source: src/linux/universal/linux-4.9/fs/ext4/inode.c @ 31885

Last change on this file since 31885 was 31885, checked in by brainslayer, 2 months ago

update

File size: 168.5 KB
Line 
1/*
2 *  linux/fs/ext4/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 *  from
10 *
11 *  linux/fs/minix/inode.c
12 *
13 *  Copyright (C) 1991, 1992  Linus Torvalds
14 *
15 *  64-bit file support on 64-bit platforms by Jakub Jelinek
16 *      (jj@sunsite.ms.mff.cuni.cz)
17 *
18 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
19 */
20
21#include <linux/fs.h>
22#include <linux/time.h>
23#include <linux/highuid.h>
24#include <linux/pagemap.h>
25#include <linux/dax.h>
26#include <linux/quotaops.h>
27#include <linux/string.h>
28#include <linux/buffer_head.h>
29#include <linux/writeback.h>
30#include <linux/pagevec.h>
31#include <linux/mpage.h>
32#include <linux/namei.h>
33#include <linux/uio.h>
34#include <linux/bio.h>
35#include <linux/workqueue.h>
36#include <linux/kernel.h>
37#include <linux/printk.h>
38#include <linux/slab.h>
39#include <linux/bitops.h>
40
41#include "ext4_jbd2.h"
42#include "xattr.h"
43#include "acl.h"
44#include "truncate.h"
45
46#include <trace/events/ext4.h>
47
48#define MPAGE_DA_EXTENT_TAIL 0x01
49
50static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51                              struct ext4_inode_info *ei)
52{
53        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
54        __u32 csum;
55        __u16 dummy_csum = 0;
56        int offset = offsetof(struct ext4_inode, i_checksum_lo);
57        unsigned int csum_size = sizeof(dummy_csum);
58
59        csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
60        csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
61        offset += csum_size;
62        csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
63                           EXT4_GOOD_OLD_INODE_SIZE - offset);
64
65        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
66                offset = offsetof(struct ext4_inode, i_checksum_hi);
67                csum = ext4_chksum(sbi, csum, (__u8 *)raw +
68                                   EXT4_GOOD_OLD_INODE_SIZE,
69                                   offset - EXT4_GOOD_OLD_INODE_SIZE);
70                if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
71                        csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
72                                           csum_size);
73                        offset += csum_size;
74                }
75                csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
76                                   EXT4_INODE_SIZE(inode->i_sb) - offset);
77        }
78
79        return csum;
80}
81
82static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
83                                  struct ext4_inode_info *ei)
84{
85        __u32 provided, calculated;
86
87        if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
88            cpu_to_le32(EXT4_OS_LINUX) ||
89            !ext4_has_metadata_csum(inode->i_sb))
90                return 1;
91
92        provided = le16_to_cpu(raw->i_checksum_lo);
93        calculated = ext4_inode_csum(inode, raw, ei);
94        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
95            EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
96                provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
97        else
98                calculated &= 0xFFFF;
99
100        return provided == calculated;
101}
102
103static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
104                                struct ext4_inode_info *ei)
105{
106        __u32 csum;
107
108        if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
109            cpu_to_le32(EXT4_OS_LINUX) ||
110            !ext4_has_metadata_csum(inode->i_sb))
111                return;
112
113        csum = ext4_inode_csum(inode, raw, ei);
114        raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
115        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
116            EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
117                raw->i_checksum_hi = cpu_to_le16(csum >> 16);
118}
119
120static inline int ext4_begin_ordered_truncate(struct inode *inode,
121                                              loff_t new_size)
122{
123        trace_ext4_begin_ordered_truncate(inode, new_size);
124        /*
125         * If jinode is zero, then we never opened the file for
126         * writing, so there's no need to call
127         * jbd2_journal_begin_ordered_truncate() since there's no
128         * outstanding writes we need to flush.
129         */
130        if (!EXT4_I(inode)->jinode)
131                return 0;
132        return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
133                                                   EXT4_I(inode)->jinode,
134                                                   new_size);
135}
136
137static void ext4_invalidatepage(struct page *page, unsigned int offset,
138                                unsigned int length);
139static int __ext4_journalled_writepage(struct page *page, unsigned int len);
140static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
141static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
142                                  int pextents);
143
144/*
145 * Test whether an inode is a fast symlink.
146 */
147int ext4_inode_is_fast_symlink(struct inode *inode)
148{
149        int ea_blocks = EXT4_I(inode)->i_file_acl ?
150                EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
151
152        if (ext4_has_inline_data(inode))
153                return 0;
154
155        return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
156}
157
158/*
159 * Restart the transaction associated with *handle.  This does a commit,
160 * so before we call here everything must be consistently dirtied against
161 * this transaction.
162 */
163int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
164                                 int nblocks)
165{
166        int ret;
167
168        /*
169         * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
170         * moment, get_block can be called only for blocks inside i_size since
171         * page cache has been already dropped and writes are blocked by
172         * i_mutex. So we can safely drop the i_data_sem here.
173         */
174        BUG_ON(EXT4_JOURNAL(inode) == NULL);
175        jbd_debug(2, "restarting handle %p\n", handle);
176        up_write(&EXT4_I(inode)->i_data_sem);
177        ret = ext4_journal_restart(handle, nblocks);
178        down_write(&EXT4_I(inode)->i_data_sem);
179        ext4_discard_preallocations(inode);
180
181        return ret;
182}
183
184/*
185 * Called at the last iput() if i_nlink is zero.
186 */
187void ext4_evict_inode(struct inode *inode)
188{
189        handle_t *handle;
190        int err;
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                        journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
217                        tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
218
219                        jbd2_complete_transaction(journal, commit_tid);
220                        filemap_write_and_wait(&inode->i_data);
221                }
222                truncate_inode_pages_final(&inode->i_data);
223
224                goto no_delete;
225        }
226
227        if (is_bad_inode(inode))
228                goto no_delete;
229        dquot_initialize(inode);
230
231        if (ext4_should_order_data(inode))
232                ext4_begin_ordered_truncate(inode, 0);
233        truncate_inode_pages_final(&inode->i_data);
234
235        /*
236         * Protect us against freezing - iput() caller didn't have to have any
237         * protection against it
238         */
239        sb_start_intwrite(inode->i_sb);
240        handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
241                                    ext4_blocks_for_truncate(inode)+3);
242        if (IS_ERR(handle)) {
243                ext4_std_error(inode->i_sb, PTR_ERR(handle));
244                /*
245                 * If we're going to skip the normal cleanup, we still need to
246                 * make sure that the in-core orphan linked list is properly
247                 * cleaned up.
248                 */
249                ext4_orphan_del(NULL, inode);
250                sb_end_intwrite(inode->i_sb);
251                goto no_delete;
252        }
253
254        if (IS_SYNC(inode))
255                ext4_handle_sync(handle);
256        inode->i_size = 0;
257        err = ext4_mark_inode_dirty(handle, inode);
258        if (err) {
259                ext4_warning(inode->i_sb,
260                             "couldn't mark inode dirty (err %d)", err);
261                goto stop_handle;
262        }
263        if (inode->i_blocks)
264                ext4_truncate(inode);
265
266        /*
267         * ext4_ext_truncate() doesn't reserve any slop when it
268         * restarts journal transactions; therefore there may not be
269         * enough credits left in the handle to remove the inode from
270         * the orphan list and set the dtime field.
271         */
272        if (!ext4_handle_has_enough_credits(handle, 3)) {
273                err = ext4_journal_extend(handle, 3);
274                if (err > 0)
275                        err = ext4_journal_restart(handle, 3);
276                if (err != 0) {
277                        ext4_warning(inode->i_sb,
278                                     "couldn't extend journal (err %d)", err);
279                stop_handle:
280                        ext4_journal_stop(handle);
281                        ext4_orphan_del(NULL, inode);
282                        sb_end_intwrite(inode->i_sb);
283                        goto no_delete;
284                }
285        }
286
287        /*
288         * Kill off the orphan record which ext4_truncate created.
289         * AKPM: I think this can be inside the above `if'.
290         * Note that ext4_orphan_del() has to be able to cope with the
291         * deletion of a non-existent orphan - this is because we don't
292         * know if ext4_truncate() actually created an orphan record.
293         * (Well, we could do this if we need to, but heck - it works)
294         */
295        ext4_orphan_del(handle, inode);
296        EXT4_I(inode)->i_dtime  = get_seconds();
297
298        /*
299         * One subtle ordering requirement: if anything has gone wrong
300         * (transaction abort, IO errors, whatever), then we can still
301         * do these next steps (the fs will already have been marked as
302         * having errors), but we can't free the inode if the mark_dirty
303         * fails.
304         */
305        if (ext4_mark_inode_dirty(handle, inode))
306                /* If that failed, just do the required in-core inode clear. */
307                ext4_clear_inode(inode);
308        else
309                ext4_free_inode(handle, inode);
310        ext4_journal_stop(handle);
311        sb_end_intwrite(inode->i_sb);
312        return;
313no_delete:
314        ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
315}
316
317#ifdef CONFIG_QUOTA
318qsize_t *ext4_get_reserved_space(struct inode *inode)
319{
320        return &EXT4_I(inode)->i_reserved_quota;
321}
322#endif
323
324/*
325 * Called with i_data_sem down, which is important since we can call
326 * ext4_discard_preallocations() from here.
327 */
328void ext4_da_update_reserve_space(struct inode *inode,
329                                        int used, int quota_claim)
330{
331        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
332        struct ext4_inode_info *ei = EXT4_I(inode);
333
334        spin_lock(&ei->i_block_reservation_lock);
335        trace_ext4_da_update_reserve_space(inode, used, quota_claim);
336        if (unlikely(used > ei->i_reserved_data_blocks)) {
337                ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
338                         "with only %d reserved data blocks",
339                         __func__, inode->i_ino, used,
340                         ei->i_reserved_data_blocks);
341                WARN_ON(1);
342                used = ei->i_reserved_data_blocks;
343        }
344
345        /* Update per-inode reservations */
346        ei->i_reserved_data_blocks -= used;
347        percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
348
349        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
350
351        /* Update quota subsystem for data blocks */
352        if (quota_claim)
353                dquot_claim_block(inode, EXT4_C2B(sbi, used));
354        else {
355                /*
356                 * We did fallocate with an offset that is already delayed
357                 * allocated. So on delayed allocated writeback we should
358                 * not re-claim the quota for fallocated blocks.
359                 */
360                dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
361        }
362
363        /*
364         * If we have done all the pending block allocations and if
365         * there aren't any writers on the inode, we can discard the
366         * inode's preallocations.
367         */
368        if ((ei->i_reserved_data_blocks == 0) &&
369            (atomic_read(&inode->i_writecount) == 0))
370                ext4_discard_preallocations(inode);
371}
372
373static int __check_block_validity(struct inode *inode, const char *func,
374                                unsigned int line,
375                                struct ext4_map_blocks *map)
376{
377        if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
378                                   map->m_len)) {
379                ext4_error_inode(inode, func, line, map->m_pblk,
380                                 "lblock %lu mapped to illegal pblock "
381                                 "(length %d)", (unsigned long) map->m_lblk,
382                                 map->m_len);
383                return -EFSCORRUPTED;
384        }
385        return 0;
386}
387
388int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
389                       ext4_lblk_t len)
390{
391        int ret;
392
393        if (ext4_encrypted_inode(inode))
394                return fscrypt_zeroout_range(inode, lblk, pblk, len);
395
396        ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
397        if (ret > 0)
398                ret = 0;
399
400        return ret;
401}
402
403#define check_block_validity(inode, map)        \
404        __check_block_validity((inode), __func__, __LINE__, (map))
405
406#ifdef ES_AGGRESSIVE_TEST
407static void ext4_map_blocks_es_recheck(handle_t *handle,
408                                       struct inode *inode,
409                                       struct ext4_map_blocks *es_map,
410                                       struct ext4_map_blocks *map,
411                                       int flags)
412{
413        int retval;
414
415        map->m_flags = 0;
416        /*
417         * There is a race window that the result is not the same.
418         * e.g. xfstests #223 when dioread_nolock enables.  The reason
419         * is that we lookup a block mapping in extent status tree with
420         * out taking i_data_sem.  So at the time the unwritten extent
421         * could be converted.
422         */
423        down_read(&EXT4_I(inode)->i_data_sem);
424        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
425                retval = ext4_ext_map_blocks(handle, inode, map, flags &
426                                             EXT4_GET_BLOCKS_KEEP_SIZE);
427        } else {
428                retval = ext4_ind_map_blocks(handle, inode, map, flags &
429                                             EXT4_GET_BLOCKS_KEEP_SIZE);
430        }
431        up_read((&EXT4_I(inode)->i_data_sem));
432
433        /*
434         * We don't check m_len because extent will be collpased in status
435         * tree.  So the m_len might not equal.
436         */
437        if (es_map->m_lblk != map->m_lblk ||
438            es_map->m_flags != map->m_flags ||
439            es_map->m_pblk != map->m_pblk) {
440                printk("ES cache assertion failed for inode: %lu "
441                       "es_cached ex [%d/%d/%llu/%x] != "
442                       "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
443                       inode->i_ino, es_map->m_lblk, es_map->m_len,
444                       es_map->m_pblk, es_map->m_flags, map->m_lblk,
445                       map->m_len, map->m_pblk, map->m_flags,
446                       retval, flags);
447        }
448}
449#endif /* ES_AGGRESSIVE_TEST */
450
451/*
452 * The ext4_map_blocks() function tries to look up the requested blocks,
453 * and returns if the blocks are already mapped.
454 *
455 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
456 * and store the allocated blocks in the result buffer head and mark it
457 * mapped.
458 *
459 * If file type is extents based, it will call ext4_ext_map_blocks(),
460 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
461 * based files
462 *
463 * On success, it returns the number of blocks being mapped or allocated.  if
464 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
465 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
466 *
467 * It returns 0 if plain look up failed (blocks have not been allocated), in
468 * that case, @map is returned as unmapped but we still do fill map->m_len to
469 * indicate the length of a hole starting at map->m_lblk.
470 *
471 * It returns the error in case of allocation failure.
472 */
473int ext4_map_blocks(handle_t *handle, struct inode *inode,
474                    struct ext4_map_blocks *map, int flags)
475{
476        struct extent_status es;
477        int retval;
478        int ret = 0;
479#ifdef ES_AGGRESSIVE_TEST
480        struct ext4_map_blocks orig_map;
481
482        memcpy(&orig_map, map, sizeof(*map));
483#endif
484
485        map->m_flags = 0;
486        ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
487                  "logical block %lu\n", inode->i_ino, flags, map->m_len,
488                  (unsigned long) map->m_lblk);
489
490        /*
491         * ext4_map_blocks returns an int, and m_len is an unsigned int
492         */
493        if (unlikely(map->m_len > INT_MAX))
494                map->m_len = INT_MAX;
495
496        /* We can handle the block number less than EXT_MAX_BLOCKS */
497        if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
498                return -EFSCORRUPTED;
499
500        /* Lookup extent status tree firstly */
501        if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
502                if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
503                        map->m_pblk = ext4_es_pblock(&es) +
504                                        map->m_lblk - es.es_lblk;
505                        map->m_flags |= ext4_es_is_written(&es) ?
506                                        EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
507                        retval = es.es_len - (map->m_lblk - es.es_lblk);
508                        if (retval > map->m_len)
509                                retval = map->m_len;
510                        map->m_len = retval;
511                } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
512                        map->m_pblk = 0;
513                        retval = es.es_len - (map->m_lblk - es.es_lblk);
514                        if (retval > map->m_len)
515                                retval = map->m_len;
516                        map->m_len = retval;
517                        retval = 0;
518                } else {
519                        BUG_ON(1);
520                }
521#ifdef ES_AGGRESSIVE_TEST
522                ext4_map_blocks_es_recheck(handle, inode, map,
523                                           &orig_map, flags);
524#endif
525                goto found;
526        }
527
528        /*
529         * Try to see if we can get the block without requesting a new
530         * file system block.
531         */
532        down_read(&EXT4_I(inode)->i_data_sem);
533        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
534                retval = ext4_ext_map_blocks(handle, inode, map, flags &
535                                             EXT4_GET_BLOCKS_KEEP_SIZE);
536        } else {
537                retval = ext4_ind_map_blocks(handle, inode, map, flags &
538                                             EXT4_GET_BLOCKS_KEEP_SIZE);
539        }
540        if (retval > 0) {
541                unsigned int status;
542
543                if (unlikely(retval != map->m_len)) {
544                        ext4_warning(inode->i_sb,
545                                     "ES len assertion failed for inode "
546                                     "%lu: retval %d != map->m_len %d",
547                                     inode->i_ino, retval, map->m_len);
548                        WARN_ON(1);
549                }
550
551                status = map->m_flags & EXT4_MAP_UNWRITTEN ?
552                                EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
553                if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
554                    !(status & EXTENT_STATUS_WRITTEN) &&
555                    ext4_find_delalloc_range(inode, map->m_lblk,
556                                             map->m_lblk + map->m_len - 1))
557                        status |= EXTENT_STATUS_DELAYED;
558                ret = ext4_es_insert_extent(inode, map->m_lblk,
559                                            map->m_len, map->m_pblk, status);
560                if (ret < 0)
561                        retval = ret;
562        }
563        up_read((&EXT4_I(inode)->i_data_sem));
564
565found:
566        if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
567                ret = check_block_validity(inode, map);
568                if (ret != 0)
569                        return ret;
570        }
571
572        /* If it is only a block(s) look up */
573        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
574                return retval;
575
576        /*
577         * Returns if the blocks have already allocated
578         *
579         * Note that if blocks have been preallocated
580         * ext4_ext_get_block() returns the create = 0
581         * with buffer head unmapped.
582         */
583        if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
584                /*
585                 * If we need to convert extent to unwritten
586                 * we continue and do the actual work in
587                 * ext4_ext_map_blocks()
588                 */
589                if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
590                        return retval;
591
592        /*
593         * Here we clear m_flags because after allocating an new extent,
594         * it will be set again.
595         */
596        map->m_flags &= ~EXT4_MAP_FLAGS;
597
598        /*
599         * New blocks allocate and/or writing to unwritten extent
600         * will possibly result in updating i_data, so we take
601         * the write lock of i_data_sem, and call get_block()
602         * with create == 1 flag.
603         */
604        down_write(&EXT4_I(inode)->i_data_sem);
605
606        /*
607         * We need to check for EXT4 here because migrate
608         * could have changed the inode type in between
609         */
610        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
611                retval = ext4_ext_map_blocks(handle, inode, map, flags);
612        } else {
613                retval = ext4_ind_map_blocks(handle, inode, map, flags);
614
615                if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
616                        /*
617                         * We allocated new blocks which will result in
618                         * i_data's format changing.  Force the migrate
619                         * to fail by clearing migrate flags
620                         */
621                        ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
622                }
623
624                /*
625                 * Update reserved blocks/metadata blocks after successful
626                 * block allocation which had been deferred till now. We don't
627                 * support fallocate for non extent files. So we can update
628                 * reserve space here.
629                 */
630                if ((retval > 0) &&
631                        (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
632                        ext4_da_update_reserve_space(inode, retval, 1);
633        }
634
635        if (retval > 0) {
636                unsigned int status;
637
638                if (unlikely(retval != map->m_len)) {
639                        ext4_warning(inode->i_sb,
640                                     "ES len assertion failed for inode "
641                                     "%lu: retval %d != map->m_len %d",
642                                     inode->i_ino, retval, map->m_len);
643                        WARN_ON(1);
644                }
645
646                /*
647                 * We have to zeroout blocks before inserting them into extent
648                 * status tree. Otherwise someone could look them up there and
649                 * use them before they are really zeroed. We also have to
650                 * unmap metadata before zeroing as otherwise writeback can
651                 * overwrite zeros with stale data from block device.
652                 */
653                if (flags & EXT4_GET_BLOCKS_ZERO &&
654                    map->m_flags & EXT4_MAP_MAPPED &&
655                    map->m_flags & EXT4_MAP_NEW) {
656                        ext4_lblk_t i;
657
658                        for (i = 0; i < map->m_len; i++) {
659                                unmap_underlying_metadata(inode->i_sb->s_bdev,
660                                                          map->m_pblk + i);
661                        }
662                        ret = ext4_issue_zeroout(inode, map->m_lblk,
663                                                 map->m_pblk, map->m_len);
664                        if (ret) {
665                                retval = ret;
666                                goto out_sem;
667                        }
668                }
669
670                /*
671                 * If the extent has been zeroed out, we don't need to update
672                 * extent status tree.
673                 */
674                if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
675                    ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
676                        if (ext4_es_is_written(&es))
677                                goto out_sem;
678                }
679                status = map->m_flags & EXT4_MAP_UNWRITTEN ?
680                                EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
681                if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
682                    !(status & EXTENT_STATUS_WRITTEN) &&
683                    ext4_find_delalloc_range(inode, map->m_lblk,
684                                             map->m_lblk + map->m_len - 1))
685                        status |= EXTENT_STATUS_DELAYED;
686                ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
687                                            map->m_pblk, status);
688                if (ret < 0) {
689                        retval = ret;
690                        goto out_sem;
691                }
692        }
693
694out_sem:
695        up_write((&EXT4_I(inode)->i_data_sem));
696        if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
697                ret = check_block_validity(inode, map);
698                if (ret != 0)
699                        return ret;
700
701                /*
702                 * Inodes with freshly allocated blocks where contents will be
703                 * visible after transaction commit must be on transaction's
704                 * ordered data list.
705                 */
706                if (map->m_flags & EXT4_MAP_NEW &&
707                    !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
708                    !(flags & EXT4_GET_BLOCKS_ZERO) &&
709                    !IS_NOQUOTA(inode) &&
710                    ext4_should_order_data(inode)) {
711                        if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
712                                ret = ext4_jbd2_inode_add_wait(handle, inode);
713                        else
714                                ret = ext4_jbd2_inode_add_write(handle, inode);
715                        if (ret)
716                                return ret;
717                }
718        }
719        return retval;
720}
721
722/*
723 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
724 * we have to be careful as someone else may be manipulating b_state as well.
725 */
726static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
727{
728        unsigned long old_state;
729        unsigned long new_state;
730
731        flags &= EXT4_MAP_FLAGS;
732
733        /* Dummy buffer_head? Set non-atomically. */
734        if (!bh->b_page) {
735                bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
736                return;
737        }
738        /*
739         * Someone else may be modifying b_state. Be careful! This is ugly but
740         * once we get rid of using bh as a container for mapping information
741         * to pass to / from get_block functions, this can go away.
742         */
743        do {
744                old_state = READ_ONCE(bh->b_state);
745                new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
746        } while (unlikely(
747                 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
748}
749
750static int _ext4_get_block(struct inode *inode, sector_t iblock,
751                           struct buffer_head *bh, int flags)
752{
753        struct ext4_map_blocks map;
754        int ret = 0;
755
756        if (ext4_has_inline_data(inode))
757                return -ERANGE;
758
759        map.m_lblk = iblock;
760        map.m_len = bh->b_size >> inode->i_blkbits;
761
762        ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
763                              flags);
764        if (ret > 0) {
765                map_bh(bh, inode->i_sb, map.m_pblk);
766                ext4_update_bh_state(bh, map.m_flags);
767                bh->b_size = inode->i_sb->s_blocksize * map.m_len;
768                ret = 0;
769        }
770        return ret;
771}
772
773int ext4_get_block(struct inode *inode, sector_t iblock,
774                   struct buffer_head *bh, int create)
775{
776        return _ext4_get_block(inode, iblock, bh,
777                               create ? EXT4_GET_BLOCKS_CREATE : 0);
778}
779
780/*
781 * Get block function used when preparing for buffered write if we require
782 * creating an unwritten extent if blocks haven't been allocated.  The extent
783 * will be converted to written after the IO is complete.
784 */
785int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
786                             struct buffer_head *bh_result, int create)
787{
788        ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
789                   inode->i_ino, create);
790        return _ext4_get_block(inode, iblock, bh_result,
791                               EXT4_GET_BLOCKS_IO_CREATE_EXT);
792}
793
794/* Maximum number of blocks we map for direct IO at once. */
795#define DIO_MAX_BLOCKS 4096
796
797/*
798 * Get blocks function for the cases that need to start a transaction -
799 * generally difference cases of direct IO and DAX IO. It also handles retries
800 * in case of ENOSPC.
801 */
802static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
803                                struct buffer_head *bh_result, int flags)
804{
805        int dio_credits;
806        handle_t *handle;
807        int retries = 0;
808        int ret;
809
810        /* Trim mapping request to maximum we can map at once for DIO */
811        if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
812                bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
813        dio_credits = ext4_chunk_trans_blocks(inode,
814                                      bh_result->b_size >> inode->i_blkbits);
815retry:
816        handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
817        if (IS_ERR(handle))
818                return PTR_ERR(handle);
819
820        ret = _ext4_get_block(inode, iblock, bh_result, flags);
821        ext4_journal_stop(handle);
822
823        if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
824                goto retry;
825        return ret;
826}
827
828/* Get block function for DIO reads and writes to inodes without extents */
829int ext4_dio_get_block(struct inode *inode, sector_t iblock,
830                       struct buffer_head *bh, int create)
831{
832        /* We don't expect handle for direct IO */
833        WARN_ON_ONCE(ext4_journal_current_handle());
834
835        if (!create)
836                return _ext4_get_block(inode, iblock, bh, 0);
837        return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
838}
839
840/*
841 * Get block function for AIO DIO writes when we create unwritten extent if
842 * blocks are not allocated yet. The extent will be converted to written
843 * after IO is complete.
844 */
845static int ext4_dio_get_block_unwritten_async(struct inode *inode,
846                sector_t iblock, struct buffer_head *bh_result, int create)
847{
848        int ret;
849
850        /* We don't expect handle for direct IO */
851        WARN_ON_ONCE(ext4_journal_current_handle());
852
853        ret = ext4_get_block_trans(inode, iblock, bh_result,
854                                   EXT4_GET_BLOCKS_IO_CREATE_EXT);
855
856        /*
857         * When doing DIO using unwritten extents, we need io_end to convert
858         * unwritten extents to written on IO completion. We allocate io_end
859         * once we spot unwritten extent and store it in b_private. Generic
860         * DIO code keeps b_private set and furthermore passes the value to
861         * our completion callback in 'private' argument.
862         */
863        if (!ret && buffer_unwritten(bh_result)) {
864                if (!bh_result->b_private) {
865                        ext4_io_end_t *io_end;
866
867                        io_end = ext4_init_io_end(inode, GFP_KERNEL);
868                        if (!io_end)
869                                return -ENOMEM;
870                        bh_result->b_private = io_end;
871                        ext4_set_io_unwritten_flag(inode, io_end);
872                }
873                set_buffer_defer_completion(bh_result);
874        }
875
876        return ret;
877}
878
879/*
880 * Get block function for non-AIO DIO writes when we create unwritten extent if
881 * blocks are not allocated yet. The extent will be converted to written
882 * after IO is complete from ext4_ext_direct_IO() function.
883 */
884static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
885                sector_t iblock, struct buffer_head *bh_result, int create)
886{
887        int ret;
888
889        /* We don't expect handle for direct IO */
890        WARN_ON_ONCE(ext4_journal_current_handle());
891
892        ret = ext4_get_block_trans(inode, iblock, bh_result,
893                                   EXT4_GET_BLOCKS_IO_CREATE_EXT);
894
895        /*
896         * Mark inode as having pending DIO writes to unwritten extents.
897         * ext4_ext_direct_IO() checks this flag and converts extents to
898         * written.
899         */
900        if (!ret && buffer_unwritten(bh_result))
901                ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
902
903        return ret;
904}
905
906static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
907                   struct buffer_head *bh_result, int create)
908{
909        int ret;
910
911        ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
912                   inode->i_ino, create);
913        /* We don't expect handle for direct IO */
914        WARN_ON_ONCE(ext4_journal_current_handle());
915
916        ret = _ext4_get_block(inode, iblock, bh_result, 0);
917        /*
918         * Blocks should have been preallocated! ext4_file_write_iter() checks
919         * that.
920         */
921        WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
922
923        return ret;
924}
925
926
927/*
928 * `handle' can be NULL if create is zero
929 */
930struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
931                                ext4_lblk_t block, int map_flags)
932{
933        struct ext4_map_blocks map;
934        struct buffer_head *bh;
935        int create = map_flags & EXT4_GET_BLOCKS_CREATE;
936        int err;
937
938        J_ASSERT(handle != NULL || create == 0);
939
940        map.m_lblk = block;
941        map.m_len = 1;
942        err = ext4_map_blocks(handle, inode, &map, map_flags);
943
944        if (err == 0)
945                return create ? ERR_PTR(-ENOSPC) : NULL;
946        if (err < 0)
947                return ERR_PTR(err);
948
949        bh = sb_getblk(inode->i_sb, map.m_pblk);
950        if (unlikely(!bh))
951                return ERR_PTR(-ENOMEM);
952        if (map.m_flags & EXT4_MAP_NEW) {
953                J_ASSERT(create != 0);
954                J_ASSERT(handle != NULL);
955
956                /*
957                 * Now that we do not always journal data, we should
958                 * keep in mind whether this should always journal the
959                 * new buffer as metadata.  For now, regular file
960                 * writes use ext4_get_block instead, so it's not a
961                 * problem.
962                 */
963                lock_buffer(bh);
964                BUFFER_TRACE(bh, "call get_create_access");
965                err = ext4_journal_get_create_access(handle, bh);
966                if (unlikely(err)) {
967                        unlock_buffer(bh);
968                        goto errout;
969                }
970                if (!buffer_uptodate(bh)) {
971                        memset(bh->b_data, 0, inode->i_sb->s_blocksize);
972                        set_buffer_uptodate(bh);
973                }
974                unlock_buffer(bh);
975                BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
976                err = ext4_handle_dirty_metadata(handle, inode, bh);
977                if (unlikely(err))
978                        goto errout;
979        } else
980                BUFFER_TRACE(bh, "not a new buffer");
981        return bh;
982errout:
983        brelse(bh);
984        return ERR_PTR(err);
985}
986
987struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
988                               ext4_lblk_t block, int map_flags)
989{
990        struct buffer_head *bh;
991
992        bh = ext4_getblk(handle, inode, block, map_flags);
993        if (IS_ERR(bh))
994                return bh;
995        if (!bh || buffer_uptodate(bh))
996                return bh;
997        ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
998        wait_on_buffer(bh);
999        if (buffer_uptodate(bh))
1000                return bh;
1001        put_bh(bh);
1002        return ERR_PTR(-EIO);
1003}
1004
1005int ext4_walk_page_buffers(handle_t *handle,
1006                           struct buffer_head *head,
1007                           unsigned from,
1008                           unsigned to,
1009                           int *partial,
1010                           int (*fn)(handle_t *handle,
1011                                     struct buffer_head *bh))
1012{
1013        struct buffer_head *bh;
1014        unsigned block_start, block_end;
1015        unsigned blocksize = head->b_size;
1016        int err, ret = 0;
1017        struct buffer_head *next;
1018
1019        for (bh = head, block_start = 0;
1020             ret == 0 && (bh != head || !block_start);
1021             block_start = block_end, bh = next) {
1022                next = bh->b_this_page;
1023                block_end = block_start + blocksize;
1024                if (block_end <= from || block_start >= to) {
1025                        if (partial && !buffer_uptodate(bh))
1026                                *partial = 1;
1027                        continue;
1028                }
1029                err = (*fn)(handle, bh);
1030                if (!ret)
1031                        ret = err;
1032        }
1033        return ret;
1034}
1035
1036/*
1037 * To preserve ordering, it is essential that the hole instantiation and
1038 * the data write be encapsulated in a single transaction.  We cannot
1039 * close off a transaction and start a new one between the ext4_get_block()
1040 * and the commit_write().  So doing the jbd2_journal_start at the start of
1041 * prepare_write() is the right place.
1042 *
1043 * Also, this function can nest inside ext4_writepage().  In that case, we
1044 * *know* that ext4_writepage() has generated enough buffer credits to do the
1045 * whole page.  So we won't block on the journal in that case, which is good,
1046 * because the caller may be PF_MEMALLOC.
1047 *
1048 * By accident, ext4 can be reentered when a transaction is open via
1049 * quota file writes.  If we were to commit the transaction while thus
1050 * reentered, there can be a deadlock - we would be holding a quota
1051 * lock, and the commit would never complete if another thread had a
1052 * transaction open and was blocking on the quota lock - a ranking
1053 * violation.
1054 *
1055 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1056 * will _not_ run commit under these circumstances because handle->h_ref
1057 * is elevated.  We'll still have enough credits for the tiny quotafile
1058 * write.
1059 */
1060int do_journal_get_write_access(handle_t *handle,
1061                                struct buffer_head *bh)
1062{
1063        int dirty = buffer_dirty(bh);
1064        int ret;
1065
1066        if (!buffer_mapped(bh) || buffer_freed(bh))
1067                return 0;
1068        /*
1069         * __block_write_begin() could have dirtied some buffers. Clean
1070         * the dirty bit as jbd2_journal_get_write_access() could complain
1071         * otherwise about fs integrity issues. Setting of the dirty bit
1072         * by __block_write_begin() isn't a real problem here as we clear
1073         * the bit before releasing a page lock and thus writeback cannot
1074         * ever write the buffer.
1075         */
1076        if (dirty)
1077                clear_buffer_dirty(bh);
1078        BUFFER_TRACE(bh, "get write access");
1079        ret = ext4_journal_get_write_access(handle, bh);
1080        if (!ret && dirty)
1081                ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1082        return ret;
1083}
1084
1085#ifdef CONFIG_EXT4_FS_ENCRYPTION
1086static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1087                                  get_block_t *get_block)
1088{
1089        unsigned from = pos & (PAGE_SIZE - 1);
1090        unsigned to = from + len;
1091        struct inode *inode = page->mapping->host;
1092        unsigned block_start, block_end;
1093        sector_t block;
1094        int err = 0;
1095        unsigned blocksize = inode->i_sb->s_blocksize;
1096        unsigned bbits;
1097        struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1098        bool decrypt = false;
1099
1100        BUG_ON(!PageLocked(page));
1101        BUG_ON(from > PAGE_SIZE);
1102        BUG_ON(to > PAGE_SIZE);
1103        BUG_ON(from > to);
1104
1105        if (!page_has_buffers(page))
1106                create_empty_buffers(page, blocksize, 0);
1107        head = page_buffers(page);
1108        bbits = ilog2(blocksize);
1109        block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1110
1111        for (bh = head, block_start = 0; bh != head || !block_start;
1112            block++, block_start = block_end, bh = bh->b_this_page) {
1113                block_end = block_start + blocksize;
1114                if (block_end <= from || block_start >= to) {
1115                        if (PageUptodate(page)) {
1116                                if (!buffer_uptodate(bh))
1117                                        set_buffer_uptodate(bh);
1118                        }
1119                        continue;
1120                }
1121                if (buffer_new(bh))
1122                        clear_buffer_new(bh);
1123                if (!buffer_mapped(bh)) {
1124                        WARN_ON(bh->b_size != blocksize);
1125                        err = get_block(inode, block, bh, 1);
1126                        if (err)
1127                                break;
1128                        if (buffer_new(bh)) {
1129                                unmap_underlying_metadata(bh->b_bdev,
1130                                                          bh->b_blocknr);
1131                                if (PageUptodate(page)) {
1132                                        clear_buffer_new(bh);
1133                                        set_buffer_uptodate(bh);
1134                                        mark_buffer_dirty(bh);
1135                                        continue;
1136                                }
1137                                if (block_end > to || block_start < from)
1138                                        zero_user_segments(page, to, block_end,
1139                                                           block_start, from);
1140                                continue;
1141                        }
1142                }
1143                if (PageUptodate(page)) {
1144                        if (!buffer_uptodate(bh))
1145                                set_buffer_uptodate(bh);
1146                        continue;
1147                }
1148                if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1149                    !buffer_unwritten(bh) &&
1150                    (block_start < from || block_end > to)) {
1151                        ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1152                        *wait_bh++ = bh;
1153                        decrypt = ext4_encrypted_inode(inode) &&
1154                                S_ISREG(inode->i_mode);
1155                }
1156        }
1157        /*
1158         * If we issued read requests, let them complete.
1159         */
1160        while (wait_bh > wait) {
1161                wait_on_buffer(*--wait_bh);
1162                if (!buffer_uptodate(*wait_bh))
1163                        err = -EIO;
1164        }
1165        if (unlikely(err))
1166                page_zero_new_buffers(page, from, to);
1167        else if (decrypt)
1168                err = fscrypt_decrypt_page(page);
1169        return err;
1170}
1171#endif
1172
1173static int ext4_write_begin(struct file *file, struct address_space *mapping,
1174                            loff_t pos, unsigned len, unsigned flags,
1175                            struct page **pagep, void **fsdata)
1176{
1177        struct inode *inode = mapping->host;
1178        int ret, needed_blocks;
1179        handle_t *handle;
1180        int retries = 0;
1181        struct page *page;
1182        pgoff_t index;
1183        unsigned from, to;
1184
1185        trace_ext4_write_begin(inode, pos, len, flags);
1186        /*
1187         * Reserve one block more for addition to orphan list in case
1188         * we allocate blocks but write fails for some reason
1189         */
1190        needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1191        index = pos >> PAGE_SHIFT;
1192        from = pos & (PAGE_SIZE - 1);
1193        to = from + len;
1194
1195        if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1196                ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1197                                                    flags, pagep);
1198                if (ret < 0)
1199                        return ret;
1200                if (ret == 1)
1201                        return 0;
1202        }
1203
1204        /*
1205         * grab_cache_page_write_begin() can take a long time if the
1206         * system is thrashing due to memory pressure, or if the page
1207         * is being written back.  So grab it first before we start
1208         * the transaction handle.  This also allows us to allocate
1209         * the page (if needed) without using GFP_NOFS.
1210         */
1211retry_grab:
1212        page = grab_cache_page_write_begin(mapping, index, flags);
1213        if (!page)
1214                return -ENOMEM;
1215        unlock_page(page);
1216
1217retry_journal:
1218        handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1219        if (IS_ERR(handle)) {
1220                put_page(page);
1221                return PTR_ERR(handle);
1222        }
1223
1224        lock_page(page);
1225        if (page->mapping != mapping) {
1226                /* The page got truncated from under us */
1227                unlock_page(page);
1228                put_page(page);
1229                ext4_journal_stop(handle);
1230                goto retry_grab;
1231        }
1232        /* In case writeback began while the page was unlocked */
1233        wait_for_stable_page(page);
1234
1235#ifdef CONFIG_EXT4_FS_ENCRYPTION
1236        if (ext4_should_dioread_nolock(inode))
1237                ret = ext4_block_write_begin(page, pos, len,
1238                                             ext4_get_block_unwritten);
1239        else
1240                ret = ext4_block_write_begin(page, pos, len,
1241                                             ext4_get_block);
1242#else
1243        if (ext4_should_dioread_nolock(inode))
1244                ret = __block_write_begin(page, pos, len,
1245                                          ext4_get_block_unwritten);
1246        else
1247                ret = __block_write_begin(page, pos, len, ext4_get_block);
1248#endif
1249        if (!ret && ext4_should_journal_data(inode)) {
1250                ret = ext4_walk_page_buffers(handle, page_buffers(page),
1251                                             from, to, NULL,
1252                                             do_journal_get_write_access);
1253        }
1254
1255        if (ret) {
1256                unlock_page(page);
1257                /*
1258                 * __block_write_begin may have instantiated a few blocks
1259                 * outside i_size.  Trim these off again. Don't need
1260                 * i_size_read because we hold i_mutex.
1261                 *
1262                 * Add inode to orphan list in case we crash before
1263                 * truncate finishes
1264                 */
1265                if (pos + len > inode->i_size && ext4_can_truncate(inode))
1266                        ext4_orphan_add(handle, inode);
1267
1268                ext4_journal_stop(handle);
1269                if (pos + len > inode->i_size) {
1270                        ext4_truncate_failed_write(inode);
1271                        /*
1272                         * If truncate failed early the inode might
1273                         * still be on the orphan list; we need to
1274                         * make sure the inode is removed from the
1275                         * orphan list in that case.
1276                         */
1277                        if (inode->i_nlink)
1278                                ext4_orphan_del(NULL, inode);
1279                }
1280
1281                if (ret == -ENOSPC &&
1282                    ext4_should_retry_alloc(inode->i_sb, &retries))
1283                        goto retry_journal;
1284                put_page(page);
1285                return ret;
1286        }
1287        *pagep = page;
1288        return ret;
1289}
1290
1291/* For write_end() in data=journal mode */
1292static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1293{
1294        int ret;
1295        if (!buffer_mapped(bh) || buffer_freed(bh))
1296                return 0;
1297        set_buffer_uptodate(bh);
1298        ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1299        clear_buffer_meta(bh);
1300        clear_buffer_prio(bh);
1301        return ret;
1302}
1303
1304/*
1305 * We need to pick up the new inode size which generic_commit_write gave us
1306 * `file' can be NULL - eg, when called from page_symlink().
1307 *
1308 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1309 * buffers are managed internally.
1310 */
1311static int ext4_write_end(struct file *file,
1312                          struct address_space *mapping,
1313                          loff_t pos, unsigned len, unsigned copied,
1314                          struct page *page, void *fsdata)
1315{
1316        handle_t *handle = ext4_journal_current_handle();
1317        struct inode *inode = mapping->host;
1318        loff_t old_size = inode->i_size;
1319        int ret = 0, ret2;
1320        int i_size_changed = 0;
1321
1322        trace_ext4_write_end(inode, pos, len, copied);
1323        if (ext4_has_inline_data(inode)) {
1324                ret = ext4_write_inline_data_end(inode, pos, len,
1325                                                 copied, page);
1326                if (ret < 0) {
1327                        unlock_page(page);
1328                        put_page(page);
1329                        goto errout;
1330                }
1331                copied = ret;
1332        } else
1333                copied = block_write_end(file, mapping, pos,
1334                                         len, copied, page, fsdata);
1335        /*
1336         * it's important to update i_size while still holding page lock:
1337         * page writeout could otherwise come in and zero beyond i_size.
1338         */
1339        i_size_changed = ext4_update_inode_size(inode, pos + copied);
1340        unlock_page(page);
1341        put_page(page);
1342
1343        if (old_size < pos)
1344                pagecache_isize_extended(inode, old_size, pos);
1345        /*
1346         * Don't mark the inode dirty under page lock. First, it unnecessarily
1347         * makes the holding time of page lock longer. Second, it forces lock
1348         * ordering of page lock and transaction start for journaling
1349         * filesystems.
1350         */
1351        if (i_size_changed)
1352                ext4_mark_inode_dirty(handle, inode);
1353
1354        if (pos + len > inode->i_size && ext4_can_truncate(inode))
1355                /* if we have allocated more blocks and copied
1356                 * less. We will have blocks allocated outside
1357                 * inode->i_size. So truncate them
1358                 */
1359                ext4_orphan_add(handle, inode);
1360errout:
1361        ret2 = ext4_journal_stop(handle);
1362        if (!ret)
1363                ret = ret2;
1364
1365        if (pos + len > inode->i_size) {
1366                ext4_truncate_failed_write(inode);
1367                /*
1368                 * If truncate failed early the inode might still be
1369                 * on the orphan list; we need to make sure the inode
1370                 * is removed from the orphan list in that case.
1371                 */
1372                if (inode->i_nlink)
1373                        ext4_orphan_del(NULL, inode);
1374        }
1375
1376        return ret ? ret : copied;
1377}
1378
1379/*
1380 * This is a private version of page_zero_new_buffers() which doesn't
1381 * set the buffer to be dirty, since in data=journalled mode we need
1382 * to call ext4_handle_dirty_metadata() instead.
1383 */
1384static void ext4_journalled_zero_new_buffers(handle_t *handle,
1385                                            struct page *page,
1386                                            unsigned from, unsigned to)
1387{
1388        unsigned int block_start = 0, block_end;
1389        struct buffer_head *head, *bh;
1390
1391        bh = head = page_buffers(page);
1392        do {
1393                block_end = block_start + bh->b_size;
1394                if (buffer_new(bh)) {
1395                        if (block_end > from && block_start < to) {
1396                                if (!PageUptodate(page)) {
1397                                        unsigned start, size;
1398
1399                                        start = max(from, block_start);
1400                                        size = min(to, block_end) - start;
1401
1402                                        zero_user(page, start, size);
1403                                        write_end_fn(handle, bh);
1404                                }
1405                                clear_buffer_new(bh);
1406                        }
1407                }
1408                block_start = block_end;
1409                bh = bh->b_this_page;
1410        } while (bh != head);
1411}
1412
1413static int ext4_journalled_write_end(struct file *file,
1414                                     struct address_space *mapping,
1415                                     loff_t pos, unsigned len, unsigned copied,
1416                                     struct page *page, void *fsdata)
1417{
1418        handle_t *handle = ext4_journal_current_handle();
1419        struct inode *inode = mapping->host;
1420        loff_t old_size = inode->i_size;
1421        int ret = 0, ret2;
1422        int partial = 0;
1423        unsigned from, to;
1424        int size_changed = 0;
1425
1426        trace_ext4_journalled_write_end(inode, pos, len, copied);
1427        from = pos & (PAGE_SIZE - 1);
1428        to = from + len;
1429
1430        BUG_ON(!ext4_handle_valid(handle));
1431
1432        if (ext4_has_inline_data(inode)) {
1433                ret = ext4_write_inline_data_end(inode, pos, len,
1434                                                 copied, page);
1435                if (ret < 0) {
1436                        unlock_page(page);
1437                        put_page(page);
1438                        goto errout;
1439                }
1440                copied = ret;
1441        } else if (unlikely(copied < len) && !PageUptodate(page)) {
1442                copied = 0;
1443                ext4_journalled_zero_new_buffers(handle, page, from, to);
1444        } else {
1445                if (unlikely(copied < len))
1446                        ext4_journalled_zero_new_buffers(handle, page,
1447                                                         from + copied, to);
1448                ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1449                                             from + copied, &partial,
1450                                             write_end_fn);
1451                if (!partial)
1452                        SetPageUptodate(page);
1453        }
1454        size_changed = ext4_update_inode_size(inode, pos + copied);
1455        ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1456        EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1457        unlock_page(page);
1458        put_page(page);
1459
1460        if (old_size < pos)
1461                pagecache_isize_extended(inode, old_size, pos);
1462
1463        if (size_changed) {
1464                ret2 = ext4_mark_inode_dirty(handle, inode);
1465                if (!ret)
1466                        ret = ret2;
1467        }
1468
1469        if (pos + len > inode->i_size && ext4_can_truncate(inode))
1470                /* if we have allocated more blocks and copied
1471                 * less. We will have blocks allocated outside
1472                 * inode->i_size. So truncate them
1473                 */
1474                ext4_orphan_add(handle, inode);
1475
1476errout:
1477        ret2 = ext4_journal_stop(handle);
1478        if (!ret)
1479                ret = ret2;
1480        if (pos + len > inode->i_size) {
1481                ext4_truncate_failed_write(inode);
1482                /*
1483                 * If truncate failed early the inode might still be
1484                 * on the orphan list; we need to make sure the inode
1485                 * is removed from the orphan list in that case.
1486                 */
1487                if (inode->i_nlink)
1488                        ext4_orphan_del(NULL, inode);
1489        }
1490
1491        return ret ? ret : copied;
1492}
1493
1494/*
1495 * Reserve space for a single cluster
1496 */
1497static int ext4_da_reserve_space(struct inode *inode)
1498{
1499        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1500        struct ext4_inode_info *ei = EXT4_I(inode);
1501        int ret;
1502
1503        /*
1504         * We will charge metadata quota at writeout time; this saves
1505         * us from metadata over-estimation, though we may go over by
1506         * a small amount in the end.  Here we just reserve for data.
1507         */
1508        ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1509        if (ret)
1510                return ret;
1511
1512        spin_lock(&ei->i_block_reservation_lock);
1513        if (ext4_claim_free_clusters(sbi, 1, 0)) {
1514                spin_unlock(&ei->i_block_reservation_lock);
1515                dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1516                return -ENOSPC;
1517        }
1518        ei->i_reserved_data_blocks++;
1519        trace_ext4_da_reserve_space(inode);
1520        spin_unlock(&ei->i_block_reservation_lock);
1521
1522        return 0;       /* success */
1523}
1524
1525static void ext4_da_release_space(struct inode *inode, int to_free)
1526{
1527        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1528        struct ext4_inode_info *ei = EXT4_I(inode);
1529
1530        if (!to_free)
1531                return;         /* Nothing to release, exit */
1532
1533        spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1534
1535        trace_ext4_da_release_space(inode, to_free);
1536        if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1537                /*
1538                 * if there aren't enough reserved blocks, then the
1539                 * counter is messed up somewhere.  Since this
1540                 * function is called from invalidate page, it's
1541                 * harmless to return without any action.
1542                 */
1543                ext4_warning(inode->i_sb, "ext4_da_release_space: "
1544                         "ino %lu, to_free %d with only %d reserved "
1545                         "data blocks", inode->i_ino, to_free,
1546                         ei->i_reserved_data_blocks);
1547                WARN_ON(1);
1548                to_free = ei->i_reserved_data_blocks;
1549        }
1550        ei->i_reserved_data_blocks -= to_free;
1551
1552        /* update fs dirty data blocks counter */
1553        percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1554
1555        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1556
1557        dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1558}
1559
1560static void ext4_da_page_release_reservation(struct page *page,
1561                                             unsigned int offset,
1562                                             unsigned int length)
1563{
1564        int to_release = 0, contiguous_blks = 0;
1565        struct buffer_head *head, *bh;
1566        unsigned int curr_off = 0;
1567        struct inode *inode = page->mapping->host;
1568        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1569        unsigned int stop = offset + length;
1570        int num_clusters;
1571        ext4_fsblk_t lblk;
1572
1573        BUG_ON(stop > PAGE_SIZE || stop < length);
1574
1575        head = page_buffers(page);
1576        bh = head;
1577        do {
1578                unsigned int next_off = curr_off + bh->b_size;
1579
1580                if (next_off > stop)
1581                        break;
1582
1583                if ((offset <= curr_off) && (buffer_delay(bh))) {
1584                        to_release++;
1585                        contiguous_blks++;
1586                        clear_buffer_delay(bh);
1587                } else if (contiguous_blks) {
1588                        lblk = page->index <<
1589                               (PAGE_SHIFT - inode->i_blkbits);
1590                        lblk += (curr_off >> inode->i_blkbits) -
1591                                contiguous_blks;
1592                        ext4_es_remove_extent(inode, lblk, contiguous_blks);
1593                        contiguous_blks = 0;
1594                }
1595                curr_off = next_off;
1596        } while ((bh = bh->b_this_page) != head);
1597
1598        if (contiguous_blks) {
1599                lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1600                lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1601                ext4_es_remove_extent(inode, lblk, contiguous_blks);
1602        }
1603
1604        /* If we have released all the blocks belonging to a cluster, then we
1605         * need to release the reserved space for that cluster. */
1606        num_clusters = EXT4_NUM_B2C(sbi, to_release);
1607        while (num_clusters > 0) {
1608                lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1609                        ((num_clusters - 1) << sbi->s_cluster_bits);
1610                if (sbi->s_cluster_ratio == 1 ||
1611                    !ext4_find_delalloc_cluster(inode, lblk))
1612                        ext4_da_release_space(inode, 1);
1613
1614                num_clusters--;
1615        }
1616}
1617
1618/*
1619 * Delayed allocation stuff
1620 */
1621
1622struct mpage_da_data {
1623        struct inode *inode;
1624        struct writeback_control *wbc;
1625
1626        pgoff_t first_page;     /* The first page to write */
1627        pgoff_t next_page;      /* Current page to examine */
1628        pgoff_t last_page;      /* Last page to examine */
1629        /*
1630         * Extent to map - this can be after first_page because that can be
1631         * fully mapped. We somewhat abuse m_flags to store whether the extent
1632         * is delalloc or unwritten.
1633         */
1634        struct ext4_map_blocks map;
1635        struct ext4_io_submit io_submit;        /* IO submission data */
1636};
1637
1638static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1639                                       bool invalidate)
1640{
1641        int nr_pages, i;
1642        pgoff_t index, end;
1643        struct pagevec pvec;
1644        struct inode *inode = mpd->inode;
1645        struct address_space *mapping = inode->i_mapping;
1646
1647        /* This is necessary when next_page == 0. */
1648        if (mpd->first_page >= mpd->next_page)
1649                return;
1650
1651        index = mpd->first_page;
1652        end   = mpd->next_page - 1;
1653        if (invalidate) {
1654                ext4_lblk_t start, last;
1655                start = index << (PAGE_SHIFT - inode->i_blkbits);
1656                last = end << (PAGE_SHIFT - inode->i_blkbits);
1657                ext4_es_remove_extent(inode, start, last - start + 1);
1658        }
1659
1660        pagevec_init(&pvec, 0);
1661        while (index <= end) {
1662                nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1663                if (nr_pages == 0)
1664                        break;
1665                for (i = 0; i < nr_pages; i++) {
1666                        struct page *page = pvec.pages[i];
1667                        if (page->index > end)
1668                                break;
1669                        BUG_ON(!PageLocked(page));
1670                        BUG_ON(PageWriteback(page));
1671                        if (invalidate) {
1672                                if (page_mapped(page))
1673                                        clear_page_dirty_for_io(page);
1674                                block_invalidatepage(page, 0, PAGE_SIZE);
1675                                ClearPageUptodate(page);
1676                        }
1677                        unlock_page(page);
1678                }
1679                index = pvec.pages[nr_pages - 1]->index + 1;
1680                pagevec_release(&pvec);
1681        }
1682}
1683
1684static void ext4_print_free_blocks(struct inode *inode)
1685{
1686        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1687        struct super_block *sb = inode->i_sb;
1688        struct ext4_inode_info *ei = EXT4_I(inode);
1689
1690        ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1691               EXT4_C2B(EXT4_SB(inode->i_sb),
1692                        ext4_count_free_clusters(sb)));
1693        ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1694        ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1695               (long long) EXT4_C2B(EXT4_SB(sb),
1696                percpu_counter_sum(&sbi->s_freeclusters_counter)));
1697        ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1698               (long long) EXT4_C2B(EXT4_SB(sb),
1699                percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1700        ext4_msg(sb, KERN_CRIT, "Block reservation details");
1701        ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1702                 ei->i_reserved_data_blocks);
1703        return;
1704}
1705
1706static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1707{
1708        return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1709}
1710
1711/*
1712 * This function is grabs code from the very beginning of
1713 * ext4_map_blocks, but assumes that the caller is from delayed write
1714 * time. This function looks up the requested blocks and sets the
1715 * buffer delay bit under the protection of i_data_sem.
1716 */
1717static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1718                              struct ext4_map_blocks *map,
1719                              struct buffer_head *bh)
1720{
1721        struct extent_status es;
1722        int retval;
1723        sector_t invalid_block = ~((sector_t) 0xffff);
1724#ifdef ES_AGGRESSIVE_TEST
1725        struct ext4_map_blocks orig_map;
1726
1727        memcpy(&orig_map, map, sizeof(*map));
1728#endif
1729
1730        if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1731                invalid_block = ~0;
1732
1733        map->m_flags = 0;
1734        ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1735                  "logical block %lu\n", inode->i_ino, map->m_len,
1736                  (unsigned long) map->m_lblk);
1737
1738        /* Lookup extent status tree firstly */
1739        if (ext4_es_lookup_extent(inode, iblock, &es)) {
1740                if (ext4_es_is_hole(&es)) {
1741                        retval = 0;
1742                        down_read(&EXT4_I(inode)->i_data_sem);
1743                        goto add_delayed;
1744                }
1745
1746                /*
1747                 * Delayed extent could be allocated by fallocate.
1748                 * So we need to check it.
1749                 */
1750                if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1751                        map_bh(bh, inode->i_sb, invalid_block);
1752                        set_buffer_new(bh);
1753                        set_buffer_delay(bh);
1754                        return 0;
1755                }
1756
1757                map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1758                retval = es.es_len - (iblock - es.es_lblk);
1759                if (retval > map->m_len)
1760                        retval = map->m_len;
1761                map->m_len = retval;
1762                if (ext4_es_is_written(&es))
1763                        map->m_flags |= EXT4_MAP_MAPPED;
1764                else if (ext4_es_is_unwritten(&es))
1765                        map->m_flags |= EXT4_MAP_UNWRITTEN;
1766                else
1767                        BUG_ON(1);
1768
1769#ifdef ES_AGGRESSIVE_TEST
1770                ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1771#endif
1772                return retval;
1773        }
1774
1775        /*
1776         * Try to see if we can get the block without requesting a new
1777         * file system block.
1778         */
1779        down_read(&EXT4_I(inode)->i_data_sem);
1780        if (ext4_has_inline_data(inode))
1781                retval = 0;
1782        else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1783                retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1784        else
1785                retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1786
1787add_delayed:
1788        if (retval == 0) {
1789                int ret;
1790                /*
1791                 * XXX: __block_prepare_write() unmaps passed block,
1792                 * is it OK?
1793                 */
1794                /*
1795                 * If the block was allocated from previously allocated cluster,
1796                 * then we don't need to reserve it again. However we still need
1797                 * to reserve metadata for every block we're going to write.
1798                 */
1799                if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1800                    !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1801                        ret = ext4_da_reserve_space(inode);
1802                        if (ret) {
1803                                /* not enough space to reserve */
1804                                retval = ret;
1805                                goto out_unlock;
1806                        }
1807                }
1808
1809                ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1810                                            ~0, EXTENT_STATUS_DELAYED);
1811                if (ret) {
1812                        retval = ret;
1813                        goto out_unlock;
1814                }
1815
1816                map_bh(bh, inode->i_sb, invalid_block);
1817                set_buffer_new(bh);
1818                set_buffer_delay(bh);
1819        } else if (retval > 0) {
1820                int ret;
1821                unsigned int status;
1822
1823                if (unlikely(retval != map->m_len)) {
1824                        ext4_warning(inode->i_sb,
1825                                     "ES len assertion failed for inode "
1826                                     "%lu: retval %d != map->m_len %d",
1827                                     inode->i_ino, retval, map->m_len);
1828                        WARN_ON(1);
1829                }
1830
1831                status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1832                                EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1833                ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1834                                            map->m_pblk, status);
1835                if (ret != 0)
1836                        retval = ret;
1837        }
1838
1839out_unlock:
1840        up_read((&EXT4_I(inode)->i_data_sem));
1841
1842        return retval;
1843}
1844
1845/*
1846 * This is a special get_block_t callback which is used by
1847 * ext4_da_write_begin().  It will either return mapped block or
1848 * reserve space for a single block.
1849 *
1850 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1851 * We also have b_blocknr = -1 and b_bdev initialized properly
1852 *
1853 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1854 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1855 * initialized properly.
1856 */
1857int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1858                           struct buffer_head *bh, int create)
1859{
1860        struct ext4_map_blocks map;
1861        int ret = 0;
1862
1863        BUG_ON(create == 0);
1864        BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1865
1866        map.m_lblk = iblock;
1867        map.m_len = 1;
1868
1869        /*
1870         * first, we need to know whether the block is allocated already
1871         * preallocated blocks are unmapped but should treated
1872         * the same as allocated blocks.
1873         */
1874        ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1875        if (ret <= 0)
1876                return ret;
1877
1878        map_bh(bh, inode->i_sb, map.m_pblk);
1879        ext4_update_bh_state(bh, map.m_flags);
1880
1881        if (buffer_unwritten(bh)) {
1882                /* A delayed write to unwritten bh should be marked
1883                 * new and mapped.  Mapped ensures that we don't do
1884                 * get_block multiple times when we write to the same
1885                 * offset and new ensures that we do proper zero out
1886                 * for partial write.
1887                 */
1888                set_buffer_new(bh);
1889                set_buffer_mapped(bh);
1890        }
1891        return 0;
1892}
1893
1894static int bget_one(handle_t *handle, struct buffer_head *bh)
1895{
1896        get_bh(bh);
1897        return 0;
1898}
1899
1900static int bput_one(handle_t *handle, struct buffer_head *bh)
1901{
1902        put_bh(bh);
1903        return 0;
1904}
1905
1906static int __ext4_journalled_writepage(struct page *page,
1907                                       unsigned int len)
1908{
1909        struct address_space *mapping = page->mapping;
1910        struct inode *inode = mapping->host;
1911        struct buffer_head *page_bufs = NULL;
1912        handle_t *handle = NULL;
1913        int ret = 0, err = 0;
1914        int inline_data = ext4_has_inline_data(inode);
1915        struct buffer_head *inode_bh = NULL;
1916
1917        ClearPageChecked(page);
1918
1919        if (inline_data) {
1920                BUG_ON(page->index != 0);
1921                BUG_ON(len > ext4_get_max_inline_size(inode));
1922                inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1923                if (inode_bh == NULL)
1924                        goto out;
1925        } else {
1926                page_bufs = page_buffers(page);
1927                if (!page_bufs) {
1928                        BUG();
1929                        goto out;
1930                }
1931                ext4_walk_page_buffers(handle, page_bufs, 0, len,
1932                                       NULL, bget_one);
1933        }
1934        /*
1935         * We need to release the page lock before we start the
1936         * journal, so grab a reference so the page won't disappear
1937         * out from under us.
1938         */
1939        get_page(page);
1940        unlock_page(page);
1941
1942        handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1943                                    ext4_writepage_trans_blocks(inode));
1944        if (IS_ERR(handle)) {
1945                ret = PTR_ERR(handle);
1946                put_page(page);
1947                goto out_no_pagelock;
1948        }
1949        BUG_ON(!ext4_handle_valid(handle));
1950
1951        lock_page(page);
1952        put_page(page);
1953        if (page->mapping != mapping) {
1954                /* The page got truncated from under us */
1955                ext4_journal_stop(handle);
1956                ret = 0;
1957                goto out;
1958        }
1959
1960        if (inline_data) {
1961                BUFFER_TRACE(inode_bh, "get write access");
1962                ret = ext4_journal_get_write_access(handle, inode_bh);
1963
1964                err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1965
1966        } else {
1967                ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1968                                             do_journal_get_write_access);
1969
1970                err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1971                                             write_end_fn);
1972        }
1973        if (ret == 0)
1974                ret = err;
1975        EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1976        err = ext4_journal_stop(handle);
1977        if (!ret)
1978                ret = err;
1979
1980        if (!ext4_has_inline_data(inode))
1981                ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1982                                       NULL, bput_one);
1983        ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1984out:
1985        unlock_page(page);
1986out_no_pagelock:
1987        brelse(inode_bh);
1988        return ret;
1989}
1990
1991/*
1992 * Note that we don't need to start a transaction unless we're journaling data
1993 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1994 * need to file the inode to the transaction's list in ordered mode because if
1995 * we are writing back data added by write(), the inode is already there and if
1996 * we are writing back data modified via mmap(), no one guarantees in which
1997 * transaction the data will hit the disk. In case we are journaling data, we
1998 * cannot start transaction directly because transaction start ranks above page
1999 * lock so we have to do some magic.
2000 *
2001 * This function can get called via...
2002 *   - ext4_writepages after taking page lock (have journal handle)
2003 *   - journal_submit_inode_data_buffers (no journal handle)
2004 *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2005 *   - grab_page_cache when doing write_begin (have journal handle)
2006 *
2007 * We don't do any block allocation in this function. If we have page with
2008 * multiple blocks we need to write those buffer_heads that are mapped. This
2009 * is important for mmaped based write. So if we do with blocksize 1K
2010 * truncate(f, 1024);
2011 * a = mmap(f, 0, 4096);
2012 * a[0] = 'a';
2013 * truncate(f, 4096);
2014 * we have in the page first buffer_head mapped via page_mkwrite call back
2015 * but other buffer_heads would be unmapped but dirty (dirty done via the
2016 * do_wp_page). So writepage should write the first block. If we modify
2017 * the mmap area beyond 1024 we will again get a page_fault and the
2018 * page_mkwrite callback will do the block allocation and mark the
2019 * buffer_heads mapped.
2020 *
2021 * We redirty the page if we have any buffer_heads that is either delay or
2022 * unwritten in the page.
2023 *
2024 * We can get recursively called as show below.
2025 *
2026 *      ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2027 *              ext4_writepage()
2028 *
2029 * But since we don't do any block allocation we should not deadlock.
2030 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2031 */
2032static int ext4_writepage(struct page *page,
2033                          struct writeback_control *wbc)
2034{
2035        int ret = 0;
2036        loff_t size;
2037        unsigned int len;
2038        struct buffer_head *page_bufs = NULL;
2039        struct inode *inode = page->mapping->host;
2040        struct ext4_io_submit io_submit;
2041        bool keep_towrite = false;
2042
2043        trace_ext4_writepage(page);
2044        size = i_size_read(inode);
2045        if (page->index == size >> PAGE_SHIFT)
2046                len = size & ~PAGE_MASK;
2047        else
2048                len = PAGE_SIZE;
2049
2050        page_bufs = page_buffers(page);
2051        /*
2052         * We cannot do block allocation or other extent handling in this
2053         * function. If there are buffers needing that, we have to redirty
2054         * the page. But we may reach here when we do a journal commit via
2055         * journal_submit_inode_data_buffers() and in that case we must write
2056         * allocated buffers to achieve data=ordered mode guarantees.
2057         *
2058         * Also, if there is only one buffer per page (the fs block
2059         * size == the page size), if one buffer needs block
2060         * allocation or needs to modify the extent tree to clear the
2061         * unwritten flag, we know that the page can't be written at
2062         * all, so we might as well refuse the write immediately.
2063         * Unfortunately if the block size != page size, we can't as
2064         * easily detect this case using ext4_walk_page_buffers(), but
2065         * for the extremely common case, this is an optimization that
2066         * skips a useless round trip through ext4_bio_write_page().
2067         */
2068        if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2069                                   ext4_bh_delay_or_unwritten)) {
2070                redirty_page_for_writepage(wbc, page);
2071                if ((current->flags & PF_MEMALLOC) ||
2072                    (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2073                        /*
2074                         * For memory cleaning there's no point in writing only
2075                         * some buffers. So just bail out. Warn if we came here
2076                         * from direct reclaim.
2077                         */
2078                        WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2079                                                        == PF_MEMALLOC);
2080                        unlock_page(page);
2081                        return 0;
2082                }
2083                keep_towrite = true;
2084        }
2085
2086        if (PageChecked(page) && ext4_should_journal_data(inode))
2087                /*
2088                 * It's mmapped pagecache.  Add buffers and journal it.  There
2089                 * doesn't seem much point in redirtying the page here.
2090                 */
2091                return __ext4_journalled_writepage(page, len);
2092
2093        ext4_io_submit_init(&io_submit, wbc);
2094        io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2095        if (!io_submit.io_end) {
2096                redirty_page_for_writepage(wbc, page);
2097                unlock_page(page);
2098                return -ENOMEM;
2099        }
2100        ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2101        ext4_io_submit(&io_submit);
2102        /* Drop io_end reference we got from init */
2103        ext4_put_io_end_defer(io_submit.io_end);
2104        return ret;
2105}
2106
2107static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2108{
2109        int len;
2110        loff_t size = i_size_read(mpd->inode);
2111        int err;
2112
2113        BUG_ON(page->index != mpd->first_page);
2114        if (page->index == size >> PAGE_SHIFT)
2115                len = size & ~PAGE_MASK;
2116        else
2117                len = PAGE_SIZE;
2118        clear_page_dirty_for_io(page);
2119        err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2120        if (!err)
2121                mpd->wbc->nr_to_write--;
2122        mpd->first_page++;
2123
2124        return err;
2125}
2126
2127#define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2128
2129/*
2130 * mballoc gives us at most this number of blocks...
2131 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2132 * The rest of mballoc seems to handle chunks up to full group size.
2133 */
2134#define MAX_WRITEPAGES_EXTENT_LEN 2048
2135
2136/*
2137 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2138 *
2139 * @mpd - extent of blocks
2140 * @lblk - logical number of the block in the file
2141 * @bh - buffer head we want to add to the extent
2142 *
2143 * The function is used to collect contig. blocks in the same state. If the
2144 * buffer doesn't require mapping for writeback and we haven't started the
2145 * extent of buffers to map yet, the function returns 'true' immediately - the
2146 * caller can write the buffer right away. Otherwise the function returns true
2147 * if the block has been added to the extent, false if the block couldn't be
2148 * added.
2149 */
2150static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2151                                   struct buffer_head *bh)
2152{
2153        struct ext4_map_blocks *map = &mpd->map;
2154
2155        /* Buffer that doesn't need mapping for writeback? */
2156        if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2157            (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2158                /* So far no extent to map => we write the buffer right away */
2159                if (map->m_len == 0)
2160                        return true;
2161                return false;
2162        }
2163
2164        /* First block in the extent? */
2165        if (map->m_len == 0) {
2166                map->m_lblk = lblk;
2167                map->m_len = 1;
2168                map->m_flags = bh->b_state & BH_FLAGS;
2169                return true;
2170        }
2171
2172        /* Don't go larger than mballoc is willing to allocate */
2173        if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2174                return false;
2175
2176        /* Can we merge the block to our big extent? */
2177        if (lblk == map->m_lblk + map->m_len &&
2178            (bh->b_state & BH_FLAGS) == map->m_flags) {
2179                map->m_len++;
2180                return true;
2181        }
2182        return false;
2183}
2184
2185/*
2186 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2187 *
2188 * @mpd - extent of blocks for mapping
2189 * @head - the first buffer in the page
2190 * @bh - buffer we should start processing from
2191 * @lblk - logical number of the block in the file corresponding to @bh
2192 *
2193 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2194 * the page for IO if all buffers in this page were mapped and there's no
2195 * accumulated extent of buffers to map or add buffers in the page to the
2196 * extent of buffers to map. The function returns 1 if the caller can continue
2197 * by processing the next page, 0 if it should stop adding buffers to the
2198 * extent to map because we cannot extend it anymore. It can also return value
2199 * < 0 in case of error during IO submission.
2200 */
2201static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2202                                   struct buffer_head *head,
2203                                   struct buffer_head *bh,
2204                                   ext4_lblk_t lblk)
2205{
2206        struct inode *inode = mpd->inode;
2207        int err;
2208        ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2209                                                        >> inode->i_blkbits;
2210
2211        do {
2212                BUG_ON(buffer_locked(bh));
2213
2214                if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2215                        /* Found extent to map? */
2216                        if (mpd->map.m_len)
2217                                return 0;
2218                        /* Everything mapped so far and we hit EOF */
2219                        break;
2220                }
2221        } while (lblk++, (bh = bh->b_this_page) != head);
2222        /* So far everything mapped? Submit the page for IO. */
2223        if (mpd->map.m_len == 0) {
2224                err = mpage_submit_page(mpd, head->b_page);
2225                if (err < 0)
2226                        return err;
2227        }
2228        return lblk < blocks;
2229}
2230
2231/*
2232 * mpage_map_buffers - update buffers corresponding to changed extent and
2233 *                     submit fully mapped pages for IO
2234 *
2235 * @mpd - description of extent to map, on return next extent to map
2236 *
2237 * Scan buffers corresponding to changed extent (we expect corresponding pages
2238 * to be already locked) and update buffer state according to new extent state.
2239 * We map delalloc buffers to their physical location, clear unwritten bits,
2240 * and mark buffers as uninit when we perform writes to unwritten extents
2241 * and do extent conversion after IO is finished. If the last page is not fully
2242 * mapped, we update @map to the next extent in the last page that needs
2243 * mapping. Otherwise we submit the page for IO.
2244 */
2245static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2246{
2247        struct pagevec pvec;
2248        int nr_pages, i;
2249        struct inode *inode = mpd->inode;
2250        struct buffer_head *head, *bh;
2251        int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2252        pgoff_t start, end;
2253        ext4_lblk_t lblk;
2254        sector_t pblock;
2255        int err;
2256
2257        start = mpd->map.m_lblk >> bpp_bits;
2258        end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2259        lblk = start << bpp_bits;
2260        pblock = mpd->map.m_pblk;
2261
2262        pagevec_init(&pvec, 0);
2263        while (start <= end) {
2264                nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2265                                          PAGEVEC_SIZE);
2266                if (nr_pages == 0)
2267                        break;
2268                for (i = 0; i < nr_pages; i++) {
2269                        struct page *page = pvec.pages[i];
2270
2271                        if (page->index > end)
2272                                break;
2273                        /* Up to 'end' pages must be contiguous */
2274                        BUG_ON(page->index != start);
2275                        bh = head = page_buffers(page);
2276                        do {
2277                                if (lblk < mpd->map.m_lblk)
2278                                        continue;
2279                                if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2280                                        /*
2281                                         * Buffer after end of mapped extent.
2282                                         * Find next buffer in the page to map.
2283                                         */
2284                                        mpd->map.m_len = 0;
2285                                        mpd->map.m_flags = 0;
2286                                        /*
2287                                         * FIXME: If dioread_nolock supports
2288                                         * blocksize < pagesize, we need to make
2289                                         * sure we add size mapped so far to
2290                                         * io_end->size as the following call
2291                                         * can submit the page for IO.
2292                                         */
2293                                        err = mpage_process_page_bufs(mpd, head,
2294                                                                      bh, lblk);
2295                                        pagevec_release(&pvec);
2296                                        if (err > 0)
2297                                                err = 0;
2298                                        return err;
2299                                }
2300                                if (buffer_delay(bh)) {
2301                                        clear_buffer_delay(bh);
2302                                        bh->b_blocknr = pblock++;
2303                                }
2304                                clear_buffer_unwritten(bh);
2305                        } while (lblk++, (bh = bh->b_this_page) != head);
2306
2307                        /*
2308                         * FIXME: This is going to break if dioread_nolock
2309                         * supports blocksize < pagesize as we will try to
2310                         * convert potentially unmapped parts of inode.
2311                         */
2312                        mpd->io_submit.io_end->size += PAGE_SIZE;
2313                        /* Page fully mapped - let IO run! */
2314                        err = mpage_submit_page(mpd, page);
2315                        if (err < 0) {
2316                                pagevec_release(&pvec);
2317                                return err;
2318                        }
2319                        start++;
2320                }
2321                pagevec_release(&pvec);
2322        }
2323        /* Extent fully mapped and matches with page boundary. We are done. */
2324        mpd->map.m_len = 0;
2325        mpd->map.m_flags = 0;
2326        return 0;
2327}
2328
2329static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2330{
2331        struct inode *inode = mpd->inode;
2332        struct ext4_map_blocks *map = &mpd->map;
2333        int get_blocks_flags;
2334        int err, dioread_nolock;
2335
2336        trace_ext4_da_write_pages_extent(inode, map);
2337        /*
2338         * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2339         * to convert an unwritten extent to be initialized (in the case
2340         * where we have written into one or more preallocated blocks).  It is
2341         * possible that we're going to need more metadata blocks than
2342         * previously reserved. However we must not fail because we're in
2343         * writeback and there is nothing we can do about it so it might result
2344         * in data loss.  So use reserved blocks to allocate metadata if
2345         * possible.
2346         *
2347         * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2348         * the blocks in question are delalloc blocks.  This indicates
2349         * that the blocks and quotas has already been checked when
2350         * the data was copied into the page cache.
2351         */
2352        get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2353                           EXT4_GET_BLOCKS_METADATA_NOFAIL |
2354                           EXT4_GET_BLOCKS_IO_SUBMIT;
2355        dioread_nolock = ext4_should_dioread_nolock(inode);
2356        if (dioread_nolock)
2357                get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2358        if (map->m_flags & (1 << BH_Delay))
2359                get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2360
2361        err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2362        if (err < 0)
2363                return err;
2364        if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2365                if (!mpd->io_submit.io_end->handle &&
2366                    ext4_handle_valid(handle)) {
2367                        mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2368                        handle->h_rsv_handle = NULL;
2369                }
2370                ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2371        }
2372
2373        BUG_ON(map->m_len == 0);
2374        if (map->m_flags & EXT4_MAP_NEW) {
2375                struct block_device *bdev = inode->i_sb->s_bdev;
2376                int i;
2377
2378                for (i = 0; i < map->m_len; i++)
2379                        unmap_underlying_metadata(bdev, map->m_pblk + i);
2380        }
2381        return 0;
2382}
2383
2384/*
2385 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2386 *                               mpd->len and submit pages underlying it for IO
2387 *
2388 * @handle - handle for journal operations
2389 * @mpd - extent to map
2390 * @give_up_on_write - we set this to true iff there is a fatal error and there
2391 *                     is no hope of writing the data. The caller should discard
2392 *                     dirty pages to avoid infinite loops.
2393 *
2394 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2395 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2396 * them to initialized or split the described range from larger unwritten
2397 * extent. Note that we need not map all the described range since allocation
2398 * can return less blocks or the range is covered by more unwritten extents. We
2399 * cannot map more because we are limited by reserved transaction credits. On
2400 * the other hand we always make sure that the last touched page is fully
2401 * mapped so that it can be written out (and thus forward progress is
2402 * guaranteed). After mapping we submit all mapped pages for IO.
2403 */
2404static int mpage_map_and_submit_extent(handle_t *handle,
2405                                       struct mpage_da_data *mpd,
2406                                       bool *give_up_on_write)
2407{
2408        struct inode *inode = mpd->inode;
2409        struct ext4_map_blocks *map = &mpd->map;
2410        int err;
2411        loff_t disksize;
2412        int progress = 0;
2413
2414        mpd->io_submit.io_end->offset =
2415                                ((loff_t)map->m_lblk) << inode->i_blkbits;
2416        do {
2417                err = mpage_map_one_extent(handle, mpd);
2418                if (err < 0) {
2419                        struct super_block *sb = inode->i_sb;
2420
2421                        if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2422                                goto invalidate_dirty_pages;
2423                        /*
2424                         * Let the uper layers retry transient errors.
2425                         * In the case of ENOSPC, if ext4_count_free_blocks()
2426                         * is non-zero, a commit should free up blocks.
2427                         */
2428                        if ((err == -ENOMEM) ||
2429                            (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2430                                if (progress)
2431                                        goto update_disksize;
2432                                return err;
2433                        }
2434                        ext4_msg(sb, KERN_CRIT,
2435                                 "Delayed block allocation failed for "
2436                                 "inode %lu at logical offset %llu with"
2437                                 " max blocks %u with error %d",
2438                                 inode->i_ino,
2439                                 (unsigned long long)map->m_lblk,
2440                                 (unsigned)map->m_len, -err);
2441                        ext4_msg(sb, KERN_CRIT,
2442                                 "This should not happen!! Data will "
2443                                 "be lost\n");
2444                        if (err == -ENOSPC)
2445                                ext4_print_free_blocks(inode);
2446                invalidate_dirty_pages:
2447                        *give_up_on_write = true;
2448                        return err;
2449                }
2450                progress = 1;
2451                /*
2452                 * Update buffer state, submit mapped pages, and get us new
2453                 * extent to map
2454                 */
2455                err = mpage_map_and_submit_buffers(mpd);
2456                if (err < 0)
2457                        goto update_disksize;
2458        } while (map->m_len);
2459
2460update_disksize:
2461        /*
2462         * Update on-disk size after IO is submitted.  Races with
2463         * truncate are avoided by checking i_size under i_data_sem.
2464         */
2465        disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2466        if (disksize > EXT4_I(inode)->i_disksize) {
2467                int err2;
2468                loff_t i_size;
2469
2470                down_write(&EXT4_I(inode)->i_data_sem);
2471                i_size = i_size_read(inode);
2472                if (disksize > i_size)
2473                        disksize = i_size;
2474                if (disksize > EXT4_I(inode)->i_disksize)
2475                        EXT4_I(inode)->i_disksize = disksize;
2476                err2 = ext4_mark_inode_dirty(handle, inode);
2477                up_write(&EXT4_I(inode)->i_data_sem);
2478                if (err2)
2479                        ext4_error(inode->i_sb,
2480                                   "Failed to mark inode %lu dirty",
2481                                   inode->i_ino);
2482                if (!err)
2483                        err = err2;
2484        }
2485        return err;
2486}
2487
2488/*
2489 * Calculate the total number of credits to reserve for one writepages
2490 * iteration. This is called from ext4_writepages(). We map an extent of
2491 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2492 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2493 * bpp - 1 blocks in bpp different extents.
2494 */
2495static int ext4_da_writepages_trans_blocks(struct inode *inode)
2496{
2497        int bpp = ext4_journal_blocks_per_page(inode);
2498
2499        return ext4_meta_trans_blocks(inode,
2500                                MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2501}
2502
2503/*
2504 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2505 *                               and underlying extent to map
2506 *
2507 * @mpd - where to look for pages
2508 *
2509 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2510 * IO immediately. When we find a page which isn't mapped we start accumulating
2511 * extent of buffers underlying these pages that needs mapping (formed by
2512 * either delayed or unwritten buffers). We also lock the pages containing
2513 * these buffers. The extent found is returned in @mpd structure (starting at
2514 * mpd->lblk with length mpd->len blocks).
2515 *
2516 * Note that this function can attach bios to one io_end structure which are
2517 * neither logically nor physically contiguous. Although it may seem as an
2518 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2519 * case as we need to track IO to all buffers underlying a page in one io_end.
2520 */
2521static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2522{
2523        struct address_space *mapping = mpd->inode->i_mapping;
2524        struct pagevec pvec;
2525        unsigned int nr_pages;
2526        long left = mpd->wbc->nr_to_write;
2527        pgoff_t index = mpd->first_page;
2528        pgoff_t end = mpd->last_page;
2529        int tag;
2530        int i, err = 0;
2531        int blkbits = mpd->inode->i_blkbits;
2532        ext4_lblk_t lblk;
2533        struct buffer_head *head;
2534
2535        if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2536                tag = PAGECACHE_TAG_TOWRITE;
2537        else
2538                tag = PAGECACHE_TAG_DIRTY;
2539
2540        pagevec_init(&pvec, 0);
2541        mpd->map.m_len = 0;
2542        mpd->next_page = index;
2543        while (index <= end) {
2544                nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2545                              min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2546                if (nr_pages == 0)
2547                        goto out;
2548
2549                for (i = 0; i < nr_pages; i++) {
2550                        struct page *page = pvec.pages[i];
2551
2552                        /*
2553                         * At this point, the page may be truncated or
2554                         * invalidated (changing page->mapping to NULL), or
2555                         * even swizzled back from swapper_space to tmpfs file
2556                         * mapping. However, page->index will not change
2557                         * because we have a reference on the page.
2558                         */
2559                        if (page->index > end)
2560                                goto out;
2561
2562                        /*
2563                         * Accumulated enough dirty pages? This doesn't apply
2564                         * to WB_SYNC_ALL mode. For integrity sync we have to
2565                         * keep going because someone may be concurrently
2566                         * dirtying pages, and we might have synced a lot of
2567                         * newly appeared dirty pages, but have not synced all
2568                         * of the old dirty pages.
2569                         */
2570                        if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2571                                goto out;
2572
2573                        /* If we can't merge this page, we are done. */
2574                        if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2575                                goto out;
2576
2577                        lock_page(page);
2578                        /*
2579                         * If the page is no longer dirty, or its mapping no
2580                         * longer corresponds to inode we are writing (which
2581                         * means it has been truncated or invalidated), or the
2582                         * page is already under writeback and we are not doing
2583                         * a data integrity writeback, skip the page
2584                         */
2585                        if (!PageDirty(page) ||
2586                            (PageWriteback(page) &&
2587                             (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2588                            unlikely(page->mapping != mapping)) {
2589                                unlock_page(page);
2590                                continue;
2591                        }
2592
2593                        wait_on_page_writeback(page);
2594                        BUG_ON(PageWriteback(page));
2595
2596                        if (mpd->map.m_len == 0)
2597                                mpd->first_page = page->index;
2598                        mpd->next_page = page->index + 1;
2599                        /* Add all dirty buffers to mpd */
2600                        lblk = ((ext4_lblk_t)page->index) <<
2601                                (PAGE_SHIFT - blkbits);
2602                        head = page_buffers(page);
2603                        err = mpage_process_page_bufs(mpd, head, head, lblk);
2604                        if (err <= 0)
2605                                goto out;
2606                        err = 0;
2607                        left--;
2608                }
2609                pagevec_release(&pvec);
2610                cond_resched();
2611        }
2612        return 0;
2613out:
2614        pagevec_release(&pvec);
2615        return err;
2616}
2617
2618static int __writepage(struct page *page, struct writeback_control *wbc,
2619                       void *data)
2620{
2621        struct address_space *mapping = data;
2622        int ret = ext4_writepage(page, wbc);
2623        mapping_set_error(mapping, ret);
2624        return ret;
2625}
2626
2627static int ext4_writepages(struct address_space *mapping,
2628                           struct writeback_control *wbc)
2629{
2630        pgoff_t writeback_index = 0;
2631        long nr_to_write = wbc->nr_to_write;
2632        int range_whole = 0;
2633        int cycled = 1;
2634        handle_t *handle = NULL;
2635        struct mpage_da_data mpd;
2636        struct inode *inode = mapping->host;
2637        int needed_blocks, rsv_blocks = 0, ret = 0;
2638        struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2639        bool done;
2640        struct blk_plug plug;
2641        bool give_up_on_write = false;
2642
2643        percpu_down_read(&sbi->s_journal_flag_rwsem);
2644        trace_ext4_writepages(inode, wbc);
2645
2646        if (dax_mapping(mapping)) {
2647                ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2648                                                  wbc);
2649                goto out_writepages;
2650        }
2651
2652        /*
2653         * No pages to write? This is mainly a kludge to avoid starting
2654         * a transaction for special inodes like journal inode on last iput()
2655         * because that could violate lock ordering on umount
2656         */
2657        if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2658                goto out_writepages;
2659
2660        if (ext4_should_journal_data(inode)) {
2661                struct blk_plug plug;
2662
2663                blk_start_plug(&plug);
2664                ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2665                blk_finish_plug(&plug);
2666                goto out_writepages;
2667        }
2668
2669        /*
2670         * If the filesystem has aborted, it is read-only, so return
2671         * right away instead of dumping stack traces later on that
2672         * will obscure the real source of the problem.  We test
2673         * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2674         * the latter could be true if the filesystem is mounted
2675         * read-only, and in that case, ext4_writepages should
2676         * *never* be called, so if that ever happens, we would want
2677         * the stack trace.
2678         */
2679        if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2680                ret = -EROFS;
2681                goto out_writepages;
2682        }
2683
2684        if (ext4_should_dioread_nolock(inode)) {
2685                /*
2686                 * We may need to convert up to one extent per block in
2687                 * the page and we may dirty the inode.
2688                 */
2689                rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2690        }
2691
2692        /*
2693         * If we have inline data and arrive here, it means that
2694         * we will soon create the block for the 1st page, so
2695         * we'd better clear the inline data here.
2696         */
2697        if (ext4_has_inline_data(inode)) {
2698                /* Just inode will be modified... */
2699                handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2700                if (IS_ERR(handle)) {
2701                        ret = PTR_ERR(handle);
2702                        goto out_writepages;
2703                }
2704                BUG_ON(ext4_test_inode_state(inode,
2705                                EXT4_STATE_MAY_INLINE_DATA));
2706                ext4_destroy_inline_data(handle, inode);
2707                ext4_journal_stop(handle);
2708        }
2709
2710        if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2711                range_whole = 1;
2712
2713        if (wbc->range_cyclic) {
2714                writeback_index = mapping->writeback_index;
2715                if (writeback_index)
2716                        cycled = 0;
2717                mpd.first_page = writeback_index;
2718                mpd.last_page = -1;
2719        } else {
2720                mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2721                mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2722        }
2723
2724        mpd.inode = inode;
2725        mpd.wbc = wbc;
2726        ext4_io_submit_init(&mpd.io_submit, wbc);
2727retry:
2728        if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2729                tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2730        done = false;
2731        blk_start_plug(&plug);
2732        while (!done && mpd.first_page <= mpd.last_page) {
2733                /* For each extent of pages we use new io_end */
2734                mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2735                if (!mpd.io_submit.io_end) {
2736                        ret = -ENOMEM;
2737                        break;
2738                }
2739
2740                /*
2741                 * We have two constraints: We find one extent to map and we
2742                 * must always write out whole page (makes a difference when
2743                 * blocksize < pagesize) so that we don't block on IO when we
2744                 * try to write out the rest of the page. Journalled mode is
2745                 * not supported by delalloc.
2746                 */
2747                BUG_ON(ext4_should_journal_data(inode));
2748                needed_blocks = ext4_da_writepages_trans_blocks(inode);
2749
2750                /* start a new transaction */
2751                handle = ext4_journal_start_with_reserve(inode,
2752                                EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2753                if (IS_ERR(handle)) {
2754                        ret = PTR_ERR(handle);
2755                        ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2756                               "%ld pages, ino %lu; err %d", __func__,
2757                                wbc->nr_to_write, inode->i_ino, ret);
2758                        /* Release allocated io_end */
2759                        ext4_put_io_end(mpd.io_submit.io_end);
2760                        break;
2761                }
2762
2763                trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2764                ret = mpage_prepare_extent_to_map(&mpd);
2765                if (!ret) {
2766                        if (mpd.map.m_len)
2767                                ret = mpage_map_and_submit_extent(handle, &mpd,
2768                                        &give_up_on_write);
2769                        else {
2770                                /*
2771                                 * We scanned the whole range (or exhausted
2772                                 * nr_to_write), submitted what was mapped and
2773                                 * didn't find anything needing mapping. We are
2774                                 * done.
2775                                 */
2776                                done = true;
2777                        }
2778                }
2779                /*
2780                 * Caution: If the handle is synchronous,
2781                 * ext4_journal_stop() can wait for transaction commit
2782                 * to finish which may depend on writeback of pages to
2783                 * complete or on page lock to be released.  In that
2784                 * case, we have to wait until after after we have
2785                 * submitted all the IO, released page locks we hold,
2786                 * and dropped io_end reference (for extent conversion
2787                 * to be able to complete) before stopping the handle.
2788                 */
2789                if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2790                        ext4_journal_stop(handle);
2791                        handle = NULL;
2792                }
2793                /* Submit prepared bio */
2794                ext4_io_submit(&mpd.io_submit);
2795                /* Unlock pages we didn't use */
2796                mpage_release_unused_pages(&mpd, give_up_on_write);
2797                /*
2798                 * Drop our io_end reference we got from init. We have
2799                 * to be careful and use deferred io_end finishing if
2800                 * we are still holding the transaction as we can
2801                 * release the last reference to io_end which may end
2802                 * up doing unwritten extent conversion.
2803                 */
2804                if (handle) {
2805                        ext4_put_io_end_defer(mpd.io_submit.io_end);
2806                        ext4_journal_stop(handle);
2807                } else
2808                        ext4_put_io_end(mpd.io_submit.io_end);
2809
2810                if (ret == -ENOSPC && sbi->s_journal) {
2811                        /*
2812                         * Commit the transaction which would
2813                         * free blocks released in the transaction
2814                         * and try again
2815                         */
2816                        jbd2_journal_force_commit_nested(sbi->s_journal);
2817                        ret = 0;
2818                        continue;
2819                }
2820                /* Fatal error - ENOMEM, EIO... */
2821                if (ret)
2822                        break;
2823        }
2824        blk_finish_plug(&plug);
2825        if (!ret && !cycled && wbc->nr_to_write > 0) {
2826                cycled = 1;
2827                mpd.last_page = writeback_index - 1;
2828                mpd.first_page = 0;
2829                goto retry;
2830        }
2831
2832        /* Update index */
2833        if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2834                /*
2835                 * Set the writeback_index so that range_cyclic
2836                 * mode will write it back later
2837                 */
2838                mapping->writeback_index = mpd.first_page;
2839
2840out_writepages:
2841        trace_ext4_writepages_result(inode, wbc, ret,
2842                                     nr_to_write - wbc->nr_to_write);
2843        percpu_up_read(&sbi->s_journal_flag_rwsem);
2844        return ret;
2845}
2846
2847static int ext4_nonda_switch(struct super_block *sb)
2848{
2849        s64 free_clusters, dirty_clusters;
2850        struct ext4_sb_info *sbi = EXT4_SB(sb);
2851
2852        /*
2853         * switch to non delalloc mode if we are running low
2854         * on free block. The free block accounting via percpu
2855         * counters can get slightly wrong with percpu_counter_batch getting
2856         * accumulated on each CPU without updating global counters
2857         * Delalloc need an accurate free block accounting. So switch
2858         * to non delalloc when we are near to error range.
2859         */
2860        free_clusters =
2861                percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2862        dirty_clusters =
2863                percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2864        /*
2865         * Start pushing delalloc when 1/2 of free blocks are dirty.
2866         */
2867        if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2868                try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2869
2870        if (2 * free_clusters < 3 * dirty_clusters ||
2871            free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2872                /*
2873                 * free block count is less than 150% of dirty blocks
2874                 * or free blocks is less than watermark
2875                 */
2876                return 1;
2877        }
2878        return 0;
2879}
2880
2881/* We always reserve for an inode update; the superblock could be there too */
2882static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2883{
2884        if (likely(ext4_has_feature_large_file(inode->i_sb)))
2885                return 1;
2886
2887        if (pos + len <= 0x7fffffffULL)
2888                return 1;
2889
2890        /* We might need to update the superblock to set LARGE_FILE */
2891        return 2;
2892}
2893
2894static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2895                               loff_t pos, unsigned len, unsigned flags,
2896                               struct page **pagep, void **fsdata)
2897{
2898        int ret, retries = 0;
2899        struct page *page;
2900        pgoff_t index;
2901        struct inode *inode = mapping->host;
2902        handle_t *handle;
2903
2904        index = pos >> PAGE_SHIFT;
2905
2906        if (ext4_nonda_switch(inode->i_sb)) {
2907                *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2908                return ext4_write_begin(file, mapping, pos,
2909                                        len, flags, pagep, fsdata);
2910        }
2911        *fsdata = (void *)0;
2912        trace_ext4_da_write_begin(inode, pos, len, flags);
2913
2914        if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2915                ret = ext4_da_write_inline_data_begin(mapping, inode,
2916                                                      pos, len, flags,
2917                                                      pagep, fsdata);
2918                if (ret < 0)
2919                        return ret;
2920                if (ret == 1)
2921                        return 0;
2922        }
2923
2924        /*
2925         * grab_cache_page_write_begin() can take a long time if the
2926         * system is thrashing due to memory pressure, or if the page
2927         * is being written back.  So grab it first before we start
2928         * the transaction handle.  This also allows us to allocate
2929         * the page (if needed) without using GFP_NOFS.
2930         */
2931retry_grab:
2932        page = grab_cache_page_write_begin(mapping, index, flags);
2933        if (!page)
2934                return -ENOMEM;
2935        unlock_page(page);
2936
2937        /*
2938         * With delayed allocation, we don't log the i_disksize update
2939         * if there is delayed block allocation. But we still need
2940         * to journalling the i_disksize update if writes to the end
2941         * of file which has an already mapped buffer.
2942         */
2943retry_journal:
2944        handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2945                                ext4_da_write_credits(inode, pos, len));
2946        if (IS_ERR(handle)) {
2947                put_page(page);
2948                return PTR_ERR(handle);
2949        }
2950
2951        lock_page(page);
2952        if (page->mapping != mapping) {
2953                /* The page got truncated from under us */
2954                unlock_page(page);
2955                put_page(page);
2956                ext4_journal_stop(handle);
2957                goto retry_grab;
2958        }
2959        /* In case writeback began while the page was unlocked */
2960        wait_for_stable_page(page);
2961
2962#ifdef CONFIG_EXT4_FS_ENCRYPTION
2963        ret = ext4_block_write_begin(page, pos, len,
2964                                     ext4_da_get_block_prep);
2965#else
2966        ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2967#endif
2968        if (ret < 0) {
2969                unlock_page(page);
2970                ext4_journal_stop(handle);
2971                /*
2972                 * block_write_begin may have instantiated a few blocks
2973                 * outside i_size.  Trim these off again. Don't need
2974                 * i_size_read because we hold i_mutex.
2975                 */
2976                if (pos + len > inode->i_size)
2977                        ext4_truncate_failed_write(inode);
2978
2979                if (ret == -ENOSPC &&
2980                    ext4_should_retry_alloc(inode->i_sb, &retries))
2981                        goto retry_journal;
2982
2983                put_page(page);
2984                return ret;
2985        }
2986
2987        *pagep = page;
2988        return ret;
2989}
2990
2991/*
2992 * Check if we should update i_disksize
2993 * when write to the end of file but not require block allocation
2994 */
2995static int ext4_da_should_update_i_disksize(struct page *page,
2996                                            unsigned long offset)
2997{
2998        struct buffer_head *bh;
2999        struct inode *inode = page->mapping->host;
3000        unsigned int idx;
3001        int i;
3002
3003        bh = page_buffers(page);
3004        idx = offset >> inode->i_blkbits;
3005
3006        for (i = 0; i < idx; i++)
3007                bh = bh->b_this_page;
3008
3009        if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3010                return 0;
3011        return 1;
3012}
3013
3014static int ext4_da_write_end(struct file *file,
3015                             struct address_space *mapping,
3016                             loff_t pos, unsigned len, unsigned copied,
3017                             struct page *page, void *fsdata)
3018{
3019        struct inode *inode = mapping->host;
3020        int ret = 0, ret2;
3021        handle_t *handle = ext4_journal_current_handle();
3022        loff_t new_i_size;
3023        unsigned long start, end;
3024        int write_mode = (int)(unsigned long)fsdata;
3025
3026        if (write_mode == FALL_BACK_TO_NONDELALLOC)
3027                return ext4_write_end(file, mapping, pos,
3028                                      len, copied, page, fsdata);
3029
3030        trace_ext4_da_write_end(inode, pos, len, copied);
3031        start = pos & (PAGE_SIZE - 1);
3032        end = start + copied - 1;
3033
3034        /*
3035         * generic_write_end() will run mark_inode_dirty() if i_size
3036         * changes.  So let's piggyback the i_disksize mark_inode_dirty
3037         * into that.
3038         */
3039        new_i_size = pos + copied;
3040        if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3041                if (ext4_has_inline_data(inode) ||
3042                    ext4_da_should_update_i_disksize(page, end)) {
3043                        ext4_update_i_disksize(inode, new_i_size);
3044                        /* We need to mark inode dirty even if
3045                         * new_i_size is less that inode->i_size
3046                         * bu greater than i_disksize.(hint delalloc)
3047                         */
3048                        ext4_mark_inode_dirty(handle, inode);
3049                }
3050        }
3051
3052        if (write_mode != CONVERT_INLINE_DATA &&
3053            ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3054            ext4_has_inline_data(inode))
3055                ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3056                                                     page);
3057        else
3058                ret2 = generic_write_end(file, mapping, pos, len, copied,
3059                                                        page, fsdata);
3060
3061        copied = ret2;
3062        if (ret2 < 0)
3063                ret = ret2;
3064        ret2 = ext4_journal_stop(handle);
3065        if (!ret)
3066                ret = ret2;
3067
3068        return ret ? ret : copied;
3069}
3070
3071static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3072                                   unsigned int length)
3073{
3074        /*
3075         * Drop reserved blocks
3076         */
3077        BUG_ON(!PageLocked(page));
3078        if (!page_has_buffers(page))
3079                goto out;
3080
3081        ext4_da_page_release_reservation(page, offset, length);
3082
3083out:
3084        ext4_invalidatepage(page, offset, length);
3085
3086        return;
3087}
3088
3089/*
3090 * Force all delayed allocation blocks to be allocated for a given inode.
3091 */
3092int ext4_alloc_da_blocks(struct inode *inode)
3093{
3094        trace_ext4_alloc_da_blocks(inode);
3095
3096        if (!EXT4_I(inode)->i_reserved_data_blocks)
3097                return 0;
3098
3099        /*
3100         * We do something simple for now.  The filemap_flush() will
3101         * also start triggering a write of the data blocks, which is
3102         * not strictly speaking necessary (and for users of
3103         * laptop_mode, not even desirable).  However, to do otherwise
3104         * would require replicating code paths in:
3105         *
3106         * ext4_writepages() ->
3107         *    write_cache_pages() ---> (via passed in callback function)
3108         *        __mpage_da_writepage() -->
3109         *           mpage_add_bh_to_extent()
3110         *           mpage_da_map_blocks()
3111         *
3112         * The problem is that write_cache_pages(), located in
3113         * mm/page-writeback.c, marks pages clean in preparation for
3114         * doing I/O, which is not desirable if we're not planning on
3115         * doing I/O at all.
3116         *
3117         * We could call write_cache_pages(), and then redirty all of
3118         * the pages by calling redirty_page_for_writepage() but that
3119         * would be ugly in the extreme.  So instead we would need to
3120         * replicate parts of the code in the above functions,
3121         * simplifying them because we wouldn't actually intend to
3122         * write out the pages, but rather only collect contiguous
3123         * logical block extents, call the multi-block allocator, and
3124         * then update the buffer heads with the block allocations.
3125         *
3126         * For now, though, we'll cheat by calling filemap_flush(),
3127         * which will map the blocks, and start the I/O, but not
3128         * actually wait for the I/O to complete.
3129         */
3130        return filemap_flush(inode->i_mapping);
3131}
3132
3133/*
3134 * bmap() is special.  It gets used by applications such as lilo and by
3135 * the swapper to find the on-disk block of a specific piece of data.
3136 *
3137 * Naturally, this is dangerous if the block concerned is still in the
3138 * journal.  If somebody makes a swapfile on an ext4 data-journaling
3139 * filesystem and enables swap, then they may get a nasty shock when the
3140 * data getting swapped to that swapfile suddenly gets overwritten by
3141 * the original zero's written out previously to the journal and
3142 * awaiting writeback in the kernel's buffer cache.
3143 *
3144 * So, if we see any bmap calls here on a modified, data-journaled file,
3145 * take extra steps to flush any blocks which might be in the cache.
3146 */
3147static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3148{
3149        struct inode *inode = mapping->host;
3150        journal_t *journal;
3151        int err;
3152
3153        /*
3154         * We can get here for an inline file via the FIBMAP ioctl
3155         */
3156        if (ext4_has_inline_data(inode))
3157                return 0;
3158
3159        if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3160                        test_opt(inode->i_sb, DELALLOC)) {
3161                /*
3162                 * With delalloc we want to sync the file
3163                 * so that we can make sure we allocate
3164                 * blocks for file
3165                 */
3166                filemap_write_and_wait(mapping);
3167        }
3168
3169        if (EXT4_JOURNAL(inode) &&
3170            ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3171                /*
3172                 * This is a REALLY heavyweight approach, but the use of
3173                 * bmap on dirty files is expected to be extremely rare:
3174                 * only if we run lilo or swapon on a freshly made file
3175                 * do we expect this to happen.
3176                 *
3177                 * (bmap requires CAP_SYS_RAWIO so this does not
3178                 * represent an unprivileged user DOS attack --- we'd be
3179                 * in trouble if mortal users could trigger this path at
3180                 * will.)
3181                 *
3182                 * NB. EXT4_STATE_JDATA is not set on files other than
3183                 * regular files.  If somebody wants to bmap a directory
3184                 * or symlink and gets confused because the buffer
3185                 * hasn't yet been flushed to disk, they deserve
3186                 * everything they get.
3187                 */
3188
3189                ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3190                journal = EXT4_JOURNAL(inode);
3191                jbd2_journal_lock_updates(journal);
3192                err = jbd2_journal_flush(journal);
3193                jbd2_journal_unlock_updates(journal);
3194
3195                if (err)
3196                        return 0;
3197        }
3198
3199        return generic_block_bmap(mapping, block, ext4_get_block);
3200}
3201
3202static int ext4_readpage(struct file *file, struct page *page)
3203{
3204        int ret = -EAGAIN;
3205        struct inode *inode = page->mapping->host;
3206
3207        trace_ext4_readpage(page);
3208
3209        if (ext4_has_inline_data(inode))
3210                ret = ext4_readpage_inline(inode, page);
3211
3212        if (ret == -EAGAIN)
3213                return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3214
3215        return ret;
3216}
3217
3218static int
3219ext4_readpages(struct file *file, struct address_space *mapping,
3220                struct list_head *pages, unsigned nr_pages)
3221{
3222        struct inode *inode = mapping->host;
3223
3224        /* If the file has inline data, no need to do readpages. */
3225        if (ext4_has_inline_data(inode))
3226                return 0;
3227
3228        return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3229}
3230
3231static void ext4_invalidatepage(struct page *page, unsigned int offset,
3232                                unsigned int length)
3233{
3234        trace_ext4_invalidatepage(page, offset, length);
3235
3236        /* No journalling happens on data buffers when this function is used */
3237        WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3238
3239        block_invalidatepage(page, offset, length);
3240}
3241
3242static int __ext4_journalled_invalidatepage(struct page *page,
3243                                            unsigned int offset,
3244                                            unsigned int length)
3245{
3246        journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3247
3248        trace_ext4_journalled_invalidatepage(page, offset, length);
3249
3250        /*
3251         * If it's a full truncate we just forget about the pending dirtying
3252         */
3253        if (offset == 0 && length == PAGE_SIZE)
3254                ClearPageChecked(page);
3255
3256        return jbd2_journal_invalidatepage(journal, page, offset, length);
3257}
3258
3259/* Wrapper for aops... */
3260static void ext4_journalled_invalidatepage(struct page *page,
3261                                           unsigned int offset,
3262                                           unsigned int length)
3263{
3264        WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3265}
3266
3267static int ext4_releasepage(struct page *page, gfp_t wait)
3268{
3269        journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3270
3271        trace_ext4_releasepage(page);
3272
3273        /* Page has dirty journalled data -> cannot release */
3274        if (PageChecked(page))
3275                return 0;
3276        if (journal)
3277                return jbd2_journal_try_to_free_buffers(journal, page, wait);
3278        else
3279                return try_to_free_buffers(page);
3280}
3281
3282#ifdef CONFIG_FS_DAX
3283/*
3284 * Get block function for DAX IO and mmap faults. It takes care of converting
3285 * unwritten extents to written ones and initializes new / converted blocks
3286 * to zeros.
3287 */
3288int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3289                       struct buffer_head *bh_result, int create)
3290{
3291        int ret;
3292
3293        ext4_debug("inode %lu, create flag %d\n", inode->i_ino, create);
3294        if (!create)
3295                return _ext4_get_block(inode, iblock, bh_result, 0);
3296
3297        ret = ext4_get_block_trans(inode, iblock, bh_result,
3298                                   EXT4_GET_BLOCKS_PRE_IO |
3299                                   EXT4_GET_BLOCKS_CREATE_ZERO);
3300        if (ret < 0)
3301                return ret;
3302
3303        if (buffer_unwritten(bh_result)) {
3304                /*
3305                 * We are protected by i_mmap_sem or i_mutex so we know block
3306                 * cannot go away from under us even though we dropped
3307                 * i_data_sem. Convert extent to written and write zeros there.
3308                 */
3309                ret = ext4_get_block_trans(inode, iblock, bh_result,
3310                                           EXT4_GET_BLOCKS_CONVERT |
3311                                           EXT4_GET_BLOCKS_CREATE_ZERO);
3312                if (ret < 0)
3313                        return ret;
3314        }
3315        /*
3316         * At least for now we have to clear BH_New so that DAX code
3317         * doesn't attempt to zero blocks again in a racy way.
3318         */
3319        clear_buffer_new(bh_result);
3320        return 0;
3321}
3322#else
3323/* Just define empty function, it will never get called. */
3324int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3325                       struct buffer_head *bh_result, int create)
3326{
3327        BUG();
3328        return 0;
3329}
3330#endif
3331
3332static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3333                            ssize_t size, void *private)
3334{
3335        ext4_io_end_t *io_end = private;
3336
3337        /* if not async direct IO just return */
3338        if (!io_end)
3339                return 0;
3340
3341        ext_debug("ext4_end_io_dio(): io_end 0x%p "
3342                  "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3343                  io_end, io_end->inode->i_ino, iocb, offset, size);
3344
3345        /*
3346         * Error during AIO DIO. We cannot convert unwritten extents as the
3347         * data was not written. Just clear the unwritten flag and drop io_end.
3348         */
3349        if (size <= 0) {
3350                ext4_clear_io_unwritten_flag(io_end);
3351                size = 0;
3352        }
3353        io_end->offset = offset;
3354        io_end->size = size;
3355        ext4_put_io_end(io_end);
3356
3357        return 0;
3358}
3359
3360/*
3361 * Handling of direct IO writes.
3362 *
3363 * For ext4 extent files, ext4 will do direct-io write even to holes,
3364 * preallocated extents, and those write extend the file, no need to
3365 * fall back to buffered IO.
3366 *
3367 * For holes, we fallocate those blocks, mark them as unwritten
3368 * If those blocks were preallocated, we mark sure they are split, but
3369 * still keep the range to write as unwritten.
3370 *
3371 * The unwritten extents will be converted to written when DIO is completed.
3372 * For async direct IO, since the IO may still pending when return, we
3373 * set up an end_io call back function, which will do the conversion
3374 * when async direct IO completed.
3375 *
3376 * If the O_DIRECT write will extend the file then add this inode to the
3377 * orphan list.  So recovery will truncate it back to the original size
3378 * if the machine crashes during the write.
3379 *
3380 */
3381static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3382{
3383        struct file *file = iocb->ki_filp;
3384        struct inode *inode = file->f_mapping->host;
3385        struct ext4_inode_info *ei = EXT4_I(inode);
3386        ssize_t ret;
3387        loff_t offset = iocb->ki_pos;
3388        size_t count = iov_iter_count(iter);
3389        int overwrite = 0;
3390        get_block_t *get_block_func = NULL;
3391        int dio_flags = 0;
3392        loff_t final_size = offset + count;
3393        int orphan = 0;
3394        handle_t *handle;
3395
3396        if (final_size > inode->i_size) {
3397                /* Credits for sb + inode write */
3398                handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3399                if (IS_ERR(handle)) {
3400                        ret = PTR_ERR(handle);
3401                        goto out;
3402                }
3403                ret = ext4_orphan_add(handle, inode);
3404                if (ret) {
3405                        ext4_journal_stop(handle);
3406                        goto out;
3407                }
3408                orphan = 1;
3409                ei->i_disksize = inode->i_size;
3410                ext4_journal_stop(handle);
3411        }
3412
3413        BUG_ON(iocb->private == NULL);
3414
3415        /*
3416         * Make all waiters for direct IO properly wait also for extent
3417         * conversion. This also disallows race between truncate() and
3418         * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3419         */
3420        inode_dio_begin(inode);
3421
3422        /* If we do a overwrite dio, i_mutex locking can be released */
3423        overwrite = *((int *)iocb->private);
3424
3425        if (overwrite)
3426                inode_unlock(inode);
3427
3428        /*
3429         * For extent mapped files we could direct write to holes and fallocate.
3430         *
3431         * Allocated blocks to fill the hole are marked as unwritten to prevent
3432         * parallel buffered read to expose the stale data before DIO complete
3433         * the data IO.
3434         *
3435         * As to previously fallocated extents, ext4 get_block will just simply
3436         * mark the buffer mapped but still keep the extents unwritten.
3437         *
3438         * For non AIO case, we will convert those unwritten extents to written
3439         * after return back from blockdev_direct_IO. That way we save us from
3440         * allocating io_end structure and also the overhead of offloading
3441         * the extent convertion to a workqueue.
3442         *
3443         * For async DIO, the conversion needs to be deferred when the
3444         * IO is completed. The ext4 end_io callback function will be
3445         * called to take care of the conversion work.  Here for async
3446         * case, we allocate an io_end structure to hook to the iocb.
3447         */
3448        iocb->private = NULL;
3449        if (overwrite)
3450                get_block_func = ext4_dio_get_block_overwrite;
3451        else if (IS_DAX(inode)) {
3452                /*
3453                 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3454                 * writes need zeroing either because they can race with page
3455                 * faults or because they use partial blocks.
3456                 */
3457                if (round_down(offset, 1<<inode->i_blkbits) >= inode->i_size &&
3458                    ext4_aligned_io(inode, offset, count))
3459                        get_block_func = ext4_dio_get_block;
3460                else
3461                        get_block_func = ext4_dax_get_block;
3462                dio_flags = DIO_LOCKING;
3463        } else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3464                   round_down(offset, 1 << inode->i_blkbits) >= inode->i_size) {
3465                get_block_func = ext4_dio_get_block;
3466                dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3467        } else if (is_sync_kiocb(iocb)) {
3468                get_block_func = ext4_dio_get_block_unwritten_sync;
3469                dio_flags = DIO_LOCKING;
3470        } else {
3471                get_block_func = ext4_dio_get_block_unwritten_async;
3472                dio_flags = DIO_LOCKING;
3473        }
3474#ifdef CONFIG_EXT4_FS_ENCRYPTION
3475        BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3476#endif
3477        if (IS_DAX(inode)) {
3478                ret = dax_do_io(iocb, inode, iter, get_block_func,
3479                                ext4_end_io_dio, dio_flags);
3480        } else
3481                ret = __blockdev_direct_IO(iocb, inode,
3482                                           inode->i_sb->s_bdev, iter,
3483                                           get_block_func,
3484                                           ext4_end_io_dio, NULL, dio_flags);
3485
3486        if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3487                                                EXT4_STATE_DIO_UNWRITTEN)) {
3488                int err;
3489                /*
3490                 * for non AIO case, since the IO is already
3491                 * completed, we could do the conversion right here
3492                 */
3493                err = ext4_convert_unwritten_extents(NULL, inode,
3494                                                     offset, ret);
3495                if (err < 0)
3496                        ret = err;
3497                ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3498        }
3499
3500        inode_dio_end(inode);
3501        /* take i_mutex locking again if we do a ovewrite dio */
3502        if (overwrite)
3503                inode_lock(inode);
3504
3505        if (ret < 0 && final_size > inode->i_size)
3506                ext4_truncate_failed_write(inode);
3507
3508        /* Handle extending of i_size after direct IO write */
3509        if (orphan) {
3510                int err;
3511
3512                /* Credits for sb + inode write */
3513                handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3514                if (IS_ERR(handle)) {
3515                        /* This is really bad luck. We've written the data
3516                         * but cannot extend i_size. Bail out and pretend
3517                         * the write failed... */
3518                        ret = PTR_ERR(handle);
3519                        if (inode->i_nlink)
3520                                ext4_orphan_del(NULL, inode);
3521
3522                        goto out;
3523                }
3524                if (inode->i_nlink)
3525                        ext4_orphan_del(handle, inode);
3526                if (ret > 0) {
3527                        loff_t end = offset + ret;
3528                        if (end > inode->i_size) {
3529                                ei->i_disksize = end;
3530                                i_size_write(inode, end);
3531                                /*
3532                                 * We're going to return a positive `ret'
3533                                 * here due to non-zero-length I/O, so there's
3534                                 * no way of reporting error returns from
3535                                 * ext4_mark_inode_dirty() to userspace.  So
3536                                 * ignore it.
3537                                 */
3538                                ext4_mark_inode_dirty(handle, inode);
3539                        }
3540                }
3541                err = ext4_journal_stop(handle);
3542                if (ret == 0)
3543                        ret = err;
3544        }
3545out:
3546        return ret;
3547}
3548
3549static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3550{
3551        struct address_space *mapping = iocb->ki_filp->f_mapping;
3552        struct inode *inode = mapping->host;
3553        ssize_t ret;
3554
3555        /*
3556         * Shared inode_lock is enough for us - it protects against concurrent
3557         * writes & truncates and since we take care of writing back page cache,
3558         * we are protected against page writeback as well.
3559         */
3560        inode_lock_shared(inode);
3561        if (IS_DAX(inode)) {
3562                ret = dax_do_io(iocb, inode, iter, ext4_dio_get_block, NULL, 0);
3563        } else {
3564                size_t count = iov_iter_count(iter);
3565
3566                ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3567                                                   iocb->ki_pos + count);
3568                if (ret)
3569                        goto out_unlock;
3570                ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3571                                           iter, ext4_dio_get_block,
3572                                           NULL, NULL, 0);
3573        }
3574out_unlock:
3575        inode_unlock_shared(inode);
3576        return ret;
3577}
3578
3579static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3580{
3581        struct file *file = iocb->ki_filp;
3582        struct inode *inode = file->f_mapping->host;
3583        size_t count = iov_iter_count(iter);
3584        loff_t offset = iocb->ki_pos;
3585        ssize_t ret;
3586
3587#ifdef CONFIG_EXT4_FS_ENCRYPTION
3588        if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3589                return 0;
3590#endif
3591
3592        /*
3593         * If we are doing data journalling we don't support O_DIRECT
3594         */
3595        if (ext4_should_journal_data(inode))
3596                return 0;
3597
3598        /* Let buffer I/O handle the inline data case. */
3599        if (ext4_has_inline_data(inode))
3600                return 0;
3601
3602        trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3603        if (iov_iter_rw(iter) == READ)
3604                ret = ext4_direct_IO_read(iocb, iter);
3605        else
3606                ret = ext4_direct_IO_write(iocb, iter);
3607        trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3608        return ret;
3609}
3610
3611/*
3612 * Pages can be marked dirty completely asynchronously from ext4's journalling
3613 * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3614 * much here because ->set_page_dirty is called under VFS locks.  The page is
3615 * not necessarily locked.
3616 *
3617 * We cannot just dirty the page and leave attached buffers clean, because the
3618 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3619 * or jbddirty because all the journalling code will explode.
3620 *
3621 * So what we do is to mark the page "pending dirty" and next time writepage
3622 * is called, propagate that into the buffers appropriately.
3623 */
3624static int ext4_journalled_set_page_dirty(struct page *page)
3625{
3626        SetPageChecked(page);
3627        return __set_page_dirty_nobuffers(page);
3628}
3629
3630static const struct address_space_operations ext4_aops = {
3631        .readpage               = ext4_readpage,
3632        .readpages              = ext4_readpages,
3633        .writepage              = ext4_writepage,
3634        .writepages             = ext4_writepages,
3635        .write_begin            = ext4_write_begin,
3636        .write_end              = ext4_write_end,
3637        .bmap                   = ext4_bmap,
3638        .invalidatepage         = ext4_invalidatepage,
3639        .releasepage            = ext4_releasepage,
3640        .direct_IO              = ext4_direct_IO,
3641        .migratepage            = buffer_migrate_page,
3642        .is_partially_uptodate  = block_is_partially_uptodate,
3643        .error_remove_page      = generic_error_remove_page,
3644};
3645
3646static const struct address_space_operations ext4_journalled_aops = {
3647        .readpage               = ext4_readpage,
3648        .readpages              = ext4_readpages,
3649        .writepage              = ext4_writepage,
3650        .writepages             = ext4_writepages,
3651        .write_begin            = ext4_write_begin,
3652        .write_end              = ext4_journalled_write_end,
3653        .set_page_dirty         = ext4_journalled_set_page_dirty,
3654        .bmap                   = ext4_bmap,
3655        .invalidatepage         = ext4_journalled_invalidatepage,
3656        .releasepage            = ext4_releasepage,
3657        .direct_IO              = ext4_direct_IO,
3658        .is_partially_uptodate  = block_is_partially_uptodate,
3659        .error_remove_page      = generic_error_remove_page,
3660};
3661
3662static const struct address_space_operations ext4_da_aops = {
3663        .readpage               = ext4_readpage,
3664        .readpages              = ext4_readpages,
3665        .writepage              = ext4_writepage,
3666        .writepages             = ext4_writepages,
3667        .write_begin            = ext4_da_write_begin,
3668        .write_end              = ext4_da_write_end,
3669        .bmap                   = ext4_bmap,
3670        .invalidatepage         = ext4_da_invalidatepage,
3671        .releasepage            = ext4_releasepage,
3672        .direct_IO              = ext4_direct_IO,
3673        .migratepage            = buffer_migrate_page,
3674        .is_partially_uptodate  = block_is_partially_uptodate,
3675        .error_remove_page      = generic_error_remove_page,
3676};
3677
3678void ext4_set_aops(struct inode *inode)
3679{
3680        switch (ext4_inode_journal_mode(inode)) {
3681        case EXT4_INODE_ORDERED_DATA_MODE:
3682        case EXT4_INODE_WRITEBACK_DATA_MODE:
3683                break;
3684        case EXT4_INODE_JOURNAL_DATA_MODE:
3685                inode->i_mapping->a_ops = &ext4_journalled_aops;
3686                return;
3687        default:
3688                BUG();
3689        }
3690        if (test_opt(inode->i_sb, DELALLOC))
3691                inode->i_mapping->a_ops = &ext4_da_aops;
3692        else
3693                inode->i_mapping->a_ops = &ext4_aops;
3694}
3695
3696static int __ext4_block_zero_page_range(handle_t *handle,
3697                struct address_space *mapping, loff_t from, loff_t length)
3698{
3699        ext4_fsblk_t index = from >> PAGE_SHIFT;
3700        unsigned offset = from & (PAGE_SIZE-1);
3701        unsigned blocksize, pos;
3702        ext4_lblk_t iblock;
3703        struct inode *inode = mapping->host;
3704        struct buffer_head *bh;
3705        struct page *page;
3706        int err = 0;
3707
3708        page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3709                                   mapping_gfp_constraint(mapping, ~__GFP_FS));
3710        if (!page)
3711                return -ENOMEM;
3712
3713        blocksize = inode->i_sb->s_blocksize;
3714
3715        iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3716
3717        if (!page_has_buffers(page))
3718                create_empty_buffers(page, blocksize, 0);
3719
3720        /* Find the buffer that contains "offset" */
3721        bh = page_buffers(page);
3722        pos = blocksize;
3723        while (offset >= pos) {
3724                bh = bh->b_this_page;
3725                iblock++;
3726                pos += blocksize;
3727        }
3728        if (buffer_freed(bh)) {
3729                BUFFER_TRACE(bh, "freed: skip");
3730                goto unlock;
3731        }
3732        if (!buffer_mapped(bh)) {
3733                BUFFER_TRACE(bh, "unmapped");
3734                ext4_get_block(inode, iblock, bh, 0);
3735                /* unmapped? It's a hole - nothing to do */
3736                if (!buffer_mapped(bh)) {
3737                        BUFFER_TRACE(bh, "still unmapped");
3738                        goto unlock;
3739                }
3740        }
3741
3742        /* Ok, it's mapped. Make sure it's up-to-date */
3743        if (PageUptodate(page))
3744                set_buffer_uptodate(bh);
3745
3746        if (!buffer_uptodate(bh)) {
3747                err = -EIO;
3748                ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3749                wait_on_buffer(bh);
3750                /* Uhhuh. Read error. Complain and punt. */
3751                if (!buffer_uptodate(bh))
3752                        goto unlock;
3753                if (S_ISREG(inode->i_mode) &&
3754                    ext4_encrypted_inode(inode)) {
3755                        /* We expect the key to be set. */
3756                        BUG_ON(!fscrypt_has_encryption_key(inode));
3757                        BUG_ON(blocksize != PAGE_SIZE);
3758                        WARN_ON_ONCE(fscrypt_decrypt_page(page));
3759                }
3760        }
3761        if (ext4_should_journal_data(inode)) {
3762                BUFFER_TRACE(bh, "get write access");
3763                err = ext4_journal_get_write_access(handle, bh);
3764                if (err)
3765                        goto unlock;
3766        }
3767        zero_user(page, offset, length);
3768        BUFFER_TRACE(bh, "zeroed end of block");
3769
3770        if (ext4_should_journal_data(inode)) {
3771                err = ext4_handle_dirty_metadata(handle, inode, bh);
3772        } else {
3773                err = 0;
3774                mark_buffer_dirty(bh);
3775                if (ext4_should_order_data(inode))
3776                        err = ext4_jbd2_inode_add_write(handle, inode);
3777        }
3778
3779unlock:
3780        unlock_page(page);
3781        put_page(page);
3782        return err;
3783}
3784
3785/*
3786 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3787 * starting from file offset 'from'.  The range to be zero'd must
3788 * be contained with in one block.  If the specified range exceeds
3789 * the end of the block it will be shortened to end of the block
3790 * that cooresponds to 'from'
3791 */
3792static int ext4_block_zero_page_range(handle_t *handle,
3793                struct address_space *mapping, loff_t from, loff_t length)
3794{
3795        struct inode *inode = mapping->host;
3796        unsigned offset = from & (PAGE_SIZE-1);
3797        unsigned blocksize = inode->i_sb->s_blocksize;
3798        unsigned max = blocksize - (offset & (blocksize - 1));
3799
3800        /*
3801         * correct length if it does not fall between
3802         * 'from' and the end of the block
3803         */
3804        if (length > max || length < 0)
3805                length = max;
3806
3807        if (IS_DAX(inode))
3808                return dax_zero_page_range(inode, from, length, ext4_get_block);
3809        return __ext4_block_zero_page_range(handle, mapping, from, length);
3810}
3811
3812/*
3813 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3814 * up to the end of the block which corresponds to `from'.
3815 * This required during truncate. We need to physically zero the tail end
3816 * of that block so it doesn't yield old data if the file is later grown.
3817 */
3818static int ext4_block_truncate_page(handle_t *handle,
3819                struct address_space *mapping, loff_t from)
3820{
3821        unsigned offset = from & (PAGE_SIZE-1);
3822        unsigned length;
3823        unsigned blocksize;
3824        struct inode *inode = mapping->host;
3825
3826        /* If we are processing an encrypted inode during orphan list handling */
3827        if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
3828                return 0;
3829
3830        blocksize = inode->i_sb->s_blocksize;
3831        length = blocksize - (offset & (blocksize - 1));
3832
3833        return ext4_block_zero_page_range(handle, mapping, from, length);
3834}
3835
3836int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3837                             loff_t lstart, loff_t length)
3838{
3839        struct super_block *sb = inode->i_sb;
3840        struct address_space *mapping = inode->i_mapping;
3841        unsigned partial_start, partial_end;
3842        ext4_fsblk_t start, end;
3843        loff_t byte_end = (lstart + length - 1);
3844        int err = 0;
3845
3846        partial_start = lstart & (sb->s_blocksize - 1);
3847        partial_end = byte_end & (sb->s_blocksize - 1);
3848
3849        start = lstart >> sb->s_blocksize_bits;
3850        end = byte_end >> sb->s_blocksize_bits;
3851
3852        /* Handle partial zero within the single block */
3853        if (start == end &&
3854            (partial_start || (partial_end != sb->s_blocksize - 1))) {
3855                err = ext4_block_zero_page_range(handle, mapping,
3856                                                 lstart, length);
3857                return err;
3858        }
3859        /* Handle partial zero out on the start of the range */
3860        if (partial_start) {
3861                err = ext4_block_zero_page_range(handle, mapping,
3862                                                 lstart, sb->s_blocksize);
3863                if (err)
3864                        return err;
3865        }
3866        /* Handle partial zero out on the end of the range */
3867        if (partial_end != sb->s_blocksize - 1)
3868                err = ext4_block_zero_page_range(handle, mapping,
3869                                                 byte_end - partial_end,
3870                                                 partial_end + 1);
3871        return err;
3872}
3873
3874int ext4_can_truncate(struct inode *inode)
3875{
3876        if (S_ISREG(inode->i_mode))
3877                return 1;
3878        if (S_ISDIR(inode->i_mode))
3879                return 1;
3880        if (S_ISLNK(inode->i_mode))
3881                return !ext4_inode_is_fast_symlink(inode);
3882        return 0;
3883}
3884
3885/*
3886 * We have to make sure i_disksize gets properly updated before we truncate
3887 * page cache due to hole punching or zero range. Otherwise i_disksize update
3888 * can get lost as it may have been postponed to submission of writeback but
3889 * that will never happen after we truncate page cache.
3890 */
3891int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3892                                      loff_t len)
3893{
3894        handle_t *handle;
3895        loff_t size = i_size_read(inode);
3896
3897        WARN_ON(!inode_is_locked(inode));
3898        if (offset > size || offset + len < size)
3899                return 0;
3900
3901        if (EXT4_I(inode)->i_disksize >= size)
3902                return 0;
3903
3904        handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3905        if (IS_ERR(handle))
3906                return PTR_ERR(handle);
3907        ext4_update_i_disksize(inode, size);
3908        ext4_mark_inode_dirty(handle, inode);
3909        ext4_journal_stop(handle);
3910
3911        return 0;
3912}
3913
3914/*
3915 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3916 * associated with the given offset and length
3917 *
3918 * @inode:  File inode
3919 * @offset: The offset where the hole will begin
3920 * @len:    The length of the hole
3921 *
3922 * Returns: 0 on success or negative on failure
3923 */
3924
3925int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3926{
3927        struct super_block *sb = inode->i_sb;
3928        ext4_lblk_t first_block, stop_block;
3929        struct address_space *mapping = inode->i_mapping;
3930        loff_t first_block_offset, last_block_offset;
3931        handle_t *handle;
3932        unsigned int credits;
3933        int ret = 0;
3934
3935        if (!S_ISREG(inode->i_mode))
3936                return -EOPNOTSUPP;
3937
3938        trace_ext4_punch_hole(inode, offset, length, 0);
3939
3940        /*
3941         * Write out all dirty pages to avoid race conditions
3942         * Then release them.
3943         */
3944        if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3945                ret = filemap_write_and_wait_range(mapping, offset,
3946                                                   offset + length - 1);
3947                if (ret)
3948                        return ret;
3949        }
3950
3951        inode_lock(inode);
3952
3953        /* No need to punch hole beyond i_size */
3954        if (offset >= inode->i_size)
3955                goto out_mutex;
3956
3957        /*
3958         * If the hole extends beyond i_size, set the hole
3959         * to end after the page that contains i_size
3960         */
3961        if (offset + length > inode->i_size) {
3962                length = inode->i_size +
3963                   PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3964                   offset;
3965        }
3966
3967        if (offset & (sb->s_blocksize - 1) ||
3968            (offset + length) & (sb->s_blocksize - 1)) {
3969                /*
3970                 * Attach jinode to inode for jbd2 if we do any zeroing of
3971                 * partial block
3972                 */
3973                ret = ext4_inode_attach_jinode(inode);
3974                if (ret < 0)
3975                        goto out_mutex;
3976
3977        }
3978
3979        /* Wait all existing dio workers, newcomers will block on i_mutex */
3980        ext4_inode_block_unlocked_dio(inode);
3981        inode_dio_wait(inode);
3982
3983        /*
3984         * Prevent page faults from reinstantiating pages we have released from
3985         * page cache.
3986         */
3987        down_write(&EXT4_I(inode)->i_mmap_sem);
3988        first_block_offset = round_up(offset, sb->s_blocksize);
3989        last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3990
3991        /* Now release the pages and zero block aligned part of pages*/
3992        if (last_block_offset > first_block_offset) {
3993                ret = ext4_update_disksize_before_punch(inode, offset, length);
3994                if (ret)
3995                        goto out_dio;
3996                truncate_pagecache_range(inode, first_block_offset,
3997                                         last_block_offset);
3998        }
3999
4000        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4001                credits = ext4_writepage_trans_blocks(inode);
4002        else
4003                credits = ext4_blocks_for_truncate(inode);
4004        handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4005        if (IS_ERR(handle)) {
4006                ret = PTR_ERR(handle);
4007                ext4_std_error(sb, ret);
4008                goto out_dio;
4009        }
4010
4011        ret = ext4_zero_partial_blocks(handle, inode, offset,
4012                                       length);
4013        if (ret)
4014                goto out_stop;
4015
4016        first_block = (offset + sb->s_blocksize - 1) >>
4017                EXT4_BLOCK_SIZE_BITS(sb);
4018        stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4019
4020        /* If there are no blocks to remove, return now */
4021        if (first_block >= stop_block)
4022                goto out_stop;
4023
4024        down_write(&EXT4_I(inode)->i_data_sem);
4025        ext4_discard_preallocations(inode);
4026
4027        ret = ext4_es_remove_extent(inode, first_block,
4028                                    stop_block - first_block);
4029        if (ret) {
4030                up_write(&EXT4_I(inode)->i_data_sem);
4031                goto out_stop;
4032        }
4033
4034        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4035                ret = ext4_ext_remove_space(inode, first_block,
4036                                            stop_block - 1);
4037        else
4038                ret = ext4_ind_remove_space(handle, inode, first_block,
4039                                            stop_block);
4040
4041        up_write(&EXT4_I(inode)->i_data_sem);
4042        if (IS_SYNC(inode))
4043                ext4_handle_sync(handle);
4044
4045        inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4046        ext4_mark_inode_dirty(handle, inode);
4047out_stop:
4048        ext4_journal_stop(handle);
4049out_dio:
4050        up_write(&EXT4_I(inode)->i_mmap_sem);
4051        ext4_inode_resume_unlocked_dio(inode);
4052out_mutex:
4053        inode_unlock(inode);
4054        return ret;
4055}
4056
4057int ext4_inode_attach_jinode(struct inode *inode)
4058{
4059        struct ext4_inode_info *ei = EXT4_I(inode);
4060        struct jbd2_inode *jinode;
4061
4062        if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4063                return 0;
4064
4065        jinode = jbd2_alloc_inode(GFP_KERNEL);
4066        spin_lock(&inode->i_lock);
4067        if (!ei->jinode) {
4068                if (!jinode) {
4069                        spin_unlock(&inode->i_lock);
4070                        return -ENOMEM;
4071                }
4072                ei->jinode = jinode;
4073                jbd2_journal_init_jbd_inode(ei->jinode, inode);
4074                jinode = NULL;
4075        }
4076        spin_unlock(&inode->i_lock);
4077        if (unlikely(jinode != NULL))
4078                jbd2_free_inode(jinode);
4079        return 0;
4080}
4081
4082/*
4083 * ext4_truncate()
4084 *
4085 * We block out ext4_get_block() block instantiations across the entire
4086 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4087 * simultaneously on behalf of the same inode.
4088 *
4089 * As we work through the truncate and commit bits of it to the journal there
4090 * is one core, guiding principle: the file's tree must always be consistent on
4091 * disk.  We must be able to restart the truncate after a crash.
4092 *
4093 * The file's tree may be transiently inconsistent in memory (although it
4094 * probably isn't), but whenever we close off and commit a journal transaction,
4095 * the contents of (the filesystem + the journal) must be consistent and
4096 * restartable.  It's pretty simple, really: bottom up, right to left (although
4097 * left-to-right works OK too).
4098 *
4099 * Note that at recovery time, journal replay occurs *before* the restart of
4100 * truncate against the orphan inode list.
4101 *
4102 * The committed inode has the new, desired i_size (which is the same as
4103 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4104 * that this inode's truncate did not complete and it will again call
4105 * ext4_truncate() to have another go.  So there will be instantiated blocks
4106 * to the right of the truncation point in a crashed ext4 filesystem.  But
4107 * that's fine - as long as they are linked from the inode, the post-crash
4108 * ext4_truncate() run will find them and release them.
4109 */
4110void ext4_truncate(struct inode *inode)
4111{
4112        struct ext4_inode_info *ei = EXT4_I(inode);
4113        unsigned int credits;
4114        handle_t *handle;
4115        struct address_space *mapping = inode->i_mapping;
4116
4117        /*
4118         * There is a possibility that we're either freeing the inode
4119         * or it's a completely new inode. In those cases we might not
4120         * have i_mutex locked because it's not necessary.
4121         */
4122        if (!(inode->i_state & (I_NEW|I_FREEING)))
4123                WARN_ON(!inode_is_locked(inode));
4124        trace_ext4_truncate_enter(inode);
4125
4126        if (!ext4_can_truncate(inode))
4127                return;
4128
4129        ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4130
4131        if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4132                ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4133
4134        if (ext4_has_inline_data(inode)) {
4135                int has_inline = 1;
4136
4137                ext4_inline_data_truncate(inode, &has_inline);
4138                if (has_inline)
4139                        return;
4140        }
4141
4142        /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4143        if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4144                if (ext4_inode_attach_jinode(inode) < 0)
4145                        return;
4146        }
4147
4148        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4149                credits = ext4_writepage_trans_blocks(inode);
4150        else
4151                credits = ext4_blocks_for_truncate(inode);
4152
4153        handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4154        if (IS_ERR(handle)) {
4155                ext4_std_error(inode->i_sb, PTR_ERR(handle));
4156                return;
4157        }
4158
4159        if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4160                ext4_block_truncate_page(handle, mapping, inode->i_size);
4161
4162        /*
4163         * We add the inode to the orphan list, so that if this
4164         * truncate spans multiple transactions, and we crash, we will
4165         * resume the truncate when the filesystem recovers.  It also
4166         * marks the inode dirty, to catch the new size.
4167         *
4168         * Implication: the file must always be in a sane, consistent
4169         * truncatable state while each transaction commits.
4170         */
4171        if (ext4_orphan_add(handle, inode))
4172                goto out_stop;
4173
4174        down_write(&EXT4_I(inode)->i_data_sem);
4175
4176        ext4_discard_preallocations(inode);
4177
4178        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4179                ext4_ext_truncate(handle, inode);
4180        else
4181                ext4_ind_truncate(handle, inode);
4182
4183        up_write(&ei->i_data_sem);
4184
4185        if (IS_SYNC(inode))
4186                ext4_handle_sync(handle);
4187
4188out_stop:
4189        /*
4190         * If this was a simple ftruncate() and the file will remain alive,
4191         * then we need to clear up the orphan record which we created above.
4192         * However, if this was a real unlink then we were called by
4193         * ext4_evict_inode(), and we allow that function to clean up the
4194         * orphan info for us.
4195         */
4196        if (inode->i_nlink)
4197                ext4_orphan_del(handle, inode);
4198
4199        inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4200        ext4_mark_inode_dirty(handle, inode);
4201        ext4_journal_stop(handle);
4202
4203        trace_ext4_truncate_exit(inode);
4204}
4205
4206/*
4207 * ext4_get_inode_loc returns with an extra refcount against the inode's
4208 * underlying buffer_head on success. If 'in_mem' is true, we have all
4209 * data in memory that is needed to recreate the on-disk version of this
4210 * inode.
4211 */
4212static int __ext4_get_inode_loc(struct inode *inode,
4213                                struct ext4_iloc *iloc, int in_mem)
4214{
4215        struct ext4_group_desc  *gdp;
4216        struct buffer_head      *bh;
4217        struct super_block      *sb = inode->i_sb;
4218        ext4_fsblk_t            block;
4219        int                     inodes_per_block, inode_offset;
4220
4221        iloc->bh = NULL;
4222        if (!ext4_valid_inum(sb, inode->i_ino))
4223                return -EFSCORRUPTED;
4224
4225        iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4226        gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4227        if (!gdp)
4228                return -EIO;
4229
4230        /*
4231         * Figure out the offset within the block group inode table
4232         */
4233        inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4234        inode_offset = ((inode->i_ino - 1) %
4235                        EXT4_INODES_PER_GROUP(sb));
4236        block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4237        iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4238
4239        bh = sb_getblk(sb, block);
4240        if (unlikely(!bh))
4241                return -ENOMEM;
4242        if (!buffer_uptodate(bh)) {
4243                lock_buffer(bh);
4244
4245                /*
4246                 * If the buffer has the write error flag, we have failed
4247                 * to write out another inode in the same block.  In this
4248                 * case, we don't have to read the block because we may
4249                 * read the old inode data successfully.
4250                 */
4251                if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4252                        set_buffer_uptodate(bh);
4253
4254                if (buffer_uptodate(bh)) {
4255                        /* someone brought it uptodate while we waited */
4256                        unlock_buffer(bh);
4257                        goto has_buffer;
4258                }
4259
4260                /*
4261                 * If we have all information of the inode in memory and this
4262                 * is the only valid inode in the block, we need not read the
4263                 * block.
4264                 */
4265                if (in_mem) {
4266                        struct buffer_head *bitmap_bh;
4267                        int i, start;
4268
4269                        start = inode_offset & ~(inodes_per_block - 1);
4270
4271                        /* Is the inode bitmap in cache? */
4272                        bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4273                        if (unlikely(!bitmap_bh))
4274                                goto make_io;
4275
4276                        /*
4277                         * If the inode bitmap isn't in cache then the
4278                         * optimisation may end up performing two reads instead
4279                         * of one, so skip it.
4280                         */
4281                        if (!buffer_uptodate(bitmap_bh)) {
4282                                brelse(bitmap_bh);
4283                                goto make_io;
4284                        }
4285                        for (i = start; i < start + inodes_per_block; i++) {
4286                                if (i == inode_offset)
4287                                        continue;
4288                                if (ext4_test_bit(i, bitmap_bh->b_data))
4289                                        break;
4290                        }
4291                        brelse(bitmap_bh);
4292                        if (i == start + inodes_per_block) {
4293                                /* all other inodes are free, so skip I/O */
4294                                memset(bh->b_data, 0, bh->b_size);
4295                                set_buffer_uptodate(bh);
4296                                unlock_buffer(bh);
4297                                goto has_buffer;
4298                        }
4299                }
4300
4301make_io:
4302                /*
4303                 * If we need to do any I/O, try to pre-readahead extra
4304                 * blocks from the inode table.
4305                 */
4306                if (EXT4_SB(sb)->s_inode_readahead_blks) {
4307                        ext4_fsblk_t b, end, table;
4308                        unsigned num;
4309                        __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4310
4311                        table = ext4_inode_table(sb, gdp);
4312                        /* s_inode_readahead_blks is always a power of 2 */
4313                        b = block & ~((ext4_fsblk_t) ra_blks - 1);
4314                        if (table > b)
4315                                b = table;
4316                        end = b + ra_blks;
4317                        num = EXT4_INODES_PER_GROUP(sb);
4318                        if (ext4_has_group_desc_csum(sb))
4319                                num -= ext4_itable_unused_count(sb, gdp);
4320                        table += num / inodes_per_block;
4321                        if (end > table)
4322                                end = table;
4323                        while (b <= end)
4324                                sb_breadahead(sb, b++);
4325                }
4326
4327                /*
4328                 * There are other valid inodes in the buffer, this inode
4329                 * has in-inode xattrs, or we don't have this inode in memory.
4330                 * Read the block from disk.
4331                 */
4332                trace_ext4_load_inode(inode);
4333                get_bh(bh);
4334                bh->b_end_io = end_buffer_read_sync;
4335                submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4336                wait_on_buffer(bh);
4337                if (!buffer_uptodate(bh)) {
4338                        EXT4_ERROR_INODE_BLOCK(inode, block,
4339                                               "unable to read itable block");
4340                        brelse(bh);
4341                        return -EIO;
4342                }
4343        }
4344has_buffer:
4345        iloc->bh = bh;
4346        return 0;
4347}
4348
4349int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4350{
4351        /* We have all inode data except xattrs in memory here. */
4352        return __ext4_get_inode_loc(inode, iloc,
4353                !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4354}
4355
4356void ext4_set_inode_flags(struct inode *inode)
4357{
4358        unsigned int flags = EXT4_I(inode)->i_flags;
4359        unsigned int new_fl = 0;
4360
4361        if (flags & EXT4_SYNC_FL)
4362                new_fl |= S_SYNC;
4363        if (flags & EXT4_APPEND_FL)
4364                new_fl |= S_APPEND;
4365        if (flags & EXT4_IMMUTABLE_FL)
4366                new_fl |= S_IMMUTABLE;
4367        if (flags & EXT4_NOATIME_FL)
4368                new_fl |= S_NOATIME;
4369        if (flags & EXT4_DIRSYNC_FL)
4370                new_fl |= S_DIRSYNC;
4371        if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
4372                new_fl |= S_DAX;
4373        inode_set_flags(inode, new_fl,
4374                        S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4375}
4376
4377/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4378void ext4_get_inode_flags(struct ext4_inode_info *ei)
4379{
4380        unsigned int vfs_fl;
4381        unsigned long old_fl, new_fl;
4382
4383        do {
4384                vfs_fl = ei->vfs_inode.i_flags;
4385                old_fl = ei->i_flags;
4386                new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4387                                EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4388                                EXT4_DIRSYNC_FL);
4389                if (vfs_fl & S_SYNC)
4390                        new_fl |= EXT4_SYNC_FL;
4391                if (vfs_fl & S_APPEND)
4392                        new_fl |= EXT4_APPEND_FL;
4393                if (vfs_fl & S_IMMUTABLE)
4394                        new_fl |= EXT4_IMMUTABLE_FL;
4395                if (vfs_fl & S_NOATIME)
4396                        new_fl |= EXT4_NOATIME_FL;
4397                if (vfs_fl & S_DIRSYNC)
4398                        new_fl |= EXT4_DIRSYNC_FL;
4399        } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4400}
4401
4402static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4403                                  struct ext4_inode_info *ei)
4404{
4405        blkcnt_t i_blocks ;
4406        struct inode *inode = &(ei->vfs_inode);
4407        struct super_block *sb = inode->i_sb;
4408
4409        if (ext4_has_feature_huge_file(sb)) {
4410                /* we are using combined 48 bit field */
4411                i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4412                                        le32_to_cpu(raw_inode->i_blocks_lo);
4413                if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4414                        /* i_blocks represent file system block size */
4415                        return i_blocks  << (inode->i_blkbits - 9);
4416                } else {
4417                        return i_blocks;
4418                }
4419        } else {
4420                return le32_to_cpu(raw_inode->i_blocks_lo);
4421        }
4422}
4423
4424static inline void ext4_iget_extra_inode(struct inode *inode,
4425                                         struct ext4_inode *raw_inode,
4426                                         struct ext4_inode_info *ei)
4427{
4428        __le32 *magic = (void *)raw_inode +
4429                        EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4430        if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4431                ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4432                ext4_find_inline_data_nolock(inode);
4433        } else
4434                EXT4_I(inode)->i_inline_off = 0;
4435}
4436
4437int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4438{
4439        if (!ext4_has_feature_project(inode->i_sb))
4440                return -EOPNOTSUPP;
4441        *projid = EXT4_I(inode)->i_projid;
4442        return 0;
4443}
4444
4445struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4446{
4447        struct ext4_iloc iloc;
4448        struct ext4_inode *raw_inode;
4449        struct ext4_inode_info *ei;
4450        struct inode *inode;
4451        journal_t *journal = EXT4_SB(sb)->s_journal;
4452        long ret;
4453        loff_t size;
4454        int block;
4455        uid_t i_uid;
4456        gid_t i_gid;
4457        projid_t i_projid;
4458
4459        inode = iget_locked(sb, ino);
4460        if (!inode)
4461                return ERR_PTR(-ENOMEM);
4462        if (!(inode->i_state & I_NEW))
4463                return inode;
4464
4465        ei = EXT4_I(inode);
4466        iloc.bh = NULL;
4467
4468        ret = __ext4_get_inode_loc(inode, &iloc, 0);
4469        if (ret < 0)
4470                goto bad_inode;
4471        raw_inode = ext4_raw_inode(&iloc);
4472
4473        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4474                ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4475                if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4476                    EXT4_INODE_SIZE(inode->i_sb)) {
4477                        EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4478                                EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4479                                EXT4_INODE_SIZE(inode->i_sb));
4480                        ret = -EFSCORRUPTED;
4481                        goto bad_inode;
4482                }
4483        } else
4484                ei->i_extra_isize = 0;
4485
4486        /* Precompute checksum seed for inode metadata */
4487        if (ext4_has_metadata_csum(sb)) {
4488                struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4489                __u32 csum;
4490                __le32 inum = cpu_to_le32(inode->i_ino);
4491                __le32 gen = raw_inode->i_generation;
4492                csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4493                                   sizeof(inum));
4494                ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4495                                              sizeof(gen));
4496        }
4497
4498        if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4499                EXT4_ERROR_INODE(inode, "checksum invalid");
4500                ret = -EFSBADCRC;
4501                goto bad_inode;
4502        }
4503
4504        inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4505        i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4506        i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4507        if (ext4_has_feature_project(sb) &&
4508            EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4509            EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4510                i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4511        else
4512                i_projid = EXT4_DEF_PROJID;
4513
4514        if (!(test_opt(inode->i_sb, NO_UID32))) {
4515                i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4516                i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4517        }
4518        i_uid_write(inode, i_uid);
4519        i_gid_write(inode, i_gid);
4520        ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4521        set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4522
4523        ext4_clear_state_flags(ei);     /* Only relevant on 32-bit archs */
4524        ei->i_inline_off = 0;
4525        ei->i_dir_start_lookup = 0;
4526        ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4527        /* We now have enough fields to check if the inode was active or not.
4528         * This is needed because nfsd might try to access dead inodes
4529         * the test is that same one that e2fsck uses
4530         * NeilBrown 1999oct15
4531         */
4532        if (inode->i_nlink == 0) {
4533                if ((inode->i_mode == 0 ||
4534                     !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4535                    ino != EXT4_BOOT_LOADER_INO) {
4536                        /* this inode is deleted */
4537                        ret = -ESTALE;
4538                        goto bad_inode;
4539                }
4540                /* The only unlinked inodes we let through here have
4541                 * valid i_mode and are being read by the orphan
4542                 * recovery code: that's fine, we're about to complete
4543                 * the process of deleting those.
4544                 * OR it is the EXT4_BOOT_LOADER_INO which is
4545                 * not initialized on a new filesystem. */
4546        }
4547        ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4548        inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4549        ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4550        if (ext4_has_feature_64bit(sb))
4551                ei->i_file_acl |=
4552                        ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4553        inode->i_size = ext4_isize(raw_inode);
4554        if ((size = i_size_read(inode)) < 0) {
4555                EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4556                ret = -EFSCORRUPTED;
4557                goto bad_inode;
4558        }
4559        ei->i_disksize = inode->i_size;
4560#ifdef CONFIG_QUOTA
4561        ei->i_reserved_quota = 0;
4562#endif
4563        inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4564        ei->i_block_group = iloc.block_group;
4565        ei->i_last_alloc_group = ~0;
4566        /*
4567         * NOTE! The in-memory inode i_data array is in little-endian order
4568         * even on big-endian machines: we do NOT byteswap the block numbers!
4569         */
4570        for (block = 0; block < EXT4_N_BLOCKS; block++)
4571                ei->i_data[block] = raw_inode->i_block[block];
4572        INIT_LIST_HEAD(&ei->i_orphan);
4573
4574        /*
4575         * Set transaction id's of transactions that have to be committed
4576         * to finish f[data]sync. We set them to currently running transaction
4577         * as we cannot be sure that the inode or some of its metadata isn't
4578         * part of the transaction - the inode could have been reclaimed and
4579         * now it is reread from disk.
4580         */
4581        if (journal) {
4582                transaction_t *transaction;
4583                tid_t tid;
4584
4585                read_lock(&journal->j_state_lock);
4586                if (journal->j_running_transaction)
4587                        transaction = journal->j_running_transaction;
4588                else
4589                        transaction = journal->j_committing_transaction;
4590                if (transaction)
4591                        tid = transaction->t_tid;
4592                else
4593                        tid = journal->j_commit_sequence;
4594                read_unlock(&journal->j_state_lock);
4595                ei->i_sync_tid = tid;
4596                ei->i_datasync_tid = tid;
4597        }
4598
4599        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4600                if (ei->i_extra_isize == 0) {
4601                        /* The extra space is currently unused. Use it. */
4602                        ei->i_extra_isize = sizeof(struct ext4_inode) -
4603                                            EXT4_GOOD_OLD_INODE_SIZE;
4604                } else {
4605                        ext4_iget_extra_inode(inode, raw_inode, ei);
4606                }
4607        }
4608
4609        EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4610        EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4611        EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4612        EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4613
4614        if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4615                inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4616                if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4617                        if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4618                                inode->i_version |=
4619                    (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4620                }
4621        }
4622
4623        ret = 0;
4624        if (ei->i_file_acl &&
4625            !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4626                EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4627                                 ei->i_file_acl);
4628                ret = -EFSCORRUPTED;
4629                goto bad_inode;
4630        } else if (!ext4_has_inline_data(inode)) {
4631                if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4632                        if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4633                            (S_ISLNK(inode->i_mode) &&
4634                             !ext4_inode_is_fast_symlink(inode))))
4635                                /* Validate extent which is part of inode */
4636                                ret = ext4_ext_check_inode(inode);
4637                } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4638                           (S_ISLNK(inode->i_mode) &&
4639                            !ext4_inode_is_fast_symlink(inode))) {
4640                        /* Validate block references which are part of inode */
4641                        ret = ext4_ind_check_inode(inode);
4642                }
4643        }
4644        if (ret)
4645                goto bad_inode;
4646
4647        if (S_ISREG(inode->i_mode)) {
4648                inode->i_op = &ext4_file_inode_operations;
4649                inode->i_fop = &ext4_file_operations;
4650                ext4_set_aops(inode);
4651        } else if (S_ISDIR(inode->i_mode)) {
4652                inode->i_op = &ext4_dir_inode_operations;
4653                inode->i_fop = &ext4_dir_operations;
4654        } else if (S_ISLNK(inode->i_mode)) {
4655                if (ext4_encrypted_inode(inode)) {
4656                        inode->i_op = &ext4_encrypted_symlink_inode_operations;
4657                        ext4_set_aops(inode);
4658                } else if (ext4_inode_is_fast_symlink(inode)) {
4659                        inode->i_link = (char *)ei->i_data;
4660                        inode->i_op = &ext4_fast_symlink_inode_operations;
4661                        nd_terminate_link(ei->i_data, inode->i_size,
4662                                sizeof(ei->i_data) - 1);
4663                } else {
4664                        inode->i_op = &ext4_symlink_inode_operations;
4665                        ext4_set_aops(inode);
4666                }
4667                inode_nohighmem(inode);
4668        } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4669              S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4670                inode->i_op = &ext4_special_inode_operations;
4671                if (raw_inode->i_block[0])
4672                        init_special_inode(inode, inode->i_mode,
4673                           old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4674                else
4675                        init_special_inode(inode, inode->i_mode,
4676                           new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4677        } else if (ino == EXT4_BOOT_LOADER_INO) {
4678                make_bad_inode(inode);
4679        } else {
4680                ret = -EFSCORRUPTED;
4681                EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4682                goto bad_inode;
4683        }
4684        brelse(iloc.bh);
4685        ext4_set_inode_flags(inode);
4686        unlock_new_inode(inode);
4687        return inode;
4688
4689bad_inode:
4690        brelse(iloc.bh);
4691        iget_failed(inode);
4692        return ERR_PTR(ret);
4693}
4694
4695struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4696{
4697        if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4698                return ERR_PTR(-EFSCORRUPTED);
4699        return ext4_iget(sb, ino);
4700}
4701
4702static int ext4_inode_blocks_set(handle_t *handle,
4703                                struct ext4_inode *raw_inode,
4704                                struct ext4_inode_info *ei)
4705{
4706        struct inode *inode = &(ei->vfs_inode);
4707        u64 i_blocks = inode->i_blocks;
4708        struct super_block *sb = inode->i_sb;
4709
4710        if (i_blocks <= ~0U) {
4711                /*
4712                 * i_blocks can be represented in a 32 bit variable
4713                 * as multiple of 512 bytes
4714                 */
4715                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4716                raw_inode->i_blocks_high = 0;
4717                ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4718                return 0;
4719        }
4720        if (!ext4_has_feature_huge_file(sb))
4721                return -EFBIG;
4722
4723        if (i_blocks <= 0xffffffffffffULL) {
4724                /*
4725                 * i_blocks can be represented in a 48 bit variable
4726                 * as multiple of 512 bytes
4727                 */
4728                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4729                raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4730                ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4731        } else {
4732                ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4733                /* i_block is stored in file system block size */
4734                i_blocks = i_blocks >> (inode->i_blkbits - 9);
4735                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4736                raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4737        }
4738        return 0;
4739}
4740
4741struct other_inode {
4742        unsigned long           orig_ino;
4743        struct ext4_inode       *raw_inode;
4744};
4745
4746static int other_inode_match(struct inode * inode, unsigned long ino,
4747                             void *data)
4748{
4749        struct other_inode *oi = (struct other_inode *) data;
4750
4751        if ((inode->i_ino != ino) ||
4752            (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4753                               I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4754            ((inode->i_state & I_DIRTY_TIME) == 0))
4755                return 0;
4756        spin_lock(&inode->i_lock);
4757        if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4758                                I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4759            (inode->i_state & I_DIRTY_TIME)) {
4760                struct ext4_inode_info  *ei = EXT4_I(inode);
4761
4762                inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4763                spin_unlock(&inode->i_lock);
4764
4765                spin_lock(&ei->i_raw_lock);
4766                EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4767                EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4768                EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4769                ext4_inode_csum_set(inode, oi->raw_inode, ei);
4770                spin_unlock(&ei->i_raw_lock);
4771                trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4772                return -1;
4773        }
4774        spin_unlock(&inode->i_lock);
4775        return -1;
4776}
4777
4778/*
4779 * Opportunistically update the other time fields for other inodes in
4780 * the same inode table block.
4781 */
4782static void ext4_update_other_inodes_time(struct super_block *sb,
4783                                          unsigned long orig_ino, char *buf)
4784{
4785        struct other_inode oi;
4786        unsigned long ino;
4787        int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4788        int inode_size = EXT4_INODE_SIZE(sb);
4789
4790        oi.orig_ino = orig_ino;
4791        /*
4792         * Calculate the first inode in the inode table block.  Inode
4793         * numbers are one-based.  That is, the first inode in a block
4794         * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4795         */
4796        ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4797        for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4798                if (ino == orig_ino)
4799                        continue;
4800                oi.raw_inode = (struct ext4_inode *) buf;
4801                (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4802        }
4803}
4804
4805/*
4806 * Post the struct inode info into an on-disk inode location in the
4807 * buffer-cache.  This gobbles the caller's reference to the
4808 * buffer_head in the inode location struct.
4809 *
4810 * The caller must have write access to iloc->bh.
4811 */
4812static int ext4_do_update_inode(handle_t *handle,
4813                                struct inode *inode,
4814                                struct ext4_iloc *iloc)
4815{
4816        struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4817        struct ext4_inode_info *ei = EXT4_I(inode);
4818        struct buffer_head *bh = iloc->bh;
4819        struct super_block *sb = inode->i_sb;
4820        int err = 0, rc, block;
4821        int need_datasync = 0, set_large_file = 0;
4822        uid_t i_uid;
4823        gid_t i_gid;
4824        projid_t i_projid;
4825
4826        spin_lock(&ei->i_raw_lock);
4827
4828        /* For fields not tracked in the in-memory inode,
4829         * initialise them to zero for new inodes. */
4830        if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4831                memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4832
4833        ext4_get_inode_flags(ei);
4834        raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4835        i_uid = i_uid_read(inode);
4836        i_gid = i_gid_read(inode);
4837        i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4838        if (!(test_opt(inode->i_sb, NO_UID32))) {
4839                raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4840                raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4841/*
4842 * Fix up interoperability with old kernels. Otherwise, old inodes get
4843 * re-used with the upper 16 bits of the uid/gid intact
4844 */
4845                if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4846                        raw_inode->i_uid_high = 0;
4847                        raw_inode->i_gid_high = 0;
4848                } else {
4849                        raw_inode->i_uid_high =
4850                                cpu_to_le16(high_16_bits(i_uid));
4851                        raw_inode->i_gid_high =
4852                                cpu_to_le16(high_16_bits(i_gid));
4853                }
4854        } else {
4855                raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4856                raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4857                raw_inode->i_uid_high = 0;
4858                raw_inode->i_gid_high = 0;
4859        }
4860        raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4861
4862        EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4863        EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4864        EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4865        EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4866
4867        err = ext4_inode_blocks_set(handle, raw_inode, ei);
4868        if (err) {
4869                spin_unlock(&ei->i_raw_lock);
4870                goto out_brelse;
4871        }
4872        raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4873        raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4874        if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4875                raw_inode->i_file_acl_high =
4876                        cpu_to_le16(ei->i_file_acl >> 32);
4877        raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4878        if (ei->i_disksize != ext4_isize(raw_inode)) {
4879                ext4_isize_set(raw_inode, ei->i_disksize);
4880                need_datasync = 1;
4881        }
4882        if (ei->i_disksize > 0x7fffffffULL) {
4883                if (!ext4_has_feature_large_file(sb) ||
4884                                EXT4_SB(sb)->s_es->s_rev_level ==
4885                    cpu_to_le32(EXT4_GOOD_OLD_REV))
4886                        set_large_file = 1;
4887        }
4888        raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4889        if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4890                if (old_valid_dev(inode->i_rdev)) {
4891                        raw_inode->i_block[0] =
4892                                cpu_to_le32(old_encode_dev(inode->i_rdev));
4893                        raw_inode->i_block[1] = 0;
4894                } else {
4895                        raw_inode->i_block[0] = 0;
4896                        raw_inode->i_block[1] =
4897                                cpu_to_le32(new_encode_dev(inode->i_rdev));
4898                        raw_inode->i_block[2] = 0;
4899                }
4900        } else if (!ext4_has_inline_data(inode)) {
4901                for (block = 0; block < EXT4_N_BLOCKS; block++)
4902                        raw_inode->i_block[block] = ei->i_data[block];
4903        }
4904
4905        if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4906                raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4907                if (ei->i_extra_isize) {
4908                        if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4909                                raw_inode->i_version_hi =
4910                                        cpu_to_le32(inode->i_version >> 32);
4911                        raw_inode->i_extra_isize =
4912                                cpu_to_le16(ei->i_extra_isize);
4913                }
4914        }
4915
4916        BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
4917               i_projid != EXT4_DEF_PROJID);
4918
4919        if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4920            EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4921                raw_inode->i_projid = cpu_to_le32(i_projid);
4922
4923        ext4_inode_csum_set(inode, raw_inode, ei);
4924        spin_unlock(&ei->i_raw_lock);
4925        if (inode->i_sb->s_flags & MS_LAZYTIME)
4926                ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4927                                              bh->b_data);
4928
4929        BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4930        rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4931        if (!err)
4932                err = rc;
4933        ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4934        if (set_large_file) {
4935                BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4936                err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4937                if (err)
4938                        goto out_brelse;
4939                ext4_update_dynamic_rev(sb);
4940                ext4_set_feature_large_file(sb);
4941                ext4_handle_sync(handle);
4942                err = ext4_handle_dirty_super(handle, sb);
4943        }
4944        ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4945out_brelse:
4946        brelse(bh);
4947        ext4_std_error(inode->i_sb, err);
4948        return err;
4949}
4950
4951/*
4952 * ext4_write_inode()
4953 *
4954 * We are called from a few places:
4955 *
4956 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4957 *   Here, there will be no transaction running. We wait for any running
4958 *   transaction to commit.
4959 *
4960 * - Within flush work (sys_sync(), kupdate and such).
4961 *   We wait on commit, if told to.
4962 *
4963 * - Within iput_final() -> write_inode_now()
4964 *   We wait on commit, if told to.
4965 *
4966 * In all cases it is actually safe for us to return without doing anything,
4967 * because the inode has been copied into a raw inode buffer in
4968 * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
4969 * writeback.
4970 *
4971 * Note that we are absolutely dependent upon all inode dirtiers doing the
4972 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4973 * which we are interested.
4974 *
4975 * It would be a bug for them to not do this.  The code:
4976 *
4977 *      mark_inode_dirty(inode)
4978 *      stuff();
4979 *      inode->i_size = expr;
4980 *
4981 * is in error because write_inode() could occur while `stuff()' is running,
4982 * and the new i_size will be lost.  Plus the inode will no longer be on the
4983 * superblock's dirty inode list.
4984 */
4985int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4986{
4987        int err;
4988
4989        if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4990                return 0;
4991
4992        if (EXT4_SB(inode->i_sb)->s_journal) {
4993                if (ext4_journal_current_handle()) {
4994                        jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4995                        dump_stack();
4996                        return -EIO;
4997                }
4998
4999                /*
5000                 * No need to force transaction in WB_SYNC_NONE mode. Also
5001                 * ext4_sync_fs() will force the commit after everything is
5002                 * written.
5003                 */
5004                if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5005                        return 0;
5006
5007                err = ext4_force_commit(inode->i_sb);
5008        } else {
5009                struct ext4_iloc iloc;
5010
5011                err = __ext4_get_inode_loc(inode, &iloc, 0);
5012                if (err)
5013                        return err;
5014                /*
5015                 * sync(2) will flush the whole buffer cache. No need to do
5016                 * it here separately for each inode.
5017                 */
5018                if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5019                        sync_dirty_buffer(iloc.bh);
5020                if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5021                        EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5022                                         "IO error syncing inode");
5023                        err = -EIO;
5024                }
5025                brelse(iloc.bh);
5026        }
5027        return err;
5028}
5029
5030/*
5031 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5032 * buffers that are attached to a page stradding i_size and are undergoing
5033 * commit. In that case we have to wait for commit to finish and try again.
5034 */
5035static void ext4_wait_for_tail_page_commit(struct inode *inode)
5036{
5037        struct page *page;
5038        unsigned offset;
5039        journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5040        tid_t commit_tid = 0;
5041        int ret;
5042
5043        offset = inode->i_size & (PAGE_SIZE - 1);
5044        /*
5045         * All buffers in the last page remain valid? Then there's nothing to
5046         * do. We do the check mainly to optimize the common PAGE_SIZE ==
5047         * blocksize case
5048         */
5049        if (offset > PAGE_SIZE - (1 << inode->i_blkbits))
5050                return;
5051        while (1) {
5052                page = find_lock_page(inode->i_mapping,
5053                                      inode->i_size >> PAGE_SHIFT);
5054                if (!page)
5055                        return;
5056                ret = __ext4_journalled_invalidatepage(page, offset,
5057                                                PAGE_SIZE - offset);
5058                unlock_page(page);
5059                put_page(page);
5060                if (ret != -EBUSY)
5061                        return;
5062                commit_tid = 0;
5063                read_lock(&journal->j_state_lock);
5064                if (journal->j_committing_transaction)
5065                        commit_tid = journal->j_committing_transaction->t_tid;
5066                read_unlock(&journal->j_state_lock);
5067                if (commit_tid)
5068                        jbd2_log_wait_commit(journal, commit_tid);
5069        }
5070}
5071
5072/*
5073 * ext4_setattr()
5074 *
5075 * Called from notify_change.
5076 *
5077 * We want to trap VFS attempts to truncate the file as soon as
5078 * possible.  In particular, we want to make sure that when the VFS
5079 * shrinks i_size, we put the inode on the orphan list and modify
5080 * i_disksize immediately, so that during the subsequent flushing of
5081 * dirty pages and freeing of disk blocks, we can guarantee that any
5082 * commit will leave the blocks being flushed in an unused state on
5083 * disk.  (On recovery, the inode will get truncated and the blocks will
5084 * be freed, so we have a strong guarantee that no future commit will
5085 * leave these blocks visible to the user.)
5086 *
5087 * Another thing we have to assure is that if we are in ordered mode
5088 * and inode is still attached to the committing transaction, we must
5089 * we start writeout of all the dirty pages which are being truncated.
5090 * This way we are sure that all the data written in the previous
5091 * transaction are already on disk (truncate waits for pages under
5092 * writeback).
5093 *
5094 * Called with inode->i_mutex down.
5095 */
5096int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5097{
5098        struct inode *inode = d_inode(dentry);
5099        int error, rc = 0;
5100        int orphan = 0;
5101        const unsigned int ia_valid = attr->ia_valid;
5102
5103        error = setattr_prepare(dentry, attr);
5104        if (error)
5105                return error;
5106
5107        if (is_quota_modification(inode, attr)) {
5108                error = dquot_initialize(inode);
5109                if (error)
5110                        return error;
5111        }
5112        if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5113            (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5114                handle_t *handle;
5115
5116                /* (user+group)*(old+new) structure, inode write (sb,
5117                 * inode block, ? - but truncate inode update has it) */
5118                handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5119                        (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5120                         EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5121                if (IS_ERR(handle)) {
5122                        error = PTR_ERR(handle);
5123                        goto err_out;
5124                }
5125                error = dquot_transfer(inode, attr);
5126                if (error) {
5127                        ext4_journal_stop(handle);
5128                        return error;
5129                }
5130                /* Update corresponding info in inode so that everything is in
5131                 * one transaction */
5132                if (attr->ia_valid & ATTR_UID)
5133                        inode->i_uid = attr->ia_uid;
5134                if (attr->ia_valid & ATTR_GID)
5135                        inode->i_gid = attr->ia_gid;
5136                error = ext4_mark_inode_dirty(handle, inode);
5137                ext4_journal_stop(handle);
5138        }
5139
5140        if (attr->ia_valid & ATTR_SIZE) {
5141                handle_t *handle;
5142                loff_t oldsize = inode->i_size;
5143                int shrink = (attr->ia_size <= inode->i_size);
5144
5145                if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5146                        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5147
5148                        if (attr->ia_size > sbi->s_bitmap_maxbytes)
5149                                return -EFBIG;
5150                }
5151                if (!S_ISREG(inode->i_mode))
5152                        return -EINVAL;
5153
5154                if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5155                        inode_inc_iversion(inode);
5156
5157                if (ext4_should_order_data(inode) &&
5158                    (attr->ia_size < inode->i_size)) {
5159                        error = ext4_begin_ordered_truncate(inode,
5160                                                            attr->ia_size);
5161                        if (error)
5162                                goto err_out;
5163                }
5164                if (attr->ia_size != inode->i_size) {
5165                        handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5166                        if (IS_ERR(handle)) {
5167                                error = PTR_ERR(handle);
5168                                goto err_out;
5169                        }
5170                        if (ext4_handle_valid(handle) && shrink) {
5171                                error = ext4_orphan_add(handle, inode);
5172                                orphan = 1;
5173                        }
5174                        /*
5175                         * Update c/mtime on truncate up, ext4_truncate() will
5176                         * update c/mtime in shrink case below
5177                         */
5178                        if (!shrink) {
5179                                inode->i_mtime = ext4_current_time(inode);
5180                                inode->i_ctime = inode->i_mtime;
5181                        }
5182                        down_write(&EXT4_I(inode)->i_data_sem);
5183                        EXT4_I(inode)->i_disksize = attr->ia_size;
5184                        rc = ext4_mark_inode_dirty(handle, inode);
5185                        if (!error)
5186                                error = rc;
5187                        /*
5188                         * We have to update i_size under i_data_sem together
5189                         * with i_disksize to avoid races with writeback code
5190                         * running ext4_wb_update_i_disksize().
5191                         */
5192                        if (!error)
5193                                i_size_write(inode, attr->ia_size);
5194                        up_write(&EXT4_I(inode)->i_data_sem);
5195                        ext4_journal_stop(handle);
5196                        if (error) {
5197                                if (orphan)
5198                                        ext4_orphan_del(NULL, inode);
5199                                goto err_out;
5200                        }
5201                }
5202                if (!shrink)
5203                        pagecache_isize_extended(inode, oldsize, inode->i_size);
5204
5205                /*
5206                 * Blocks are going to be removed from the inode. Wait
5207                 * for dio in flight.  Temporarily disable
5208                 * dioread_nolock to prevent livelock.
5209                 */
5210                if (orphan) {
5211                        if (!ext4_should_journal_data(inode)) {
5212                                ext4_inode_block_unlocked_dio(inode);
5213                                inode_dio_wait(inode);
5214                                ext4_inode_resume_unlocked_dio(inode);
5215                        } else
5216                                ext4_wait_for_tail_page_commit(inode);
5217                }
5218                down_write(&EXT4_I(inode)->i_mmap_sem);
5219                /*
5220                 * Truncate pagecache after we've waited for commit
5221                 * in data=journal mode to make pages freeable.
5222                 */
5223                truncate_pagecache(inode, inode->i_size);
5224                if (shrink)
5225                        ext4_truncate(inode);
5226                up_write(&EXT4_I(inode)->i_mmap_sem);
5227        }
5228
5229        if (!rc) {
5230                setattr_copy(inode, attr);
5231                mark_inode_dirty(inode);
5232        }
5233
5234        /*
5235         * If the call to ext4_truncate failed to get a transaction handle at
5236         * all, we need to clean up the in-core orphan list manually.
5237         */
5238        if (orphan && inode->i_nlink)
5239                ext4_orphan_del(NULL, inode);
5240
5241        if (!rc && (ia_valid & ATTR_MODE))
5242                rc = posix_acl_chmod(inode, inode->i_mode);
5243
5244err_out:
5245        ext4_std_error(inode->i_sb, error);
5246        if (!error)
5247                error = rc;
5248        return error;
5249}
5250
5251int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5252                 struct kstat *stat)
5253{
5254        struct inode *inode;
5255        unsigned long long delalloc_blocks;
5256
5257        inode = d_inode(dentry);
5258        generic_fillattr(inode, stat);
5259
5260        /*
5261         * If there is inline data in the inode, the inode will normally not
5262         * have data blocks allocated (it may have an external xattr block).
5263         * Report at least one sector for such files, so tools like tar, rsync,
5264         * others doen't incorrectly think the file is completely sparse.
5265         */
5266        if (unlikely(ext4_has_inline_data(inode)))
5267                stat->blocks += (stat->size + 511) >> 9;
5268
5269        /*
5270         * We can't update i_blocks if the block allocation is delayed
5271         * otherwise in the case of system crash before the real block
5272         * allocation is done, we will have i_blocks inconsistent with
5273         * on-disk file blocks.
5274         * We always keep i_blocks updated together with real
5275         * allocation. But to not confuse with user, stat
5276         * will return the blocks that include the delayed allocation
5277         * blocks for this file.
5278         */
5279        delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5280                                   EXT4_I(inode)->i_reserved_data_blocks);
5281        stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5282        return 0;
5283}
5284
5285static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5286                                   int pextents)
5287{
5288        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5289                return ext4_ind_trans_blocks(inode, lblocks);
5290        return ext4_ext_index_trans_blocks(inode, pextents);
5291}
5292
5293/*
5294 * Account for index blocks, block groups bitmaps and block group
5295 * descriptor blocks if modify datablocks and index blocks
5296 * worse case, the indexs blocks spread over different block groups
5297 *
5298 * If datablocks are discontiguous, they are possible to spread over
5299 * different block groups too. If they are contiguous, with flexbg,
5300 * they could still across block group boundary.
5301 *
5302 * Also account for superblock, inode, quota and xattr blocks
5303 */
5304static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5305                                  int pextents)
5306{
5307        ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5308        int gdpblocks;
5309        int idxblocks;
5310        int ret = 0;
5311
5312        /*
5313         * How many index blocks need to touch to map @lblocks logical blocks
5314         * to @pextents physical extents?
5315         */
5316        idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5317
5318        ret = idxblocks;
5319
5320        /*
5321         * Now let's see how many group bitmaps and group descriptors need
5322         * to account
5323         */
5324        groups = idxblocks + pextents;
5325        gdpblocks = groups;
5326        if (groups > ngroups)
5327                groups = ngroups;
5328        if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5329                gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5330
5331        /* bitmaps and block group descriptor blocks */
5332        ret += groups + gdpblocks;
5333
5334        /* Blocks for super block, inode, quota and xattr blocks */
5335        ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5336
5337        return ret;
5338}
5339
5340/*
5341 * Calculate the total number of credits to reserve to fit
5342 * the modification of a single pages into a single transaction,
5343 * which may include multiple chunks of block allocations.
5344 *
5345 * This could be called via ext4_write_begin()
5346 *
5347 * We need to consider the worse case, when
5348 * one new block per extent.
5349 */
5350int ext4_writepage_trans_blocks(struct inode *inode)
5351{
5352        int bpp = ext4_journal_blocks_per_page(inode);
5353        int ret;
5354
5355        ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5356
5357        /* Account for data blocks for journalled mode */
5358        if (ext4_should_journal_data(inode))
5359                ret += bpp;
5360        return ret;
5361}
5362
5363/*
5364 * Calculate the journal credits for a chunk of data modification.
5365 *
5366 * This is called from DIO, fallocate or whoever calling
5367 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5368 *
5369 * journal buffers for data blocks are not included here, as DIO
5370 * and fallocate do no need to journal data buffers.
5371 */
5372int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5373{
5374        return ext4_meta_trans_blocks(inode, nrblocks, 1);
5375}
5376
5377/*
5378 * The caller must have previously called ext4_reserve_inode_write().
5379 * Give this, we know that the caller already has write access to iloc->bh.
5380 */
5381int ext4_mark_iloc_dirty(handle_t *handle,
5382                         struct inode *inode, struct ext4_iloc *iloc)
5383{
5384        int err = 0;
5385
5386        if (IS_I_VERSION(inode))
5387                inode_inc_iversion(inode);
5388
5389        /* the do_update_inode consumes one bh->b_count */
5390        get_bh(iloc->bh);
5391
5392        /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5393        err = ext4_do_update_inode(handle, inode, iloc);
5394        put_bh(iloc->bh);
5395        return err;
5396}
5397
5398/*
5399 * On success, We end up with an outstanding reference count against
5400 * iloc->bh.  This _must_ be cleaned up later.
5401 */
5402
5403int
5404ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5405                         struct ext4_iloc *iloc)
5406{
5407        int err;
5408
5409        err = ext4_get_inode_loc(inode, iloc);
5410        if (!err) {
5411                BUFFER_TRACE(iloc->bh, "get_write_access");
5412                err = ext4_journal_get_write_access(handle, iloc->bh);
5413                if (err) {
5414                        brelse(iloc->bh);
5415                        iloc->bh = NULL;
5416                }
5417        }
5418        ext4_std_error(inode->i_sb, err);
5419        return err;
5420}
5421
5422/*
5423 * Expand an inode by new_extra_isize bytes.
5424 * Returns 0 on success or negative error number on failure.
5425 */
5426static int ext4_expand_extra_isize(struct inode *inode,
5427                                   unsigned int new_extra_isize,
5428                                   struct ext4_iloc iloc,
5429                                   handle_t *handle)
5430{
5431        struct ext4_inode *raw_inode;
5432        struct ext4_xattr_ibody_header *header;
5433
5434        if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5435                return 0;
5436
5437        raw_inode = ext4_raw_inode(&iloc);
5438
5439        header = IHDR(inode, raw_inode);
5440
5441        /* No extended attributes present */
5442        if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5443            header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5444                memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5445                        new_extra_isize);
5446                EXT4_I(inode)->i_extra_isize = new_extra_isize;
5447                return 0;
5448        }
5449
5450        /* try to expand with EAs present */
5451        return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5452                                          raw_inode, handle);
5453}
5454
5455/*
5456 * What we do here is to mark the in-core inode as clean with respect to inode
5457 * dirtiness (it may still be data-dirty).
5458 * This means that the in-core inode may be reaped by prune_icache
5459 * without having to perform any I/O.  This is a very good thing,
5460 * because *any* task may call prune_icache - even ones which
5461 * have a transaction open against a different journal.
5462 *
5463 * Is this cheating?  Not really.  Sure, we haven't written the
5464 * inode out, but prune_icache isn't a user-visible syncing function.
5465 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5466 * we start and wait on commits.
5467 */
5468int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5469{
5470        struct ext4_iloc iloc;
5471        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5472        static unsigned int mnt_count;
5473        int err, ret;
5474
5475        might_sleep();
5476        trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5477        err = ext4_reserve_inode_write(handle, inode, &iloc);
5478        if (err)
5479                return err;
5480        if (ext4_handle_valid(handle) &&
5481            EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5482            !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5483                /*
5484                 * We need extra buffer credits since we may write into EA block
5485                 * with this same handle. If journal_extend fails, then it will
5486                 * only result in a minor loss of functionality for that inode.
5487                 * If this is felt to be critical, then e2fsck should be run to
5488                 * force a large enough s_min_extra_isize.
5489                 */
5490                if ((jbd2_journal_extend(handle,
5491                             EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5492                        ret = ext4_expand_extra_isize(inode,
5493                                                      sbi->s_want_extra_isize,
5494                                                      iloc, handle);
5495                        if (ret) {
5496                                if (mnt_count !=
5497                                        le16_to_cpu(sbi->s_es->s_mnt_count)) {
5498                                        ext4_warning(inode->i_sb,
5499                                        "Unable to expand inode %lu. Delete"
5500                                        " some EAs or run e2fsck.",
5501                                        inode->i_ino);
5502                                        mnt_count =
5503                                          le16_to_cpu(sbi->s_es->s_mnt_count);
5504                                }
5505                        }
5506                }
5507        }
5508        return ext4_mark_iloc_dirty(handle, inode, &iloc);
5509}
5510
5511/*
5512 * ext4_dirty_inode() is called from __mark_inode_dirty()
5513 *
5514 * We're really interested in the case where a file is being extended.
5515 * i_size has been changed by generic_commit_write() and we thus need
5516 * to include the updated inode in the current transaction.
5517 *
5518 * Also, dquot_alloc_block() will always dirty the inode when blocks
5519 * are allocated to the file.
5520 *
5521 * If the inode is marked synchronous, we don't honour that here - doing
5522 * so would cause a commit on atime updates, which we don't bother doing.
5523 * We handle synchronous inodes at the highest possible level.
5524 *
5525 * If only the I_DIRTY_TIME flag is set, we can skip everything.  If
5526 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5527 * to copy into the on-disk inode structure are the timestamp files.
5528 */
5529void ext4_dirty_inode(struct inode *inode, int flags)
5530{
5531        handle_t *handle;
5532
5533        if (flags == I_DIRTY_TIME)
5534                return;
5535        handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5536        if (IS_ERR(handle))
5537                goto out;
5538
5539        ext4_mark_inode_dirty(handle, inode);
5540
5541        ext4_journal_stop(handle);
5542out:
5543        return;
5544}
5545
5546#if 0
5547/*
5548 * Bind an inode's backing buffer_head into this transaction, to prevent
5549 * it from being flushed to disk early.  Unlike
5550 * ext4_reserve_inode_write, this leaves behind no bh reference and
5551 * returns no iloc structure, so the caller needs to repeat the iloc
5552 * lookup to mark the inode dirty later.
5553 */
5554static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5555{
5556        struct ext4_iloc iloc;
5557
5558        int err = 0;
5559        if (handle) {
5560                err = ext4_get_inode_loc(inode, &iloc);
5561                if (!err) {
5562                        BUFFER_TRACE(iloc.bh, "get_write_access");
5563                        err = jbd2_journal_get_write_access(handle, iloc.bh);
5564                        if (!err)
5565                                err = ext4_handle_dirty_metadata(handle,
5566                                                                 NULL,
5567                                                                 iloc.bh);
5568                        brelse(iloc.bh);
5569                }
5570        }
5571        ext4_std_error(inode->i_sb, err);
5572        return err;
5573}
5574#endif
5575
5576int ext4_change_inode_journal_flag(struct inode *inode, int val)
5577{
5578        journal_t *journal;
5579        handle_t *handle;
5580        int err;
5581        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5582
5583        /*
5584         * We have to be very careful here: changing a data block's
5585         * journaling status dynamically is dangerous.  If we write a
5586         * data block to the journal, change the status and then delete
5587         * that block, we risk forgetting to revoke the old log record
5588         * from the journal and so a subsequent replay can corrupt data.
5589         * So, first we make sure that the journal is empty and that
5590         * nobody is changing anything.
5591         */
5592
5593        journal = EXT4_JOURNAL(inode);
5594        if (!journal)
5595                return 0;
5596        if (is_journal_aborted(journal))
5597                return -EROFS;
5598
5599        /* Wait for all existing dio workers */
5600        ext4_inode_block_unlocked_dio(inode);
5601        inode_dio_wait(inode);
5602
5603        /*
5604         * Before flushing the journal and switching inode's aops, we have
5605         * to flush all dirty data the inode has. There can be outstanding
5606         * delayed allocations, there can be unwritten extents created by
5607         * fallocate or buffered writes in dioread_nolock mode covered by
5608         * dirty data which can be converted only after flushing the dirty
5609         * data (and journalled aops don't know how to handle these cases).
5610         */
5611        if (val) {
5612                down_write(&EXT4_I(inode)->i_mmap_sem);
5613                err = filemap_write_and_wait(inode->i_mapping);
5614                if (err < 0) {
5615                        up_write(&EXT4_I(inode)->i_mmap_sem);
5616                        ext4_inode_resume_unlocked_dio(inode);
5617                        return err;
5618                }
5619        }
5620
5621        percpu_down_write(&sbi->s_journal_flag_rwsem);
5622        jbd2_journal_lock_updates(journal);
5623
5624        /*
5625         * OK, there are no updates running now, and all cached data is
5626         * synced to disk.  We are now in a completely consistent state
5627         * which doesn't have anything in the journal, and we know that
5628         * no filesystem updates are running, so it is safe to modify
5629         * the inode's in-core data-journaling state flag now.
5630         */
5631
5632        if (val)
5633                ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5634        else {
5635                err = jbd2_journal_flush(journal);
5636                if (err < 0) {
5637                        jbd2_journal_unlock_updates(journal);
5638                        percpu_up_write(&sbi->s_journal_flag_rwsem);
5639                        ext4_inode_resume_unlocked_dio(inode);
5640                        return err;
5641                }
5642                ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5643        }
5644        ext4_set_aops(inode);
5645
5646        jbd2_journal_unlock_updates(journal);
5647        percpu_up_write(&sbi->s_journal_flag_rwsem);
5648
5649        if (val)
5650                up_write(&EXT4_I(inode)->i_mmap_sem);
5651        ext4_inode_resume_unlocked_dio(inode);
5652
5653        /* Finally we can mark the inode as dirty. */
5654
5655        handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5656        if (IS_ERR(handle))
5657                return PTR_ERR(handle);
5658
5659        err = ext4_mark_inode_dirty(handle, inode);
5660        ext4_handle_sync(handle);
5661        ext4_journal_stop(handle);
5662        ext4_std_error(inode->i_sb, err);
5663
5664        return err;
5665}
5666
5667static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5668{
5669        return !buffer_mapped(bh);
5670}
5671
5672int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5673{
5674        struct page *page = vmf->page;
5675        loff_t size;
5676        unsigned long len;
5677        int ret;
5678        struct file *file = vma->vm_file;
5679        struct inode *inode = file_inode(file);
5680        struct address_space *mapping = inode->i_mapping;
5681        handle_t *handle;
5682        get_block_t *get_block;
5683        int retries = 0;
5684
5685        sb_start_pagefault(inode->i_sb);
5686        file_update_time(vma->vm_file);
5687
5688        down_read(&EXT4_I(inode)->i_mmap_sem);
5689        /* Delalloc case is easy... */
5690        if (test_opt(inode->i_sb, DELALLOC) &&
5691            !ext4_should_journal_data(inode) &&
5692            !ext4_nonda_switch(inode->i_sb)) {
5693                do {
5694                        ret = block_page_mkwrite(vma, vmf,
5695                                                   ext4_da_get_block_prep);
5696                } while (ret == -ENOSPC &&
5697                       ext4_should_retry_alloc(inode->i_sb, &retries));
5698                goto out_ret;
5699        }
5700
5701        lock_page(page);
5702        size = i_size_read(inode);
5703        /* Page got truncated from under us? */
5704        if (page->mapping != mapping || page_offset(page) > size) {
5705                unlock_page(page);
5706                ret = VM_FAULT_NOPAGE;
5707                goto out;
5708        }
5709
5710        if (page->index == size >> PAGE_SHIFT)
5711                len = size & ~PAGE_MASK;
5712        else
5713                len = PAGE_SIZE;
5714        /*
5715         * Return if we have all the buffers mapped. This avoids the need to do
5716         * journal_start/journal_stop which can block and take a long time
5717         */
5718        if (page_has_buffers(page)) {
5719                if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5720                                            0, len, NULL,
5721                                            ext4_bh_unmapped)) {
5722                        /* Wait so that we don't change page under IO */
5723                        wait_for_stable_page(page);
5724                        ret = VM_FAULT_LOCKED;
5725                        goto out;
5726                }
5727        }
5728        unlock_page(page);
5729        /* OK, we need to fill the hole... */
5730        if (ext4_should_dioread_nolock(inode))
5731                get_block = ext4_get_block_unwritten;
5732        else
5733                get_block = ext4_get_block;
5734retry_alloc:
5735        handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5736                                    ext4_writepage_trans_blocks(inode));
5737        if (IS_ERR(handle)) {
5738                ret = VM_FAULT_SIGBUS;
5739                goto out;
5740        }
5741        ret = block_page_mkwrite(vma, vmf, get_block);
5742        if (!ret && ext4_should_journal_data(inode)) {
5743                if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5744                          PAGE_SIZE, NULL, do_journal_get_write_access)) {
5745                        unlock_page(page);
5746                        ret = VM_FAULT_SIGBUS;
5747                        ext4_journal_stop(handle);
5748                        goto out;
5749                }
5750                ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5751        }
5752        ext4_journal_stop(handle);
5753        if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5754                goto retry_alloc;
5755out_ret:
5756        ret = block_page_mkwrite_return(ret);
5757out:
5758        up_read(&EXT4_I(inode)->i_mmap_sem);
5759        sb_end_pagefault(inode->i_sb);
5760        return ret;
5761}
5762
5763int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5764{
5765        struct inode *inode = file_inode(vma->vm_file);
5766        int err;
5767
5768        down_read(&EXT4_I(inode)->i_mmap_sem);
5769        err = filemap_fault(vma, vmf);
5770        up_read(&EXT4_I(inode)->i_mmap_sem);
5771
5772        return err;
5773}
5774
5775/*
5776 * Find the first extent at or after @lblk in an inode that is not a hole.
5777 * Search for @map_len blocks at most. The extent is returned in @result.
5778 *
5779 * The function returns 1 if we found an extent. The function returns 0 in
5780 * case there is no extent at or after @lblk and in that case also sets
5781 * @result->es_len to 0. In case of error, the error code is returned.
5782 */
5783int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
5784                         unsigned int map_len, struct extent_status *result)
5785{
5786        struct ext4_map_blocks map;
5787        struct extent_status es = {};
5788        int ret;
5789
5790        map.m_lblk = lblk;
5791        map.m_len = map_len;
5792
5793        /*
5794         * For non-extent based files this loop may iterate several times since
5795         * we do not determine full hole size.
5796         */
5797        while (map.m_len > 0) {
5798                ret = ext4_map_blocks(NULL, inode, &map, 0);
5799                if (ret < 0)
5800                        return ret;
5801                /* There's extent covering m_lblk? Just return it. */
5802                if (ret > 0) {
5803                        int status;
5804
5805                        ext4_es_store_pblock(result, map.m_pblk);
5806                        result->es_lblk = map.m_lblk;
5807                        result->es_len = map.m_len;
5808                        if (map.m_flags & EXT4_MAP_UNWRITTEN)
5809                                status = EXTENT_STATUS_UNWRITTEN;
5810                        else
5811                                status = EXTENT_STATUS_WRITTEN;
5812                        ext4_es_store_status(result, status);
5813                        return 1;
5814                }
5815                ext4_es_find_delayed_extent_range(inode, map.m_lblk,
5816                                                  map.m_lblk + map.m_len - 1,
5817                                                  &es);
5818                /* Is delalloc data before next block in extent tree? */
5819                if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
5820                        ext4_lblk_t offset = 0;
5821
5822                        if (es.es_lblk < lblk)
5823                                offset = lblk - es.es_lblk;
5824                        result->es_lblk = es.es_lblk + offset;
5825                        ext4_es_store_pblock(result,
5826                                             ext4_es_pblock(&es) + offset);
5827                        result->es_len = es.es_len - offset;
5828                        ext4_es_store_status(result, ext4_es_status(&es));
5829
5830                        return 1;
5831                }
5832                /* There's a hole at m_lblk, advance us after it */
5833                map.m_lblk += map.m_len;
5834                map_len -= map.m_len;
5835                map.m_len = map_len;
5836                cond_resched();
5837        }
5838        result->es_len = 0;
5839        return 0;
5840}
Note: See TracBrowser for help on using the repository browser.