aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ext4
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2010-05-27 13:26:37 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2010-05-27 13:26:37 -0400
commite4ce30f3779c2ddaa7dfaa4042209e5dbacbada5 (patch)
treecc64c1dcd16b5dbf71ebc8338b339e6fb04abaee /fs/ext4
parentb899ebeb05da4287ce845976727e3e83dadd25d5 (diff)
parent14ece1028b3ed53ffec1b1213ffc6acaf79ad77c (diff)
Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (40 commits) ext4: Make fsync sync new parent directories in no-journal mode ext4: Drop whitespace at end of lines ext4: Fix compat EXT4_IOC_ADD_GROUP ext4: Conditionally define compat ioctl numbers tracing: Convert more ext4 events to DEFINE_EVENT ext4: Add new tracepoints to track mballoc's buddy bitmap loads ext4: Add a missing trace hook ext4: restart ext4_ext_remove_space() after transaction restart ext4: Clear the EXT4_EOFBLOCKS_FL flag only when warranted ext4: Avoid crashing on NULL ptr dereference on a filesystem error ext4: Use bitops to read/modify i_flags in struct ext4_inode_info ext4: Convert calls of ext4_error() to EXT4_ERROR_INODE() ext4: Convert callers of ext4_get_blocks() to use ext4_map_blocks() ext4: Add new abstraction ext4_map_blocks() underneath ext4_get_blocks() ext4: Use our own write_cache_pages() ext4: Show journal_checksum option ext4: Fix for ext4_mb_collect_stats() ext4: check for a good block group before loading buddy pages ext4: Prevent creation of files larger than RLIMIT_FSIZE using fallocate ext4: Remove extraneous newlines in ext4_msg() calls ... Fixed up trivial conflict in fs/ext4/fsync.c
Diffstat (limited to 'fs/ext4')
-rw-r--r--fs/ext4/balloc.c5
-rw-r--r--fs/ext4/block_validity.c4
-rw-r--r--fs/ext4/dir.c26
-rw-r--r--fs/ext4/ext4.h167
-rw-r--r--fs/ext4/ext4_jbd2.h8
-rw-r--r--fs/ext4/extents.c417
-rw-r--r--fs/ext4/file.c2
-rw-r--r--fs/ext4/fsync.c35
-rw-r--r--fs/ext4/ialloc.c89
-rw-r--r--fs/ext4/inode.c723
-rw-r--r--fs/ext4/ioctl.c27
-rw-r--r--fs/ext4/mballoc.c120
-rw-r--r--fs/ext4/migrate.c2
-rw-r--r--fs/ext4/move_extent.c13
-rw-r--r--fs/ext4/namei.c61
-rw-r--r--fs/ext4/resize.c3
-rw-r--r--fs/ext4/super.c80
-rw-r--r--fs/ext4/symlink.c2
-rw-r--r--fs/ext4/xattr.c39
19 files changed, 1052 insertions, 771 deletions
diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index d2f37a5516c7..95b7594c76f9 100644
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -591,14 +591,15 @@ ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
591 ret = ext4_mb_new_blocks(handle, &ar, errp); 591 ret = ext4_mb_new_blocks(handle, &ar, errp);
592 if (count) 592 if (count)
593 *count = ar.len; 593 *count = ar.len;
594
595 /* 594 /*
596 * Account for the allocated meta blocks 595 * Account for the allocated meta blocks. We will never
596 * fail EDQUOT for metdata, but we do account for it.
597 */ 597 */
598 if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) { 598 if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
599 spin_lock(&EXT4_I(inode)->i_block_reservation_lock); 599 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
600 EXT4_I(inode)->i_allocated_meta_blocks += ar.len; 600 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
601 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 601 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
602 dquot_alloc_block_nofail(inode, ar.len);
602 } 603 }
603 return ret; 604 return ret;
604} 605}
diff --git a/fs/ext4/block_validity.c b/fs/ext4/block_validity.c
index 538c48655084..5b6973fbf1bd 100644
--- a/fs/ext4/block_validity.c
+++ b/fs/ext4/block_validity.c
@@ -72,9 +72,9 @@ static int add_system_zone(struct ext4_sb_info *sbi,
72 else if (start_blk >= (entry->start_blk + entry->count)) 72 else if (start_blk >= (entry->start_blk + entry->count))
73 n = &(*n)->rb_right; 73 n = &(*n)->rb_right;
74 else { 74 else {
75 if (start_blk + count > (entry->start_blk + 75 if (start_blk + count > (entry->start_blk +
76 entry->count)) 76 entry->count))
77 entry->count = (start_blk + count - 77 entry->count = (start_blk + count -
78 entry->start_blk); 78 entry->start_blk);
79 new_node = *n; 79 new_node = *n;
80 new_entry = rb_entry(new_node, struct ext4_system_zone, 80 new_entry = rb_entry(new_node, struct ext4_system_zone,
diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
index 86cb6d86a048..ea5e6cb7e2a5 100644
--- a/fs/ext4/dir.c
+++ b/fs/ext4/dir.c
@@ -83,11 +83,10 @@ int ext4_check_dir_entry(const char *function, struct inode *dir,
83 error_msg = "inode out of bounds"; 83 error_msg = "inode out of bounds";
84 84
85 if (error_msg != NULL) 85 if (error_msg != NULL)
86 __ext4_error(dir->i_sb, function, 86 ext4_error_inode(function, dir,
87 "bad entry in directory #%lu: %s - block=%llu" 87 "bad entry in directory: %s - block=%llu"
88 "offset=%u(%u), inode=%u, rec_len=%d, name_len=%d", 88 "offset=%u(%u), inode=%u, rec_len=%d, name_len=%d",
89 dir->i_ino, error_msg, 89 error_msg, (unsigned long long) bh->b_blocknr,
90 (unsigned long long) bh->b_blocknr,
91 (unsigned) (offset%bh->b_size), offset, 90 (unsigned) (offset%bh->b_size), offset,
92 le32_to_cpu(de->inode), 91 le32_to_cpu(de->inode),
93 rlen, de->name_len); 92 rlen, de->name_len);
@@ -111,7 +110,7 @@ static int ext4_readdir(struct file *filp,
111 110
112 if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb, 111 if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
113 EXT4_FEATURE_COMPAT_DIR_INDEX) && 112 EXT4_FEATURE_COMPAT_DIR_INDEX) &&
114 ((EXT4_I(inode)->i_flags & EXT4_INDEX_FL) || 113 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
115 ((inode->i_size >> sb->s_blocksize_bits) == 1))) { 114 ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
116 err = ext4_dx_readdir(filp, dirent, filldir); 115 err = ext4_dx_readdir(filp, dirent, filldir);
117 if (err != ERR_BAD_DX_DIR) { 116 if (err != ERR_BAD_DX_DIR) {
@@ -122,20 +121,20 @@ static int ext4_readdir(struct file *filp,
122 * We don't set the inode dirty flag since it's not 121 * We don't set the inode dirty flag since it's not
123 * critical that it get flushed back to the disk. 122 * critical that it get flushed back to the disk.
124 */ 123 */
125 EXT4_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT4_INDEX_FL; 124 ext4_clear_inode_flag(filp->f_path.dentry->d_inode, EXT4_INODE_INDEX);
126 } 125 }
127 stored = 0; 126 stored = 0;
128 offset = filp->f_pos & (sb->s_blocksize - 1); 127 offset = filp->f_pos & (sb->s_blocksize - 1);
129 128
130 while (!error && !stored && filp->f_pos < inode->i_size) { 129 while (!error && !stored && filp->f_pos < inode->i_size) {
131 ext4_lblk_t blk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb); 130 struct ext4_map_blocks map;
132 struct buffer_head map_bh;
133 struct buffer_head *bh = NULL; 131 struct buffer_head *bh = NULL;
134 132
135 map_bh.b_state = 0; 133 map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
136 err = ext4_get_blocks(NULL, inode, blk, 1, &map_bh, 0); 134 map.m_len = 1;
135 err = ext4_map_blocks(NULL, inode, &map, 0);
137 if (err > 0) { 136 if (err > 0) {
138 pgoff_t index = map_bh.b_blocknr >> 137 pgoff_t index = map.m_pblk >>
139 (PAGE_CACHE_SHIFT - inode->i_blkbits); 138 (PAGE_CACHE_SHIFT - inode->i_blkbits);
140 if (!ra_has_index(&filp->f_ra, index)) 139 if (!ra_has_index(&filp->f_ra, index))
141 page_cache_sync_readahead( 140 page_cache_sync_readahead(
@@ -143,7 +142,7 @@ static int ext4_readdir(struct file *filp,
143 &filp->f_ra, filp, 142 &filp->f_ra, filp,
144 index, 1); 143 index, 1);
145 filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; 144 filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
146 bh = ext4_bread(NULL, inode, blk, 0, &err); 145 bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
147 } 146 }
148 147
149 /* 148 /*
@@ -152,9 +151,8 @@ static int ext4_readdir(struct file *filp,
152 */ 151 */
153 if (!bh) { 152 if (!bh) {
154 if (!dir_has_error) { 153 if (!dir_has_error) {
155 ext4_error(sb, "directory #%lu " 154 EXT4_ERROR_INODE(inode, "directory "
156 "contains a hole at offset %Lu", 155 "contains a hole at offset %Lu",
157 inode->i_ino,
158 (unsigned long long) filp->f_pos); 156 (unsigned long long) filp->f_pos);
159 dir_has_error = 1; 157 dir_has_error = 1;
160 } 158 }
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index bf938cf7c5f0..60bd31026e7c 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -29,6 +29,9 @@
29#include <linux/wait.h> 29#include <linux/wait.h>
30#include <linux/blockgroup_lock.h> 30#include <linux/blockgroup_lock.h>
31#include <linux/percpu_counter.h> 31#include <linux/percpu_counter.h>
32#ifdef __KERNEL__
33#include <linux/compat.h>
34#endif
32 35
33/* 36/*
34 * The fourth extended filesystem constants/structures 37 * The fourth extended filesystem constants/structures
@@ -54,10 +57,10 @@
54#endif 57#endif
55 58
56#define EXT4_ERROR_INODE(inode, fmt, a...) \ 59#define EXT4_ERROR_INODE(inode, fmt, a...) \
57 ext4_error_inode(__func__, (inode), (fmt), ## a); 60 ext4_error_inode(__func__, (inode), (fmt), ## a)
58 61
59#define EXT4_ERROR_FILE(file, fmt, a...) \ 62#define EXT4_ERROR_FILE(file, fmt, a...) \
60 ext4_error_file(__func__, (file), (fmt), ## a); 63 ext4_error_file(__func__, (file), (fmt), ## a)
61 64
62/* data type for block offset of block group */ 65/* data type for block offset of block group */
63typedef int ext4_grpblk_t; 66typedef int ext4_grpblk_t;
@@ -72,7 +75,7 @@ typedef __u32 ext4_lblk_t;
72typedef unsigned int ext4_group_t; 75typedef unsigned int ext4_group_t;
73 76
74/* 77/*
75 * Flags used in mballoc's allocation_context flags field. 78 * Flags used in mballoc's allocation_context flags field.
76 * 79 *
77 * Also used to show what's going on for debugging purposes when the 80 * Also used to show what's going on for debugging purposes when the
78 * flag field is exported via the traceport interface 81 * flag field is exported via the traceport interface
@@ -126,6 +129,29 @@ struct ext4_allocation_request {
126}; 129};
127 130
128/* 131/*
132 * Logical to physical block mapping, used by ext4_map_blocks()
133 *
134 * This structure is used to pass requests into ext4_map_blocks() as
135 * well as to store the information returned by ext4_map_blocks(). It
136 * takes less room on the stack than a struct buffer_head.
137 */
138#define EXT4_MAP_NEW (1 << BH_New)
139#define EXT4_MAP_MAPPED (1 << BH_Mapped)
140#define EXT4_MAP_UNWRITTEN (1 << BH_Unwritten)
141#define EXT4_MAP_BOUNDARY (1 << BH_Boundary)
142#define EXT4_MAP_UNINIT (1 << BH_Uninit)
143#define EXT4_MAP_FLAGS (EXT4_MAP_NEW | EXT4_MAP_MAPPED |\
144 EXT4_MAP_UNWRITTEN | EXT4_MAP_BOUNDARY |\
145 EXT4_MAP_UNINIT)
146
147struct ext4_map_blocks {
148 ext4_fsblk_t m_pblk;
149 ext4_lblk_t m_lblk;
150 unsigned int m_len;
151 unsigned int m_flags;
152};
153
154/*
129 * For delayed allocation tracking 155 * For delayed allocation tracking
130 */ 156 */
131struct mpage_da_data { 157struct mpage_da_data {
@@ -321,6 +347,83 @@ static inline __u32 ext4_mask_flags(umode_t mode, __u32 flags)
321 return flags & EXT4_OTHER_FLMASK; 347 return flags & EXT4_OTHER_FLMASK;
322} 348}
323 349
350/*
351 * Inode flags used for atomic set/get
352 */
353enum {
354 EXT4_INODE_SECRM = 0, /* Secure deletion */
355 EXT4_INODE_UNRM = 1, /* Undelete */
356 EXT4_INODE_COMPR = 2, /* Compress file */
357 EXT4_INODE_SYNC = 3, /* Synchronous updates */
358 EXT4_INODE_IMMUTABLE = 4, /* Immutable file */
359 EXT4_INODE_APPEND = 5, /* writes to file may only append */
360 EXT4_INODE_NODUMP = 6, /* do not dump file */
361 EXT4_INODE_NOATIME = 7, /* do not update atime */
362/* Reserved for compression usage... */
363 EXT4_INODE_DIRTY = 8,
364 EXT4_INODE_COMPRBLK = 9, /* One or more compressed clusters */
365 EXT4_INODE_NOCOMPR = 10, /* Don't compress */
366 EXT4_INODE_ECOMPR = 11, /* Compression error */
367/* End compression flags --- maybe not all used */
368 EXT4_INODE_INDEX = 12, /* hash-indexed directory */
369 EXT4_INODE_IMAGIC = 13, /* AFS directory */
370 EXT4_INODE_JOURNAL_DATA = 14, /* file data should be journaled */
371 EXT4_INODE_NOTAIL = 15, /* file tail should not be merged */
372 EXT4_INODE_DIRSYNC = 16, /* dirsync behaviour (directories only) */
373 EXT4_INODE_TOPDIR = 17, /* Top of directory hierarchies*/
374 EXT4_INODE_HUGE_FILE = 18, /* Set to each huge file */
375 EXT4_INODE_EXTENTS = 19, /* Inode uses extents */
376 EXT4_INODE_EA_INODE = 21, /* Inode used for large EA */
377 EXT4_INODE_EOFBLOCKS = 22, /* Blocks allocated beyond EOF */
378 EXT4_INODE_RESERVED = 31, /* reserved for ext4 lib */
379};
380
381#define TEST_FLAG_VALUE(FLAG) (EXT4_##FLAG##_FL == (1 << EXT4_INODE_##FLAG))
382#define CHECK_FLAG_VALUE(FLAG) if (!TEST_FLAG_VALUE(FLAG)) { \
383 printk(KERN_EMERG "EXT4 flag fail: " #FLAG ": %d %d\n", \
384 EXT4_##FLAG##_FL, EXT4_INODE_##FLAG); BUG_ON(1); }
385
386/*
387 * Since it's pretty easy to mix up bit numbers and hex values, and we
388 * can't do a compile-time test for ENUM values, we use a run-time
389 * test to make sure that EXT4_XXX_FL is consistent with respect to
390 * EXT4_INODE_XXX. If all is well the printk and BUG_ON will all drop
391 * out so it won't cost any extra space in the compiled kernel image.
392 * But it's important that these values are the same, since we are
393 * using EXT4_INODE_XXX to test for the flag values, but EXT4_XX_FL
394 * must be consistent with the values of FS_XXX_FL defined in
395 * include/linux/fs.h and the on-disk values found in ext2, ext3, and
396 * ext4 filesystems, and of course the values defined in e2fsprogs.
397 *
398 * It's not paranoia if the Murphy's Law really *is* out to get you. :-)
399 */
400static inline void ext4_check_flag_values(void)
401{
402 CHECK_FLAG_VALUE(SECRM);
403 CHECK_FLAG_VALUE(UNRM);
404 CHECK_FLAG_VALUE(COMPR);
405 CHECK_FLAG_VALUE(SYNC);
406 CHECK_FLAG_VALUE(IMMUTABLE);
407 CHECK_FLAG_VALUE(APPEND);
408 CHECK_FLAG_VALUE(NODUMP);
409 CHECK_FLAG_VALUE(NOATIME);
410 CHECK_FLAG_VALUE(DIRTY);
411 CHECK_FLAG_VALUE(COMPRBLK);
412 CHECK_FLAG_VALUE(NOCOMPR);
413 CHECK_FLAG_VALUE(ECOMPR);
414 CHECK_FLAG_VALUE(INDEX);
415 CHECK_FLAG_VALUE(IMAGIC);
416 CHECK_FLAG_VALUE(JOURNAL_DATA);
417 CHECK_FLAG_VALUE(NOTAIL);
418 CHECK_FLAG_VALUE(DIRSYNC);
419 CHECK_FLAG_VALUE(TOPDIR);
420 CHECK_FLAG_VALUE(HUGE_FILE);
421 CHECK_FLAG_VALUE(EXTENTS);
422 CHECK_FLAG_VALUE(EA_INODE);
423 CHECK_FLAG_VALUE(EOFBLOCKS);
424 CHECK_FLAG_VALUE(RESERVED);
425}
426
324/* Used to pass group descriptor data when online resize is done */ 427/* Used to pass group descriptor data when online resize is done */
325struct ext4_new_group_input { 428struct ext4_new_group_input {
326 __u32 group; /* Group number for this data */ 429 __u32 group; /* Group number for this data */
@@ -332,6 +435,18 @@ struct ext4_new_group_input {
332 __u16 unused; 435 __u16 unused;
333}; 436};
334 437
438#if defined(__KERNEL__) && defined(CONFIG_COMPAT)
439struct compat_ext4_new_group_input {
440 u32 group;
441 compat_u64 block_bitmap;
442 compat_u64 inode_bitmap;
443 compat_u64 inode_table;
444 u32 blocks_count;
445 u16 reserved_blocks;
446 u16 unused;
447};
448#endif
449
335/* The struct ext4_new_group_input in kernel space, with free_blocks_count */ 450/* The struct ext4_new_group_input in kernel space, with free_blocks_count */
336struct ext4_new_group_data { 451struct ext4_new_group_data {
337 __u32 group; 452 __u32 group;
@@ -355,7 +470,7 @@ struct ext4_new_group_data {
355#define EXT4_GET_BLOCKS_CREATE_UNINIT_EXT (EXT4_GET_BLOCKS_UNINIT_EXT|\ 470#define EXT4_GET_BLOCKS_CREATE_UNINIT_EXT (EXT4_GET_BLOCKS_UNINIT_EXT|\
356 EXT4_GET_BLOCKS_CREATE) 471 EXT4_GET_BLOCKS_CREATE)
357 /* Caller is from the delayed allocation writeout path, 472 /* Caller is from the delayed allocation writeout path,
358 so set the magic i_delalloc_reserve_flag after taking the 473 so set the magic i_delalloc_reserve_flag after taking the
359 inode allocation semaphore for */ 474 inode allocation semaphore for */
360#define EXT4_GET_BLOCKS_DELALLOC_RESERVE 0x0004 475#define EXT4_GET_BLOCKS_DELALLOC_RESERVE 0x0004
361 /* caller is from the direct IO path, request to creation of an 476 /* caller is from the direct IO path, request to creation of an
@@ -398,6 +513,7 @@ struct ext4_new_group_data {
398#define EXT4_IOC_ALLOC_DA_BLKS _IO('f', 12) 513#define EXT4_IOC_ALLOC_DA_BLKS _IO('f', 12)
399#define EXT4_IOC_MOVE_EXT _IOWR('f', 15, struct move_extent) 514#define EXT4_IOC_MOVE_EXT _IOWR('f', 15, struct move_extent)
400 515
516#if defined(__KERNEL__) && defined(CONFIG_COMPAT)
401/* 517/*
402 * ioctl commands in 32 bit emulation 518 * ioctl commands in 32 bit emulation
403 */ 519 */
@@ -408,11 +524,13 @@ struct ext4_new_group_data {
408#define EXT4_IOC32_GETRSVSZ _IOR('f', 5, int) 524#define EXT4_IOC32_GETRSVSZ _IOR('f', 5, int)
409#define EXT4_IOC32_SETRSVSZ _IOW('f', 6, int) 525#define EXT4_IOC32_SETRSVSZ _IOW('f', 6, int)
410#define EXT4_IOC32_GROUP_EXTEND _IOW('f', 7, unsigned int) 526#define EXT4_IOC32_GROUP_EXTEND _IOW('f', 7, unsigned int)
527#define EXT4_IOC32_GROUP_ADD _IOW('f', 8, struct compat_ext4_new_group_input)
411#ifdef CONFIG_JBD2_DEBUG 528#ifdef CONFIG_JBD2_DEBUG
412#define EXT4_IOC32_WAIT_FOR_READONLY _IOR('f', 99, int) 529#define EXT4_IOC32_WAIT_FOR_READONLY _IOR('f', 99, int)
413#endif 530#endif
414#define EXT4_IOC32_GETVERSION_OLD FS_IOC32_GETVERSION 531#define EXT4_IOC32_GETVERSION_OLD FS_IOC32_GETVERSION
415#define EXT4_IOC32_SETVERSION_OLD FS_IOC32_SETVERSION 532#define EXT4_IOC32_SETVERSION_OLD FS_IOC32_SETVERSION
533#endif
416 534
417 535
418/* 536/*
@@ -616,9 +734,8 @@ struct ext4_ext_cache {
616 */ 734 */
617struct ext4_inode_info { 735struct ext4_inode_info {
618 __le32 i_data[15]; /* unconverted */ 736 __le32 i_data[15]; /* unconverted */
619 __u32 i_flags;
620 ext4_fsblk_t i_file_acl;
621 __u32 i_dtime; 737 __u32 i_dtime;
738 ext4_fsblk_t i_file_acl;
622 739
623 /* 740 /*
624 * i_block_group is the number of the block group which contains 741 * i_block_group is the number of the block group which contains
@@ -629,6 +746,7 @@ struct ext4_inode_info {
629 */ 746 */
630 ext4_group_t i_block_group; 747 ext4_group_t i_block_group;
631 unsigned long i_state_flags; /* Dynamic state flags */ 748 unsigned long i_state_flags; /* Dynamic state flags */
749 unsigned long i_flags;
632 750
633 ext4_lblk_t i_dir_start_lookup; 751 ext4_lblk_t i_dir_start_lookup;
634#ifdef CONFIG_EXT4_FS_XATTR 752#ifdef CONFIG_EXT4_FS_XATTR
@@ -1062,22 +1180,25 @@ enum {
1062 EXT4_STATE_DA_ALLOC_CLOSE, /* Alloc DA blks on close */ 1180 EXT4_STATE_DA_ALLOC_CLOSE, /* Alloc DA blks on close */
1063 EXT4_STATE_EXT_MIGRATE, /* Inode is migrating */ 1181 EXT4_STATE_EXT_MIGRATE, /* Inode is migrating */
1064 EXT4_STATE_DIO_UNWRITTEN, /* need convert on dio done*/ 1182 EXT4_STATE_DIO_UNWRITTEN, /* need convert on dio done*/
1183 EXT4_STATE_NEWENTRY, /* File just added to dir */
1065}; 1184};
1066 1185
1067static inline int ext4_test_inode_state(struct inode *inode, int bit) 1186#define EXT4_INODE_BIT_FNS(name, field) \
1068{ 1187static inline int ext4_test_inode_##name(struct inode *inode, int bit) \
1069 return test_bit(bit, &EXT4_I(inode)->i_state_flags); 1188{ \
1070} 1189 return test_bit(bit, &EXT4_I(inode)->i_##field); \
1071 1190} \
1072static inline void ext4_set_inode_state(struct inode *inode, int bit) 1191static inline void ext4_set_inode_##name(struct inode *inode, int bit) \
1073{ 1192{ \
1074 set_bit(bit, &EXT4_I(inode)->i_state_flags); 1193 set_bit(bit, &EXT4_I(inode)->i_##field); \
1194} \
1195static inline void ext4_clear_inode_##name(struct inode *inode, int bit) \
1196{ \
1197 clear_bit(bit, &EXT4_I(inode)->i_##field); \
1075} 1198}
1076 1199
1077static inline void ext4_clear_inode_state(struct inode *inode, int bit) 1200EXT4_INODE_BIT_FNS(flag, flags)
1078{ 1201EXT4_INODE_BIT_FNS(state, state_flags)
1079 clear_bit(bit, &EXT4_I(inode)->i_state_flags);
1080}
1081#else 1202#else
1082/* Assume that user mode programs are passing in an ext4fs superblock, not 1203/* Assume that user mode programs are passing in an ext4fs superblock, not
1083 * a kernel struct super_block. This will allow us to call the feature-test 1204 * a kernel struct super_block. This will allow us to call the feature-test
@@ -1264,7 +1385,7 @@ struct ext4_dir_entry_2 {
1264 1385
1265#define is_dx(dir) (EXT4_HAS_COMPAT_FEATURE(dir->i_sb, \ 1386#define is_dx(dir) (EXT4_HAS_COMPAT_FEATURE(dir->i_sb, \
1266 EXT4_FEATURE_COMPAT_DIR_INDEX) && \ 1387 EXT4_FEATURE_COMPAT_DIR_INDEX) && \
1267 (EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) 1388 ext4_test_inode_flag((dir), EXT4_INODE_INDEX))
1268#define EXT4_DIR_LINK_MAX(dir) (!is_dx(dir) && (dir)->i_nlink >= EXT4_LINK_MAX) 1389#define EXT4_DIR_LINK_MAX(dir) (!is_dx(dir) && (dir)->i_nlink >= EXT4_LINK_MAX)
1269#define EXT4_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2 || (dir)->i_nlink == 1) 1390#define EXT4_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2 || (dir)->i_nlink == 1)
1270 1391
@@ -1678,6 +1799,7 @@ struct ext4_group_info {
1678 ext4_grpblk_t bb_first_free; /* first free block */ 1799 ext4_grpblk_t bb_first_free; /* first free block */
1679 ext4_grpblk_t bb_free; /* total free blocks */ 1800 ext4_grpblk_t bb_free; /* total free blocks */
1680 ext4_grpblk_t bb_fragments; /* nr of freespace fragments */ 1801 ext4_grpblk_t bb_fragments; /* nr of freespace fragments */
1802 ext4_grpblk_t bb_largest_free_order;/* order of largest frag in BG */
1681 struct list_head bb_prealloc_list; 1803 struct list_head bb_prealloc_list;
1682#ifdef DOUBLE_CHECK 1804#ifdef DOUBLE_CHECK
1683 void *bb_bitmap; 1805 void *bb_bitmap;
@@ -1772,9 +1894,8 @@ extern int ext4_ext_tree_init(handle_t *handle, struct inode *);
1772extern int ext4_ext_writepage_trans_blocks(struct inode *, int); 1894extern int ext4_ext_writepage_trans_blocks(struct inode *, int);
1773extern int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, 1895extern int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks,
1774 int chunk); 1896 int chunk);
1775extern int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, 1897extern int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
1776 ext4_lblk_t iblock, unsigned int max_blocks, 1898 struct ext4_map_blocks *map, int flags);
1777 struct buffer_head *bh_result, int flags);
1778extern void ext4_ext_truncate(struct inode *); 1899extern void ext4_ext_truncate(struct inode *);
1779extern void ext4_ext_init(struct super_block *); 1900extern void ext4_ext_init(struct super_block *);
1780extern void ext4_ext_release(struct super_block *); 1901extern void ext4_ext_release(struct super_block *);
@@ -1782,6 +1903,8 @@ extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset,
1782 loff_t len); 1903 loff_t len);
1783extern int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, 1904extern int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
1784 ssize_t len); 1905 ssize_t len);
1906extern int ext4_map_blocks(handle_t *handle, struct inode *inode,
1907 struct ext4_map_blocks *map, int flags);
1785extern int ext4_get_blocks(handle_t *handle, struct inode *inode, 1908extern int ext4_get_blocks(handle_t *handle, struct inode *inode,
1786 sector_t block, unsigned int max_blocks, 1909 sector_t block, unsigned int max_blocks,
1787 struct buffer_head *bh, int flags); 1910 struct buffer_head *bh, int flags);
diff --git a/fs/ext4/ext4_jbd2.h b/fs/ext4/ext4_jbd2.h
index b79ad5126468..dade0c024797 100644
--- a/fs/ext4/ext4_jbd2.h
+++ b/fs/ext4/ext4_jbd2.h
@@ -273,7 +273,7 @@ static inline int ext4_should_journal_data(struct inode *inode)
273 return 1; 273 return 1;
274 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) 274 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
275 return 1; 275 return 1;
276 if (EXT4_I(inode)->i_flags & EXT4_JOURNAL_DATA_FL) 276 if (ext4_test_inode_flag(inode, EXT4_INODE_JOURNAL_DATA))
277 return 1; 277 return 1;
278 return 0; 278 return 0;
279} 279}
@@ -284,7 +284,7 @@ static inline int ext4_should_order_data(struct inode *inode)
284 return 0; 284 return 0;
285 if (!S_ISREG(inode->i_mode)) 285 if (!S_ISREG(inode->i_mode))
286 return 0; 286 return 0;
287 if (EXT4_I(inode)->i_flags & EXT4_JOURNAL_DATA_FL) 287 if (ext4_test_inode_flag(inode, EXT4_INODE_JOURNAL_DATA))
288 return 0; 288 return 0;
289 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) 289 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
290 return 1; 290 return 1;
@@ -297,7 +297,7 @@ static inline int ext4_should_writeback_data(struct inode *inode)
297 return 0; 297 return 0;
298 if (EXT4_JOURNAL(inode) == NULL) 298 if (EXT4_JOURNAL(inode) == NULL)
299 return 1; 299 return 1;
300 if (EXT4_I(inode)->i_flags & EXT4_JOURNAL_DATA_FL) 300 if (ext4_test_inode_flag(inode, EXT4_INODE_JOURNAL_DATA))
301 return 0; 301 return 0;
302 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 302 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
303 return 1; 303 return 1;
@@ -321,7 +321,7 @@ static inline int ext4_should_dioread_nolock(struct inode *inode)
321 return 0; 321 return 0;
322 if (!S_ISREG(inode->i_mode)) 322 if (!S_ISREG(inode->i_mode))
323 return 0; 323 return 0;
324 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 324 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
325 return 0; 325 return 0;
326 if (ext4_should_journal_data(inode)) 326 if (ext4_should_journal_data(inode))
327 return 0; 327 return 0;
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 236b834b4ca8..377309c1af65 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -107,11 +107,8 @@ static int ext4_ext_truncate_extend_restart(handle_t *handle,
107 if (err <= 0) 107 if (err <= 0)
108 return err; 108 return err;
109 err = ext4_truncate_restart_trans(handle, inode, needed); 109 err = ext4_truncate_restart_trans(handle, inode, needed);
110 /* 110 if (err == 0)
111 * We have dropped i_data_sem so someone might have cached again 111 err = -EAGAIN;
112 * an extent we are going to truncate.
113 */
114 ext4_ext_invalidate_cache(inode);
115 112
116 return err; 113 return err;
117} 114}
@@ -185,10 +182,10 @@ static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
185 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { 182 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
186 /* 183 /*
187 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME 184 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
188 * block groups per flexgroup, reserve the first block 185 * block groups per flexgroup, reserve the first block
189 * group for directories and special files. Regular 186 * group for directories and special files. Regular
190 * files will start at the second block group. This 187 * files will start at the second block group. This
191 * tends to speed up directory access and improves 188 * tends to speed up directory access and improves
192 * fsck times. 189 * fsck times.
193 */ 190 */
194 block_group &= ~(flex_size-1); 191 block_group &= ~(flex_size-1);
@@ -439,10 +436,10 @@ static int __ext4_ext_check(const char *function, struct inode *inode,
439 return 0; 436 return 0;
440 437
441corrupted: 438corrupted:
442 __ext4_error(inode->i_sb, function, 439 ext4_error_inode(function, inode,
443 "bad header/extent in inode #%lu: %s - magic %x, " 440 "bad header/extent: %s - magic %x, "
444 "entries %u, max %u(%u), depth %u(%u)", 441 "entries %u, max %u(%u), depth %u(%u)",
445 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), 442 error_msg, le16_to_cpu(eh->eh_magic),
446 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), 443 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
447 max, le16_to_cpu(eh->eh_depth), depth); 444 max, le16_to_cpu(eh->eh_depth), depth);
448 445
@@ -1622,9 +1619,7 @@ int ext4_ext_try_to_merge(struct inode *inode,
1622 merge_done = 1; 1619 merge_done = 1;
1623 WARN_ON(eh->eh_entries == 0); 1620 WARN_ON(eh->eh_entries == 0);
1624 if (!eh->eh_entries) 1621 if (!eh->eh_entries)
1625 ext4_error(inode->i_sb, 1622 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1626 "inode#%lu, eh->eh_entries = 0!",
1627 inode->i_ino);
1628 } 1623 }
1629 1624
1630 return merge_done; 1625 return merge_done;
@@ -2039,7 +2034,7 @@ ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2039 struct ext4_ext_cache *cex; 2034 struct ext4_ext_cache *cex;
2040 int ret = EXT4_EXT_CACHE_NO; 2035 int ret = EXT4_EXT_CACHE_NO;
2041 2036
2042 /* 2037 /*
2043 * We borrow i_block_reservation_lock to protect i_cached_extent 2038 * We borrow i_block_reservation_lock to protect i_cached_extent
2044 */ 2039 */
2045 spin_lock(&EXT4_I(inode)->i_block_reservation_lock); 2040 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
@@ -2361,7 +2356,7 @@ static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2361 int depth = ext_depth(inode); 2356 int depth = ext_depth(inode);
2362 struct ext4_ext_path *path; 2357 struct ext4_ext_path *path;
2363 handle_t *handle; 2358 handle_t *handle;
2364 int i = 0, err = 0; 2359 int i, err;
2365 2360
2366 ext_debug("truncate since %u\n", start); 2361 ext_debug("truncate since %u\n", start);
2367 2362
@@ -2370,23 +2365,26 @@ static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2370 if (IS_ERR(handle)) 2365 if (IS_ERR(handle))
2371 return PTR_ERR(handle); 2366 return PTR_ERR(handle);
2372 2367
2368again:
2373 ext4_ext_invalidate_cache(inode); 2369 ext4_ext_invalidate_cache(inode);
2374 2370
2375 /* 2371 /*
2376 * We start scanning from right side, freeing all the blocks 2372 * We start scanning from right side, freeing all the blocks
2377 * after i_size and walking into the tree depth-wise. 2373 * after i_size and walking into the tree depth-wise.
2378 */ 2374 */
2375 depth = ext_depth(inode);
2379 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); 2376 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2380 if (path == NULL) { 2377 if (path == NULL) {
2381 ext4_journal_stop(handle); 2378 ext4_journal_stop(handle);
2382 return -ENOMEM; 2379 return -ENOMEM;
2383 } 2380 }
2381 path[0].p_depth = depth;
2384 path[0].p_hdr = ext_inode_hdr(inode); 2382 path[0].p_hdr = ext_inode_hdr(inode);
2385 if (ext4_ext_check(inode, path[0].p_hdr, depth)) { 2383 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2386 err = -EIO; 2384 err = -EIO;
2387 goto out; 2385 goto out;
2388 } 2386 }
2389 path[0].p_depth = depth; 2387 i = err = 0;
2390 2388
2391 while (i >= 0 && err == 0) { 2389 while (i >= 0 && err == 0) {
2392 if (i == depth) { 2390 if (i == depth) {
@@ -2480,6 +2478,8 @@ static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2480out: 2478out:
2481 ext4_ext_drop_refs(path); 2479 ext4_ext_drop_refs(path);
2482 kfree(path); 2480 kfree(path);
2481 if (err == -EAGAIN)
2482 goto again;
2483 ext4_journal_stop(handle); 2483 ext4_journal_stop(handle);
2484 2484
2485 return err; 2485 return err;
@@ -2544,7 +2544,7 @@ static void bi_complete(struct bio *bio, int error)
2544/* FIXME!! we need to try to merge to left or right after zero-out */ 2544/* FIXME!! we need to try to merge to left or right after zero-out */
2545static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) 2545static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2546{ 2546{
2547 int ret = -EIO; 2547 int ret;
2548 struct bio *bio; 2548 struct bio *bio;
2549 int blkbits, blocksize; 2549 int blkbits, blocksize;
2550 sector_t ee_pblock; 2550 sector_t ee_pblock;
@@ -2568,6 +2568,9 @@ static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2568 len = ee_len; 2568 len = ee_len;
2569 2569
2570 bio = bio_alloc(GFP_NOIO, len); 2570 bio = bio_alloc(GFP_NOIO, len);
2571 if (!bio)
2572 return -ENOMEM;
2573
2571 bio->bi_sector = ee_pblock; 2574 bio->bi_sector = ee_pblock;
2572 bio->bi_bdev = inode->i_sb->s_bdev; 2575 bio->bi_bdev = inode->i_sb->s_bdev;
2573 2576
@@ -2595,22 +2598,20 @@ static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2595 submit_bio(WRITE, bio); 2598 submit_bio(WRITE, bio);
2596 wait_for_completion(&event); 2599 wait_for_completion(&event);
2597 2600
2598 if (test_bit(BIO_UPTODATE, &bio->bi_flags)) 2601 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2599 ret = 0; 2602 bio_put(bio);
2600 else { 2603 return -EIO;
2601 ret = -EIO;
2602 break;
2603 } 2604 }
2604 bio_put(bio); 2605 bio_put(bio);
2605 ee_len -= done; 2606 ee_len -= done;
2606 ee_pblock += done << (blkbits - 9); 2607 ee_pblock += done << (blkbits - 9);
2607 } 2608 }
2608 return ret; 2609 return 0;
2609} 2610}
2610 2611
2611#define EXT4_EXT_ZERO_LEN 7 2612#define EXT4_EXT_ZERO_LEN 7
2612/* 2613/*
2613 * This function is called by ext4_ext_get_blocks() if someone tries to write 2614 * This function is called by ext4_ext_map_blocks() if someone tries to write
2614 * to an uninitialized extent. It may result in splitting the uninitialized 2615 * to an uninitialized extent. It may result in splitting the uninitialized
2615 * extent into multiple extents (upto three - one initialized and two 2616 * extent into multiple extents (upto three - one initialized and two
2616 * uninitialized). 2617 * uninitialized).
@@ -2620,39 +2621,55 @@ static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2620 * c> Splits in three extents: Somone is writing in middle of the extent 2621 * c> Splits in three extents: Somone is writing in middle of the extent
2621 */ 2622 */
2622static int ext4_ext_convert_to_initialized(handle_t *handle, 2623static int ext4_ext_convert_to_initialized(handle_t *handle,
2623 struct inode *inode, 2624 struct inode *inode,
2624 struct ext4_ext_path *path, 2625 struct ext4_map_blocks *map,
2625 ext4_lblk_t iblock, 2626 struct ext4_ext_path *path)
2626 unsigned int max_blocks)
2627{ 2627{
2628 struct ext4_extent *ex, newex, orig_ex; 2628 struct ext4_extent *ex, newex, orig_ex;
2629 struct ext4_extent *ex1 = NULL; 2629 struct ext4_extent *ex1 = NULL;
2630 struct ext4_extent *ex2 = NULL; 2630 struct ext4_extent *ex2 = NULL;
2631 struct ext4_extent *ex3 = NULL; 2631 struct ext4_extent *ex3 = NULL;
2632 struct ext4_extent_header *eh; 2632 struct ext4_extent_header *eh;
2633 ext4_lblk_t ee_block; 2633 ext4_lblk_t ee_block, eof_block;
2634 unsigned int allocated, ee_len, depth; 2634 unsigned int allocated, ee_len, depth;
2635 ext4_fsblk_t newblock; 2635 ext4_fsblk_t newblock;
2636 int err = 0; 2636 int err = 0;
2637 int ret = 0; 2637 int ret = 0;
2638 int may_zeroout;
2639
2640 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2641 "block %llu, max_blocks %u\n", inode->i_ino,
2642 (unsigned long long)map->m_lblk, map->m_len);
2643
2644 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2645 inode->i_sb->s_blocksize_bits;
2646 if (eof_block < map->m_lblk + map->m_len)
2647 eof_block = map->m_lblk + map->m_len;
2638 2648
2639 depth = ext_depth(inode); 2649 depth = ext_depth(inode);
2640 eh = path[depth].p_hdr; 2650 eh = path[depth].p_hdr;
2641 ex = path[depth].p_ext; 2651 ex = path[depth].p_ext;
2642 ee_block = le32_to_cpu(ex->ee_block); 2652 ee_block = le32_to_cpu(ex->ee_block);
2643 ee_len = ext4_ext_get_actual_len(ex); 2653 ee_len = ext4_ext_get_actual_len(ex);
2644 allocated = ee_len - (iblock - ee_block); 2654 allocated = ee_len - (map->m_lblk - ee_block);
2645 newblock = iblock - ee_block + ext_pblock(ex); 2655 newblock = map->m_lblk - ee_block + ext_pblock(ex);
2656
2646 ex2 = ex; 2657 ex2 = ex;
2647 orig_ex.ee_block = ex->ee_block; 2658 orig_ex.ee_block = ex->ee_block;
2648 orig_ex.ee_len = cpu_to_le16(ee_len); 2659 orig_ex.ee_len = cpu_to_le16(ee_len);
2649 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); 2660 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2650 2661
2662 /*
2663 * It is safe to convert extent to initialized via explicit
2664 * zeroout only if extent is fully insde i_size or new_size.
2665 */
2666 may_zeroout = ee_block + ee_len <= eof_block;
2667
2651 err = ext4_ext_get_access(handle, inode, path + depth); 2668 err = ext4_ext_get_access(handle, inode, path + depth);
2652 if (err) 2669 if (err)
2653 goto out; 2670 goto out;
2654 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ 2671 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2655 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) { 2672 if (ee_len <= 2*EXT4_EXT_ZERO_LEN && may_zeroout) {
2656 err = ext4_ext_zeroout(inode, &orig_ex); 2673 err = ext4_ext_zeroout(inode, &orig_ex);
2657 if (err) 2674 if (err)
2658 goto fix_extent_len; 2675 goto fix_extent_len;
@@ -2665,10 +2682,10 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2665 return allocated; 2682 return allocated;
2666 } 2683 }
2667 2684
2668 /* ex1: ee_block to iblock - 1 : uninitialized */ 2685 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2669 if (iblock > ee_block) { 2686 if (map->m_lblk > ee_block) {
2670 ex1 = ex; 2687 ex1 = ex;
2671 ex1->ee_len = cpu_to_le16(iblock - ee_block); 2688 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2672 ext4_ext_mark_uninitialized(ex1); 2689 ext4_ext_mark_uninitialized(ex1);
2673 ex2 = &newex; 2690 ex2 = &newex;
2674 } 2691 }
@@ -2677,15 +2694,15 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2677 * we insert ex3, if ex1 is NULL. This is to avoid temporary 2694 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2678 * overlap of blocks. 2695 * overlap of blocks.
2679 */ 2696 */
2680 if (!ex1 && allocated > max_blocks) 2697 if (!ex1 && allocated > map->m_len)
2681 ex2->ee_len = cpu_to_le16(max_blocks); 2698 ex2->ee_len = cpu_to_le16(map->m_len);
2682 /* ex3: to ee_block + ee_len : uninitialised */ 2699 /* ex3: to ee_block + ee_len : uninitialised */
2683 if (allocated > max_blocks) { 2700 if (allocated > map->m_len) {
2684 unsigned int newdepth; 2701 unsigned int newdepth;
2685 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */ 2702 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2686 if (allocated <= EXT4_EXT_ZERO_LEN) { 2703 if (allocated <= EXT4_EXT_ZERO_LEN && may_zeroout) {
2687 /* 2704 /*
2688 * iblock == ee_block is handled by the zerouout 2705 * map->m_lblk == ee_block is handled by the zerouout
2689 * at the beginning. 2706 * at the beginning.
2690 * Mark first half uninitialized. 2707 * Mark first half uninitialized.
2691 * Mark second half initialized and zero out the 2708 * Mark second half initialized and zero out the
@@ -2698,7 +2715,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2698 ext4_ext_dirty(handle, inode, path + depth); 2715 ext4_ext_dirty(handle, inode, path + depth);
2699 2716
2700 ex3 = &newex; 2717 ex3 = &newex;
2701 ex3->ee_block = cpu_to_le32(iblock); 2718 ex3->ee_block = cpu_to_le32(map->m_lblk);
2702 ext4_ext_store_pblock(ex3, newblock); 2719 ext4_ext_store_pblock(ex3, newblock);
2703 ex3->ee_len = cpu_to_le16(allocated); 2720 ex3->ee_len = cpu_to_le16(allocated);
2704 err = ext4_ext_insert_extent(handle, inode, path, 2721 err = ext4_ext_insert_extent(handle, inode, path,
@@ -2711,7 +2728,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2711 ex->ee_len = orig_ex.ee_len; 2728 ex->ee_len = orig_ex.ee_len;
2712 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2729 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2713 ext4_ext_dirty(handle, inode, path + depth); 2730 ext4_ext_dirty(handle, inode, path + depth);
2714 /* blocks available from iblock */ 2731 /* blocks available from map->m_lblk */
2715 return allocated; 2732 return allocated;
2716 2733
2717 } else if (err) 2734 } else if (err)
@@ -2733,8 +2750,8 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2733 */ 2750 */
2734 depth = ext_depth(inode); 2751 depth = ext_depth(inode);
2735 ext4_ext_drop_refs(path); 2752 ext4_ext_drop_refs(path);
2736 path = ext4_ext_find_extent(inode, 2753 path = ext4_ext_find_extent(inode, map->m_lblk,
2737 iblock, path); 2754 path);
2738 if (IS_ERR(path)) { 2755 if (IS_ERR(path)) {
2739 err = PTR_ERR(path); 2756 err = PTR_ERR(path);
2740 return err; 2757 return err;
@@ -2754,12 +2771,12 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2754 return allocated; 2771 return allocated;
2755 } 2772 }
2756 ex3 = &newex; 2773 ex3 = &newex;
2757 ex3->ee_block = cpu_to_le32(iblock + max_blocks); 2774 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2758 ext4_ext_store_pblock(ex3, newblock + max_blocks); 2775 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2759 ex3->ee_len = cpu_to_le16(allocated - max_blocks); 2776 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2760 ext4_ext_mark_uninitialized(ex3); 2777 ext4_ext_mark_uninitialized(ex3);
2761 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0); 2778 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2762 if (err == -ENOSPC) { 2779 if (err == -ENOSPC && may_zeroout) {
2763 err = ext4_ext_zeroout(inode, &orig_ex); 2780 err = ext4_ext_zeroout(inode, &orig_ex);
2764 if (err) 2781 if (err)
2765 goto fix_extent_len; 2782 goto fix_extent_len;
@@ -2769,7 +2786,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2769 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2786 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2770 ext4_ext_dirty(handle, inode, path + depth); 2787 ext4_ext_dirty(handle, inode, path + depth);
2771 /* zeroed the full extent */ 2788 /* zeroed the full extent */
2772 /* blocks available from iblock */ 2789 /* blocks available from map->m_lblk */
2773 return allocated; 2790 return allocated;
2774 2791
2775 } else if (err) 2792 } else if (err)
@@ -2783,11 +2800,13 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2783 * update the extent length after successful insert of the 2800 * update the extent length after successful insert of the
2784 * split extent 2801 * split extent
2785 */ 2802 */
2786 orig_ex.ee_len = cpu_to_le16(ee_len - 2803 ee_len -= ext4_ext_get_actual_len(ex3);
2787 ext4_ext_get_actual_len(ex3)); 2804 orig_ex.ee_len = cpu_to_le16(ee_len);
2805 may_zeroout = ee_block + ee_len <= eof_block;
2806
2788 depth = newdepth; 2807 depth = newdepth;
2789 ext4_ext_drop_refs(path); 2808 ext4_ext_drop_refs(path);
2790 path = ext4_ext_find_extent(inode, iblock, path); 2809 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2791 if (IS_ERR(path)) { 2810 if (IS_ERR(path)) {
2792 err = PTR_ERR(path); 2811 err = PTR_ERR(path);
2793 goto out; 2812 goto out;
@@ -2801,14 +2820,14 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2801 if (err) 2820 if (err)
2802 goto out; 2821 goto out;
2803 2822
2804 allocated = max_blocks; 2823 allocated = map->m_len;
2805 2824
2806 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying 2825 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2807 * to insert a extent in the middle zerout directly 2826 * to insert a extent in the middle zerout directly
2808 * otherwise give the extent a chance to merge to left 2827 * otherwise give the extent a chance to merge to left
2809 */ 2828 */
2810 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN && 2829 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2811 iblock != ee_block) { 2830 map->m_lblk != ee_block && may_zeroout) {
2812 err = ext4_ext_zeroout(inode, &orig_ex); 2831 err = ext4_ext_zeroout(inode, &orig_ex);
2813 if (err) 2832 if (err)
2814 goto fix_extent_len; 2833 goto fix_extent_len;
@@ -2818,7 +2837,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2818 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2837 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2819 ext4_ext_dirty(handle, inode, path + depth); 2838 ext4_ext_dirty(handle, inode, path + depth);
2820 /* zero out the first half */ 2839 /* zero out the first half */
2821 /* blocks available from iblock */ 2840 /* blocks available from map->m_lblk */
2822 return allocated; 2841 return allocated;
2823 } 2842 }
2824 } 2843 }
@@ -2829,12 +2848,12 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2829 */ 2848 */
2830 if (ex1 && ex1 != ex) { 2849 if (ex1 && ex1 != ex) {
2831 ex1 = ex; 2850 ex1 = ex;
2832 ex1->ee_len = cpu_to_le16(iblock - ee_block); 2851 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2833 ext4_ext_mark_uninitialized(ex1); 2852 ext4_ext_mark_uninitialized(ex1);
2834 ex2 = &newex; 2853 ex2 = &newex;
2835 } 2854 }
2836 /* ex2: iblock to iblock + maxblocks-1 : initialised */ 2855 /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2837 ex2->ee_block = cpu_to_le32(iblock); 2856 ex2->ee_block = cpu_to_le32(map->m_lblk);
2838 ext4_ext_store_pblock(ex2, newblock); 2857 ext4_ext_store_pblock(ex2, newblock);
2839 ex2->ee_len = cpu_to_le16(allocated); 2858 ex2->ee_len = cpu_to_le16(allocated);
2840 if (ex2 != ex) 2859 if (ex2 != ex)
@@ -2877,7 +2896,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2877 goto out; 2896 goto out;
2878insert: 2897insert:
2879 err = ext4_ext_insert_extent(handle, inode, path, &newex, 0); 2898 err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2880 if (err == -ENOSPC) { 2899 if (err == -ENOSPC && may_zeroout) {
2881 err = ext4_ext_zeroout(inode, &orig_ex); 2900 err = ext4_ext_zeroout(inode, &orig_ex);
2882 if (err) 2901 if (err)
2883 goto fix_extent_len; 2902 goto fix_extent_len;
@@ -2904,7 +2923,7 @@ fix_extent_len:
2904} 2923}
2905 2924
2906/* 2925/*
2907 * This function is called by ext4_ext_get_blocks() from 2926 * This function is called by ext4_ext_map_blocks() from
2908 * ext4_get_blocks_dio_write() when DIO to write 2927 * ext4_get_blocks_dio_write() when DIO to write
2909 * to an uninitialized extent. 2928 * to an uninitialized extent.
2910 * 2929 *
@@ -2927,9 +2946,8 @@ fix_extent_len:
2927 */ 2946 */
2928static int ext4_split_unwritten_extents(handle_t *handle, 2947static int ext4_split_unwritten_extents(handle_t *handle,
2929 struct inode *inode, 2948 struct inode *inode,
2949 struct ext4_map_blocks *map,
2930 struct ext4_ext_path *path, 2950 struct ext4_ext_path *path,
2931 ext4_lblk_t iblock,
2932 unsigned int max_blocks,
2933 int flags) 2951 int flags)
2934{ 2952{
2935 struct ext4_extent *ex, newex, orig_ex; 2953 struct ext4_extent *ex, newex, orig_ex;
@@ -2937,41 +2955,55 @@ static int ext4_split_unwritten_extents(handle_t *handle,
2937 struct ext4_extent *ex2 = NULL; 2955 struct ext4_extent *ex2 = NULL;
2938 struct ext4_extent *ex3 = NULL; 2956 struct ext4_extent *ex3 = NULL;
2939 struct ext4_extent_header *eh; 2957 struct ext4_extent_header *eh;
2940 ext4_lblk_t ee_block; 2958 ext4_lblk_t ee_block, eof_block;
2941 unsigned int allocated, ee_len, depth; 2959 unsigned int allocated, ee_len, depth;
2942 ext4_fsblk_t newblock; 2960 ext4_fsblk_t newblock;
2943 int err = 0; 2961 int err = 0;
2962 int may_zeroout;
2963
2964 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2965 "block %llu, max_blocks %u\n", inode->i_ino,
2966 (unsigned long long)map->m_lblk, map->m_len);
2967
2968 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2969 inode->i_sb->s_blocksize_bits;
2970 if (eof_block < map->m_lblk + map->m_len)
2971 eof_block = map->m_lblk + map->m_len;
2944 2972
2945 ext_debug("ext4_split_unwritten_extents: inode %lu,"
2946 "iblock %llu, max_blocks %u\n", inode->i_ino,
2947 (unsigned long long)iblock, max_blocks);
2948 depth = ext_depth(inode); 2973 depth = ext_depth(inode);
2949 eh = path[depth].p_hdr; 2974 eh = path[depth].p_hdr;
2950 ex = path[depth].p_ext; 2975 ex = path[depth].p_ext;
2951 ee_block = le32_to_cpu(ex->ee_block); 2976 ee_block = le32_to_cpu(ex->ee_block);
2952 ee_len = ext4_ext_get_actual_len(ex); 2977 ee_len = ext4_ext_get_actual_len(ex);
2953 allocated = ee_len - (iblock - ee_block); 2978 allocated = ee_len - (map->m_lblk - ee_block);
2954 newblock = iblock - ee_block + ext_pblock(ex); 2979 newblock = map->m_lblk - ee_block + ext_pblock(ex);
2980
2955 ex2 = ex; 2981 ex2 = ex;
2956 orig_ex.ee_block = ex->ee_block; 2982 orig_ex.ee_block = ex->ee_block;
2957 orig_ex.ee_len = cpu_to_le16(ee_len); 2983 orig_ex.ee_len = cpu_to_le16(ee_len);
2958 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); 2984 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2959 2985
2960 /* 2986 /*
2987 * It is safe to convert extent to initialized via explicit
2988 * zeroout only if extent is fully insde i_size or new_size.
2989 */
2990 may_zeroout = ee_block + ee_len <= eof_block;
2991
2992 /*
2961 * If the uninitialized extent begins at the same logical 2993 * If the uninitialized extent begins at the same logical
2962 * block where the write begins, and the write completely 2994 * block where the write begins, and the write completely
2963 * covers the extent, then we don't need to split it. 2995 * covers the extent, then we don't need to split it.
2964 */ 2996 */
2965 if ((iblock == ee_block) && (allocated <= max_blocks)) 2997 if ((map->m_lblk == ee_block) && (allocated <= map->m_len))
2966 return allocated; 2998 return allocated;
2967 2999
2968 err = ext4_ext_get_access(handle, inode, path + depth); 3000 err = ext4_ext_get_access(handle, inode, path + depth);
2969 if (err) 3001 if (err)
2970 goto out; 3002 goto out;
2971 /* ex1: ee_block to iblock - 1 : uninitialized */ 3003 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2972 if (iblock > ee_block) { 3004 if (map->m_lblk > ee_block) {
2973 ex1 = ex; 3005 ex1 = ex;
2974 ex1->ee_len = cpu_to_le16(iblock - ee_block); 3006 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2975 ext4_ext_mark_uninitialized(ex1); 3007 ext4_ext_mark_uninitialized(ex1);
2976 ex2 = &newex; 3008 ex2 = &newex;
2977 } 3009 }
@@ -2980,18 +3012,18 @@ static int ext4_split_unwritten_extents(handle_t *handle,
2980 * we insert ex3, if ex1 is NULL. This is to avoid temporary 3012 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2981 * overlap of blocks. 3013 * overlap of blocks.
2982 */ 3014 */
2983 if (!ex1 && allocated > max_blocks) 3015 if (!ex1 && allocated > map->m_len)
2984 ex2->ee_len = cpu_to_le16(max_blocks); 3016 ex2->ee_len = cpu_to_le16(map->m_len);
2985 /* ex3: to ee_block + ee_len : uninitialised */ 3017 /* ex3: to ee_block + ee_len : uninitialised */
2986 if (allocated > max_blocks) { 3018 if (allocated > map->m_len) {
2987 unsigned int newdepth; 3019 unsigned int newdepth;
2988 ex3 = &newex; 3020 ex3 = &newex;
2989 ex3->ee_block = cpu_to_le32(iblock + max_blocks); 3021 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2990 ext4_ext_store_pblock(ex3, newblock + max_blocks); 3022 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2991 ex3->ee_len = cpu_to_le16(allocated - max_blocks); 3023 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2992 ext4_ext_mark_uninitialized(ex3); 3024 ext4_ext_mark_uninitialized(ex3);
2993 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags); 3025 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2994 if (err == -ENOSPC) { 3026 if (err == -ENOSPC && may_zeroout) {
2995 err = ext4_ext_zeroout(inode, &orig_ex); 3027 err = ext4_ext_zeroout(inode, &orig_ex);
2996 if (err) 3028 if (err)
2997 goto fix_extent_len; 3029 goto fix_extent_len;
@@ -3001,7 +3033,7 @@ static int ext4_split_unwritten_extents(handle_t *handle,
3001 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 3033 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3002 ext4_ext_dirty(handle, inode, path + depth); 3034 ext4_ext_dirty(handle, inode, path + depth);
3003 /* zeroed the full extent */ 3035 /* zeroed the full extent */
3004 /* blocks available from iblock */ 3036 /* blocks available from map->m_lblk */
3005 return allocated; 3037 return allocated;
3006 3038
3007 } else if (err) 3039 } else if (err)
@@ -3015,11 +3047,13 @@ static int ext4_split_unwritten_extents(handle_t *handle,
3015 * update the extent length after successful insert of the 3047 * update the extent length after successful insert of the
3016 * split extent 3048 * split extent
3017 */ 3049 */
3018 orig_ex.ee_len = cpu_to_le16(ee_len - 3050 ee_len -= ext4_ext_get_actual_len(ex3);
3019 ext4_ext_get_actual_len(ex3)); 3051 orig_ex.ee_len = cpu_to_le16(ee_len);
3052 may_zeroout = ee_block + ee_len <= eof_block;
3053
3020 depth = newdepth; 3054 depth = newdepth;
3021 ext4_ext_drop_refs(path); 3055 ext4_ext_drop_refs(path);
3022 path = ext4_ext_find_extent(inode, iblock, path); 3056 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3023 if (IS_ERR(path)) { 3057 if (IS_ERR(path)) {
3024 err = PTR_ERR(path); 3058 err = PTR_ERR(path);
3025 goto out; 3059 goto out;
@@ -3033,7 +3067,7 @@ static int ext4_split_unwritten_extents(handle_t *handle,
3033 if (err) 3067 if (err)
3034 goto out; 3068 goto out;
3035 3069
3036 allocated = max_blocks; 3070 allocated = map->m_len;
3037 } 3071 }
3038 /* 3072 /*
3039 * If there was a change of depth as part of the 3073 * If there was a change of depth as part of the
@@ -3042,15 +3076,15 @@ static int ext4_split_unwritten_extents(handle_t *handle,
3042 */ 3076 */
3043 if (ex1 && ex1 != ex) { 3077 if (ex1 && ex1 != ex) {
3044 ex1 = ex; 3078 ex1 = ex;
3045 ex1->ee_len = cpu_to_le16(iblock - ee_block); 3079 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
3046 ext4_ext_mark_uninitialized(ex1); 3080 ext4_ext_mark_uninitialized(ex1);
3047 ex2 = &newex; 3081 ex2 = &newex;
3048 } 3082 }
3049 /* 3083 /*
3050 * ex2: iblock to iblock + maxblocks-1 : to be direct IO written, 3084 * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
3051 * uninitialised still. 3085 * using direct I/O, uninitialised still.
3052 */ 3086 */
3053 ex2->ee_block = cpu_to_le32(iblock); 3087 ex2->ee_block = cpu_to_le32(map->m_lblk);
3054 ext4_ext_store_pblock(ex2, newblock); 3088 ext4_ext_store_pblock(ex2, newblock);
3055 ex2->ee_len = cpu_to_le16(allocated); 3089 ex2->ee_len = cpu_to_le16(allocated);
3056 ext4_ext_mark_uninitialized(ex2); 3090 ext4_ext_mark_uninitialized(ex2);
@@ -3062,7 +3096,7 @@ static int ext4_split_unwritten_extents(handle_t *handle,
3062 goto out; 3096 goto out;
3063insert: 3097insert:
3064 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); 3098 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3065 if (err == -ENOSPC) { 3099 if (err == -ENOSPC && may_zeroout) {
3066 err = ext4_ext_zeroout(inode, &orig_ex); 3100 err = ext4_ext_zeroout(inode, &orig_ex);
3067 if (err) 3101 if (err)
3068 goto fix_extent_len; 3102 goto fix_extent_len;
@@ -3152,10 +3186,9 @@ static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3152 3186
3153static int 3187static int
3154ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, 3188ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3155 ext4_lblk_t iblock, unsigned int max_blocks, 3189 struct ext4_map_blocks *map,
3156 struct ext4_ext_path *path, int flags, 3190 struct ext4_ext_path *path, int flags,
3157 unsigned int allocated, struct buffer_head *bh_result, 3191 unsigned int allocated, ext4_fsblk_t newblock)
3158 ext4_fsblk_t newblock)
3159{ 3192{
3160 int ret = 0; 3193 int ret = 0;
3161 int err = 0; 3194 int err = 0;
@@ -3163,15 +3196,14 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3163 3196
3164 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical" 3197 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3165 "block %llu, max_blocks %u, flags %d, allocated %u", 3198 "block %llu, max_blocks %u, flags %d, allocated %u",
3166 inode->i_ino, (unsigned long long)iblock, max_blocks, 3199 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3167 flags, allocated); 3200 flags, allocated);
3168 ext4_ext_show_leaf(inode, path); 3201 ext4_ext_show_leaf(inode, path);
3169 3202
3170 /* get_block() before submit the IO, split the extent */ 3203 /* get_block() before submit the IO, split the extent */
3171 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { 3204 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3172 ret = ext4_split_unwritten_extents(handle, 3205 ret = ext4_split_unwritten_extents(handle, inode, map,
3173 inode, path, iblock, 3206 path, flags);
3174 max_blocks, flags);
3175 /* 3207 /*
3176 * Flag the inode(non aio case) or end_io struct (aio case) 3208 * Flag the inode(non aio case) or end_io struct (aio case)
3177 * that this IO needs to convertion to written when IO is 3209 * that this IO needs to convertion to written when IO is
@@ -3182,7 +3214,7 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3182 else 3214 else
3183 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); 3215 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3184 if (ext4_should_dioread_nolock(inode)) 3216 if (ext4_should_dioread_nolock(inode))
3185 set_buffer_uninit(bh_result); 3217 map->m_flags |= EXT4_MAP_UNINIT;
3186 goto out; 3218 goto out;
3187 } 3219 }
3188 /* IO end_io complete, convert the filled extent to written */ 3220 /* IO end_io complete, convert the filled extent to written */
@@ -3210,14 +3242,12 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3210 * the buffer head will be unmapped so that 3242 * the buffer head will be unmapped so that
3211 * a read from the block returns 0s. 3243 * a read from the block returns 0s.
3212 */ 3244 */
3213 set_buffer_unwritten(bh_result); 3245 map->m_flags |= EXT4_MAP_UNWRITTEN;
3214 goto out1; 3246 goto out1;
3215 } 3247 }
3216 3248
3217 /* buffered write, writepage time, convert*/ 3249 /* buffered write, writepage time, convert*/
3218 ret = ext4_ext_convert_to_initialized(handle, inode, 3250 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3219 path, iblock,
3220 max_blocks);
3221 if (ret >= 0) 3251 if (ret >= 0)
3222 ext4_update_inode_fsync_trans(handle, inode, 1); 3252 ext4_update_inode_fsync_trans(handle, inode, 1);
3223out: 3253out:
@@ -3226,7 +3256,7 @@ out:
3226 goto out2; 3256 goto out2;
3227 } else 3257 } else
3228 allocated = ret; 3258 allocated = ret;
3229 set_buffer_new(bh_result); 3259 map->m_flags |= EXT4_MAP_NEW;
3230 /* 3260 /*
3231 * if we allocated more blocks than requested 3261 * if we allocated more blocks than requested
3232 * we need to make sure we unmap the extra block 3262 * we need to make sure we unmap the extra block
@@ -3234,11 +3264,11 @@ out:
3234 * unmapped later when we find the buffer_head marked 3264 * unmapped later when we find the buffer_head marked
3235 * new. 3265 * new.
3236 */ 3266 */
3237 if (allocated > max_blocks) { 3267 if (allocated > map->m_len) {
3238 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, 3268 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3239 newblock + max_blocks, 3269 newblock + map->m_len,
3240 allocated - max_blocks); 3270 allocated - map->m_len);
3241 allocated = max_blocks; 3271 allocated = map->m_len;
3242 } 3272 }
3243 3273
3244 /* 3274 /*
@@ -3252,13 +3282,13 @@ out:
3252 ext4_da_update_reserve_space(inode, allocated, 0); 3282 ext4_da_update_reserve_space(inode, allocated, 0);
3253 3283
3254map_out: 3284map_out:
3255 set_buffer_mapped(bh_result); 3285 map->m_flags |= EXT4_MAP_MAPPED;
3256out1: 3286out1:
3257 if (allocated > max_blocks) 3287 if (allocated > map->m_len)
3258 allocated = max_blocks; 3288 allocated = map->m_len;
3259 ext4_ext_show_leaf(inode, path); 3289 ext4_ext_show_leaf(inode, path);
3260 bh_result->b_bdev = inode->i_sb->s_bdev; 3290 map->m_pblk = newblock;
3261 bh_result->b_blocknr = newblock; 3291 map->m_len = allocated;
3262out2: 3292out2:
3263 if (path) { 3293 if (path) {
3264 ext4_ext_drop_refs(path); 3294 ext4_ext_drop_refs(path);
@@ -3284,26 +3314,23 @@ out2:
3284 * 3314 *
3285 * return < 0, error case. 3315 * return < 0, error case.
3286 */ 3316 */
3287int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, 3317int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3288 ext4_lblk_t iblock, 3318 struct ext4_map_blocks *map, int flags)
3289 unsigned int max_blocks, struct buffer_head *bh_result,
3290 int flags)
3291{ 3319{
3292 struct ext4_ext_path *path = NULL; 3320 struct ext4_ext_path *path = NULL;
3293 struct ext4_extent_header *eh; 3321 struct ext4_extent_header *eh;
3294 struct ext4_extent newex, *ex, *last_ex; 3322 struct ext4_extent newex, *ex, *last_ex;
3295 ext4_fsblk_t newblock; 3323 ext4_fsblk_t newblock;
3296 int err = 0, depth, ret, cache_type; 3324 int i, err = 0, depth, ret, cache_type;
3297 unsigned int allocated = 0; 3325 unsigned int allocated = 0;
3298 struct ext4_allocation_request ar; 3326 struct ext4_allocation_request ar;
3299 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; 3327 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3300 3328
3301 __clear_bit(BH_New, &bh_result->b_state);
3302 ext_debug("blocks %u/%u requested for inode %lu\n", 3329 ext_debug("blocks %u/%u requested for inode %lu\n",
3303 iblock, max_blocks, inode->i_ino); 3330 map->m_lblk, map->m_len, inode->i_ino);
3304 3331
3305 /* check in cache */ 3332 /* check in cache */
3306 cache_type = ext4_ext_in_cache(inode, iblock, &newex); 3333 cache_type = ext4_ext_in_cache(inode, map->m_lblk, &newex);
3307 if (cache_type) { 3334 if (cache_type) {
3308 if (cache_type == EXT4_EXT_CACHE_GAP) { 3335 if (cache_type == EXT4_EXT_CACHE_GAP) {
3309 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { 3336 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
@@ -3316,12 +3343,12 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3316 /* we should allocate requested block */ 3343 /* we should allocate requested block */
3317 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) { 3344 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3318 /* block is already allocated */ 3345 /* block is already allocated */
3319 newblock = iblock 3346 newblock = map->m_lblk
3320 - le32_to_cpu(newex.ee_block) 3347 - le32_to_cpu(newex.ee_block)
3321 + ext_pblock(&newex); 3348 + ext_pblock(&newex);
3322 /* number of remaining blocks in the extent */ 3349 /* number of remaining blocks in the extent */
3323 allocated = ext4_ext_get_actual_len(&newex) - 3350 allocated = ext4_ext_get_actual_len(&newex) -
3324 (iblock - le32_to_cpu(newex.ee_block)); 3351 (map->m_lblk - le32_to_cpu(newex.ee_block));
3325 goto out; 3352 goto out;
3326 } else { 3353 } else {
3327 BUG(); 3354 BUG();
@@ -3329,7 +3356,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3329 } 3356 }
3330 3357
3331 /* find extent for this block */ 3358 /* find extent for this block */
3332 path = ext4_ext_find_extent(inode, iblock, NULL); 3359 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3333 if (IS_ERR(path)) { 3360 if (IS_ERR(path)) {
3334 err = PTR_ERR(path); 3361 err = PTR_ERR(path);
3335 path = NULL; 3362 path = NULL;
@@ -3345,8 +3372,9 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3345 */ 3372 */
3346 if (unlikely(path[depth].p_ext == NULL && depth != 0)) { 3373 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3347 EXT4_ERROR_INODE(inode, "bad extent address " 3374 EXT4_ERROR_INODE(inode, "bad extent address "
3348 "iblock: %d, depth: %d pblock %lld", 3375 "lblock: %lu, depth: %d pblock %lld",
3349 iblock, depth, path[depth].p_block); 3376 (unsigned long) map->m_lblk, depth,
3377 path[depth].p_block);
3350 err = -EIO; 3378 err = -EIO;
3351 goto out2; 3379 goto out2;
3352 } 3380 }
@@ -3364,12 +3392,12 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3364 */ 3392 */
3365 ee_len = ext4_ext_get_actual_len(ex); 3393 ee_len = ext4_ext_get_actual_len(ex);
3366 /* if found extent covers block, simply return it */ 3394 /* if found extent covers block, simply return it */
3367 if (in_range(iblock, ee_block, ee_len)) { 3395 if (in_range(map->m_lblk, ee_block, ee_len)) {
3368 newblock = iblock - ee_block + ee_start; 3396 newblock = map->m_lblk - ee_block + ee_start;
3369 /* number of remaining blocks in the extent */ 3397 /* number of remaining blocks in the extent */
3370 allocated = ee_len - (iblock - ee_block); 3398 allocated = ee_len - (map->m_lblk - ee_block);
3371 ext_debug("%u fit into %u:%d -> %llu\n", iblock, 3399 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3372 ee_block, ee_len, newblock); 3400 ee_block, ee_len, newblock);
3373 3401
3374 /* Do not put uninitialized extent in the cache */ 3402 /* Do not put uninitialized extent in the cache */
3375 if (!ext4_ext_is_uninitialized(ex)) { 3403 if (!ext4_ext_is_uninitialized(ex)) {
@@ -3379,8 +3407,8 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3379 goto out; 3407 goto out;
3380 } 3408 }
3381 ret = ext4_ext_handle_uninitialized_extents(handle, 3409 ret = ext4_ext_handle_uninitialized_extents(handle,
3382 inode, iblock, max_blocks, path, 3410 inode, map, path, flags, allocated,
3383 flags, allocated, bh_result, newblock); 3411 newblock);
3384 return ret; 3412 return ret;
3385 } 3413 }
3386 } 3414 }
@@ -3394,7 +3422,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3394 * put just found gap into cache to speed up 3422 * put just found gap into cache to speed up
3395 * subsequent requests 3423 * subsequent requests
3396 */ 3424 */
3397 ext4_ext_put_gap_in_cache(inode, path, iblock); 3425 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3398 goto out2; 3426 goto out2;
3399 } 3427 }
3400 /* 3428 /*
@@ -3402,11 +3430,11 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3402 */ 3430 */
3403 3431
3404 /* find neighbour allocated blocks */ 3432 /* find neighbour allocated blocks */
3405 ar.lleft = iblock; 3433 ar.lleft = map->m_lblk;
3406 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); 3434 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3407 if (err) 3435 if (err)
3408 goto out2; 3436 goto out2;
3409 ar.lright = iblock; 3437 ar.lright = map->m_lblk;
3410 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright); 3438 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3411 if (err) 3439 if (err)
3412 goto out2; 3440 goto out2;
@@ -3417,26 +3445,26 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3417 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is 3445 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3418 * EXT_UNINIT_MAX_LEN. 3446 * EXT_UNINIT_MAX_LEN.
3419 */ 3447 */
3420 if (max_blocks > EXT_INIT_MAX_LEN && 3448 if (map->m_len > EXT_INIT_MAX_LEN &&
3421 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) 3449 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3422 max_blocks = EXT_INIT_MAX_LEN; 3450 map->m_len = EXT_INIT_MAX_LEN;
3423 else if (max_blocks > EXT_UNINIT_MAX_LEN && 3451 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3424 (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) 3452 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3425 max_blocks = EXT_UNINIT_MAX_LEN; 3453 map->m_len = EXT_UNINIT_MAX_LEN;
3426 3454
3427 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ 3455 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3428 newex.ee_block = cpu_to_le32(iblock); 3456 newex.ee_block = cpu_to_le32(map->m_lblk);
3429 newex.ee_len = cpu_to_le16(max_blocks); 3457 newex.ee_len = cpu_to_le16(map->m_len);
3430 err = ext4_ext_check_overlap(inode, &newex, path); 3458 err = ext4_ext_check_overlap(inode, &newex, path);
3431 if (err) 3459 if (err)
3432 allocated = ext4_ext_get_actual_len(&newex); 3460 allocated = ext4_ext_get_actual_len(&newex);
3433 else 3461 else
3434 allocated = max_blocks; 3462 allocated = map->m_len;
3435 3463
3436 /* allocate new block */ 3464 /* allocate new block */
3437 ar.inode = inode; 3465 ar.inode = inode;
3438 ar.goal = ext4_ext_find_goal(inode, path, iblock); 3466 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3439 ar.logical = iblock; 3467 ar.logical = map->m_lblk;
3440 ar.len = allocated; 3468 ar.len = allocated;
3441 if (S_ISREG(inode->i_mode)) 3469 if (S_ISREG(inode->i_mode))
3442 ar.flags = EXT4_MB_HINT_DATA; 3470 ar.flags = EXT4_MB_HINT_DATA;
@@ -3470,21 +3498,33 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3470 EXT4_STATE_DIO_UNWRITTEN); 3498 EXT4_STATE_DIO_UNWRITTEN);
3471 } 3499 }
3472 if (ext4_should_dioread_nolock(inode)) 3500 if (ext4_should_dioread_nolock(inode))
3473 set_buffer_uninit(bh_result); 3501 map->m_flags |= EXT4_MAP_UNINIT;
3474 } 3502 }
3475 3503
3476 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL)) { 3504 if (unlikely(ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))) {
3477 if (unlikely(!eh->eh_entries)) { 3505 if (unlikely(!eh->eh_entries)) {
3478 EXT4_ERROR_INODE(inode, 3506 EXT4_ERROR_INODE(inode,
3479 "eh->eh_entries == 0 ee_block %d", 3507 "eh->eh_entries == 0 and "
3480 ex->ee_block); 3508 "EOFBLOCKS_FL set");
3481 err = -EIO; 3509 err = -EIO;
3482 goto out2; 3510 goto out2;
3483 } 3511 }
3484 last_ex = EXT_LAST_EXTENT(eh); 3512 last_ex = EXT_LAST_EXTENT(eh);
3485 if (iblock + ar.len > le32_to_cpu(last_ex->ee_block) 3513 /*
3486 + ext4_ext_get_actual_len(last_ex)) 3514 * If the current leaf block was reached by looking at
3487 EXT4_I(inode)->i_flags &= ~EXT4_EOFBLOCKS_FL; 3515 * the last index block all the way down the tree, and
3516 * we are extending the inode beyond the last extent
3517 * in the current leaf block, then clear the
3518 * EOFBLOCKS_FL flag.
3519 */
3520 for (i = depth-1; i >= 0; i--) {
3521 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3522 break;
3523 }
3524 if ((i < 0) &&
3525 (map->m_lblk + ar.len > le32_to_cpu(last_ex->ee_block) +
3526 ext4_ext_get_actual_len(last_ex)))
3527 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3488 } 3528 }
3489 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); 3529 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3490 if (err) { 3530 if (err) {
@@ -3500,9 +3540,9 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3500 /* previous routine could use block we allocated */ 3540 /* previous routine could use block we allocated */
3501 newblock = ext_pblock(&newex); 3541 newblock = ext_pblock(&newex);
3502 allocated = ext4_ext_get_actual_len(&newex); 3542 allocated = ext4_ext_get_actual_len(&newex);
3503 if (allocated > max_blocks) 3543 if (allocated > map->m_len)
3504 allocated = max_blocks; 3544 allocated = map->m_len;
3505 set_buffer_new(bh_result); 3545 map->m_flags |= EXT4_MAP_NEW;
3506 3546
3507 /* 3547 /*
3508 * Update reserved blocks/metadata blocks after successful 3548 * Update reserved blocks/metadata blocks after successful
@@ -3516,18 +3556,18 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3516 * when it is _not_ an uninitialized extent. 3556 * when it is _not_ an uninitialized extent.
3517 */ 3557 */
3518 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { 3558 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3519 ext4_ext_put_in_cache(inode, iblock, allocated, newblock, 3559 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock,
3520 EXT4_EXT_CACHE_EXTENT); 3560 EXT4_EXT_CACHE_EXTENT);
3521 ext4_update_inode_fsync_trans(handle, inode, 1); 3561 ext4_update_inode_fsync_trans(handle, inode, 1);
3522 } else 3562 } else
3523 ext4_update_inode_fsync_trans(handle, inode, 0); 3563 ext4_update_inode_fsync_trans(handle, inode, 0);
3524out: 3564out:
3525 if (allocated > max_blocks) 3565 if (allocated > map->m_len)
3526 allocated = max_blocks; 3566 allocated = map->m_len;
3527 ext4_ext_show_leaf(inode, path); 3567 ext4_ext_show_leaf(inode, path);
3528 set_buffer_mapped(bh_result); 3568 map->m_flags |= EXT4_MAP_MAPPED;
3529 bh_result->b_bdev = inode->i_sb->s_bdev; 3569 map->m_pblk = newblock;
3530 bh_result->b_blocknr = newblock; 3570 map->m_len = allocated;
3531out2: 3571out2:
3532 if (path) { 3572 if (path) {
3533 ext4_ext_drop_refs(path); 3573 ext4_ext_drop_refs(path);
@@ -3625,7 +3665,7 @@ static void ext4_falloc_update_inode(struct inode *inode,
3625 * can proceed even if the new size is the same as i_size. 3665 * can proceed even if the new size is the same as i_size.
3626 */ 3666 */
3627 if (new_size > i_size_read(inode)) 3667 if (new_size > i_size_read(inode))
3628 EXT4_I(inode)->i_flags |= EXT4_EOFBLOCKS_FL; 3668 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3629 } 3669 }
3630 3670
3631} 3671}
@@ -3640,55 +3680,57 @@ static void ext4_falloc_update_inode(struct inode *inode,
3640long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len) 3680long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3641{ 3681{
3642 handle_t *handle; 3682 handle_t *handle;
3643 ext4_lblk_t block;
3644 loff_t new_size; 3683 loff_t new_size;
3645 unsigned int max_blocks; 3684 unsigned int max_blocks;
3646 int ret = 0; 3685 int ret = 0;
3647 int ret2 = 0; 3686 int ret2 = 0;
3648 int retries = 0; 3687 int retries = 0;
3649 struct buffer_head map_bh; 3688 struct ext4_map_blocks map;
3650 unsigned int credits, blkbits = inode->i_blkbits; 3689 unsigned int credits, blkbits = inode->i_blkbits;
3651 3690
3652 /* 3691 /*
3653 * currently supporting (pre)allocate mode for extent-based 3692 * currently supporting (pre)allocate mode for extent-based
3654 * files _only_ 3693 * files _only_
3655 */ 3694 */
3656 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 3695 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3657 return -EOPNOTSUPP; 3696 return -EOPNOTSUPP;
3658 3697
3659 /* preallocation to directories is currently not supported */ 3698 /* preallocation to directories is currently not supported */
3660 if (S_ISDIR(inode->i_mode)) 3699 if (S_ISDIR(inode->i_mode))
3661 return -ENODEV; 3700 return -ENODEV;
3662 3701
3663 block = offset >> blkbits; 3702 map.m_lblk = offset >> blkbits;
3664 /* 3703 /*
3665 * We can't just convert len to max_blocks because 3704 * We can't just convert len to max_blocks because
3666 * If blocksize = 4096 offset = 3072 and len = 2048 3705 * If blocksize = 4096 offset = 3072 and len = 2048
3667 */ 3706 */
3668 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) 3707 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3669 - block; 3708 - map.m_lblk;
3670 /* 3709 /*
3671 * credits to insert 1 extent into extent tree 3710 * credits to insert 1 extent into extent tree
3672 */ 3711 */
3673 credits = ext4_chunk_trans_blocks(inode, max_blocks); 3712 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3674 mutex_lock(&inode->i_mutex); 3713 mutex_lock(&inode->i_mutex);
3714 ret = inode_newsize_ok(inode, (len + offset));
3715 if (ret) {
3716 mutex_unlock(&inode->i_mutex);
3717 return ret;
3718 }
3675retry: 3719retry:
3676 while (ret >= 0 && ret < max_blocks) { 3720 while (ret >= 0 && ret < max_blocks) {
3677 block = block + ret; 3721 map.m_lblk = map.m_lblk + ret;
3678 max_blocks = max_blocks - ret; 3722 map.m_len = max_blocks = max_blocks - ret;
3679 handle = ext4_journal_start(inode, credits); 3723 handle = ext4_journal_start(inode, credits);
3680 if (IS_ERR(handle)) { 3724 if (IS_ERR(handle)) {
3681 ret = PTR_ERR(handle); 3725 ret = PTR_ERR(handle);
3682 break; 3726 break;
3683 } 3727 }
3684 map_bh.b_state = 0; 3728 ret = ext4_map_blocks(handle, inode, &map,
3685 ret = ext4_get_blocks(handle, inode, block,
3686 max_blocks, &map_bh,
3687 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT); 3729 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3688 if (ret <= 0) { 3730 if (ret <= 0) {
3689#ifdef EXT4FS_DEBUG 3731#ifdef EXT4FS_DEBUG
3690 WARN_ON(ret <= 0); 3732 WARN_ON(ret <= 0);
3691 printk(KERN_ERR "%s: ext4_ext_get_blocks " 3733 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3692 "returned error inode#%lu, block=%u, " 3734 "returned error inode#%lu, block=%u, "
3693 "max_blocks=%u", __func__, 3735 "max_blocks=%u", __func__,
3694 inode->i_ino, block, max_blocks); 3736 inode->i_ino, block, max_blocks);
@@ -3697,14 +3739,14 @@ retry:
3697 ret2 = ext4_journal_stop(handle); 3739 ret2 = ext4_journal_stop(handle);
3698 break; 3740 break;
3699 } 3741 }
3700 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len, 3742 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3701 blkbits) >> blkbits)) 3743 blkbits) >> blkbits))
3702 new_size = offset + len; 3744 new_size = offset + len;
3703 else 3745 else
3704 new_size = (block + ret) << blkbits; 3746 new_size = (map.m_lblk + ret) << blkbits;
3705 3747
3706 ext4_falloc_update_inode(inode, mode, new_size, 3748 ext4_falloc_update_inode(inode, mode, new_size,
3707 buffer_new(&map_bh)); 3749 (map.m_flags & EXT4_MAP_NEW));
3708 ext4_mark_inode_dirty(handle, inode); 3750 ext4_mark_inode_dirty(handle, inode);
3709 ret2 = ext4_journal_stop(handle); 3751 ret2 = ext4_journal_stop(handle);
3710 if (ret2) 3752 if (ret2)
@@ -3733,42 +3775,39 @@ int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3733 ssize_t len) 3775 ssize_t len)
3734{ 3776{
3735 handle_t *handle; 3777 handle_t *handle;
3736 ext4_lblk_t block;
3737 unsigned int max_blocks; 3778 unsigned int max_blocks;
3738 int ret = 0; 3779 int ret = 0;
3739 int ret2 = 0; 3780 int ret2 = 0;
3740 struct buffer_head map_bh; 3781 struct ext4_map_blocks map;
3741 unsigned int credits, blkbits = inode->i_blkbits; 3782 unsigned int credits, blkbits = inode->i_blkbits;
3742 3783
3743 block = offset >> blkbits; 3784 map.m_lblk = offset >> blkbits;
3744 /* 3785 /*
3745 * We can't just convert len to max_blocks because 3786 * We can't just convert len to max_blocks because
3746 * If blocksize = 4096 offset = 3072 and len = 2048 3787 * If blocksize = 4096 offset = 3072 and len = 2048
3747 */ 3788 */
3748 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) 3789 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3749 - block; 3790 map.m_lblk);
3750 /* 3791 /*
3751 * credits to insert 1 extent into extent tree 3792 * credits to insert 1 extent into extent tree
3752 */ 3793 */
3753 credits = ext4_chunk_trans_blocks(inode, max_blocks); 3794 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3754 while (ret >= 0 && ret < max_blocks) { 3795 while (ret >= 0 && ret < max_blocks) {
3755 block = block + ret; 3796 map.m_lblk += ret;
3756 max_blocks = max_blocks - ret; 3797 map.m_len = (max_blocks -= ret);
3757 handle = ext4_journal_start(inode, credits); 3798 handle = ext4_journal_start(inode, credits);
3758 if (IS_ERR(handle)) { 3799 if (IS_ERR(handle)) {
3759 ret = PTR_ERR(handle); 3800 ret = PTR_ERR(handle);
3760 break; 3801 break;
3761 } 3802 }
3762 map_bh.b_state = 0; 3803 ret = ext4_map_blocks(handle, inode, &map,
3763 ret = ext4_get_blocks(handle, inode, block,
3764 max_blocks, &map_bh,
3765 EXT4_GET_BLOCKS_IO_CONVERT_EXT); 3804 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3766 if (ret <= 0) { 3805 if (ret <= 0) {
3767 WARN_ON(ret <= 0); 3806 WARN_ON(ret <= 0);
3768 printk(KERN_ERR "%s: ext4_ext_get_blocks " 3807 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3769 "returned error inode#%lu, block=%u, " 3808 "returned error inode#%lu, block=%u, "
3770 "max_blocks=%u", __func__, 3809 "max_blocks=%u", __func__,
3771 inode->i_ino, block, max_blocks); 3810 inode->i_ino, map.m_lblk, map.m_len);
3772 } 3811 }
3773 ext4_mark_inode_dirty(handle, inode); 3812 ext4_mark_inode_dirty(handle, inode);
3774 ret2 = ext4_journal_stop(handle); 3813 ret2 = ext4_journal_stop(handle);
@@ -3898,7 +3937,7 @@ int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3898 int error = 0; 3937 int error = 0;
3899 3938
3900 /* fallback to generic here if not in extents fmt */ 3939 /* fallback to generic here if not in extents fmt */
3901 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 3940 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3902 return generic_block_fiemap(inode, fieinfo, start, len, 3941 return generic_block_fiemap(inode, fieinfo, start, len,
3903 ext4_get_block); 3942 ext4_get_block);
3904 3943
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index d0776e410f34..5313ae4cda2d 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -66,7 +66,7 @@ ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
66 * is smaller than s_maxbytes, which is for extent-mapped files. 66 * is smaller than s_maxbytes, which is for extent-mapped files.
67 */ 67 */
68 68
69 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) { 69 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
70 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 70 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
71 size_t length = iov_length(iov, nr_segs); 71 size_t length = iov_length(iov, nr_segs);
72 72
diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index ef3d980e67cb..b6a74f991bf4 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -35,6 +35,29 @@
35#include <trace/events/ext4.h> 35#include <trace/events/ext4.h>
36 36
37/* 37/*
38 * If we're not journaling and this is a just-created file, we have to
39 * sync our parent directory (if it was freshly created) since
40 * otherwise it will only be written by writeback, leaving a huge
41 * window during which a crash may lose the file. This may apply for
42 * the parent directory's parent as well, and so on recursively, if
43 * they are also freshly created.
44 */
45static void ext4_sync_parent(struct inode *inode)
46{
47 struct dentry *dentry = NULL;
48
49 while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
50 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
51 dentry = list_entry(inode->i_dentry.next,
52 struct dentry, d_alias);
53 if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
54 break;
55 inode = dentry->d_parent->d_inode;
56 sync_mapping_buffers(inode->i_mapping);
57 }
58}
59
60/*
38 * akpm: A new design for ext4_sync_file(). 61 * akpm: A new design for ext4_sync_file().
39 * 62 *
40 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync(). 63 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
@@ -66,9 +89,13 @@ int ext4_sync_file(struct file *file, struct dentry *dentry, int datasync)
66 ret = flush_completed_IO(inode); 89 ret = flush_completed_IO(inode);
67 if (ret < 0) 90 if (ret < 0)
68 return ret; 91 return ret;
69 92
70 if (!journal) 93 if (!journal) {
71 return simple_fsync(file, dentry, datasync); 94 ret = simple_fsync(file, dentry, datasync);
95 if (!ret && !list_empty(&inode->i_dentry))
96 ext4_sync_parent(inode);
97 return ret;
98 }
72 99
73 /* 100 /*
74 * data=writeback,ordered: 101 * data=writeback,ordered:
@@ -102,7 +129,7 @@ int ext4_sync_file(struct file *file, struct dentry *dentry, int datasync)
102 (journal->j_flags & JBD2_BARRIER)) 129 (journal->j_flags & JBD2_BARRIER))
103 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, 130 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL,
104 NULL, BLKDEV_IFL_WAIT); 131 NULL, BLKDEV_IFL_WAIT);
105 jbd2_log_wait_commit(journal, commit_tid); 132 ret = jbd2_log_wait_commit(journal, commit_tid);
106 } else if (journal->j_flags & JBD2_BARRIER) 133 } else if (journal->j_flags & JBD2_BARRIER)
107 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL, 134 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL,
108 BLKDEV_IFL_WAIT); 135 BLKDEV_IFL_WAIT);
diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index 1a0e183a2f04..25c4b3173fd9 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -240,56 +240,49 @@ void ext4_free_inode(handle_t *handle, struct inode *inode)
240 if (fatal) 240 if (fatal)
241 goto error_return; 241 goto error_return;
242 242
243 /* Ok, now we can actually update the inode bitmaps.. */ 243 fatal = -ESRCH;
244 cleared = ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group), 244 gdp = ext4_get_group_desc(sb, block_group, &bh2);
245 bit, bitmap_bh->b_data); 245 if (gdp) {
246 if (!cleared)
247 ext4_error(sb, "bit already cleared for inode %lu", ino);
248 else {
249 gdp = ext4_get_group_desc(sb, block_group, &bh2);
250
251 BUFFER_TRACE(bh2, "get_write_access"); 246 BUFFER_TRACE(bh2, "get_write_access");
252 fatal = ext4_journal_get_write_access(handle, bh2); 247 fatal = ext4_journal_get_write_access(handle, bh2);
253 if (fatal) goto error_return; 248 }
254 249 ext4_lock_group(sb, block_group);
255 if (gdp) { 250 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
256 ext4_lock_group(sb, block_group); 251 if (fatal || !cleared) {
257 count = ext4_free_inodes_count(sb, gdp) + 1; 252 ext4_unlock_group(sb, block_group);
258 ext4_free_inodes_set(sb, gdp, count); 253 goto out;
259 if (is_directory) { 254 }
260 count = ext4_used_dirs_count(sb, gdp) - 1;
261 ext4_used_dirs_set(sb, gdp, count);
262 if (sbi->s_log_groups_per_flex) {
263 ext4_group_t f;
264
265 f = ext4_flex_group(sbi, block_group);
266 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
267 }
268 255
269 } 256 count = ext4_free_inodes_count(sb, gdp) + 1;
270 gdp->bg_checksum = ext4_group_desc_csum(sbi, 257 ext4_free_inodes_set(sb, gdp, count);
271 block_group, gdp); 258 if (is_directory) {
272 ext4_unlock_group(sb, block_group); 259 count = ext4_used_dirs_count(sb, gdp) - 1;
273 percpu_counter_inc(&sbi->s_freeinodes_counter); 260 ext4_used_dirs_set(sb, gdp, count);
274 if (is_directory) 261 percpu_counter_dec(&sbi->s_dirs_counter);
275 percpu_counter_dec(&sbi->s_dirs_counter);
276
277 if (sbi->s_log_groups_per_flex) {
278 ext4_group_t f;
279
280 f = ext4_flex_group(sbi, block_group);
281 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
282 }
283 }
284 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
285 err = ext4_handle_dirty_metadata(handle, NULL, bh2);
286 if (!fatal) fatal = err;
287 } 262 }
288 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); 263 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
289 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); 264 ext4_unlock_group(sb, block_group);
290 if (!fatal) 265
291 fatal = err; 266 percpu_counter_inc(&sbi->s_freeinodes_counter);
292 sb->s_dirt = 1; 267 if (sbi->s_log_groups_per_flex) {
268 ext4_group_t f = ext4_flex_group(sbi, block_group);
269
270 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
271 if (is_directory)
272 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
273 }
274 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
275 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
276out:
277 if (cleared) {
278 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
279 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
280 if (!fatal)
281 fatal = err;
282 sb->s_dirt = 1;
283 } else
284 ext4_error(sb, "bit already cleared for inode %lu", ino);
285
293error_return: 286error_return:
294 brelse(bitmap_bh); 287 brelse(bitmap_bh);
295 ext4_std_error(sb, fatal); 288 ext4_std_error(sb, fatal);
@@ -499,7 +492,7 @@ static int find_group_orlov(struct super_block *sb, struct inode *parent,
499 492
500 if (S_ISDIR(mode) && 493 if (S_ISDIR(mode) &&
501 ((parent == sb->s_root->d_inode) || 494 ((parent == sb->s_root->d_inode) ||
502 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) { 495 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
503 int best_ndir = inodes_per_group; 496 int best_ndir = inodes_per_group;
504 int ret = -1; 497 int ret = -1;
505 498
@@ -1041,7 +1034,7 @@ got:
1041 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { 1034 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1042 /* set extent flag only for directory, file and normal symlink*/ 1035 /* set extent flag only for directory, file and normal symlink*/
1043 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { 1036 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1044 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; 1037 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1045 ext4_ext_tree_init(handle, inode); 1038 ext4_ext_tree_init(handle, inode);
1046 } 1039 }
1047 } 1040 }
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 3e0f6af9d08d..19df61c321fd 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -149,7 +149,7 @@ int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
149 int ret; 149 int ret;
150 150
151 /* 151 /*
152 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this 152 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
153 * moment, get_block can be called only for blocks inside i_size since 153 * moment, get_block can be called only for blocks inside i_size since
154 * page cache has been already dropped and writes are blocked by 154 * page cache has been already dropped and writes are blocked by
155 * i_mutex. So we can safely drop the i_data_sem here. 155 * i_mutex. So we can safely drop the i_data_sem here.
@@ -348,9 +348,8 @@ static int __ext4_check_blockref(const char *function, struct inode *inode,
348 if (blk && 348 if (blk &&
349 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb), 349 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
350 blk, 1))) { 350 blk, 1))) {
351 __ext4_error(inode->i_sb, function, 351 ext4_error_inode(function, inode,
352 "invalid block reference %u " 352 "invalid block reference %u", blk);
353 "in inode #%lu", blk, inode->i_ino);
354 return -EIO; 353 return -EIO;
355 } 354 }
356 } 355 }
@@ -785,7 +784,7 @@ failed:
785 /* Allocation failed, free what we already allocated */ 784 /* Allocation failed, free what we already allocated */
786 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0); 785 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
787 for (i = 1; i <= n ; i++) { 786 for (i = 1; i <= n ; i++) {
788 /* 787 /*
789 * branch[i].bh is newly allocated, so there is no 788 * branch[i].bh is newly allocated, so there is no
790 * need to revoke the block, which is why we don't 789 * need to revoke the block, which is why we don't
791 * need to set EXT4_FREE_BLOCKS_METADATA. 790 * need to set EXT4_FREE_BLOCKS_METADATA.
@@ -875,7 +874,7 @@ static int ext4_splice_branch(handle_t *handle, struct inode *inode,
875 874
876err_out: 875err_out:
877 for (i = 1; i <= num; i++) { 876 for (i = 1; i <= num; i++) {
878 /* 877 /*
879 * branch[i].bh is newly allocated, so there is no 878 * branch[i].bh is newly allocated, so there is no
880 * need to revoke the block, which is why we don't 879 * need to revoke the block, which is why we don't
881 * need to set EXT4_FREE_BLOCKS_METADATA. 880 * need to set EXT4_FREE_BLOCKS_METADATA.
@@ -890,9 +889,9 @@ err_out:
890} 889}
891 890
892/* 891/*
893 * The ext4_ind_get_blocks() function handles non-extents inodes 892 * The ext4_ind_map_blocks() function handles non-extents inodes
894 * (i.e., using the traditional indirect/double-indirect i_blocks 893 * (i.e., using the traditional indirect/double-indirect i_blocks
895 * scheme) for ext4_get_blocks(). 894 * scheme) for ext4_map_blocks().
896 * 895 *
897 * Allocation strategy is simple: if we have to allocate something, we will 896 * Allocation strategy is simple: if we have to allocate something, we will
898 * have to go the whole way to leaf. So let's do it before attaching anything 897 * have to go the whole way to leaf. So let's do it before attaching anything
@@ -917,9 +916,8 @@ err_out:
917 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system 916 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
918 * blocks. 917 * blocks.
919 */ 918 */
920static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode, 919static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
921 ext4_lblk_t iblock, unsigned int maxblocks, 920 struct ext4_map_blocks *map,
922 struct buffer_head *bh_result,
923 int flags) 921 int flags)
924{ 922{
925 int err = -EIO; 923 int err = -EIO;
@@ -933,9 +931,9 @@ static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
933 int count = 0; 931 int count = 0;
934 ext4_fsblk_t first_block = 0; 932 ext4_fsblk_t first_block = 0;
935 933
936 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); 934 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
937 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); 935 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
938 depth = ext4_block_to_path(inode, iblock, offsets, 936 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
939 &blocks_to_boundary); 937 &blocks_to_boundary);
940 938
941 if (depth == 0) 939 if (depth == 0)
@@ -946,10 +944,9 @@ static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
946 /* Simplest case - block found, no allocation needed */ 944 /* Simplest case - block found, no allocation needed */
947 if (!partial) { 945 if (!partial) {
948 first_block = le32_to_cpu(chain[depth - 1].key); 946 first_block = le32_to_cpu(chain[depth - 1].key);
949 clear_buffer_new(bh_result);
950 count++; 947 count++;
951 /*map more blocks*/ 948 /*map more blocks*/
952 while (count < maxblocks && count <= blocks_to_boundary) { 949 while (count < map->m_len && count <= blocks_to_boundary) {
953 ext4_fsblk_t blk; 950 ext4_fsblk_t blk;
954 951
955 blk = le32_to_cpu(*(chain[depth-1].p + count)); 952 blk = le32_to_cpu(*(chain[depth-1].p + count));
@@ -969,7 +966,7 @@ static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
969 /* 966 /*
970 * Okay, we need to do block allocation. 967 * Okay, we need to do block allocation.
971 */ 968 */
972 goal = ext4_find_goal(inode, iblock, partial); 969 goal = ext4_find_goal(inode, map->m_lblk, partial);
973 970
974 /* the number of blocks need to allocate for [d,t]indirect blocks */ 971 /* the number of blocks need to allocate for [d,t]indirect blocks */
975 indirect_blks = (chain + depth) - partial - 1; 972 indirect_blks = (chain + depth) - partial - 1;
@@ -979,11 +976,11 @@ static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
979 * direct blocks to allocate for this branch. 976 * direct blocks to allocate for this branch.
980 */ 977 */
981 count = ext4_blks_to_allocate(partial, indirect_blks, 978 count = ext4_blks_to_allocate(partial, indirect_blks,
982 maxblocks, blocks_to_boundary); 979 map->m_len, blocks_to_boundary);
983 /* 980 /*
984 * Block out ext4_truncate while we alter the tree 981 * Block out ext4_truncate while we alter the tree
985 */ 982 */
986 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks, 983 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
987 &count, goal, 984 &count, goal,
988 offsets + (partial - chain), partial); 985 offsets + (partial - chain), partial);
989 986
@@ -995,18 +992,20 @@ static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
995 * may need to return -EAGAIN upwards in the worst case. --sct 992 * may need to return -EAGAIN upwards in the worst case. --sct
996 */ 993 */
997 if (!err) 994 if (!err)
998 err = ext4_splice_branch(handle, inode, iblock, 995 err = ext4_splice_branch(handle, inode, map->m_lblk,
999 partial, indirect_blks, count); 996 partial, indirect_blks, count);
1000 if (err) 997 if (err)
1001 goto cleanup; 998 goto cleanup;
1002 999
1003 set_buffer_new(bh_result); 1000 map->m_flags |= EXT4_MAP_NEW;
1004 1001
1005 ext4_update_inode_fsync_trans(handle, inode, 1); 1002 ext4_update_inode_fsync_trans(handle, inode, 1);
1006got_it: 1003got_it:
1007 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); 1004 map->m_flags |= EXT4_MAP_MAPPED;
1005 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1006 map->m_len = count;
1008 if (count > blocks_to_boundary) 1007 if (count > blocks_to_boundary)
1009 set_buffer_boundary(bh_result); 1008 map->m_flags |= EXT4_MAP_BOUNDARY;
1010 err = count; 1009 err = count;
1011 /* Clean up and exit */ 1010 /* Clean up and exit */
1012 partial = chain + depth - 1; /* the whole chain */ 1011 partial = chain + depth - 1; /* the whole chain */
@@ -1016,7 +1015,6 @@ cleanup:
1016 brelse(partial->bh); 1015 brelse(partial->bh);
1017 partial--; 1016 partial--;
1018 } 1017 }
1019 BUFFER_TRACE(bh_result, "returned");
1020out: 1018out:
1021 return err; 1019 return err;
1022} 1020}
@@ -1061,7 +1059,7 @@ static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1061 */ 1059 */
1062static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock) 1060static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1063{ 1061{
1064 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) 1062 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1065 return ext4_ext_calc_metadata_amount(inode, lblock); 1063 return ext4_ext_calc_metadata_amount(inode, lblock);
1066 1064
1067 return ext4_indirect_calc_metadata_amount(inode, lblock); 1065 return ext4_indirect_calc_metadata_amount(inode, lblock);
@@ -1076,7 +1074,6 @@ void ext4_da_update_reserve_space(struct inode *inode,
1076{ 1074{
1077 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1075 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1078 struct ext4_inode_info *ei = EXT4_I(inode); 1076 struct ext4_inode_info *ei = EXT4_I(inode);
1079 int mdb_free = 0, allocated_meta_blocks = 0;
1080 1077
1081 spin_lock(&ei->i_block_reservation_lock); 1078 spin_lock(&ei->i_block_reservation_lock);
1082 trace_ext4_da_update_reserve_space(inode, used); 1079 trace_ext4_da_update_reserve_space(inode, used);
@@ -1091,11 +1088,10 @@ void ext4_da_update_reserve_space(struct inode *inode,
1091 1088
1092 /* Update per-inode reservations */ 1089 /* Update per-inode reservations */
1093 ei->i_reserved_data_blocks -= used; 1090 ei->i_reserved_data_blocks -= used;
1094 used += ei->i_allocated_meta_blocks;
1095 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; 1091 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1096 allocated_meta_blocks = ei->i_allocated_meta_blocks; 1092 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1093 used + ei->i_allocated_meta_blocks);
1097 ei->i_allocated_meta_blocks = 0; 1094 ei->i_allocated_meta_blocks = 0;
1098 percpu_counter_sub(&sbi->s_dirtyblocks_counter, used);
1099 1095
1100 if (ei->i_reserved_data_blocks == 0) { 1096 if (ei->i_reserved_data_blocks == 0) {
1101 /* 1097 /*
@@ -1103,30 +1099,23 @@ void ext4_da_update_reserve_space(struct inode *inode,
1103 * only when we have written all of the delayed 1099 * only when we have written all of the delayed
1104 * allocation blocks. 1100 * allocation blocks.
1105 */ 1101 */
1106 mdb_free = ei->i_reserved_meta_blocks; 1102 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1103 ei->i_reserved_meta_blocks);
1107 ei->i_reserved_meta_blocks = 0; 1104 ei->i_reserved_meta_blocks = 0;
1108 ei->i_da_metadata_calc_len = 0; 1105 ei->i_da_metadata_calc_len = 0;
1109 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1110 } 1106 }
1111 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 1107 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1112 1108
1113 /* Update quota subsystem */ 1109 /* Update quota subsystem for data blocks */
1114 if (quota_claim) { 1110 if (quota_claim)
1115 dquot_claim_block(inode, used); 1111 dquot_claim_block(inode, used);
1116 if (mdb_free) 1112 else {
1117 dquot_release_reservation_block(inode, mdb_free);
1118 } else {
1119 /* 1113 /*
1120 * We did fallocate with an offset that is already delayed 1114 * We did fallocate with an offset that is already delayed
1121 * allocated. So on delayed allocated writeback we should 1115 * allocated. So on delayed allocated writeback we should
1122 * not update the quota for allocated blocks. But then 1116 * not re-claim the quota for fallocated blocks.
1123 * converting an fallocate region to initialized region would
1124 * have caused a metadata allocation. So claim quota for
1125 * that
1126 */ 1117 */
1127 if (allocated_meta_blocks) 1118 dquot_release_reservation_block(inode, used);
1128 dquot_claim_block(inode, allocated_meta_blocks);
1129 dquot_release_reservation_block(inode, mdb_free + used);
1130 } 1119 }
1131 1120
1132 /* 1121 /*
@@ -1139,15 +1128,15 @@ void ext4_da_update_reserve_space(struct inode *inode,
1139 ext4_discard_preallocations(inode); 1128 ext4_discard_preallocations(inode);
1140} 1129}
1141 1130
1142static int check_block_validity(struct inode *inode, const char *msg, 1131static int check_block_validity(struct inode *inode, const char *func,
1143 sector_t logical, sector_t phys, int len) 1132 struct ext4_map_blocks *map)
1144{ 1133{
1145 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) { 1134 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1146 __ext4_error(inode->i_sb, msg, 1135 map->m_len)) {
1147 "inode #%lu logical block %llu mapped to %llu " 1136 ext4_error_inode(func, inode,
1148 "(size %d)", inode->i_ino, 1137 "lblock %lu mapped to illegal pblock %llu "
1149 (unsigned long long) logical, 1138 "(length %d)", (unsigned long) map->m_lblk,
1150 (unsigned long long) phys, len); 1139 map->m_pblk, map->m_len);
1151 return -EIO; 1140 return -EIO;
1152 } 1141 }
1153 return 0; 1142 return 0;
@@ -1212,15 +1201,15 @@ static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1212} 1201}
1213 1202
1214/* 1203/*
1215 * The ext4_get_blocks() function tries to look up the requested blocks, 1204 * The ext4_map_blocks() function tries to look up the requested blocks,
1216 * and returns if the blocks are already mapped. 1205 * and returns if the blocks are already mapped.
1217 * 1206 *
1218 * Otherwise it takes the write lock of the i_data_sem and allocate blocks 1207 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1219 * and store the allocated blocks in the result buffer head and mark it 1208 * and store the allocated blocks in the result buffer head and mark it
1220 * mapped. 1209 * mapped.
1221 * 1210 *
1222 * If file type is extents based, it will call ext4_ext_get_blocks(), 1211 * If file type is extents based, it will call ext4_ext_map_blocks(),
1223 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping 1212 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1224 * based files 1213 * based files
1225 * 1214 *
1226 * On success, it returns the number of blocks being mapped or allocate. 1215 * On success, it returns the number of blocks being mapped or allocate.
@@ -1233,35 +1222,29 @@ static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1233 * 1222 *
1234 * It returns the error in case of allocation failure. 1223 * It returns the error in case of allocation failure.
1235 */ 1224 */
1236int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block, 1225int ext4_map_blocks(handle_t *handle, struct inode *inode,
1237 unsigned int max_blocks, struct buffer_head *bh, 1226 struct ext4_map_blocks *map, int flags)
1238 int flags)
1239{ 1227{
1240 int retval; 1228 int retval;
1241 1229
1242 clear_buffer_mapped(bh); 1230 map->m_flags = 0;
1243 clear_buffer_unwritten(bh); 1231 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1244 1232 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1245 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u," 1233 (unsigned long) map->m_lblk);
1246 "logical block %lu\n", inode->i_ino, flags, max_blocks,
1247 (unsigned long)block);
1248 /* 1234 /*
1249 * Try to see if we can get the block without requesting a new 1235 * Try to see if we can get the block without requesting a new
1250 * file system block. 1236 * file system block.
1251 */ 1237 */
1252 down_read((&EXT4_I(inode)->i_data_sem)); 1238 down_read((&EXT4_I(inode)->i_data_sem));
1253 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { 1239 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1254 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, 1240 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1255 bh, 0);
1256 } else { 1241 } else {
1257 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks, 1242 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1258 bh, 0);
1259 } 1243 }
1260 up_read((&EXT4_I(inode)->i_data_sem)); 1244 up_read((&EXT4_I(inode)->i_data_sem));
1261 1245
1262 if (retval > 0 && buffer_mapped(bh)) { 1246 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1263 int ret = check_block_validity(inode, "file system corruption", 1247 int ret = check_block_validity(inode, __func__, map);
1264 block, bh->b_blocknr, retval);
1265 if (ret != 0) 1248 if (ret != 0)
1266 return ret; 1249 return ret;
1267 } 1250 }
@@ -1277,7 +1260,7 @@ int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1277 * ext4_ext_get_block() returns th create = 0 1260 * ext4_ext_get_block() returns th create = 0
1278 * with buffer head unmapped. 1261 * with buffer head unmapped.
1279 */ 1262 */
1280 if (retval > 0 && buffer_mapped(bh)) 1263 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1281 return retval; 1264 return retval;
1282 1265
1283 /* 1266 /*
@@ -1290,7 +1273,7 @@ int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1290 * of BH_Unwritten and BH_Mapped flags being simultaneously 1273 * of BH_Unwritten and BH_Mapped flags being simultaneously
1291 * set on the buffer_head. 1274 * set on the buffer_head.
1292 */ 1275 */
1293 clear_buffer_unwritten(bh); 1276 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1294 1277
1295 /* 1278 /*
1296 * New blocks allocate and/or writing to uninitialized extent 1279 * New blocks allocate and/or writing to uninitialized extent
@@ -1312,14 +1295,12 @@ int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1312 * We need to check for EXT4 here because migrate 1295 * We need to check for EXT4 here because migrate
1313 * could have changed the inode type in between 1296 * could have changed the inode type in between
1314 */ 1297 */
1315 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { 1298 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1316 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, 1299 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1317 bh, flags);
1318 } else { 1300 } else {
1319 retval = ext4_ind_get_blocks(handle, inode, block, 1301 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1320 max_blocks, bh, flags);
1321 1302
1322 if (retval > 0 && buffer_new(bh)) { 1303 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1323 /* 1304 /*
1324 * We allocated new blocks which will result in 1305 * We allocated new blocks which will result in
1325 * i_data's format changing. Force the migrate 1306 * i_data's format changing. Force the migrate
@@ -1342,10 +1323,10 @@ int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1342 EXT4_I(inode)->i_delalloc_reserved_flag = 0; 1323 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1343 1324
1344 up_write((&EXT4_I(inode)->i_data_sem)); 1325 up_write((&EXT4_I(inode)->i_data_sem));
1345 if (retval > 0 && buffer_mapped(bh)) { 1326 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1346 int ret = check_block_validity(inode, "file system " 1327 int ret = check_block_validity(inode,
1347 "corruption after allocation", 1328 "ext4_map_blocks_after_alloc",
1348 block, bh->b_blocknr, retval); 1329 map);
1349 if (ret != 0) 1330 if (ret != 0)
1350 return ret; 1331 return ret;
1351 } 1332 }
@@ -1355,109 +1336,109 @@ int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1355/* Maximum number of blocks we map for direct IO at once. */ 1336/* Maximum number of blocks we map for direct IO at once. */
1356#define DIO_MAX_BLOCKS 4096 1337#define DIO_MAX_BLOCKS 4096
1357 1338
1358int ext4_get_block(struct inode *inode, sector_t iblock, 1339static int _ext4_get_block(struct inode *inode, sector_t iblock,
1359 struct buffer_head *bh_result, int create) 1340 struct buffer_head *bh, int flags)
1360{ 1341{
1361 handle_t *handle = ext4_journal_current_handle(); 1342 handle_t *handle = ext4_journal_current_handle();
1343 struct ext4_map_blocks map;
1362 int ret = 0, started = 0; 1344 int ret = 0, started = 0;
1363 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1364 int dio_credits; 1345 int dio_credits;
1365 1346
1366 if (create && !handle) { 1347 map.m_lblk = iblock;
1348 map.m_len = bh->b_size >> inode->i_blkbits;
1349
1350 if (flags && !handle) {
1367 /* Direct IO write... */ 1351 /* Direct IO write... */
1368 if (max_blocks > DIO_MAX_BLOCKS) 1352 if (map.m_len > DIO_MAX_BLOCKS)
1369 max_blocks = DIO_MAX_BLOCKS; 1353 map.m_len = DIO_MAX_BLOCKS;
1370 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks); 1354 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1371 handle = ext4_journal_start(inode, dio_credits); 1355 handle = ext4_journal_start(inode, dio_credits);
1372 if (IS_ERR(handle)) { 1356 if (IS_ERR(handle)) {
1373 ret = PTR_ERR(handle); 1357 ret = PTR_ERR(handle);
1374 goto out; 1358 return ret;
1375 } 1359 }
1376 started = 1; 1360 started = 1;
1377 } 1361 }
1378 1362
1379 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result, 1363 ret = ext4_map_blocks(handle, inode, &map, flags);
1380 create ? EXT4_GET_BLOCKS_CREATE : 0);
1381 if (ret > 0) { 1364 if (ret > 0) {
1382 bh_result->b_size = (ret << inode->i_blkbits); 1365 map_bh(bh, inode->i_sb, map.m_pblk);
1366 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1367 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1383 ret = 0; 1368 ret = 0;
1384 } 1369 }
1385 if (started) 1370 if (started)
1386 ext4_journal_stop(handle); 1371 ext4_journal_stop(handle);
1387out:
1388 return ret; 1372 return ret;
1389} 1373}
1390 1374
1375int ext4_get_block(struct inode *inode, sector_t iblock,
1376 struct buffer_head *bh, int create)
1377{
1378 return _ext4_get_block(inode, iblock, bh,
1379 create ? EXT4_GET_BLOCKS_CREATE : 0);
1380}
1381
1391/* 1382/*
1392 * `handle' can be NULL if create is zero 1383 * `handle' can be NULL if create is zero
1393 */ 1384 */
1394struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, 1385struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1395 ext4_lblk_t block, int create, int *errp) 1386 ext4_lblk_t block, int create, int *errp)
1396{ 1387{
1397 struct buffer_head dummy; 1388 struct ext4_map_blocks map;
1389 struct buffer_head *bh;
1398 int fatal = 0, err; 1390 int fatal = 0, err;
1399 int flags = 0;
1400 1391
1401 J_ASSERT(handle != NULL || create == 0); 1392 J_ASSERT(handle != NULL || create == 0);
1402 1393
1403 dummy.b_state = 0; 1394 map.m_lblk = block;
1404 dummy.b_blocknr = -1000; 1395 map.m_len = 1;
1405 buffer_trace_init(&dummy.b_history); 1396 err = ext4_map_blocks(handle, inode, &map,
1406 if (create) 1397 create ? EXT4_GET_BLOCKS_CREATE : 0);
1407 flags |= EXT4_GET_BLOCKS_CREATE; 1398
1408 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags); 1399 if (err < 0)
1409 /* 1400 *errp = err;
1410 * ext4_get_blocks() returns number of blocks mapped. 0 in 1401 if (err <= 0)
1411 * case of a HOLE. 1402 return NULL;
1412 */ 1403 *errp = 0;
1413 if (err > 0) { 1404
1414 if (err > 1) 1405 bh = sb_getblk(inode->i_sb, map.m_pblk);
1415 WARN_ON(1); 1406 if (!bh) {
1416 err = 0; 1407 *errp = -EIO;
1408 return NULL;
1417 } 1409 }
1418 *errp = err; 1410 if (map.m_flags & EXT4_MAP_NEW) {
1419 if (!err && buffer_mapped(&dummy)) { 1411 J_ASSERT(create != 0);
1420 struct buffer_head *bh; 1412 J_ASSERT(handle != NULL);
1421 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1422 if (!bh) {
1423 *errp = -EIO;
1424 goto err;
1425 }
1426 if (buffer_new(&dummy)) {
1427 J_ASSERT(create != 0);
1428 J_ASSERT(handle != NULL);
1429 1413
1430 /* 1414 /*
1431 * Now that we do not always journal data, we should 1415 * Now that we do not always journal data, we should
1432 * keep in mind whether this should always journal the 1416 * keep in mind whether this should always journal the
1433 * new buffer as metadata. For now, regular file 1417 * new buffer as metadata. For now, regular file
1434 * writes use ext4_get_block instead, so it's not a 1418 * writes use ext4_get_block instead, so it's not a
1435 * problem. 1419 * problem.
1436 */ 1420 */
1437 lock_buffer(bh); 1421 lock_buffer(bh);
1438 BUFFER_TRACE(bh, "call get_create_access"); 1422 BUFFER_TRACE(bh, "call get_create_access");
1439 fatal = ext4_journal_get_create_access(handle, bh); 1423 fatal = ext4_journal_get_create_access(handle, bh);
1440 if (!fatal && !buffer_uptodate(bh)) { 1424 if (!fatal && !buffer_uptodate(bh)) {
1441 memset(bh->b_data, 0, inode->i_sb->s_blocksize); 1425 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1442 set_buffer_uptodate(bh); 1426 set_buffer_uptodate(bh);
1443 }
1444 unlock_buffer(bh);
1445 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1446 err = ext4_handle_dirty_metadata(handle, inode, bh);
1447 if (!fatal)
1448 fatal = err;
1449 } else {
1450 BUFFER_TRACE(bh, "not a new buffer");
1451 }
1452 if (fatal) {
1453 *errp = fatal;
1454 brelse(bh);
1455 bh = NULL;
1456 } 1427 }
1457 return bh; 1428 unlock_buffer(bh);
1429 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1430 err = ext4_handle_dirty_metadata(handle, inode, bh);
1431 if (!fatal)
1432 fatal = err;
1433 } else {
1434 BUFFER_TRACE(bh, "not a new buffer");
1458 } 1435 }
1459err: 1436 if (fatal) {
1460 return NULL; 1437 *errp = fatal;
1438 brelse(bh);
1439 bh = NULL;
1440 }
1441 return bh;
1461} 1442}
1462 1443
1463struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, 1444struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
@@ -1860,7 +1841,7 @@ static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1860 int retries = 0; 1841 int retries = 0;
1861 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1842 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1862 struct ext4_inode_info *ei = EXT4_I(inode); 1843 struct ext4_inode_info *ei = EXT4_I(inode);
1863 unsigned long md_needed, md_reserved; 1844 unsigned long md_needed;
1864 int ret; 1845 int ret;
1865 1846
1866 /* 1847 /*
@@ -1870,22 +1851,24 @@ static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1870 */ 1851 */
1871repeat: 1852repeat:
1872 spin_lock(&ei->i_block_reservation_lock); 1853 spin_lock(&ei->i_block_reservation_lock);
1873 md_reserved = ei->i_reserved_meta_blocks;
1874 md_needed = ext4_calc_metadata_amount(inode, lblock); 1854 md_needed = ext4_calc_metadata_amount(inode, lblock);
1875 trace_ext4_da_reserve_space(inode, md_needed); 1855 trace_ext4_da_reserve_space(inode, md_needed);
1876 spin_unlock(&ei->i_block_reservation_lock); 1856 spin_unlock(&ei->i_block_reservation_lock);
1877 1857
1878 /* 1858 /*
1879 * Make quota reservation here to prevent quota overflow 1859 * We will charge metadata quota at writeout time; this saves
1880 * later. Real quota accounting is done at pages writeout 1860 * us from metadata over-estimation, though we may go over by
1881 * time. 1861 * a small amount in the end. Here we just reserve for data.
1882 */ 1862 */
1883 ret = dquot_reserve_block(inode, md_needed + 1); 1863 ret = dquot_reserve_block(inode, 1);
1884 if (ret) 1864 if (ret)
1885 return ret; 1865 return ret;
1886 1866 /*
1867 * We do still charge estimated metadata to the sb though;
1868 * we cannot afford to run out of free blocks.
1869 */
1887 if (ext4_claim_free_blocks(sbi, md_needed + 1)) { 1870 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1888 dquot_release_reservation_block(inode, md_needed + 1); 1871 dquot_release_reservation_block(inode, 1);
1889 if (ext4_should_retry_alloc(inode->i_sb, &retries)) { 1872 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1890 yield(); 1873 yield();
1891 goto repeat; 1874 goto repeat;
@@ -1910,6 +1893,7 @@ static void ext4_da_release_space(struct inode *inode, int to_free)
1910 1893
1911 spin_lock(&EXT4_I(inode)->i_block_reservation_lock); 1894 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1912 1895
1896 trace_ext4_da_release_space(inode, to_free);
1913 if (unlikely(to_free > ei->i_reserved_data_blocks)) { 1897 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1914 /* 1898 /*
1915 * if there aren't enough reserved blocks, then the 1899 * if there aren't enough reserved blocks, then the
@@ -1932,12 +1916,13 @@ static void ext4_da_release_space(struct inode *inode, int to_free)
1932 * only when we have written all of the delayed 1916 * only when we have written all of the delayed
1933 * allocation blocks. 1917 * allocation blocks.
1934 */ 1918 */
1935 to_free += ei->i_reserved_meta_blocks; 1919 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1920 ei->i_reserved_meta_blocks);
1936 ei->i_reserved_meta_blocks = 0; 1921 ei->i_reserved_meta_blocks = 0;
1937 ei->i_da_metadata_calc_len = 0; 1922 ei->i_da_metadata_calc_len = 0;
1938 } 1923 }
1939 1924
1940 /* update fs dirty blocks counter */ 1925 /* update fs dirty data blocks counter */
1941 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free); 1926 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1942 1927
1943 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 1928 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
@@ -2042,28 +2027,23 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd)
2042/* 2027/*
2043 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers 2028 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2044 * 2029 *
2045 * @mpd->inode - inode to walk through
2046 * @exbh->b_blocknr - first block on a disk
2047 * @exbh->b_size - amount of space in bytes
2048 * @logical - first logical block to start assignment with
2049 *
2050 * the function goes through all passed space and put actual disk 2030 * the function goes through all passed space and put actual disk
2051 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten 2031 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2052 */ 2032 */
2053static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical, 2033static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd,
2054 struct buffer_head *exbh) 2034 struct ext4_map_blocks *map)
2055{ 2035{
2056 struct inode *inode = mpd->inode; 2036 struct inode *inode = mpd->inode;
2057 struct address_space *mapping = inode->i_mapping; 2037 struct address_space *mapping = inode->i_mapping;
2058 int blocks = exbh->b_size >> inode->i_blkbits; 2038 int blocks = map->m_len;
2059 sector_t pblock = exbh->b_blocknr, cur_logical; 2039 sector_t pblock = map->m_pblk, cur_logical;
2060 struct buffer_head *head, *bh; 2040 struct buffer_head *head, *bh;
2061 pgoff_t index, end; 2041 pgoff_t index, end;
2062 struct pagevec pvec; 2042 struct pagevec pvec;
2063 int nr_pages, i; 2043 int nr_pages, i;
2064 2044
2065 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); 2045 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2066 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits); 2046 end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2067 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 2047 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2068 2048
2069 pagevec_init(&pvec, 0); 2049 pagevec_init(&pvec, 0);
@@ -2090,17 +2070,16 @@ static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2090 2070
2091 /* skip blocks out of the range */ 2071 /* skip blocks out of the range */
2092 do { 2072 do {
2093 if (cur_logical >= logical) 2073 if (cur_logical >= map->m_lblk)
2094 break; 2074 break;
2095 cur_logical++; 2075 cur_logical++;
2096 } while ((bh = bh->b_this_page) != head); 2076 } while ((bh = bh->b_this_page) != head);
2097 2077
2098 do { 2078 do {
2099 if (cur_logical >= logical + blocks) 2079 if (cur_logical >= map->m_lblk + blocks)
2100 break; 2080 break;
2101 2081
2102 if (buffer_delay(bh) || 2082 if (buffer_delay(bh) || buffer_unwritten(bh)) {
2103 buffer_unwritten(bh)) {
2104 2083
2105 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev); 2084 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2106 2085
@@ -2119,7 +2098,7 @@ static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2119 } else if (buffer_mapped(bh)) 2098 } else if (buffer_mapped(bh))
2120 BUG_ON(bh->b_blocknr != pblock); 2099 BUG_ON(bh->b_blocknr != pblock);
2121 2100
2122 if (buffer_uninit(exbh)) 2101 if (map->m_flags & EXT4_MAP_UNINIT)
2123 set_buffer_uninit(bh); 2102 set_buffer_uninit(bh);
2124 cur_logical++; 2103 cur_logical++;
2125 pblock++; 2104 pblock++;
@@ -2130,21 +2109,6 @@ static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2130} 2109}
2131 2110
2132 2111
2133/*
2134 * __unmap_underlying_blocks - just a helper function to unmap
2135 * set of blocks described by @bh
2136 */
2137static inline void __unmap_underlying_blocks(struct inode *inode,
2138 struct buffer_head *bh)
2139{
2140 struct block_device *bdev = inode->i_sb->s_bdev;
2141 int blocks, i;
2142
2143 blocks = bh->b_size >> inode->i_blkbits;
2144 for (i = 0; i < blocks; i++)
2145 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2146}
2147
2148static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, 2112static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2149 sector_t logical, long blk_cnt) 2113 sector_t logical, long blk_cnt)
2150{ 2114{
@@ -2206,7 +2170,7 @@ static void ext4_print_free_blocks(struct inode *inode)
2206static int mpage_da_map_blocks(struct mpage_da_data *mpd) 2170static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2207{ 2171{
2208 int err, blks, get_blocks_flags; 2172 int err, blks, get_blocks_flags;
2209 struct buffer_head new; 2173 struct ext4_map_blocks map;
2210 sector_t next = mpd->b_blocknr; 2174 sector_t next = mpd->b_blocknr;
2211 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; 2175 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2212 loff_t disksize = EXT4_I(mpd->inode)->i_disksize; 2176 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
@@ -2247,15 +2211,15 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2247 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting 2211 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2248 * variables are updated after the blocks have been allocated. 2212 * variables are updated after the blocks have been allocated.
2249 */ 2213 */
2250 new.b_state = 0; 2214 map.m_lblk = next;
2215 map.m_len = max_blocks;
2251 get_blocks_flags = EXT4_GET_BLOCKS_CREATE; 2216 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2252 if (ext4_should_dioread_nolock(mpd->inode)) 2217 if (ext4_should_dioread_nolock(mpd->inode))
2253 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; 2218 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2254 if (mpd->b_state & (1 << BH_Delay)) 2219 if (mpd->b_state & (1 << BH_Delay))
2255 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; 2220 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2256 2221
2257 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks, 2222 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2258 &new, get_blocks_flags);
2259 if (blks < 0) { 2223 if (blks < 0) {
2260 err = blks; 2224 err = blks;
2261 /* 2225 /*
@@ -2282,7 +2246,7 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2282 ext4_msg(mpd->inode->i_sb, KERN_CRIT, 2246 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2283 "delayed block allocation failed for inode %lu at " 2247 "delayed block allocation failed for inode %lu at "
2284 "logical offset %llu with max blocks %zd with " 2248 "logical offset %llu with max blocks %zd with "
2285 "error %d\n", mpd->inode->i_ino, 2249 "error %d", mpd->inode->i_ino,
2286 (unsigned long long) next, 2250 (unsigned long long) next,
2287 mpd->b_size >> mpd->inode->i_blkbits, err); 2251 mpd->b_size >> mpd->inode->i_blkbits, err);
2288 printk(KERN_CRIT "This should not happen!! " 2252 printk(KERN_CRIT "This should not happen!! "
@@ -2297,10 +2261,13 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2297 } 2261 }
2298 BUG_ON(blks == 0); 2262 BUG_ON(blks == 0);
2299 2263
2300 new.b_size = (blks << mpd->inode->i_blkbits); 2264 if (map.m_flags & EXT4_MAP_NEW) {
2265 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2266 int i;
2301 2267
2302 if (buffer_new(&new)) 2268 for (i = 0; i < map.m_len; i++)
2303 __unmap_underlying_blocks(mpd->inode, &new); 2269 unmap_underlying_metadata(bdev, map.m_pblk + i);
2270 }
2304 2271
2305 /* 2272 /*
2306 * If blocks are delayed marked, we need to 2273 * If blocks are delayed marked, we need to
@@ -2308,7 +2275,7 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2308 */ 2275 */
2309 if ((mpd->b_state & (1 << BH_Delay)) || 2276 if ((mpd->b_state & (1 << BH_Delay)) ||
2310 (mpd->b_state & (1 << BH_Unwritten))) 2277 (mpd->b_state & (1 << BH_Unwritten)))
2311 mpage_put_bnr_to_bhs(mpd, next, &new); 2278 mpage_put_bnr_to_bhs(mpd, &map);
2312 2279
2313 if (ext4_should_order_data(mpd->inode)) { 2280 if (ext4_should_order_data(mpd->inode)) {
2314 err = ext4_jbd2_file_inode(handle, mpd->inode); 2281 err = ext4_jbd2_file_inode(handle, mpd->inode);
@@ -2349,8 +2316,17 @@ static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2349 sector_t next; 2316 sector_t next;
2350 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; 2317 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2351 2318
2319 /*
2320 * XXX Don't go larger than mballoc is willing to allocate
2321 * This is a stopgap solution. We eventually need to fold
2322 * mpage_da_submit_io() into this function and then call
2323 * ext4_get_blocks() multiple times in a loop
2324 */
2325 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2326 goto flush_it;
2327
2352 /* check if thereserved journal credits might overflow */ 2328 /* check if thereserved journal credits might overflow */
2353 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) { 2329 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2354 if (nrblocks >= EXT4_MAX_TRANS_DATA) { 2330 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2355 /* 2331 /*
2356 * With non-extent format we are limited by the journal 2332 * With non-extent format we are limited by the journal
@@ -2423,17 +2399,6 @@ static int __mpage_da_writepage(struct page *page,
2423 struct buffer_head *bh, *head; 2399 struct buffer_head *bh, *head;
2424 sector_t logical; 2400 sector_t logical;
2425 2401
2426 if (mpd->io_done) {
2427 /*
2428 * Rest of the page in the page_vec
2429 * redirty then and skip then. We will
2430 * try to write them again after
2431 * starting a new transaction
2432 */
2433 redirty_page_for_writepage(wbc, page);
2434 unlock_page(page);
2435 return MPAGE_DA_EXTENT_TAIL;
2436 }
2437 /* 2402 /*
2438 * Can we merge this page to current extent? 2403 * Can we merge this page to current extent?
2439 */ 2404 */
@@ -2528,8 +2493,9 @@ static int __mpage_da_writepage(struct page *page,
2528 * initialized properly. 2493 * initialized properly.
2529 */ 2494 */
2530static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, 2495static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2531 struct buffer_head *bh_result, int create) 2496 struct buffer_head *bh, int create)
2532{ 2497{
2498 struct ext4_map_blocks map;
2533 int ret = 0; 2499 int ret = 0;
2534 sector_t invalid_block = ~((sector_t) 0xffff); 2500 sector_t invalid_block = ~((sector_t) 0xffff);
2535 2501
@@ -2537,16 +2503,22 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2537 invalid_block = ~0; 2503 invalid_block = ~0;
2538 2504
2539 BUG_ON(create == 0); 2505 BUG_ON(create == 0);
2540 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); 2506 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2507
2508 map.m_lblk = iblock;
2509 map.m_len = 1;
2541 2510
2542 /* 2511 /*
2543 * first, we need to know whether the block is allocated already 2512 * first, we need to know whether the block is allocated already
2544 * preallocated blocks are unmapped but should treated 2513 * preallocated blocks are unmapped but should treated
2545 * the same as allocated blocks. 2514 * the same as allocated blocks.
2546 */ 2515 */
2547 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0); 2516 ret = ext4_map_blocks(NULL, inode, &map, 0);
2548 if ((ret == 0) && !buffer_delay(bh_result)) { 2517 if (ret < 0)
2549 /* the block isn't (pre)allocated yet, let's reserve space */ 2518 return ret;
2519 if (ret == 0) {
2520 if (buffer_delay(bh))
2521 return 0; /* Not sure this could or should happen */
2550 /* 2522 /*
2551 * XXX: __block_prepare_write() unmaps passed block, 2523 * XXX: __block_prepare_write() unmaps passed block,
2552 * is it OK? 2524 * is it OK?
@@ -2556,26 +2528,26 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2556 /* not enough space to reserve */ 2528 /* not enough space to reserve */
2557 return ret; 2529 return ret;
2558 2530
2559 map_bh(bh_result, inode->i_sb, invalid_block); 2531 map_bh(bh, inode->i_sb, invalid_block);
2560 set_buffer_new(bh_result); 2532 set_buffer_new(bh);
2561 set_buffer_delay(bh_result); 2533 set_buffer_delay(bh);
2562 } else if (ret > 0) { 2534 return 0;
2563 bh_result->b_size = (ret << inode->i_blkbits);
2564 if (buffer_unwritten(bh_result)) {
2565 /* A delayed write to unwritten bh should
2566 * be marked new and mapped. Mapped ensures
2567 * that we don't do get_block multiple times
2568 * when we write to the same offset and new
2569 * ensures that we do proper zero out for
2570 * partial write.
2571 */
2572 set_buffer_new(bh_result);
2573 set_buffer_mapped(bh_result);
2574 }
2575 ret = 0;
2576 } 2535 }
2577 2536
2578 return ret; 2537 map_bh(bh, inode->i_sb, map.m_pblk);
2538 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2539
2540 if (buffer_unwritten(bh)) {
2541 /* A delayed write to unwritten bh should be marked
2542 * new and mapped. Mapped ensures that we don't do
2543 * get_block multiple times when we write to the same
2544 * offset and new ensures that we do proper zero out
2545 * for partial write.
2546 */
2547 set_buffer_new(bh);
2548 set_buffer_mapped(bh);
2549 }
2550 return 0;
2579} 2551}
2580 2552
2581/* 2553/*
@@ -2597,21 +2569,8 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2597static int noalloc_get_block_write(struct inode *inode, sector_t iblock, 2569static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2598 struct buffer_head *bh_result, int create) 2570 struct buffer_head *bh_result, int create)
2599{ 2571{
2600 int ret = 0;
2601 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2602
2603 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); 2572 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2604 2573 return _ext4_get_block(inode, iblock, bh_result, 0);
2605 /*
2606 * we don't want to do block allocation in writepage
2607 * so call get_block_wrap with create = 0
2608 */
2609 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2610 if (ret > 0) {
2611 bh_result->b_size = (ret << inode->i_blkbits);
2612 ret = 0;
2613 }
2614 return ret;
2615} 2574}
2616 2575
2617static int bget_one(handle_t *handle, struct buffer_head *bh) 2576static int bget_one(handle_t *handle, struct buffer_head *bh)
@@ -2821,13 +2780,131 @@ static int ext4_da_writepages_trans_blocks(struct inode *inode)
2821 * number of contiguous block. So we will limit 2780 * number of contiguous block. So we will limit
2822 * number of contiguous block to a sane value 2781 * number of contiguous block to a sane value
2823 */ 2782 */
2824 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) && 2783 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2825 (max_blocks > EXT4_MAX_TRANS_DATA)) 2784 (max_blocks > EXT4_MAX_TRANS_DATA))
2826 max_blocks = EXT4_MAX_TRANS_DATA; 2785 max_blocks = EXT4_MAX_TRANS_DATA;
2827 2786
2828 return ext4_chunk_trans_blocks(inode, max_blocks); 2787 return ext4_chunk_trans_blocks(inode, max_blocks);
2829} 2788}
2830 2789
2790/*
2791 * write_cache_pages_da - walk the list of dirty pages of the given
2792 * address space and call the callback function (which usually writes
2793 * the pages).
2794 *
2795 * This is a forked version of write_cache_pages(). Differences:
2796 * Range cyclic is ignored.
2797 * no_nrwrite_index_update is always presumed true
2798 */
2799static int write_cache_pages_da(struct address_space *mapping,
2800 struct writeback_control *wbc,
2801 struct mpage_da_data *mpd)
2802{
2803 int ret = 0;
2804 int done = 0;
2805 struct pagevec pvec;
2806 int nr_pages;
2807 pgoff_t index;
2808 pgoff_t end; /* Inclusive */
2809 long nr_to_write = wbc->nr_to_write;
2810
2811 pagevec_init(&pvec, 0);
2812 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2813 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2814
2815 while (!done && (index <= end)) {
2816 int i;
2817
2818 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2819 PAGECACHE_TAG_DIRTY,
2820 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2821 if (nr_pages == 0)
2822 break;
2823
2824 for (i = 0; i < nr_pages; i++) {
2825 struct page *page = pvec.pages[i];
2826
2827 /*
2828 * At this point, the page may be truncated or
2829 * invalidated (changing page->mapping to NULL), or
2830 * even swizzled back from swapper_space to tmpfs file
2831 * mapping. However, page->index will not change
2832 * because we have a reference on the page.
2833 */
2834 if (page->index > end) {
2835 done = 1;
2836 break;
2837 }
2838
2839 lock_page(page);
2840
2841 /*
2842 * Page truncated or invalidated. We can freely skip it
2843 * then, even for data integrity operations: the page
2844 * has disappeared concurrently, so there could be no
2845 * real expectation of this data interity operation
2846 * even if there is now a new, dirty page at the same
2847 * pagecache address.
2848 */
2849 if (unlikely(page->mapping != mapping)) {
2850continue_unlock:
2851 unlock_page(page);
2852 continue;
2853 }
2854
2855 if (!PageDirty(page)) {
2856 /* someone wrote it for us */
2857 goto continue_unlock;
2858 }
2859
2860 if (PageWriteback(page)) {
2861 if (wbc->sync_mode != WB_SYNC_NONE)
2862 wait_on_page_writeback(page);
2863 else
2864 goto continue_unlock;
2865 }
2866
2867 BUG_ON(PageWriteback(page));
2868 if (!clear_page_dirty_for_io(page))
2869 goto continue_unlock;
2870
2871 ret = __mpage_da_writepage(page, wbc, mpd);
2872 if (unlikely(ret)) {
2873 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2874 unlock_page(page);
2875 ret = 0;
2876 } else {
2877 done = 1;
2878 break;
2879 }
2880 }
2881
2882 if (nr_to_write > 0) {
2883 nr_to_write--;
2884 if (nr_to_write == 0 &&
2885 wbc->sync_mode == WB_SYNC_NONE) {
2886 /*
2887 * We stop writing back only if we are
2888 * not doing integrity sync. In case of
2889 * integrity sync we have to keep going
2890 * because someone may be concurrently
2891 * dirtying pages, and we might have
2892 * synced a lot of newly appeared dirty
2893 * pages, but have not synced all of the
2894 * old dirty pages.
2895 */
2896 done = 1;
2897 break;
2898 }
2899 }
2900 }
2901 pagevec_release(&pvec);
2902 cond_resched();
2903 }
2904 return ret;
2905}
2906
2907
2831static int ext4_da_writepages(struct address_space *mapping, 2908static int ext4_da_writepages(struct address_space *mapping,
2832 struct writeback_control *wbc) 2909 struct writeback_control *wbc)
2833{ 2910{
@@ -2836,7 +2913,6 @@ static int ext4_da_writepages(struct address_space *mapping,
2836 handle_t *handle = NULL; 2913 handle_t *handle = NULL;
2837 struct mpage_da_data mpd; 2914 struct mpage_da_data mpd;
2838 struct inode *inode = mapping->host; 2915 struct inode *inode = mapping->host;
2839 int no_nrwrite_index_update;
2840 int pages_written = 0; 2916 int pages_written = 0;
2841 long pages_skipped; 2917 long pages_skipped;
2842 unsigned int max_pages; 2918 unsigned int max_pages;
@@ -2916,12 +2992,6 @@ static int ext4_da_writepages(struct address_space *mapping,
2916 mpd.wbc = wbc; 2992 mpd.wbc = wbc;
2917 mpd.inode = mapping->host; 2993 mpd.inode = mapping->host;
2918 2994
2919 /*
2920 * we don't want write_cache_pages to update
2921 * nr_to_write and writeback_index
2922 */
2923 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2924 wbc->no_nrwrite_index_update = 1;
2925 pages_skipped = wbc->pages_skipped; 2995 pages_skipped = wbc->pages_skipped;
2926 2996
2927retry: 2997retry:
@@ -2941,7 +3011,7 @@ retry:
2941 if (IS_ERR(handle)) { 3011 if (IS_ERR(handle)) {
2942 ret = PTR_ERR(handle); 3012 ret = PTR_ERR(handle);
2943 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " 3013 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2944 "%ld pages, ino %lu; err %d\n", __func__, 3014 "%ld pages, ino %lu; err %d", __func__,
2945 wbc->nr_to_write, inode->i_ino, ret); 3015 wbc->nr_to_write, inode->i_ino, ret);
2946 goto out_writepages; 3016 goto out_writepages;
2947 } 3017 }
@@ -2963,8 +3033,7 @@ retry:
2963 mpd.io_done = 0; 3033 mpd.io_done = 0;
2964 mpd.pages_written = 0; 3034 mpd.pages_written = 0;
2965 mpd.retval = 0; 3035 mpd.retval = 0;
2966 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, 3036 ret = write_cache_pages_da(mapping, wbc, &mpd);
2967 &mpd);
2968 /* 3037 /*
2969 * If we have a contiguous extent of pages and we 3038 * If we have a contiguous extent of pages and we
2970 * haven't done the I/O yet, map the blocks and submit 3039 * haven't done the I/O yet, map the blocks and submit
@@ -3016,7 +3085,7 @@ retry:
3016 if (pages_skipped != wbc->pages_skipped) 3085 if (pages_skipped != wbc->pages_skipped)
3017 ext4_msg(inode->i_sb, KERN_CRIT, 3086 ext4_msg(inode->i_sb, KERN_CRIT,
3018 "This should not happen leaving %s " 3087 "This should not happen leaving %s "
3019 "with nr_to_write = %ld ret = %d\n", 3088 "with nr_to_write = %ld ret = %d",
3020 __func__, wbc->nr_to_write, ret); 3089 __func__, wbc->nr_to_write, ret);
3021 3090
3022 /* Update index */ 3091 /* Update index */
@@ -3030,8 +3099,6 @@ retry:
3030 mapping->writeback_index = index; 3099 mapping->writeback_index = index;
3031 3100
3032out_writepages: 3101out_writepages:
3033 if (!no_nrwrite_index_update)
3034 wbc->no_nrwrite_index_update = 0;
3035 wbc->nr_to_write -= nr_to_writebump; 3102 wbc->nr_to_write -= nr_to_writebump;
3036 wbc->range_start = range_start; 3103 wbc->range_start = range_start;
3037 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); 3104 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
@@ -3076,7 +3143,7 @@ static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3076 loff_t pos, unsigned len, unsigned flags, 3143 loff_t pos, unsigned len, unsigned flags,
3077 struct page **pagep, void **fsdata) 3144 struct page **pagep, void **fsdata)
3078{ 3145{
3079 int ret, retries = 0, quota_retries = 0; 3146 int ret, retries = 0;
3080 struct page *page; 3147 struct page *page;
3081 pgoff_t index; 3148 pgoff_t index;
3082 unsigned from, to; 3149 unsigned from, to;
@@ -3135,22 +3202,6 @@ retry:
3135 3202
3136 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 3203 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3137 goto retry; 3204 goto retry;
3138
3139 if ((ret == -EDQUOT) &&
3140 EXT4_I(inode)->i_reserved_meta_blocks &&
3141 (quota_retries++ < 3)) {
3142 /*
3143 * Since we often over-estimate the number of meta
3144 * data blocks required, we may sometimes get a
3145 * spurios out of quota error even though there would
3146 * be enough space once we write the data blocks and
3147 * find out how many meta data blocks were _really_
3148 * required. So try forcing the inode write to see if
3149 * that helps.
3150 */
3151 write_inode_now(inode, (quota_retries == 3));
3152 goto retry;
3153 }
3154out: 3205out:
3155 return ret; 3206 return ret;
3156} 3207}
@@ -3546,46 +3597,18 @@ out:
3546 return ret; 3597 return ret;
3547} 3598}
3548 3599
3600/*
3601 * ext4_get_block used when preparing for a DIO write or buffer write.
3602 * We allocate an uinitialized extent if blocks haven't been allocated.
3603 * The extent will be converted to initialized after the IO is complete.
3604 */
3549static int ext4_get_block_write(struct inode *inode, sector_t iblock, 3605static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3550 struct buffer_head *bh_result, int create) 3606 struct buffer_head *bh_result, int create)
3551{ 3607{
3552 handle_t *handle = ext4_journal_current_handle();
3553 int ret = 0;
3554 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3555 int dio_credits;
3556 int started = 0;
3557
3558 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", 3608 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3559 inode->i_ino, create); 3609 inode->i_ino, create);
3560 /* 3610 return _ext4_get_block(inode, iblock, bh_result,
3561 * ext4_get_block in prepare for a DIO write or buffer write. 3611 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3562 * We allocate an uinitialized extent if blocks haven't been allocated.
3563 * The extent will be converted to initialized after IO complete.
3564 */
3565 create = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3566
3567 if (!handle) {
3568 if (max_blocks > DIO_MAX_BLOCKS)
3569 max_blocks = DIO_MAX_BLOCKS;
3570 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3571 handle = ext4_journal_start(inode, dio_credits);
3572 if (IS_ERR(handle)) {
3573 ret = PTR_ERR(handle);
3574 goto out;
3575 }
3576 started = 1;
3577 }
3578
3579 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3580 create);
3581 if (ret > 0) {
3582 bh_result->b_size = (ret << inode->i_blkbits);
3583 ret = 0;
3584 }
3585 if (started)
3586 ext4_journal_stop(handle);
3587out:
3588 return ret;
3589} 3612}
3590 3613
3591static void dump_completed_IO(struct inode * inode) 3614static void dump_completed_IO(struct inode * inode)
@@ -3973,7 +3996,7 @@ static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3973 struct file *file = iocb->ki_filp; 3996 struct file *file = iocb->ki_filp;
3974 struct inode *inode = file->f_mapping->host; 3997 struct inode *inode = file->f_mapping->host;
3975 3998
3976 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) 3999 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3977 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); 4000 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3978 4001
3979 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); 4002 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
@@ -4302,10 +4325,9 @@ static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4302 4325
4303 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, 4326 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4304 count)) { 4327 count)) {
4305 ext4_error(inode->i_sb, "inode #%lu: " 4328 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4306 "attempt to clear blocks %llu len %lu, invalid", 4329 "blocks %llu len %lu",
4307 inode->i_ino, (unsigned long long) block_to_free, 4330 (unsigned long long) block_to_free, count);
4308 count);
4309 return 1; 4331 return 1;
4310 } 4332 }
4311 4333
@@ -4410,11 +4432,10 @@ static void ext4_free_data(handle_t *handle, struct inode *inode,
4410 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) 4432 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4411 ext4_handle_dirty_metadata(handle, inode, this_bh); 4433 ext4_handle_dirty_metadata(handle, inode, this_bh);
4412 else 4434 else
4413 ext4_error(inode->i_sb, 4435 EXT4_ERROR_INODE(inode,
4414 "circular indirect block detected, " 4436 "circular indirect block detected at "
4415 "inode=%lu, block=%llu", 4437 "block %llu",
4416 inode->i_ino, 4438 (unsigned long long) this_bh->b_blocknr);
4417 (unsigned long long) this_bh->b_blocknr);
4418 } 4439 }
4419} 4440}
4420 4441
@@ -4452,11 +4473,10 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
4452 4473
4453 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), 4474 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4454 nr, 1)) { 4475 nr, 1)) {
4455 ext4_error(inode->i_sb, 4476 EXT4_ERROR_INODE(inode,
4456 "indirect mapped block in inode " 4477 "invalid indirect mapped "
4457 "#%lu invalid (level %d, blk #%lu)", 4478 "block %lu (level %d)",
4458 inode->i_ino, depth, 4479 (unsigned long) nr, depth);
4459 (unsigned long) nr);
4460 break; 4480 break;
4461 } 4481 }
4462 4482
@@ -4468,9 +4488,9 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
4468 * (should be rare). 4488 * (should be rare).
4469 */ 4489 */
4470 if (!bh) { 4490 if (!bh) {
4471 ext4_error(inode->i_sb, 4491 EXT4_ERROR_INODE(inode,
4472 "Read failure, inode=%lu, block=%llu", 4492 "Read failure block=%llu",
4473 inode->i_ino, nr); 4493 (unsigned long long) nr);
4474 continue; 4494 continue;
4475 } 4495 }
4476 4496
@@ -4612,12 +4632,12 @@ void ext4_truncate(struct inode *inode)
4612 if (!ext4_can_truncate(inode)) 4632 if (!ext4_can_truncate(inode))
4613 return; 4633 return;
4614 4634
4615 EXT4_I(inode)->i_flags &= ~EXT4_EOFBLOCKS_FL; 4635 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4616 4636
4617 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) 4637 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4618 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 4638 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4619 4639
4620 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { 4640 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4621 ext4_ext_truncate(inode); 4641 ext4_ext_truncate(inode);
4622 return; 4642 return;
4623 } 4643 }
@@ -4785,8 +4805,8 @@ static int __ext4_get_inode_loc(struct inode *inode,
4785 4805
4786 bh = sb_getblk(sb, block); 4806 bh = sb_getblk(sb, block);
4787 if (!bh) { 4807 if (!bh) {
4788 ext4_error(sb, "unable to read inode block - " 4808 EXT4_ERROR_INODE(inode, "unable to read inode block - "
4789 "inode=%lu, block=%llu", inode->i_ino, block); 4809 "block %llu", block);
4790 return -EIO; 4810 return -EIO;
4791 } 4811 }
4792 if (!buffer_uptodate(bh)) { 4812 if (!buffer_uptodate(bh)) {
@@ -4884,8 +4904,8 @@ make_io:
4884 submit_bh(READ_META, bh); 4904 submit_bh(READ_META, bh);
4885 wait_on_buffer(bh); 4905 wait_on_buffer(bh);
4886 if (!buffer_uptodate(bh)) { 4906 if (!buffer_uptodate(bh)) {
4887 ext4_error(sb, "unable to read inode block - inode=%lu," 4907 EXT4_ERROR_INODE(inode, "unable to read inode "
4888 " block=%llu", inode->i_ino, block); 4908 "block %llu", block);
4889 brelse(bh); 4909 brelse(bh);
4890 return -EIO; 4910 return -EIO;
4891 } 4911 }
@@ -5096,8 +5116,8 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
5096 ret = 0; 5116 ret = 0;
5097 if (ei->i_file_acl && 5117 if (ei->i_file_acl &&
5098 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { 5118 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5099 ext4_error(sb, "bad extended attribute block %llu inode #%lu", 5119 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5100 ei->i_file_acl, inode->i_ino); 5120 ei->i_file_acl);
5101 ret = -EIO; 5121 ret = -EIO;
5102 goto bad_inode; 5122 goto bad_inode;
5103 } else if (ei->i_flags & EXT4_EXTENTS_FL) { 5123 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
@@ -5142,8 +5162,7 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
5142 new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); 5162 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5143 } else { 5163 } else {
5144 ret = -EIO; 5164 ret = -EIO;
5145 ext4_error(inode->i_sb, "bogus i_mode (%o) for inode=%lu", 5165 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5146 inode->i_mode, inode->i_ino);
5147 goto bad_inode; 5166 goto bad_inode;
5148 } 5167 }
5149 brelse(iloc.bh); 5168 brelse(iloc.bh);
@@ -5381,9 +5400,9 @@ int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5381 if (wbc->sync_mode == WB_SYNC_ALL) 5400 if (wbc->sync_mode == WB_SYNC_ALL)
5382 sync_dirty_buffer(iloc.bh); 5401 sync_dirty_buffer(iloc.bh);
5383 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { 5402 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5384 ext4_error(inode->i_sb, "IO error syncing inode, " 5403 EXT4_ERROR_INODE(inode,
5385 "inode=%lu, block=%llu", inode->i_ino, 5404 "IO error syncing inode (block=%llu)",
5386 (unsigned long long)iloc.bh->b_blocknr); 5405 (unsigned long long) iloc.bh->b_blocknr);
5387 err = -EIO; 5406 err = -EIO;
5388 } 5407 }
5389 brelse(iloc.bh); 5408 brelse(iloc.bh);
@@ -5455,7 +5474,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5455 } 5474 }
5456 5475
5457 if (attr->ia_valid & ATTR_SIZE) { 5476 if (attr->ia_valid & ATTR_SIZE) {
5458 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) { 5477 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5459 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 5478 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5460 5479
5461 if (attr->ia_size > sbi->s_bitmap_maxbytes) { 5480 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
@@ -5468,7 +5487,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5468 if (S_ISREG(inode->i_mode) && 5487 if (S_ISREG(inode->i_mode) &&
5469 attr->ia_valid & ATTR_SIZE && 5488 attr->ia_valid & ATTR_SIZE &&
5470 (attr->ia_size < inode->i_size || 5489 (attr->ia_size < inode->i_size ||
5471 (EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL))) { 5490 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5472 handle_t *handle; 5491 handle_t *handle;
5473 5492
5474 handle = ext4_journal_start(inode, 3); 5493 handle = ext4_journal_start(inode, 3);
@@ -5500,7 +5519,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5500 } 5519 }
5501 } 5520 }
5502 /* ext4_truncate will clear the flag */ 5521 /* ext4_truncate will clear the flag */
5503 if ((EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL)) 5522 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5504 ext4_truncate(inode); 5523 ext4_truncate(inode);
5505 } 5524 }
5506 5525
@@ -5576,7 +5595,7 @@ static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5576 5595
5577static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) 5596static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5578{ 5597{
5579 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 5598 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5580 return ext4_indirect_trans_blocks(inode, nrblocks, chunk); 5599 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5581 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); 5600 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5582} 5601}
@@ -5911,9 +5930,9 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
5911 */ 5930 */
5912 5931
5913 if (val) 5932 if (val)
5914 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; 5933 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5915 else 5934 else
5916 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; 5935 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5917 ext4_set_aops(inode); 5936 ext4_set_aops(inode);
5918 5937
5919 jbd2_journal_unlock_updates(journal); 5938 jbd2_journal_unlock_updates(journal);
diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
index 016d0249294f..bf5ae883b1bd 100644
--- a/fs/ext4/ioctl.c
+++ b/fs/ext4/ioctl.c
@@ -258,7 +258,7 @@ setversion_out:
258 if (me.moved_len > 0) 258 if (me.moved_len > 0)
259 file_remove_suid(donor_filp); 259 file_remove_suid(donor_filp);
260 260
261 if (copy_to_user((struct move_extent __user *)arg, 261 if (copy_to_user((struct move_extent __user *)arg,
262 &me, sizeof(me))) 262 &me, sizeof(me)))
263 err = -EFAULT; 263 err = -EFAULT;
264mext_out: 264mext_out:
@@ -373,7 +373,30 @@ long ext4_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
373 case EXT4_IOC32_SETRSVSZ: 373 case EXT4_IOC32_SETRSVSZ:
374 cmd = EXT4_IOC_SETRSVSZ; 374 cmd = EXT4_IOC_SETRSVSZ;
375 break; 375 break;
376 case EXT4_IOC_GROUP_ADD: 376 case EXT4_IOC32_GROUP_ADD: {
377 struct compat_ext4_new_group_input __user *uinput;
378 struct ext4_new_group_input input;
379 mm_segment_t old_fs;
380 int err;
381
382 uinput = compat_ptr(arg);
383 err = get_user(input.group, &uinput->group);
384 err |= get_user(input.block_bitmap, &uinput->block_bitmap);
385 err |= get_user(input.inode_bitmap, &uinput->inode_bitmap);
386 err |= get_user(input.inode_table, &uinput->inode_table);
387 err |= get_user(input.blocks_count, &uinput->blocks_count);
388 err |= get_user(input.reserved_blocks,
389 &uinput->reserved_blocks);
390 if (err)
391 return -EFAULT;
392 old_fs = get_fs();
393 set_fs(KERNEL_DS);
394 err = ext4_ioctl(file, EXT4_IOC_GROUP_ADD,
395 (unsigned long) &input);
396 set_fs(old_fs);
397 return err;
398 }
399 case EXT4_IOC_MOVE_EXT:
377 break; 400 break;
378 default: 401 default:
379 return -ENOIOCTLCMD; 402 return -ENOIOCTLCMD;
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index b423a364dca3..12b3bc026a68 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -658,6 +658,27 @@ static void ext4_mb_mark_free_simple(struct super_block *sb,
658 } 658 }
659} 659}
660 660
661/*
662 * Cache the order of the largest free extent we have available in this block
663 * group.
664 */
665static void
666mb_set_largest_free_order(struct super_block *sb, struct ext4_group_info *grp)
667{
668 int i;
669 int bits;
670
671 grp->bb_largest_free_order = -1; /* uninit */
672
673 bits = sb->s_blocksize_bits + 1;
674 for (i = bits; i >= 0; i--) {
675 if (grp->bb_counters[i] > 0) {
676 grp->bb_largest_free_order = i;
677 break;
678 }
679 }
680}
681
661static noinline_for_stack 682static noinline_for_stack
662void ext4_mb_generate_buddy(struct super_block *sb, 683void ext4_mb_generate_buddy(struct super_block *sb,
663 void *buddy, void *bitmap, ext4_group_t group) 684 void *buddy, void *bitmap, ext4_group_t group)
@@ -700,6 +721,7 @@ void ext4_mb_generate_buddy(struct super_block *sb,
700 */ 721 */
701 grp->bb_free = free; 722 grp->bb_free = free;
702 } 723 }
724 mb_set_largest_free_order(sb, grp);
703 725
704 clear_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state)); 726 clear_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
705 727
@@ -725,6 +747,9 @@ void ext4_mb_generate_buddy(struct super_block *sb,
725 * contain blocks_per_page (PAGE_CACHE_SIZE / blocksize) blocks. 747 * contain blocks_per_page (PAGE_CACHE_SIZE / blocksize) blocks.
726 * So it can have information regarding groups_per_page which 748 * So it can have information regarding groups_per_page which
727 * is blocks_per_page/2 749 * is blocks_per_page/2
750 *
751 * Locking note: This routine takes the block group lock of all groups
752 * for this page; do not hold this lock when calling this routine!
728 */ 753 */
729 754
730static int ext4_mb_init_cache(struct page *page, char *incore) 755static int ext4_mb_init_cache(struct page *page, char *incore)
@@ -865,6 +890,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore)
865 BUG_ON(incore == NULL); 890 BUG_ON(incore == NULL);
866 mb_debug(1, "put buddy for group %u in page %lu/%x\n", 891 mb_debug(1, "put buddy for group %u in page %lu/%x\n",
867 group, page->index, i * blocksize); 892 group, page->index, i * blocksize);
893 trace_ext4_mb_buddy_bitmap_load(sb, group);
868 grinfo = ext4_get_group_info(sb, group); 894 grinfo = ext4_get_group_info(sb, group);
869 grinfo->bb_fragments = 0; 895 grinfo->bb_fragments = 0;
870 memset(grinfo->bb_counters, 0, 896 memset(grinfo->bb_counters, 0,
@@ -882,6 +908,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore)
882 BUG_ON(incore != NULL); 908 BUG_ON(incore != NULL);
883 mb_debug(1, "put bitmap for group %u in page %lu/%x\n", 909 mb_debug(1, "put bitmap for group %u in page %lu/%x\n",
884 group, page->index, i * blocksize); 910 group, page->index, i * blocksize);
911 trace_ext4_mb_bitmap_load(sb, group);
885 912
886 /* see comments in ext4_mb_put_pa() */ 913 /* see comments in ext4_mb_put_pa() */
887 ext4_lock_group(sb, group); 914 ext4_lock_group(sb, group);
@@ -910,6 +937,11 @@ out:
910 return err; 937 return err;
911} 938}
912 939
940/*
941 * Locking note: This routine calls ext4_mb_init_cache(), which takes the
942 * block group lock of all groups for this page; do not hold the BG lock when
943 * calling this routine!
944 */
913static noinline_for_stack 945static noinline_for_stack
914int ext4_mb_init_group(struct super_block *sb, ext4_group_t group) 946int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
915{ 947{
@@ -1004,6 +1036,11 @@ err:
1004 return ret; 1036 return ret;
1005} 1037}
1006 1038
1039/*
1040 * Locking note: This routine calls ext4_mb_init_cache(), which takes the
1041 * block group lock of all groups for this page; do not hold the BG lock when
1042 * calling this routine!
1043 */
1007static noinline_for_stack int 1044static noinline_for_stack int
1008ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group, 1045ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
1009 struct ext4_buddy *e4b) 1046 struct ext4_buddy *e4b)
@@ -1150,7 +1187,7 @@ err:
1150 return ret; 1187 return ret;
1151} 1188}
1152 1189
1153static void ext4_mb_release_desc(struct ext4_buddy *e4b) 1190static void ext4_mb_unload_buddy(struct ext4_buddy *e4b)
1154{ 1191{
1155 if (e4b->bd_bitmap_page) 1192 if (e4b->bd_bitmap_page)
1156 page_cache_release(e4b->bd_bitmap_page); 1193 page_cache_release(e4b->bd_bitmap_page);
@@ -1299,6 +1336,7 @@ static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1299 buddy = buddy2; 1336 buddy = buddy2;
1300 } while (1); 1337 } while (1);
1301 } 1338 }
1339 mb_set_largest_free_order(sb, e4b->bd_info);
1302 mb_check_buddy(e4b); 1340 mb_check_buddy(e4b);
1303} 1341}
1304 1342
@@ -1427,6 +1465,7 @@ static int mb_mark_used(struct ext4_buddy *e4b, struct ext4_free_extent *ex)
1427 e4b->bd_info->bb_counters[ord]++; 1465 e4b->bd_info->bb_counters[ord]++;
1428 e4b->bd_info->bb_counters[ord]++; 1466 e4b->bd_info->bb_counters[ord]++;
1429 } 1467 }
1468 mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);
1430 1469
1431 mb_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0); 1470 mb_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0);
1432 mb_check_buddy(e4b); 1471 mb_check_buddy(e4b);
@@ -1617,7 +1656,7 @@ int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
1617 } 1656 }
1618 1657
1619 ext4_unlock_group(ac->ac_sb, group); 1658 ext4_unlock_group(ac->ac_sb, group);
1620 ext4_mb_release_desc(e4b); 1659 ext4_mb_unload_buddy(e4b);
1621 1660
1622 return 0; 1661 return 0;
1623} 1662}
@@ -1672,7 +1711,7 @@ int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
1672 ext4_mb_use_best_found(ac, e4b); 1711 ext4_mb_use_best_found(ac, e4b);
1673 } 1712 }
1674 ext4_unlock_group(ac->ac_sb, group); 1713 ext4_unlock_group(ac->ac_sb, group);
1675 ext4_mb_release_desc(e4b); 1714 ext4_mb_unload_buddy(e4b);
1676 1715
1677 return 0; 1716 return 0;
1678} 1717}
@@ -1821,16 +1860,22 @@ void ext4_mb_scan_aligned(struct ext4_allocation_context *ac,
1821 } 1860 }
1822} 1861}
1823 1862
1863/* This is now called BEFORE we load the buddy bitmap. */
1824static int ext4_mb_good_group(struct ext4_allocation_context *ac, 1864static int ext4_mb_good_group(struct ext4_allocation_context *ac,
1825 ext4_group_t group, int cr) 1865 ext4_group_t group, int cr)
1826{ 1866{
1827 unsigned free, fragments; 1867 unsigned free, fragments;
1828 unsigned i, bits;
1829 int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb)); 1868 int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb));
1830 struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group); 1869 struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);
1831 1870
1832 BUG_ON(cr < 0 || cr >= 4); 1871 BUG_ON(cr < 0 || cr >= 4);
1833 BUG_ON(EXT4_MB_GRP_NEED_INIT(grp)); 1872
1873 /* We only do this if the grp has never been initialized */
1874 if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
1875 int ret = ext4_mb_init_group(ac->ac_sb, group);
1876 if (ret)
1877 return 0;
1878 }
1834 1879
1835 free = grp->bb_free; 1880 free = grp->bb_free;
1836 fragments = grp->bb_fragments; 1881 fragments = grp->bb_fragments;
@@ -1843,17 +1888,16 @@ static int ext4_mb_good_group(struct ext4_allocation_context *ac,
1843 case 0: 1888 case 0:
1844 BUG_ON(ac->ac_2order == 0); 1889 BUG_ON(ac->ac_2order == 0);
1845 1890
1891 if (grp->bb_largest_free_order < ac->ac_2order)
1892 return 0;
1893
1846 /* Avoid using the first bg of a flexgroup for data files */ 1894 /* Avoid using the first bg of a flexgroup for data files */
1847 if ((ac->ac_flags & EXT4_MB_HINT_DATA) && 1895 if ((ac->ac_flags & EXT4_MB_HINT_DATA) &&
1848 (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) && 1896 (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) &&
1849 ((group % flex_size) == 0)) 1897 ((group % flex_size) == 0))
1850 return 0; 1898 return 0;
1851 1899
1852 bits = ac->ac_sb->s_blocksize_bits + 1; 1900 return 1;
1853 for (i = ac->ac_2order; i <= bits; i++)
1854 if (grp->bb_counters[i] > 0)
1855 return 1;
1856 break;
1857 case 1: 1901 case 1:
1858 if ((free / fragments) >= ac->ac_g_ex.fe_len) 1902 if ((free / fragments) >= ac->ac_g_ex.fe_len)
1859 return 1; 1903 return 1;
@@ -1964,7 +2008,7 @@ ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
1964 sbi = EXT4_SB(sb); 2008 sbi = EXT4_SB(sb);
1965 ngroups = ext4_get_groups_count(sb); 2009 ngroups = ext4_get_groups_count(sb);
1966 /* non-extent files are limited to low blocks/groups */ 2010 /* non-extent files are limited to low blocks/groups */
1967 if (!(EXT4_I(ac->ac_inode)->i_flags & EXT4_EXTENTS_FL)) 2011 if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)))
1968 ngroups = sbi->s_blockfile_groups; 2012 ngroups = sbi->s_blockfile_groups;
1969 2013
1970 BUG_ON(ac->ac_status == AC_STATUS_FOUND); 2014 BUG_ON(ac->ac_status == AC_STATUS_FOUND);
@@ -2024,15 +2068,11 @@ repeat:
2024 group = ac->ac_g_ex.fe_group; 2068 group = ac->ac_g_ex.fe_group;
2025 2069
2026 for (i = 0; i < ngroups; group++, i++) { 2070 for (i = 0; i < ngroups; group++, i++) {
2027 struct ext4_group_info *grp;
2028 struct ext4_group_desc *desc;
2029
2030 if (group == ngroups) 2071 if (group == ngroups)
2031 group = 0; 2072 group = 0;
2032 2073
2033 /* quick check to skip empty groups */ 2074 /* This now checks without needing the buddy page */
2034 grp = ext4_get_group_info(sb, group); 2075 if (!ext4_mb_good_group(ac, group, cr))
2035 if (grp->bb_free == 0)
2036 continue; 2076 continue;
2037 2077
2038 err = ext4_mb_load_buddy(sb, group, &e4b); 2078 err = ext4_mb_load_buddy(sb, group, &e4b);
@@ -2040,15 +2080,18 @@ repeat:
2040 goto out; 2080 goto out;
2041 2081
2042 ext4_lock_group(sb, group); 2082 ext4_lock_group(sb, group);
2083
2084 /*
2085 * We need to check again after locking the
2086 * block group
2087 */
2043 if (!ext4_mb_good_group(ac, group, cr)) { 2088 if (!ext4_mb_good_group(ac, group, cr)) {
2044 /* someone did allocation from this group */
2045 ext4_unlock_group(sb, group); 2089 ext4_unlock_group(sb, group);
2046 ext4_mb_release_desc(&e4b); 2090 ext4_mb_unload_buddy(&e4b);
2047 continue; 2091 continue;
2048 } 2092 }
2049 2093
2050 ac->ac_groups_scanned++; 2094 ac->ac_groups_scanned++;
2051 desc = ext4_get_group_desc(sb, group, NULL);
2052 if (cr == 0) 2095 if (cr == 0)
2053 ext4_mb_simple_scan_group(ac, &e4b); 2096 ext4_mb_simple_scan_group(ac, &e4b);
2054 else if (cr == 1 && 2097 else if (cr == 1 &&
@@ -2058,7 +2101,7 @@ repeat:
2058 ext4_mb_complex_scan_group(ac, &e4b); 2101 ext4_mb_complex_scan_group(ac, &e4b);
2059 2102
2060 ext4_unlock_group(sb, group); 2103 ext4_unlock_group(sb, group);
2061 ext4_mb_release_desc(&e4b); 2104 ext4_mb_unload_buddy(&e4b);
2062 2105
2063 if (ac->ac_status != AC_STATUS_CONTINUE) 2106 if (ac->ac_status != AC_STATUS_CONTINUE)
2064 break; 2107 break;
@@ -2148,7 +2191,7 @@ static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
2148 ext4_lock_group(sb, group); 2191 ext4_lock_group(sb, group);
2149 memcpy(&sg, ext4_get_group_info(sb, group), i); 2192 memcpy(&sg, ext4_get_group_info(sb, group), i);
2150 ext4_unlock_group(sb, group); 2193 ext4_unlock_group(sb, group);
2151 ext4_mb_release_desc(&e4b); 2194 ext4_mb_unload_buddy(&e4b);
2152 2195
2153 seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free, 2196 seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
2154 sg.info.bb_fragments, sg.info.bb_first_free); 2197 sg.info.bb_fragments, sg.info.bb_first_free);
@@ -2255,6 +2298,7 @@ int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2255 INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list); 2298 INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
2256 init_rwsem(&meta_group_info[i]->alloc_sem); 2299 init_rwsem(&meta_group_info[i]->alloc_sem);
2257 meta_group_info[i]->bb_free_root = RB_ROOT; 2300 meta_group_info[i]->bb_free_root = RB_ROOT;
2301 meta_group_info[i]->bb_largest_free_order = -1; /* uninit */
2258 2302
2259#ifdef DOUBLE_CHECK 2303#ifdef DOUBLE_CHECK
2260 { 2304 {
@@ -2536,6 +2580,7 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
2536 entry->count, entry->group, entry); 2580 entry->count, entry->group, entry);
2537 2581
2538 if (test_opt(sb, DISCARD)) { 2582 if (test_opt(sb, DISCARD)) {
2583 int ret;
2539 ext4_fsblk_t discard_block; 2584 ext4_fsblk_t discard_block;
2540 2585
2541 discard_block = entry->start_blk + 2586 discard_block = entry->start_blk +
@@ -2543,7 +2588,12 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
2543 trace_ext4_discard_blocks(sb, 2588 trace_ext4_discard_blocks(sb,
2544 (unsigned long long)discard_block, 2589 (unsigned long long)discard_block,
2545 entry->count); 2590 entry->count);
2546 sb_issue_discard(sb, discard_block, entry->count); 2591 ret = sb_issue_discard(sb, discard_block, entry->count);
2592 if (ret == EOPNOTSUPP) {
2593 ext4_warning(sb,
2594 "discard not supported, disabling");
2595 clear_opt(EXT4_SB(sb)->s_mount_opt, DISCARD);
2596 }
2547 } 2597 }
2548 2598
2549 err = ext4_mb_load_buddy(sb, entry->group, &e4b); 2599 err = ext4_mb_load_buddy(sb, entry->group, &e4b);
@@ -2568,7 +2618,7 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
2568 } 2618 }
2569 ext4_unlock_group(sb, entry->group); 2619 ext4_unlock_group(sb, entry->group);
2570 kmem_cache_free(ext4_free_ext_cachep, entry); 2620 kmem_cache_free(ext4_free_ext_cachep, entry);
2571 ext4_mb_release_desc(&e4b); 2621 ext4_mb_unload_buddy(&e4b);
2572 } 2622 }
2573 2623
2574 mb_debug(1, "freed %u blocks in %u structures\n", count, count2); 2624 mb_debug(1, "freed %u blocks in %u structures\n", count, count2);
@@ -2641,7 +2691,7 @@ int __init init_ext4_mballoc(void)
2641 2691
2642void exit_ext4_mballoc(void) 2692void exit_ext4_mballoc(void)
2643{ 2693{
2644 /* 2694 /*
2645 * Wait for completion of call_rcu()'s on ext4_pspace_cachep 2695 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
2646 * before destroying the slab cache. 2696 * before destroying the slab cache.
2647 */ 2697 */
@@ -2981,7 +3031,7 @@ static void ext4_mb_collect_stats(struct ext4_allocation_context *ac)
2981 if (sbi->s_mb_stats && ac->ac_g_ex.fe_len > 1) { 3031 if (sbi->s_mb_stats && ac->ac_g_ex.fe_len > 1) {
2982 atomic_inc(&sbi->s_bal_reqs); 3032 atomic_inc(&sbi->s_bal_reqs);
2983 atomic_add(ac->ac_b_ex.fe_len, &sbi->s_bal_allocated); 3033 atomic_add(ac->ac_b_ex.fe_len, &sbi->s_bal_allocated);
2984 if (ac->ac_o_ex.fe_len >= ac->ac_g_ex.fe_len) 3034 if (ac->ac_b_ex.fe_len >= ac->ac_o_ex.fe_len)
2985 atomic_inc(&sbi->s_bal_success); 3035 atomic_inc(&sbi->s_bal_success);
2986 atomic_add(ac->ac_found, &sbi->s_bal_ex_scanned); 3036 atomic_add(ac->ac_found, &sbi->s_bal_ex_scanned);
2987 if (ac->ac_g_ex.fe_start == ac->ac_b_ex.fe_start && 3037 if (ac->ac_g_ex.fe_start == ac->ac_b_ex.fe_start &&
@@ -3123,7 +3173,7 @@ ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
3123 continue; 3173 continue;
3124 3174
3125 /* non-extent files can't have physical blocks past 2^32 */ 3175 /* non-extent files can't have physical blocks past 2^32 */
3126 if (!(EXT4_I(ac->ac_inode)->i_flags & EXT4_EXTENTS_FL) && 3176 if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)) &&
3127 pa->pa_pstart + pa->pa_len > EXT4_MAX_BLOCK_FILE_PHYS) 3177 pa->pa_pstart + pa->pa_len > EXT4_MAX_BLOCK_FILE_PHYS)
3128 continue; 3178 continue;
3129 3179
@@ -3280,7 +3330,7 @@ static void ext4_mb_put_pa(struct ext4_allocation_context *ac,
3280 spin_unlock(&pa->pa_lock); 3330 spin_unlock(&pa->pa_lock);
3281 3331
3282 grp_blk = pa->pa_pstart; 3332 grp_blk = pa->pa_pstart;
3283 /* 3333 /*
3284 * If doing group-based preallocation, pa_pstart may be in the 3334 * If doing group-based preallocation, pa_pstart may be in the
3285 * next group when pa is used up 3335 * next group when pa is used up
3286 */ 3336 */
@@ -3697,7 +3747,7 @@ out:
3697 ext4_unlock_group(sb, group); 3747 ext4_unlock_group(sb, group);
3698 if (ac) 3748 if (ac)
3699 kmem_cache_free(ext4_ac_cachep, ac); 3749 kmem_cache_free(ext4_ac_cachep, ac);
3700 ext4_mb_release_desc(&e4b); 3750 ext4_mb_unload_buddy(&e4b);
3701 put_bh(bitmap_bh); 3751 put_bh(bitmap_bh);
3702 return free; 3752 return free;
3703} 3753}
@@ -3801,7 +3851,7 @@ repeat:
3801 if (bitmap_bh == NULL) { 3851 if (bitmap_bh == NULL) {
3802 ext4_error(sb, "Error reading block bitmap for %u", 3852 ext4_error(sb, "Error reading block bitmap for %u",
3803 group); 3853 group);
3804 ext4_mb_release_desc(&e4b); 3854 ext4_mb_unload_buddy(&e4b);
3805 continue; 3855 continue;
3806 } 3856 }
3807 3857
@@ -3810,7 +3860,7 @@ repeat:
3810 ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac); 3860 ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac);
3811 ext4_unlock_group(sb, group); 3861 ext4_unlock_group(sb, group);
3812 3862
3813 ext4_mb_release_desc(&e4b); 3863 ext4_mb_unload_buddy(&e4b);
3814 put_bh(bitmap_bh); 3864 put_bh(bitmap_bh);
3815 3865
3816 list_del(&pa->u.pa_tmp_list); 3866 list_del(&pa->u.pa_tmp_list);
@@ -4074,7 +4124,7 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
4074 ext4_mb_release_group_pa(&e4b, pa, ac); 4124 ext4_mb_release_group_pa(&e4b, pa, ac);
4075 ext4_unlock_group(sb, group); 4125 ext4_unlock_group(sb, group);
4076 4126
4077 ext4_mb_release_desc(&e4b); 4127 ext4_mb_unload_buddy(&e4b);
4078 list_del(&pa->u.pa_tmp_list); 4128 list_del(&pa->u.pa_tmp_list);
4079 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback); 4129 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
4080 } 4130 }
@@ -4484,12 +4534,12 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
4484 if (!bh) 4534 if (!bh)
4485 tbh = sb_find_get_block(inode->i_sb, 4535 tbh = sb_find_get_block(inode->i_sb,
4486 block + i); 4536 block + i);
4487 ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA, 4537 ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
4488 inode, tbh, block + i); 4538 inode, tbh, block + i);
4489 } 4539 }
4490 } 4540 }
4491 4541
4492 /* 4542 /*
4493 * We need to make sure we don't reuse the freed block until 4543 * We need to make sure we don't reuse the freed block until
4494 * after the transaction is committed, which we can do by 4544 * after the transaction is committed, which we can do by
4495 * treating the block as metadata, below. We make an 4545 * treating the block as metadata, below. We make an
@@ -4610,7 +4660,7 @@ do_more:
4610 atomic_add(count, &sbi->s_flex_groups[flex_group].free_blocks); 4660 atomic_add(count, &sbi->s_flex_groups[flex_group].free_blocks);
4611 } 4661 }
4612 4662
4613 ext4_mb_release_desc(&e4b); 4663 ext4_mb_unload_buddy(&e4b);
4614 4664
4615 freed += count; 4665 freed += count;
4616 4666
diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
index 34dcfc52ef44..6f3a27ec30bf 100644
--- a/fs/ext4/migrate.c
+++ b/fs/ext4/migrate.c
@@ -475,7 +475,7 @@ int ext4_ext_migrate(struct inode *inode)
475 */ 475 */
476 if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb, 476 if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
477 EXT4_FEATURE_INCOMPAT_EXTENTS) || 477 EXT4_FEATURE_INCOMPAT_EXTENTS) ||
478 (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 478 (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
479 return -EINVAL; 479 return -EINVAL;
480 480
481 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0) 481 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
diff --git a/fs/ext4/move_extent.c b/fs/ext4/move_extent.c
index d1fc662cc311..3a6c92ac131c 100644
--- a/fs/ext4/move_extent.c
+++ b/fs/ext4/move_extent.c
@@ -482,6 +482,7 @@ mext_leaf_block(handle_t *handle, struct inode *orig_inode,
482 int depth = ext_depth(orig_inode); 482 int depth = ext_depth(orig_inode);
483 int ret; 483 int ret;
484 484
485 start_ext.ee_block = end_ext.ee_block = 0;
485 o_start = o_end = oext = orig_path[depth].p_ext; 486 o_start = o_end = oext = orig_path[depth].p_ext;
486 oext_alen = ext4_ext_get_actual_len(oext); 487 oext_alen = ext4_ext_get_actual_len(oext);
487 start_ext.ee_len = end_ext.ee_len = 0; 488 start_ext.ee_len = end_ext.ee_len = 0;
@@ -529,7 +530,7 @@ mext_leaf_block(handle_t *handle, struct inode *orig_inode,
529 * new_ext |-------| 530 * new_ext |-------|
530 */ 531 */
531 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) { 532 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
532 ext4_error(orig_inode->i_sb, 533 EXT4_ERROR_INODE(orig_inode,
533 "new_ext_end(%u) should be less than or equal to " 534 "new_ext_end(%u) should be less than or equal to "
534 "oext->ee_block(%u) + oext_alen(%d) - 1", 535 "oext->ee_block(%u) + oext_alen(%d) - 1",
535 new_ext_end, le32_to_cpu(oext->ee_block), 536 new_ext_end, le32_to_cpu(oext->ee_block),
@@ -692,12 +693,12 @@ mext_replace_branches(handle_t *handle, struct inode *orig_inode,
692 while (1) { 693 while (1) {
693 /* The extent for donor must be found. */ 694 /* The extent for donor must be found. */
694 if (!dext) { 695 if (!dext) {
695 ext4_error(donor_inode->i_sb, 696 EXT4_ERROR_INODE(donor_inode,
696 "The extent for donor must be found"); 697 "The extent for donor must be found");
697 *err = -EIO; 698 *err = -EIO;
698 goto out; 699 goto out;
699 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) { 700 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
700 ext4_error(donor_inode->i_sb, 701 EXT4_ERROR_INODE(donor_inode,
701 "Donor offset(%u) and the first block of donor " 702 "Donor offset(%u) and the first block of donor "
702 "extent(%u) should be equal", 703 "extent(%u) should be equal",
703 donor_off, 704 donor_off,
@@ -976,11 +977,11 @@ mext_check_arguments(struct inode *orig_inode,
976 } 977 }
977 978
978 /* Ext4 move extent supports only extent based file */ 979 /* Ext4 move extent supports only extent based file */
979 if (!(EXT4_I(orig_inode)->i_flags & EXT4_EXTENTS_FL)) { 980 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
980 ext4_debug("ext4 move extent: orig file is not extents " 981 ext4_debug("ext4 move extent: orig file is not extents "
981 "based file [ino:orig %lu]\n", orig_inode->i_ino); 982 "based file [ino:orig %lu]\n", orig_inode->i_ino);
982 return -EOPNOTSUPP; 983 return -EOPNOTSUPP;
983 } else if (!(EXT4_I(donor_inode)->i_flags & EXT4_EXTENTS_FL)) { 984 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
984 ext4_debug("ext4 move extent: donor file is not extents " 985 ext4_debug("ext4 move extent: donor file is not extents "
985 "based file [ino:donor %lu]\n", donor_inode->i_ino); 986 "based file [ino:donor %lu]\n", donor_inode->i_ino);
986 return -EOPNOTSUPP; 987 return -EOPNOTSUPP;
@@ -1354,7 +1355,7 @@ ext4_move_extents(struct file *o_filp, struct file *d_filp,
1354 if (ret1 < 0) 1355 if (ret1 < 0)
1355 break; 1356 break;
1356 if (*moved_len > len) { 1357 if (*moved_len > len) {
1357 ext4_error(orig_inode->i_sb, 1358 EXT4_ERROR_INODE(orig_inode,
1358 "We replaced blocks too much! " 1359 "We replaced blocks too much! "
1359 "sum of replaced: %llu requested: %llu", 1360 "sum of replaced: %llu requested: %llu",
1360 *moved_len, len); 1361 *moved_len, len);
diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index 0c070fabd108..a43e6617b351 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -187,7 +187,7 @@ unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize)
187 return blocksize; 187 return blocksize;
188 return (len & 65532) | ((len & 3) << 16); 188 return (len & 65532) | ((len & 3) << 16);
189} 189}
190 190
191__le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize) 191__le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
192{ 192{
193 if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3)) 193 if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3))
@@ -197,7 +197,7 @@ __le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
197 if (len == blocksize) { 197 if (len == blocksize) {
198 if (blocksize == 65536) 198 if (blocksize == 65536)
199 return cpu_to_le16(EXT4_MAX_REC_LEN); 199 return cpu_to_le16(EXT4_MAX_REC_LEN);
200 else 200 else
201 return cpu_to_le16(0); 201 return cpu_to_le16(0);
202 } 202 }
203 return cpu_to_le16((len & 65532) | ((len >> 16) & 3)); 203 return cpu_to_le16((len & 65532) | ((len >> 16) & 3));
@@ -349,7 +349,7 @@ struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
349 brelse(bh); 349 brelse(bh);
350 } 350 }
351 if (bcount) 351 if (bcount)
352 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", 352 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
353 levels ? "" : " ", names, space/bcount, 353 levels ? "" : " ", names, space/bcount,
354 (space/bcount)*100/blocksize); 354 (space/bcount)*100/blocksize);
355 return (struct stats) { names, space, bcount}; 355 return (struct stats) { names, space, bcount};
@@ -653,10 +653,10 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
653 int ret, err; 653 int ret, err;
654 __u32 hashval; 654 __u32 hashval;
655 655
656 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", 656 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
657 start_hash, start_minor_hash)); 657 start_hash, start_minor_hash));
658 dir = dir_file->f_path.dentry->d_inode; 658 dir = dir_file->f_path.dentry->d_inode;
659 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) { 659 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
660 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; 660 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
661 if (hinfo.hash_version <= DX_HASH_TEA) 661 if (hinfo.hash_version <= DX_HASH_TEA)
662 hinfo.hash_version += 662 hinfo.hash_version +=
@@ -801,7 +801,7 @@ static void ext4_update_dx_flag(struct inode *inode)
801{ 801{
802 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb, 802 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
803 EXT4_FEATURE_COMPAT_DIR_INDEX)) 803 EXT4_FEATURE_COMPAT_DIR_INDEX))
804 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL; 804 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
805} 805}
806 806
807/* 807/*
@@ -943,8 +943,8 @@ restart:
943 wait_on_buffer(bh); 943 wait_on_buffer(bh);
944 if (!buffer_uptodate(bh)) { 944 if (!buffer_uptodate(bh)) {
945 /* read error, skip block & hope for the best */ 945 /* read error, skip block & hope for the best */
946 ext4_error(sb, "reading directory #%lu offset %lu", 946 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
947 dir->i_ino, (unsigned long)block); 947 (unsigned long) block);
948 brelse(bh); 948 brelse(bh);
949 goto next; 949 goto next;
950 } 950 }
@@ -1066,15 +1066,15 @@ static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, stru
1066 __u32 ino = le32_to_cpu(de->inode); 1066 __u32 ino = le32_to_cpu(de->inode);
1067 brelse(bh); 1067 brelse(bh);
1068 if (!ext4_valid_inum(dir->i_sb, ino)) { 1068 if (!ext4_valid_inum(dir->i_sb, ino)) {
1069 ext4_error(dir->i_sb, "bad inode number: %u", ino); 1069 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1070 return ERR_PTR(-EIO); 1070 return ERR_PTR(-EIO);
1071 } 1071 }
1072 inode = ext4_iget(dir->i_sb, ino); 1072 inode = ext4_iget(dir->i_sb, ino);
1073 if (unlikely(IS_ERR(inode))) { 1073 if (unlikely(IS_ERR(inode))) {
1074 if (PTR_ERR(inode) == -ESTALE) { 1074 if (PTR_ERR(inode) == -ESTALE) {
1075 ext4_error(dir->i_sb, 1075 EXT4_ERROR_INODE(dir,
1076 "deleted inode referenced: %u", 1076 "deleted inode referenced: %u",
1077 ino); 1077 ino);
1078 return ERR_PTR(-EIO); 1078 return ERR_PTR(-EIO);
1079 } else { 1079 } else {
1080 return ERR_CAST(inode); 1080 return ERR_CAST(inode);
@@ -1104,8 +1104,8 @@ struct dentry *ext4_get_parent(struct dentry *child)
1104 brelse(bh); 1104 brelse(bh);
1105 1105
1106 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) { 1106 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1107 ext4_error(child->d_inode->i_sb, 1107 EXT4_ERROR_INODE(child->d_inode,
1108 "bad inode number: %u", ino); 1108 "bad parent inode number: %u", ino);
1109 return ERR_PTR(-EIO); 1109 return ERR_PTR(-EIO);
1110 } 1110 }
1111 1111
@@ -1141,7 +1141,7 @@ dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1141 unsigned rec_len = 0; 1141 unsigned rec_len = 0;
1142 1142
1143 while (count--) { 1143 while (count--) {
1144 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) 1144 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1145 (from + (map->offs<<2)); 1145 (from + (map->offs<<2));
1146 rec_len = EXT4_DIR_REC_LEN(de->name_len); 1146 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1147 memcpy (to, de, rec_len); 1147 memcpy (to, de, rec_len);
@@ -1404,9 +1404,7 @@ static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1404 de = (struct ext4_dir_entry_2 *)((char *)fde + 1404 de = (struct ext4_dir_entry_2 *)((char *)fde +
1405 ext4_rec_len_from_disk(fde->rec_len, blocksize)); 1405 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1406 if ((char *) de >= (((char *) root) + blocksize)) { 1406 if ((char *) de >= (((char *) root) + blocksize)) {
1407 ext4_error(dir->i_sb, 1407 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1408 "invalid rec_len for '..' in inode %lu",
1409 dir->i_ino);
1410 brelse(bh); 1408 brelse(bh);
1411 return -EIO; 1409 return -EIO;
1412 } 1410 }
@@ -1418,7 +1416,7 @@ static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1418 brelse(bh); 1416 brelse(bh);
1419 return retval; 1417 return retval;
1420 } 1418 }
1421 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL; 1419 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1422 data1 = bh2->b_data; 1420 data1 = bh2->b_data;
1423 1421
1424 memcpy (data1, de, len); 1422 memcpy (data1, de, len);
@@ -1491,7 +1489,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1491 retval = ext4_dx_add_entry(handle, dentry, inode); 1489 retval = ext4_dx_add_entry(handle, dentry, inode);
1492 if (!retval || (retval != ERR_BAD_DX_DIR)) 1490 if (!retval || (retval != ERR_BAD_DX_DIR))
1493 return retval; 1491 return retval;
1494 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL; 1492 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1495 dx_fallback++; 1493 dx_fallback++;
1496 ext4_mark_inode_dirty(handle, dir); 1494 ext4_mark_inode_dirty(handle, dir);
1497 } 1495 }
@@ -1519,6 +1517,8 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1519 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize); 1517 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1520 retval = add_dirent_to_buf(handle, dentry, inode, de, bh); 1518 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1521 brelse(bh); 1519 brelse(bh);
1520 if (retval == 0)
1521 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1522 return retval; 1522 return retval;
1523} 1523}
1524 1524
@@ -1915,9 +1915,8 @@ static int empty_dir(struct inode *inode)
1915 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) || 1915 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1916 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) { 1916 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1917 if (err) 1917 if (err)
1918 ext4_error(inode->i_sb, 1918 EXT4_ERROR_INODE(inode,
1919 "error %d reading directory #%lu offset 0", 1919 "error %d reading directory lblock 0", err);
1920 err, inode->i_ino);
1921 else 1920 else
1922 ext4_warning(inode->i_sb, 1921 ext4_warning(inode->i_sb,
1923 "bad directory (dir #%lu) - no data block", 1922 "bad directory (dir #%lu) - no data block",
@@ -1941,17 +1940,17 @@ static int empty_dir(struct inode *inode)
1941 de = ext4_next_entry(de1, sb->s_blocksize); 1940 de = ext4_next_entry(de1, sb->s_blocksize);
1942 while (offset < inode->i_size) { 1941 while (offset < inode->i_size) {
1943 if (!bh || 1942 if (!bh ||
1944 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { 1943 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1944 unsigned int lblock;
1945 err = 0; 1945 err = 0;
1946 brelse(bh); 1946 brelse(bh);
1947 bh = ext4_bread(NULL, inode, 1947 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1948 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err); 1948 bh = ext4_bread(NULL, inode, lblock, 0, &err);
1949 if (!bh) { 1949 if (!bh) {
1950 if (err) 1950 if (err)
1951 ext4_error(sb, 1951 EXT4_ERROR_INODE(inode,
1952 "error %d reading directory" 1952 "error %d reading directory "
1953 " #%lu offset %u", 1953 "lblock %u", err, lblock);
1954 err, inode->i_ino, offset);
1955 offset += sb->s_blocksize; 1954 offset += sb->s_blocksize;
1956 continue; 1955 continue;
1957 } 1956 }
@@ -2297,7 +2296,7 @@ retry:
2297 } 2296 }
2298 } else { 2297 } else {
2299 /* clear the extent format for fast symlink */ 2298 /* clear the extent format for fast symlink */
2300 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL; 2299 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2301 inode->i_op = &ext4_fast_symlink_inode_operations; 2300 inode->i_op = &ext4_fast_symlink_inode_operations;
2302 memcpy((char *)&EXT4_I(inode)->i_data, symname, l); 2301 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2303 inode->i_size = l-1; 2302 inode->i_size = l-1;
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
index 5692c48754a0..6df797eb9aeb 100644
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -911,7 +911,8 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
911 percpu_counter_add(&sbi->s_freeinodes_counter, 911 percpu_counter_add(&sbi->s_freeinodes_counter,
912 EXT4_INODES_PER_GROUP(sb)); 912 EXT4_INODES_PER_GROUP(sb));
913 913
914 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { 914 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
915 sbi->s_log_groups_per_flex) {
915 ext4_group_t flex_group; 916 ext4_group_t flex_group;
916 flex_group = ext4_flex_group(sbi, input->group); 917 flex_group = ext4_flex_group(sbi, input->group);
917 atomic_add(input->free_blocks_count, 918 atomic_add(input->free_blocks_count,
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index e14d22c170d5..49d88c0597c4 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -241,6 +241,7 @@ handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
241 if (sb->s_flags & MS_RDONLY) 241 if (sb->s_flags & MS_RDONLY)
242 return ERR_PTR(-EROFS); 242 return ERR_PTR(-EROFS);
243 243
244 vfs_check_frozen(sb, SB_FREEZE_WRITE);
244 /* Special case here: if the journal has aborted behind our 245 /* Special case here: if the journal has aborted behind our
245 * backs (eg. EIO in the commit thread), then we still need to 246 * backs (eg. EIO in the commit thread), then we still need to
246 * take the FS itself readonly cleanly. */ 247 * take the FS itself readonly cleanly. */
@@ -941,6 +942,8 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
941 seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0"); 942 seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
942 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) 943 if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
943 seq_puts(seq, ",journal_async_commit"); 944 seq_puts(seq, ",journal_async_commit");
945 else if (test_opt(sb, JOURNAL_CHECKSUM))
946 seq_puts(seq, ",journal_checksum");
944 if (test_opt(sb, NOBH)) 947 if (test_opt(sb, NOBH))
945 seq_puts(seq, ",nobh"); 948 seq_puts(seq, ",nobh");
946 if (test_opt(sb, I_VERSION)) 949 if (test_opt(sb, I_VERSION))
@@ -2213,7 +2216,7 @@ static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2213struct ext4_attr { 2216struct ext4_attr {
2214 struct attribute attr; 2217 struct attribute attr;
2215 ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *); 2218 ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2216 ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *, 2219 ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2217 const char *, size_t); 2220 const char *, size_t);
2218 int offset; 2221 int offset;
2219}; 2222};
@@ -2430,6 +2433,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2430 __releases(kernel_lock) 2433 __releases(kernel_lock)
2431 __acquires(kernel_lock) 2434 __acquires(kernel_lock)
2432{ 2435{
2436 char *orig_data = kstrdup(data, GFP_KERNEL);
2433 struct buffer_head *bh; 2437 struct buffer_head *bh;
2434 struct ext4_super_block *es = NULL; 2438 struct ext4_super_block *es = NULL;
2435 struct ext4_sb_info *sbi; 2439 struct ext4_sb_info *sbi;
@@ -2793,24 +2797,6 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2793 get_random_bytes(&sbi->s_next_generation, sizeof(u32)); 2797 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2794 spin_lock_init(&sbi->s_next_gen_lock); 2798 spin_lock_init(&sbi->s_next_gen_lock);
2795 2799
2796 err = percpu_counter_init(&sbi->s_freeblocks_counter,
2797 ext4_count_free_blocks(sb));
2798 if (!err) {
2799 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2800 ext4_count_free_inodes(sb));
2801 }
2802 if (!err) {
2803 err = percpu_counter_init(&sbi->s_dirs_counter,
2804 ext4_count_dirs(sb));
2805 }
2806 if (!err) {
2807 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2808 }
2809 if (err) {
2810 ext4_msg(sb, KERN_ERR, "insufficient memory");
2811 goto failed_mount3;
2812 }
2813
2814 sbi->s_stripe = ext4_get_stripe_size(sbi); 2800 sbi->s_stripe = ext4_get_stripe_size(sbi);
2815 sbi->s_max_writeback_mb_bump = 128; 2801 sbi->s_max_writeback_mb_bump = 128;
2816 2802
@@ -2910,6 +2896,20 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2910 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 2896 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2911 2897
2912no_journal: 2898no_journal:
2899 err = percpu_counter_init(&sbi->s_freeblocks_counter,
2900 ext4_count_free_blocks(sb));
2901 if (!err)
2902 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2903 ext4_count_free_inodes(sb));
2904 if (!err)
2905 err = percpu_counter_init(&sbi->s_dirs_counter,
2906 ext4_count_dirs(sb));
2907 if (!err)
2908 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2909 if (err) {
2910 ext4_msg(sb, KERN_ERR, "insufficient memory");
2911 goto failed_mount_wq;
2912 }
2913 if (test_opt(sb, NOBH)) { 2913 if (test_opt(sb, NOBH)) {
2914 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) { 2914 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2915 ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - " 2915 ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - "
@@ -3001,7 +3001,7 @@ no_journal:
3001 err = ext4_setup_system_zone(sb); 3001 err = ext4_setup_system_zone(sb);
3002 if (err) { 3002 if (err) {
3003 ext4_msg(sb, KERN_ERR, "failed to initialize system " 3003 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3004 "zone (%d)\n", err); 3004 "zone (%d)", err);
3005 goto failed_mount4; 3005 goto failed_mount4;
3006 } 3006 }
3007 3007
@@ -3040,9 +3040,11 @@ no_journal:
3040 } else 3040 } else
3041 descr = "out journal"; 3041 descr = "out journal";
3042 3042
3043 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s", descr); 3043 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3044 "Opts: %s", descr, orig_data);
3044 3045
3045 lock_kernel(); 3046 lock_kernel();
3047 kfree(orig_data);
3046 return 0; 3048 return 0;
3047 3049
3048cantfind_ext4: 3050cantfind_ext4:
@@ -3059,6 +3061,10 @@ failed_mount_wq:
3059 jbd2_journal_destroy(sbi->s_journal); 3061 jbd2_journal_destroy(sbi->s_journal);
3060 sbi->s_journal = NULL; 3062 sbi->s_journal = NULL;
3061 } 3063 }
3064 percpu_counter_destroy(&sbi->s_freeblocks_counter);
3065 percpu_counter_destroy(&sbi->s_freeinodes_counter);
3066 percpu_counter_destroy(&sbi->s_dirs_counter);
3067 percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3062failed_mount3: 3068failed_mount3:
3063 if (sbi->s_flex_groups) { 3069 if (sbi->s_flex_groups) {
3064 if (is_vmalloc_addr(sbi->s_flex_groups)) 3070 if (is_vmalloc_addr(sbi->s_flex_groups))
@@ -3066,10 +3072,6 @@ failed_mount3:
3066 else 3072 else
3067 kfree(sbi->s_flex_groups); 3073 kfree(sbi->s_flex_groups);
3068 } 3074 }
3069 percpu_counter_destroy(&sbi->s_freeblocks_counter);
3070 percpu_counter_destroy(&sbi->s_freeinodes_counter);
3071 percpu_counter_destroy(&sbi->s_dirs_counter);
3072 percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3073failed_mount2: 3075failed_mount2:
3074 for (i = 0; i < db_count; i++) 3076 for (i = 0; i < db_count; i++)
3075 brelse(sbi->s_group_desc[i]); 3077 brelse(sbi->s_group_desc[i]);
@@ -3089,6 +3091,7 @@ out_fail:
3089 kfree(sbi->s_blockgroup_lock); 3091 kfree(sbi->s_blockgroup_lock);
3090 kfree(sbi); 3092 kfree(sbi);
3091 lock_kernel(); 3093 lock_kernel();
3094 kfree(orig_data);
3092 return ret; 3095 return ret;
3093} 3096}
3094 3097
@@ -3380,7 +3383,7 @@ static int ext4_commit_super(struct super_block *sb, int sync)
3380 if (!(sb->s_flags & MS_RDONLY)) 3383 if (!(sb->s_flags & MS_RDONLY))
3381 es->s_wtime = cpu_to_le32(get_seconds()); 3384 es->s_wtime = cpu_to_le32(get_seconds());
3382 es->s_kbytes_written = 3385 es->s_kbytes_written =
3383 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 3386 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3384 ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) - 3387 ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3385 EXT4_SB(sb)->s_sectors_written_start) >> 1)); 3388 EXT4_SB(sb)->s_sectors_written_start) >> 1));
3386 ext4_free_blocks_count_set(es, percpu_counter_sum_positive( 3389 ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
@@ -3485,8 +3488,10 @@ int ext4_force_commit(struct super_block *sb)
3485 return 0; 3488 return 0;
3486 3489
3487 journal = EXT4_SB(sb)->s_journal; 3490 journal = EXT4_SB(sb)->s_journal;
3488 if (journal) 3491 if (journal) {
3492 vfs_check_frozen(sb, SB_FREEZE_WRITE);
3489 ret = ext4_journal_force_commit(journal); 3493 ret = ext4_journal_force_commit(journal);
3494 }
3490 3495
3491 return ret; 3496 return ret;
3492} 3497}
@@ -3535,18 +3540,16 @@ static int ext4_freeze(struct super_block *sb)
3535 * the journal. 3540 * the journal.
3536 */ 3541 */
3537 error = jbd2_journal_flush(journal); 3542 error = jbd2_journal_flush(journal);
3538 if (error < 0) { 3543 if (error < 0)
3539 out: 3544 goto out;
3540 jbd2_journal_unlock_updates(journal);
3541 return error;
3542 }
3543 3545
3544 /* Journal blocked and flushed, clear needs_recovery flag. */ 3546 /* Journal blocked and flushed, clear needs_recovery flag. */
3545 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER); 3547 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3546 error = ext4_commit_super(sb, 1); 3548 error = ext4_commit_super(sb, 1);
3547 if (error) 3549out:
3548 goto out; 3550 /* we rely on s_frozen to stop further updates */
3549 return 0; 3551 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3552 return error;
3550} 3553}
3551 3554
3552/* 3555/*
@@ -3563,7 +3566,6 @@ static int ext4_unfreeze(struct super_block *sb)
3563 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER); 3566 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3564 ext4_commit_super(sb, 1); 3567 ext4_commit_super(sb, 1);
3565 unlock_super(sb); 3568 unlock_super(sb);
3566 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3567 return 0; 3569 return 0;
3568} 3570}
3569 3571
@@ -3580,6 +3582,7 @@ static int ext4_remount(struct super_block *sb, int *flags, char *data)
3580#ifdef CONFIG_QUOTA 3582#ifdef CONFIG_QUOTA
3581 int i; 3583 int i;
3582#endif 3584#endif
3585 char *orig_data = kstrdup(data, GFP_KERNEL);
3583 3586
3584 lock_kernel(); 3587 lock_kernel();
3585 3588
@@ -3713,6 +3716,9 @@ static int ext4_remount(struct super_block *sb, int *flags, char *data)
3713#endif 3716#endif
3714 unlock_super(sb); 3717 unlock_super(sb);
3715 unlock_kernel(); 3718 unlock_kernel();
3719
3720 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
3721 kfree(orig_data);
3716 return 0; 3722 return 0;
3717 3723
3718restore_opts: 3724restore_opts:
@@ -3734,6 +3740,7 @@ restore_opts:
3734#endif 3740#endif
3735 unlock_super(sb); 3741 unlock_super(sb);
3736 unlock_kernel(); 3742 unlock_kernel();
3743 kfree(orig_data);
3737 return err; 3744 return err;
3738} 3745}
3739 3746
@@ -4141,6 +4148,7 @@ static int __init init_ext4_fs(void)
4141{ 4148{
4142 int err; 4149 int err;
4143 4150
4151 ext4_check_flag_values();
4144 err = init_ext4_system_zone(); 4152 err = init_ext4_system_zone();
4145 if (err) 4153 if (err)
4146 return err; 4154 return err;
diff --git a/fs/ext4/symlink.c b/fs/ext4/symlink.c
index 00740cb32be3..ed9354aff279 100644
--- a/fs/ext4/symlink.c
+++ b/fs/ext4/symlink.c
@@ -34,6 +34,7 @@ const struct inode_operations ext4_symlink_inode_operations = {
34 .readlink = generic_readlink, 34 .readlink = generic_readlink,
35 .follow_link = page_follow_link_light, 35 .follow_link = page_follow_link_light,
36 .put_link = page_put_link, 36 .put_link = page_put_link,
37 .setattr = ext4_setattr,
37#ifdef CONFIG_EXT4_FS_XATTR 38#ifdef CONFIG_EXT4_FS_XATTR
38 .setxattr = generic_setxattr, 39 .setxattr = generic_setxattr,
39 .getxattr = generic_getxattr, 40 .getxattr = generic_getxattr,
@@ -45,6 +46,7 @@ const struct inode_operations ext4_symlink_inode_operations = {
45const struct inode_operations ext4_fast_symlink_inode_operations = { 46const struct inode_operations ext4_fast_symlink_inode_operations = {
46 .readlink = generic_readlink, 47 .readlink = generic_readlink,
47 .follow_link = ext4_follow_link, 48 .follow_link = ext4_follow_link,
49 .setattr = ext4_setattr,
48#ifdef CONFIG_EXT4_FS_XATTR 50#ifdef CONFIG_EXT4_FS_XATTR
49 .setxattr = generic_setxattr, 51 .setxattr = generic_setxattr,
50 .getxattr = generic_getxattr, 52 .getxattr = generic_getxattr,
diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
index 2de0e9515089..04338009793a 100644
--- a/fs/ext4/xattr.c
+++ b/fs/ext4/xattr.c
@@ -228,9 +228,8 @@ ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
228 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount)); 228 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
229 if (ext4_xattr_check_block(bh)) { 229 if (ext4_xattr_check_block(bh)) {
230bad_block: 230bad_block:
231 ext4_error(inode->i_sb, 231 EXT4_ERROR_INODE(inode, "bad block %llu",
232 "inode %lu: bad block %llu", inode->i_ino, 232 EXT4_I(inode)->i_file_acl);
233 EXT4_I(inode)->i_file_acl);
234 error = -EIO; 233 error = -EIO;
235 goto cleanup; 234 goto cleanup;
236 } 235 }
@@ -372,9 +371,8 @@ ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
372 ea_bdebug(bh, "b_count=%d, refcount=%d", 371 ea_bdebug(bh, "b_count=%d, refcount=%d",
373 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount)); 372 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
374 if (ext4_xattr_check_block(bh)) { 373 if (ext4_xattr_check_block(bh)) {
375 ext4_error(inode->i_sb, 374 EXT4_ERROR_INODE(inode, "bad block %llu",
376 "inode %lu: bad block %llu", inode->i_ino, 375 EXT4_I(inode)->i_file_acl);
377 EXT4_I(inode)->i_file_acl);
378 error = -EIO; 376 error = -EIO;
379 goto cleanup; 377 goto cleanup;
380 } 378 }
@@ -666,8 +664,8 @@ ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
666 atomic_read(&(bs->bh->b_count)), 664 atomic_read(&(bs->bh->b_count)),
667 le32_to_cpu(BHDR(bs->bh)->h_refcount)); 665 le32_to_cpu(BHDR(bs->bh)->h_refcount));
668 if (ext4_xattr_check_block(bs->bh)) { 666 if (ext4_xattr_check_block(bs->bh)) {
669 ext4_error(sb, "inode %lu: bad block %llu", 667 EXT4_ERROR_INODE(inode, "bad block %llu",
670 inode->i_ino, EXT4_I(inode)->i_file_acl); 668 EXT4_I(inode)->i_file_acl);
671 error = -EIO; 669 error = -EIO;
672 goto cleanup; 670 goto cleanup;
673 } 671 }
@@ -820,7 +818,7 @@ inserted:
820 EXT4_I(inode)->i_block_group); 818 EXT4_I(inode)->i_block_group);
821 819
822 /* non-extent files can't have physical blocks past 2^32 */ 820 /* non-extent files can't have physical blocks past 2^32 */
823 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 821 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
824 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; 822 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
825 823
826 block = ext4_new_meta_blocks(handle, inode, 824 block = ext4_new_meta_blocks(handle, inode,
@@ -828,7 +826,7 @@ inserted:
828 if (error) 826 if (error)
829 goto cleanup; 827 goto cleanup;
830 828
831 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) 829 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
832 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS); 830 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
833 831
834 ea_idebug(inode, "creating block %d", block); 832 ea_idebug(inode, "creating block %d", block);
@@ -880,8 +878,8 @@ cleanup_dquot:
880 goto cleanup; 878 goto cleanup;
881 879
882bad_block: 880bad_block:
883 ext4_error(inode->i_sb, "inode %lu: bad block %llu", 881 EXT4_ERROR_INODE(inode, "bad block %llu",
884 inode->i_ino, EXT4_I(inode)->i_file_acl); 882 EXT4_I(inode)->i_file_acl);
885 goto cleanup; 883 goto cleanup;
886 884
887#undef header 885#undef header
@@ -1194,8 +1192,8 @@ retry:
1194 if (!bh) 1192 if (!bh)
1195 goto cleanup; 1193 goto cleanup;
1196 if (ext4_xattr_check_block(bh)) { 1194 if (ext4_xattr_check_block(bh)) {
1197 ext4_error(inode->i_sb, "inode %lu: bad block %llu", 1195 EXT4_ERROR_INODE(inode, "bad block %llu",
1198 inode->i_ino, EXT4_I(inode)->i_file_acl); 1196 EXT4_I(inode)->i_file_acl);
1199 error = -EIO; 1197 error = -EIO;
1200 goto cleanup; 1198 goto cleanup;
1201 } 1199 }
@@ -1372,14 +1370,14 @@ ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1372 goto cleanup; 1370 goto cleanup;
1373 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl); 1371 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1374 if (!bh) { 1372 if (!bh) {
1375 ext4_error(inode->i_sb, "inode %lu: block %llu read error", 1373 EXT4_ERROR_INODE(inode, "block %llu read error",
1376 inode->i_ino, EXT4_I(inode)->i_file_acl); 1374 EXT4_I(inode)->i_file_acl);
1377 goto cleanup; 1375 goto cleanup;
1378 } 1376 }
1379 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) || 1377 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1380 BHDR(bh)->h_blocks != cpu_to_le32(1)) { 1378 BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1381 ext4_error(inode->i_sb, "inode %lu: bad block %llu", 1379 EXT4_ERROR_INODE(inode, "bad block %llu",
1382 inode->i_ino, EXT4_I(inode)->i_file_acl); 1380 EXT4_I(inode)->i_file_acl);
1383 goto cleanup; 1381 goto cleanup;
1384 } 1382 }
1385 ext4_xattr_release_block(handle, inode, bh); 1383 ext4_xattr_release_block(handle, inode, bh);
@@ -1504,9 +1502,8 @@ again:
1504 } 1502 }
1505 bh = sb_bread(inode->i_sb, ce->e_block); 1503 bh = sb_bread(inode->i_sb, ce->e_block);
1506 if (!bh) { 1504 if (!bh) {
1507 ext4_error(inode->i_sb, 1505 EXT4_ERROR_INODE(inode, "block %lu read error",
1508 "inode %lu: block %lu read error", 1506 (unsigned long) ce->e_block);
1509 inode->i_ino, (unsigned long) ce->e_block);
1510 } else if (le32_to_cpu(BHDR(bh)->h_refcount) >= 1507 } else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1511 EXT4_XATTR_REFCOUNT_MAX) { 1508 EXT4_XATTR_REFCOUNT_MAX) {
1512 ea_idebug(inode, "block %lu refcount %d>=%d", 1509 ea_idebug(inode, "block %lu refcount %d>=%d",
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-26 11:53:13 -0500