aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2011-08-01 19:56:03 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2011-08-01 19:56:03 -0400
commit60ad4466821a96913a9b567115e194ed1087c2d7 (patch)
treecd488ba72a60f856b85a467763fb633cbe7ef2d9
parent1b8e94993c4752d98c33903aa836acc15f7e6d5c (diff)
parent79a77c5ac34cc27ccbfbdf7113b41cdd93534eab (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: (60 commits) ext4: prevent memory leaks from ext4_mb_init_backend() on error path ext4: use EXT4_BAD_INO for buddy cache to avoid colliding with valid inode # ext4: use ext4_msg() instead of printk in mballoc ext4: use ext4_kvzalloc()/ext4_kvmalloc() for s_group_desc and s_group_info ext4: introduce ext4_kvmalloc(), ext4_kzalloc(), and ext4_kvfree() ext4: use the correct error exit path in ext4_init_inode_table() ext4: add missing kfree() on error return path in add_new_gdb() ext4: change umode_t in tracepoint headers to be an explicit __u16 ext4: fix races in ext4_sync_parent() ext4: Fix overflow caused by missing cast in ext4_fallocate() ext4: add action of moving index in ext4_ext_rm_idx for Punch Hole ext4: simplify parameters of reserve_backup_gdb() ext4: simplify parameters of add_new_gdb() ext4: remove lock_buffer in bclean() and setup_new_group_blocks() ext4: simplify journal handling in setup_new_group_blocks() ext4: let setup_new_group_blocks() set multiple bits at a time ext4: fix a typo in ext4_group_extend() ext4: let ext4_group_add_blocks() handle 0 blocks quickly ext4: let ext4_group_add_blocks() return an error code ext4: rename ext4_add_groupblocks() to ext4_group_add_blocks() ... Fix up conflict in fs/ext4/inode.c: commit aacfc19c626e ("fs: simplify the blockdev_direct_IO prototype") had changed the ext4_ind_direct_IO() function for the new simplified calling convention, while commit dae1e52cb126 ("ext4: move ext4_ind_* functions from inode.c to indirect.c") moved the function to another file.
Diffstat
-rw-r--r--fs/ext4/Makefile2
-rw-r--r--fs/ext4/balloc.c48
-rw-r--r--fs/ext4/block_validity.c21
-rw-r--r--fs/ext4/ext4.h55
-rw-r--r--fs/ext4/extents.c129
-rw-r--r--fs/ext4/fsync.c26
-rw-r--r--fs/ext4/ialloc.c2
-rw-r--r--fs/ext4/indirect.c1482
-rw-r--r--fs/ext4/inode.c1596
-rw-r--r--fs/ext4/ioctl.c12
-rw-r--r--fs/ext4/mballoc.c230
-rw-r--r--fs/ext4/mballoc.h1
-rw-r--r--fs/ext4/namei.c21
-rw-r--r--fs/ext4/page-io.c6
-rw-r--r--fs/ext4/resize.c199
-rw-r--r--fs/ext4/super.c88
-rw-r--r--fs/ext4/truncate.h43
-rw-r--r--fs/jbd2/checkpoint.c5
-rw-r--r--fs/jbd2/journal.c67
-rw-r--r--include/linux/jbd2.h6
-rw-r--r--include/trace/events/ext4.h87
-rw-r--r--include/trace/events/jbd2.h36
22 files changed, 2190 insertions, 1972 deletions
diff --git a/fs/ext4/Makefile b/fs/ext4/Makefile
index 04109460ba9..56fd8f86593 100644
--- a/fs/ext4/Makefile
+++ b/fs/ext4/Makefile
@@ -7,7 +7,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
7ext4-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \ 7ext4-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
8 ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \ 8 ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
9 ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \ 9 ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
10 mmp.o 10 mmp.o indirect.o
11 11
12ext4-$(CONFIG_EXT4_FS_XATTR) += xattr.o xattr_user.o xattr_trusted.o 12ext4-$(CONFIG_EXT4_FS_XATTR) += xattr.o xattr_user.o xattr_trusted.o
13ext4-$(CONFIG_EXT4_FS_POSIX_ACL) += acl.o 13ext4-$(CONFIG_EXT4_FS_POSIX_ACL) += acl.o
diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index 264f6949511..f8224adf496 100644
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -620,3 +620,51 @@ unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
620 620
621} 621}
622 622
623/**
624 * ext4_inode_to_goal_block - return a hint for block allocation
625 * @inode: inode for block allocation
626 *
627 * Return the ideal location to start allocating blocks for a
628 * newly created inode.
629 */
630ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
631{
632 struct ext4_inode_info *ei = EXT4_I(inode);
633 ext4_group_t block_group;
634 ext4_grpblk_t colour;
635 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
636 ext4_fsblk_t bg_start;
637 ext4_fsblk_t last_block;
638
639 block_group = ei->i_block_group;
640 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
641 /*
642 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
643 * block groups per flexgroup, reserve the first block
644 * group for directories and special files. Regular
645 * files will start at the second block group. This
646 * tends to speed up directory access and improves
647 * fsck times.
648 */
649 block_group &= ~(flex_size-1);
650 if (S_ISREG(inode->i_mode))
651 block_group++;
652 }
653 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
654 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
655
656 /*
657 * If we are doing delayed allocation, we don't need take
658 * colour into account.
659 */
660 if (test_opt(inode->i_sb, DELALLOC))
661 return bg_start;
662
663 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
664 colour = (current->pid % 16) *
665 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
666 else
667 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
668 return bg_start + colour;
669}
670
diff --git a/fs/ext4/block_validity.c b/fs/ext4/block_validity.c
index fac90f3fba8..8efb2f0a344 100644
--- a/fs/ext4/block_validity.c
+++ b/fs/ext4/block_validity.c
@@ -246,3 +246,24 @@ int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
246 return 1; 246 return 1;
247} 247}
248 248
249int ext4_check_blockref(const char *function, unsigned int line,
250 struct inode *inode, __le32 *p, unsigned int max)
251{
252 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
253 __le32 *bref = p;
254 unsigned int blk;
255
256 while (bref < p+max) {
257 blk = le32_to_cpu(*bref++);
258 if (blk &&
259 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
260 blk, 1))) {
261 es->s_last_error_block = cpu_to_le64(blk);
262 ext4_error_inode(inode, function, line, blk,
263 "invalid block");
264 return -EIO;
265 }
266 }
267 return 0;
268}
269
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index fa44df87971..e717dfd2f2b 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -526,6 +526,7 @@ struct ext4_new_group_data {
526#define EXT4_FREE_BLOCKS_METADATA 0x0001 526#define EXT4_FREE_BLOCKS_METADATA 0x0001
527#define EXT4_FREE_BLOCKS_FORGET 0x0002 527#define EXT4_FREE_BLOCKS_FORGET 0x0002
528#define EXT4_FREE_BLOCKS_VALIDATED 0x0004 528#define EXT4_FREE_BLOCKS_VALIDATED 0x0004
529#define EXT4_FREE_BLOCKS_NO_QUOT_UPDATE 0x0008
529 530
530/* 531/*
531 * ioctl commands 532 * ioctl commands
@@ -939,6 +940,8 @@ struct ext4_inode_info {
939#define ext4_find_next_zero_bit find_next_zero_bit_le 940#define ext4_find_next_zero_bit find_next_zero_bit_le
940#define ext4_find_next_bit find_next_bit_le 941#define ext4_find_next_bit find_next_bit_le
941 942
943extern void ext4_set_bits(void *bm, int cur, int len);
944
942/* 945/*
943 * Maximal mount counts between two filesystem checks 946 * Maximal mount counts between two filesystem checks
944 */ 947 */
@@ -1126,7 +1129,8 @@ struct ext4_sb_info {
1126 struct journal_s *s_journal; 1129 struct journal_s *s_journal;
1127 struct list_head s_orphan; 1130 struct list_head s_orphan;
1128 struct mutex s_orphan_lock; 1131 struct mutex s_orphan_lock;
1129 struct mutex s_resize_lock; 1132 unsigned long s_resize_flags; /* Flags indicating if there
1133 is a resizer */
1130 unsigned long s_commit_interval; 1134 unsigned long s_commit_interval;
1131 u32 s_max_batch_time; 1135 u32 s_max_batch_time;
1132 u32 s_min_batch_time; 1136 u32 s_min_batch_time;
@@ -1214,6 +1218,9 @@ struct ext4_sb_info {
1214 1218
1215 /* Kernel thread for multiple mount protection */ 1219 /* Kernel thread for multiple mount protection */
1216 struct task_struct *s_mmp_tsk; 1220 struct task_struct *s_mmp_tsk;
1221
1222 /* record the last minlen when FITRIM is called. */
1223 atomic_t s_last_trim_minblks;
1217}; 1224};
1218 1225
1219static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb) 1226static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb)
@@ -1743,6 +1750,7 @@ extern unsigned ext4_init_block_bitmap(struct super_block *sb,
1743 struct ext4_group_desc *desc); 1750 struct ext4_group_desc *desc);
1744#define ext4_free_blocks_after_init(sb, group, desc) \ 1751#define ext4_free_blocks_after_init(sb, group, desc) \
1745 ext4_init_block_bitmap(sb, NULL, group, desc) 1752 ext4_init_block_bitmap(sb, NULL, group, desc)
1753ext4_fsblk_t ext4_inode_to_goal_block(struct inode *);
1746 1754
1747/* dir.c */ 1755/* dir.c */
1748extern int __ext4_check_dir_entry(const char *, unsigned int, struct inode *, 1756extern int __ext4_check_dir_entry(const char *, unsigned int, struct inode *,
@@ -1793,7 +1801,7 @@ extern void ext4_free_blocks(handle_t *handle, struct inode *inode,
1793 unsigned long count, int flags); 1801 unsigned long count, int flags);
1794extern int ext4_mb_add_groupinfo(struct super_block *sb, 1802extern int ext4_mb_add_groupinfo(struct super_block *sb,
1795 ext4_group_t i, struct ext4_group_desc *desc); 1803 ext4_group_t i, struct ext4_group_desc *desc);
1796extern void ext4_add_groupblocks(handle_t *handle, struct super_block *sb, 1804extern int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
1797 ext4_fsblk_t block, unsigned long count); 1805 ext4_fsblk_t block, unsigned long count);
1798extern int ext4_trim_fs(struct super_block *, struct fstrim_range *); 1806extern int ext4_trim_fs(struct super_block *, struct fstrim_range *);
1799 1807
@@ -1834,6 +1842,17 @@ extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf);
1834extern qsize_t *ext4_get_reserved_space(struct inode *inode); 1842extern qsize_t *ext4_get_reserved_space(struct inode *inode);
1835extern void ext4_da_update_reserve_space(struct inode *inode, 1843extern void ext4_da_update_reserve_space(struct inode *inode,
1836 int used, int quota_claim); 1844 int used, int quota_claim);
1845
1846/* indirect.c */
1847extern int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
1848 struct ext4_map_blocks *map, int flags);
1849extern ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
1850 const struct iovec *iov, loff_t offset,
1851 unsigned long nr_segs);
1852extern int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock);
1853extern int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk);
1854extern void ext4_ind_truncate(struct inode *inode);
1855
1837/* ioctl.c */ 1856/* ioctl.c */
1838extern long ext4_ioctl(struct file *, unsigned int, unsigned long); 1857extern long ext4_ioctl(struct file *, unsigned int, unsigned long);
1839extern long ext4_compat_ioctl(struct file *, unsigned int, unsigned long); 1858extern long ext4_compat_ioctl(struct file *, unsigned int, unsigned long);
@@ -1855,6 +1874,9 @@ extern int ext4_group_extend(struct super_block *sb,
1855 ext4_fsblk_t n_blocks_count); 1874 ext4_fsblk_t n_blocks_count);
1856 1875
1857/* super.c */ 1876/* super.c */
1877extern void *ext4_kvmalloc(size_t size, gfp_t flags);
1878extern void *ext4_kvzalloc(size_t size, gfp_t flags);
1879extern void ext4_kvfree(void *ptr);
1858extern void __ext4_error(struct super_block *, const char *, unsigned int, 1880extern void __ext4_error(struct super_block *, const char *, unsigned int,
1859 const char *, ...) 1881 const char *, ...)
1860 __attribute__ ((format (printf, 4, 5))); 1882 __attribute__ ((format (printf, 4, 5)));
@@ -2067,11 +2089,19 @@ struct ext4_group_info {
2067 * 5 free 8-block regions. */ 2089 * 5 free 8-block regions. */
2068}; 2090};
2069 2091
2070#define EXT4_GROUP_INFO_NEED_INIT_BIT 0 2092#define EXT4_GROUP_INFO_NEED_INIT_BIT 0
2093#define EXT4_GROUP_INFO_WAS_TRIMMED_BIT 1
2071 2094
2072#define EXT4_MB_GRP_NEED_INIT(grp) \ 2095#define EXT4_MB_GRP_NEED_INIT(grp) \
2073 (test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state))) 2096 (test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state)))
2074 2097
2098#define EXT4_MB_GRP_WAS_TRIMMED(grp) \
2099 (test_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, &((grp)->bb_state)))
2100#define EXT4_MB_GRP_SET_TRIMMED(grp) \
2101 (set_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, &((grp)->bb_state)))
2102#define EXT4_MB_GRP_CLEAR_TRIMMED(grp) \
2103 (clear_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, &((grp)->bb_state)))
2104
2075#define EXT4_MAX_CONTENTION 8 2105#define EXT4_MAX_CONTENTION 8
2076#define EXT4_CONTENTION_THRESHOLD 2 2106#define EXT4_CONTENTION_THRESHOLD 2
2077 2107
@@ -2123,6 +2153,19 @@ static inline void ext4_mark_super_dirty(struct super_block *sb)
2123} 2153}
2124 2154
2125/* 2155/*
2156 * Block validity checking
2157 */
2158#define ext4_check_indirect_blockref(inode, bh) \
2159 ext4_check_blockref(__func__, __LINE__, inode, \
2160 (__le32 *)(bh)->b_data, \
2161 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
2162
2163#define ext4_ind_check_inode(inode) \
2164 ext4_check_blockref(__func__, __LINE__, inode, \
2165 EXT4_I(inode)->i_data, \
2166 EXT4_NDIR_BLOCKS)
2167
2168/*
2126 * Inodes and files operations 2169 * Inodes and files operations
2127 */ 2170 */
2128 2171
@@ -2151,6 +2194,8 @@ extern void ext4_exit_system_zone(void);
2151extern int ext4_data_block_valid(struct ext4_sb_info *sbi, 2194extern int ext4_data_block_valid(struct ext4_sb_info *sbi,
2152 ext4_fsblk_t start_blk, 2195 ext4_fsblk_t start_blk,
2153 unsigned int count); 2196 unsigned int count);
2197extern int ext4_check_blockref(const char *, unsigned int,
2198 struct inode *, __le32 *, unsigned int);
2154 2199
2155/* extents.c */ 2200/* extents.c */
2156extern int ext4_ext_tree_init(handle_t *handle, struct inode *); 2201extern int ext4_ext_tree_init(handle_t *handle, struct inode *);
@@ -2230,6 +2275,10 @@ static inline void set_bitmap_uptodate(struct buffer_head *bh)
2230extern wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ]; 2275extern wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
2231extern struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ]; 2276extern struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
2232 2277
2278#define EXT4_RESIZING 0
2279extern int ext4_resize_begin(struct super_block *sb);
2280extern void ext4_resize_end(struct super_block *sb);
2281
2233#endif /* __KERNEL__ */ 2282#endif /* __KERNEL__ */
2234 2283
2235#endif /* _EXT4_H */ 2284#endif /* _EXT4_H */
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index f815cc81e7a..57cf568a98a 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -114,12 +114,6 @@ static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
114 struct ext4_ext_path *path, 114 struct ext4_ext_path *path,
115 ext4_lblk_t block) 115 ext4_lblk_t block)
116{ 116{
117 struct ext4_inode_info *ei = EXT4_I(inode);
118 ext4_fsblk_t bg_start;
119 ext4_fsblk_t last_block;
120 ext4_grpblk_t colour;
121 ext4_group_t block_group;
122 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
123 int depth; 117 int depth;
124 118
125 if (path) { 119 if (path) {
@@ -161,36 +155,7 @@ static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
161 } 155 }
162 156
163 /* OK. use inode's group */ 157 /* OK. use inode's group */
164 block_group = ei->i_block_group; 158 return ext4_inode_to_goal_block(inode);
165 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
166 /*
167 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
168 * block groups per flexgroup, reserve the first block
169 * group for directories and special files. Regular
170 * files will start at the second block group. This
171 * tends to speed up directory access and improves
172 * fsck times.
173 */
174 block_group &= ~(flex_size-1);
175 if (S_ISREG(inode->i_mode))
176 block_group++;
177 }
178 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
179 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
180
181 /*
182 * If we are doing delayed allocation, we don't need take
183 * colour into account.
184 */
185 if (test_opt(inode->i_sb, DELALLOC))
186 return bg_start;
187
188 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
189 colour = (current->pid % 16) *
190 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
191 else
192 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
193 return bg_start + colour + block;
194} 159}
195 160
196/* 161/*
@@ -776,6 +741,16 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
776 logical, le32_to_cpu(curp->p_idx->ei_block)); 741 logical, le32_to_cpu(curp->p_idx->ei_block));
777 return -EIO; 742 return -EIO;
778 } 743 }
744
745 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
746 >= le16_to_cpu(curp->p_hdr->eh_max))) {
747 EXT4_ERROR_INODE(inode,
748 "eh_entries %d >= eh_max %d!",
749 le16_to_cpu(curp->p_hdr->eh_entries),
750 le16_to_cpu(curp->p_hdr->eh_max));
751 return -EIO;
752 }
753
779 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; 754 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
780 if (logical > le32_to_cpu(curp->p_idx->ei_block)) { 755 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
781 /* insert after */ 756 /* insert after */
@@ -805,13 +780,6 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
805 ext4_idx_store_pblock(ix, ptr); 780 ext4_idx_store_pblock(ix, ptr);
806 le16_add_cpu(&curp->p_hdr->eh_entries, 1); 781 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
807 782
808 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
809 > le16_to_cpu(curp->p_hdr->eh_max))) {
810 EXT4_ERROR_INODE(inode,
811 "logical %d == ei_block %d!",
812 logical, le32_to_cpu(curp->p_idx->ei_block));
813 return -EIO;
814 }
815 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { 783 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
816 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); 784 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
817 return -EIO; 785 return -EIO;
@@ -1446,8 +1414,7 @@ ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1446 * ext4_ext_next_leaf_block: 1414 * ext4_ext_next_leaf_block:
1447 * returns first allocated block from next leaf or EXT_MAX_BLOCKS 1415 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1448 */ 1416 */
1449static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode, 1417static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1450 struct ext4_ext_path *path)
1451{ 1418{
1452 int depth; 1419 int depth;
1453 1420
@@ -1757,7 +1724,6 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1757 goto merge; 1724 goto merge;
1758 } 1725 }
1759 1726
1760repeat:
1761 depth = ext_depth(inode); 1727 depth = ext_depth(inode);
1762 eh = path[depth].p_hdr; 1728 eh = path[depth].p_hdr;
1763 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) 1729 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
@@ -1765,9 +1731,10 @@ repeat:
1765 1731
1766 /* probably next leaf has space for us? */ 1732 /* probably next leaf has space for us? */
1767 fex = EXT_LAST_EXTENT(eh); 1733 fex = EXT_LAST_EXTENT(eh);
1768 next = ext4_ext_next_leaf_block(inode, path); 1734 next = EXT_MAX_BLOCKS;
1769 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) 1735 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1770 && next != EXT_MAX_BLOCKS) { 1736 next = ext4_ext_next_leaf_block(path);
1737 if (next != EXT_MAX_BLOCKS) {
1771 ext_debug("next leaf block - %d\n", next); 1738 ext_debug("next leaf block - %d\n", next);
1772 BUG_ON(npath != NULL); 1739 BUG_ON(npath != NULL);
1773 npath = ext4_ext_find_extent(inode, next, NULL); 1740 npath = ext4_ext_find_extent(inode, next, NULL);
@@ -1779,7 +1746,7 @@ repeat:
1779 ext_debug("next leaf isn't full(%d)\n", 1746 ext_debug("next leaf isn't full(%d)\n",
1780 le16_to_cpu(eh->eh_entries)); 1747 le16_to_cpu(eh->eh_entries));
1781 path = npath; 1748 path = npath;
1782 goto repeat; 1749 goto has_space;
1783 } 1750 }
1784 ext_debug("next leaf has no free space(%d,%d)\n", 1751 ext_debug("next leaf has no free space(%d,%d)\n",
1785 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); 1752 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
@@ -1839,7 +1806,7 @@ has_space:
1839 ext4_ext_pblock(newext), 1806 ext4_ext_pblock(newext),
1840 ext4_ext_is_uninitialized(newext), 1807 ext4_ext_is_uninitialized(newext),
1841 ext4_ext_get_actual_len(newext), 1808 ext4_ext_get_actual_len(newext),
1842 nearex, len, nearex + 1, nearex + 2); 1809 nearex, len, nearex, nearex + 1);
1843 memmove(nearex + 1, nearex, len); 1810 memmove(nearex + 1, nearex, len);
1844 path[depth].p_ext = nearex; 1811 path[depth].p_ext = nearex;
1845 } 1812 }
@@ -2052,7 +2019,7 @@ ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2052} 2019}
2053 2020
2054/* 2021/*
2055 * ext4_ext_in_cache() 2022 * ext4_ext_check_cache()
2056 * Checks to see if the given block is in the cache. 2023 * Checks to see if the given block is in the cache.
2057 * If it is, the cached extent is stored in the given 2024 * If it is, the cached extent is stored in the given
2058 * cache extent pointer. If the cached extent is a hole, 2025 * cache extent pointer. If the cached extent is a hole,
@@ -2134,8 +2101,6 @@ ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2134/* 2101/*
2135 * ext4_ext_rm_idx: 2102 * ext4_ext_rm_idx:
2136 * removes index from the index block. 2103 * removes index from the index block.
2137 * It's used in truncate case only, thus all requests are for
2138 * last index in the block only.
2139 */ 2104 */
2140static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, 2105static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2141 struct ext4_ext_path *path) 2106 struct ext4_ext_path *path)
@@ -2153,6 +2118,13 @@ static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2153 err = ext4_ext_get_access(handle, inode, path); 2118 err = ext4_ext_get_access(handle, inode, path);
2154 if (err) 2119 if (err)
2155 return err; 2120 return err;
2121
2122 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2123 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2124 len *= sizeof(struct ext4_extent_idx);
2125 memmove(path->p_idx, path->p_idx + 1, len);
2126 }
2127
2156 le16_add_cpu(&path->p_hdr->eh_entries, -1); 2128 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2157 err = ext4_ext_dirty(handle, inode, path); 2129 err = ext4_ext_dirty(handle, inode, path);
2158 if (err) 2130 if (err)
@@ -2534,8 +2506,7 @@ ext4_ext_more_to_rm(struct ext4_ext_path *path)
2534 return 1; 2506 return 1;
2535} 2507}
2536 2508
2537static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, 2509static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2538 ext4_lblk_t end)
2539{ 2510{
2540 struct super_block *sb = inode->i_sb; 2511 struct super_block *sb = inode->i_sb;
2541 int depth = ext_depth(inode); 2512 int depth = ext_depth(inode);
@@ -2575,7 +2546,7 @@ again:
2575 if (i == depth) { 2546 if (i == depth) {
2576 /* this is leaf block */ 2547 /* this is leaf block */
2577 err = ext4_ext_rm_leaf(handle, inode, path, 2548 err = ext4_ext_rm_leaf(handle, inode, path,
2578 start, end); 2549 start, EXT_MAX_BLOCKS - 1);
2579 /* root level has p_bh == NULL, brelse() eats this */ 2550 /* root level has p_bh == NULL, brelse() eats this */
2580 brelse(path[i].p_bh); 2551 brelse(path[i].p_bh);
2581 path[i].p_bh = NULL; 2552 path[i].p_bh = NULL;
@@ -3107,12 +3078,10 @@ static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3107 struct ext4_ext_path *path) 3078 struct ext4_ext_path *path)
3108{ 3079{
3109 struct ext4_extent *ex; 3080 struct ext4_extent *ex;
3110 struct ext4_extent_header *eh;
3111 int depth; 3081 int depth;
3112 int err = 0; 3082 int err = 0;
3113 3083
3114 depth = ext_depth(inode); 3084 depth = ext_depth(inode);
3115 eh = path[depth].p_hdr;
3116 ex = path[depth].p_ext; 3085 ex = path[depth].p_ext;
3117 3086
3118 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" 3087 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
@@ -3357,8 +3326,8 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3357 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); 3326 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3358 3327
3359 /* check in cache */ 3328 /* check in cache */
3360 if (ext4_ext_in_cache(inode, map->m_lblk, &newex) && 3329 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3361 ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0)) { 3330 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3362 if (!newex.ee_start_lo && !newex.ee_start_hi) { 3331 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3363 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { 3332 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3364 /* 3333 /*
@@ -3497,8 +3466,27 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3497 3466
3498 ext4_ext_mark_uninitialized(ex); 3467 ext4_ext_mark_uninitialized(ex);
3499 3468
3500 err = ext4_ext_remove_space(inode, map->m_lblk, 3469 ext4_ext_invalidate_cache(inode);
3501 map->m_lblk + punched_out); 3470
3471 err = ext4_ext_rm_leaf(handle, inode, path,
3472 map->m_lblk, map->m_lblk + punched_out);
3473
3474 if (!err && path->p_hdr->eh_entries == 0) {
3475 /*
3476 * Punch hole freed all of this sub tree,
3477 * so we need to correct eh_depth
3478 */
3479 err = ext4_ext_get_access(handle, inode, path);
3480 if (err == 0) {
3481 ext_inode_hdr(inode)->eh_depth = 0;
3482 ext_inode_hdr(inode)->eh_max =
3483 cpu_to_le16(ext4_ext_space_root(
3484 inode, 0));
3485
3486 err = ext4_ext_dirty(
3487 handle, inode, path);
3488 }
3489 }
3502 3490
3503 goto out2; 3491 goto out2;
3504 } 3492 }
@@ -3596,17 +3584,18 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3596 } 3584 }
3597 3585
3598 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len); 3586 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3599 if (err) 3587 if (!err)
3600 goto out2; 3588 err = ext4_ext_insert_extent(handle, inode, path,
3601 3589 &newex, flags);
3602 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3603 if (err) { 3590 if (err) {
3591 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
3592 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
3604 /* free data blocks we just allocated */ 3593 /* free data blocks we just allocated */
3605 /* not a good idea to call discard here directly, 3594 /* not a good idea to call discard here directly,
3606 * but otherwise we'd need to call it every free() */ 3595 * but otherwise we'd need to call it every free() */
3607 ext4_discard_preallocations(inode); 3596 ext4_discard_preallocations(inode);
3608 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), 3597 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3609 ext4_ext_get_actual_len(&newex), 0); 3598 ext4_ext_get_actual_len(&newex), fb_flags);
3610 goto out2; 3599 goto out2;
3611 } 3600 }
3612 3601
@@ -3699,7 +3688,7 @@ void ext4_ext_truncate(struct inode *inode)
3699 3688
3700 last_block = (inode->i_size + sb->s_blocksize - 1) 3689 last_block = (inode->i_size + sb->s_blocksize - 1)
3701 >> EXT4_BLOCK_SIZE_BITS(sb); 3690 >> EXT4_BLOCK_SIZE_BITS(sb);
3702 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); 3691 err = ext4_ext_remove_space(inode, last_block);
3703 3692
3704 /* In a multi-transaction truncate, we only make the final 3693 /* In a multi-transaction truncate, we only make the final
3705 * transaction synchronous. 3694 * transaction synchronous.
@@ -3835,7 +3824,7 @@ retry:
3835 blkbits) >> blkbits)) 3824 blkbits) >> blkbits))
3836 new_size = offset + len; 3825 new_size = offset + len;
3837 else 3826 else
3838 new_size = (map.m_lblk + ret) << blkbits; 3827 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
3839 3828
3840 ext4_falloc_update_inode(inode, mode, new_size, 3829 ext4_falloc_update_inode(inode, mode, new_size,
3841 (map.m_flags & EXT4_MAP_NEW)); 3830 (map.m_flags & EXT4_MAP_NEW));
diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index da3bed3e0c2..036f78f7a1e 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -129,15 +129,30 @@ static int ext4_sync_parent(struct inode *inode)
129{ 129{
130 struct writeback_control wbc; 130 struct writeback_control wbc;
131 struct dentry *dentry = NULL; 131 struct dentry *dentry = NULL;
132 struct inode *next;
132 int ret = 0; 133 int ret = 0;
133 134
134 while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) { 135 if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
136 return 0;
137 inode = igrab(inode);
138 while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
135 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY); 139 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
136 dentry = list_entry(inode->i_dentry.next, 140 dentry = NULL;
137 struct dentry, d_alias); 141 spin_lock(&inode->i_lock);
138 if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode) 142 if (!list_empty(&inode->i_dentry)) {
143 dentry = list_first_entry(&inode->i_dentry,
144 struct dentry, d_alias);
145 dget(dentry);
146 }
147 spin_unlock(&inode->i_lock);
148 if (!dentry)
139 break; 149 break;
140 inode = dentry->d_parent->d_inode; 150 next = igrab(dentry->d_parent->d_inode);
151 dput(dentry);
152 if (!next)
153 break;
154 iput(inode);
155 inode = next;
141 ret = sync_mapping_buffers(inode->i_mapping); 156 ret = sync_mapping_buffers(inode->i_mapping);
142 if (ret) 157 if (ret)
143 break; 158 break;
@@ -148,6 +163,7 @@ static int ext4_sync_parent(struct inode *inode)
148 if (ret) 163 if (ret)
149 break; 164 break;
150 } 165 }
166 iput(inode);
151 return ret; 167 return ret;
152} 168}
153 169
diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index 21bb2f61e50..9c63f273b55 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -1287,7 +1287,7 @@ extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1287 group, used_blks, 1287 group, used_blks,
1288 ext4_itable_unused_count(sb, gdp)); 1288 ext4_itable_unused_count(sb, gdp));
1289 ret = 1; 1289 ret = 1;
1290 goto out; 1290 goto err_out;
1291 } 1291 }
1292 1292
1293 blk = ext4_inode_table(sb, gdp) + used_blks; 1293 blk = ext4_inode_table(sb, gdp) + used_blks;
diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
new file mode 100644
index 00000000000..b8602cde5b5
--- /dev/null
+++ b/fs/ext4/indirect.c
@@ -0,0 +1,1482 @@
1/*
2 * linux/fs/ext4/indirect.c
3 *
4 * from
5 *
6 * linux/fs/ext4/inode.c
7 *
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
12 *
13 * from
14 *
15 * linux/fs/minix/inode.c
16 *
17 * Copyright (C) 1991, 1992 Linus Torvalds
18 *
19 * Goal-directed block allocation by Stephen Tweedie
20 * (sct@redhat.com), 1993, 1998
21 */
22
23#include <linux/module.h>
24#include "ext4_jbd2.h"
25#include "truncate.h"
26
27#include <trace/events/ext4.h>
28
29typedef struct {
30 __le32 *p;
31 __le32 key;
32 struct buffer_head *bh;
33} Indirect;
34
35static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
36{
37 p->key = *(p->p = v);
38 p->bh = bh;
39}
40
41/**
42 * ext4_block_to_path - parse the block number into array of offsets
43 * @inode: inode in question (we are only interested in its superblock)
44 * @i_block: block number to be parsed
45 * @offsets: array to store the offsets in
46 * @boundary: set this non-zero if the referred-to block is likely to be
47 * followed (on disk) by an indirect block.
48 *
49 * To store the locations of file's data ext4 uses a data structure common
50 * for UNIX filesystems - tree of pointers anchored in the inode, with
51 * data blocks at leaves and indirect blocks in intermediate nodes.
52 * This function translates the block number into path in that tree -
53 * return value is the path length and @offsets[n] is the offset of
54 * pointer to (n+1)th node in the nth one. If @block is out of range
55 * (negative or too large) warning is printed and zero returned.
56 *
57 * Note: function doesn't find node addresses, so no IO is needed. All
58 * we need to know is the capacity of indirect blocks (taken from the
59 * inode->i_sb).
60 */
61
62/*
63 * Portability note: the last comparison (check that we fit into triple
64 * indirect block) is spelled differently, because otherwise on an
65 * architecture with 32-bit longs and 8Kb pages we might get into trouble
66 * if our filesystem had 8Kb blocks. We might use long long, but that would
67 * kill us on x86. Oh, well, at least the sign propagation does not matter -
68 * i_block would have to be negative in the very beginning, so we would not
69 * get there at all.
70 */
71
72static int ext4_block_to_path(struct inode *inode,
73 ext4_lblk_t i_block,
74 ext4_lblk_t offsets[4], int *boundary)
75{
76 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
77 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
78 const long direct_blocks = EXT4_NDIR_BLOCKS,
79 indirect_blocks = ptrs,
80 double_blocks = (1 << (ptrs_bits * 2));
81 int n = 0;
82 int final = 0;
83
84 if (i_block < direct_blocks) {
85 offsets[n++] = i_block;
86 final = direct_blocks;
87 } else if ((i_block -= direct_blocks) < indirect_blocks) {
88 offsets[n++] = EXT4_IND_BLOCK;
89 offsets[n++] = i_block;
90 final = ptrs;
91 } else if ((i_block -= indirect_blocks) < double_blocks) {
92 offsets[n++] = EXT4_DIND_BLOCK;
93 offsets[n++] = i_block >> ptrs_bits;
94 offsets[n++] = i_block & (ptrs - 1);
95 final = ptrs;
96 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
97 offsets[n++] = EXT4_TIND_BLOCK;
98 offsets[n++] = i_block >> (ptrs_bits * 2);
99 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
100 offsets[n++] = i_block & (ptrs - 1);
101 final = ptrs;
102 } else {
103 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
104 i_block + direct_blocks +
105 indirect_blocks + double_blocks, inode->i_ino);
106 }
107 if (boundary)
108 *boundary = final - 1 - (i_block & (ptrs - 1));
109 return n;
110}
111
112/**
113 * ext4_get_branch - read the chain of indirect blocks leading to data
114 * @inode: inode in question
115 * @depth: depth of the chain (1 - direct pointer, etc.)
116 * @offsets: offsets of pointers in inode/indirect blocks
117 * @chain: place to store the result
118 * @err: here we store the error value
119 *
120 * Function fills the array of triples <key, p, bh> and returns %NULL
121 * if everything went OK or the pointer to the last filled triple
122 * (incomplete one) otherwise. Upon the return chain[i].key contains
123 * the number of (i+1)-th block in the chain (as it is stored in memory,
124 * i.e. little-endian 32-bit), chain[i].p contains the address of that
125 * number (it points into struct inode for i==0 and into the bh->b_data
126 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
127 * block for i>0 and NULL for i==0. In other words, it holds the block
128 * numbers of the chain, addresses they were taken from (and where we can
129 * verify that chain did not change) and buffer_heads hosting these
130 * numbers.
131 *
132 * Function stops when it stumbles upon zero pointer (absent block)
133 * (pointer to last triple returned, *@err == 0)
134 * or when it gets an IO error reading an indirect block
135 * (ditto, *@err == -EIO)
136 * or when it reads all @depth-1 indirect blocks successfully and finds
137 * the whole chain, all way to the data (returns %NULL, *err == 0).
138 *
139 * Need to be called with
140 * down_read(&EXT4_I(inode)->i_data_sem)
141 */
142static Indirect *ext4_get_branch(struct inode *inode, int depth,
143 ext4_lblk_t *offsets,
144 Indirect chain[4], int *err)
145{
146 struct super_block *sb = inode->i_sb;
147 Indirect *p = chain;
148 struct buffer_head *bh;
149
150 *err = 0;
151 /* i_data is not going away, no lock needed */
152 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
153 if (!p->key)
154 goto no_block;
155 while (--depth) {
156 bh = sb_getblk(sb, le32_to_cpu(p->key));
157 if (unlikely(!bh))
158 goto failure;
159
160 if (!bh_uptodate_or_lock(bh)) {
161 if (bh_submit_read(bh) < 0) {
162 put_bh(bh);
163 goto failure;
164 }
165 /* validate block references */
166 if (ext4_check_indirect_blockref(inode, bh)) {
167 put_bh(bh);
168 goto failure;
169 }
170 }
171
172 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
173 /* Reader: end */
174 if (!p->key)
175 goto no_block;
176 }
177 return NULL;
178
179failure:
180 *err = -EIO;
181no_block:
182 return p;
183}
184
185/**
186 * ext4_find_near - find a place for allocation with sufficient locality
187 * @inode: owner
188 * @ind: descriptor of indirect block.
189 *
190 * This function returns the preferred place for block allocation.
191 * It is used when heuristic for sequential allocation fails.
192 * Rules are:
193 * + if there is a block to the left of our position - allocate near it.
194 * + if pointer will live in indirect block - allocate near that block.
195 * + if pointer will live in inode - allocate in the same
196 * cylinder group.
197 *
198 * In the latter case we colour the starting block by the callers PID to
199 * prevent it from clashing with concurrent allocations for a different inode
200 * in the same block group. The PID is used here so that functionally related
201 * files will be close-by on-disk.
202 *
203 * Caller must make sure that @ind is valid and will stay that way.
204 */
205static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
206{
207 struct ext4_inode_info *ei = EXT4_I(inode);
208 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
209 __le32 *p;
210
211 /* Try to find previous block */
212 for (p = ind->p - 1; p >= start; p--) {
213 if (*p)
214 return le32_to_cpu(*p);
215 }
216
217 /* No such thing, so let's try location of indirect block */
218 if (ind->bh)
219 return ind->bh->b_blocknr;
220
221 /*
222 * It is going to be referred to from the inode itself? OK, just put it
223 * into the same cylinder group then.
224 */
225 return ext4_inode_to_goal_block(inode);
226}
227
228/**
229 * ext4_find_goal - find a preferred place for allocation.
230 * @inode: owner
231 * @block: block we want
232 * @partial: pointer to the last triple within a chain
233 *
234 * Normally this function find the preferred place for block allocation,
235 * returns it.
236 * Because this is only used for non-extent files, we limit the block nr
237 * to 32 bits.
238 */
239static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
240 Indirect *partial)
241{
242 ext4_fsblk_t goal;
243
244 /*
245 * XXX need to get goal block from mballoc's data structures
246 */
247
248 goal = ext4_find_near(inode, partial);
249 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
250 return goal;
251}
252
253/**
254 * ext4_blks_to_allocate - Look up the block map and count the number
255 * of direct blocks need to be allocated for the given branch.
256 *
257 * @branch: chain of indirect blocks
258 * @k: number of blocks need for indirect blocks
259 * @blks: number of data blocks to be mapped.
260 * @blocks_to_boundary: the offset in the indirect block
261 *
262 * return the total number of blocks to be allocate, including the
263 * direct and indirect blocks.
264 */
265static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
266 int blocks_to_boundary)
267{
268 unsigned int count = 0;
269
270 /*
271 * Simple case, [t,d]Indirect block(s) has not allocated yet
272 * then it's clear blocks on that path have not allocated
273 */
274 if (k > 0) {
275 /* right now we don't handle cross boundary allocation */
276 if (blks < blocks_to_boundary + 1)
277 count += blks;
278 else
279 count += blocks_to_boundary + 1;
280 return count;
281 }
282
283 count++;
284 while (count < blks && count <= blocks_to_boundary &&
285 le32_to_cpu(*(branch[0].p + count)) == 0) {
286 count++;
287 }
288 return count;
289}
290
291/**
292 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
293 * @handle: handle for this transaction
294 * @inode: inode which needs allocated blocks
295 * @iblock: the logical block to start allocated at
296 * @goal: preferred physical block of allocation
297 * @indirect_blks: the number of blocks need to allocate for indirect
298 * blocks
299 * @blks: number of desired blocks
300 * @new_blocks: on return it will store the new block numbers for
301 * the indirect blocks(if needed) and the first direct block,
302 * @err: on return it will store the error code
303 *
304 * This function will return the number of blocks allocated as
305 * requested by the passed-in parameters.
306 */
307static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
308 ext4_lblk_t iblock, ext4_fsblk_t goal,
309 int indirect_blks, int blks,
310 ext4_fsblk_t new_blocks[4], int *err)
311{
312 struct ext4_allocation_request ar;
313 int target, i;
314 unsigned long count = 0, blk_allocated = 0;
315 int index = 0;
316 ext4_fsblk_t current_block = 0;
317 int ret = 0;
318
319 /*
320 * Here we try to allocate the requested multiple blocks at once,
321 * on a best-effort basis.
322 * To build a branch, we should allocate blocks for
323 * the indirect blocks(if not allocated yet), and at least
324 * the first direct block of this branch. That's the
325 * minimum number of blocks need to allocate(required)
326 */
327 /* first we try to allocate the indirect blocks */
328 target = indirect_blks;
329 while (target > 0) {
330 count = target;
331 /* allocating blocks for indirect blocks and direct blocks */
332 current_block = ext4_new_meta_blocks(handle, inode, goal,
333 0, &count, err);
334 if (*err)
335 goto failed_out;
336
337 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
338 EXT4_ERROR_INODE(inode,
339 "current_block %llu + count %lu > %d!",
340 current_block, count,
341 EXT4_MAX_BLOCK_FILE_PHYS);
342 *err = -EIO;
343 goto failed_out;
344 }
345
346 target -= count;
347 /* allocate blocks for indirect blocks */
348 while (index < indirect_blks && count) {
349 new_blocks[index++] = current_block++;
350 count--;
351 }
352 if (count > 0) {
353 /*
354 * save the new block number
355 * for the first direct block
356 */
357 new_blocks[index] = current_block;
358 printk(KERN_INFO "%s returned more blocks than "
359 "requested\n", __func__);
360 WARN_ON(1);
361 break;
362 }
363 }
364
365 target = blks - count ;
366 blk_allocated = count;
367 if (!target)
368 goto allocated;
369 /* Now allocate data blocks */
370 memset(&ar, 0, sizeof(ar));
371 ar.inode = inode;
372 ar.goal = goal;
373 ar.len = target;
374 ar.logical = iblock;
375 if (S_ISREG(inode->i_mode))
376 /* enable in-core preallocation only for regular files */
377 ar.flags = EXT4_MB_HINT_DATA;
378
379 current_block = ext4_mb_new_blocks(handle, &ar, err);
380 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
381 EXT4_ERROR_INODE(inode,
382 "current_block %llu + ar.len %d > %d!",
383 current_block, ar.len,
384 EXT4_MAX_BLOCK_FILE_PHYS);
385 *err = -EIO;
386 goto failed_out;
387 }
388
389 if (*err && (target == blks)) {
390 /*
391 * if the allocation failed and we didn't allocate
392 * any blocks before
393 */
394 goto failed_out;
395 }
396 if (!*err) {
397 if (target == blks) {
398 /*
399 * save the new block number
400 * for the first direct block
401 */
402 new_blocks[index] = current_block;
403 }
404 blk_allocated += ar.len;
405 }
406allocated:
407 /* total number of blocks allocated for direct blocks */
408 ret = blk_allocated;
409 *err = 0;
410 return ret;
411failed_out:
412 for (i = 0; i < index; i++)
413 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
414 return ret;
415}
416
417/**
418 * ext4_alloc_branch - allocate and set up a chain of blocks.
419 * @handle: handle for this transaction
420 * @inode: owner
421 * @indirect_blks: number of allocated indirect blocks
422 * @blks: number of allocated direct blocks
423 * @goal: preferred place for allocation
424 * @offsets: offsets (in the blocks) to store the pointers to next.
425 * @branch: place to store the chain in.
426 *
427 * This function allocates blocks, zeroes out all but the last one,
428 * links them into chain and (if we are synchronous) writes them to disk.
429 * In other words, it prepares a branch that can be spliced onto the
430 * inode. It stores the information about that chain in the branch[], in
431 * the same format as ext4_get_branch() would do. We are calling it after
432 * we had read the existing part of chain and partial points to the last
433 * triple of that (one with zero ->key). Upon the exit we have the same
434 * picture as after the successful ext4_get_block(), except that in one
435 * place chain is disconnected - *branch->p is still zero (we did not
436 * set the last link), but branch->key contains the number that should
437 * be placed into *branch->p to fill that gap.
438 *
439 * If allocation fails we free all blocks we've allocated (and forget
440 * their buffer_heads) and return the error value the from failed
441 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
442 * as described above and return 0.
443 */
444static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
445 ext4_lblk_t iblock, int indirect_blks,
446 int *blks, ext4_fsblk_t goal,
447 ext4_lblk_t *offsets, Indirect *branch)
448{
449 int blocksize = inode->i_sb->s_blocksize;
450 int i, n = 0;
451 int err = 0;
452 struct buffer_head *bh;
453 int num;
454 ext4_fsblk_t new_blocks[4];
455 ext4_fsblk_t current_block;
456
457 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
458 *blks, new_blocks, &err);
459 if (err)
460 return err;
461
462 branch[0].key = cpu_to_le32(new_blocks[0]);
463 /*
464 * metadata blocks and data blocks are allocated.
465 */
466 for (n = 1; n <= indirect_blks; n++) {
467 /*
468 * Get buffer_head for parent block, zero it out
469 * and set the pointer to new one, then send
470 * parent to disk.
471 */
472 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
473 if (unlikely(!bh)) {
474 err = -EIO;
475 goto failed;
476 }
477
478 branch[n].bh = bh;
479 lock_buffer(bh);
480 BUFFER_TRACE(bh, "call get_create_access");
481 err = ext4_journal_get_create_access(handle, bh);
482 if (err) {
483 /* Don't brelse(bh) here; it's done in
484 * ext4_journal_forget() below */
485 unlock_buffer(bh);
486 goto failed;
487 }
488
489 memset(bh->b_data, 0, blocksize);
490 branch[n].p = (__le32 *) bh->b_data + offsets[n];
491 branch[n].key = cpu_to_le32(new_blocks[n]);
492 *branch[n].p = branch[n].key;
493 if (n == indirect_blks) {
494 current_block = new_blocks[n];
495 /*
496 * End of chain, update the last new metablock of
497 * the chain to point to the new allocated
498 * data blocks numbers
499 */
500 for (i = 1; i < num; i++)
501 *(branch[n].p + i) = cpu_to_le32(++current_block);
502 }
503 BUFFER_TRACE(bh, "marking uptodate");
504 set_buffer_uptodate(bh);
505 unlock_buffer(bh);
506
507 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
508 err = ext4_handle_dirty_metadata(handle, inode, bh);
509 if (err)
510 goto failed;
511 }
512 *blks = num;
513 return err;
514failed:
515 /* Allocation failed, free what we already allocated */
516 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
517 for (i = 1; i <= n ; i++) {
518 /*
519 * branch[i].bh is newly allocated, so there is no
520 * need to revoke the block, which is why we don't
521 * need to set EXT4_FREE_BLOCKS_METADATA.
522 */
523 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
524 EXT4_FREE_BLOCKS_FORGET);
525 }
526 for (i = n+1; i < indirect_blks; i++)
527 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
528
529 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
530
531 return err;
532}
533
534/**
535 * ext4_splice_branch - splice the allocated branch onto inode.
536 * @handle: handle for this transaction
537 * @inode: owner
538 * @block: (logical) number of block we are adding
539 * @chain: chain of indirect blocks (with a missing link - see
540 * ext4_alloc_branch)
541 * @where: location of missing link
542 * @num: number of indirect blocks we are adding
543 * @blks: number of direct blocks we are adding
544 *
545 * This function fills the missing link and does all housekeeping needed in
546 * inode (->i_blocks, etc.). In case of success we end up with the full
547 * chain to new block and return 0.
548 */
549static int ext4_splice_branch(handle_t *handle, struct inode *inode,
550 ext4_lblk_t block, Indirect *where, int num,
551 int blks)
552{
553 int i;
554 int err = 0;
555 ext4_fsblk_t current_block;
556
557 /*
558 * If we're splicing into a [td]indirect block (as opposed to the
559 * inode) then we need to get write access to the [td]indirect block
560 * before the splice.
561 */
562 if (where->bh) {
563 BUFFER_TRACE(where->bh, "get_write_access");
564 err = ext4_journal_get_write_access(handle, where->bh);
565 if (err)
566 goto err_out;
567 }
568 /* That's it */
569
570 *where->p = where->key;
571
572 /*
573 * Update the host buffer_head or inode to point to more just allocated
574 * direct blocks blocks
575 */
576 if (num == 0 && blks > 1) {
577 current_block = le32_to_cpu(where->key) + 1;
578 for (i = 1; i < blks; i++)
579 *(where->p + i) = cpu_to_le32(current_block++);
580 }
581
582 /* We are done with atomic stuff, now do the rest of housekeeping */
583 /* had we spliced it onto indirect block? */
584 if (where->bh) {
585 /*
586 * If we spliced it onto an indirect block, we haven't
587 * altered the inode. Note however that if it is being spliced
588 * onto an indirect block at the very end of the file (the
589 * file is growing) then we *will* alter the inode to reflect
590 * the new i_size. But that is not done here - it is done in
591 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
592 */
593 jbd_debug(5, "splicing indirect only\n");
594 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
595 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
596 if (err)
597 goto err_out;
598 } else {
599 /*
600 * OK, we spliced it into the inode itself on a direct block.
601 */
602 ext4_mark_inode_dirty(handle, inode);
603 jbd_debug(5, "splicing direct\n");
604 }
605 return err;
606
607err_out:
608 for (i = 1; i <= num; i++) {
609 /*
610 * branch[i].bh is newly allocated, so there is no
611 * need to revoke the block, which is why we don't
612 * need to set EXT4_FREE_BLOCKS_METADATA.
613 */
614 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
615 EXT4_FREE_BLOCKS_FORGET);
616 }
617 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
618 blks, 0);
619
620 return err;
621}
622
623/*
624 * The ext4_ind_map_blocks() function handles non-extents inodes
625 * (i.e., using the traditional indirect/double-indirect i_blocks
626 * scheme) for ext4_map_blocks().
627 *
628 * Allocation strategy is simple: if we have to allocate something, we will
629 * have to go the whole way to leaf. So let's do it before attaching anything
630 * to tree, set linkage between the newborn blocks, write them if sync is
631 * required, recheck the path, free and repeat if check fails, otherwise
632 * set the last missing link (that will protect us from any truncate-generated
633 * removals - all blocks on the path are immune now) and possibly force the
634 * write on the parent block.
635 * That has a nice additional property: no special recovery from the failed
636 * allocations is needed - we simply release blocks and do not touch anything
637 * reachable from inode.
638 *
639 * `handle' can be NULL if create == 0.
640 *
641 * return > 0, # of blocks mapped or allocated.
642 * return = 0, if plain lookup failed.
643 * return < 0, error case.
644 *
645 * The ext4_ind_get_blocks() function should be called with
646 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
647 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
648 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
649 * blocks.
650 */
651int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
652 struct ext4_map_blocks *map,
653 int flags)
654{
655 int err = -EIO;
656 ext4_lblk_t offsets[4];
657 Indirect chain[4];
658 Indirect *partial;
659 ext4_fsblk_t goal;
660 int indirect_blks;
661 int blocks_to_boundary = 0;
662 int depth;
663 int count = 0;
664 ext4_fsblk_t first_block = 0;
665
666 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
667 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
668 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
669 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
670 &blocks_to_boundary);
671
672 if (depth == 0)
673 goto out;
674
675 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
676
677 /* Simplest case - block found, no allocation needed */
678 if (!partial) {
679 first_block = le32_to_cpu(chain[depth - 1].key);
680 count++;
681 /*map more blocks*/
682 while (count < map->m_len && count <= blocks_to_boundary) {
683 ext4_fsblk_t blk;
684
685 blk = le32_to_cpu(*(chain[depth-1].p + count));
686
687 if (blk == first_block + count)
688 count++;
689 else
690 break;
691 }
692 goto got_it;
693 }
694
695 /* Next simple case - plain lookup or failed read of indirect block */
696 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
697 goto cleanup;
698
699 /*
700 * Okay, we need to do block allocation.
701 */
702 goal = ext4_find_goal(inode, map->m_lblk, partial);
703
704 /* the number of blocks need to allocate for [d,t]indirect blocks */
705 indirect_blks = (chain + depth) - partial - 1;
706
707 /*
708 * Next look up the indirect map to count the totoal number of
709 * direct blocks to allocate for this branch.
710 */
711 count = ext4_blks_to_allocate(partial, indirect_blks,
712 map->m_len, blocks_to_boundary);
713 /*
714 * Block out ext4_truncate while we alter the tree
715 */
716 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
717 &count, goal,
718 offsets + (partial - chain), partial);
719
720 /*
721 * The ext4_splice_branch call will free and forget any buffers
722 * on the new chain if there is a failure, but that risks using
723 * up transaction credits, especially for bitmaps where the
724 * credits cannot be returned. Can we handle this somehow? We
725 * may need to return -EAGAIN upwards in the worst case. --sct
726 */
727 if (!err)
728 err = ext4_splice_branch(handle, inode, map->m_lblk,
729 partial, indirect_blks, count);
730 if (err)
731 goto cleanup;
732
733 map->m_flags |= EXT4_MAP_NEW;
734
735 ext4_update_inode_fsync_trans(handle, inode, 1);
736got_it:
737 map->m_flags |= EXT4_MAP_MAPPED;
738 map->m_pblk = le32_to_cpu(chain[depth-1].key);
739 map->m_len = count;
740 if (count > blocks_to_boundary)
741 map->m_flags |= EXT4_MAP_BOUNDARY;
742 err = count;
743 /* Clean up and exit */
744 partial = chain + depth - 1; /* the whole chain */
745cleanup:
746 while (partial > chain) {
747 BUFFER_TRACE(partial->bh, "call brelse");
748 brelse(partial->bh);
749 partial--;
750 }
751out:
752 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
753 map->m_pblk, map->m_len, err);
754 return err;
755}
756
757/*
758 * O_DIRECT for ext3 (or indirect map) based files
759 *
760 * If the O_DIRECT write will extend the file then add this inode to the
761 * orphan list. So recovery will truncate it back to the original size
762 * if the machine crashes during the write.
763 *
764 * If the O_DIRECT write is intantiating holes inside i_size and the machine
765 * crashes then stale disk data _may_ be exposed inside the file. But current
766 * VFS code falls back into buffered path in that case so we are safe.
767 */
768ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
769 const struct iovec *iov, loff_t offset,
770 unsigned long nr_segs)
771{
772 struct file *file = iocb->ki_filp;
773 struct inode *inode = file->f_mapping->host;
774 struct ext4_inode_info *ei = EXT4_I(inode);
775 handle_t *handle;
776 ssize_t ret;
777 int orphan = 0;
778 size_t count = iov_length(iov, nr_segs);
779 int retries = 0;
780
781 if (rw == WRITE) {
782 loff_t final_size = offset + count;
783
784 if (final_size > inode->i_size) {
785 /* Credits for sb + inode write */
786 handle = ext4_journal_start(inode, 2);
787 if (IS_ERR(handle)) {
788 ret = PTR_ERR(handle);
789 goto out;
790 }
791 ret = ext4_orphan_add(handle, inode);
792 if (ret) {
793 ext4_journal_stop(handle);
794 goto out;
795 }
796 orphan = 1;
797 ei->i_disksize = inode->i_size;
798 ext4_journal_stop(handle);
799 }
800 }
801
802retry:
803 if (rw == READ && ext4_should_dioread_nolock(inode))
804 ret = __blockdev_direct_IO(rw, iocb, inode,
805 inode->i_sb->s_bdev, iov,
806 offset, nr_segs,
807 ext4_get_block, NULL, NULL, 0);
808 else {
809 ret = blockdev_direct_IO(rw, iocb, inode, iov,
810 offset, nr_segs, ext4_get_block);
811
812 if (unlikely((rw & WRITE) && ret < 0)) {
813 loff_t isize = i_size_read(inode);
814 loff_t end = offset + iov_length(iov, nr_segs);
815
816 if (end > isize)
817 ext4_truncate_failed_write(inode);
818 }
819 }
820 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
821 goto retry;
822
823 if (orphan) {
824 int err;
825
826 /* Credits for sb + inode write */
827 handle = ext4_journal_start(inode, 2);
828 if (IS_ERR(handle)) {
829 /* This is really bad luck. We've written the data
830 * but cannot extend i_size. Bail out and pretend
831 * the write failed... */
832 ret = PTR_ERR(handle);
833 if (inode->i_nlink)
834 ext4_orphan_del(NULL, inode);
835
836 goto out;
837 }
838 if (inode->i_nlink)
839 ext4_orphan_del(handle, inode);
840 if (ret > 0) {
841 loff_t end = offset + ret;
842 if (end > inode->i_size) {
843 ei->i_disksize = end;
844 i_size_write(inode, end);
845 /*
846 * We're going to return a positive `ret'
847 * here due to non-zero-length I/O, so there's
848 * no way of reporting error returns from
849 * ext4_mark_inode_dirty() to userspace. So
850 * ignore it.
851 */
852 ext4_mark_inode_dirty(handle, inode);
853 }
854 }
855 err = ext4_journal_stop(handle);
856 if (ret == 0)
857 ret = err;
858 }
859out:
860 return ret;
861}
862
863/*
864 * Calculate the number of metadata blocks need to reserve
865 * to allocate a new block at @lblocks for non extent file based file
866 */
867int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
868{
869 struct ext4_inode_info *ei = EXT4_I(inode);
870 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
871 int blk_bits;
872
873 if (lblock < EXT4_NDIR_BLOCKS)
874 return 0;
875
876 lblock -= EXT4_NDIR_BLOCKS;
877
878 if (ei->i_da_metadata_calc_len &&
879 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
880 ei->i_da_metadata_calc_len++;
881 return 0;
882 }
883 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
884 ei->i_da_metadata_calc_len = 1;
885 blk_bits = order_base_2(lblock);
886 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
887}
888
889int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
890{
891 int indirects;
892
893 /* if nrblocks are contiguous */
894 if (chunk) {
895 /*
896 * With N contiguous data blocks, we need at most
897 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
898 * 2 dindirect blocks, and 1 tindirect block
899 */
900 return DIV_ROUND_UP(nrblocks,
901 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
902 }
903 /*
904 * if nrblocks are not contiguous, worse case, each block touch
905 * a indirect block, and each indirect block touch a double indirect
906 * block, plus a triple indirect block
907 */
908 indirects = nrblocks * 2 + 1;
909 return indirects;
910}
911
912/*
913 * Truncate transactions can be complex and absolutely huge. So we need to
914 * be able to restart the transaction at a conventient checkpoint to make
915 * sure we don't overflow the journal.
916 *
917 * start_transaction gets us a new handle for a truncate transaction,
918 * and extend_transaction tries to extend the existing one a bit. If
919 * extend fails, we need to propagate the failure up and restart the
920 * transaction in the top-level truncate loop. --sct
921 */
922static handle_t *start_transaction(struct inode *inode)
923{
924 handle_t *result;
925
926 result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode));
927 if (!IS_ERR(result))
928 return result;
929
930 ext4_std_error(inode->i_sb, PTR_ERR(result));
931 return result;
932}
933
934/*
935 * Try to extend this transaction for the purposes of truncation.
936 *
937 * Returns 0 if we managed to create more room. If we can't create more
938 * room, and the transaction must be restarted we return 1.
939 */
940static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
941{
942 if (!ext4_handle_valid(handle))
943 return 0;
944 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
945 return 0;
946 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
947 return 0;
948 return 1;
949}
950
951/*
952 * Probably it should be a library function... search for first non-zero word
953 * or memcmp with zero_page, whatever is better for particular architecture.
954 * Linus?
955 */
956static inline int all_zeroes(__le32 *p, __le32 *q)
957{
958 while (p < q)
959 if (*p++)
960 return 0;
961 return 1;
962}
963
964/**
965 * ext4_find_shared - find the indirect blocks for partial truncation.
966 * @inode: inode in question
967 * @depth: depth of the affected branch
968 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
969 * @chain: place to store the pointers to partial indirect blocks
970 * @top: place to the (detached) top of branch
971 *
972 * This is a helper function used by ext4_truncate().
973 *
974 * When we do truncate() we may have to clean the ends of several
975 * indirect blocks but leave the blocks themselves alive. Block is
976 * partially truncated if some data below the new i_size is referred
977 * from it (and it is on the path to the first completely truncated
978 * data block, indeed). We have to free the top of that path along
979 * with everything to the right of the path. Since no allocation
980 * past the truncation point is possible until ext4_truncate()
981 * finishes, we may safely do the latter, but top of branch may
982 * require special attention - pageout below the truncation point
983 * might try to populate it.
984 *
985 * We atomically detach the top of branch from the tree, store the
986 * block number of its root in *@top, pointers to buffer_heads of
987 * partially truncated blocks - in @chain[].bh and pointers to
988 * their last elements that should not be removed - in
989 * @chain[].p. Return value is the pointer to last filled element
990 * of @chain.
991 *
992 * The work left to caller to do the actual freeing of subtrees:
993 * a) free the subtree starting from *@top
994 * b) free the subtrees whose roots are stored in
995 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
996 * c) free the subtrees growing from the inode past the @chain[0].
997 * (no partially truncated stuff there). */
998
999static Indirect *ext4_find_shared(struct inode *inode, int depth,
1000 ext4_lblk_t offsets[4], Indirect chain[4],
1001 __le32 *top)
1002{
1003 Indirect *partial, *p;
1004 int k, err;
1005
1006 *top = 0;
1007 /* Make k index the deepest non-null offset + 1 */
1008 for (k = depth; k > 1 && !offsets[k-1]; k--)
1009 ;
1010 partial = ext4_get_branch(inode, k, offsets, chain, &err);
1011 /* Writer: pointers */
1012 if (!partial)
1013 partial = chain + k-1;
1014 /*
1015 * If the branch acquired continuation since we've looked at it -
1016 * fine, it should all survive and (new) top doesn't belong to us.
1017 */
1018 if (!partial->key && *partial->p)
1019 /* Writer: end */
1020 goto no_top;
1021 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
1022 ;
1023 /*
1024 * OK, we've found the last block that must survive. The rest of our
1025 * branch should be detached before unlocking. However, if that rest
1026 * of branch is all ours and does not grow immediately from the inode
1027 * it's easier to cheat and just decrement partial->p.
1028 */
1029 if (p == chain + k - 1 && p > chain) {
1030 p->p--;
1031 } else {
1032 *top = *p->p;
1033 /* Nope, don't do this in ext4. Must leave the tree intact */
1034#if 0
1035 *p->p = 0;
1036#endif
1037 }
1038 /* Writer: end */
1039
1040 while (partial > p) {
1041 brelse(partial->bh);
1042 partial--;
1043 }
1044no_top:
1045 return partial;
1046}
1047
1048/*
1049 * Zero a number of block pointers in either an inode or an indirect block.
1050 * If we restart the transaction we must again get write access to the
1051 * indirect block for further modification.
1052 *
1053 * We release `count' blocks on disk, but (last - first) may be greater
1054 * than `count' because there can be holes in there.
1055 *
1056 * Return 0 on success, 1 on invalid block range
1057 * and < 0 on fatal error.
1058 */
1059static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
1060 struct buffer_head *bh,
1061 ext4_fsblk_t block_to_free,
1062 unsigned long count, __le32 *first,
1063 __le32 *last)
1064{
1065 __le32 *p;
1066 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
1067 int err;
1068
1069 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1070 flags |= EXT4_FREE_BLOCKS_METADATA;
1071
1072 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
1073 count)) {
1074 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
1075 "blocks %llu len %lu",
1076 (unsigned long long) block_to_free, count);
1077 return 1;
1078 }
1079
1080 if (try_to_extend_transaction(handle, inode)) {
1081 if (bh) {
1082 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1083 err = ext4_handle_dirty_metadata(handle, inode, bh);
1084 if (unlikely(err))
1085 goto out_err;
1086 }
1087 err = ext4_mark_inode_dirty(handle, inode);
1088 if (unlikely(err))
1089 goto out_err;
1090 err = ext4_truncate_restart_trans(handle, inode,
1091 ext4_blocks_for_truncate(inode));
1092 if (unlikely(err))
1093 goto out_err;
1094 if (bh) {
1095 BUFFER_TRACE(bh, "retaking write access");
1096 err = ext4_journal_get_write_access(handle, bh);
1097 if (unlikely(err))
1098 goto out_err;
1099 }
1100 }
1101
1102 for (p = first; p < last; p++)
1103 *p = 0;
1104
1105 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
1106 return 0;
1107out_err:
1108 ext4_std_error(inode->i_sb, err);
1109 return err;
1110}
1111
1112/**
1113 * ext4_free_data - free a list of data blocks
1114 * @handle: handle for this transaction
1115 * @inode: inode we are dealing with
1116 * @this_bh: indirect buffer_head which contains *@first and *@last
1117 * @first: array of block numbers
1118 * @last: points immediately past the end of array
1119 *
1120 * We are freeing all blocks referred from that array (numbers are stored as
1121 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1122 *
1123 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1124 * blocks are contiguous then releasing them at one time will only affect one
1125 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1126 * actually use a lot of journal space.
1127 *
1128 * @this_bh will be %NULL if @first and @last point into the inode's direct
1129 * block pointers.
1130 */
1131static void ext4_free_data(handle_t *handle, struct inode *inode,
1132 struct buffer_head *this_bh,
1133 __le32 *first, __le32 *last)
1134{
1135 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
1136 unsigned long count = 0; /* Number of blocks in the run */
1137 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1138 corresponding to
1139 block_to_free */
1140 ext4_fsblk_t nr; /* Current block # */
1141 __le32 *p; /* Pointer into inode/ind
1142 for current block */
1143 int err = 0;
1144
1145 if (this_bh) { /* For indirect block */
1146 BUFFER_TRACE(this_bh, "get_write_access");
1147 err = ext4_journal_get_write_access(handle, this_bh);
1148 /* Important: if we can't update the indirect pointers
1149 * to the blocks, we can't free them. */
1150 if (err)
1151 return;
1152 }
1153
1154 for (p = first; p < last; p++) {
1155 nr = le32_to_cpu(*p);
1156 if (nr) {
1157 /* accumulate blocks to free if they're contiguous */
1158 if (count == 0) {
1159 block_to_free = nr;
1160 block_to_free_p = p;
1161 count = 1;
1162 } else if (nr == block_to_free + count) {
1163 count++;
1164 } else {
1165 err = ext4_clear_blocks(handle, inode, this_bh,
1166 block_to_free, count,
1167 block_to_free_p, p);
1168 if (err)
1169 break;
1170 block_to_free = nr;
1171 block_to_free_p = p;
1172 count = 1;
1173 }
1174 }
1175 }
1176
1177 if (!err && count > 0)
1178 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1179 count, block_to_free_p, p);
1180 if (err < 0)
1181 /* fatal error */
1182 return;
1183
1184 if (this_bh) {
1185 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1186
1187 /*
1188 * The buffer head should have an attached journal head at this
1189 * point. However, if the data is corrupted and an indirect
1190 * block pointed to itself, it would have been detached when
1191 * the block was cleared. Check for this instead of OOPSing.
1192 */
1193 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1194 ext4_handle_dirty_metadata(handle, inode, this_bh);
1195 else
1196 EXT4_ERROR_INODE(inode,
1197 "circular indirect block detected at "
1198 "block %llu",
1199 (unsigned long long) this_bh->b_blocknr);
1200 }
1201}
1202
1203/**
1204 * ext4_free_branches - free an array of branches
1205 * @handle: JBD handle for this transaction
1206 * @inode: inode we are dealing with
1207 * @parent_bh: the buffer_head which contains *@first and *@last
1208 * @first: array of block numbers
1209 * @last: pointer immediately past the end of array
1210 * @depth: depth of the branches to free
1211 *
1212 * We are freeing all blocks referred from these branches (numbers are
1213 * stored as little-endian 32-bit) and updating @inode->i_blocks
1214 * appropriately.
1215 */
1216static void ext4_free_branches(handle_t *handle, struct inode *inode,
1217 struct buffer_head *parent_bh,
1218 __le32 *first, __le32 *last, int depth)
1219{
1220 ext4_fsblk_t nr;
1221 __le32 *p;
1222
1223 if (ext4_handle_is_aborted(handle))
1224 return;
1225
1226 if (depth--) {
1227 struct buffer_head *bh;
1228 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1229 p = last;
1230 while (--p >= first) {
1231 nr = le32_to_cpu(*p);
1232 if (!nr)
1233 continue; /* A hole */
1234
1235 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1236 nr, 1)) {
1237 EXT4_ERROR_INODE(inode,
1238 "invalid indirect mapped "
1239 "block %lu (level %d)",
1240 (unsigned long) nr, depth);
1241 break;
1242 }
1243
1244 /* Go read the buffer for the next level down */
1245 bh = sb_bread(inode->i_sb, nr);
1246
1247 /*
1248 * A read failure? Report error and clear slot
1249 * (should be rare).
1250 */
1251 if (!bh) {
1252 EXT4_ERROR_INODE_BLOCK(inode, nr,
1253 "Read failure");
1254 continue;
1255 }
1256
1257 /* This zaps the entire block. Bottom up. */
1258 BUFFER_TRACE(bh, "free child branches");
1259 ext4_free_branches(handle, inode, bh,
1260 (__le32 *) bh->b_data,
1261 (__le32 *) bh->b_data + addr_per_block,
1262 depth);
1263 brelse(bh);
1264
1265 /*
1266 * Everything below this this pointer has been
1267 * released. Now let this top-of-subtree go.
1268 *
1269 * We want the freeing of this indirect block to be
1270 * atomic in the journal with the updating of the
1271 * bitmap block which owns it. So make some room in
1272 * the journal.
1273 *
1274 * We zero the parent pointer *after* freeing its
1275 * pointee in the bitmaps, so if extend_transaction()
1276 * for some reason fails to put the bitmap changes and
1277 * the release into the same transaction, recovery
1278 * will merely complain about releasing a free block,
1279 * rather than leaking blocks.
1280 */
1281 if (ext4_handle_is_aborted(handle))
1282 return;
1283 if (try_to_extend_transaction(handle, inode)) {
1284 ext4_mark_inode_dirty(handle, inode);
1285 ext4_truncate_restart_trans(handle, inode,
1286 ext4_blocks_for_truncate(inode));
1287 }
1288
1289 /*
1290 * The forget flag here is critical because if
1291 * we are journaling (and not doing data
1292 * journaling), we have to make sure a revoke
1293 * record is written to prevent the journal
1294 * replay from overwriting the (former)
1295 * indirect block if it gets reallocated as a
1296 * data block. This must happen in the same
1297 * transaction where the data blocks are
1298 * actually freed.
1299 */
1300 ext4_free_blocks(handle, inode, NULL, nr, 1,
1301 EXT4_FREE_BLOCKS_METADATA|
1302 EXT4_FREE_BLOCKS_FORGET);
1303
1304 if (parent_bh) {
1305 /*
1306 * The block which we have just freed is
1307 * pointed to by an indirect block: journal it
1308 */
1309 BUFFER_TRACE(parent_bh, "get_write_access");
1310 if (!ext4_journal_get_write_access(handle,
1311 parent_bh)){
1312 *p = 0;
1313 BUFFER_TRACE(parent_bh,
1314 "call ext4_handle_dirty_metadata");
1315 ext4_handle_dirty_metadata(handle,
1316 inode,
1317 parent_bh);
1318 }
1319 }
1320 }
1321 } else {
1322 /* We have reached the bottom of the tree. */
1323 BUFFER_TRACE(parent_bh, "free data blocks");
1324 ext4_free_data(handle, inode, parent_bh, first, last);
1325 }
1326}
1327
1328void ext4_ind_truncate(struct inode *inode)
1329{
1330 handle_t *handle;
1331 struct ext4_inode_info *ei = EXT4_I(inode);
1332 __le32 *i_data = ei->i_data;
1333 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1334 struct address_space *mapping = inode->i_mapping;
1335 ext4_lblk_t offsets[4];
1336 Indirect chain[4];
1337 Indirect *partial;
1338 __le32 nr = 0;
1339 int n = 0;
1340 ext4_lblk_t last_block, max_block;
1341 unsigned blocksize = inode->i_sb->s_blocksize;
1342
1343 handle = start_transaction(inode);
1344 if (IS_ERR(handle))
1345 return; /* AKPM: return what? */
1346
1347 last_block = (inode->i_size + blocksize-1)
1348 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1349 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1350 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1351
1352 if (inode->i_size & (blocksize - 1))
1353 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
1354 goto out_stop;
1355
1356 if (last_block != max_block) {
1357 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1358 if (n == 0)
1359 goto out_stop; /* error */
1360 }
1361
1362 /*
1363 * OK. This truncate is going to happen. We add the inode to the
1364 * orphan list, so that if this truncate spans multiple transactions,
1365 * and we crash, we will resume the truncate when the filesystem
1366 * recovers. It also marks the inode dirty, to catch the new size.
1367 *
1368 * Implication: the file must always be in a sane, consistent
1369 * truncatable state while each transaction commits.
1370 */
1371 if (ext4_orphan_add(handle, inode))
1372 goto out_stop;
1373
1374 /*
1375 * From here we block out all ext4_get_block() callers who want to
1376 * modify the block allocation tree.
1377 */
1378 down_write(&ei->i_data_sem);
1379
1380 ext4_discard_preallocations(inode);
1381
1382 /*
1383 * The orphan list entry will now protect us from any crash which
1384 * occurs before the truncate completes, so it is now safe to propagate
1385 * the new, shorter inode size (held for now in i_size) into the
1386 * on-disk inode. We do this via i_disksize, which is the value which
1387 * ext4 *really* writes onto the disk inode.
1388 */
1389 ei->i_disksize = inode->i_size;
1390
1391 if (last_block == max_block) {
1392 /*
1393 * It is unnecessary to free any data blocks if last_block is
1394 * equal to the indirect block limit.
1395 */
1396 goto out_unlock;
1397 } else if (n == 1) { /* direct blocks */
1398 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1399 i_data + EXT4_NDIR_BLOCKS);
1400 goto do_indirects;
1401 }
1402
1403 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1404 /* Kill the top of shared branch (not detached) */
1405 if (nr) {
1406 if (partial == chain) {
1407 /* Shared branch grows from the inode */
1408 ext4_free_branches(handle, inode, NULL,
1409 &nr, &nr+1, (chain+n-1) - partial);
1410 *partial->p = 0;
1411 /*
1412 * We mark the inode dirty prior to restart,
1413 * and prior to stop. No need for it here.
1414 */
1415 } else {
1416 /* Shared branch grows from an indirect block */
1417 BUFFER_TRACE(partial->bh, "get_write_access");
1418 ext4_free_branches(handle, inode, partial->bh,
1419 partial->p,
1420 partial->p+1, (chain+n-1) - partial);
1421 }
1422 }
1423 /* Clear the ends of indirect blocks on the shared branch */
1424 while (partial > chain) {
1425 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1426 (__le32*)partial->bh->b_data+addr_per_block,
1427 (chain+n-1) - partial);
1428 BUFFER_TRACE(partial->bh, "call brelse");
1429 brelse(partial->bh);
1430 partial--;
1431 }
1432do_indirects:
1433 /* Kill the remaining (whole) subtrees */
1434 switch (offsets[0]) {
1435 default:
1436 nr = i_data[EXT4_IND_BLOCK];
1437 if (nr) {
1438 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1439 i_data[EXT4_IND_BLOCK] = 0;
1440 }
1441 case EXT4_IND_BLOCK:
1442 nr = i_data[EXT4_DIND_BLOCK];
1443 if (nr) {
1444 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1445 i_data[EXT4_DIND_BLOCK] = 0;
1446 }
1447 case EXT4_DIND_BLOCK:
1448 nr = i_data[EXT4_TIND_BLOCK];
1449 if (nr) {
1450 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1451 i_data[EXT4_TIND_BLOCK] = 0;
1452 }
1453 case EXT4_TIND_BLOCK:
1454 ;
1455 }
1456
1457out_unlock:
1458 up_write(&ei->i_data_sem);
1459 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1460 ext4_mark_inode_dirty(handle, inode);
1461
1462 /*
1463 * In a multi-transaction truncate, we only make the final transaction
1464 * synchronous
1465 */
1466 if (IS_SYNC(inode))
1467 ext4_handle_sync(handle);
1468out_stop:
1469 /*
1470 * If this was a simple ftruncate(), and the file will remain alive
1471 * then we need to clear up the orphan record which we created above.
1472 * However, if this was a real unlink then we were called by
1473 * ext4_delete_inode(), and we allow that function to clean up the
1474 * orphan info for us.
1475 */
1476 if (inode->i_nlink)
1477 ext4_orphan_del(handle, inode);
1478
1479 ext4_journal_stop(handle);
1480 trace_ext4_truncate_exit(inode);
1481}
1482
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 3e5191f9f39..d47264cafee 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -12,10 +12,6 @@
12 * 12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds 13 * Copyright (C) 1991, 1992 Linus Torvalds
14 * 14 *
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek 15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz) 16 * (jj@sunsite.ms.mff.cuni.cz)
21 * 17 *
@@ -47,6 +43,7 @@
47#include "xattr.h" 43#include "xattr.h"
48#include "acl.h" 44#include "acl.h"
49#include "ext4_extents.h" 45#include "ext4_extents.h"
46#include "truncate.h"
50 47
51#include <trace/events/ext4.h> 48#include <trace/events/ext4.h>
52 49
@@ -89,72 +86,6 @@ static int ext4_inode_is_fast_symlink(struct inode *inode)
89} 86}
90 87
91/* 88/*
92 * Work out how many blocks we need to proceed with the next chunk of a
93 * truncate transaction.
94 */
95static unsigned long blocks_for_truncate(struct inode *inode)
96{
97 ext4_lblk_t needed;
98
99 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
100
101 /* Give ourselves just enough room to cope with inodes in which
102 * i_blocks is corrupt: we've seen disk corruptions in the past
103 * which resulted in random data in an inode which looked enough
104 * like a regular file for ext4 to try to delete it. Things
105 * will go a bit crazy if that happens, but at least we should
106 * try not to panic the whole kernel. */
107 if (needed < 2)
108 needed = 2;
109
110 /* But we need to bound the transaction so we don't overflow the
111 * journal. */
112 if (needed > EXT4_MAX_TRANS_DATA)
113 needed = EXT4_MAX_TRANS_DATA;
114
115 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
116}
117
118/*
119 * Truncate transactions can be complex and absolutely huge. So we need to
120 * be able to restart the transaction at a conventient checkpoint to make
121 * sure we don't overflow the journal.
122 *
123 * start_transaction gets us a new handle for a truncate transaction,
124 * and extend_transaction tries to extend the existing one a bit. If
125 * extend fails, we need to propagate the failure up and restart the
126 * transaction in the top-level truncate loop. --sct
127 */
128static handle_t *start_transaction(struct inode *inode)
129{
130 handle_t *result;
131
132 result = ext4_journal_start(inode, blocks_for_truncate(inode));
133 if (!IS_ERR(result))
134 return result;
135
136 ext4_std_error(inode->i_sb, PTR_ERR(result));
137 return result;
138}
139
140/*
141 * Try to extend this transaction for the purposes of truncation.
142 *
143 * Returns 0 if we managed to create more room. If we can't create more
144 * room, and the transaction must be restarted we return 1.
145 */
146static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
147{
148 if (!ext4_handle_valid(handle))
149 return 0;
150 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
151 return 0;
152 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
153 return 0;
154 return 1;
155}
156
157/*
158 * Restart the transaction associated with *handle. This does a commit, 89 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against 90 * so before we call here everything must be consistently dirtied against
160 * this transaction. 91 * this transaction.
@@ -190,6 +121,33 @@ void ext4_evict_inode(struct inode *inode)
190 121
191 trace_ext4_evict_inode(inode); 122 trace_ext4_evict_inode(inode);
192 if (inode->i_nlink) { 123 if (inode->i_nlink) {
124 /*
125 * When journalling data dirty buffers are tracked only in the
126 * journal. So although mm thinks everything is clean and
127 * ready for reaping the inode might still have some pages to
128 * write in the running transaction or waiting to be
129 * checkpointed. Thus calling jbd2_journal_invalidatepage()
130 * (via truncate_inode_pages()) to discard these buffers can
131 * cause data loss. Also even if we did not discard these
132 * buffers, we would have no way to find them after the inode
133 * is reaped and thus user could see stale data if he tries to
134 * read them before the transaction is checkpointed. So be
135 * careful and force everything to disk here... We use
136 * ei->i_datasync_tid to store the newest transaction
137 * containing inode's data.
138 *
139 * Note that directories do not have this problem because they
140 * don't use page cache.
141 */
142 if (ext4_should_journal_data(inode) &&
143 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
144 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
145 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
146
147 jbd2_log_start_commit(journal, commit_tid);
148 jbd2_log_wait_commit(journal, commit_tid);
149 filemap_write_and_wait(&inode->i_data);
150 }
193 truncate_inode_pages(&inode->i_data, 0); 151 truncate_inode_pages(&inode->i_data, 0);
194 goto no_delete; 152 goto no_delete;
195 } 153 }
@@ -204,7 +162,7 @@ void ext4_evict_inode(struct inode *inode)
204 if (is_bad_inode(inode)) 162 if (is_bad_inode(inode))
205 goto no_delete; 163 goto no_delete;
206 164
207 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); 165 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
208 if (IS_ERR(handle)) { 166 if (IS_ERR(handle)) {
209 ext4_std_error(inode->i_sb, PTR_ERR(handle)); 167 ext4_std_error(inode->i_sb, PTR_ERR(handle));
210 /* 168 /*
@@ -277,793 +235,6 @@ no_delete:
277 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ 235 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
278} 236}
279 237
280typedef struct {
281 __le32 *p;
282 __le32 key;
283 struct buffer_head *bh;
284} Indirect;
285
286static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
287{
288 p->key = *(p->p = v);
289 p->bh = bh;
290}
291
292/**
293 * ext4_block_to_path - parse the block number into array of offsets
294 * @inode: inode in question (we are only interested in its superblock)
295 * @i_block: block number to be parsed
296 * @offsets: array to store the offsets in
297 * @boundary: set this non-zero if the referred-to block is likely to be
298 * followed (on disk) by an indirect block.
299 *
300 * To store the locations of file's data ext4 uses a data structure common
301 * for UNIX filesystems - tree of pointers anchored in the inode, with
302 * data blocks at leaves and indirect blocks in intermediate nodes.
303 * This function translates the block number into path in that tree -
304 * return value is the path length and @offsets[n] is the offset of
305 * pointer to (n+1)th node in the nth one. If @block is out of range
306 * (negative or too large) warning is printed and zero returned.
307 *
308 * Note: function doesn't find node addresses, so no IO is needed. All
309 * we need to know is the capacity of indirect blocks (taken from the
310 * inode->i_sb).
311 */
312
313/*
314 * Portability note: the last comparison (check that we fit into triple
315 * indirect block) is spelled differently, because otherwise on an
316 * architecture with 32-bit longs and 8Kb pages we might get into trouble
317 * if our filesystem had 8Kb blocks. We might use long long, but that would
318 * kill us on x86. Oh, well, at least the sign propagation does not matter -
319 * i_block would have to be negative in the very beginning, so we would not
320 * get there at all.
321 */
322
323static int ext4_block_to_path(struct inode *inode,
324 ext4_lblk_t i_block,
325 ext4_lblk_t offsets[4], int *boundary)
326{
327 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
328 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
329 const long direct_blocks = EXT4_NDIR_BLOCKS,
330 indirect_blocks = ptrs,
331 double_blocks = (1 << (ptrs_bits * 2));
332 int n = 0;
333 int final = 0;
334
335 if (i_block < direct_blocks) {
336 offsets[n++] = i_block;
337 final = direct_blocks;
338 } else if ((i_block -= direct_blocks) < indirect_blocks) {
339 offsets[n++] = EXT4_IND_BLOCK;
340 offsets[n++] = i_block;
341 final = ptrs;
342 } else if ((i_block -= indirect_blocks) < double_blocks) {
343 offsets[n++] = EXT4_DIND_BLOCK;
344 offsets[n++] = i_block >> ptrs_bits;
345 offsets[n++] = i_block & (ptrs - 1);
346 final = ptrs;
347 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
348 offsets[n++] = EXT4_TIND_BLOCK;
349 offsets[n++] = i_block >> (ptrs_bits * 2);
350 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
351 offsets[n++] = i_block & (ptrs - 1);
352 final = ptrs;
353 } else {
354 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
355 i_block + direct_blocks +
356 indirect_blocks + double_blocks, inode->i_ino);
357 }
358 if (boundary)
359 *boundary = final - 1 - (i_block & (ptrs - 1));
360 return n;
361}
362
363static int __ext4_check_blockref(const char *function, unsigned int line,
364 struct inode *inode,
365 __le32 *p, unsigned int max)
366{
367 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
368 __le32 *bref = p;
369 unsigned int blk;
370
371 while (bref < p+max) {
372 blk = le32_to_cpu(*bref++);
373 if (blk &&
374 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
375 blk, 1))) {
376 es->s_last_error_block = cpu_to_le64(blk);
377 ext4_error_inode(inode, function, line, blk,
378 "invalid block");
379 return -EIO;
380 }
381 }
382 return 0;
383}
384
385
386#define ext4_check_indirect_blockref(inode, bh) \
387 __ext4_check_blockref(__func__, __LINE__, inode, \
388 (__le32 *)(bh)->b_data, \
389 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
390
391#define ext4_check_inode_blockref(inode) \
392 __ext4_check_blockref(__func__, __LINE__, inode, \
393 EXT4_I(inode)->i_data, \
394 EXT4_NDIR_BLOCKS)
395
396/**
397 * ext4_get_branch - read the chain of indirect blocks leading to data
398 * @inode: inode in question
399 * @depth: depth of the chain (1 - direct pointer, etc.)
400 * @offsets: offsets of pointers in inode/indirect blocks
401 * @chain: place to store the result
402 * @err: here we store the error value
403 *
404 * Function fills the array of triples <key, p, bh> and returns %NULL
405 * if everything went OK or the pointer to the last filled triple
406 * (incomplete one) otherwise. Upon the return chain[i].key contains
407 * the number of (i+1)-th block in the chain (as it is stored in memory,
408 * i.e. little-endian 32-bit), chain[i].p contains the address of that
409 * number (it points into struct inode for i==0 and into the bh->b_data
410 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
411 * block for i>0 and NULL for i==0. In other words, it holds the block
412 * numbers of the chain, addresses they were taken from (and where we can
413 * verify that chain did not change) and buffer_heads hosting these
414 * numbers.
415 *
416 * Function stops when it stumbles upon zero pointer (absent block)
417 * (pointer to last triple returned, *@err == 0)
418 * or when it gets an IO error reading an indirect block
419 * (ditto, *@err == -EIO)
420 * or when it reads all @depth-1 indirect blocks successfully and finds
421 * the whole chain, all way to the data (returns %NULL, *err == 0).
422 *
423 * Need to be called with
424 * down_read(&EXT4_I(inode)->i_data_sem)
425 */
426static Indirect *ext4_get_branch(struct inode *inode, int depth,
427 ext4_lblk_t *offsets,
428 Indirect chain[4], int *err)
429{
430 struct super_block *sb = inode->i_sb;
431 Indirect *p = chain;
432 struct buffer_head *bh;
433
434 *err = 0;
435 /* i_data is not going away, no lock needed */
436 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
437 if (!p->key)
438 goto no_block;
439 while (--depth) {
440 bh = sb_getblk(sb, le32_to_cpu(p->key));
441 if (unlikely(!bh))
442 goto failure;
443
444 if (!bh_uptodate_or_lock(bh)) {
445 if (bh_submit_read(bh) < 0) {
446 put_bh(bh);
447 goto failure;
448 }
449 /* validate block references */
450 if (ext4_check_indirect_blockref(inode, bh)) {
451 put_bh(bh);
452 goto failure;
453 }
454 }
455
456 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
457 /* Reader: end */
458 if (!p->key)
459 goto no_block;
460 }
461 return NULL;
462
463failure:
464 *err = -EIO;
465no_block:
466 return p;
467}
468
469/**
470 * ext4_find_near - find a place for allocation with sufficient locality
471 * @inode: owner
472 * @ind: descriptor of indirect block.
473 *
474 * This function returns the preferred place for block allocation.
475 * It is used when heuristic for sequential allocation fails.
476 * Rules are:
477 * + if there is a block to the left of our position - allocate near it.
478 * + if pointer will live in indirect block - allocate near that block.
479 * + if pointer will live in inode - allocate in the same
480 * cylinder group.
481 *
482 * In the latter case we colour the starting block by the callers PID to
483 * prevent it from clashing with concurrent allocations for a different inode
484 * in the same block group. The PID is used here so that functionally related
485 * files will be close-by on-disk.
486 *
487 * Caller must make sure that @ind is valid and will stay that way.
488 */
489static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
490{
491 struct ext4_inode_info *ei = EXT4_I(inode);
492 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
493 __le32 *p;
494 ext4_fsblk_t bg_start;
495 ext4_fsblk_t last_block;
496 ext4_grpblk_t colour;
497 ext4_group_t block_group;
498 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
499
500 /* Try to find previous block */
501 for (p = ind->p - 1; p >= start; p--) {
502 if (*p)
503 return le32_to_cpu(*p);
504 }
505
506 /* No such thing, so let's try location of indirect block */
507 if (ind->bh)
508 return ind->bh->b_blocknr;
509
510 /*
511 * It is going to be referred to from the inode itself? OK, just put it
512 * into the same cylinder group then.
513 */
514 block_group = ei->i_block_group;
515 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
516 block_group &= ~(flex_size-1);
517 if (S_ISREG(inode->i_mode))
518 block_group++;
519 }
520 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
521 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
522
523 /*
524 * If we are doing delayed allocation, we don't need take
525 * colour into account.
526 */
527 if (test_opt(inode->i_sb, DELALLOC))
528 return bg_start;
529
530 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
531 colour = (current->pid % 16) *
532 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
533 else
534 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
535 return bg_start + colour;
536}
537
538/**
539 * ext4_find_goal - find a preferred place for allocation.
540 * @inode: owner
541 * @block: block we want
542 * @partial: pointer to the last triple within a chain
543 *
544 * Normally this function find the preferred place for block allocation,
545 * returns it.
546 * Because this is only used for non-extent files, we limit the block nr
547 * to 32 bits.
548 */
549static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
550 Indirect *partial)
551{
552 ext4_fsblk_t goal;
553
554 /*
555 * XXX need to get goal block from mballoc's data structures
556 */
557
558 goal = ext4_find_near(inode, partial);
559 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
560 return goal;
561}
562
563/**
564 * ext4_blks_to_allocate - Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
566 *
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
571 *
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
574 */
575static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
576 int blocks_to_boundary)
577{
578 unsigned int count = 0;
579
580 /*
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
583 */
584 if (k > 0) {
585 /* right now we don't handle cross boundary allocation */
586 if (blks < blocks_to_boundary + 1)
587 count += blks;
588 else
589 count += blocks_to_boundary + 1;
590 return count;
591 }
592
593 count++;
594 while (count < blks && count <= blocks_to_boundary &&
595 le32_to_cpu(*(branch[0].p + count)) == 0) {
596 count++;
597 }
598 return count;
599}
600
601/**
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @handle: handle for this transaction
604 * @inode: inode which needs allocated blocks
605 * @iblock: the logical block to start allocated at
606 * @goal: preferred physical block of allocation
607 * @indirect_blks: the number of blocks need to allocate for indirect
608 * blocks
609 * @blks: number of desired blocks
610 * @new_blocks: on return it will store the new block numbers for
611 * the indirect blocks(if needed) and the first direct block,
612 * @err: on return it will store the error code
613 *
614 * This function will return the number of blocks allocated as
615 * requested by the passed-in parameters.
616 */
617static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
618 ext4_lblk_t iblock, ext4_fsblk_t goal,
619 int indirect_blks, int blks,
620 ext4_fsblk_t new_blocks[4], int *err)
621{
622 struct ext4_allocation_request ar;
623 int target, i;
624 unsigned long count = 0, blk_allocated = 0;
625 int index = 0;
626 ext4_fsblk_t current_block = 0;
627 int ret = 0;
628
629 /*
630 * Here we try to allocate the requested multiple blocks at once,
631 * on a best-effort basis.
632 * To build a branch, we should allocate blocks for
633 * the indirect blocks(if not allocated yet), and at least
634 * the first direct block of this branch. That's the
635 * minimum number of blocks need to allocate(required)
636 */
637 /* first we try to allocate the indirect blocks */
638 target = indirect_blks;
639 while (target > 0) {
640 count = target;
641 /* allocating blocks for indirect blocks and direct blocks */
642 current_block = ext4_new_meta_blocks(handle, inode, goal,
643 0, &count, err);
644 if (*err)
645 goto failed_out;
646
647 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
648 EXT4_ERROR_INODE(inode,
649 "current_block %llu + count %lu > %d!",
650 current_block, count,
651 EXT4_MAX_BLOCK_FILE_PHYS);
652 *err = -EIO;
653 goto failed_out;
654 }
655
656 target -= count;
657 /* allocate blocks for indirect blocks */
658 while (index < indirect_blks && count) {
659 new_blocks[index++] = current_block++;
660 count--;
661 }
662 if (count > 0) {
663 /*
664 * save the new block number
665 * for the first direct block
666 */
667 new_blocks[index] = current_block;
668 printk(KERN_INFO "%s returned more blocks than "
669 "requested\n", __func__);
670 WARN_ON(1);
671 break;
672 }
673 }
674
675 target = blks - count ;
676 blk_allocated = count;
677 if (!target)
678 goto allocated;
679 /* Now allocate data blocks */
680 memset(&ar, 0, sizeof(ar));
681 ar.inode = inode;
682 ar.goal = goal;
683 ar.len = target;
684 ar.logical = iblock;
685 if (S_ISREG(inode->i_mode))
686 /* enable in-core preallocation only for regular files */
687 ar.flags = EXT4_MB_HINT_DATA;
688
689 current_block = ext4_mb_new_blocks(handle, &ar, err);
690 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
691 EXT4_ERROR_INODE(inode,
692 "current_block %llu + ar.len %d > %d!",
693 current_block, ar.len,
694 EXT4_MAX_BLOCK_FILE_PHYS);
695 *err = -EIO;
696 goto failed_out;
697 }
698
699 if (*err && (target == blks)) {
700 /*
701 * if the allocation failed and we didn't allocate
702 * any blocks before
703 */
704 goto failed_out;
705 }
706 if (!*err) {
707 if (target == blks) {
708 /*
709 * save the new block number
710 * for the first direct block
711 */
712 new_blocks[index] = current_block;
713 }
714 blk_allocated += ar.len;
715 }
716allocated:
717 /* total number of blocks allocated for direct blocks */
718 ret = blk_allocated;
719 *err = 0;
720 return ret;
721failed_out:
722 for (i = 0; i < index; i++)
723 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
724 return ret;
725}
726
727/**
728 * ext4_alloc_branch - allocate and set up a chain of blocks.
729 * @handle: handle for this transaction
730 * @inode: owner
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @goal: preferred place for allocation
734 * @offsets: offsets (in the blocks) to store the pointers to next.
735 * @branch: place to store the chain in.
736 *
737 * This function allocates blocks, zeroes out all but the last one,
738 * links them into chain and (if we are synchronous) writes them to disk.
739 * In other words, it prepares a branch that can be spliced onto the
740 * inode. It stores the information about that chain in the branch[], in
741 * the same format as ext4_get_branch() would do. We are calling it after
742 * we had read the existing part of chain and partial points to the last
743 * triple of that (one with zero ->key). Upon the exit we have the same
744 * picture as after the successful ext4_get_block(), except that in one
745 * place chain is disconnected - *branch->p is still zero (we did not
746 * set the last link), but branch->key contains the number that should
747 * be placed into *branch->p to fill that gap.
748 *
749 * If allocation fails we free all blocks we've allocated (and forget
750 * their buffer_heads) and return the error value the from failed
751 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
752 * as described above and return 0.
753 */
754static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
755 ext4_lblk_t iblock, int indirect_blks,
756 int *blks, ext4_fsblk_t goal,
757 ext4_lblk_t *offsets, Indirect *branch)
758{
759 int blocksize = inode->i_sb->s_blocksize;
760 int i, n = 0;
761 int err = 0;
762 struct buffer_head *bh;
763 int num;
764 ext4_fsblk_t new_blocks[4];
765 ext4_fsblk_t current_block;
766
767 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
768 *blks, new_blocks, &err);
769 if (err)
770 return err;
771
772 branch[0].key = cpu_to_le32(new_blocks[0]);
773 /*
774 * metadata blocks and data blocks are allocated.
775 */
776 for (n = 1; n <= indirect_blks; n++) {
777 /*
778 * Get buffer_head for parent block, zero it out
779 * and set the pointer to new one, then send
780 * parent to disk.
781 */
782 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
783 if (unlikely(!bh)) {
784 err = -EIO;
785 goto failed;
786 }
787
788 branch[n].bh = bh;
789 lock_buffer(bh);
790 BUFFER_TRACE(bh, "call get_create_access");
791 err = ext4_journal_get_create_access(handle, bh);
792 if (err) {
793 /* Don't brelse(bh) here; it's done in
794 * ext4_journal_forget() below */
795 unlock_buffer(bh);
796 goto failed;
797 }
798
799 memset(bh->b_data, 0, blocksize);
800 branch[n].p = (__le32 *) bh->b_data + offsets[n];
801 branch[n].key = cpu_to_le32(new_blocks[n]);
802 *branch[n].p = branch[n].key;
803 if (n == indirect_blks) {
804 current_block = new_blocks[n];
805 /*
806 * End of chain, update the last new metablock of
807 * the chain to point to the new allocated
808 * data blocks numbers
809 */
810 for (i = 1; i < num; i++)
811 *(branch[n].p + i) = cpu_to_le32(++current_block);
812 }
813 BUFFER_TRACE(bh, "marking uptodate");
814 set_buffer_uptodate(bh);
815 unlock_buffer(bh);
816
817 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
818 err = ext4_handle_dirty_metadata(handle, inode, bh);
819 if (err)
820 goto failed;
821 }
822 *blks = num;
823 return err;
824failed:
825 /* Allocation failed, free what we already allocated */
826 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
827 for (i = 1; i <= n ; i++) {
828 /*
829 * branch[i].bh is newly allocated, so there is no
830 * need to revoke the block, which is why we don't
831 * need to set EXT4_FREE_BLOCKS_METADATA.
832 */
833 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
834 EXT4_FREE_BLOCKS_FORGET);
835 }
836 for (i = n+1; i < indirect_blks; i++)
837 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
838
839 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
840
841 return err;
842}
843
844/**
845 * ext4_splice_branch - splice the allocated branch onto inode.
846 * @handle: handle for this transaction
847 * @inode: owner
848 * @block: (logical) number of block we are adding
849 * @chain: chain of indirect blocks (with a missing link - see
850 * ext4_alloc_branch)
851 * @where: location of missing link
852 * @num: number of indirect blocks we are adding
853 * @blks: number of direct blocks we are adding
854 *
855 * This function fills the missing link and does all housekeeping needed in
856 * inode (->i_blocks, etc.). In case of success we end up with the full
857 * chain to new block and return 0.
858 */
859static int ext4_splice_branch(handle_t *handle, struct inode *inode,
860 ext4_lblk_t block, Indirect *where, int num,
861 int blks)
862{
863 int i;
864 int err = 0;
865 ext4_fsblk_t current_block;
866
867 /*
868 * If we're splicing into a [td]indirect block (as opposed to the
869 * inode) then we need to get write access to the [td]indirect block
870 * before the splice.
871 */
872 if (where->bh) {
873 BUFFER_TRACE(where->bh, "get_write_access");
874 err = ext4_journal_get_write_access(handle, where->bh);
875 if (err)
876 goto err_out;
877 }
878 /* That's it */
879
880 *where->p = where->key;
881
882 /*
883 * Update the host buffer_head or inode to point to more just allocated
884 * direct blocks blocks
885 */
886 if (num == 0 && blks > 1) {
887 current_block = le32_to_cpu(where->key) + 1;
888 for (i = 1; i < blks; i++)
889 *(where->p + i) = cpu_to_le32(current_block++);
890 }
891
892 /* We are done with atomic stuff, now do the rest of housekeeping */
893 /* had we spliced it onto indirect block? */
894 if (where->bh) {
895 /*
896 * If we spliced it onto an indirect block, we haven't
897 * altered the inode. Note however that if it is being spliced
898 * onto an indirect block at the very end of the file (the
899 * file is growing) then we *will* alter the inode to reflect
900 * the new i_size. But that is not done here - it is done in
901 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
902 */
903 jbd_debug(5, "splicing indirect only\n");
904 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
905 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
906 if (err)
907 goto err_out;
908 } else {
909 /*
910 * OK, we spliced it into the inode itself on a direct block.
911 */
912 ext4_mark_inode_dirty(handle, inode);
913 jbd_debug(5, "splicing direct\n");
914 }
915 return err;
916
917err_out:
918 for (i = 1; i <= num; i++) {
919 /*
920 * branch[i].bh is newly allocated, so there is no
921 * need to revoke the block, which is why we don't
922 * need to set EXT4_FREE_BLOCKS_METADATA.
923 */
924 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
925 EXT4_FREE_BLOCKS_FORGET);
926 }
927 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
928 blks, 0);
929
930 return err;
931}
932
933/*
934 * The ext4_ind_map_blocks() function handles non-extents inodes
935 * (i.e., using the traditional indirect/double-indirect i_blocks
936 * scheme) for ext4_map_blocks().
937 *
938 * Allocation strategy is simple: if we have to allocate something, we will
939 * have to go the whole way to leaf. So let's do it before attaching anything
940 * to tree, set linkage between the newborn blocks, write them if sync is
941 * required, recheck the path, free and repeat if check fails, otherwise
942 * set the last missing link (that will protect us from any truncate-generated
943 * removals - all blocks on the path are immune now) and possibly force the
944 * write on the parent block.
945 * That has a nice additional property: no special recovery from the failed
946 * allocations is needed - we simply release blocks and do not touch anything
947 * reachable from inode.
948 *
949 * `handle' can be NULL if create == 0.
950 *
951 * return > 0, # of blocks mapped or allocated.
952 * return = 0, if plain lookup failed.
953 * return < 0, error case.
954 *
955 * The ext4_ind_get_blocks() function should be called with
956 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
957 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
958 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
959 * blocks.
960 */
961static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
962 struct ext4_map_blocks *map,
963 int flags)
964{
965 int err = -EIO;
966 ext4_lblk_t offsets[4];
967 Indirect chain[4];
968 Indirect *partial;
969 ext4_fsblk_t goal;
970 int indirect_blks;
971 int blocks_to_boundary = 0;
972 int depth;
973 int count = 0;
974 ext4_fsblk_t first_block = 0;
975
976 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
977 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
978 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
979 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
980 &blocks_to_boundary);
981
982 if (depth == 0)
983 goto out;
984
985 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
986
987 /* Simplest case - block found, no allocation needed */
988 if (!partial) {
989 first_block = le32_to_cpu(chain[depth - 1].key);
990 count++;
991 /*map more blocks*/
992 while (count < map->m_len && count <= blocks_to_boundary) {
993 ext4_fsblk_t blk;
994
995 blk = le32_to_cpu(*(chain[depth-1].p + count));
996
997 if (blk == first_block + count)
998 count++;
999 else
1000 break;
1001 }
1002 goto got_it;
1003 }
1004
1005 /* Next simple case - plain lookup or failed read of indirect block */
1006 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
1007 goto cleanup;
1008
1009 /*
1010 * Okay, we need to do block allocation.
1011 */
1012 goal = ext4_find_goal(inode, map->m_lblk, partial);
1013
1014 /* the number of blocks need to allocate for [d,t]indirect blocks */
1015 indirect_blks = (chain + depth) - partial - 1;
1016
1017 /*
1018 * Next look up the indirect map to count the totoal number of
1019 * direct blocks to allocate for this branch.
1020 */
1021 count = ext4_blks_to_allocate(partial, indirect_blks,
1022 map->m_len, blocks_to_boundary);
1023 /*
1024 * Block out ext4_truncate while we alter the tree
1025 */
1026 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1027 &count, goal,
1028 offsets + (partial - chain), partial);
1029
1030 /*
1031 * The ext4_splice_branch call will free and forget any buffers
1032 * on the new chain if there is a failure, but that risks using
1033 * up transaction credits, especially for bitmaps where the
1034 * credits cannot be returned. Can we handle this somehow? We
1035 * may need to return -EAGAIN upwards in the worst case. --sct
1036 */
1037 if (!err)
1038 err = ext4_splice_branch(handle, inode, map->m_lblk,
1039 partial, indirect_blks, count);
1040 if (err)
1041 goto cleanup;
1042
1043 map->m_flags |= EXT4_MAP_NEW;
1044
1045 ext4_update_inode_fsync_trans(handle, inode, 1);
1046got_it:
1047 map->m_flags |= EXT4_MAP_MAPPED;
1048 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1049 map->m_len = count;
1050 if (count > blocks_to_boundary)
1051 map->m_flags |= EXT4_MAP_BOUNDARY;
1052 err = count;
1053 /* Clean up and exit */
1054 partial = chain + depth - 1; /* the whole chain */
1055cleanup:
1056 while (partial > chain) {
1057 BUFFER_TRACE(partial->bh, "call brelse");
1058 brelse(partial->bh);
1059 partial--;
1060 }
1061out:
1062 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
1063 map->m_pblk, map->m_len, err);
1064 return err;
1065}
1066
1067#ifdef CONFIG_QUOTA 238#ifdef CONFIG_QUOTA
1068qsize_t *ext4_get_reserved_space(struct inode *inode) 239qsize_t *ext4_get_reserved_space(struct inode *inode)
1069{ 240{
@@ -1073,33 +244,6 @@ qsize_t *ext4_get_reserved_space(struct inode *inode)
1073 244
1074/* 245/*
1075 * Calculate the number of metadata blocks need to reserve 246 * Calculate the number of metadata blocks need to reserve
1076 * to allocate a new block at @lblocks for non extent file based file
1077 */
1078static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1079 sector_t lblock)
1080{
1081 struct ext4_inode_info *ei = EXT4_I(inode);
1082 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1083 int blk_bits;
1084
1085 if (lblock < EXT4_NDIR_BLOCKS)
1086 return 0;
1087
1088 lblock -= EXT4_NDIR_BLOCKS;
1089
1090 if (ei->i_da_metadata_calc_len &&
1091 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1092 ei->i_da_metadata_calc_len++;
1093 return 0;
1094 }
1095 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1096 ei->i_da_metadata_calc_len = 1;
1097 blk_bits = order_base_2(lblock);
1098 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1099}
1100
1101/*
1102 * Calculate the number of metadata blocks need to reserve
1103 * to allocate a block located at @lblock 247 * to allocate a block located at @lblock
1104 */ 248 */
1105static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) 249static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
@@ -1107,7 +251,7 @@ static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
1107 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 251 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1108 return ext4_ext_calc_metadata_amount(inode, lblock); 252 return ext4_ext_calc_metadata_amount(inode, lblock);
1109 253
1110 return ext4_indirect_calc_metadata_amount(inode, lblock); 254 return ext4_ind_calc_metadata_amount(inode, lblock);
1111} 255}
1112 256
1113/* 257/*
@@ -1589,16 +733,6 @@ static int do_journal_get_write_access(handle_t *handle,
1589 return ret; 733 return ret;
1590} 734}
1591 735
1592/*
1593 * Truncate blocks that were not used by write. We have to truncate the
1594 * pagecache as well so that corresponding buffers get properly unmapped.
1595 */
1596static void ext4_truncate_failed_write(struct inode *inode)
1597{
1598 truncate_inode_pages(inode->i_mapping, inode->i_size);
1599 ext4_truncate(inode);
1600}
1601
1602static int ext4_get_block_write(struct inode *inode, sector_t iblock, 736static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1603 struct buffer_head *bh_result, int create); 737 struct buffer_head *bh_result, int create);
1604static int ext4_write_begin(struct file *file, struct address_space *mapping, 738static int ext4_write_begin(struct file *file, struct address_space *mapping,
@@ -1863,6 +997,7 @@ static int ext4_journalled_write_end(struct file *file,
1863 if (new_i_size > inode->i_size) 997 if (new_i_size > inode->i_size)
1864 i_size_write(inode, pos+copied); 998 i_size_write(inode, pos+copied);
1865 ext4_set_inode_state(inode, EXT4_STATE_JDATA); 999 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1000 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1866 if (new_i_size > EXT4_I(inode)->i_disksize) { 1001 if (new_i_size > EXT4_I(inode)->i_disksize) {
1867 ext4_update_i_disksize(inode, new_i_size); 1002 ext4_update_i_disksize(inode, new_i_size);
1868 ret2 = ext4_mark_inode_dirty(handle, inode); 1003 ret2 = ext4_mark_inode_dirty(handle, inode);
@@ -2571,6 +1706,7 @@ static int __ext4_journalled_writepage(struct page *page,
2571 write_end_fn); 1706 write_end_fn);
2572 if (ret == 0) 1707 if (ret == 0)
2573 ret = err; 1708 ret = err;
1709 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2574 err = ext4_journal_stop(handle); 1710 err = ext4_journal_stop(handle);
2575 if (!ret) 1711 if (!ret)
2576 ret = err; 1712 ret = err;
@@ -3450,112 +2586,6 @@ static int ext4_releasepage(struct page *page, gfp_t wait)
3450} 2586}
3451 2587
3452/* 2588/*
3453 * O_DIRECT for ext3 (or indirect map) based files
3454 *
3455 * If the O_DIRECT write will extend the file then add this inode to the
3456 * orphan list. So recovery will truncate it back to the original size
3457 * if the machine crashes during the write.
3458 *
3459 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3460 * crashes then stale disk data _may_ be exposed inside the file. But current
3461 * VFS code falls back into buffered path in that case so we are safe.
3462 */
3463static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3464 const struct iovec *iov, loff_t offset,
3465 unsigned long nr_segs)
3466{
3467 struct file *file = iocb->ki_filp;
3468 struct inode *inode = file->f_mapping->host;
3469 struct ext4_inode_info *ei = EXT4_I(inode);
3470 handle_t *handle;
3471 ssize_t ret;
3472 int orphan = 0;
3473 size_t count = iov_length(iov, nr_segs);
3474 int retries = 0;
3475
3476 if (rw == WRITE) {
3477 loff_t final_size = offset + count;
3478
3479 if (final_size > inode->i_size) {
3480 /* Credits for sb + inode write */
3481 handle = ext4_journal_start(inode, 2);
3482 if (IS_ERR(handle)) {
3483 ret = PTR_ERR(handle);
3484 goto out;
3485 }
3486 ret = ext4_orphan_add(handle, inode);
3487 if (ret) {
3488 ext4_journal_stop(handle);
3489 goto out;
3490 }
3491 orphan = 1;
3492 ei->i_disksize = inode->i_size;
3493 ext4_journal_stop(handle);
3494 }
3495 }
3496
3497retry:
3498 if (rw == READ && ext4_should_dioread_nolock(inode))
3499 ret = __blockdev_direct_IO(rw, iocb, inode,
3500 inode->i_sb->s_bdev, iov,
3501 offset, nr_segs,
3502 ext4_get_block, NULL, NULL, 0);
3503 else {
3504 ret = blockdev_direct_IO(rw, iocb, inode, iov,
3505 offset, nr_segs, ext4_get_block);
3506
3507 if (unlikely((rw & WRITE) && ret < 0)) {
3508 loff_t isize = i_size_read(inode);
3509 loff_t end = offset + iov_length(iov, nr_segs);
3510
3511 if (end > isize)
3512 ext4_truncate_failed_write(inode);
3513 }
3514 }
3515 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3516 goto retry;
3517
3518 if (orphan) {
3519 int err;
3520
3521 /* Credits for sb + inode write */
3522 handle = ext4_journal_start(inode, 2);
3523 if (IS_ERR(handle)) {
3524 /* This is really bad luck. We've written the data
3525 * but cannot extend i_size. Bail out and pretend
3526 * the write failed... */
3527 ret = PTR_ERR(handle);
3528 if (inode->i_nlink)
3529 ext4_orphan_del(NULL, inode);
3530
3531 goto out;
3532 }
3533 if (inode->i_nlink)
3534 ext4_orphan_del(handle, inode);
3535 if (ret > 0) {
3536 loff_t end = offset + ret;
3537 if (end > inode->i_size) {
3538 ei->i_disksize = end;
3539 i_size_write(inode, end);
3540 /*
3541 * We're going to return a positive `ret'
3542 * here due to non-zero-length I/O, so there's
3543 * no way of reporting error returns from
3544 * ext4_mark_inode_dirty() to userspace. So
3545 * ignore it.
3546 */
3547 ext4_mark_inode_dirty(handle, inode);
3548 }
3549 }
3550 err = ext4_journal_stop(handle);
3551 if (ret == 0)
3552 ret = err;
3553 }
3554out:
3555 return ret;
3556}
3557
3558/*
3559 * ext4_get_block used when preparing for a DIO write or buffer write. 2589 * ext4_get_block used when preparing for a DIO write or buffer write.
3560 * We allocate an uinitialized extent if blocks haven't been allocated. 2590 * We allocate an uinitialized extent if blocks haven't been allocated.
3561 * The extent will be converted to initialized after the IO is complete. 2591 * The extent will be converted to initialized after the IO is complete.
@@ -4033,383 +3063,6 @@ unlock:
4033 return err; 3063 return err;
4034} 3064}
4035 3065
4036/*
4037 * Probably it should be a library function... search for first non-zero word
4038 * or memcmp with zero_page, whatever is better for particular architecture.
4039 * Linus?
4040 */
4041static inline int all_zeroes(__le32 *p, __le32 *q)
4042{
4043 while (p < q)
4044 if (*p++)
4045 return 0;
4046 return 1;
4047}
4048
4049/**
4050 * ext4_find_shared - find the indirect blocks for partial truncation.
4051 * @inode: inode in question
4052 * @depth: depth of the affected branch
4053 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4054 * @chain: place to store the pointers to partial indirect blocks
4055 * @top: place to the (detached) top of branch
4056 *
4057 * This is a helper function used by ext4_truncate().
4058 *
4059 * When we do truncate() we may have to clean the ends of several
4060 * indirect blocks but leave the blocks themselves alive. Block is
4061 * partially truncated if some data below the new i_size is referred
4062 * from it (and it is on the path to the first completely truncated
4063 * data block, indeed). We have to free the top of that path along
4064 * with everything to the right of the path. Since no allocation
4065 * past the truncation point is possible until ext4_truncate()
4066 * finishes, we may safely do the latter, but top of branch may
4067 * require special attention - pageout below the truncation point
4068 * might try to populate it.
4069 *
4070 * We atomically detach the top of branch from the tree, store the
4071 * block number of its root in *@top, pointers to buffer_heads of
4072 * partially truncated blocks - in @chain[].bh and pointers to
4073 * their last elements that should not be removed - in
4074 * @chain[].p. Return value is the pointer to last filled element
4075 * of @chain.
4076 *
4077 * The work left to caller to do the actual freeing of subtrees:
4078 * a) free the subtree starting from *@top
4079 * b) free the subtrees whose roots are stored in
4080 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4081 * c) free the subtrees growing from the inode past the @chain[0].
4082 * (no partially truncated stuff there). */
4083
4084static Indirect *ext4_find_shared(struct inode *inode, int depth,
4085 ext4_lblk_t offsets[4], Indirect chain[4],
4086 __le32 *top)
4087{
4088 Indirect *partial, *p;
4089 int k, err;
4090
4091 *top = 0;
4092 /* Make k index the deepest non-null offset + 1 */
4093 for (k = depth; k > 1 && !offsets[k-1]; k--)
4094 ;
4095 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4096 /* Writer: pointers */
4097 if (!partial)
4098 partial = chain + k-1;
4099 /*
4100 * If the branch acquired continuation since we've looked at it -
4101 * fine, it should all survive and (new) top doesn't belong to us.
4102 */
4103 if (!partial->key && *partial->p)
4104 /* Writer: end */
4105 goto no_top;
4106 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4107 ;
4108 /*
4109 * OK, we've found the last block that must survive. The rest of our
4110 * branch should be detached before unlocking. However, if that rest
4111 * of branch is all ours and does not grow immediately from the inode
4112 * it's easier to cheat and just decrement partial->p.
4113 */
4114 if (p == chain + k - 1 && p > chain) {
4115 p->p--;
4116 } else {
4117 *top = *p->p;
4118 /* Nope, don't do this in ext4. Must leave the tree intact */
4119#if 0
4120 *p->p = 0;
4121#endif
4122 }
4123 /* Writer: end */
4124
4125 while (partial > p) {
4126 brelse(partial->bh);
4127 partial--;
4128 }
4129no_top:
4130 return partial;
4131}
4132
4133/*
4134 * Zero a number of block pointers in either an inode or an indirect block.
4135 * If we restart the transaction we must again get write access to the
4136 * indirect block for further modification.
4137 *
4138 * We release `count' blocks on disk, but (last - first) may be greater
4139 * than `count' because there can be holes in there.
4140 *
4141 * Return 0 on success, 1 on invalid block range
4142 * and < 0 on fatal error.
4143 */
4144static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4145 struct buffer_head *bh,
4146 ext4_fsblk_t block_to_free,
4147 unsigned long count, __le32 *first,
4148 __le32 *last)
4149{
4150 __le32 *p;
4151 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4152 int err;
4153
4154 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4155 flags |= EXT4_FREE_BLOCKS_METADATA;
4156
4157 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4158 count)) {
4159 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4160 "blocks %llu len %lu",
4161 (unsigned long long) block_to_free, count);
4162 return 1;
4163 }
4164
4165 if (try_to_extend_transaction(handle, inode)) {
4166 if (bh) {
4167 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4168 err = ext4_handle_dirty_metadata(handle, inode, bh);
4169 if (unlikely(err))
4170 goto out_err;
4171 }
4172 err = ext4_mark_inode_dirty(handle, inode);
4173 if (unlikely(err))
4174 goto out_err;
4175 err = ext4_truncate_restart_trans(handle, inode,
4176 blocks_for_truncate(inode));
4177 if (unlikely(err))
4178 goto out_err;
4179 if (bh) {
4180 BUFFER_TRACE(bh, "retaking write access");
4181 err = ext4_journal_get_write_access(handle, bh);
4182 if (unlikely(err))
4183 goto out_err;
4184 }
4185 }
4186
4187 for (p = first; p < last; p++)
4188 *p = 0;
4189
4190 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
4191 return 0;
4192out_err:
4193 ext4_std_error(inode->i_sb, err);
4194 return err;
4195}
4196
4197/**
4198 * ext4_free_data - free a list of data blocks
4199 * @handle: handle for this transaction
4200 * @inode: inode we are dealing with
4201 * @this_bh: indirect buffer_head which contains *@first and *@last
4202 * @first: array of block numbers
4203 * @last: points immediately past the end of array
4204 *
4205 * We are freeing all blocks referred from that array (numbers are stored as
4206 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4207 *
4208 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4209 * blocks are contiguous then releasing them at one time will only affect one
4210 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4211 * actually use a lot of journal space.
4212 *
4213 * @this_bh will be %NULL if @first and @last point into the inode's direct
4214 * block pointers.
4215 */
4216static void ext4_free_data(handle_t *handle, struct inode *inode,
4217 struct buffer_head *this_bh,
4218 __le32 *first, __le32 *last)
4219{
4220 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4221 unsigned long count = 0; /* Number of blocks in the run */
4222 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4223 corresponding to
4224 block_to_free */
4225 ext4_fsblk_t nr; /* Current block # */
4226 __le32 *p; /* Pointer into inode/ind
4227 for current block */
4228 int err = 0;
4229
4230 if (this_bh) { /* For indirect block */
4231 BUFFER_TRACE(this_bh, "get_write_access");
4232 err = ext4_journal_get_write_access(handle, this_bh);
4233 /* Important: if we can't update the indirect pointers
4234 * to the blocks, we can't free them. */
4235 if (err)
4236 return;
4237 }
4238
4239 for (p = first; p < last; p++) {
4240 nr = le32_to_cpu(*p);
4241 if (nr) {
4242 /* accumulate blocks to free if they're contiguous */
4243 if (count == 0) {
4244 block_to_free = nr;
4245 block_to_free_p = p;
4246 count = 1;
4247 } else if (nr == block_to_free + count) {
4248 count++;
4249 } else {
4250 err = ext4_clear_blocks(handle, inode, this_bh,
4251 block_to_free, count,
4252 block_to_free_p, p);
4253 if (err)
4254 break;
4255 block_to_free = nr;
4256 block_to_free_p = p;
4257 count = 1;
4258 }
4259 }
4260 }
4261
4262 if (!err && count > 0)
4263 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4264 count, block_to_free_p, p);
4265 if (err < 0)
4266 /* fatal error */
4267 return;
4268
4269 if (this_bh) {
4270 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4271
4272 /*
4273 * The buffer head should have an attached journal head at this
4274 * point. However, if the data is corrupted and an indirect
4275 * block pointed to itself, it would have been detached when
4276 * the block was cleared. Check for this instead of OOPSing.
4277 */
4278 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4279 ext4_handle_dirty_metadata(handle, inode, this_bh);
4280 else
4281 EXT4_ERROR_INODE(inode,
4282 "circular indirect block detected at "
4283 "block %llu",
4284 (unsigned long long) this_bh->b_blocknr);
4285 }
4286}
4287
4288/**
4289 * ext4_free_branches - free an array of branches
4290 * @handle: JBD handle for this transaction
4291 * @inode: inode we are dealing with
4292 * @parent_bh: the buffer_head which contains *@first and *@last
4293 * @first: array of block numbers
4294 * @last: pointer immediately past the end of array
4295 * @depth: depth of the branches to free
4296 *
4297 * We are freeing all blocks referred from these branches (numbers are
4298 * stored as little-endian 32-bit) and updating @inode->i_blocks
4299 * appropriately.
4300 */
4301static void ext4_free_branches(handle_t *handle, struct inode *inode,
4302 struct buffer_head *parent_bh,
4303 __le32 *first, __le32 *last, int depth)
4304{
4305 ext4_fsblk_t nr;
4306 __le32 *p;
4307
4308 if (ext4_handle_is_aborted(handle))
4309 return;
4310
4311 if (depth--) {
4312 struct buffer_head *bh;
4313 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4314 p = last;
4315 while (--p >= first) {
4316 nr = le32_to_cpu(*p);
4317 if (!nr)
4318 continue; /* A hole */
4319
4320 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4321 nr, 1)) {
4322 EXT4_ERROR_INODE(inode,
4323 "invalid indirect mapped "
4324 "block %lu (level %d)",
4325 (unsigned long) nr, depth);
4326 break;
4327 }
4328
4329 /* Go read the buffer for the next level down */
4330 bh = sb_bread(inode->i_sb, nr);
4331
4332 /*
4333 * A read failure? Report error and clear slot
4334 * (should be rare).
4335 */
4336 if (!bh) {
4337 EXT4_ERROR_INODE_BLOCK(inode, nr,
4338 "Read failure");
4339 continue;
4340 }
4341
4342 /* This zaps the entire block. Bottom up. */
4343 BUFFER_TRACE(bh, "free child branches");
4344 ext4_free_branches(handle, inode, bh,
4345 (__le32 *) bh->b_data,
4346 (__le32 *) bh->b_data + addr_per_block,
4347 depth);
4348 brelse(bh);
4349
4350 /*
4351 * Everything below this this pointer has been
4352 * released. Now let this top-of-subtree go.
4353 *
4354 * We want the freeing of this indirect block to be
4355 * atomic in the journal with the updating of the
4356 * bitmap block which owns it. So make some room in
4357 * the journal.
4358 *
4359 * We zero the parent pointer *after* freeing its
4360 * pointee in the bitmaps, so if extend_transaction()
4361 * for some reason fails to put the bitmap changes and
4362 * the release into the same transaction, recovery
4363 * will merely complain about releasing a free block,
4364 * rather than leaking blocks.
4365 */
4366 if (ext4_handle_is_aborted(handle))
4367 return;
4368 if (try_to_extend_transaction(handle, inode)) {
4369 ext4_mark_inode_dirty(handle, inode);
4370 ext4_truncate_restart_trans(handle, inode,
4371 blocks_for_truncate(inode));
4372 }
4373
4374 /*
4375 * The forget flag here is critical because if
4376 * we are journaling (and not doing data
4377 * journaling), we have to make sure a revoke
4378 * record is written to prevent the journal
4379 * replay from overwriting the (former)
4380 * indirect block if it gets reallocated as a
4381 * data block. This must happen in the same
4382 * transaction where the data blocks are
4383 * actually freed.
4384 */
4385 ext4_free_blocks(handle, inode, NULL, nr, 1,
4386 EXT4_FREE_BLOCKS_METADATA|
4387 EXT4_FREE_BLOCKS_FORGET);
4388
4389 if (parent_bh) {
4390 /*
4391 * The block which we have just freed is
4392 * pointed to by an indirect block: journal it
4393 */
4394 BUFFER_TRACE(parent_bh, "get_write_access");
4395 if (!ext4_journal_get_write_access(handle,
4396 parent_bh)){
4397 *p = 0;
4398 BUFFER_TRACE(parent_bh,
4399 "call ext4_handle_dirty_metadata");
4400 ext4_handle_dirty_metadata(handle,
4401 inode,
4402 parent_bh);
4403 }
4404 }
4405 }
4406 } else {
4407 /* We have reached the bottom of the tree. */
4408 BUFFER_TRACE(parent_bh, "free data blocks");
4409 ext4_free_data(handle, inode, parent_bh, first, last);
4410 }
4411}
4412
4413int ext4_can_truncate(struct inode *inode) 3066int ext4_can_truncate(struct inode *inode)
4414{ 3067{
4415 if (S_ISREG(inode->i_mode)) 3068 if (S_ISREG(inode->i_mode))
@@ -4476,19 +3129,6 @@ int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4476 */ 3129 */
4477void ext4_truncate(struct inode *inode) 3130void ext4_truncate(struct inode *inode)
4478{ 3131{
4479 handle_t *handle;
4480 struct ext4_inode_info *ei = EXT4_I(inode);
4481 __le32 *i_data = ei->i_data;
4482 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4483 struct address_space *mapping = inode->i_mapping;
4484 ext4_lblk_t offsets[4];
4485 Indirect chain[4];
4486 Indirect *partial;
4487 __le32 nr = 0;
4488 int n = 0;
4489 ext4_lblk_t last_block, max_block;
4490 unsigned blocksize = inode->i_sb->s_blocksize;
4491
4492 trace_ext4_truncate_enter(inode); 3132 trace_ext4_truncate_enter(inode);
4493 3133
4494 if (!ext4_can_truncate(inode)) 3134 if (!ext4_can_truncate(inode))
@@ -4499,149 +3139,11 @@ void ext4_truncate(struct inode *inode)
4499 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) 3139 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4500 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 3140 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4501 3141
4502 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 3142 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4503 ext4_ext_truncate(inode); 3143 ext4_ext_truncate(inode);
4504 trace_ext4_truncate_exit(inode); 3144 else
4505 return; 3145 ext4_ind_truncate(inode);
4506 }
4507
4508 handle = start_transaction(inode);
4509 if (IS_ERR(handle))
4510 return; /* AKPM: return what? */
4511
4512 last_block = (inode->i_size + blocksize-1)
4513 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4514 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
4515 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4516
4517 if (inode->i_size & (blocksize - 1))
4518 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4519 goto out_stop;
4520
4521 if (last_block != max_block) {
4522 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4523 if (n == 0)
4524 goto out_stop; /* error */
4525 }
4526
4527 /*
4528 * OK. This truncate is going to happen. We add the inode to the
4529 * orphan list, so that if this truncate spans multiple transactions,
4530 * and we crash, we will resume the truncate when the filesystem
4531 * recovers. It also marks the inode dirty, to catch the new size.
4532 *
4533 * Implication: the file must always be in a sane, consistent
4534 * truncatable state while each transaction commits.
4535 */
4536 if (ext4_orphan_add(handle, inode))
4537 goto out_stop;
4538
4539 /*
4540 * From here we block out all ext4_get_block() callers who want to
4541 * modify the block allocation tree.
4542 */
4543 down_write(&ei->i_data_sem);
4544
4545 ext4_discard_preallocations(inode);
4546
4547 /*
4548 * The orphan list entry will now protect us from any crash which
4549 * occurs before the truncate completes, so it is now safe to propagate
4550 * the new, shorter inode size (held for now in i_size) into the
4551 * on-disk inode. We do this via i_disksize, which is the value which
4552 * ext4 *really* writes onto the disk inode.
4553 */
4554 ei->i_disksize = inode->i_size;
4555
4556 if (last_block == max_block) {
4557 /*
4558 * It is unnecessary to free any data blocks if last_block is
4559 * equal to the indirect block limit.
4560 */
4561 goto out_unlock;
4562 } else if (n == 1) { /* direct blocks */
4563 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4564 i_data + EXT4_NDIR_BLOCKS);
4565 goto do_indirects;
4566 }
4567
4568 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4569 /* Kill the top of shared branch (not detached) */
4570 if (nr) {
4571 if (partial == chain) {
4572 /* Shared branch grows from the inode */
4573 ext4_free_branches(handle, inode, NULL,
4574 &nr, &nr+1, (chain+n-1) - partial);
4575 *partial->p = 0;
4576 /*
4577 * We mark the inode dirty prior to restart,
4578 * and prior to stop. No need for it here.
4579 */
4580 } else {
4581 /* Shared branch grows from an indirect block */
4582 BUFFER_TRACE(partial->bh, "get_write_access");
4583 ext4_free_branches(handle, inode, partial->bh,
4584 partial->p,
4585 partial->p+1, (chain+n-1) - partial);
4586 }
4587 }
4588 /* Clear the ends of indirect blocks on the shared branch */
4589 while (partial > chain) {
4590 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4591 (__le32*)partial->bh->b_data+addr_per_block,
4592 (chain+n-1) - partial);
4593 BUFFER_TRACE(partial->bh, "call brelse");
4594 brelse(partial->bh);
4595 partial--;
4596 }
4597do_indirects:
4598 /* Kill the remaining (whole) subtrees */
4599 switch (offsets[0]) {
4600 default:
4601 nr = i_data[EXT4_IND_BLOCK];
4602 if (nr) {
4603 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4604 i_data[EXT4_IND_BLOCK] = 0;
4605 }
4606 case EXT4_IND_BLOCK:
4607 nr = i_data[EXT4_DIND_BLOCK];
4608 if (nr) {
4609 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4610 i_data[EXT4_DIND_BLOCK] = 0;
4611 }
4612 case EXT4_DIND_BLOCK:
4613 nr = i_data[EXT4_TIND_BLOCK];
4614 if (nr) {
4615 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4616 i_data[EXT4_TIND_BLOCK] = 0;
4617 }
4618 case EXT4_TIND_BLOCK:
4619 ;
4620 }
4621
4622out_unlock:
4623 up_write(&ei->i_data_sem);
4624 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4625 ext4_mark_inode_dirty(handle, inode);
4626
4627 /*
4628 * In a multi-transaction truncate, we only make the final transaction
4629 * synchronous
4630 */
4631 if (IS_SYNC(inode))
4632 ext4_handle_sync(handle);
4633out_stop:
4634 /*
4635 * If this was a simple ftruncate(), and the file will remain alive
4636 * then we need to clear up the orphan record which we created above.
4637 * However, if this was a real unlink then we were called by
4638 * ext4_delete_inode(), and we allow that function to clean up the
4639 * orphan info for us.
4640 */
4641 if (inode->i_nlink)
4642 ext4_orphan_del(handle, inode);
4643 3146
4644 ext4_journal_stop(handle);
4645 trace_ext4_truncate_exit(inode); 3147 trace_ext4_truncate_exit(inode);
4646} 3148}
4647 3149
@@ -5012,7 +3514,7 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
5012 (S_ISLNK(inode->i_mode) && 3514 (S_ISLNK(inode->i_mode) &&
5013 !ext4_inode_is_fast_symlink(inode))) { 3515 !ext4_inode_is_fast_symlink(inode))) {
5014 /* Validate block references which are part of inode */ 3516 /* Validate block references which are part of inode */
5015 ret = ext4_check_inode_blockref(inode); 3517 ret = ext4_ind_check_inode(inode);
5016 } 3518 }
5017 if (ret) 3519 if (ret)
5018 goto bad_inode; 3520 goto bad_inode;
@@ -5459,34 +3961,10 @@ int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5459 return 0; 3961 return 0;
5460} 3962}
5461 3963
5462static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5463 int chunk)
5464{
5465 int indirects;
5466
5467 /* if nrblocks are contiguous */
5468 if (chunk) {
5469 /*
5470 * With N contiguous data blocks, we need at most
5471 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
5472 * 2 dindirect blocks, and 1 tindirect block
5473 */
5474 return DIV_ROUND_UP(nrblocks,
5475 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
5476 }
5477 /*
5478 * if nrblocks are not contiguous, worse case, each block touch
5479 * a indirect block, and each indirect block touch a double indirect
5480 * block, plus a triple indirect block
5481 */
5482 indirects = nrblocks * 2 + 1;
5483 return indirects;
5484}
5485
5486static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) 3964static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5487{ 3965{
5488 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 3966 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5489 return ext4_indirect_trans_blocks(inode, nrblocks, chunk); 3967 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
5490 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); 3968 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5491} 3969}
5492 3970
diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
index 808c554e773..f18bfe37aff 100644
--- a/fs/ext4/ioctl.c
+++ b/fs/ext4/ioctl.c
@@ -202,8 +202,9 @@ setversion_out:
202 struct super_block *sb = inode->i_sb; 202 struct super_block *sb = inode->i_sb;
203 int err, err2=0; 203 int err, err2=0;
204 204
205 if (!capable(CAP_SYS_RESOURCE)) 205 err = ext4_resize_begin(sb);
206 return -EPERM; 206 if (err)
207 return err;
207 208
208 if (get_user(n_blocks_count, (__u32 __user *)arg)) 209 if (get_user(n_blocks_count, (__u32 __user *)arg))
209 return -EFAULT; 210 return -EFAULT;
@@ -221,6 +222,7 @@ setversion_out:
221 if (err == 0) 222 if (err == 0)
222 err = err2; 223 err = err2;
223 mnt_drop_write(filp->f_path.mnt); 224 mnt_drop_write(filp->f_path.mnt);
225 ext4_resize_end(sb);
224 226
225 return err; 227 return err;
226 } 228 }
@@ -271,8 +273,9 @@ mext_out:
271 struct super_block *sb = inode->i_sb; 273 struct super_block *sb = inode->i_sb;
272 int err, err2=0; 274 int err, err2=0;
273 275
274 if (!capable(CAP_SYS_RESOURCE)) 276 err = ext4_resize_begin(sb);
275 return -EPERM; 277 if (err)
278 return err;
276 279
277 if (copy_from_user(&input, (struct ext4_new_group_input __user *)arg, 280 if (copy_from_user(&input, (struct ext4_new_group_input __user *)arg,
278 sizeof(input))) 281 sizeof(input)))
@@ -291,6 +294,7 @@ mext_out:
291 if (err == 0) 294 if (err == 0)
292 err = err2; 295 err = err2;
293 mnt_drop_write(filp->f_path.mnt); 296 mnt_drop_write(filp->f_path.mnt);
297 ext4_resize_end(sb);
294 298
295 return err; 299 return err;
296 } 300 }
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 6ed859d5685..17a5a57c415 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -75,8 +75,8 @@
75 * 75 *
76 * The inode preallocation space is used looking at the _logical_ start 76 * The inode preallocation space is used looking at the _logical_ start
77 * block. If only the logical file block falls within the range of prealloc 77 * block. If only the logical file block falls within the range of prealloc
78 * space we will consume the particular prealloc space. This make sure that 78 * space we will consume the particular prealloc space. This makes sure that
79 * that the we have contiguous physical blocks representing the file blocks 79 * we have contiguous physical blocks representing the file blocks
80 * 80 *
81 * The important thing to be noted in case of inode prealloc space is that 81 * The important thing to be noted in case of inode prealloc space is that
82 * we don't modify the values associated to inode prealloc space except 82 * we don't modify the values associated to inode prealloc space except
@@ -84,7 +84,7 @@
84 * 84 *
85 * If we are not able to find blocks in the inode prealloc space and if we 85 * If we are not able to find blocks in the inode prealloc space and if we
86 * have the group allocation flag set then we look at the locality group 86 * have the group allocation flag set then we look at the locality group
87 * prealloc space. These are per CPU prealloc list repreasented as 87 * prealloc space. These are per CPU prealloc list represented as
88 * 88 *
89 * ext4_sb_info.s_locality_groups[smp_processor_id()] 89 * ext4_sb_info.s_locality_groups[smp_processor_id()]
90 * 90 *
@@ -128,12 +128,13 @@
128 * we are doing a group prealloc we try to normalize the request to 128 * we are doing a group prealloc we try to normalize the request to
129 * sbi->s_mb_group_prealloc. Default value of s_mb_group_prealloc is 129 * sbi->s_mb_group_prealloc. Default value of s_mb_group_prealloc is
130 * 512 blocks. This can be tuned via 130 * 512 blocks. This can be tuned via
131 * /sys/fs/ext4/<partition/mb_group_prealloc. The value is represented in 131 * /sys/fs/ext4/<partition>/mb_group_prealloc. The value is represented in
132 * terms of number of blocks. If we have mounted the file system with -O 132 * terms of number of blocks. If we have mounted the file system with -O
133 * stripe=<value> option the group prealloc request is normalized to the 133 * stripe=<value> option the group prealloc request is normalized to the
134 * stripe value (sbi->s_stripe) 134 * the smallest multiple of the stripe value (sbi->s_stripe) which is
135 * greater than the default mb_group_prealloc.
135 * 136 *
136 * The regular allocator(using the buddy cache) supports few tunables. 137 * The regular allocator (using the buddy cache) supports a few tunables.
137 * 138 *
138 * /sys/fs/ext4/<partition>/mb_min_to_scan 139 * /sys/fs/ext4/<partition>/mb_min_to_scan
139 * /sys/fs/ext4/<partition>/mb_max_to_scan 140 * /sys/fs/ext4/<partition>/mb_max_to_scan
@@ -152,7 +153,7 @@
152 * best extent in the found extents. Searching for the blocks starts with 153 * best extent in the found extents. Searching for the blocks starts with
153 * the group specified as the goal value in allocation context via 154 * the group specified as the goal value in allocation context via
154 * ac_g_ex. Each group is first checked based on the criteria whether it 155 * ac_g_ex. Each group is first checked based on the criteria whether it
155 * can used for allocation. ext4_mb_good_group explains how the groups are 156 * can be used for allocation. ext4_mb_good_group explains how the groups are
156 * checked. 157 * checked.
157 * 158 *
158 * Both the prealloc space are getting populated as above. So for the first 159 * Both the prealloc space are getting populated as above. So for the first
@@ -492,10 +493,11 @@ static void mb_cmp_bitmaps(struct ext4_buddy *e4b, void *bitmap)
492 b2 = (unsigned char *) bitmap; 493 b2 = (unsigned char *) bitmap;
493 for (i = 0; i < e4b->bd_sb->s_blocksize; i++) { 494 for (i = 0; i < e4b->bd_sb->s_blocksize; i++) {
494 if (b1[i] != b2[i]) { 495 if (b1[i] != b2[i]) {
495 printk(KERN_ERR "corruption in group %u " 496 ext4_msg(e4b->bd_sb, KERN_ERR,
496 "at byte %u(%u): %x in copy != %x " 497 "corruption in group %u "
497 "on disk/prealloc\n", 498 "at byte %u(%u): %x in copy != %x "
498 e4b->bd_group, i, i * 8, b1[i], b2[i]); 499 "on disk/prealloc",
500 e4b->bd_group, i, i * 8, b1[i], b2[i]);
499 BUG(); 501 BUG();
500 } 502 }
501 } 503 }
@@ -1125,7 +1127,7 @@ ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
1125 grp = ext4_get_group_info(sb, group); 1127 grp = ext4_get_group_info(sb, group);
1126 1128
1127 e4b->bd_blkbits = sb->s_blocksize_bits; 1129 e4b->bd_blkbits = sb->s_blocksize_bits;
1128 e4b->bd_info = ext4_get_group_info(sb, group); 1130 e4b->bd_info = grp;
1129 e4b->bd_sb = sb; 1131 e4b->bd_sb = sb;
1130 e4b->bd_group = group; 1132 e4b->bd_group = group;
1131 e4b->bd_buddy_page = NULL; 1133 e4b->bd_buddy_page = NULL;
@@ -1281,7 +1283,7 @@ static void mb_clear_bits(void *bm, int cur, int len)
1281 } 1283 }
1282} 1284}
1283 1285
1284static void mb_set_bits(void *bm, int cur, int len) 1286void ext4_set_bits(void *bm, int cur, int len)
1285{ 1287{
1286 __u32 *addr; 1288 __u32 *addr;
1287 1289
@@ -1510,7 +1512,7 @@ static int mb_mark_used(struct ext4_buddy *e4b, struct ext4_free_extent *ex)
1510 } 1512 }
1511 mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info); 1513 mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);
1512 1514
1513 mb_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0); 1515 ext4_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0);
1514 mb_check_buddy(e4b); 1516 mb_check_buddy(e4b);
1515 1517
1516 return ret; 1518 return ret;
@@ -2223,8 +2225,8 @@ int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2223 EXT4_DESC_PER_BLOCK_BITS(sb); 2225 EXT4_DESC_PER_BLOCK_BITS(sb);
2224 meta_group_info = kmalloc(metalen, GFP_KERNEL); 2226 meta_group_info = kmalloc(metalen, GFP_KERNEL);
2225 if (meta_group_info == NULL) { 2227 if (meta_group_info == NULL) {
2226 printk(KERN_ERR "EXT4-fs: can't allocate mem for a " 2228 ext4_msg(sb, KERN_ERR, "EXT4-fs: can't allocate mem "
2227 "buddy group\n"); 2229 "for a buddy group");
2228 goto exit_meta_group_info; 2230 goto exit_meta_group_info;
2229 } 2231 }
2230 sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)] = 2232 sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)] =
@@ -2237,7 +2239,7 @@ int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2237 2239
2238 meta_group_info[i] = kmem_cache_alloc(cachep, GFP_KERNEL); 2240 meta_group_info[i] = kmem_cache_alloc(cachep, GFP_KERNEL);
2239 if (meta_group_info[i] == NULL) { 2241 if (meta_group_info[i] == NULL) {
2240 printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n"); 2242 ext4_msg(sb, KERN_ERR, "EXT4-fs: can't allocate buddy mem");
2241 goto exit_group_info; 2243 goto exit_group_info;
2242 } 2244 }
2243 memset(meta_group_info[i], 0, kmem_cache_size(cachep)); 2245 memset(meta_group_info[i], 0, kmem_cache_size(cachep));
@@ -2279,8 +2281,10 @@ int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2279 2281
2280exit_group_info: 2282exit_group_info:
2281 /* If a meta_group_info table has been allocated, release it now */ 2283 /* If a meta_group_info table has been allocated, release it now */
2282 if (group % EXT4_DESC_PER_BLOCK(sb) == 0) 2284 if (group % EXT4_DESC_PER_BLOCK(sb) == 0) {
2283 kfree(sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)]); 2285 kfree(sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)]);
2286 sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)] = NULL;
2287 }
2284exit_meta_group_info: 2288exit_meta_group_info:
2285 return -ENOMEM; 2289 return -ENOMEM;
2286} /* ext4_mb_add_groupinfo */ 2290} /* ext4_mb_add_groupinfo */
@@ -2328,23 +2332,26 @@ static int ext4_mb_init_backend(struct super_block *sb)
2328 /* An 8TB filesystem with 64-bit pointers requires a 4096 byte 2332 /* An 8TB filesystem with 64-bit pointers requires a 4096 byte
2329 * kmalloc. A 128kb malloc should suffice for a 256TB filesystem. 2333 * kmalloc. A 128kb malloc should suffice for a 256TB filesystem.
2330 * So a two level scheme suffices for now. */ 2334 * So a two level scheme suffices for now. */
2331 sbi->s_group_info = kzalloc(array_size, GFP_KERNEL); 2335 sbi->s_group_info = ext4_kvzalloc(array_size, GFP_KERNEL);
2332 if (sbi->s_group_info == NULL) { 2336 if (sbi->s_group_info == NULL) {
2333 printk(KERN_ERR "EXT4-fs: can't allocate buddy meta group\n"); 2337 ext4_msg(sb, KERN_ERR, "can't allocate buddy meta group");
2334 return -ENOMEM; 2338 return -ENOMEM;
2335 } 2339 }
2336 sbi->s_buddy_cache = new_inode(sb); 2340 sbi->s_buddy_cache = new_inode(sb);
2337 if (sbi->s_buddy_cache == NULL) { 2341 if (sbi->s_buddy_cache == NULL) {
2338 printk(KERN_ERR "EXT4-fs: can't get new inode\n"); 2342 ext4_msg(sb, KERN_ERR, "can't get new inode");
2339 goto err_freesgi; 2343 goto err_freesgi;
2340 } 2344 }
2341 sbi->s_buddy_cache->i_ino = get_next_ino(); 2345 /* To avoid potentially colliding with an valid on-disk inode number,
2346 * use EXT4_BAD_INO for the buddy cache inode number. This inode is
2347 * not in the inode hash, so it should never be found by iget(), but
2348 * this will avoid confusion if it ever shows up during debugging. */
2349 sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
2342 EXT4_I(sbi->s_buddy_cache)->i_disksize = 0; 2350 EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
2343 for (i = 0; i < ngroups; i++) { 2351 for (i = 0; i < ngroups; i++) {
2344 desc = ext4_get_group_desc(sb, i, NULL); 2352 desc = ext4_get_group_desc(sb, i, NULL);
2345 if (desc == NULL) { 2353 if (desc == NULL) {
2346 printk(KERN_ERR 2354 ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
2347 "EXT4-fs: can't read descriptor %u\n", i);
2348 goto err_freebuddy; 2355 goto err_freebuddy;
2349 } 2356 }
2350 if (ext4_mb_add_groupinfo(sb, i, desc) != 0) 2357 if (ext4_mb_add_groupinfo(sb, i, desc) != 0)
@@ -2362,7 +2369,7 @@ err_freebuddy:
2362 kfree(sbi->s_group_info[i]); 2369 kfree(sbi->s_group_info[i]);
2363 iput(sbi->s_buddy_cache); 2370 iput(sbi->s_buddy_cache);
2364err_freesgi: 2371err_freesgi:
2365 kfree(sbi->s_group_info); 2372 ext4_kvfree(sbi->s_group_info);
2366 return -ENOMEM; 2373 return -ENOMEM;
2367} 2374}
2368 2375
@@ -2404,14 +2411,15 @@ static int ext4_groupinfo_create_slab(size_t size)
2404 slab_size, 0, SLAB_RECLAIM_ACCOUNT, 2411 slab_size, 0, SLAB_RECLAIM_ACCOUNT,
2405 NULL); 2412 NULL);
2406 2413
2414 ext4_groupinfo_caches[cache_index] = cachep;
2415
2407 mutex_unlock(&ext4_grpinfo_slab_create_mutex); 2416 mutex_unlock(&ext4_grpinfo_slab_create_mutex);
2408 if (!cachep) { 2417 if (!cachep) {
2409 printk(KERN_EMERG "EXT4: no memory for groupinfo slab cache\n"); 2418 printk(KERN_EMERG
2419 "EXT4-fs: no memory for groupinfo slab cache\n");
2410 return -ENOMEM; 2420 return -ENOMEM;
2411 } 2421 }
2412 2422
2413 ext4_groupinfo_caches[cache_index] = cachep;
2414
2415 return 0; 2423 return 0;
2416} 2424}
2417 2425
@@ -2457,12 +2465,6 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2457 i++; 2465 i++;
2458 } while (i <= sb->s_blocksize_bits + 1); 2466 } while (i <= sb->s_blocksize_bits + 1);
2459 2467
2460 /* init file for buddy data */
2461 ret = ext4_mb_init_backend(sb);
2462 if (ret != 0) {
2463 goto out;
2464 }
2465
2466 spin_lock_init(&sbi->s_md_lock); 2468 spin_lock_init(&sbi->s_md_lock);
2467 spin_lock_init(&sbi->s_bal_lock); 2469 spin_lock_init(&sbi->s_bal_lock);
2468 2470
@@ -2472,6 +2474,18 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2472 sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD; 2474 sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
2473 sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS; 2475 sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
2474 sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC; 2476 sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;
2477 /*
2478 * If there is a s_stripe > 1, then we set the s_mb_group_prealloc
2479 * to the lowest multiple of s_stripe which is bigger than
2480 * the s_mb_group_prealloc as determined above. We want
2481 * the preallocation size to be an exact multiple of the
2482 * RAID stripe size so that preallocations don't fragment
2483 * the stripes.
2484 */
2485 if (sbi->s_stripe > 1) {
2486 sbi->s_mb_group_prealloc = roundup(
2487 sbi->s_mb_group_prealloc, sbi->s_stripe);
2488 }
2475 2489
2476 sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group); 2490 sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
2477 if (sbi->s_locality_groups == NULL) { 2491 if (sbi->s_locality_groups == NULL) {
@@ -2487,6 +2501,12 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2487 spin_lock_init(&lg->lg_prealloc_lock); 2501 spin_lock_init(&lg->lg_prealloc_lock);
2488 } 2502 }
2489 2503
2504 /* init file for buddy data */
2505 ret = ext4_mb_init_backend(sb);
2506 if (ret != 0) {
2507 goto out;
2508 }
2509
2490 if (sbi->s_proc) 2510 if (sbi->s_proc)
2491 proc_create_data("mb_groups", S_IRUGO, sbi->s_proc, 2511 proc_create_data("mb_groups", S_IRUGO, sbi->s_proc,
2492 &ext4_mb_seq_groups_fops, sb); 2512 &ext4_mb_seq_groups_fops, sb);
@@ -2544,32 +2564,32 @@ int ext4_mb_release(struct super_block *sb)
2544 EXT4_DESC_PER_BLOCK_BITS(sb); 2564 EXT4_DESC_PER_BLOCK_BITS(sb);
2545 for (i = 0; i < num_meta_group_infos; i++) 2565 for (i = 0; i < num_meta_group_infos; i++)
2546 kfree(sbi->s_group_info[i]); 2566 kfree(sbi->s_group_info[i]);
2547 kfree(sbi->s_group_info); 2567 ext4_kvfree(sbi->s_group_info);
2548 } 2568 }
2549 kfree(sbi->s_mb_offsets); 2569 kfree(sbi->s_mb_offsets);
2550 kfree(sbi->s_mb_maxs); 2570 kfree(sbi->s_mb_maxs);
2551 if (sbi->s_buddy_cache) 2571 if (sbi->s_buddy_cache)
2552 iput(sbi->s_buddy_cache); 2572 iput(sbi->s_buddy_cache);
2553 if (sbi->s_mb_stats) { 2573 if (sbi->s_mb_stats) {
2554 printk(KERN_INFO 2574 ext4_msg(sb, KERN_INFO,
2555 "EXT4-fs: mballoc: %u blocks %u reqs (%u success)\n", 2575 "mballoc: %u blocks %u reqs (%u success)",
2556 atomic_read(&sbi->s_bal_allocated), 2576 atomic_read(&sbi->s_bal_allocated),
2557 atomic_read(&sbi->s_bal_reqs), 2577 atomic_read(&sbi->s_bal_reqs),
2558 atomic_read(&sbi->s_bal_success)); 2578 atomic_read(&sbi->s_bal_success));
2559 printk(KERN_INFO 2579 ext4_msg(sb, KERN_INFO,
2560 "EXT4-fs: mballoc: %u extents scanned, %u goal hits, " 2580 "mballoc: %u extents scanned, %u goal hits, "
2561 "%u 2^N hits, %u breaks, %u lost\n", 2581 "%u 2^N hits, %u breaks, %u lost",
2562 atomic_read(&sbi->s_bal_ex_scanned), 2582 atomic_read(&sbi->s_bal_ex_scanned),
2563 atomic_read(&sbi->s_bal_goals), 2583 atomic_read(&sbi->s_bal_goals),
2564 atomic_read(&sbi->s_bal_2orders), 2584 atomic_read(&sbi->s_bal_2orders),
2565 atomic_read(&sbi->s_bal_breaks), 2585 atomic_read(&sbi->s_bal_breaks),
2566 atomic_read(&sbi->s_mb_lost_chunks)); 2586 atomic_read(&sbi->s_mb_lost_chunks));
2567 printk(KERN_INFO 2587 ext4_msg(sb, KERN_INFO,
2568 "EXT4-fs: mballoc: %lu generated and it took %Lu\n", 2588 "mballoc: %lu generated and it took %Lu",
2569 sbi->s_mb_buddies_generated++, 2589 sbi->s_mb_buddies_generated,
2570 sbi->s_mb_generation_time); 2590 sbi->s_mb_generation_time);
2571 printk(KERN_INFO 2591 ext4_msg(sb, KERN_INFO,
2572 "EXT4-fs: mballoc: %u preallocated, %u discarded\n", 2592 "mballoc: %u preallocated, %u discarded",
2573 atomic_read(&sbi->s_mb_preallocated), 2593 atomic_read(&sbi->s_mb_preallocated),
2574 atomic_read(&sbi->s_mb_discarded)); 2594 atomic_read(&sbi->s_mb_discarded));
2575 } 2595 }
@@ -2628,6 +2648,15 @@ static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
2628 rb_erase(&entry->node, &(db->bb_free_root)); 2648 rb_erase(&entry->node, &(db->bb_free_root));
2629 mb_free_blocks(NULL, &e4b, entry->start_blk, entry->count); 2649 mb_free_blocks(NULL, &e4b, entry->start_blk, entry->count);
2630 2650
2651 /*
2652 * Clear the trimmed flag for the group so that the next
2653 * ext4_trim_fs can trim it.
2654 * If the volume is mounted with -o discard, online discard
2655 * is supported and the free blocks will be trimmed online.
2656 */
2657 if (!test_opt(sb, DISCARD))
2658 EXT4_MB_GRP_CLEAR_TRIMMED(db);
2659
2631 if (!db->bb_free_root.rb_node) { 2660 if (!db->bb_free_root.rb_node) {
2632 /* No more items in the per group rb tree 2661 /* No more items in the per group rb tree
2633 * balance refcounts from ext4_mb_free_metadata() 2662 * balance refcounts from ext4_mb_free_metadata()
@@ -2771,8 +2800,8 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2771 * We leak some of the blocks here. 2800 * We leak some of the blocks here.
2772 */ 2801 */
2773 ext4_lock_group(sb, ac->ac_b_ex.fe_group); 2802 ext4_lock_group(sb, ac->ac_b_ex.fe_group);
2774 mb_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start, 2803 ext4_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,
2775 ac->ac_b_ex.fe_len); 2804 ac->ac_b_ex.fe_len);
2776 ext4_unlock_group(sb, ac->ac_b_ex.fe_group); 2805 ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
2777 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); 2806 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2778 if (!err) 2807 if (!err)
@@ -2790,7 +2819,8 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2790 } 2819 }
2791 } 2820 }
2792#endif 2821#endif
2793 mb_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,ac->ac_b_ex.fe_len); 2822 ext4_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,
2823 ac->ac_b_ex.fe_len);
2794 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 2824 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
2795 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); 2825 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
2796 ext4_free_blks_set(sb, gdp, 2826 ext4_free_blks_set(sb, gdp,
@@ -2830,8 +2860,9 @@ out_err:
2830 2860
2831/* 2861/*
2832 * here we normalize request for locality group 2862 * here we normalize request for locality group
2833 * Group request are normalized to s_strip size if we set the same via mount 2863 * Group request are normalized to s_mb_group_prealloc, which goes to
2834 * option. If not we set it to s_mb_group_prealloc which can be configured via 2864 * s_strip if we set the same via mount option.
2865 * s_mb_group_prealloc can be configured via
2835 * /sys/fs/ext4/<partition>/mb_group_prealloc 2866 * /sys/fs/ext4/<partition>/mb_group_prealloc
2836 * 2867 *
2837 * XXX: should we try to preallocate more than the group has now? 2868 * XXX: should we try to preallocate more than the group has now?
@@ -2842,10 +2873,7 @@ static void ext4_mb_normalize_group_request(struct ext4_allocation_context *ac)
2842 struct ext4_locality_group *lg = ac->ac_lg; 2873 struct ext4_locality_group *lg = ac->ac_lg;
2843 2874
2844 BUG_ON(lg == NULL); 2875 BUG_ON(lg == NULL);
2845 if (EXT4_SB(sb)->s_stripe) 2876 ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
2846 ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_stripe;
2847 else
2848 ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
2849 mb_debug(1, "#%u: goal %u blocks for locality group\n", 2877 mb_debug(1, "#%u: goal %u blocks for locality group\n",
2850 current->pid, ac->ac_g_ex.fe_len); 2878 current->pid, ac->ac_g_ex.fe_len);
2851} 2879}
@@ -3001,9 +3029,10 @@ ext4_mb_normalize_request(struct ext4_allocation_context *ac,
3001 3029
3002 if (start + size <= ac->ac_o_ex.fe_logical && 3030 if (start + size <= ac->ac_o_ex.fe_logical &&
3003 start > ac->ac_o_ex.fe_logical) { 3031 start > ac->ac_o_ex.fe_logical) {
3004 printk(KERN_ERR "start %lu, size %lu, fe_logical %lu\n", 3032 ext4_msg(ac->ac_sb, KERN_ERR,
3005 (unsigned long) start, (unsigned long) size, 3033 "start %lu, size %lu, fe_logical %lu",
3006 (unsigned long) ac->ac_o_ex.fe_logical); 3034 (unsigned long) start, (unsigned long) size,
3035 (unsigned long) ac->ac_o_ex.fe_logical);
3007 } 3036 }
3008 BUG_ON(start + size <= ac->ac_o_ex.fe_logical && 3037 BUG_ON(start + size <= ac->ac_o_ex.fe_logical &&
3009 start > ac->ac_o_ex.fe_logical); 3038 start > ac->ac_o_ex.fe_logical);
@@ -3262,7 +3291,7 @@ static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
3262 3291
3263 while (n) { 3292 while (n) {
3264 entry = rb_entry(n, struct ext4_free_data, node); 3293 entry = rb_entry(n, struct ext4_free_data, node);
3265 mb_set_bits(bitmap, entry->start_blk, entry->count); 3294 ext4_set_bits(bitmap, entry->start_blk, entry->count);
3266 n = rb_next(n); 3295 n = rb_next(n);
3267 } 3296 }
3268 return; 3297 return;
@@ -3304,7 +3333,7 @@ void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
3304 if (unlikely(len == 0)) 3333 if (unlikely(len == 0))
3305 continue; 3334 continue;
3306 BUG_ON(groupnr != group); 3335 BUG_ON(groupnr != group);
3307 mb_set_bits(bitmap, start, len); 3336 ext4_set_bits(bitmap, start, len);
3308 preallocated += len; 3337 preallocated += len;
3309 count++; 3338 count++;
3310 } 3339 }
@@ -3584,10 +3613,11 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3584 bit = next + 1; 3613 bit = next + 1;
3585 } 3614 }
3586 if (free != pa->pa_free) { 3615 if (free != pa->pa_free) {
3587 printk(KERN_CRIT "pa %p: logic %lu, phys. %lu, len %lu\n", 3616 ext4_msg(e4b->bd_sb, KERN_CRIT,
3588 pa, (unsigned long) pa->pa_lstart, 3617 "pa %p: logic %lu, phys. %lu, len %lu",
3589 (unsigned long) pa->pa_pstart, 3618 pa, (unsigned long) pa->pa_lstart,
3590 (unsigned long) pa->pa_len); 3619 (unsigned long) pa->pa_pstart,
3620 (unsigned long) pa->pa_len);
3591 ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u", 3621 ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
3592 free, pa->pa_free); 3622 free, pa->pa_free);
3593 /* 3623 /*
@@ -3775,7 +3805,8 @@ repeat:
3775 * use preallocation while we're discarding it */ 3805 * use preallocation while we're discarding it */
3776 spin_unlock(&pa->pa_lock); 3806 spin_unlock(&pa->pa_lock);
3777 spin_unlock(&ei->i_prealloc_lock); 3807 spin_unlock(&ei->i_prealloc_lock);
3778 printk(KERN_ERR "uh-oh! used pa while discarding\n"); 3808 ext4_msg(sb, KERN_ERR,
3809 "uh-oh! used pa while discarding");
3779 WARN_ON(1); 3810 WARN_ON(1);
3780 schedule_timeout_uninterruptible(HZ); 3811 schedule_timeout_uninterruptible(HZ);
3781 goto repeat; 3812 goto repeat;
@@ -3852,12 +3883,13 @@ static void ext4_mb_show_ac(struct ext4_allocation_context *ac)
3852 (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) 3883 (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED))
3853 return; 3884 return;
3854 3885
3855 printk(KERN_ERR "EXT4-fs: Can't allocate:" 3886 ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: Can't allocate:"
3856 " Allocation context details:\n"); 3887 " Allocation context details:");
3857 printk(KERN_ERR "EXT4-fs: status %d flags %d\n", 3888 ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: status %d flags %d",
3858 ac->ac_status, ac->ac_flags); 3889 ac->ac_status, ac->ac_flags);
3859 printk(KERN_ERR "EXT4-fs: orig %lu/%lu/%lu@%lu, goal %lu/%lu/%lu@%lu, " 3890 ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: orig %lu/%lu/%lu@%lu, "
3860 "best %lu/%lu/%lu@%lu cr %d\n", 3891 "goal %lu/%lu/%lu@%lu, "
3892 "best %lu/%lu/%lu@%lu cr %d",
3861 (unsigned long)ac->ac_o_ex.fe_group, 3893 (unsigned long)ac->ac_o_ex.fe_group,
3862 (unsigned long)ac->ac_o_ex.fe_start, 3894 (unsigned long)ac->ac_o_ex.fe_start,
3863 (unsigned long)ac->ac_o_ex.fe_len, 3895 (unsigned long)ac->ac_o_ex.fe_len,
@@ -3871,9 +3903,9 @@ static void ext4_mb_show_ac(struct ext4_allocation_context *ac)
3871 (unsigned long)ac->ac_b_ex.fe_len, 3903 (unsigned long)ac->ac_b_ex.fe_len,
3872 (unsigned long)ac->ac_b_ex.fe_logical, 3904 (unsigned long)ac->ac_b_ex.fe_logical,
3873 (int)ac->ac_criteria); 3905 (int)ac->ac_criteria);
3874 printk(KERN_ERR "EXT4-fs: %lu scanned, %d found\n", ac->ac_ex_scanned, 3906 ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: %lu scanned, %d found",
3875 ac->ac_found); 3907 ac->ac_ex_scanned, ac->ac_found);
3876 printk(KERN_ERR "EXT4-fs: groups: \n"); 3908 ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: groups: ");
3877 ngroups = ext4_get_groups_count(sb); 3909 ngroups = ext4_get_groups_count(sb);
3878 for (i = 0; i < ngroups; i++) { 3910 for (i = 0; i < ngroups; i++) {
3879 struct ext4_group_info *grp = ext4_get_group_info(sb, i); 3911 struct ext4_group_info *grp = ext4_get_group_info(sb, i);
@@ -4637,7 +4669,7 @@ do_more:
4637 } 4669 }
4638 ext4_mark_super_dirty(sb); 4670 ext4_mark_super_dirty(sb);
4639error_return: 4671error_return:
4640 if (freed) 4672 if (freed && !(flags & EXT4_FREE_BLOCKS_NO_QUOT_UPDATE))
4641 dquot_free_block(inode, freed); 4673 dquot_free_block(inode, freed);
4642 brelse(bitmap_bh); 4674 brelse(bitmap_bh);
4643 ext4_std_error(sb, err); 4675 ext4_std_error(sb, err);
@@ -4645,7 +4677,7 @@ error_return:
4645} 4677}
4646 4678
4647/** 4679/**
4648 * ext4_add_groupblocks() -- Add given blocks to an existing group 4680 * ext4_group_add_blocks() -- Add given blocks to an existing group
4649 * @handle: handle to this transaction 4681 * @handle: handle to this transaction
4650 * @sb: super block 4682 * @sb: super block
4651 * @block: start physcial block to add to the block group 4683 * @block: start physcial block to add to the block group
@@ -4653,7 +4685,7 @@ error_return:
4653 * 4685 *
4654 * This marks the blocks as free in the bitmap and buddy. 4686 * This marks the blocks as free in the bitmap and buddy.
4655 */ 4687 */
4656void ext4_add_groupblocks(handle_t *handle, struct super_block *sb, 4688int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
4657 ext4_fsblk_t block, unsigned long count) 4689 ext4_fsblk_t block, unsigned long count)
4658{ 4690{
4659 struct buffer_head *bitmap_bh = NULL; 4691 struct buffer_head *bitmap_bh = NULL;
@@ -4666,25 +4698,35 @@ void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
4666 struct ext4_buddy e4b; 4698 struct ext4_buddy e4b;
4667 int err = 0, ret, blk_free_count; 4699 int err = 0, ret, blk_free_count;
4668 ext4_grpblk_t blocks_freed; 4700 ext4_grpblk_t blocks_freed;
4669 struct ext4_group_info *grp;
4670 4701
4671 ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1); 4702 ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
4672 4703
4704 if (count == 0)
4705 return 0;
4706
4673 ext4_get_group_no_and_offset(sb, block, &block_group, &bit); 4707 ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
4674 grp = ext4_get_group_info(sb, block_group);
4675 /* 4708 /*
4676 * Check to see if we are freeing blocks across a group 4709 * Check to see if we are freeing blocks across a group
4677 * boundary. 4710 * boundary.
4678 */ 4711 */
4679 if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) 4712 if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
4713 ext4_warning(sb, "too much blocks added to group %u\n",
4714 block_group);
4715 err = -EINVAL;
4680 goto error_return; 4716 goto error_return;
4717 }
4681 4718
4682 bitmap_bh = ext4_read_block_bitmap(sb, block_group); 4719 bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4683 if (!bitmap_bh) 4720 if (!bitmap_bh) {
4721 err = -EIO;
4684 goto error_return; 4722 goto error_return;
4723 }
4724
4685 desc = ext4_get_group_desc(sb, block_group, &gd_bh); 4725 desc = ext4_get_group_desc(sb, block_group, &gd_bh);
4686 if (!desc) 4726 if (!desc) {
4727 err = -EIO;
4687 goto error_return; 4728 goto error_return;
4729 }
4688 4730
4689 if (in_range(ext4_block_bitmap(sb, desc), block, count) || 4731 if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
4690 in_range(ext4_inode_bitmap(sb, desc), block, count) || 4732 in_range(ext4_inode_bitmap(sb, desc), block, count) ||
@@ -4694,6 +4736,7 @@ void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
4694 ext4_error(sb, "Adding blocks in system zones - " 4736 ext4_error(sb, "Adding blocks in system zones - "
4695 "Block = %llu, count = %lu", 4737 "Block = %llu, count = %lu",
4696 block, count); 4738 block, count);
4739 err = -EINVAL;
4697 goto error_return; 4740 goto error_return;
4698 } 4741 }
4699 4742
@@ -4762,7 +4805,7 @@ void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
4762error_return: 4805error_return:
4763 brelse(bitmap_bh); 4806 brelse(bitmap_bh);
4764 ext4_std_error(sb, err); 4807 ext4_std_error(sb, err);
4765 return; 4808 return err;
4766} 4809}
4767 4810
4768/** 4811/**
@@ -4782,6 +4825,8 @@ static void ext4_trim_extent(struct super_block *sb, int start, int count,
4782{ 4825{
4783 struct ext4_free_extent ex; 4826 struct ext4_free_extent ex;
4784 4827
4828 trace_ext4_trim_extent(sb, group, start, count);
4829
4785 assert_spin_locked(ext4_group_lock_ptr(sb, group)); 4830 assert_spin_locked(ext4_group_lock_ptr(sb, group));
4786 4831
4787 ex.fe_start = start; 4832 ex.fe_start = start;
@@ -4802,7 +4847,7 @@ static void ext4_trim_extent(struct super_block *sb, int start, int count,
4802/** 4847/**
4803 * ext4_trim_all_free -- function to trim all free space in alloc. group 4848 * ext4_trim_all_free -- function to trim all free space in alloc. group
4804 * @sb: super block for file system 4849 * @sb: super block for file system
4805 * @e4b: ext4 buddy 4850 * @group: group to be trimmed
4806 * @start: first group block to examine 4851 * @start: first group block to examine
4807 * @max: last group block to examine 4852 * @max: last group block to examine
4808 * @minblocks: minimum extent block count 4853 * @minblocks: minimum extent block count
@@ -4823,10 +4868,12 @@ ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
4823 ext4_grpblk_t minblocks) 4868 ext4_grpblk_t minblocks)
4824{ 4869{
4825 void *bitmap; 4870 void *bitmap;
4826 ext4_grpblk_t next, count = 0; 4871 ext4_grpblk_t next, count = 0, free_count = 0;
4827 struct ext4_buddy e4b; 4872 struct ext4_buddy e4b;
4828 int ret; 4873 int ret;
4829 4874
4875 trace_ext4_trim_all_free(sb, group, start, max);
4876
4830 ret = ext4_mb_load_buddy(sb, group, &e4b); 4877 ret = ext4_mb_load_buddy(sb, group, &e4b);
4831 if (ret) { 4878 if (ret) {
4832 ext4_error(sb, "Error in loading buddy " 4879 ext4_error(sb, "Error in loading buddy "
@@ -4836,6 +4883,10 @@ ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
4836 bitmap = e4b.bd_bitmap; 4883 bitmap = e4b.bd_bitmap;
4837 4884
4838 ext4_lock_group(sb, group); 4885 ext4_lock_group(sb, group);
4886 if (EXT4_MB_GRP_WAS_TRIMMED(e4b.bd_info) &&
4887 minblocks >= atomic_read(&EXT4_SB(sb)->s_last_trim_minblks))
4888 goto out;
4889
4839 start = (e4b.bd_info->bb_first_free > start) ? 4890 start = (e4b.bd_info->bb_first_free > start) ?
4840 e4b.bd_info->bb_first_free : start; 4891 e4b.bd_info->bb_first_free : start;
4841 4892
@@ -4850,6 +4901,7 @@ ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
4850 next - start, group, &e4b); 4901 next - start, group, &e4b);
4851 count += next - start; 4902 count += next - start;
4852 } 4903 }
4904 free_count += next - start;
4853 start = next + 1; 4905 start = next + 1;
4854 4906
4855 if (fatal_signal_pending(current)) { 4907 if (fatal_signal_pending(current)) {
@@ -4863,9 +4915,13 @@ ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
4863 ext4_lock_group(sb, group); 4915 ext4_lock_group(sb, group);
4864 } 4916 }
4865 4917
4866 if ((e4b.bd_info->bb_free - count) < minblocks) 4918 if ((e4b.bd_info->bb_free - free_count) < minblocks)
4867 break; 4919 break;
4868 } 4920 }
4921
4922 if (!ret)
4923 EXT4_MB_GRP_SET_TRIMMED(e4b.bd_info);
4924out:
4869 ext4_unlock_group(sb, group); 4925 ext4_unlock_group(sb, group);
4870 ext4_mb_unload_buddy(&e4b); 4926 ext4_mb_unload_buddy(&e4b);
4871 4927
@@ -4904,6 +4960,8 @@ int ext4_trim_fs(struct super_block *sb, struct fstrim_range *range)
4904 4960
4905 if (unlikely(minlen > EXT4_BLOCKS_PER_GROUP(sb))) 4961 if (unlikely(minlen > EXT4_BLOCKS_PER_GROUP(sb)))
4906 return -EINVAL; 4962 return -EINVAL;
4963 if (start + len <= first_data_blk)
4964 goto out;
4907 if (start < first_data_blk) { 4965 if (start < first_data_blk) {
4908 len -= first_data_blk - start; 4966 len -= first_data_blk - start;
4909 start = first_data_blk; 4967 start = first_data_blk;
@@ -4952,5 +5010,9 @@ int ext4_trim_fs(struct super_block *sb, struct fstrim_range *range)
4952 } 5010 }
4953 range->len = trimmed * sb->s_blocksize; 5011 range->len = trimmed * sb->s_blocksize;
4954 5012
5013 if (!ret)
5014 atomic_set(&EXT4_SB(sb)->s_last_trim_minblks, minlen);
5015
5016out:
4955 return ret; 5017 return ret;
4956} 5018}
diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h
index 20b5e7bfebd..9d4a636b546 100644
--- a/fs/ext4/mballoc.h
+++ b/fs/ext4/mballoc.h
@@ -187,7 +187,6 @@ struct ext4_allocation_context {
187 __u16 ac_flags; /* allocation hints */ 187 __u16 ac_flags; /* allocation hints */
188 __u8 ac_status; 188 __u8 ac_status;
189 __u8 ac_criteria; 189 __u8 ac_criteria;
190 __u8 ac_repeats;
191 __u8 ac_2order; /* if request is to allocate 2^N blocks and 190 __u8 ac_2order; /* if request is to allocate 2^N blocks and
192 * N > 0, the field stores N, otherwise 0 */ 191 * N > 0, the field stores N, otherwise 0 */
193 __u8 ac_op; /* operation, for history only */ 192 __u8 ac_op; /* operation, for history only */
diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index 8c9babac43d..565a154e22d 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -289,7 +289,7 @@ static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_ent
289 while (len--) printk("%c", *name++); 289 while (len--) printk("%c", *name++);
290 ext4fs_dirhash(de->name, de->name_len, &h); 290 ext4fs_dirhash(de->name, de->name_len, &h);
291 printk(":%x.%u ", h.hash, 291 printk(":%x.%u ", h.hash,
292 ((char *) de - base)); 292 (unsigned) ((char *) de - base));
293 } 293 }
294 space += EXT4_DIR_REC_LEN(de->name_len); 294 space += EXT4_DIR_REC_LEN(de->name_len);
295 names++; 295 names++;
@@ -1013,7 +1013,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct q
1013 1013
1014 *err = -ENOENT; 1014 *err = -ENOENT;
1015errout: 1015errout:
1016 dxtrace(printk(KERN_DEBUG "%s not found\n", name)); 1016 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1017 dx_release (frames); 1017 dx_release (frames);
1018 return NULL; 1018 return NULL;
1019} 1019}
@@ -1985,18 +1985,11 @@ int ext4_orphan_add(handle_t *handle, struct inode *inode)
1985 if (!list_empty(&EXT4_I(inode)->i_orphan)) 1985 if (!list_empty(&EXT4_I(inode)->i_orphan))
1986 goto out_unlock; 1986 goto out_unlock;
1987 1987
1988 /* Orphan handling is only valid for files with data blocks 1988 /*
1989 * being truncated, or files being unlinked. */ 1989 * Orphan handling is only valid for files with data blocks
1990 1990 * being truncated, or files being unlinked. Note that we either
1991 /* @@@ FIXME: Observation from aviro: 1991 * hold i_mutex, or the inode can not be referenced from outside,
1992 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block 1992 * so i_nlink should not be bumped due to race
1993 * here (on s_orphan_lock), so race with ext4_link() which might bump
1994 * ->i_nlink. For, say it, character device. Not a regular file,
1995 * not a directory, not a symlink and ->i_nlink > 0.
1996 *
1997 * tytso, 4/25/2009: I'm not sure how that could happen;
1998 * shouldn't the fs core protect us from these sort of
1999 * unlink()/link() races?
2000 */ 1993 */
2001 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1994 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2002 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); 1995 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
diff --git a/fs/ext4/page-io.c b/fs/ext4/page-io.c
index 7bb8f76d470..430c401d089 100644
--- a/fs/ext4/page-io.c
+++ b/fs/ext4/page-io.c
@@ -285,11 +285,7 @@ static int io_submit_init(struct ext4_io_submit *io,
285 io_end = ext4_init_io_end(inode, GFP_NOFS); 285 io_end = ext4_init_io_end(inode, GFP_NOFS);
286 if (!io_end) 286 if (!io_end)
287 return -ENOMEM; 287 return -ENOMEM;
288 do { 288 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
289 bio = bio_alloc(GFP_NOIO, nvecs);
290 nvecs >>= 1;
291 } while (bio == NULL);
292
293 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); 289 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
294 bio->bi_bdev = bh->b_bdev; 290 bio->bi_bdev = bh->b_bdev;
295 bio->bi_private = io->io_end = io_end; 291 bio->bi_private = io->io_end = io_end;
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
index 80bbc9c60c2..707d3f16f7c 100644
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -16,6 +16,35 @@
16 16
17#include "ext4_jbd2.h" 17#include "ext4_jbd2.h"
18 18
19int ext4_resize_begin(struct super_block *sb)
20{
21 int ret = 0;
22
23 if (!capable(CAP_SYS_RESOURCE))
24 return -EPERM;
25
26 /*
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
29 */
30 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 ext4_warning(sb, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
33 return -EPERM;
34 }
35
36 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
37 ret = -EBUSY;
38
39 return ret;
40}
41
42void ext4_resize_end(struct super_block *sb)
43{
44 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 smp_mb__after_clear_bit();
46}
47
19#define outside(b, first, last) ((b) < (first) || (b) >= (last)) 48#define outside(b, first, last) ((b) < (first) || (b) >= (last))
20#define inside(b, first, last) ((b) >= (first) && (b) < (last)) 49#define inside(b, first, last) ((b) >= (first) && (b) < (last))
21 50
@@ -118,10 +147,8 @@ static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
118 brelse(bh); 147 brelse(bh);
119 bh = ERR_PTR(err); 148 bh = ERR_PTR(err);
120 } else { 149 } else {
121 lock_buffer(bh);
122 memset(bh->b_data, 0, sb->s_blocksize); 150 memset(bh->b_data, 0, sb->s_blocksize);
123 set_buffer_uptodate(bh); 151 set_buffer_uptodate(bh);
124 unlock_buffer(bh);
125 } 152 }
126 153
127 return bh; 154 return bh;
@@ -132,8 +159,7 @@ static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
132 * If that fails, restart the transaction & regain write access for the 159 * If that fails, restart the transaction & regain write access for the
133 * buffer head which is used for block_bitmap modifications. 160 * buffer head which is used for block_bitmap modifications.
134 */ 161 */
135static int extend_or_restart_transaction(handle_t *handle, int thresh, 162static int extend_or_restart_transaction(handle_t *handle, int thresh)
136 struct buffer_head *bh)
137{ 163{
138 int err; 164 int err;
139 165
@@ -144,9 +170,8 @@ static int extend_or_restart_transaction(handle_t *handle, int thresh,
144 if (err < 0) 170 if (err < 0)
145 return err; 171 return err;
146 if (err) { 172 if (err) {
147 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA))) 173 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
148 return err; 174 if (err)
149 if ((err = ext4_journal_get_write_access(handle, bh)))
150 return err; 175 return err;
151 } 176 }
152 177
@@ -181,21 +206,7 @@ static int setup_new_group_blocks(struct super_block *sb,
181 if (IS_ERR(handle)) 206 if (IS_ERR(handle))
182 return PTR_ERR(handle); 207 return PTR_ERR(handle);
183 208
184 mutex_lock(&sbi->s_resize_lock); 209 BUG_ON(input->group != sbi->s_groups_count);
185 if (input->group != sbi->s_groups_count) {
186 err = -EBUSY;
187 goto exit_journal;
188 }
189
190 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
191 err = PTR_ERR(bh);
192 goto exit_journal;
193 }
194
195 if (ext4_bg_has_super(sb, input->group)) {
196 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
197 ext4_set_bit(0, bh->b_data);
198 }
199 210
200 /* Copy all of the GDT blocks into the backup in this group */ 211 /* Copy all of the GDT blocks into the backup in this group */
201 for (i = 0, bit = 1, block = start + 1; 212 for (i = 0, bit = 1, block = start + 1;
@@ -203,29 +214,26 @@ static int setup_new_group_blocks(struct super_block *sb,
203 struct buffer_head *gdb; 214 struct buffer_head *gdb;
204 215
205 ext4_debug("update backup group %#04llx (+%d)\n", block, bit); 216 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
206 217 err = extend_or_restart_transaction(handle, 1);
207 if ((err = extend_or_restart_transaction(handle, 1, bh))) 218 if (err)
208 goto exit_bh; 219 goto exit_journal;
209 220
210 gdb = sb_getblk(sb, block); 221 gdb = sb_getblk(sb, block);
211 if (!gdb) { 222 if (!gdb) {
212 err = -EIO; 223 err = -EIO;
213 goto exit_bh; 224 goto exit_journal;
214 } 225 }
215 if ((err = ext4_journal_get_write_access(handle, gdb))) { 226 if ((err = ext4_journal_get_write_access(handle, gdb))) {
216 brelse(gdb); 227 brelse(gdb);
217 goto exit_bh; 228 goto exit_journal;
218 } 229 }
219 lock_buffer(gdb);
220 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size); 230 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
221 set_buffer_uptodate(gdb); 231 set_buffer_uptodate(gdb);
222 unlock_buffer(gdb);
223 err = ext4_handle_dirty_metadata(handle, NULL, gdb); 232 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
224 if (unlikely(err)) { 233 if (unlikely(err)) {
225 brelse(gdb); 234 brelse(gdb);
226 goto exit_bh; 235 goto exit_journal;
227 } 236 }
228 ext4_set_bit(bit, bh->b_data);
229 brelse(gdb); 237 brelse(gdb);
230 } 238 }
231 239
@@ -235,9 +243,22 @@ static int setup_new_group_blocks(struct super_block *sb,
235 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb, 243 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
236 GFP_NOFS); 244 GFP_NOFS);
237 if (err) 245 if (err)
238 goto exit_bh; 246 goto exit_journal;
239 for (i = 0, bit = gdblocks + 1; i < reserved_gdb; i++, bit++) 247
240 ext4_set_bit(bit, bh->b_data); 248 err = extend_or_restart_transaction(handle, 2);
249 if (err)
250 goto exit_journal;
251
252 bh = bclean(handle, sb, input->block_bitmap);
253 if (IS_ERR(bh)) {
254 err = PTR_ERR(bh);
255 goto exit_journal;
256 }
257
258 if (ext4_bg_has_super(sb, input->group)) {
259 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
260 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
261 }
241 262
242 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap, 263 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
243 input->block_bitmap - start); 264 input->block_bitmap - start);
@@ -253,12 +274,9 @@ static int setup_new_group_blocks(struct super_block *sb,
253 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS); 274 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
254 if (err) 275 if (err)
255 goto exit_bh; 276 goto exit_bh;
256 for (i = 0, bit = input->inode_table - start; 277 ext4_set_bits(bh->b_data, input->inode_table - start,
257 i < sbi->s_itb_per_group; i++, bit++) 278 sbi->s_itb_per_group);
258 ext4_set_bit(bit, bh->b_data);
259 279
260 if ((err = extend_or_restart_transaction(handle, 2, bh)))
261 goto exit_bh;
262 280
263 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8, 281 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
264 bh->b_data); 282 bh->b_data);
@@ -285,7 +303,6 @@ exit_bh:
285 brelse(bh); 303 brelse(bh);
286 304
287exit_journal: 305exit_journal:
288 mutex_unlock(&sbi->s_resize_lock);
289 if ((err2 = ext4_journal_stop(handle)) && !err) 306 if ((err2 = ext4_journal_stop(handle)) && !err)
290 err = err2; 307 err = err2;
291 308
@@ -377,15 +394,15 @@ static int verify_reserved_gdb(struct super_block *sb,
377 * fail once we start modifying the data on disk, because JBD has no rollback. 394 * fail once we start modifying the data on disk, because JBD has no rollback.
378 */ 395 */
379static int add_new_gdb(handle_t *handle, struct inode *inode, 396static int add_new_gdb(handle_t *handle, struct inode *inode,
380 struct ext4_new_group_data *input, 397 ext4_group_t group)
381 struct buffer_head **primary)
382{ 398{
383 struct super_block *sb = inode->i_sb; 399 struct super_block *sb = inode->i_sb;
384 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 400 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
385 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb); 401 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
386 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; 402 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
387 struct buffer_head **o_group_desc, **n_group_desc; 403 struct buffer_head **o_group_desc, **n_group_desc;
388 struct buffer_head *dind; 404 struct buffer_head *dind;
405 struct buffer_head *gdb_bh;
389 int gdbackups; 406 int gdbackups;
390 struct ext4_iloc iloc; 407 struct ext4_iloc iloc;
391 __le32 *data; 408 __le32 *data;
@@ -408,11 +425,12 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
408 return -EPERM; 425 return -EPERM;
409 } 426 }
410 427
411 *primary = sb_bread(sb, gdblock); 428 gdb_bh = sb_bread(sb, gdblock);
412 if (!*primary) 429 if (!gdb_bh)
413 return -EIO; 430 return -EIO;
414 431
415 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) { 432 gdbackups = verify_reserved_gdb(sb, gdb_bh);
433 if (gdbackups < 0) {
416 err = gdbackups; 434 err = gdbackups;
417 goto exit_bh; 435 goto exit_bh;
418 } 436 }
@@ -427,7 +445,7 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
427 data = (__le32 *)dind->b_data; 445 data = (__le32 *)dind->b_data;
428 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) { 446 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
429 ext4_warning(sb, "new group %u GDT block %llu not reserved", 447 ext4_warning(sb, "new group %u GDT block %llu not reserved",
430 input->group, gdblock); 448 group, gdblock);
431 err = -EINVAL; 449 err = -EINVAL;
432 goto exit_dind; 450 goto exit_dind;
433 } 451 }
@@ -436,7 +454,7 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
436 if (unlikely(err)) 454 if (unlikely(err))
437 goto exit_dind; 455 goto exit_dind;
438 456
439 err = ext4_journal_get_write_access(handle, *primary); 457 err = ext4_journal_get_write_access(handle, gdb_bh);
440 if (unlikely(err)) 458 if (unlikely(err))
441 goto exit_sbh; 459 goto exit_sbh;
442 460
@@ -449,12 +467,13 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
449 if (unlikely(err)) 467 if (unlikely(err))
450 goto exit_dindj; 468 goto exit_dindj;
451 469
452 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *), 470 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
453 GFP_NOFS); 471 sizeof(struct buffer_head *),
472 GFP_NOFS);
454 if (!n_group_desc) { 473 if (!n_group_desc) {
455 err = -ENOMEM; 474 err = -ENOMEM;
456 ext4_warning(sb, 475 ext4_warning(sb, "not enough memory for %lu groups",
457 "not enough memory for %lu groups", gdb_num + 1); 476 gdb_num + 1);
458 goto exit_inode; 477 goto exit_inode;
459 } 478 }
460 479
@@ -475,8 +494,8 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
475 } 494 }
476 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9; 495 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
477 ext4_mark_iloc_dirty(handle, inode, &iloc); 496 ext4_mark_iloc_dirty(handle, inode, &iloc);
478 memset((*primary)->b_data, 0, sb->s_blocksize); 497 memset(gdb_bh->b_data, 0, sb->s_blocksize);
479 err = ext4_handle_dirty_metadata(handle, NULL, *primary); 498 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
480 if (unlikely(err)) { 499 if (unlikely(err)) {
481 ext4_std_error(sb, err); 500 ext4_std_error(sb, err);
482 goto exit_inode; 501 goto exit_inode;
@@ -486,10 +505,10 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
486 o_group_desc = EXT4_SB(sb)->s_group_desc; 505 o_group_desc = EXT4_SB(sb)->s_group_desc;
487 memcpy(n_group_desc, o_group_desc, 506 memcpy(n_group_desc, o_group_desc,
488 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); 507 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
489 n_group_desc[gdb_num] = *primary; 508 n_group_desc[gdb_num] = gdb_bh;
490 EXT4_SB(sb)->s_group_desc = n_group_desc; 509 EXT4_SB(sb)->s_group_desc = n_group_desc;
491 EXT4_SB(sb)->s_gdb_count++; 510 EXT4_SB(sb)->s_gdb_count++;
492 kfree(o_group_desc); 511 ext4_kvfree(o_group_desc);
493 512
494 le16_add_cpu(&es->s_reserved_gdt_blocks, -1); 513 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
495 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); 514 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
@@ -499,6 +518,7 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
499 return err; 518 return err;
500 519
501exit_inode: 520exit_inode:
521 ext4_kvfree(n_group_desc);
502 /* ext4_handle_release_buffer(handle, iloc.bh); */ 522 /* ext4_handle_release_buffer(handle, iloc.bh); */
503 brelse(iloc.bh); 523 brelse(iloc.bh);
504exit_dindj: 524exit_dindj:
@@ -508,7 +528,7 @@ exit_sbh:
508exit_dind: 528exit_dind:
509 brelse(dind); 529 brelse(dind);
510exit_bh: 530exit_bh:
511 brelse(*primary); 531 brelse(gdb_bh);
512 532
513 ext4_debug("leaving with error %d\n", err); 533 ext4_debug("leaving with error %d\n", err);
514 return err; 534 return err;
@@ -528,7 +548,7 @@ exit_bh:
528 * backup GDT blocks are stored in their reserved primary GDT block. 548 * backup GDT blocks are stored in their reserved primary GDT block.
529 */ 549 */
530static int reserve_backup_gdb(handle_t *handle, struct inode *inode, 550static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
531 struct ext4_new_group_data *input) 551 ext4_group_t group)
532{ 552{
533 struct super_block *sb = inode->i_sb; 553 struct super_block *sb = inode->i_sb;
534 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); 554 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
@@ -599,7 +619,7 @@ static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
599 * Finally we can add each of the reserved backup GDT blocks from 619 * Finally we can add each of the reserved backup GDT blocks from
600 * the new group to its reserved primary GDT block. 620 * the new group to its reserved primary GDT block.
601 */ 621 */
602 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb); 622 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
603 for (i = 0; i < reserved_gdb; i++) { 623 for (i = 0; i < reserved_gdb; i++) {
604 int err2; 624 int err2;
605 data = (__le32 *)primary[i]->b_data; 625 data = (__le32 *)primary[i]->b_data;
@@ -799,13 +819,6 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
799 goto exit_put; 819 goto exit_put;
800 } 820 }
801 821
802 mutex_lock(&sbi->s_resize_lock);
803 if (input->group != sbi->s_groups_count) {
804 ext4_warning(sb, "multiple resizers run on filesystem!");
805 err = -EBUSY;
806 goto exit_journal;
807 }
808
809 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh))) 822 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
810 goto exit_journal; 823 goto exit_journal;
811 824
@@ -820,16 +833,25 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
820 if ((err = ext4_journal_get_write_access(handle, primary))) 833 if ((err = ext4_journal_get_write_access(handle, primary)))
821 goto exit_journal; 834 goto exit_journal;
822 835
823 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) && 836 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group)) {
824 (err = reserve_backup_gdb(handle, inode, input))) 837 err = reserve_backup_gdb(handle, inode, input->group);
838 if (err)
839 goto exit_journal;
840 }
841 } else {
842 /*
843 * Note that we can access new group descriptor block safely
844 * only if add_new_gdb() succeeds.
845 */
846 err = add_new_gdb(handle, inode, input->group);
847 if (err)
825 goto exit_journal; 848 goto exit_journal;
826 } else if ((err = add_new_gdb(handle, inode, input, &primary))) 849 primary = sbi->s_group_desc[gdb_num];
827 goto exit_journal; 850 }
828 851
829 /* 852 /*
830 * OK, now we've set up the new group. Time to make it active. 853 * OK, now we've set up the new group. Time to make it active.
831 * 854 *
832 * We do not lock all allocations via s_resize_lock
833 * so we have to be safe wrt. concurrent accesses the group 855 * so we have to be safe wrt. concurrent accesses the group
834 * data. So we need to be careful to set all of the relevant 856 * data. So we need to be careful to set all of the relevant
835 * group descriptor data etc. *before* we enable the group. 857 * group descriptor data etc. *before* we enable the group.
@@ -886,13 +908,9 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
886 * 908 *
887 * The precise rules we use are: 909 * The precise rules we use are:
888 * 910 *
889 * * Writers of s_groups_count *must* hold s_resize_lock
890 * AND
891 * * Writers must perform a smp_wmb() after updating all dependent 911 * * Writers must perform a smp_wmb() after updating all dependent
892 * data and before modifying the groups count 912 * data and before modifying the groups count
893 * 913 *
894 * * Readers must hold s_resize_lock over the access
895 * OR
896 * * Readers must perform an smp_rmb() after reading the groups count 914 * * Readers must perform an smp_rmb() after reading the groups count
897 * and before reading any dependent data. 915 * and before reading any dependent data.
898 * 916 *
@@ -937,10 +955,9 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
937 ext4_handle_dirty_super(handle, sb); 955 ext4_handle_dirty_super(handle, sb);
938 956
939exit_journal: 957exit_journal:
940 mutex_unlock(&sbi->s_resize_lock);
941 if ((err2 = ext4_journal_stop(handle)) && !err) 958 if ((err2 = ext4_journal_stop(handle)) && !err)
942 err = err2; 959 err = err2;
943 if (!err) { 960 if (!err && primary) {
944 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es, 961 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
945 sizeof(struct ext4_super_block)); 962 sizeof(struct ext4_super_block));
946 update_backups(sb, primary->b_blocknr, primary->b_data, 963 update_backups(sb, primary->b_blocknr, primary->b_data,
@@ -969,16 +986,13 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
969 ext4_grpblk_t add; 986 ext4_grpblk_t add;
970 struct buffer_head *bh; 987 struct buffer_head *bh;
971 handle_t *handle; 988 handle_t *handle;
972 int err; 989 int err, err2;
973 ext4_group_t group; 990 ext4_group_t group;
974 991
975 /* We don't need to worry about locking wrt other resizers just
976 * yet: we're going to revalidate es->s_blocks_count after
977 * taking the s_resize_lock below. */
978 o_blocks_count = ext4_blocks_count(es); 992 o_blocks_count = ext4_blocks_count(es);
979 993
980 if (test_opt(sb, DEBUG)) 994 if (test_opt(sb, DEBUG))
981 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n", 995 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
982 o_blocks_count, n_blocks_count); 996 o_blocks_count, n_blocks_count);
983 997
984 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) 998 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
@@ -995,7 +1009,7 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
995 1009
996 if (n_blocks_count < o_blocks_count) { 1010 if (n_blocks_count < o_blocks_count) {
997 ext4_warning(sb, "can't shrink FS - resize aborted"); 1011 ext4_warning(sb, "can't shrink FS - resize aborted");
998 return -EBUSY; 1012 return -EINVAL;
999 } 1013 }
1000 1014
1001 /* Handle the remaining blocks in the last group only. */ 1015 /* Handle the remaining blocks in the last group only. */
@@ -1038,32 +1052,25 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1038 goto exit_put; 1052 goto exit_put;
1039 } 1053 }
1040 1054
1041 mutex_lock(&EXT4_SB(sb)->s_resize_lock);
1042 if (o_blocks_count != ext4_blocks_count(es)) {
1043 ext4_warning(sb, "multiple resizers run on filesystem!");
1044 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1045 ext4_journal_stop(handle);
1046 err = -EBUSY;
1047 goto exit_put;
1048 }
1049
1050 if ((err = ext4_journal_get_write_access(handle, 1055 if ((err = ext4_journal_get_write_access(handle,
1051 EXT4_SB(sb)->s_sbh))) { 1056 EXT4_SB(sb)->s_sbh))) {
1052 ext4_warning(sb, "error %d on journal write access", err); 1057 ext4_warning(sb, "error %d on journal write access", err);
1053 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1054 ext4_journal_stop(handle); 1058 ext4_journal_stop(handle);
1055 goto exit_put; 1059 goto exit_put;
1056 } 1060 }
1057 ext4_blocks_count_set(es, o_blocks_count + add); 1061 ext4_blocks_count_set(es, o_blocks_count + add);
1058 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1059 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, 1062 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1060 o_blocks_count + add); 1063 o_blocks_count + add);
1061 /* We add the blocks to the bitmap and set the group need init bit */ 1064 /* We add the blocks to the bitmap and set the group need init bit */
1062 ext4_add_groupblocks(handle, sb, o_blocks_count, add); 1065 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1063 ext4_handle_dirty_super(handle, sb); 1066 ext4_handle_dirty_super(handle, sb);
1064 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count, 1067 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1065 o_blocks_count + add); 1068 o_blocks_count + add);
1066 if ((err = ext4_journal_stop(handle))) 1069 err2 = ext4_journal_stop(handle);
1070 if (!err && err2)
1071 err = err2;
1072
1073 if (err)
1067 goto exit_put; 1074 goto exit_put;
1068 1075
1069 if (test_opt(sb, DEBUG)) 1076 if (test_opt(sb, DEBUG))
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index 9ea71aa864b..e2d88baf91d 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -110,6 +110,35 @@ static struct file_system_type ext3_fs_type = {
110#define IS_EXT3_SB(sb) (0) 110#define IS_EXT3_SB(sb) (0)
111#endif 111#endif
112 112
113void *ext4_kvmalloc(size_t size, gfp_t flags)
114{
115 void *ret;
116
117 ret = kmalloc(size, flags);
118 if (!ret)
119 ret = __vmalloc(size, flags, PAGE_KERNEL);
120 return ret;
121}
122
123void *ext4_kvzalloc(size_t size, gfp_t flags)
124{
125 void *ret;
126
127 ret = kmalloc(size, flags);
128 if (!ret)
129 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
130 return ret;
131}
132
133void ext4_kvfree(void *ptr)
134{
135 if (is_vmalloc_addr(ptr))
136 vfree(ptr);
137 else
138 kfree(ptr);
139
140}
141
113ext4_fsblk_t ext4_block_bitmap(struct super_block *sb, 142ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
114 struct ext4_group_desc *bg) 143 struct ext4_group_desc *bg)
115{ 144{
@@ -269,6 +298,7 @@ handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
269 journal_t *journal; 298 journal_t *journal;
270 handle_t *handle; 299 handle_t *handle;
271 300
301 trace_ext4_journal_start(sb, nblocks, _RET_IP_);
272 if (sb->s_flags & MS_RDONLY) 302 if (sb->s_flags & MS_RDONLY)
273 return ERR_PTR(-EROFS); 303 return ERR_PTR(-EROFS);
274 304
@@ -789,11 +819,8 @@ static void ext4_put_super(struct super_block *sb)
789 819
790 for (i = 0; i < sbi->s_gdb_count; i++) 820 for (i = 0; i < sbi->s_gdb_count; i++)
791 brelse(sbi->s_group_desc[i]); 821 brelse(sbi->s_group_desc[i]);
792 kfree(sbi->s_group_desc); 822 ext4_kvfree(sbi->s_group_desc);
793 if (is_vmalloc_addr(sbi->s_flex_groups)) 823 ext4_kvfree(sbi->s_flex_groups);
794 vfree(sbi->s_flex_groups);
795 else
796 kfree(sbi->s_flex_groups);
797 percpu_counter_destroy(&sbi->s_freeblocks_counter); 824 percpu_counter_destroy(&sbi->s_freeblocks_counter);
798 percpu_counter_destroy(&sbi->s_freeinodes_counter); 825 percpu_counter_destroy(&sbi->s_freeinodes_counter);
799 percpu_counter_destroy(&sbi->s_dirs_counter); 826 percpu_counter_destroy(&sbi->s_dirs_counter);
@@ -1976,15 +2003,11 @@ static int ext4_fill_flex_info(struct super_block *sb)
1976 ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) << 2003 ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1977 EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex; 2004 EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1978 size = flex_group_count * sizeof(struct flex_groups); 2005 size = flex_group_count * sizeof(struct flex_groups);
1979 sbi->s_flex_groups = kzalloc(size, GFP_KERNEL); 2006 sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1980 if (sbi->s_flex_groups == NULL) { 2007 if (sbi->s_flex_groups == NULL) {
1981 sbi->s_flex_groups = vzalloc(size); 2008 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1982 if (sbi->s_flex_groups == NULL) { 2009 flex_group_count);
1983 ext4_msg(sb, KERN_ERR, 2010 goto failed;
1984 "not enough memory for %u flex groups",
1985 flex_group_count);
1986 goto failed;
1987 }
1988 } 2011 }
1989 2012
1990 for (i = 0; i < sbi->s_groups_count; i++) { 2013 for (i = 0; i < sbi->s_groups_count; i++) {
@@ -2383,17 +2406,25 @@ static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2383 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride); 2406 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2384 unsigned long stripe_width = 2407 unsigned long stripe_width =
2385 le32_to_cpu(sbi->s_es->s_raid_stripe_width); 2408 le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2409 int ret;
2386 2410
2387 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group) 2411 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2388 return sbi->s_stripe; 2412 ret = sbi->s_stripe;
2389 2413 else if (stripe_width <= sbi->s_blocks_per_group)
2390 if (stripe_width <= sbi->s_blocks_per_group) 2414 ret = stripe_width;
2391 return stripe_width; 2415 else if (stride <= sbi->s_blocks_per_group)
2416 ret = stride;
2417 else
2418 ret = 0;
2392 2419
2393 if (stride <= sbi->s_blocks_per_group) 2420 /*
2394 return stride; 2421 * If the stripe width is 1, this makes no sense and
2422 * we set it to 0 to turn off stripe handling code.
2423 */
2424 if (ret <= 1)
2425 ret = 0;
2395 2426
2396 return 0; 2427 return ret;
2397} 2428}
2398 2429
2399/* sysfs supprt */ 2430/* sysfs supprt */
@@ -3408,8 +3439,9 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3408 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); 3439 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3409 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) / 3440 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3410 EXT4_DESC_PER_BLOCK(sb); 3441 EXT4_DESC_PER_BLOCK(sb);
3411 sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *), 3442 sbi->s_group_desc = ext4_kvmalloc(db_count *
3412 GFP_KERNEL); 3443 sizeof(struct buffer_head *),
3444 GFP_KERNEL);
3413 if (sbi->s_group_desc == NULL) { 3445 if (sbi->s_group_desc == NULL) {
3414 ext4_msg(sb, KERN_ERR, "not enough memory"); 3446 ext4_msg(sb, KERN_ERR, "not enough memory");
3415 goto failed_mount; 3447 goto failed_mount;
@@ -3491,7 +3523,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3491 3523
3492 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ 3524 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3493 mutex_init(&sbi->s_orphan_lock); 3525 mutex_init(&sbi->s_orphan_lock);
3494 mutex_init(&sbi->s_resize_lock); 3526 sbi->s_resize_flags = 0;
3495 3527
3496 sb->s_root = NULL; 3528 sb->s_root = NULL;
3497 3529
@@ -3741,12 +3773,8 @@ failed_mount_wq:
3741 } 3773 }
3742failed_mount3: 3774failed_mount3:
3743 del_timer(&sbi->s_err_report); 3775 del_timer(&sbi->s_err_report);
3744 if (sbi->s_flex_groups) { 3776 if (sbi->s_flex_groups)
3745 if (is_vmalloc_addr(sbi->s_flex_groups)) 3777 ext4_kvfree(sbi->s_flex_groups);
3746 vfree(sbi->s_flex_groups);
3747 else
3748 kfree(sbi->s_flex_groups);
3749 }
3750 percpu_counter_destroy(&sbi->s_freeblocks_counter); 3778 percpu_counter_destroy(&sbi->s_freeblocks_counter);
3751 percpu_counter_destroy(&sbi->s_freeinodes_counter); 3779 percpu_counter_destroy(&sbi->s_freeinodes_counter);
3752 percpu_counter_destroy(&sbi->s_dirs_counter); 3780 percpu_counter_destroy(&sbi->s_dirs_counter);
@@ -3756,7 +3784,7 @@ failed_mount3:
3756failed_mount2: 3784failed_mount2:
3757 for (i = 0; i < db_count; i++) 3785 for (i = 0; i < db_count; i++)
3758 brelse(sbi->s_group_desc[i]); 3786 brelse(sbi->s_group_desc[i]);
3759 kfree(sbi->s_group_desc); 3787 ext4_kvfree(sbi->s_group_desc);
3760failed_mount: 3788failed_mount:
3761 if (sbi->s_proc) { 3789 if (sbi->s_proc) {
3762 remove_proc_entry(sb->s_id, ext4_proc_root); 3790 remove_proc_entry(sb->s_id, ext4_proc_root);
diff --git a/fs/ext4/truncate.h b/fs/ext4/truncate.h
new file mode 100644
index 00000000000..011ba6670d9
--- /dev/null
+++ b/fs/ext4/truncate.h
@@ -0,0 +1,43 @@
1/*
2 * linux/fs/ext4/truncate.h
3 *
4 * Common inline functions needed for truncate support
5 */
6
7/*
8 * Truncate blocks that were not used by write. We have to truncate the
9 * pagecache as well so that corresponding buffers get properly unmapped.
10 */
11static inline void ext4_truncate_failed_write(struct inode *inode)
12{
13 truncate_inode_pages(inode->i_mapping, inode->i_size);
14 ext4_truncate(inode);
15}
16
17/*
18 * Work out how many blocks we need to proceed with the next chunk of a
19 * truncate transaction.
20 */
21static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
22{
23 ext4_lblk_t needed;
24
25 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
26
27 /* Give ourselves just enough room to cope with inodes in which
28 * i_blocks is corrupt: we've seen disk corruptions in the past
29 * which resulted in random data in an inode which looked enough
30 * like a regular file for ext4 to try to delete it. Things
31 * will go a bit crazy if that happens, but at least we should
32 * try not to panic the whole kernel. */
33 if (needed < 2)
34 needed = 2;
35
36 /* But we need to bound the transaction so we don't overflow the
37 * journal. */
38 if (needed > EXT4_MAX_TRANS_DATA)
39 needed = EXT4_MAX_TRANS_DATA;
40
41 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
42}
43
diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c
index 2c62c5aae82..16a698bd906 100644
--- a/fs/jbd2/checkpoint.c
+++ b/fs/jbd2/checkpoint.c
@@ -257,9 +257,12 @@ static void
257__flush_batch(journal_t *journal, int *batch_count) 257__flush_batch(journal_t *journal, int *batch_count)
258{ 258{
259 int i; 259 int i;
260 struct blk_plug plug;
260 261
262 blk_start_plug(&plug);
261 for (i = 0; i < *batch_count; i++) 263 for (i = 0; i < *batch_count; i++)
262 write_dirty_buffer(journal->j_chkpt_bhs[i], WRITE); 264 write_dirty_buffer(journal->j_chkpt_bhs[i], WRITE_SYNC);
265 blk_finish_plug(&plug);
263 266
264 for (i = 0; i < *batch_count; i++) { 267 for (i = 0; i < *batch_count; i++) {
265 struct buffer_head *bh = journal->j_chkpt_bhs[i]; 268 struct buffer_head *bh = journal->j_chkpt_bhs[i];
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index 0dfa5b598e6..f24df13adc4 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -2390,73 +2390,6 @@ static void __exit journal_exit(void)
2390 jbd2_journal_destroy_caches(); 2390 jbd2_journal_destroy_caches();
2391} 2391}
2392 2392
2393/*
2394 * jbd2_dev_to_name is a utility function used by the jbd2 and ext4
2395 * tracing infrastructure to map a dev_t to a device name.
2396 *
2397 * The caller should use rcu_read_lock() in order to make sure the
2398 * device name stays valid until its done with it. We use
2399 * rcu_read_lock() as well to make sure we're safe in case the caller
2400 * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2401 * nested.
2402 */
2403struct devname_cache {
2404 struct rcu_head rcu;
2405 dev_t device;
2406 char devname[BDEVNAME_SIZE];
2407};
2408#define CACHE_SIZE_BITS 6
2409static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2410static DEFINE_SPINLOCK(devname_cache_lock);
2411
2412static void free_devcache(struct rcu_head *rcu)
2413{
2414 kfree(rcu);
2415}
2416
2417const char *jbd2_dev_to_name(dev_t device)
2418{
2419 int i = hash_32(device, CACHE_SIZE_BITS);
2420 char *ret;
2421 struct block_device *bd;
2422 static struct devname_cache *new_dev;
2423
2424 rcu_read_lock();
2425 if (devcache[i] && devcache[i]->device == device) {
2426 ret = devcache[i]->devname;
2427 rcu_read_unlock();
2428 return ret;
2429 }
2430 rcu_read_unlock();
2431
2432 new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2433 if (!new_dev)
2434 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2435 bd = bdget(device);
2436 spin_lock(&devname_cache_lock);
2437 if (devcache[i]) {
2438 if (devcache[i]->device == device) {
2439 kfree(new_dev);
2440 bdput(bd);
2441 ret = devcache[i]->devname;
2442 spin_unlock(&devname_cache_lock);
2443 return ret;
2444 }
2445 call_rcu(&devcache[i]->rcu, free_devcache);
2446 }
2447 devcache[i] = new_dev;
2448 devcache[i]->device = device;
2449 if (bd) {
2450 bdevname(bd, devcache[i]->devname);
2451 bdput(bd);
2452 } else
2453 __bdevname(device, devcache[i]->devname);
2454 ret = devcache[i]->devname;
2455 spin_unlock(&devname_cache_lock);
2456 return ret;
2457}
2458EXPORT_SYMBOL(jbd2_dev_to_name);
2459
2460MODULE_LICENSE("GPL"); 2393MODULE_LICENSE("GPL");
2461module_init(journal_init); 2394module_init(journal_init);
2462module_exit(journal_exit); 2395module_exit(journal_exit);
diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h
index d087c2e7b2a..38f307b8c33 100644
--- a/include/linux/jbd2.h
+++ b/include/linux/jbd2.h
@@ -1329,12 +1329,6 @@ extern int jbd_blocks_per_page(struct inode *inode);
1329#define BUFFER_TRACE2(bh, bh2, info) do {} while (0) 1329#define BUFFER_TRACE2(bh, bh2, info) do {} while (0)
1330#define JBUFFER_TRACE(jh, info) do {} while (0) 1330#define JBUFFER_TRACE(jh, info) do {} while (0)
1331 1331
1332/*
1333 * jbd2_dev_to_name is a utility function used by the jbd2 and ext4
1334 * tracing infrastructure to map a dev_t to a device name.
1335 */
1336extern const char *jbd2_dev_to_name(dev_t device);
1337
1338#endif /* __KERNEL__ */ 1332#endif /* __KERNEL__ */
1339 1333
1340#endif /* _LINUX_JBD2_H */ 1334#endif /* _LINUX_JBD2_H */
diff --git a/include/trace/events/ext4.h b/include/trace/events/ext4.h
index 6363193a341..b50a5473624 100644
--- a/include/trace/events/ext4.h
+++ b/include/trace/events/ext4.h
@@ -23,7 +23,7 @@ TRACE_EVENT(ext4_free_inode,
23 TP_STRUCT__entry( 23 TP_STRUCT__entry(
24 __field( dev_t, dev ) 24 __field( dev_t, dev )
25 __field( ino_t, ino ) 25 __field( ino_t, ino )
26 __field( umode_t, mode ) 26 __field( __u16, mode )
27 __field( uid_t, uid ) 27 __field( uid_t, uid )
28 __field( gid_t, gid ) 28 __field( gid_t, gid )
29 __field( __u64, blocks ) 29 __field( __u64, blocks )
@@ -52,7 +52,7 @@ TRACE_EVENT(ext4_request_inode,
52 TP_STRUCT__entry( 52 TP_STRUCT__entry(
53 __field( dev_t, dev ) 53 __field( dev_t, dev )
54 __field( ino_t, dir ) 54 __field( ino_t, dir )
55 __field( umode_t, mode ) 55 __field( __u16, mode )
56 ), 56 ),
57 57
58 TP_fast_assign( 58 TP_fast_assign(
@@ -75,7 +75,7 @@ TRACE_EVENT(ext4_allocate_inode,
75 __field( dev_t, dev ) 75 __field( dev_t, dev )
76 __field( ino_t, ino ) 76 __field( ino_t, ino )
77 __field( ino_t, dir ) 77 __field( ino_t, dir )
78 __field( umode_t, mode ) 78 __field( __u16, mode )
79 ), 79 ),
80 80
81 TP_fast_assign( 81 TP_fast_assign(
@@ -725,7 +725,7 @@ TRACE_EVENT(ext4_free_blocks,
725 TP_STRUCT__entry( 725 TP_STRUCT__entry(
726 __field( dev_t, dev ) 726 __field( dev_t, dev )
727 __field( ino_t, ino ) 727 __field( ino_t, ino )
728 __field( umode_t, mode ) 728 __field( __u16, mode )
729 __field( __u64, block ) 729 __field( __u64, block )
730 __field( unsigned long, count ) 730 __field( unsigned long, count )
731 __field( int, flags ) 731 __field( int, flags )
@@ -1012,7 +1012,7 @@ TRACE_EVENT(ext4_forget,
1012 TP_STRUCT__entry( 1012 TP_STRUCT__entry(
1013 __field( dev_t, dev ) 1013 __field( dev_t, dev )
1014 __field( ino_t, ino ) 1014 __field( ino_t, ino )
1015 __field( umode_t, mode ) 1015 __field( __u16, mode )
1016 __field( int, is_metadata ) 1016 __field( int, is_metadata )
1017 __field( __u64, block ) 1017 __field( __u64, block )
1018 ), 1018 ),
@@ -1039,7 +1039,7 @@ TRACE_EVENT(ext4_da_update_reserve_space,
1039 TP_STRUCT__entry( 1039 TP_STRUCT__entry(
1040 __field( dev_t, dev ) 1040 __field( dev_t, dev )
1041 __field( ino_t, ino ) 1041 __field( ino_t, ino )
1042 __field( umode_t, mode ) 1042 __field( __u16, mode )
1043 __field( __u64, i_blocks ) 1043 __field( __u64, i_blocks )
1044 __field( int, used_blocks ) 1044 __field( int, used_blocks )
1045 __field( int, reserved_data_blocks ) 1045 __field( int, reserved_data_blocks )
@@ -1076,7 +1076,7 @@ TRACE_EVENT(ext4_da_reserve_space,
1076 TP_STRUCT__entry( 1076 TP_STRUCT__entry(
1077 __field( dev_t, dev ) 1077 __field( dev_t, dev )
1078 __field( ino_t, ino ) 1078 __field( ino_t, ino )
1079 __field( umode_t, mode ) 1079 __field( __u16, mode )
1080 __field( __u64, i_blocks ) 1080 __field( __u64, i_blocks )
1081 __field( int, md_needed ) 1081 __field( int, md_needed )
1082 __field( int, reserved_data_blocks ) 1082 __field( int, reserved_data_blocks )
@@ -1110,7 +1110,7 @@ TRACE_EVENT(ext4_da_release_space,
1110 TP_STRUCT__entry( 1110 TP_STRUCT__entry(
1111 __field( dev_t, dev ) 1111 __field( dev_t, dev )
1112 __field( ino_t, ino ) 1112 __field( ino_t, ino )
1113 __field( umode_t, mode ) 1113 __field( __u16, mode )
1114 __field( __u64, i_blocks ) 1114 __field( __u64, i_blocks )
1115 __field( int, freed_blocks ) 1115 __field( int, freed_blocks )
1116 __field( int, reserved_data_blocks ) 1116 __field( int, reserved_data_blocks )
@@ -1518,6 +1518,77 @@ TRACE_EVENT(ext4_load_inode,
1518 (unsigned long) __entry->ino) 1518 (unsigned long) __entry->ino)
1519); 1519);
1520 1520
1521TRACE_EVENT(ext4_journal_start,
1522 TP_PROTO(struct super_block *sb, int nblocks, unsigned long IP),
1523
1524 TP_ARGS(sb, nblocks, IP),
1525
1526 TP_STRUCT__entry(
1527 __field( dev_t, dev )
1528 __field( int, nblocks )
1529 __field(unsigned long, ip )
1530 ),
1531
1532 TP_fast_assign(
1533 __entry->dev = sb->s_dev;
1534 __entry->nblocks = nblocks;
1535 __entry->ip = IP;
1536 ),
1537
1538 TP_printk("dev %d,%d nblocks %d caller %pF",
1539 MAJOR(__entry->dev), MINOR(__entry->dev),
1540 __entry->nblocks, (void *)__entry->ip)
1541);
1542
1543DECLARE_EVENT_CLASS(ext4__trim,
1544 TP_PROTO(struct super_block *sb,
1545 ext4_group_t group,
1546 ext4_grpblk_t start,
1547 ext4_grpblk_t len),
1548
1549 TP_ARGS(sb, group, start, len),
1550
1551 TP_STRUCT__entry(
1552 __field( int, dev_major )
1553 __field( int, dev_minor )
1554 __field( __u32, group )
1555 __field( int, start )
1556 __field( int, len )
1557 ),
1558
1559 TP_fast_assign(
1560 __entry->dev_major = MAJOR(sb->s_dev);
1561 __entry->dev_minor = MINOR(sb->s_dev);
1562 __entry->group = group;
1563 __entry->start = start;
1564 __entry->len = len;
1565 ),
1566
1567 TP_printk("dev %d,%d group %u, start %d, len %d",
1568 __entry->dev_major, __entry->dev_minor,
1569 __entry->group, __entry->start, __entry->len)
1570);
1571
1572DEFINE_EVENT(ext4__trim, ext4_trim_extent,
1573
1574 TP_PROTO(struct super_block *sb,
1575 ext4_group_t group,
1576 ext4_grpblk_t start,
1577 ext4_grpblk_t len),
1578
1579 TP_ARGS(sb, group, start, len)
1580);
1581
1582DEFINE_EVENT(ext4__trim, ext4_trim_all_free,
1583
1584 TP_PROTO(struct super_block *sb,
1585 ext4_group_t group,
1586 ext4_grpblk_t start,
1587 ext4_grpblk_t len),
1588
1589 TP_ARGS(sb, group, start, len)
1590);
1591
1521#endif /* _TRACE_EXT4_H */ 1592#endif /* _TRACE_EXT4_H */
1522 1593
1523/* This part must be outside protection */ 1594/* This part must be outside protection */
diff --git a/include/trace/events/jbd2.h b/include/trace/events/jbd2.h
index bf16545cc97..75964412ddb 100644
--- a/include/trace/events/jbd2.h
+++ b/include/trace/events/jbd2.h
@@ -26,8 +26,8 @@ TRACE_EVENT(jbd2_checkpoint,
26 __entry->result = result; 26 __entry->result = result;
27 ), 27 ),
28 28
29 TP_printk("dev %s result %d", 29 TP_printk("dev %d,%d result %d",
30 jbd2_dev_to_name(__entry->dev), __entry->result) 30 MAJOR(__entry->dev), MINOR(__entry->dev), __entry->result)
31); 31);
32 32
33DECLARE_EVENT_CLASS(jbd2_commit, 33DECLARE_EVENT_CLASS(jbd2_commit,
@@ -48,9 +48,9 @@ DECLARE_EVENT_CLASS(jbd2_commit,
48 __entry->transaction = commit_transaction->t_tid; 48 __entry->transaction = commit_transaction->t_tid;
49 ), 49 ),
50 50
51 TP_printk("dev %s transaction %d sync %d", 51 TP_printk("dev %d,%d transaction %d sync %d",
52 jbd2_dev_to_name(__entry->dev), __entry->transaction, 52 MAJOR(__entry->dev), MINOR(__entry->dev),
53 __entry->sync_commit) 53 __entry->transaction, __entry->sync_commit)
54); 54);
55 55
56DEFINE_EVENT(jbd2_commit, jbd2_start_commit, 56DEFINE_EVENT(jbd2_commit, jbd2_start_commit,
@@ -100,9 +100,9 @@ TRACE_EVENT(jbd2_end_commit,
100 __entry->head = journal->j_tail_sequence; 100 __entry->head = journal->j_tail_sequence;
101 ), 101 ),
102 102
103 TP_printk("dev %s transaction %d sync %d head %d", 103 TP_printk("dev %d,%d transaction %d sync %d head %d",
104 jbd2_dev_to_name(__entry->dev), __entry->transaction, 104 MAJOR(__entry->dev), MINOR(__entry->dev),
105 __entry->sync_commit, __entry->head) 105 __entry->transaction, __entry->sync_commit, __entry->head)
106); 106);
107 107
108TRACE_EVENT(jbd2_submit_inode_data, 108TRACE_EVENT(jbd2_submit_inode_data,
@@ -120,8 +120,9 @@ TRACE_EVENT(jbd2_submit_inode_data,
120 __entry->ino = inode->i_ino; 120 __entry->ino = inode->i_ino;
121 ), 121 ),
122 122
123 TP_printk("dev %s ino %lu", 123 TP_printk("dev %d,%d ino %lu",
124 jbd2_dev_to_name(__entry->dev), (unsigned long) __entry->ino) 124 MAJOR(__entry->dev), MINOR(__entry->dev),
125 (unsigned long) __entry->ino)
125); 126);
126 127
127TRACE_EVENT(jbd2_run_stats, 128TRACE_EVENT(jbd2_run_stats,
@@ -156,9 +157,9 @@ TRACE_EVENT(jbd2_run_stats,
156 __entry->blocks_logged = stats->rs_blocks_logged; 157 __entry->blocks_logged = stats->rs_blocks_logged;
157 ), 158 ),
158 159
159 TP_printk("dev %s tid %lu wait %u running %u locked %u flushing %u " 160 TP_printk("dev %d,%d tid %lu wait %u running %u locked %u flushing %u "
160 "logging %u handle_count %u blocks %u blocks_logged %u", 161 "logging %u handle_count %u blocks %u blocks_logged %u",
161 jbd2_dev_to_name(__entry->dev), __entry->tid, 162 MAJOR(__entry->dev), MINOR(__entry->dev), __entry->tid,
162 jiffies_to_msecs(__entry->wait), 163 jiffies_to_msecs(__entry->wait),
163 jiffies_to_msecs(__entry->running), 164 jiffies_to_msecs(__entry->running),
164 jiffies_to_msecs(__entry->locked), 165 jiffies_to_msecs(__entry->locked),
@@ -192,9 +193,9 @@ TRACE_EVENT(jbd2_checkpoint_stats,
192 __entry->dropped = stats->cs_dropped; 193 __entry->dropped = stats->cs_dropped;
193 ), 194 ),
194 195
195 TP_printk("dev %s tid %lu chp_time %u forced_to_close %u " 196 TP_printk("dev %d,%d tid %lu chp_time %u forced_to_close %u "
196 "written %u dropped %u", 197 "written %u dropped %u",
197 jbd2_dev_to_name(__entry->dev), __entry->tid, 198 MAJOR(__entry->dev), MINOR(__entry->dev), __entry->tid,
198 jiffies_to_msecs(__entry->chp_time), 199 jiffies_to_msecs(__entry->chp_time),
199 __entry->forced_to_close, __entry->written, __entry->dropped) 200 __entry->forced_to_close, __entry->written, __entry->dropped)
200); 201);
@@ -222,9 +223,10 @@ TRACE_EVENT(jbd2_cleanup_journal_tail,
222 __entry->freed = freed; 223 __entry->freed = freed;
223 ), 224 ),
224 225
225 TP_printk("dev %s from %u to %u offset %lu freed %lu", 226 TP_printk("dev %d,%d from %u to %u offset %lu freed %lu",
226 jbd2_dev_to_name(__entry->dev), __entry->tail_sequence, 227 MAJOR(__entry->dev), MINOR(__entry->dev),
227 __entry->first_tid, __entry->block_nr, __entry->freed) 228 __entry->tail_sequence, __entry->first_tid,
229 __entry->block_nr, __entry->freed)
228); 230);
229 231
230#endif /* _TRACE_JBD2_H */ 232#endif /* _TRACE_JBD2_H */
generated by cgit v1.2.2 (git 2.25.0) at 2024-10-19 09:31:22 -0400