if (journal->j_running_transaction)
4847 			transaction = journal->j_running_transaction;
4848 		else
4849 			transaction = journal->j_committing_transaction;
4850 		if (transaction)
4851 			tid = transaction->t_tid;
4852 		else
4853 			tid = journal->j_commit_sequence;
4854 		read_unlock(&journal->j_state_lock);
4855 		ei->i_sync_tid = tid;
4856 		ei->i_datasync_tid = tid;
4857 	}
4858 
4859 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4860 		if (ei->i_extra_isize == 0) {
4861 			/* The extra space is currently unused. Use it. */
4862 			BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4863 			ei->i_extra_isize = sizeof(struct ext4_inode) -
4864 					    EXT4_GOOD_OLD_INODE_SIZE;
4865 		} else {
4866 			ext4_iget_extra_inode(inode, raw_inode, ei);
4867 		}
4868 	}
4869 
4870 	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4871 	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4872 	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4873 	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4874 
4875 	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4876 		inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4877 		if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4878 			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4879 				inode->i_version |=
4880 		    (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4881 		}
4882 	}
4883 
4884 	ret = 0;
4885 	if (ei->i_file_acl &&
4886 	    !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4887 		EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4888 				 ei->i_file_acl);
4889 		ret = -EFSCORRUPTED;
4890 		goto bad_inode;
4891 	} else if (!ext4_has_inline_data(inode)) {
4892 		if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4893 			if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4894 			    (S_ISLNK(inode->i_mode) &&
4895 			     !ext4_inode_is_fast_symlink(inode))))
4896 				/* Validate extent which is part of inode */
4897 				ret = ext4_ext_check_inode(inode);
4898 		} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4899 			   (S_ISLNK(inode->i_mode) &&
4900 			    !ext4_inode_is_fast_symlink(inode))) {
4901 			/* Validate block references which are part of inode */
4902 			ret = ext4_ind_check_inode(inode);
4903 		}
4904 	}
4905 	if (ret)
4906 		goto bad_inode;
4907 
4908 	if (S_ISREG(inode->i_mode)) {
4909 		inode->i_op = &ext4_file_inode_operations;
4910 		inode->i_fop = &ext4_file_operations;
4911 		ext4_set_aops(inode);
4912 	} else if (S_ISDIR(inode->i_mode)) {
4913 		inode->i_op = &ext4_dir_inode_operations;
4914 		inode->i_fop = &ext4_dir_operations;
4915 	} else if (S_ISLNK(inode->i_mode)) {
4916 		if (ext4_encrypted_inode(inode)) {
4917 			inode->i_op = &ext4_encrypted_symlink_inode_operations;
4918 			ext4_set_aops(inode);
4919 		} else if (ext4_inode_is_fast_symlink(inode)) {
4920 			inode->i_link = (char *)ei->i_data;
4921 			inode->i_op = &ext4_fast_symlink_inode_operations;
4922 			nd_terminate_link(ei->i_data, inode->i_size,
4923 				sizeof(ei->i_data) - 1);
4924 		} else {
4925 			inode->i_op = &ext4_symlink_inode_operations;
4926 			ext4_set_aops(inode);
4927 		}
4928 		inode_nohighmem(inode);
4929 	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4930 	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4931 		inode->i_op = &ext4_special_inode_operations;
4932 		if (raw_inode->i_block[0])
4933 			init_special_inode(inode, inode->i_mode,
4934 			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4935 		else
4936 			init_special_inode(inode, inode->i_mode,
4937 			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4938 	} else if (ino == EXT4_BOOT_LOADER_INO) {
4939 		make_bad_inode(inode);
4940 	} else {
4941 		ret = -EFSCORRUPTED;
4942 		EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4943 		goto bad_inode;
4944 	}
4945 	brelse(iloc.bh);
4946 	ext4_set_inode_flags(inode);
4947 
4948 	unlock_new_inode(inode);
4949 	return inode;
4950 
4951 bad_inode:
4952 	brelse(iloc.bh);
4953 	iget_failed(inode);
4954 	return ERR_PTR(ret);
4955 }
4956 
4957 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4958 {
4959 	if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4960 		return ERR_PTR(-EFSCORRUPTED);
4961 	return ext4_iget(sb, ino);
4962 }
4963 
4964 static int ext4_inode_blocks_set(handle_t *handle,
4965 				struct ext4_inode *raw_inode,
4966 				struct ext4_inode_info *ei)
4967 {
4968 	struct inode *inode = &(ei->vfs_inode);
4969 	u64 i_blocks = inode->i_blocks;
4970 	struct super_block *sb = inode->i_sb;
4971 
4972 	if (i_blocks <= ~0U) {
4973 		/*
4974 		 * i_blocks can be represented in a 32 bit variable
4975 		 * as multiple of 512 bytes
4976 		 */
4977 		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4978 		raw_inode->i_blocks_high = 0;
4979 		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4980 		return 0;
4981 	}
4982 	if (!ext4_has_feature_huge_file(sb))
4983 		return -EFBIG;
4984 
4985 	if (i_blocks <= 0xffffffffffffULL) {
4986 		/*
4987 		 * i_blocks can be represented in a 48 bit variable
4988 		 * as multiple of 512 bytes
4989 		 */
4990 		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4991 		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4992 		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4993 	} else {
4994 		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4995 		/* i_block is stored in file system block size */
4996 		i_blocks = i_blocks >> (inode->i_blkbits - 9);
4997 		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4998 		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4999 	}
5000 	return 0;
5001 }
5002 
5003 struct other_inode {
5004 	unsigned long		orig_ino;
5005 	struct ext4_inode	*raw_inode;
5006 };
5007 
5008 static int other_inode_match(struct inode * inode, unsigned long ino,
5009 			     void *data)
5010 {
5011 	struct other_inode *oi = (struct other_inode *) data;
5012 
5013 	if ((inode->i_ino != ino) ||
5014 	    (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5015 			       I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
5016 	    ((inode->i_state & I_DIRTY_TIME) == 0))
5017 		return 0;
5018 	spin_lock(&inode->i_lock);
5019 	if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5020 				I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
5021 	    (inode->i_state & I_DIRTY_TIME)) {
5022 		struct ext4_inode_info	*ei = EXT4_I(inode);
5023 
5024 		inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5025 		spin_unlock(&inode->i_lock);
5026 
5027 		spin_lock(&ei->i_raw_lock);
5028 		EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5029 		EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5030 		EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5031 		ext4_inode_csum_set(inode, oi->raw_inode, ei);
5032 		spin_unlock(&ei->i_raw_lock);
5033 		trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5034 		return -1;
5035 	}
5036 	spin_unlock(&inode->i_lock);
5037 	return -1;
5038 }
5039 
5040 /*
5041  * Opportunistically update the other time fields for other inodes in
5042  * the same inode table block.
5043  */
5044 static void ext4_update_other_inodes_time(struct super_block *sb,
5045 					  unsigned long orig_ino, char *buf)
5046 {
5047 	struct other_inode oi;
5048 	unsigned long ino;
5049 	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5050 	int inode_size = EXT4_INODE_SIZE(sb);
5051 
5052 	oi.orig_ino = orig_ino;
5053 	/*
5054 	 * Calculate the first inode in the inode table block.  Inode
5055 	 * numbers are one-based.  That is, the first inode in a block
5056 	 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5057 	 */
5058 	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5059 	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5060 		if (ino == orig_ino)
5061 			continue;
5062 		oi.raw_inode = (struct ext4_inode *) buf;
5063 		(void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5064 	}
5065 }
5066 
5067 /*
5068  * Post the struct inode info into an on-disk inode location in the
5069  * buffer-cache.  This gobbles the caller's reference to the
5070  * buffer_head in the inode location struct.
5071  *
5072  * The caller must have write access to iloc->bh.
5073  */
5074 static int ext4_do_update_inode(handle_t *handle,
5075 				struct inode *inode,
5076 				struct ext4_iloc *iloc)
5077 {
5078 	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5079 	struct ext4_inode_info *ei = EXT4_I(inode);
5080 	struct buffer_head *bh = iloc->bh;
5081 	struct super_block *sb = inode->i_sb;
5082 	int err = 0, rc, block;
5083 	int need_datasync = 0, set_large_file = 0;
5084 	uid_t i_uid;
5085 	gid_t i_gid;
5086 	projid_t i_projid;
5087 
5088 	spin_lock(&ei->i_raw_lock);
5089 
5090 	/* For fields not tracked in the in-memory inode,
5091 	 * initialise them to zero for new inodes. */
5092 	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5093 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5094 
5095 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5096 	i_uid = i_uid_read(inode);
5097 	i_gid = i_gid_read(inode);
5098 	i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5099 	if (!(test_opt(inode->i_sb, NO_UID32))) {
5100 		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5101 		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5102 /*
5103  * Fix up interoperability with old kernels. Otherwise, old inodes get
5104  * re-used with the upper 16 bits of the uid/gid intact
5105  */
5106 		if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5107 			raw_inode->i_uid_high = 0;
5108 			raw_inode->i_gid_high = 0;
5109 		} else {
5110 			raw_inode->i_uid_high =
5111 				cpu_to_le16(high_16_bits(i_uid));
5112 			raw_inode->i_gid_high =
5113 				cpu_to_le16(high_16_bits(i_gid));
5114 		}
5115 	} else {
5116 		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5117 		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5118 		raw_inode->i_uid_high = 0;
5119 		raw_inode->i_gid_high = 0;
5120 	}
5121 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5122 
5123 	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5124 	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5125 	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5126 	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5127 
5128 	err = ext4_inode_blocks_set(handle, raw_inode, ei);
5129 	if (err) {
5130 		spin_unlock(&ei->i_raw_lock);
5131 		goto out_brelse;
5132 	}
5133 	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5134 	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5135 	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5136 		raw_inode->i_file_acl_high =
5137 			cpu_to_le16(ei->i_file_acl >> 32);
5138 	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5139 	if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5140 		ext4_isize_set(raw_inode, ei->i_disksize);
5141 		need_datasync = 1;
5142 	}
5143 	if (ei->i_disksize > 0x7fffffffULL) {
5144 		if (!ext4_has_feature_large_file(sb) ||
5145 				EXT4_SB(sb)->s_es->s_rev_level ==
5146 		    cpu_to_le32(EXT4_GOOD_OLD_REV))
5147 			set_large_file = 1;
5148 	}
5149 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5150 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5151 		if (old_valid_dev(inode->i_rdev)) {
5152 			raw_inode->i_block[0] =
5153 				cpu_to_le32(old_encode_dev(inode->i_rdev));
5154 			raw_inode->i_block[1] = 0;
5155 		} else {
5156 			raw_inode->i_block[0] = 0;
5157 			raw_inode->i_block[1] =
5158 				cpu_to_le32(new_encode_dev(inode->i_rdev));
5159 			raw_inode->i_block[2] = 0;
5160 		}
5161 	} else if (!ext4_has_inline_data(inode)) {
5162 		for (block = 0; block < EXT4_N_BLOCKS; block++)
5163 			raw_inode->i_block[block] = ei->i_data[block];
5164 	}
5165 
5166 	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5167 		raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5168 		if (ei->i_extra_isize) {
5169 			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5170 				raw_inode->i_version_hi =
5171 					cpu_to_le32(inode->i_version >> 32);
5172 			raw_inode->i_extra_isize =
5173 				cpu_to_le16(ei->i_extra_isize);
5174 		}
5175 	}
5176 
5177 	BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5178 	       i_projid != EXT4_DEF_PROJID);
5179 
5180 	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5181 	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5182 		raw_inode->i_projid = cpu_to_le32(i_projid);
5183 
5184 	ext4_inode_csum_set(inode, raw_inode, ei);
5185 	spin_unlock(&ei->i_raw_lock);
5186 	if (inode->i_sb->s_flags & MS_LAZYTIME)
5187 		ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5188 					      bh->b_data);
5189 
5190 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5191 	rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5192 	if (!err)
5193 		err = rc;
5194 	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5195 	if (set_large_file) {
5196 		BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5197 		err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5198 		if (err)
5199 			goto out_brelse;
5200 		ext4_update_dynamic_rev(sb);
5201 		ext4_set_feature_large_file(sb);
5202 		ext4_handle_sync(handle);
5203 		err = ext4_handle_dirty_super(handle, sb);
5204 	}
5205 	ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5206 out_brelse:
5207 	brelse(bh);
5208 	ext4_std_error(inode->i_sb, err);
5209 	return err;
5210 }
5211 
5212 /*
5213  * ext4_write_inode()
5214  *
5215  * We are called from a few places:
5216  *
5217  * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5218  *   Here, there will be no transaction running. We wait for any running
5219  *   transaction to commit.
5220  *
5221  * - Within flush work (sys_sync(), kupdate and such).
5222  *   We wait on commit, if told to.
5223  *
5224  * - Within iput_final() -> write_inode_now()
5225  *   We wait on commit, if told to.
5226  *
5227  * In all cases it is actually safe for us to return without doing anything,
5228  * because the inode has been copied into a raw inode buffer in
5229  * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5230  * writeback.
5231  *
5232  * Note that we are absolutely dependent upon all inode dirtiers doing the
5233  * right thing: they *must* call mark_inode_dirty() after dirtying info in
5234  * which we are interested.
5235  *
5236  * It would be a bug for them to not do this.  The code:
5237  *
5238  *	mark_inode_dirty(inode)
5239  *	stuff();
5240  *	inode->i_size = expr;
5241  *
5242  * is in error because write_inode() could occur while `stuff()' is running,
5243  * and the new i_size will be lost.  Plus the inode will no longer be on the
5244  * superblock's dirty inode list.
5245  */
5246 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5247 {
5248 	int err;
5249 
5250 	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5251 		return 0;
5252 
5253 	if (EXT4_SB(inode->i_sb)->s_journal) {
5254 		if (ext4_journal_current_handle()) {
5255 			jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5256 			dump_stack();
5257 			return -EIO;
5258 		}
5259 
5260 		/*
5261 		 * No need to force transaction in WB_SYNC_NONE mode. Also
5262 		 * ext4_sync_fs() will force the commit after everything is
5263 		 * written.
5264 		 */
5265 		if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5266 			return 0;
5267 
5268 		err = ext4_force_commit(inode->i_sb);
5269 	} else {
5270 		struct ext4_iloc iloc;
5271 
5272 		err = __ext4_get_inode_loc(inode, &iloc, 0);
5273 		if (err)
5274 			return err;
5275 		/*
5276 		 * sync(2) will flush the whole buffer cache. No need to do
5277 		 * it here separately for each inode.
5278 		 */
5279 		if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5280 			sync_dirty_buffer(iloc.bh);
5281 		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5282 			EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5283 					 "IO error syncing inode");
5284 			err = -EIO;
5285 		}
5286 		brelse(iloc.bh);
5287 	}
5288 	return err;
5289 }
5290 
5291 /*
5292  * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5293  * buffers that are attached to a page stradding i_size and are undergoing
5294  * commit. In that case we have to wait for commit to finish and try again.
5295  */
5296 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5297 {
5298 	struct page *page;
5299 	unsigned offset;
5300 	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5301 	tid_t commit_tid = 0;
5302 	int ret;
5303 
5304 	offset = inode->i_size & (PAGE_SIZE - 1);
5305 	/*
5306 	 * All buffers in the last page remain valid? Then there's nothing to
5307 	 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5308 	 * blocksize case
5309 	 */
5310 	if (offset > PAGE_SIZE - i_blocksize(inode))
5311 		return;
5312 	while (1) {
5313 		page = find_lock_page(inode->i_mapping,
5314 				      inode->i_size >> PAGE_SHIFT);
5315 		if (!page)
5316 			return;
5317 		ret = __ext4_journalled_invalidatepage(page, offset,
5318 						PAGE_SIZE - offset);
5319 		unlock_page(page);
5320 		put_page(page);
5321 		if (ret != -EBUSY)
5322 			return;
5323 		commit_tid = 0;
5324 		read_lock(&journal->j_state_lock);
5325 		if (journal->j_committing_transaction)
5326 			commit_tid = journal->j_committing_transaction->t_tid;
5327 		read_unlock(&journal->j_state_lock);
5328 		if (commit_tid)
5329 			jbd2_log_wait_commit(journal, commit_tid);
5330 	}
5331 }
5332 
5333 /*
5334  * ext4_setattr()
5335  *
5336  * Called from notify_change.
5337  *
5338  * We want to trap VFS attempts to truncate the file as soon as
5339  * possible.  In particular, we want to make sure that when the VFS
5340  * shrinks i_size, we put the inode on the orphan list and modify
5341  * i_disksize immediately, so that during the subsequent flushing of
5342  * dirty pages and freeing of disk blocks, we can guarantee that any
5343  * commit will leave the blocks being flushed in an unused state on
5344  * disk.  (On recovery, the inode will get truncated and the blocks will
5345  * be freed, so we have a strong guarantee that no future commit will
5346  * leave these blocks visible to the user.)
5347  *
5348  * Another thing we have to assure is that if we are in ordered mode
5349  * and inode is still attached to the committing transaction, we must
5350  * we start writeout of all the dirty pages which are being truncated.
5351  * This way we are sure that all the data written in the previous
5352  * transaction are already on disk (truncate waits for pages under
5353  * writeback).
5354  *
5355  * Called with inode->i_mutex down.
5356  */
5357 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5358 {
5359 	struct inode *inode = d_inode(dentry);
5360 	int error, rc = 0;
5361 	int orphan = 0;
5362 	const unsigned int ia_valid = attr->ia_valid;
5363 
5364 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5365 		return -EIO;
5366 
5367 	error = setattr_prepare(dentry, attr);
5368 	if (error)
5369 		return error;
5370 
5371 	error = fscrypt_prepare_setattr(dentry, attr);
5372 	if (error)
5373 		return error;
5374 
5375 	if (is_quota_modification(inode, attr)) {
5376 		error = dquot_initialize(inode);
5377 		if (error)
5378 			return error;
5379 	}
5380 	if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5381 	    (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5382 		handle_t *handle;
5383 
5384 		/* (user+group)*(old+new) structure, inode write (sb,
5385 		 * inode block, ? - but truncate inode update has it) */
5386 		handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5387 			(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5388 			 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5389 		if (IS_ERR(handle)) {
5390 			error = PTR_ERR(handle);
5391 			goto err_out;
5392 		}
5393 
5394 		/* dquot_transfer() calls back ext4_get_inode_usage() which
5395 		 * counts xattr inode references.
5396 		 */
5397 		down_read(&EXT4_I(inode)->xattr_sem);
5398 		error = dquot_transfer(inode, attr);
5399 		up_read(&EXT4_I(inode)->xattr_sem);
5400 
5401 		if (error) {
5402 			ext4_journal_stop(handle);
5403 			return error;
5404 		}
5405 		/* Update corresponding info in inode so that everything is in
5406 		 * one transaction */
5407 		if (attr->ia_valid & ATTR_UID)
5408 			inode->i_uid = attr->ia_uid;
5409 		if (attr->ia_valid & ATTR_GID)
5410 			inode->i_gid = attr->ia_gid;
5411 		error = ext4_mark_inode_dirty(handle, inode);
5412 		ext4_journal_stop(handle);
5413 	}
5414 
5415 	if (attr->ia_valid & ATTR_SIZE) {
5416 		handle_t *handle;
5417 		loff_t oldsize = inode->i_size;
5418 		int shrink = (attr->ia_size <= inode->i_size);
5419 
5420 		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5421 			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5422 
5423 			if (attr->ia_size > sbi->s_bitmap_maxbytes)
5424 				return -EFBIG;
5425 		}
5426 		if (!S_ISREG(inode->i_mode))
5427 			return -EINVAL;
5428 
5429 		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5430 			inode_inc_iversion(inode);
5431 
5432 		if (ext4_should_order_data(inode) &&
5433 		    (attr->ia_size < inode->i_size)) {
5434 			error = ext4_begin_ordered_truncate(inode,
5435 							    attr->ia_size);
5436 			if (error)
5437 				goto err_out;
5438 		}
5439 		if (attr->ia_size != inode->i_size) {
5440 			handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5441 			if (IS_ERR(handle)) {
5442 				error = PTR_ERR(handle);
5443 				goto err_out;
5444 			}
5445 			if (ext4_handle_valid(handle) && shrink) {
5446 				error = ext4_orphan_add(handle, inode);
5447 				orphan = 1;
5448 			}
5449 			/*
5450 			 * Update c/mtime on truncate up, ext4_truncate() will
5451 			 * update c/mtime in shrink case below
5452 			 */
5453 			if (!shrink) {
5454 				inode->i_mtime = current_time(inode);
5455 				inode->i_ctime = inode->i_mtime;
5456 			}
5457 			down_write(&EXT4_I(inode)->i_data_sem);
5458 			EXT4_I(inode)->i_disksize = attr->ia_size;
5459 			rc = ext4_mark_inode_dirty(handle, inode);
5460 			if (!error)
5461 				error = rc;
5462 			/*
5463 			 * We have to update i_size under i_data_sem together
5464 			 * with i_disksize to avoid races with writeback code
5465 			 * running ext4_wb_update_i_disksize().
5466 			 */
5467 			if (!error)
5468 				i_size_write(inode, attr->ia_size);
5469 			up_write(&EXT4_I(inode)->i_data_sem);
5470 			ext4_journal_stop(handle);
5471 			if (error) {
5472 				if (orphan)
5473 					ext4_orphan_del(NULL, inode);
5474 				goto err_out;
5475 			}
5476 		}
5477 		if (!shrink)
5478 			pagecache_isize_extended(inode, oldsize, inode->i_size);
5479 
5480 		/*
5481 		 * Blocks are going to be removed from the inode. Wait
5482 		 * for dio in flight.  Temporarily disable
5483 		 * dioread_nolock to prevent livelock.
5484 		 */
5485 		if (orphan) {
5486 			if (!ext4_should_journal_data(inode)) {
5487 				ext4_inode_block_unlocked_dio(inode);
5488 				inode_dio_wait(inode);
5489 				ext4_inode_resume_unlocked_dio(inode);
5490 			} else
5491 				ext4_wait_for_tail_page_commit(inode);
5492 		}
5493 		down_write(&EXT4_I(inode)->i_mmap_sem);
5494 		/*
5495 		 * Truncate pagecache after we've waited for commit
5496 		 * in data=journal mode to make pages freeable.
5497 		 */
5498 		truncate_pagecache(inode, inode->i_size);
5499 		if (shrink) {
5500 			rc = ext4_truncate(inode);
5501 			if (rc)
5502 				error = rc;
5503 		}
5504 		up_write(&EXT4_I(inode)->i_mmap_sem);
5505 	}
5506 
5507 	if (!error) {
5508 		setattr_copy(inode, attr);
5509 		mark_inode_dirty(inode);
5510 	}
5511 
5512 	/*
5513 	 * If the call to ext4_truncate failed to get a transaction handle at
5514 	 * all, we need to clean up the in-core orphan list manually.
5515 	 */
5516 	if (orphan && inode->i_nlink)
5517 		ext4_orphan_del(NULL, inode);
5518 
5519 	if (!error && (ia_valid & ATTR_MODE))
5520 		rc = posix_acl_chmod(inode, inode->i_mode);
5521 
5522 err_out:
5523 	ext4_std_error(inode->i_sb, error);
5524 	if (!error)
5525 		error = rc;
5526 	return error;
5527 }
5528 
5529 int ext4_getattr(const struct path *path, struct kstat *stat,
5530 		 u32 request_mask, unsigned int query_flags)
5531 {
5532 	struct inode *inode = d_inode(path->dentry);
5533 	struct ext4_inode *raw_inode;
5534 	struct ext4_inode_info *ei = EXT4_I(inode);
5535 	unsigned int flags;
5536 
5537 	if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5538 		stat->result_mask |= STATX_BTIME;
5539 		stat->btime.tv_sec = ei->i_crtime.tv_sec;
5540 		stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5541 	}
5542 
5543 	flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5544 	if (flags & EXT4_APPEND_FL)
5545 		stat->attributes |= STATX_ATTR_APPEND;
5546 	if (flags & EXT4_COMPR_FL)
5547 		stat->attributes |= STATX_ATTR_COMPRESSED;
5548 	if (flags & EXT4_ENCRYPT_FL)
5549 		stat->attributes |= STATX_ATTR_ENCRYPTED;
5550 	if (flags & EXT4_IMMUTABLE_FL)
5551 		stat->attributes |= STATX_ATTR_IMMUTABLE;
5552 	if (flags & EXT4_NODUMP_FL)
5553 		stat->attributes |= STATX_ATTR_NODUMP;
5554 
5555 	stat->attributes_mask |= (STATX_ATTR_APPEND |
5556 				  STATX_ATTR_COMPRESSED |
5557 				  STATX_ATTR_ENCRYPTED |
5558 				  STATX_ATTR_IMMUTABLE |
5559 				  STATX_ATTR_NODUMP);
5560 
5561 	generic_fillattr(inode, stat);
5562 	return 0;
5563 }
5564 
5565 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5566 		      u32 request_mask, unsigned int query_flags)
5567 {
5568 	struct inode *inode = d_inode(path->dentry);
5569 	u64 delalloc_blocks;
5570 
5571 	ext4_getattr(path, stat, request_mask, query_flags);
5572 
5573 	/*
5574 	 * If there is inline data in the inode, the inode will normally not
5575 	 * have data blocks allocated (it may have an external xattr block).
5576 	 * Report at least one sector for such files, so tools like tar, rsync,
5577 	 * others don't incorrectly think the file is completely sparse.
5578 	 */
5579 	if (unlikely(ext4_has_inline_data(inode)))
5580 		stat->blocks += (stat->size + 511) >> 9;
5581 
5582 	/*
5583 	 * We can't update i_blocks if the block allocation is delayed
5584 	 * otherwise in the case of system crash before the real block
5585 	 * allocation is done, we will have i_blocks inconsistent with
5586 	 * on-disk file blocks.
5587 	 * We always keep i_blocks updated together with real
5588 	 * allocation. But to not confuse with user, stat
5589 	 * will return the blocks that include the delayed allocation
5590 	 * blocks for this file.
5591 	 */
5592 	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5593 				   EXT4_I(inode)->i_reserved_data_blocks);
5594 	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5595 	return 0;
5596 }
5597 
5598 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5599 				   int pextents)
5600 {
5601 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5602 		return ext4_ind_trans_blocks(inode, lblocks);
5603 	return ext4_ext_index_trans_blocks(inode, pextents);
5604 }
5605 
5606 /*
5607  * Account for index blocks, block groups bitmaps and block group
5608  * descriptor blocks if modify datablocks and index blocks
5609  * worse case, the indexs blocks spread over different block groups
5610  *
5611  * If datablocks are discontiguous, they are possible to spread over
5612  * different block groups too. If they are contiguous, with flexbg,
5613  * they could still across block group boundary.
5614  *
5615  * Also account for superblock, inode, quota and xattr blocks
5616  */
5617 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5618 				  int pextents)
5619 {
5620 	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5621 	int gdpblocks;
5622 	int idxblocks;
5623 	int ret = 0;
5624 
5625 	/*
5626 	 * How many index blocks need to touch to map @lblocks logical blocks
5627 	 * to @pextents physical extents?
5628 	 */
5629 	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5630 
5631 	ret = idxblocks;
5632 
5633 	/*
5634 	 * Now let's see how many group bitmaps and group descriptors need
5635 	 * to account
5636 	 */
5637 	groups = idxblocks + pextents;
5638 	gdpblocks = groups;
5639 	if (groups > ngroups)
5640 		groups = ngroups;
5641 	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5642 		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5643 
5644 	/* bitmaps and block group descriptor blocks */
5645 	ret += groups + gdpblocks;
5646 
5647 	/* Blocks for super block, inode, quota and xattr blocks */
5648 	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5649 
5650 	return ret;
5651 }
5652 
5653 /*
5654  * Calculate the total number of credits to reserve to fit
5655  * the modification of a single pages into a single transaction,
5656  * which may include multiple chunks of block allocations.
5657  *
5658  * This could be called via ext4_write_begin()
5659  *
5660  * We need to consider the worse case, when
5661  * one new block per extent.
5662  */
5663 int ext4_writepage_trans_blocks(struct inode *inode)
5664 {
5665 	int bpp = ext4_journal_blocks_per_page(inode);
5666 	int ret;
5667 
5668 	ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5669 
5670 	/* Account for data blocks for journalled mode */
5671 	if (ext4_should_journal_data(inode))
5672 		ret += bpp;
5673 	return ret;
5674 }
5675 
5676 /*
5677  * Calculate the journal credits for a chunk of data modification.
5678  *
5679  * This is called from DIO, fallocate or whoever calling
5680  * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5681  *
5682  * journal buffers for data blocks are not included here, as DIO
5683  * and fallocate do no need to journal data buffers.
5684  */
5685 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5686 {
5687 	return ext4_meta_trans_blocks(inode, nrblocks, 1);
5688 }
5689 
5690 /*
5691  * The caller must have previously called ext4_reserve_inode_write().
5692  * Give this, we know that the caller already has write access to iloc->bh.
5693  */
5694 int ext4_mark_iloc_dirty(handle_t *handle,
5695 			 struct inode *inode, struct ext4_iloc *iloc)
5696 {
5697 	int err = 0;
5698 
5699 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5700 		return -EIO;
5701 
5702 	if (IS_I_VERSION(inode))
5703 		inode_inc_iversion(inode);
5704 
5705 	/* the do_update_inode consumes one bh->b_count */
5706 	get_bh(iloc->bh);
5707 
5708 	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5709 	err = ext4_do_update_inode(handle, inode, iloc);
5710 	put_bh(iloc->bh);
5711 	return err;
5712 }
5713 
5714 /*
5715  * On success, We end up with an outstanding reference count against
5716  * iloc->bh.  This _must_ be cleaned up later.
5717  */
5718 
5719 int
5720 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5721 			 struct ext4_iloc *iloc)
5722 {
5723 	int err;
5724 
5725 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5726 		return -EIO;
5727 
5728 	err = ext4_get_inode_loc(inode, iloc);
5729 	if (!err) {
5730 		BUFFER_TRACE(iloc->bh, "get_write_access");
5731 		err = ext4_journal_get_write_access(handle, iloc->bh);
5732 		if (err) {
5733 			brelse(iloc->bh);
5734 			iloc->bh = NULL;
5735 		}
5736 	}
5737 	ext4_std_error(inode->i_sb, err);
5738 	return err;
5739 }
5740 
5741 static int __ext4_expand_extra_isize(struct inode *inode,
5742 				     unsigned int new_extra_isize,
5743 				     struct ext4_iloc *iloc,
5744 				     handle_t *handle, int *no_expand)
5745 {
5746 	struct ext4_inode *raw_inode;
5747 	struct ext4_xattr_ibody_header *header;
5748 	int error;
5749 
5750 	raw_inode = ext4_raw_inode(iloc);
5751 
5752 	header = IHDR(inode, raw_inode);
5753 
5754 	/* No extended attributes present */
5755 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5756 	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5757 		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5758 		       EXT4_I(inode)->i_extra_isize, 0,
5759 		       new_extra_isize - EXT4_I(inode)->i_extra_isize);
5760 		EXT4_I(inode)->i_extra_isize = new_extra_isize;
5761 		return 0;
5762 	}
5763 
5764 	/* try to expand with EAs present */
5765 	error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5766 					   raw_inode, handle);
5767 	if (error) {
5768 		/*
5769 		 * Inode size expansion failed; don't try again
5770 		 */
5771 		*no_expand = 1;
5772 	}
5773 
5774 	return error;
5775 }
5776 
5777 /*
5778  * Expand an inode by new_extra_isize bytes.
5779  * Returns 0 on success or negative error number on failure.
5780  */
5781 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5782 					  unsigned int new_extra_isize,
5783 					  struct ext4_iloc iloc,
5784 					  handle_t *handle)
5785 {
5786 	int no_expand;
5787 	int error;
5788 
5789 	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5790 		return -EOVERFLOW;
5791 
5792 	/*
5793 	 * In nojournal mode, we can immediately attempt to expand
5794 	 * the inode.  When journaled, we first need to obtain extra
5795 	 * buffer credits since we may write into the EA block
5796 	 * with this same handle. If journal_extend fails, then it will
5797 	 * only result in a minor loss of functionality for that inode.
5798 	 * If this is felt to be critical, then e2fsck should be run to
5799 	 * force a large enough s_min_extra_isize.
5800 	 */
5801 	if (ext4_handle_valid(handle) &&
5802 	    jbd2_journal_extend(handle,
5803 				EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5804 		return -ENOSPC;
5805 
5806 	if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5807 		return -EBUSY;
5808 
5809 	error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5810 					  handle, &no_expand);
5811 	ext4_write_unlock_xattr(inode, &no_expand);
5812 
5813 	return error;
5814 }
5815 
5816 int ext4_expand_extra_isize(struct inode *inode,
5817 			    unsigned int new_extra_isize,
5818 			    struct ext4_iloc *iloc)
5819 {
5820 	handle_t *handle;
5821 	int no_expand;
5822 	int error, rc;
5823 
5824 	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5825 		brelse(iloc->bh);
5826 		return -EOVERFLOW;
5827 	}
5828 
5829 	handle = ext4_journal_start(inode, EXT4_HT_INODE,
5830 				    EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5831 	if (IS_ERR(handle)) {
5832 		error = PTR_ERR(handle);
5833 		brelse(iloc->bh);
5834 		return error;
5835 	}
5836 
5837 	ext4_write_lock_xattr(inode, &no_expand);
5838 
5839 	BUFFER_TRACE(iloc.bh, "get_write_access");
5840 	error = ext4_journal_get_write_access(handle, iloc->bh);
5841 	if (error) {
5842 		brelse(iloc->bh);
5843 		goto out_stop;
5844 	}
5845 
5846 	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5847 					  handle, &no_expand);
5848 
5849 	rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5850 	if (!error)
5851 		error = rc;
5852 
5853 	ext4_write_unlock_xattr(inode, &no_expand);
5854 out_stop:
5855 	ext4_journal_stop(handle);
5856 	return error;
5857 }
5858 
5859 /*
5860  * What we do here is to mark the in-core inode as clean with respect to inode
5861  * dirtiness (it may still be data-dirty).
5862  * This means that the in-core inode may be reaped by prune_icache
5863  * without having to perform any I/O.  This is a very good thing,
5864  * because *any* task may call prune_icache - even ones which
5865  * have a transaction open against a different journal.
5866  *
5867  * Is this cheating?  Not really.  Sure, we haven't written the
5868  * inode out, but prune_icache isn't a user-visible syncing function.
5869  * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5870  * we start and wait on commits.
5871  */
5872 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5873 {
5874 	struct ext4_iloc iloc;
5875 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5876 	int err;
5877 
5878 	might_sleep();
5879 	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5880 	err = ext4_reserve_inode_write(handle, inode, &iloc);
5881 	if (err)
5882 		return err;
5883 
5884 	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5885 		ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5886 					       iloc, handle);
5887 
5888 	return ext4_mark_iloc_dirty(handle, inode, &iloc);
5889 }
5890 
5891 /*
5892  * ext4_dirty_inode() is called from __mark_inode_dirty()
5893  *
5894  * We're really interested in the case where a file is being extended.
5895  * i_size has been changed by generic_commit_write() and we thus need
5896  * to include the updated inode in the current transaction.
5897  *
5898  * Also, dquot_alloc_block() will always dirty the inode when blocks
5899  * are allocated to the file.
5900  *
5901  * If the inode is marked synchronous, we don't honour that here - doing
5902  * so would cause a commit on atime updates, which we don't bother doing.
5903  * We handle synchronous inodes at the highest possible level.
5904  *
5905  * If only the I_DIRTY_TIME flag is set, we can skip everything.  If
5906  * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5907  * to copy into the on-disk inode structure are the timestamp files.
5908  */
5909 void ext4_dirty_inode(struct inode *inode, int flags)
5910 {
5911 	handle_t *handle;
5912 
5913 	if (flags == I_DIRTY_TIME)
5914 		return;
5915 	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5916 	if (IS_ERR(handle))
5917 		goto out;
5918 
5919 	ext4_mark_inode_dirty(handle, inode);
5920 
5921 	ext4_journal_stop(handle);
5922 out:
5923 	return;
5924 }
5925 
5926 #if 0
5927 /*
5928  * Bind an inode's backing buffer_head into this transaction, to prevent
5929  * it from being flushed to disk early.  Unlike
5930  * ext4_reserve_inode_write, this leaves behind no bh reference and
5931  * returns no iloc structure, so the caller needs to repeat the iloc
5932  * lookup to mark the inode dirty later.
5933  */
5934 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5935 {
5936 	struct ext4_iloc iloc;
5937 
5938 	int err = 0;
5939 	if (handle) {
5940 		err = ext4_get_inode_loc(inode, &iloc);
5941 		if (!err) {
5942 			BUFFER_TRACE(iloc.bh, "get_write_access");
5943 			err = jbd2_journal_get_write_access(handle, iloc.bh);
5944 			if (!err)
5945 				err = ext4_handle_dirty_metadata(handle,
5946 								 NULL,
5947 								 iloc.bh);
5948 			brelse(iloc.bh);
5949 		}
5950 	}
5951 	ext4_std_error(inode->i_sb, err);
5952 	return err;
5953 }
5954 #endif
5955 
5956 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5957 {
5958 	journal_t *journal;
5959 	handle_t *handle;
5960 	int err;
5961 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5962 
5963 	/*
5964 	 * We have to be very careful here: changing a data block's
5965 	 * journaling status dynamically is dangerous.  If we write a
5966 	 * data block to the journal, change the status and then delete
5967 	 * that block, we risk forgetting to revoke the old log record
5968 	 * from the journal and so a subsequent replay can corrupt data.
5969 	 * So, first we make sure that the journal is empty and that
5970 	 * nobody is changing anything.
5971 	 */
5972 
5973 	journal = EXT4_JOURNAL(inode);
5974 	if (!journal)
5975 		return 0;
5976 	if (is_journal_aborted(journal))
5977 		return -EROFS;
5978 
5979 	/* Wait for all existing dio workers */
5980 	ext4_inode_block_unlocked_dio(inode);
5981 	inode_dio_wait(inode);
5982 
5983 	/*
5984 	 * Before flushing the journal and switching inode's aops, we have
5985 	 * to flush all dirty data the inode has. There can be outstanding
5986 	 * delayed allocations, there can be unwritten extents created by
5987 	 * fallocate or buffered writes in dioread_nolock mode covered by
5988 	 * dirty data which can be converted only after flushing the dirty
5989 	 * data (and journalled aops don't know how to handle these cases).
5990 	 */
5991 	if (val) {
5992 		down_write(&EXT4_I(inode)->i_mmap_sem);
5993 		err = filemap_write_and_wait(inode->i_mapping);
5994 		if (err < 0) {
5995 			up_write(&EXT4_I(inode)->i_mmap_sem);
5996 			ext4_inode_resume_unlocked_dio(inode);
5997 			return err;
5998 		}
5999 	}
6000 
6001 	percpu_down_write(&sbi->s_journal_flag_rwsem);
6002 	jbd2_journal_lock_updates(journal);
6003 
6004 	/*
6005 	 * OK, there are no updates running now, and all cached data is
6006 	 * synced to disk.  We are now in a completely consistent state
6007 	 * which doesn't have anything in the journal, and we know that
6008 	 * no filesystem updates are running, so it is safe to modify
6009 	 * the inode's in-core data-journaling state flag now.
6010 	 */
6011 
6012 	if (val)
6013 		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6014 	else {
6015 		err = jbd2_journal_flush(journal);
6016 		if (err < 0) {
6017 			jbd2_journal_unlock_updates(journal);
6018 			percpu_up_write(&sbi->s_journal_flag_rwsem);
6019 			ext4_inode_resume_unlocked_dio(inode);
6020 			return err;
6021 		}
6022 		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6023 	}
6024 	ext4_set_aops(inode);
6025 
6026 	jbd2_journal_unlock_updates(journal);
6027 	percpu_up_write(&sbi->s_journal_flag_rwsem);
6028 
6029 	if (val)
6030 		up_write(&EXT4_I(inode)->i_mmap_sem);
6031 	ext4_inode_resume_unlocked_dio(inode);
6032 
6033 	/* Finally we can mark the inode as dirty. */
6034 
6035 	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6036 	if (IS_ERR(handle))
6037 		return PTR_ERR(handle);
6038 
6039 	err = ext4_mark_inode_dirty(handle, inode);
6040 	ext4_handle_sync(handle);
6041 	ext4_journal_stop(handle);
6042 	ext4_std_error(inode->i_sb, err);
6043 
6044 	return err;
6045 }
6046 
6047 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6048 {
6049 	return !buffer_mapped(bh);
6050 }
6051 
6052 int ext4_page_mkwrite(struct vm_fault *vmf)
6053 {
6054 	struct vm_area_struct *vma = vmf->vma;
6055 	struct page *page = vmf->page;
6056 	loff_t size;
6057 	unsigned long len;
6058 	int ret;
6059 	struct file *file = vma->vm_file;
6060 	struct inode *inode = file_inode(file);
6061 	struct address_space *mapping = inode->i_mapping;
6062 	handle_t *handle;
6063 	get_block_t *get_block;
6064 	int retries = 0;
6065 
6066 	sb_start_pagefault(inode->i_sb);
6067 	file_update_time(vma->vm_file);
6068 
6069 	down_read(&EXT4_I(inode)->i_mmap_sem);
6070 
6071 	ret = ext4_convert_inline_data(inode);
6072 	if (ret)
6073 		goto out_ret;
6074 
6075 	/* Delalloc case is easy... */
6076 	if (test_opt(inode->i_sb, DELALLOC) &&
6077 	    !ext4_should_journal_data(inode) &&
6078 	    !ext4_nonda_switch(inode->i_sb)) {
6079 		do {
6080 			ret = block_page_mkwrite(vma, vmf,
6081 						   ext4_da_get_block_prep);
6082 		} while (ret == -ENOSPC &&
6083 		       ext4_should_retry_alloc(inode->i_sb, &retries));
6084 		goto out_ret;
6085 	}
6086 
6087 	lock_page(page);
6088 	size = i_size_read(inode);
6089 	/* Page got truncated from under us? */
6090 	if (page->mapping != mapping || page_offset(page) > size) {
6091 		unlock_page(page);
6092 		ret = VM_FAULT_NOPAGE;
6093 		goto out;
6094 	}
6095 
6096 	if (page->index == size >> PAGE_SHIFT)
6097 		len = size & ~PAGE_MASK;
6098 	else
6099 		len = PAGE_SIZE;
6100 	/*
6101 	 * Return if we have all the buffers mapped. This avoids the need to do
6102 	 * journal_start/journal_stop which can block and take a long time
6103 	 */
6104 	if (page_has_buffers(page)) {
6105 		if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6106 					    0, len, NULL,
6107 					    ext4_bh_unmapped)) {
6108 			/* Wait so that we don't change page under IO */
6109 			wait_for_stable_page(page);
6110 			ret = VM_FAULT_LOCKED;
6111 			goto out;
6112 		}
6113 	}
6114 	unlock_page(page);
6115 	/* OK, we need to fill the hole... */
6116 	if (ext4_should_dioread_nolock(inode))
6117 		get_block = ext4_get_block_unwritten;
6118 	else
6119 		get_block = ext4_get_block;
6120 retry_alloc:
6121 	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6122 				    ext4_writepage_trans_blocks(inode));
6123 	if (IS_ERR(handle)) {
6124 		ret = VM_FAULT_SIGBUS;
6125 		goto out;
6126 	}
6127 	ret = block_page_mkwrite(vma, vmf, get_block);
6128 	if (!ret && ext4_should_journal_data(inode)) {
6129 		if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6130 			  PAGE_SIZE, NULL, do_journal_get_write_access)) {
6131 			unlock_page(page);
6132 			ret = VM_FAULT_SIGBUS;
6133 			ext4_journal_stop(handle);
6134 			goto out;
6135 		}
6136 		ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6137 	}
6138 	ext4_journal_stop(handle);
6139 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6140 		goto retry_alloc;
6141 out_ret:
6142 	ret = block_page_mkwrite_return(ret);
6143 out:
6144 	up_read(&EXT4_I(inode)->i_mmap_sem);
6145 	sb_end_pagefault(inode->i_sb);
6146 	return ret;
6147 }
6148 
6149 int ext4_filemap_fault(struct vm_fault *vmf)
6150 {
6151 	struct inode *inode = file_inode(vmf->vma->vm_file);
6152 	int err;
6153 
6154 	down_read(&EXT4_I(inode)->i_mmap_sem);
6155 	err = filemap_fault(vmf);
6156 	up_read(&EXT4_I(inode)->i_mmap_sem);
6157 
6158 	return err;
6159 }
6160