aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ext4
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ext4')
-rw-r--r--fs/ext4/Makefile2
-rw-r--r--fs/ext4/balloc.c5
-rw-r--r--fs/ext4/block_validity.c7
-rw-r--r--fs/ext4/dir.c2
-rw-r--r--fs/ext4/ext4.h112
-rw-r--r--fs/ext4/ext4_extents.h65
-rw-r--r--fs/ext4/extents.c368
-rw-r--r--fs/ext4/file.c44
-rw-r--r--fs/ext4/fsync.c83
-rw-r--r--fs/ext4/ialloc.c135
-rw-r--r--fs/ext4/inode.c599
-rw-r--r--fs/ext4/ioctl.c24
-rw-r--r--fs/ext4/mballoc.c553
-rw-r--r--fs/ext4/migrate.c2
-rw-r--r--fs/ext4/move_extent.c22
-rw-r--r--fs/ext4/namei.c65
-rw-r--r--fs/ext4/page-io.c431
-rw-r--r--fs/ext4/resize.c52
-rw-r--r--fs/ext4/super.c620
-rw-r--r--fs/ext4/xattr.c4
-rw-r--r--fs/ext4/xattr.h10
21 files changed, 2215 insertions, 990 deletions
diff --git a/fs/ext4/Makefile b/fs/ext4/Makefile
index 8867b2a1e5fe..c947e36eda6c 100644
--- a/fs/ext4/Makefile
+++ b/fs/ext4/Makefile
@@ -4,7 +4,7 @@
4 4
5obj-$(CONFIG_EXT4_FS) += ext4.o 5obj-$(CONFIG_EXT4_FS) += ext4.o
6 6
7ext4-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.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 10
diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index bd30799a43ed..14c3af26c671 100644
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -171,7 +171,8 @@ unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
171 * less than the blocksize * 8 ( which is the size 171 * less than the blocksize * 8 ( which is the size
172 * of bitmap ), set rest of the block bitmap to 1 172 * of bitmap ), set rest of the block bitmap to 1
173 */ 173 */
174 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data); 174 ext4_mark_bitmap_end(group_blocks, sb->s_blocksize * 8,
175 bh->b_data);
175 } 176 }
176 return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp); 177 return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
177} 178}
@@ -489,7 +490,7 @@ error_return:
489 * Check if filesystem has nblocks free & available for allocation. 490 * Check if filesystem has nblocks free & available for allocation.
490 * On success return 1, return 0 on failure. 491 * On success return 1, return 0 on failure.
491 */ 492 */
492int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks) 493static int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
493{ 494{
494 s64 free_blocks, dirty_blocks, root_blocks; 495 s64 free_blocks, dirty_blocks, root_blocks;
495 struct percpu_counter *fbc = &sbi->s_freeblocks_counter; 496 struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
diff --git a/fs/ext4/block_validity.c b/fs/ext4/block_validity.c
index 3db5084db9bd..fac90f3fba80 100644
--- a/fs/ext4/block_validity.c
+++ b/fs/ext4/block_validity.c
@@ -29,16 +29,15 @@ struct ext4_system_zone {
29 29
30static struct kmem_cache *ext4_system_zone_cachep; 30static struct kmem_cache *ext4_system_zone_cachep;
31 31
32int __init init_ext4_system_zone(void) 32int __init ext4_init_system_zone(void)
33{ 33{
34 ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 34 ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
35 SLAB_RECLAIM_ACCOUNT);
36 if (ext4_system_zone_cachep == NULL) 35 if (ext4_system_zone_cachep == NULL)
37 return -ENOMEM; 36 return -ENOMEM;
38 return 0; 37 return 0;
39} 38}
40 39
41void exit_ext4_system_zone(void) 40void ext4_exit_system_zone(void)
42{ 41{
43 kmem_cache_destroy(ext4_system_zone_cachep); 42 kmem_cache_destroy(ext4_system_zone_cachep);
44} 43}
diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
index 374510f72baa..ece76fb6a40c 100644
--- a/fs/ext4/dir.c
+++ b/fs/ext4/dir.c
@@ -39,7 +39,7 @@ static int ext4_release_dir(struct inode *inode,
39 struct file *filp); 39 struct file *filp);
40 40
41const struct file_operations ext4_dir_operations = { 41const struct file_operations ext4_dir_operations = {
42 .llseek = generic_file_llseek, 42 .llseek = ext4_llseek,
43 .read = generic_read_dir, 43 .read = generic_read_dir,
44 .readdir = ext4_readdir, /* we take BKL. needed?*/ 44 .readdir = ext4_readdir, /* we take BKL. needed?*/
45 .unlocked_ioctl = ext4_ioctl, 45 .unlocked_ioctl = ext4_ioctl,
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 889ec9d5e6ad..6a5edea2d70b 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -168,7 +168,20 @@ struct mpage_da_data {
168 int pages_written; 168 int pages_written;
169 int retval; 169 int retval;
170}; 170};
171#define EXT4_IO_UNWRITTEN 0x1 171
172/*
173 * Flags for ext4_io_end->flags
174 */
175#define EXT4_IO_END_UNWRITTEN 0x0001
176#define EXT4_IO_END_ERROR 0x0002
177
178struct ext4_io_page {
179 struct page *p_page;
180 atomic_t p_count;
181};
182
183#define MAX_IO_PAGES 128
184
172typedef struct ext4_io_end { 185typedef struct ext4_io_end {
173 struct list_head list; /* per-file finished IO list */ 186 struct list_head list; /* per-file finished IO list */
174 struct inode *inode; /* file being written to */ 187 struct inode *inode; /* file being written to */
@@ -179,8 +192,18 @@ typedef struct ext4_io_end {
179 struct work_struct work; /* data work queue */ 192 struct work_struct work; /* data work queue */
180 struct kiocb *iocb; /* iocb struct for AIO */ 193 struct kiocb *iocb; /* iocb struct for AIO */
181 int result; /* error value for AIO */ 194 int result; /* error value for AIO */
195 int num_io_pages;
196 struct ext4_io_page *pages[MAX_IO_PAGES];
182} ext4_io_end_t; 197} ext4_io_end_t;
183 198
199struct ext4_io_submit {
200 int io_op;
201 struct bio *io_bio;
202 ext4_io_end_t *io_end;
203 struct ext4_io_page *io_page;
204 sector_t io_next_block;
205};
206
184/* 207/*
185 * Special inodes numbers 208 * Special inodes numbers
186 */ 209 */
@@ -205,6 +228,7 @@ typedef struct ext4_io_end {
205#define EXT4_MIN_BLOCK_SIZE 1024 228#define EXT4_MIN_BLOCK_SIZE 1024
206#define EXT4_MAX_BLOCK_SIZE 65536 229#define EXT4_MAX_BLOCK_SIZE 65536
207#define EXT4_MIN_BLOCK_LOG_SIZE 10 230#define EXT4_MIN_BLOCK_LOG_SIZE 10
231#define EXT4_MAX_BLOCK_LOG_SIZE 16
208#ifdef __KERNEL__ 232#ifdef __KERNEL__
209# define EXT4_BLOCK_SIZE(s) ((s)->s_blocksize) 233# define EXT4_BLOCK_SIZE(s) ((s)->s_blocksize)
210#else 234#else
@@ -834,6 +858,7 @@ struct ext4_inode_info {
834 spinlock_t i_completed_io_lock; 858 spinlock_t i_completed_io_lock;
835 /* current io_end structure for async DIO write*/ 859 /* current io_end structure for async DIO write*/
836 ext4_io_end_t *cur_aio_dio; 860 ext4_io_end_t *cur_aio_dio;
861 atomic_t i_ioend_count; /* Number of outstanding io_end structs */
837 862
838 /* 863 /*
839 * Transactions that contain inode's metadata needed to complete 864 * Transactions that contain inode's metadata needed to complete
@@ -889,6 +914,7 @@ struct ext4_inode_info {
889#define EXT4_MOUNT_DATA_ERR_ABORT 0x10000000 /* Abort on file data write */ 914#define EXT4_MOUNT_DATA_ERR_ABORT 0x10000000 /* Abort on file data write */
890#define EXT4_MOUNT_BLOCK_VALIDITY 0x20000000 /* Block validity checking */ 915#define EXT4_MOUNT_BLOCK_VALIDITY 0x20000000 /* Block validity checking */
891#define EXT4_MOUNT_DISCARD 0x40000000 /* Issue DISCARD requests */ 916#define EXT4_MOUNT_DISCARD 0x40000000 /* Issue DISCARD requests */
917#define EXT4_MOUNT_INIT_INODE_TABLE 0x80000000 /* Initialize uninitialized itables */
892 918
893#define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt 919#define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt
894#define set_opt(o, opt) o |= EXT4_MOUNT_##opt 920#define set_opt(o, opt) o |= EXT4_MOUNT_##opt
@@ -1087,7 +1113,6 @@ struct ext4_sb_info {
1087 struct completion s_kobj_unregister; 1113 struct completion s_kobj_unregister;
1088 1114
1089 /* Journaling */ 1115 /* Journaling */
1090 struct inode *s_journal_inode;
1091 struct journal_s *s_journal; 1116 struct journal_s *s_journal;
1092 struct list_head s_orphan; 1117 struct list_head s_orphan;
1093 struct mutex s_orphan_lock; 1118 struct mutex s_orphan_lock;
@@ -1120,10 +1145,7 @@ struct ext4_sb_info {
1120 /* for buddy allocator */ 1145 /* for buddy allocator */
1121 struct ext4_group_info ***s_group_info; 1146 struct ext4_group_info ***s_group_info;
1122 struct inode *s_buddy_cache; 1147 struct inode *s_buddy_cache;
1123 long s_blocks_reserved;
1124 spinlock_t s_reserve_lock;
1125 spinlock_t s_md_lock; 1148 spinlock_t s_md_lock;
1126 tid_t s_last_transaction;
1127 unsigned short *s_mb_offsets; 1149 unsigned short *s_mb_offsets;
1128 unsigned int *s_mb_maxs; 1150 unsigned int *s_mb_maxs;
1129 1151
@@ -1141,7 +1163,6 @@ struct ext4_sb_info {
1141 unsigned long s_mb_last_start; 1163 unsigned long s_mb_last_start;
1142 1164
1143 /* stats for buddy allocator */ 1165 /* stats for buddy allocator */
1144 spinlock_t s_mb_pa_lock;
1145 atomic_t s_bal_reqs; /* number of reqs with len > 1 */ 1166 atomic_t s_bal_reqs; /* number of reqs with len > 1 */
1146 atomic_t s_bal_success; /* we found long enough chunks */ 1167 atomic_t s_bal_success; /* we found long enough chunks */
1147 atomic_t s_bal_allocated; /* in blocks */ 1168 atomic_t s_bal_allocated; /* in blocks */
@@ -1172,6 +1193,11 @@ struct ext4_sb_info {
1172 1193
1173 /* timer for periodic error stats printing */ 1194 /* timer for periodic error stats printing */
1174 struct timer_list s_err_report; 1195 struct timer_list s_err_report;
1196
1197 /* Lazy inode table initialization info */
1198 struct ext4_li_request *s_li_request;
1199 /* Wait multiplier for lazy initialization thread */
1200 unsigned int s_li_wait_mult;
1175}; 1201};
1176 1202
1177static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb) 1203static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb)
@@ -1533,7 +1559,42 @@ ext4_group_first_block_no(struct super_block *sb, ext4_group_t group_no)
1533void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr, 1559void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
1534 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp); 1560 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp);
1535 1561
1536extern struct proc_dir_entry *ext4_proc_root; 1562/*
1563 * Timeout and state flag for lazy initialization inode thread.
1564 */
1565#define EXT4_DEF_LI_WAIT_MULT 10
1566#define EXT4_DEF_LI_MAX_START_DELAY 5
1567#define EXT4_LAZYINIT_QUIT 0x0001
1568#define EXT4_LAZYINIT_RUNNING 0x0002
1569
1570/*
1571 * Lazy inode table initialization info
1572 */
1573struct ext4_lazy_init {
1574 unsigned long li_state;
1575
1576 wait_queue_head_t li_wait_daemon;
1577 wait_queue_head_t li_wait_task;
1578 struct timer_list li_timer;
1579 struct task_struct *li_task;
1580
1581 struct list_head li_request_list;
1582 struct mutex li_list_mtx;
1583};
1584
1585struct ext4_li_request {
1586 struct super_block *lr_super;
1587 struct ext4_sb_info *lr_sbi;
1588 ext4_group_t lr_next_group;
1589 struct list_head lr_request;
1590 unsigned long lr_next_sched;
1591 unsigned long lr_timeout;
1592};
1593
1594struct ext4_features {
1595 struct kobject f_kobj;
1596 struct completion f_kobj_unregister;
1597};
1537 1598
1538/* 1599/*
1539 * Function prototypes 1600 * Function prototypes
@@ -1561,7 +1622,6 @@ extern unsigned long ext4_bg_num_gdb(struct super_block *sb,
1561extern ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode, 1622extern ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
1562 ext4_fsblk_t goal, unsigned long *count, int *errp); 1623 ext4_fsblk_t goal, unsigned long *count, int *errp);
1563extern int ext4_claim_free_blocks(struct ext4_sb_info *sbi, s64 nblocks); 1624extern int ext4_claim_free_blocks(struct ext4_sb_info *sbi, s64 nblocks);
1564extern int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks);
1565extern void ext4_add_groupblocks(handle_t *handle, struct super_block *sb, 1625extern void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
1566 ext4_fsblk_t block, unsigned long count); 1626 ext4_fsblk_t block, unsigned long count);
1567extern ext4_fsblk_t ext4_count_free_blocks(struct super_block *); 1627extern ext4_fsblk_t ext4_count_free_blocks(struct super_block *);
@@ -1605,11 +1665,9 @@ extern struct inode * ext4_orphan_get(struct super_block *, unsigned long);
1605extern unsigned long ext4_count_free_inodes(struct super_block *); 1665extern unsigned long ext4_count_free_inodes(struct super_block *);
1606extern unsigned long ext4_count_dirs(struct super_block *); 1666extern unsigned long ext4_count_dirs(struct super_block *);
1607extern void ext4_check_inodes_bitmap(struct super_block *); 1667extern void ext4_check_inodes_bitmap(struct super_block *);
1608extern unsigned ext4_init_inode_bitmap(struct super_block *sb, 1668extern void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap);
1609 struct buffer_head *bh, 1669extern int ext4_init_inode_table(struct super_block *sb,
1610 ext4_group_t group, 1670 ext4_group_t group, int barrier);
1611 struct ext4_group_desc *desc);
1612extern void mark_bitmap_end(int start_bit, int end_bit, char *bitmap);
1613 1671
1614/* mballoc.c */ 1672/* mballoc.c */
1615extern long ext4_mb_stats; 1673extern long ext4_mb_stats;
@@ -1620,16 +1678,15 @@ extern ext4_fsblk_t ext4_mb_new_blocks(handle_t *,
1620 struct ext4_allocation_request *, int *); 1678 struct ext4_allocation_request *, int *);
1621extern int ext4_mb_reserve_blocks(struct super_block *, int); 1679extern int ext4_mb_reserve_blocks(struct super_block *, int);
1622extern void ext4_discard_preallocations(struct inode *); 1680extern void ext4_discard_preallocations(struct inode *);
1623extern int __init init_ext4_mballoc(void); 1681extern int __init ext4_init_mballoc(void);
1624extern void exit_ext4_mballoc(void); 1682extern void ext4_exit_mballoc(void);
1625extern void ext4_free_blocks(handle_t *handle, struct inode *inode, 1683extern void ext4_free_blocks(handle_t *handle, struct inode *inode,
1626 struct buffer_head *bh, ext4_fsblk_t block, 1684 struct buffer_head *bh, ext4_fsblk_t block,
1627 unsigned long count, int flags); 1685 unsigned long count, int flags);
1628extern int ext4_mb_add_groupinfo(struct super_block *sb, 1686extern int ext4_mb_add_groupinfo(struct super_block *sb,
1629 ext4_group_t i, struct ext4_group_desc *desc); 1687 ext4_group_t i, struct ext4_group_desc *desc);
1630extern int ext4_mb_get_buddy_cache_lock(struct super_block *, ext4_group_t); 1688extern int ext4_trim_fs(struct super_block *, struct fstrim_range *);
1631extern void ext4_mb_put_buddy_cache_lock(struct super_block *, 1689
1632 ext4_group_t, int);
1633/* inode.c */ 1690/* inode.c */
1634struct buffer_head *ext4_getblk(handle_t *, struct inode *, 1691struct buffer_head *ext4_getblk(handle_t *, struct inode *,
1635 ext4_lblk_t, int, int *); 1692 ext4_lblk_t, int, int *);
@@ -1657,13 +1714,11 @@ extern void ext4_get_inode_flags(struct ext4_inode_info *);
1657extern int ext4_alloc_da_blocks(struct inode *inode); 1714extern int ext4_alloc_da_blocks(struct inode *inode);
1658extern void ext4_set_aops(struct inode *inode); 1715extern void ext4_set_aops(struct inode *inode);
1659extern int ext4_writepage_trans_blocks(struct inode *); 1716extern int ext4_writepage_trans_blocks(struct inode *);
1660extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int idxblocks);
1661extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks); 1717extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
1662extern int ext4_block_truncate_page(handle_t *handle, 1718extern int ext4_block_truncate_page(handle_t *handle,
1663 struct address_space *mapping, loff_t from); 1719 struct address_space *mapping, loff_t from);
1664extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf); 1720extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf);
1665extern qsize_t *ext4_get_reserved_space(struct inode *inode); 1721extern qsize_t *ext4_get_reserved_space(struct inode *inode);
1666extern int flush_completed_IO(struct inode *inode);
1667extern void ext4_da_update_reserve_space(struct inode *inode, 1722extern void ext4_da_update_reserve_space(struct inode *inode,
1668 int used, int quota_claim); 1723 int used, int quota_claim);
1669/* ioctl.c */ 1724/* ioctl.c */
@@ -1960,6 +2015,7 @@ extern const struct file_operations ext4_dir_operations;
1960/* file.c */ 2015/* file.c */
1961extern const struct inode_operations ext4_file_inode_operations; 2016extern const struct inode_operations ext4_file_inode_operations;
1962extern const struct file_operations ext4_file_operations; 2017extern const struct file_operations ext4_file_operations;
2018extern loff_t ext4_llseek(struct file *file, loff_t offset, int origin);
1963 2019
1964/* namei.c */ 2020/* namei.c */
1965extern const struct inode_operations ext4_dir_inode_operations; 2021extern const struct inode_operations ext4_dir_inode_operations;
@@ -1973,8 +2029,8 @@ extern const struct inode_operations ext4_fast_symlink_inode_operations;
1973/* block_validity */ 2029/* block_validity */
1974extern void ext4_release_system_zone(struct super_block *sb); 2030extern void ext4_release_system_zone(struct super_block *sb);
1975extern int ext4_setup_system_zone(struct super_block *sb); 2031extern int ext4_setup_system_zone(struct super_block *sb);
1976extern int __init init_ext4_system_zone(void); 2032extern int __init ext4_init_system_zone(void);
1977extern void exit_ext4_system_zone(void); 2033extern void ext4_exit_system_zone(void);
1978extern int ext4_data_block_valid(struct ext4_sb_info *sbi, 2034extern int ext4_data_block_valid(struct ext4_sb_info *sbi,
1979 ext4_fsblk_t start_blk, 2035 ext4_fsblk_t start_blk,
1980 unsigned int count); 2036 unsigned int count);
@@ -2002,6 +2058,18 @@ extern int ext4_move_extents(struct file *o_filp, struct file *d_filp,
2002 __u64 start_orig, __u64 start_donor, 2058 __u64 start_orig, __u64 start_donor,
2003 __u64 len, __u64 *moved_len); 2059 __u64 len, __u64 *moved_len);
2004 2060
2061/* page-io.c */
2062extern int __init ext4_init_pageio(void);
2063extern void ext4_exit_pageio(void);
2064extern void ext4_ioend_wait(struct inode *);
2065extern void ext4_free_io_end(ext4_io_end_t *io);
2066extern ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags);
2067extern int ext4_end_io_nolock(ext4_io_end_t *io);
2068extern void ext4_io_submit(struct ext4_io_submit *io);
2069extern int ext4_bio_write_page(struct ext4_io_submit *io,
2070 struct page *page,
2071 int len,
2072 struct writeback_control *wbc);
2005 2073
2006/* BH_Uninit flag: blocks are allocated but uninitialized on disk */ 2074/* BH_Uninit flag: blocks are allocated but uninitialized on disk */
2007enum ext4_state_bits { 2075enum ext4_state_bits {
diff --git a/fs/ext4/ext4_extents.h b/fs/ext4/ext4_extents.h
index bdb6ce7e2eb4..28ce70fd9cd0 100644
--- a/fs/ext4/ext4_extents.h
+++ b/fs/ext4/ext4_extents.h
@@ -225,11 +225,60 @@ static inline void ext4_ext_mark_initialized(struct ext4_extent *ext)
225 ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext)); 225 ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext));
226} 226}
227 227
228/*
229 * ext4_ext_pblock:
230 * combine low and high parts of physical block number into ext4_fsblk_t
231 */
232static inline ext4_fsblk_t ext4_ext_pblock(struct ext4_extent *ex)
233{
234 ext4_fsblk_t block;
235
236 block = le32_to_cpu(ex->ee_start_lo);
237 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
238 return block;
239}
240
241/*
242 * ext4_idx_pblock:
243 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
244 */
245static inline ext4_fsblk_t ext4_idx_pblock(struct ext4_extent_idx *ix)
246{
247 ext4_fsblk_t block;
248
249 block = le32_to_cpu(ix->ei_leaf_lo);
250 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
251 return block;
252}
253
254/*
255 * ext4_ext_store_pblock:
256 * stores a large physical block number into an extent struct,
257 * breaking it into parts
258 */
259static inline void ext4_ext_store_pblock(struct ext4_extent *ex,
260 ext4_fsblk_t pb)
261{
262 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
263 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
264 0xffff);
265}
266
267/*
268 * ext4_idx_store_pblock:
269 * stores a large physical block number into an index struct,
270 * breaking it into parts
271 */
272static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix,
273 ext4_fsblk_t pb)
274{
275 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
276 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
277 0xffff);
278}
279
228extern int ext4_ext_calc_metadata_amount(struct inode *inode, 280extern int ext4_ext_calc_metadata_amount(struct inode *inode,
229 sector_t lblocks); 281 sector_t lblocks);
230extern ext4_fsblk_t ext_pblock(struct ext4_extent *ex);
231extern ext4_fsblk_t idx_pblock(struct ext4_extent_idx *);
232extern void ext4_ext_store_pblock(struct ext4_extent *, ext4_fsblk_t);
233extern int ext4_extent_tree_init(handle_t *, struct inode *); 282extern int ext4_extent_tree_init(handle_t *, struct inode *);
234extern int ext4_ext_calc_credits_for_single_extent(struct inode *inode, 283extern int ext4_ext_calc_credits_for_single_extent(struct inode *inode,
235 int num, 284 int num,
@@ -237,19 +286,9 @@ extern int ext4_ext_calc_credits_for_single_extent(struct inode *inode,
237extern int ext4_can_extents_be_merged(struct inode *inode, 286extern int ext4_can_extents_be_merged(struct inode *inode,
238 struct ext4_extent *ex1, 287 struct ext4_extent *ex1,
239 struct ext4_extent *ex2); 288 struct ext4_extent *ex2);
240extern int ext4_ext_try_to_merge(struct inode *inode,
241 struct ext4_ext_path *path,
242 struct ext4_extent *);
243extern unsigned int ext4_ext_check_overlap(struct inode *, struct ext4_extent *, struct ext4_ext_path *);
244extern int ext4_ext_insert_extent(handle_t *, struct inode *, struct ext4_ext_path *, struct ext4_extent *, int); 289extern int ext4_ext_insert_extent(handle_t *, struct inode *, struct ext4_ext_path *, struct ext4_extent *, int);
245extern int ext4_ext_walk_space(struct inode *, ext4_lblk_t, ext4_lblk_t,
246 ext_prepare_callback, void *);
247extern struct ext4_ext_path *ext4_ext_find_extent(struct inode *, ext4_lblk_t, 290extern struct ext4_ext_path *ext4_ext_find_extent(struct inode *, ext4_lblk_t,
248 struct ext4_ext_path *); 291 struct ext4_ext_path *);
249extern int ext4_ext_search_left(struct inode *, struct ext4_ext_path *,
250 ext4_lblk_t *, ext4_fsblk_t *);
251extern int ext4_ext_search_right(struct inode *, struct ext4_ext_path *,
252 ext4_lblk_t *, ext4_fsblk_t *);
253extern void ext4_ext_drop_refs(struct ext4_ext_path *); 292extern void ext4_ext_drop_refs(struct ext4_ext_path *);
254extern int ext4_ext_check_inode(struct inode *inode); 293extern int ext4_ext_check_inode(struct inode *inode);
255#endif /* _EXT4_EXTENTS */ 294#endif /* _EXT4_EXTENTS */
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 06328d3e5717..0554c48cb1fd 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -44,55 +44,6 @@
44#include "ext4_jbd2.h" 44#include "ext4_jbd2.h"
45#include "ext4_extents.h" 45#include "ext4_extents.h"
46 46
47
48/*
49 * ext_pblock:
50 * combine low and high parts of physical block number into ext4_fsblk_t
51 */
52ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
53{
54 ext4_fsblk_t block;
55
56 block = le32_to_cpu(ex->ee_start_lo);
57 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
58 return block;
59}
60
61/*
62 * idx_pblock:
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64 */
65ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
66{
67 ext4_fsblk_t block;
68
69 block = le32_to_cpu(ix->ei_leaf_lo);
70 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
71 return block;
72}
73
74/*
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
78 */
79void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
80{
81 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
82 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
83}
84
85/*
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
89 */
90static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
91{
92 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
93 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
94}
95
96static int ext4_ext_truncate_extend_restart(handle_t *handle, 47static int ext4_ext_truncate_extend_restart(handle_t *handle,
97 struct inode *inode, 48 struct inode *inode,
98 int needed) 49 int needed)
@@ -169,7 +120,8 @@ static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
169 /* try to predict block placement */ 120 /* try to predict block placement */
170 ex = path[depth].p_ext; 121 ex = path[depth].p_ext;
171 if (ex) 122 if (ex)
172 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block)); 123 return (ext4_ext_pblock(ex) +
124 (block - le32_to_cpu(ex->ee_block)));
173 125
174 /* it looks like index is empty; 126 /* it looks like index is empty;
175 * try to find starting block from index itself */ 127 * try to find starting block from index itself */
@@ -354,7 +306,7 @@ ext4_ext_max_entries(struct inode *inode, int depth)
354 306
355static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) 307static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
356{ 308{
357 ext4_fsblk_t block = ext_pblock(ext); 309 ext4_fsblk_t block = ext4_ext_pblock(ext);
358 int len = ext4_ext_get_actual_len(ext); 310 int len = ext4_ext_get_actual_len(ext);
359 311
360 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); 312 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
@@ -363,7 +315,7 @@ static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
363static int ext4_valid_extent_idx(struct inode *inode, 315static int ext4_valid_extent_idx(struct inode *inode,
364 struct ext4_extent_idx *ext_idx) 316 struct ext4_extent_idx *ext_idx)
365{ 317{
366 ext4_fsblk_t block = idx_pblock(ext_idx); 318 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
367 319
368 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); 320 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
369} 321}
@@ -463,13 +415,13 @@ static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
463 for (k = 0; k <= l; k++, path++) { 415 for (k = 0; k <= l; k++, path++) {
464 if (path->p_idx) { 416 if (path->p_idx) {
465 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), 417 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
466 idx_pblock(path->p_idx)); 418 ext4_idx_pblock(path->p_idx));
467 } else if (path->p_ext) { 419 } else if (path->p_ext) {
468 ext_debug(" %d:[%d]%d:%llu ", 420 ext_debug(" %d:[%d]%d:%llu ",
469 le32_to_cpu(path->p_ext->ee_block), 421 le32_to_cpu(path->p_ext->ee_block),
470 ext4_ext_is_uninitialized(path->p_ext), 422 ext4_ext_is_uninitialized(path->p_ext),
471 ext4_ext_get_actual_len(path->p_ext), 423 ext4_ext_get_actual_len(path->p_ext),
472 ext_pblock(path->p_ext)); 424 ext4_ext_pblock(path->p_ext));
473 } else 425 } else
474 ext_debug(" []"); 426 ext_debug(" []");
475 } 427 }
@@ -494,7 +446,7 @@ static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
494 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { 446 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
495 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), 447 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
496 ext4_ext_is_uninitialized(ex), 448 ext4_ext_is_uninitialized(ex),
497 ext4_ext_get_actual_len(ex), ext_pblock(ex)); 449 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
498 } 450 }
499 ext_debug("\n"); 451 ext_debug("\n");
500} 452}
@@ -545,7 +497,7 @@ ext4_ext_binsearch_idx(struct inode *inode,
545 497
546 path->p_idx = l - 1; 498 path->p_idx = l - 1;
547 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), 499 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
548 idx_pblock(path->p_idx)); 500 ext4_idx_pblock(path->p_idx));
549 501
550#ifdef CHECK_BINSEARCH 502#ifdef CHECK_BINSEARCH
551 { 503 {
@@ -614,7 +566,7 @@ ext4_ext_binsearch(struct inode *inode,
614 path->p_ext = l - 1; 566 path->p_ext = l - 1;
615 ext_debug(" -> %d:%llu:[%d]%d ", 567 ext_debug(" -> %d:%llu:[%d]%d ",
616 le32_to_cpu(path->p_ext->ee_block), 568 le32_to_cpu(path->p_ext->ee_block),
617 ext_pblock(path->p_ext), 569 ext4_ext_pblock(path->p_ext),
618 ext4_ext_is_uninitialized(path->p_ext), 570 ext4_ext_is_uninitialized(path->p_ext),
619 ext4_ext_get_actual_len(path->p_ext)); 571 ext4_ext_get_actual_len(path->p_ext));
620 572
@@ -682,7 +634,7 @@ ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
682 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); 634 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
683 635
684 ext4_ext_binsearch_idx(inode, path + ppos, block); 636 ext4_ext_binsearch_idx(inode, path + ppos, block);
685 path[ppos].p_block = idx_pblock(path[ppos].p_idx); 637 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
686 path[ppos].p_depth = i; 638 path[ppos].p_depth = i;
687 path[ppos].p_ext = NULL; 639 path[ppos].p_ext = NULL;
688 640
@@ -721,7 +673,7 @@ ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
721 ext4_ext_binsearch(inode, path + ppos, block); 673 ext4_ext_binsearch(inode, path + ppos, block);
722 /* if not an empty leaf */ 674 /* if not an empty leaf */
723 if (path[ppos].p_ext) 675 if (path[ppos].p_ext)
724 path[ppos].p_block = ext_pblock(path[ppos].p_ext); 676 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
725 677
726 ext4_ext_show_path(inode, path); 678 ext4_ext_show_path(inode, path);
727 679
@@ -739,9 +691,9 @@ err:
739 * insert new index [@logical;@ptr] into the block at @curp; 691 * insert new index [@logical;@ptr] into the block at @curp;
740 * check where to insert: before @curp or after @curp 692 * check where to insert: before @curp or after @curp
741 */ 693 */
742int ext4_ext_insert_index(handle_t *handle, struct inode *inode, 694static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
743 struct ext4_ext_path *curp, 695 struct ext4_ext_path *curp,
744 int logical, ext4_fsblk_t ptr) 696 int logical, ext4_fsblk_t ptr)
745{ 697{
746 struct ext4_extent_idx *ix; 698 struct ext4_extent_idx *ix;
747 int len, err; 699 int len, err;
@@ -917,7 +869,7 @@ static int ext4_ext_split(handle_t *handle, struct inode *inode,
917 EXT_MAX_EXTENT(path[depth].p_hdr)) { 869 EXT_MAX_EXTENT(path[depth].p_hdr)) {
918 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", 870 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
919 le32_to_cpu(path[depth].p_ext->ee_block), 871 le32_to_cpu(path[depth].p_ext->ee_block),
920 ext_pblock(path[depth].p_ext), 872 ext4_ext_pblock(path[depth].p_ext),
921 ext4_ext_is_uninitialized(path[depth].p_ext), 873 ext4_ext_is_uninitialized(path[depth].p_ext),
922 ext4_ext_get_actual_len(path[depth].p_ext), 874 ext4_ext_get_actual_len(path[depth].p_ext),
923 newblock); 875 newblock);
@@ -1007,7 +959,7 @@ static int ext4_ext_split(handle_t *handle, struct inode *inode,
1007 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { 959 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
1008 ext_debug("%d: move %d:%llu in new index %llu\n", i, 960 ext_debug("%d: move %d:%llu in new index %llu\n", i,
1009 le32_to_cpu(path[i].p_idx->ei_block), 961 le32_to_cpu(path[i].p_idx->ei_block),
1010 idx_pblock(path[i].p_idx), 962 ext4_idx_pblock(path[i].p_idx),
1011 newblock); 963 newblock);
1012 /*memmove(++fidx, path[i].p_idx++, 964 /*memmove(++fidx, path[i].p_idx++,
1013 sizeof(struct ext4_extent_idx)); 965 sizeof(struct ext4_extent_idx));
@@ -1146,7 +1098,7 @@ static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1146 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", 1098 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1147 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), 1099 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1148 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), 1100 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1149 idx_pblock(EXT_FIRST_INDEX(neh))); 1101 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1150 1102
1151 neh->eh_depth = cpu_to_le16(path->p_depth + 1); 1103 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1152 err = ext4_ext_dirty(handle, inode, curp); 1104 err = ext4_ext_dirty(handle, inode, curp);
@@ -1232,9 +1184,9 @@ out:
1232 * returns 0 at @phys 1184 * returns 0 at @phys
1233 * return value contains 0 (success) or error code 1185 * return value contains 0 (success) or error code
1234 */ 1186 */
1235int 1187static int ext4_ext_search_left(struct inode *inode,
1236ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path, 1188 struct ext4_ext_path *path,
1237 ext4_lblk_t *logical, ext4_fsblk_t *phys) 1189 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1238{ 1190{
1239 struct ext4_extent_idx *ix; 1191 struct ext4_extent_idx *ix;
1240 struct ext4_extent *ex; 1192 struct ext4_extent *ex;
@@ -1286,7 +1238,7 @@ ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1286 } 1238 }
1287 1239
1288 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; 1240 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1289 *phys = ext_pblock(ex) + ee_len - 1; 1241 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1290 return 0; 1242 return 0;
1291} 1243}
1292 1244
@@ -1297,9 +1249,9 @@ ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1297 * returns 0 at @phys 1249 * returns 0 at @phys
1298 * return value contains 0 (success) or error code 1250 * return value contains 0 (success) or error code
1299 */ 1251 */
1300int 1252static int ext4_ext_search_right(struct inode *inode,
1301ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path, 1253 struct ext4_ext_path *path,
1302 ext4_lblk_t *logical, ext4_fsblk_t *phys) 1254 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1303{ 1255{
1304 struct buffer_head *bh = NULL; 1256 struct buffer_head *bh = NULL;
1305 struct ext4_extent_header *eh; 1257 struct ext4_extent_header *eh;
@@ -1342,7 +1294,7 @@ ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1342 } 1294 }
1343 } 1295 }
1344 *logical = le32_to_cpu(ex->ee_block); 1296 *logical = le32_to_cpu(ex->ee_block);
1345 *phys = ext_pblock(ex); 1297 *phys = ext4_ext_pblock(ex);
1346 return 0; 1298 return 0;
1347 } 1299 }
1348 1300
@@ -1357,7 +1309,7 @@ ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1357 /* next allocated block in this leaf */ 1309 /* next allocated block in this leaf */
1358 ex++; 1310 ex++;
1359 *logical = le32_to_cpu(ex->ee_block); 1311 *logical = le32_to_cpu(ex->ee_block);
1360 *phys = ext_pblock(ex); 1312 *phys = ext4_ext_pblock(ex);
1361 return 0; 1313 return 0;
1362 } 1314 }
1363 1315
@@ -1376,7 +1328,7 @@ got_index:
1376 * follow it and find the closest allocated 1328 * follow it and find the closest allocated
1377 * block to the right */ 1329 * block to the right */
1378 ix++; 1330 ix++;
1379 block = idx_pblock(ix); 1331 block = ext4_idx_pblock(ix);
1380 while (++depth < path->p_depth) { 1332 while (++depth < path->p_depth) {
1381 bh = sb_bread(inode->i_sb, block); 1333 bh = sb_bread(inode->i_sb, block);
1382 if (bh == NULL) 1334 if (bh == NULL)
@@ -1388,7 +1340,7 @@ got_index:
1388 return -EIO; 1340 return -EIO;
1389 } 1341 }
1390 ix = EXT_FIRST_INDEX(eh); 1342 ix = EXT_FIRST_INDEX(eh);
1391 block = idx_pblock(ix); 1343 block = ext4_idx_pblock(ix);
1392 put_bh(bh); 1344 put_bh(bh);
1393 } 1345 }
1394 1346
@@ -1402,7 +1354,7 @@ got_index:
1402 } 1354 }
1403 ex = EXT_FIRST_EXTENT(eh); 1355 ex = EXT_FIRST_EXTENT(eh);
1404 *logical = le32_to_cpu(ex->ee_block); 1356 *logical = le32_to_cpu(ex->ee_block);
1405 *phys = ext_pblock(ex); 1357 *phys = ext4_ext_pblock(ex);
1406 put_bh(bh); 1358 put_bh(bh);
1407 return 0; 1359 return 0;
1408} 1360}
@@ -1573,7 +1525,7 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1573 return 0; 1525 return 0;
1574#endif 1526#endif
1575 1527
1576 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2)) 1528 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1577 return 1; 1529 return 1;
1578 return 0; 1530 return 0;
1579} 1531}
@@ -1585,9 +1537,9 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1585 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns 1537 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1586 * 1 if they got merged. 1538 * 1 if they got merged.
1587 */ 1539 */
1588int ext4_ext_try_to_merge(struct inode *inode, 1540static int ext4_ext_try_to_merge(struct inode *inode,
1589 struct ext4_ext_path *path, 1541 struct ext4_ext_path *path,
1590 struct ext4_extent *ex) 1542 struct ext4_extent *ex)
1591{ 1543{
1592 struct ext4_extent_header *eh; 1544 struct ext4_extent_header *eh;
1593 unsigned int depth, len; 1545 unsigned int depth, len;
@@ -1632,9 +1584,9 @@ int ext4_ext_try_to_merge(struct inode *inode,
1632 * such that there will be no overlap, and then returns 1. 1584 * such that there will be no overlap, and then returns 1.
1633 * If there is no overlap found, it returns 0. 1585 * If there is no overlap found, it returns 0.
1634 */ 1586 */
1635unsigned int ext4_ext_check_overlap(struct inode *inode, 1587static unsigned int ext4_ext_check_overlap(struct inode *inode,
1636 struct ext4_extent *newext, 1588 struct ext4_extent *newext,
1637 struct ext4_ext_path *path) 1589 struct ext4_ext_path *path)
1638{ 1590{
1639 ext4_lblk_t b1, b2; 1591 ext4_lblk_t b1, b2;
1640 unsigned int depth, len1; 1592 unsigned int depth, len1;
@@ -1706,11 +1658,12 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1706 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) 1658 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1707 && ext4_can_extents_be_merged(inode, ex, newext)) { 1659 && ext4_can_extents_be_merged(inode, ex, newext)) {
1708 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n", 1660 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1709 ext4_ext_is_uninitialized(newext), 1661 ext4_ext_is_uninitialized(newext),
1710 ext4_ext_get_actual_len(newext), 1662 ext4_ext_get_actual_len(newext),
1711 le32_to_cpu(ex->ee_block), 1663 le32_to_cpu(ex->ee_block),
1712 ext4_ext_is_uninitialized(ex), 1664 ext4_ext_is_uninitialized(ex),
1713 ext4_ext_get_actual_len(ex), ext_pblock(ex)); 1665 ext4_ext_get_actual_len(ex),
1666 ext4_ext_pblock(ex));
1714 err = ext4_ext_get_access(handle, inode, path + depth); 1667 err = ext4_ext_get_access(handle, inode, path + depth);
1715 if (err) 1668 if (err)
1716 return err; 1669 return err;
@@ -1780,7 +1733,7 @@ has_space:
1780 /* there is no extent in this leaf, create first one */ 1733 /* there is no extent in this leaf, create first one */
1781 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n", 1734 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1782 le32_to_cpu(newext->ee_block), 1735 le32_to_cpu(newext->ee_block),
1783 ext_pblock(newext), 1736 ext4_ext_pblock(newext),
1784 ext4_ext_is_uninitialized(newext), 1737 ext4_ext_is_uninitialized(newext),
1785 ext4_ext_get_actual_len(newext)); 1738 ext4_ext_get_actual_len(newext));
1786 path[depth].p_ext = EXT_FIRST_EXTENT(eh); 1739 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
@@ -1794,7 +1747,7 @@ has_space:
1794 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " 1747 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1795 "move %d from 0x%p to 0x%p\n", 1748 "move %d from 0x%p to 0x%p\n",
1796 le32_to_cpu(newext->ee_block), 1749 le32_to_cpu(newext->ee_block),
1797 ext_pblock(newext), 1750 ext4_ext_pblock(newext),
1798 ext4_ext_is_uninitialized(newext), 1751 ext4_ext_is_uninitialized(newext),
1799 ext4_ext_get_actual_len(newext), 1752 ext4_ext_get_actual_len(newext),
1800 nearex, len, nearex + 1, nearex + 2); 1753 nearex, len, nearex + 1, nearex + 2);
@@ -1808,7 +1761,7 @@ has_space:
1808 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " 1761 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1809 "move %d from 0x%p to 0x%p\n", 1762 "move %d from 0x%p to 0x%p\n",
1810 le32_to_cpu(newext->ee_block), 1763 le32_to_cpu(newext->ee_block),
1811 ext_pblock(newext), 1764 ext4_ext_pblock(newext),
1812 ext4_ext_is_uninitialized(newext), 1765 ext4_ext_is_uninitialized(newext),
1813 ext4_ext_get_actual_len(newext), 1766 ext4_ext_get_actual_len(newext),
1814 nearex, len, nearex + 1, nearex + 2); 1767 nearex, len, nearex + 1, nearex + 2);
@@ -1819,7 +1772,7 @@ has_space:
1819 le16_add_cpu(&eh->eh_entries, 1); 1772 le16_add_cpu(&eh->eh_entries, 1);
1820 nearex = path[depth].p_ext; 1773 nearex = path[depth].p_ext;
1821 nearex->ee_block = newext->ee_block; 1774 nearex->ee_block = newext->ee_block;
1822 ext4_ext_store_pblock(nearex, ext_pblock(newext)); 1775 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1823 nearex->ee_len = newext->ee_len; 1776 nearex->ee_len = newext->ee_len;
1824 1777
1825merge: 1778merge:
@@ -1845,9 +1798,9 @@ cleanup:
1845 return err; 1798 return err;
1846} 1799}
1847 1800
1848int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, 1801static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1849 ext4_lblk_t num, ext_prepare_callback func, 1802 ext4_lblk_t num, ext_prepare_callback func,
1850 void *cbdata) 1803 void *cbdata)
1851{ 1804{
1852 struct ext4_ext_path *path = NULL; 1805 struct ext4_ext_path *path = NULL;
1853 struct ext4_ext_cache cbex; 1806 struct ext4_ext_cache cbex;
@@ -1923,7 +1876,7 @@ int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1923 } else { 1876 } else {
1924 cbex.ec_block = le32_to_cpu(ex->ee_block); 1877 cbex.ec_block = le32_to_cpu(ex->ee_block);
1925 cbex.ec_len = ext4_ext_get_actual_len(ex); 1878 cbex.ec_len = ext4_ext_get_actual_len(ex);
1926 cbex.ec_start = ext_pblock(ex); 1879 cbex.ec_start = ext4_ext_pblock(ex);
1927 cbex.ec_type = EXT4_EXT_CACHE_EXTENT; 1880 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1928 } 1881 }
1929 1882
@@ -2073,7 +2026,7 @@ static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2073 2026
2074 /* free index block */ 2027 /* free index block */
2075 path--; 2028 path--;
2076 leaf = idx_pblock(path->p_idx); 2029 leaf = ext4_idx_pblock(path->p_idx);
2077 if (unlikely(path->p_hdr->eh_entries == 0)) { 2030 if (unlikely(path->p_hdr->eh_entries == 0)) {
2078 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); 2031 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2079 return -EIO; 2032 return -EIO;
@@ -2181,7 +2134,7 @@ static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2181 ext4_fsblk_t start; 2134 ext4_fsblk_t start;
2182 2135
2183 num = le32_to_cpu(ex->ee_block) + ee_len - from; 2136 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2184 start = ext_pblock(ex) + ee_len - num; 2137 start = ext4_ext_pblock(ex) + ee_len - num;
2185 ext_debug("free last %u blocks starting %llu\n", num, start); 2138 ext_debug("free last %u blocks starting %llu\n", num, start);
2186 ext4_free_blocks(handle, inode, 0, start, num, flags); 2139 ext4_free_blocks(handle, inode, 0, start, num, flags);
2187 } else if (from == le32_to_cpu(ex->ee_block) 2140 } else if (from == le32_to_cpu(ex->ee_block)
@@ -2310,7 +2263,7 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2310 goto out; 2263 goto out;
2311 2264
2312 ext_debug("new extent: %u:%u:%llu\n", block, num, 2265 ext_debug("new extent: %u:%u:%llu\n", block, num,
2313 ext_pblock(ex)); 2266 ext4_ext_pblock(ex));
2314 ex--; 2267 ex--;
2315 ex_ee_block = le32_to_cpu(ex->ee_block); 2268 ex_ee_block = le32_to_cpu(ex->ee_block);
2316 ex_ee_len = ext4_ext_get_actual_len(ex); 2269 ex_ee_len = ext4_ext_get_actual_len(ex);
@@ -2421,9 +2374,9 @@ again:
2421 struct buffer_head *bh; 2374 struct buffer_head *bh;
2422 /* go to the next level */ 2375 /* go to the next level */
2423 ext_debug("move to level %d (block %llu)\n", 2376 ext_debug("move to level %d (block %llu)\n",
2424 i + 1, idx_pblock(path[i].p_idx)); 2377 i + 1, ext4_idx_pblock(path[i].p_idx));
2425 memset(path + i + 1, 0, sizeof(*path)); 2378 memset(path + i + 1, 0, sizeof(*path));
2426 bh = sb_bread(sb, idx_pblock(path[i].p_idx)); 2379 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2427 if (!bh) { 2380 if (!bh) {
2428 /* should we reset i_size? */ 2381 /* should we reset i_size? */
2429 err = -EIO; 2382 err = -EIO;
@@ -2535,77 +2488,21 @@ void ext4_ext_release(struct super_block *sb)
2535#endif 2488#endif
2536} 2489}
2537 2490
2538static void bi_complete(struct bio *bio, int error)
2539{
2540 complete((struct completion *)bio->bi_private);
2541}
2542
2543/* FIXME!! we need to try to merge to left or right after zero-out */ 2491/* FIXME!! we need to try to merge to left or right after zero-out */
2544static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) 2492static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2545{ 2493{
2494 ext4_fsblk_t ee_pblock;
2495 unsigned int ee_len;
2546 int ret; 2496 int ret;
2547 struct bio *bio;
2548 int blkbits, blocksize;
2549 sector_t ee_pblock;
2550 struct completion event;
2551 unsigned int ee_len, len, done, offset;
2552 2497
2553
2554 blkbits = inode->i_blkbits;
2555 blocksize = inode->i_sb->s_blocksize;
2556 ee_len = ext4_ext_get_actual_len(ex); 2498 ee_len = ext4_ext_get_actual_len(ex);
2557 ee_pblock = ext_pblock(ex); 2499 ee_pblock = ext4_ext_pblock(ex);
2558
2559 /* convert ee_pblock to 512 byte sectors */
2560 ee_pblock = ee_pblock << (blkbits - 9);
2561
2562 while (ee_len > 0) {
2563
2564 if (ee_len > BIO_MAX_PAGES)
2565 len = BIO_MAX_PAGES;
2566 else
2567 len = ee_len;
2568
2569 bio = bio_alloc(GFP_NOIO, len);
2570 if (!bio)
2571 return -ENOMEM;
2572
2573 bio->bi_sector = ee_pblock;
2574 bio->bi_bdev = inode->i_sb->s_bdev;
2575
2576 done = 0;
2577 offset = 0;
2578 while (done < len) {
2579 ret = bio_add_page(bio, ZERO_PAGE(0),
2580 blocksize, offset);
2581 if (ret != blocksize) {
2582 /*
2583 * We can't add any more pages because of
2584 * hardware limitations. Start a new bio.
2585 */
2586 break;
2587 }
2588 done++;
2589 offset += blocksize;
2590 if (offset >= PAGE_CACHE_SIZE)
2591 offset = 0;
2592 }
2593 2500
2594 init_completion(&event); 2501 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2595 bio->bi_private = &event; 2502 if (ret > 0)
2596 bio->bi_end_io = bi_complete; 2503 ret = 0;
2597 submit_bio(WRITE, bio);
2598 wait_for_completion(&event);
2599 2504
2600 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { 2505 return ret;
2601 bio_put(bio);
2602 return -EIO;
2603 }
2604 bio_put(bio);
2605 ee_len -= done;
2606 ee_pblock += done << (blkbits - 9);
2607 }
2608 return 0;
2609} 2506}
2610 2507
2611#define EXT4_EXT_ZERO_LEN 7 2508#define EXT4_EXT_ZERO_LEN 7
@@ -2651,12 +2548,12 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2651 ee_block = le32_to_cpu(ex->ee_block); 2548 ee_block = le32_to_cpu(ex->ee_block);
2652 ee_len = ext4_ext_get_actual_len(ex); 2549 ee_len = ext4_ext_get_actual_len(ex);
2653 allocated = ee_len - (map->m_lblk - ee_block); 2550 allocated = ee_len - (map->m_lblk - ee_block);
2654 newblock = map->m_lblk - ee_block + ext_pblock(ex); 2551 newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2655 2552
2656 ex2 = ex; 2553 ex2 = ex;
2657 orig_ex.ee_block = ex->ee_block; 2554 orig_ex.ee_block = ex->ee_block;
2658 orig_ex.ee_len = cpu_to_le16(ee_len); 2555 orig_ex.ee_len = cpu_to_le16(ee_len);
2659 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); 2556 ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2660 2557
2661 /* 2558 /*
2662 * It is safe to convert extent to initialized via explicit 2559 * It is safe to convert extent to initialized via explicit
@@ -2675,7 +2572,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2675 /* update the extent length and mark as initialized */ 2572 /* update the extent length and mark as initialized */
2676 ex->ee_block = orig_ex.ee_block; 2573 ex->ee_block = orig_ex.ee_block;
2677 ex->ee_len = orig_ex.ee_len; 2574 ex->ee_len = orig_ex.ee_len;
2678 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2575 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2679 ext4_ext_dirty(handle, inode, path + depth); 2576 ext4_ext_dirty(handle, inode, path + depth);
2680 /* zeroed the full extent */ 2577 /* zeroed the full extent */
2681 return allocated; 2578 return allocated;
@@ -2710,7 +2607,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2710 ex->ee_block = orig_ex.ee_block; 2607 ex->ee_block = orig_ex.ee_block;
2711 ex->ee_len = cpu_to_le16(ee_len - allocated); 2608 ex->ee_len = cpu_to_le16(ee_len - allocated);
2712 ext4_ext_mark_uninitialized(ex); 2609 ext4_ext_mark_uninitialized(ex);
2713 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2610 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2714 ext4_ext_dirty(handle, inode, path + depth); 2611 ext4_ext_dirty(handle, inode, path + depth);
2715 2612
2716 ex3 = &newex; 2613 ex3 = &newex;
@@ -2725,7 +2622,8 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2725 goto fix_extent_len; 2622 goto fix_extent_len;
2726 ex->ee_block = orig_ex.ee_block; 2623 ex->ee_block = orig_ex.ee_block;
2727 ex->ee_len = orig_ex.ee_len; 2624 ex->ee_len = orig_ex.ee_len;
2728 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2625 ext4_ext_store_pblock(ex,
2626 ext4_ext_pblock(&orig_ex));
2729 ext4_ext_dirty(handle, inode, path + depth); 2627 ext4_ext_dirty(handle, inode, path + depth);
2730 /* blocks available from map->m_lblk */ 2628 /* blocks available from map->m_lblk */
2731 return allocated; 2629 return allocated;
@@ -2782,7 +2680,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2782 /* update the extent length and mark as initialized */ 2680 /* update the extent length and mark as initialized */
2783 ex->ee_block = orig_ex.ee_block; 2681 ex->ee_block = orig_ex.ee_block;
2784 ex->ee_len = orig_ex.ee_len; 2682 ex->ee_len = orig_ex.ee_len;
2785 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2683 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2786 ext4_ext_dirty(handle, inode, path + depth); 2684 ext4_ext_dirty(handle, inode, path + depth);
2787 /* zeroed the full extent */ 2685 /* zeroed the full extent */
2788 /* blocks available from map->m_lblk */ 2686 /* blocks available from map->m_lblk */
@@ -2833,7 +2731,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle,
2833 /* update the extent length and mark as initialized */ 2731 /* update the extent length and mark as initialized */
2834 ex->ee_block = orig_ex.ee_block; 2732 ex->ee_block = orig_ex.ee_block;
2835 ex->ee_len = orig_ex.ee_len; 2733 ex->ee_len = orig_ex.ee_len;
2836 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2734 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2837 ext4_ext_dirty(handle, inode, path + depth); 2735 ext4_ext_dirty(handle, inode, path + depth);
2838 /* zero out the first half */ 2736 /* zero out the first half */
2839 /* blocks available from map->m_lblk */ 2737 /* blocks available from map->m_lblk */
@@ -2902,7 +2800,7 @@ insert:
2902 /* update the extent length and mark as initialized */ 2800 /* update the extent length and mark as initialized */
2903 ex->ee_block = orig_ex.ee_block; 2801 ex->ee_block = orig_ex.ee_block;
2904 ex->ee_len = orig_ex.ee_len; 2802 ex->ee_len = orig_ex.ee_len;
2905 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2803 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2906 ext4_ext_dirty(handle, inode, path + depth); 2804 ext4_ext_dirty(handle, inode, path + depth);
2907 /* zero out the first half */ 2805 /* zero out the first half */
2908 return allocated; 2806 return allocated;
@@ -2915,7 +2813,7 @@ out:
2915fix_extent_len: 2813fix_extent_len:
2916 ex->ee_block = orig_ex.ee_block; 2814 ex->ee_block = orig_ex.ee_block;
2917 ex->ee_len = orig_ex.ee_len; 2815 ex->ee_len = orig_ex.ee_len;
2918 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2816 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2919 ext4_ext_mark_uninitialized(ex); 2817 ext4_ext_mark_uninitialized(ex);
2920 ext4_ext_dirty(handle, inode, path + depth); 2818 ext4_ext_dirty(handle, inode, path + depth);
2921 return err; 2819 return err;
@@ -2973,12 +2871,12 @@ static int ext4_split_unwritten_extents(handle_t *handle,
2973 ee_block = le32_to_cpu(ex->ee_block); 2871 ee_block = le32_to_cpu(ex->ee_block);
2974 ee_len = ext4_ext_get_actual_len(ex); 2872 ee_len = ext4_ext_get_actual_len(ex);
2975 allocated = ee_len - (map->m_lblk - ee_block); 2873 allocated = ee_len - (map->m_lblk - ee_block);
2976 newblock = map->m_lblk - ee_block + ext_pblock(ex); 2874 newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2977 2875
2978 ex2 = ex; 2876 ex2 = ex;
2979 orig_ex.ee_block = ex->ee_block; 2877 orig_ex.ee_block = ex->ee_block;
2980 orig_ex.ee_len = cpu_to_le16(ee_len); 2878 orig_ex.ee_len = cpu_to_le16(ee_len);
2981 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); 2879 ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2982 2880
2983 /* 2881 /*
2984 * It is safe to convert extent to initialized via explicit 2882 * It is safe to convert extent to initialized via explicit
@@ -3027,7 +2925,7 @@ static int ext4_split_unwritten_extents(handle_t *handle,
3027 /* update the extent length and mark as initialized */ 2925 /* update the extent length and mark as initialized */
3028 ex->ee_block = orig_ex.ee_block; 2926 ex->ee_block = orig_ex.ee_block;
3029 ex->ee_len = orig_ex.ee_len; 2927 ex->ee_len = orig_ex.ee_len;
3030 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 2928 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3031 ext4_ext_dirty(handle, inode, path + depth); 2929 ext4_ext_dirty(handle, inode, path + depth);
3032 /* zeroed the full extent */ 2930 /* zeroed the full extent */
3033 /* blocks available from map->m_lblk */ 2931 /* blocks available from map->m_lblk */
@@ -3099,7 +2997,7 @@ insert:
3099 /* update the extent length and mark as initialized */ 2997 /* update the extent length and mark as initialized */
3100 ex->ee_block = orig_ex.ee_block; 2998 ex->ee_block = orig_ex.ee_block;
3101 ex->ee_len = orig_ex.ee_len; 2999 ex->ee_len = orig_ex.ee_len;
3102 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 3000 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3103 ext4_ext_dirty(handle, inode, path + depth); 3001 ext4_ext_dirty(handle, inode, path + depth);
3104 /* zero out the first half */ 3002 /* zero out the first half */
3105 return allocated; 3003 return allocated;
@@ -3112,7 +3010,7 @@ out:
3112fix_extent_len: 3010fix_extent_len:
3113 ex->ee_block = orig_ex.ee_block; 3011 ex->ee_block = orig_ex.ee_block;
3114 ex->ee_len = orig_ex.ee_len; 3012 ex->ee_len = orig_ex.ee_len;
3115 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); 3013 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3116 ext4_ext_mark_uninitialized(ex); 3014 ext4_ext_mark_uninitialized(ex);
3117 ext4_ext_dirty(handle, inode, path + depth); 3015 ext4_ext_dirty(handle, inode, path + depth);
3118 return err; 3016 return err;
@@ -3180,6 +3078,57 @@ static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3180 unmap_underlying_metadata(bdev, block + i); 3078 unmap_underlying_metadata(bdev, block + i);
3181} 3079}
3182 3080
3081/*
3082 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3083 */
3084static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3085 struct ext4_map_blocks *map,
3086 struct ext4_ext_path *path,
3087 unsigned int len)
3088{
3089 int i, depth;
3090 struct ext4_extent_header *eh;
3091 struct ext4_extent *ex, *last_ex;
3092
3093 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3094 return 0;
3095
3096 depth = ext_depth(inode);
3097 eh = path[depth].p_hdr;
3098 ex = path[depth].p_ext;
3099
3100 if (unlikely(!eh->eh_entries)) {
3101 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3102 "EOFBLOCKS_FL set");
3103 return -EIO;
3104 }
3105 last_ex = EXT_LAST_EXTENT(eh);
3106 /*
3107 * We should clear the EOFBLOCKS_FL flag if we are writing the
3108 * last block in the last extent in the file. We test this by
3109 * first checking to see if the caller to
3110 * ext4_ext_get_blocks() was interested in the last block (or
3111 * a block beyond the last block) in the current extent. If
3112 * this turns out to be false, we can bail out from this
3113 * function immediately.
3114 */
3115 if (map->m_lblk + len < le32_to_cpu(last_ex->ee_block) +
3116 ext4_ext_get_actual_len(last_ex))
3117 return 0;
3118 /*
3119 * If the caller does appear to be planning to write at or
3120 * beyond the end of the current extent, we then test to see
3121 * if the current extent is the last extent in the file, by
3122 * checking to make sure it was reached via the rightmost node
3123 * at each level of the tree.
3124 */
3125 for (i = depth-1; i >= 0; i--)
3126 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3127 return 0;
3128 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3129 return ext4_mark_inode_dirty(handle, inode);
3130}
3131
3183static int 3132static int
3184ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, 3133ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3185 struct ext4_map_blocks *map, 3134 struct ext4_map_blocks *map,
@@ -3206,7 +3155,7 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3206 * completed 3155 * completed
3207 */ 3156 */
3208 if (io) 3157 if (io)
3209 io->flag = EXT4_IO_UNWRITTEN; 3158 io->flag = EXT4_IO_END_UNWRITTEN;
3210 else 3159 else
3211 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); 3160 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3212 if (ext4_should_dioread_nolock(inode)) 3161 if (ext4_should_dioread_nolock(inode))
@@ -3217,8 +3166,12 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3217 if ((flags & EXT4_GET_BLOCKS_CONVERT)) { 3166 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3218 ret = ext4_convert_unwritten_extents_endio(handle, inode, 3167 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3219 path); 3168 path);
3220 if (ret >= 0) 3169 if (ret >= 0) {
3221 ext4_update_inode_fsync_trans(handle, inode, 1); 3170 ext4_update_inode_fsync_trans(handle, inode, 1);
3171 err = check_eofblocks_fl(handle, inode, map, path,
3172 map->m_len);
3173 } else
3174 err = ret;
3222 goto out2; 3175 goto out2;
3223 } 3176 }
3224 /* buffered IO case */ 3177 /* buffered IO case */
@@ -3244,8 +3197,13 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3244 3197
3245 /* buffered write, writepage time, convert*/ 3198 /* buffered write, writepage time, convert*/
3246 ret = ext4_ext_convert_to_initialized(handle, inode, map, path); 3199 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3247 if (ret >= 0) 3200 if (ret >= 0) {
3248 ext4_update_inode_fsync_trans(handle, inode, 1); 3201 ext4_update_inode_fsync_trans(handle, inode, 1);
3202 err = check_eofblocks_fl(handle, inode, map, path, map->m_len);
3203 if (err < 0)
3204 goto out2;
3205 }
3206
3249out: 3207out:
3250 if (ret <= 0) { 3208 if (ret <= 0) {
3251 err = ret; 3209 err = ret;
@@ -3292,6 +3250,7 @@ out2:
3292 } 3250 }
3293 return err ? err : allocated; 3251 return err ? err : allocated;
3294} 3252}
3253
3295/* 3254/*
3296 * Block allocation/map/preallocation routine for extents based files 3255 * Block allocation/map/preallocation routine for extents based files
3297 * 3256 *
@@ -3315,9 +3274,9 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3315{ 3274{
3316 struct ext4_ext_path *path = NULL; 3275 struct ext4_ext_path *path = NULL;
3317 struct ext4_extent_header *eh; 3276 struct ext4_extent_header *eh;
3318 struct ext4_extent newex, *ex, *last_ex; 3277 struct ext4_extent newex, *ex;
3319 ext4_fsblk_t newblock; 3278 ext4_fsblk_t newblock;
3320 int i, err = 0, depth, ret, cache_type; 3279 int err = 0, depth, ret, cache_type;
3321 unsigned int allocated = 0; 3280 unsigned int allocated = 0;
3322 struct ext4_allocation_request ar; 3281 struct ext4_allocation_request ar;
3323 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; 3282 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
@@ -3341,7 +3300,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3341 /* block is already allocated */ 3300 /* block is already allocated */
3342 newblock = map->m_lblk 3301 newblock = map->m_lblk
3343 - le32_to_cpu(newex.ee_block) 3302 - le32_to_cpu(newex.ee_block)
3344 + ext_pblock(&newex); 3303 + ext4_ext_pblock(&newex);
3345 /* number of remaining blocks in the extent */ 3304 /* number of remaining blocks in the extent */
3346 allocated = ext4_ext_get_actual_len(&newex) - 3305 allocated = ext4_ext_get_actual_len(&newex) -
3347 (map->m_lblk - le32_to_cpu(newex.ee_block)); 3306 (map->m_lblk - le32_to_cpu(newex.ee_block));
@@ -3379,7 +3338,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3379 ex = path[depth].p_ext; 3338 ex = path[depth].p_ext;
3380 if (ex) { 3339 if (ex) {
3381 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); 3340 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3382 ext4_fsblk_t ee_start = ext_pblock(ex); 3341 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3383 unsigned short ee_len; 3342 unsigned short ee_len;
3384 3343
3385 /* 3344 /*
@@ -3488,7 +3447,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3488 */ 3447 */
3489 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { 3448 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3490 if (io) 3449 if (io)
3491 io->flag = EXT4_IO_UNWRITTEN; 3450 io->flag = EXT4_IO_END_UNWRITTEN;
3492 else 3451 else
3493 ext4_set_inode_state(inode, 3452 ext4_set_inode_state(inode,
3494 EXT4_STATE_DIO_UNWRITTEN); 3453 EXT4_STATE_DIO_UNWRITTEN);
@@ -3497,44 +3456,23 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3497 map->m_flags |= EXT4_MAP_UNINIT; 3456 map->m_flags |= EXT4_MAP_UNINIT;
3498 } 3457 }
3499 3458
3500 if (unlikely(ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))) { 3459 err = check_eofblocks_fl(handle, inode, map, path, ar.len);
3501 if (unlikely(!eh->eh_entries)) { 3460 if (err)
3502 EXT4_ERROR_INODE(inode, 3461 goto out2;
3503 "eh->eh_entries == 0 and " 3462
3504 "EOFBLOCKS_FL set");
3505 err = -EIO;
3506 goto out2;
3507 }
3508 last_ex = EXT_LAST_EXTENT(eh);
3509 /*
3510 * If the current leaf block was reached by looking at
3511 * the last index block all the way down the tree, and
3512 * we are extending the inode beyond the last extent
3513 * in the current leaf block, then clear the
3514 * EOFBLOCKS_FL flag.
3515 */
3516 for (i = depth-1; i >= 0; i--) {
3517 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3518 break;
3519 }
3520 if ((i < 0) &&
3521 (map->m_lblk + ar.len > le32_to_cpu(last_ex->ee_block) +
3522 ext4_ext_get_actual_len(last_ex)))
3523 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3524 }
3525 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); 3463 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3526 if (err) { 3464 if (err) {
3527 /* free data blocks we just allocated */ 3465 /* free data blocks we just allocated */
3528 /* not a good idea to call discard here directly, 3466 /* not a good idea to call discard here directly,
3529 * but otherwise we'd need to call it every free() */ 3467 * but otherwise we'd need to call it every free() */
3530 ext4_discard_preallocations(inode); 3468 ext4_discard_preallocations(inode);
3531 ext4_free_blocks(handle, inode, 0, ext_pblock(&newex), 3469 ext4_free_blocks(handle, inode, 0, ext4_ext_pblock(&newex),
3532 ext4_ext_get_actual_len(&newex), 0); 3470 ext4_ext_get_actual_len(&newex), 0);
3533 goto out2; 3471 goto out2;
3534 } 3472 }
3535 3473
3536 /* previous routine could use block we allocated */ 3474 /* previous routine could use block we allocated */
3537 newblock = ext_pblock(&newex); 3475 newblock = ext4_ext_pblock(&newex);
3538 allocated = ext4_ext_get_actual_len(&newex); 3476 allocated = ext4_ext_get_actual_len(&newex);
3539 if (allocated > map->m_len) 3477 if (allocated > map->m_len)
3540 allocated = map->m_len; 3478 allocated = map->m_len;
@@ -3729,7 +3667,7 @@ retry:
3729 printk(KERN_ERR "%s: ext4_ext_map_blocks " 3667 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3730 "returned error inode#%lu, block=%u, " 3668 "returned error inode#%lu, block=%u, "
3731 "max_blocks=%u", __func__, 3669 "max_blocks=%u", __func__,
3732 inode->i_ino, block, max_blocks); 3670 inode->i_ino, map.m_lblk, max_blocks);
3733#endif 3671#endif
3734 ext4_mark_inode_dirty(handle, inode); 3672 ext4_mark_inode_dirty(handle, inode);
3735 ret2 = ext4_journal_stop(handle); 3673 ret2 = ext4_journal_stop(handle);
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index ee92b66d4558..5a5c55ddceef 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -130,8 +130,50 @@ static int ext4_file_open(struct inode * inode, struct file * filp)
130 return dquot_file_open(inode, filp); 130 return dquot_file_open(inode, filp);
131} 131}
132 132
133/*
134 * ext4_llseek() copied from generic_file_llseek() to handle both
135 * block-mapped and extent-mapped maxbytes values. This should
136 * otherwise be identical with generic_file_llseek().
137 */
138loff_t ext4_llseek(struct file *file, loff_t offset, int origin)
139{
140 struct inode *inode = file->f_mapping->host;
141 loff_t maxbytes;
142
143 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
144 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
145 else
146 maxbytes = inode->i_sb->s_maxbytes;
147 mutex_lock(&inode->i_mutex);
148 switch (origin) {
149 case SEEK_END:
150 offset += inode->i_size;
151 break;
152 case SEEK_CUR:
153 if (offset == 0) {
154 mutex_unlock(&inode->i_mutex);
155 return file->f_pos;
156 }
157 offset += file->f_pos;
158 break;
159 }
160
161 if (offset < 0 || offset > maxbytes) {
162 mutex_unlock(&inode->i_mutex);
163 return -EINVAL;
164 }
165
166 if (offset != file->f_pos) {
167 file->f_pos = offset;
168 file->f_version = 0;
169 }
170 mutex_unlock(&inode->i_mutex);
171
172 return offset;
173}
174
133const struct file_operations ext4_file_operations = { 175const struct file_operations ext4_file_operations = {
134 .llseek = generic_file_llseek, 176 .llseek = ext4_llseek,
135 .read = do_sync_read, 177 .read = do_sync_read,
136 .write = do_sync_write, 178 .write = do_sync_write,
137 .aio_read = generic_file_aio_read, 179 .aio_read = generic_file_aio_read,
diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index 3f3ff5ee8f9d..c1a7bc923cf6 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -34,6 +34,89 @@
34 34
35#include <trace/events/ext4.h> 35#include <trace/events/ext4.h>
36 36
37static void dump_completed_IO(struct inode * inode)
38{
39#ifdef EXT4_DEBUG
40 struct list_head *cur, *before, *after;
41 ext4_io_end_t *io, *io0, *io1;
42 unsigned long flags;
43
44 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
45 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
46 return;
47 }
48
49 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
50 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
51 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
52 cur = &io->list;
53 before = cur->prev;
54 io0 = container_of(before, ext4_io_end_t, list);
55 after = cur->next;
56 io1 = container_of(after, ext4_io_end_t, list);
57
58 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
59 io, inode->i_ino, io0, io1);
60 }
61 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
62#endif
63}
64
65/*
66 * This function is called from ext4_sync_file().
67 *
68 * When IO is completed, the work to convert unwritten extents to
69 * written is queued on workqueue but may not get immediately
70 * scheduled. When fsync is called, we need to ensure the
71 * conversion is complete before fsync returns.
72 * The inode keeps track of a list of pending/completed IO that
73 * might needs to do the conversion. This function walks through
74 * the list and convert the related unwritten extents for completed IO
75 * to written.
76 * The function return the number of pending IOs on success.
77 */
78static int flush_completed_IO(struct inode *inode)
79{
80 ext4_io_end_t *io;
81 struct ext4_inode_info *ei = EXT4_I(inode);
82 unsigned long flags;
83 int ret = 0;
84 int ret2 = 0;
85
86 if (list_empty(&ei->i_completed_io_list))
87 return ret;
88
89 dump_completed_IO(inode);
90 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
91 while (!list_empty(&ei->i_completed_io_list)){
92 io = list_entry(ei->i_completed_io_list.next,
93 ext4_io_end_t, list);
94 /*
95 * Calling ext4_end_io_nolock() to convert completed
96 * IO to written.
97 *
98 * When ext4_sync_file() is called, run_queue() may already
99 * about to flush the work corresponding to this io structure.
100 * It will be upset if it founds the io structure related
101 * to the work-to-be schedule is freed.
102 *
103 * Thus we need to keep the io structure still valid here after
104 * convertion finished. The io structure has a flag to
105 * avoid double converting from both fsync and background work
106 * queue work.
107 */
108 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
109 ret = ext4_end_io_nolock(io);
110 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
111 if (ret < 0)
112 ret2 = ret;
113 else
114 list_del_init(&io->list);
115 }
116 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
117 return (ret2 < 0) ? ret2 : 0;
118}
119
37/* 120/*
38 * If we're not journaling and this is a just-created file, we have to 121 * If we're not journaling and this is a just-created file, we have to
39 * sync our parent directory (if it was freshly created) since 122 * sync our parent directory (if it was freshly created) since
diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index 45853e0d1f21..1ce240a23ebb 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -50,7 +50,7 @@
50 * need to use it within a single byte (to ensure we get endianness right). 50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users. 51 * We can use memset for the rest of the bitmap as there are no other users.
52 */ 52 */
53void mark_bitmap_end(int start_bit, int end_bit, char *bitmap) 53void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
54{ 54{
55 int i; 55 int i;
56 56
@@ -65,9 +65,10 @@ void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
65} 65}
66 66
67/* Initializes an uninitialized inode bitmap */ 67/* Initializes an uninitialized inode bitmap */
68unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh, 68static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 ext4_group_t block_group, 69 struct buffer_head *bh,
70 struct ext4_group_desc *gdp) 70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
71{ 72{
72 struct ext4_sb_info *sbi = EXT4_SB(sb); 73 struct ext4_sb_info *sbi = EXT4_SB(sb);
73 74
@@ -85,7 +86,7 @@ unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
85 } 86 }
86 87
87 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); 88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
88 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, 89 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
89 bh->b_data); 90 bh->b_data);
90 91
91 return EXT4_INODES_PER_GROUP(sb); 92 return EXT4_INODES_PER_GROUP(sb);
@@ -107,6 +108,7 @@ ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
107 desc = ext4_get_group_desc(sb, block_group, NULL); 108 desc = ext4_get_group_desc(sb, block_group, NULL);
108 if (!desc) 109 if (!desc)
109 return NULL; 110 return NULL;
111
110 bitmap_blk = ext4_inode_bitmap(sb, desc); 112 bitmap_blk = ext4_inode_bitmap(sb, desc);
111 bh = sb_getblk(sb, bitmap_blk); 113 bh = sb_getblk(sb, bitmap_blk);
112 if (unlikely(!bh)) { 114 if (unlikely(!bh)) {
@@ -123,6 +125,7 @@ ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
123 unlock_buffer(bh); 125 unlock_buffer(bh);
124 return bh; 126 return bh;
125 } 127 }
128
126 ext4_lock_group(sb, block_group); 129 ext4_lock_group(sb, block_group);
127 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { 130 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
128 ext4_init_inode_bitmap(sb, bh, block_group, desc); 131 ext4_init_inode_bitmap(sb, bh, block_group, desc);
@@ -133,6 +136,7 @@ ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
133 return bh; 136 return bh;
134 } 137 }
135 ext4_unlock_group(sb, block_group); 138 ext4_unlock_group(sb, block_group);
139
136 if (buffer_uptodate(bh)) { 140 if (buffer_uptodate(bh)) {
137 /* 141 /*
138 * if not uninit if bh is uptodate, 142 * if not uninit if bh is uptodate,
@@ -411,8 +415,8 @@ struct orlov_stats {
411 * for a particular block group or flex_bg. If flex_size is 1, then g 415 * for a particular block group or flex_bg. If flex_size is 1, then g
412 * is a block group number; otherwise it is flex_bg number. 416 * is a block group number; otherwise it is flex_bg number.
413 */ 417 */
414void get_orlov_stats(struct super_block *sb, ext4_group_t g, 418static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
415 int flex_size, struct orlov_stats *stats) 419 int flex_size, struct orlov_stats *stats)
416{ 420{
417 struct ext4_group_desc *desc; 421 struct ext4_group_desc *desc;
418 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; 422 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
@@ -712,8 +716,17 @@ static int ext4_claim_inode(struct super_block *sb,
712{ 716{
713 int free = 0, retval = 0, count; 717 int free = 0, retval = 0, count;
714 struct ext4_sb_info *sbi = EXT4_SB(sb); 718 struct ext4_sb_info *sbi = EXT4_SB(sb);
719 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
715 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 720 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
716 721
722 /*
723 * We have to be sure that new inode allocation does not race with
724 * inode table initialization, because otherwise we may end up
725 * allocating and writing new inode right before sb_issue_zeroout
726 * takes place and overwriting our new inode with zeroes. So we
727 * take alloc_sem to prevent it.
728 */
729 down_read(&grp->alloc_sem);
717 ext4_lock_group(sb, group); 730 ext4_lock_group(sb, group);
718 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { 731 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
719 /* not a free inode */ 732 /* not a free inode */
@@ -724,6 +737,7 @@ static int ext4_claim_inode(struct super_block *sb,
724 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || 737 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
725 ino > EXT4_INODES_PER_GROUP(sb)) { 738 ino > EXT4_INODES_PER_GROUP(sb)) {
726 ext4_unlock_group(sb, group); 739 ext4_unlock_group(sb, group);
740 up_read(&grp->alloc_sem);
727 ext4_error(sb, "reserved inode or inode > inodes count - " 741 ext4_error(sb, "reserved inode or inode > inodes count - "
728 "block_group = %u, inode=%lu", group, 742 "block_group = %u, inode=%lu", group,
729 ino + group * EXT4_INODES_PER_GROUP(sb)); 743 ino + group * EXT4_INODES_PER_GROUP(sb));
@@ -772,6 +786,7 @@ static int ext4_claim_inode(struct super_block *sb,
772 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); 786 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
773err_ret: 787err_ret:
774 ext4_unlock_group(sb, group); 788 ext4_unlock_group(sb, group);
789 up_read(&grp->alloc_sem);
775 return retval; 790 return retval;
776} 791}
777 792
@@ -1205,3 +1220,109 @@ unsigned long ext4_count_dirs(struct super_block * sb)
1205 } 1220 }
1206 return count; 1221 return count;
1207} 1222}
1223
1224/*
1225 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1226 * inode table. Must be called without any spinlock held. The only place
1227 * where it is called from on active part of filesystem is ext4lazyinit
1228 * thread, so we do not need any special locks, however we have to prevent
1229 * inode allocation from the current group, so we take alloc_sem lock, to
1230 * block ext4_claim_inode until we are finished.
1231 */
1232extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1233 int barrier)
1234{
1235 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1236 struct ext4_sb_info *sbi = EXT4_SB(sb);
1237 struct ext4_group_desc *gdp = NULL;
1238 struct buffer_head *group_desc_bh;
1239 handle_t *handle;
1240 ext4_fsblk_t blk;
1241 int num, ret = 0, used_blks = 0;
1242
1243 /* This should not happen, but just to be sure check this */
1244 if (sb->s_flags & MS_RDONLY) {
1245 ret = 1;
1246 goto out;
1247 }
1248
1249 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1250 if (!gdp)
1251 goto out;
1252
1253 /*
1254 * We do not need to lock this, because we are the only one
1255 * handling this flag.
1256 */
1257 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1258 goto out;
1259
1260 handle = ext4_journal_start_sb(sb, 1);
1261 if (IS_ERR(handle)) {
1262 ret = PTR_ERR(handle);
1263 goto out;
1264 }
1265
1266 down_write(&grp->alloc_sem);
1267 /*
1268 * If inode bitmap was already initialized there may be some
1269 * used inodes so we need to skip blocks with used inodes in
1270 * inode table.
1271 */
1272 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1273 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1274 ext4_itable_unused_count(sb, gdp)),
1275 sbi->s_inodes_per_block);
1276
1277 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1278 ext4_error(sb, "Something is wrong with group %u\n"
1279 "Used itable blocks: %d"
1280 "itable unused count: %u\n",
1281 group, used_blks,
1282 ext4_itable_unused_count(sb, gdp));
1283 ret = 1;
1284 goto out;
1285 }
1286
1287 blk = ext4_inode_table(sb, gdp) + used_blks;
1288 num = sbi->s_itb_per_group - used_blks;
1289
1290 BUFFER_TRACE(group_desc_bh, "get_write_access");
1291 ret = ext4_journal_get_write_access(handle,
1292 group_desc_bh);
1293 if (ret)
1294 goto err_out;
1295
1296 /*
1297 * Skip zeroout if the inode table is full. But we set the ZEROED
1298 * flag anyway, because obviously, when it is full it does not need
1299 * further zeroing.
1300 */
1301 if (unlikely(num == 0))
1302 goto skip_zeroout;
1303
1304 ext4_debug("going to zero out inode table in group %d\n",
1305 group);
1306 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1307 if (ret < 0)
1308 goto err_out;
1309 if (barrier)
1310 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1311
1312skip_zeroout:
1313 ext4_lock_group(sb, group);
1314 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1315 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1316 ext4_unlock_group(sb, group);
1317
1318 BUFFER_TRACE(group_desc_bh,
1319 "call ext4_handle_dirty_metadata");
1320 ret = ext4_handle_dirty_metadata(handle, NULL,
1321 group_desc_bh);
1322
1323err_out:
1324 up_write(&grp->alloc_sem);
1325 ext4_journal_stop(handle);
1326out:
1327 return ret;
1328}
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 4b8debeb3965..bdbe69902207 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -53,6 +53,7 @@
53static inline int ext4_begin_ordered_truncate(struct inode *inode, 53static inline int ext4_begin_ordered_truncate(struct inode *inode,
54 loff_t new_size) 54 loff_t new_size)
55{ 55{
56 trace_ext4_begin_ordered_truncate(inode, new_size);
56 return jbd2_journal_begin_ordered_truncate( 57 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode->i_sb)->s_journal, 58 EXT4_SB(inode->i_sb)->s_journal,
58 &EXT4_I(inode)->jinode, 59 &EXT4_I(inode)->jinode,
@@ -60,6 +61,12 @@ static inline int ext4_begin_ordered_truncate(struct inode *inode,
60} 61}
61 62
62static void ext4_invalidatepage(struct page *page, unsigned long offset); 63static void ext4_invalidatepage(struct page *page, unsigned long offset);
64static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
65 struct buffer_head *bh_result, int create);
66static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
67static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
68static int __ext4_journalled_writepage(struct page *page, unsigned int len);
69static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
63 70
64/* 71/*
65 * Test whether an inode is a fast symlink. 72 * Test whether an inode is a fast symlink.
@@ -172,6 +179,7 @@ void ext4_evict_inode(struct inode *inode)
172 handle_t *handle; 179 handle_t *handle;
173 int err; 180 int err;
174 181
182 trace_ext4_evict_inode(inode);
175 if (inode->i_nlink) { 183 if (inode->i_nlink) {
176 truncate_inode_pages(&inode->i_data, 0); 184 truncate_inode_pages(&inode->i_data, 0);
177 goto no_delete; 185 goto no_delete;
@@ -755,6 +763,11 @@ static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
755 * parent to disk. 763 * parent to disk.
756 */ 764 */
757 bh = sb_getblk(inode->i_sb, new_blocks[n-1]); 765 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
766 if (unlikely(!bh)) {
767 err = -EIO;
768 goto failed;
769 }
770
758 branch[n].bh = bh; 771 branch[n].bh = bh;
759 lock_buffer(bh); 772 lock_buffer(bh);
760 BUFFER_TRACE(bh, "call get_create_access"); 773 BUFFER_TRACE(bh, "call get_create_access");
@@ -1207,8 +1220,10 @@ static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1207 break; 1220 break;
1208 idx++; 1221 idx++;
1209 num++; 1222 num++;
1210 if (num >= max_pages) 1223 if (num >= max_pages) {
1224 done = 1;
1211 break; 1225 break;
1226 }
1212 } 1227 }
1213 pagevec_release(&pvec); 1228 pagevec_release(&pvec);
1214 } 1229 }
@@ -1538,10 +1553,10 @@ static int do_journal_get_write_access(handle_t *handle,
1538 if (!buffer_mapped(bh) || buffer_freed(bh)) 1553 if (!buffer_mapped(bh) || buffer_freed(bh))
1539 return 0; 1554 return 0;
1540 /* 1555 /*
1541 * __block_prepare_write() could have dirtied some buffers. Clean 1556 * __block_write_begin() could have dirtied some buffers. Clean
1542 * the dirty bit as jbd2_journal_get_write_access() could complain 1557 * the dirty bit as jbd2_journal_get_write_access() could complain
1543 * otherwise about fs integrity issues. Setting of the dirty bit 1558 * otherwise about fs integrity issues. Setting of the dirty bit
1544 * by __block_prepare_write() isn't a real problem here as we clear 1559 * by __block_write_begin() isn't a real problem here as we clear
1545 * the bit before releasing a page lock and thus writeback cannot 1560 * the bit before releasing a page lock and thus writeback cannot
1546 * ever write the buffer. 1561 * ever write the buffer.
1547 */ 1562 */
@@ -1995,16 +2010,23 @@ static void ext4_da_page_release_reservation(struct page *page,
1995 * 2010 *
1996 * As pages are already locked by write_cache_pages(), we can't use it 2011 * As pages are already locked by write_cache_pages(), we can't use it
1997 */ 2012 */
1998static int mpage_da_submit_io(struct mpage_da_data *mpd) 2013static int mpage_da_submit_io(struct mpage_da_data *mpd,
2014 struct ext4_map_blocks *map)
1999{ 2015{
2000 long pages_skipped;
2001 struct pagevec pvec; 2016 struct pagevec pvec;
2002 unsigned long index, end; 2017 unsigned long index, end;
2003 int ret = 0, err, nr_pages, i; 2018 int ret = 0, err, nr_pages, i;
2004 struct inode *inode = mpd->inode; 2019 struct inode *inode = mpd->inode;
2005 struct address_space *mapping = inode->i_mapping; 2020 struct address_space *mapping = inode->i_mapping;
2021 loff_t size = i_size_read(inode);
2022 unsigned int len, block_start;
2023 struct buffer_head *bh, *page_bufs = NULL;
2024 int journal_data = ext4_should_journal_data(inode);
2025 sector_t pblock = 0, cur_logical = 0;
2026 struct ext4_io_submit io_submit;
2006 2027
2007 BUG_ON(mpd->next_page <= mpd->first_page); 2028 BUG_ON(mpd->next_page <= mpd->first_page);
2029 memset(&io_submit, 0, sizeof(io_submit));
2008 /* 2030 /*
2009 * We need to start from the first_page to the next_page - 1 2031 * We need to start from the first_page to the next_page - 1
2010 * to make sure we also write the mapped dirty buffer_heads. 2032 * to make sure we also write the mapped dirty buffer_heads.
@@ -2020,122 +2042,108 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd)
2020 if (nr_pages == 0) 2042 if (nr_pages == 0)
2021 break; 2043 break;
2022 for (i = 0; i < nr_pages; i++) { 2044 for (i = 0; i < nr_pages; i++) {
2045 int commit_write = 0, redirty_page = 0;
2023 struct page *page = pvec.pages[i]; 2046 struct page *page = pvec.pages[i];
2024 2047
2025 index = page->index; 2048 index = page->index;
2026 if (index > end) 2049 if (index > end)
2027 break; 2050 break;
2051
2052 if (index == size >> PAGE_CACHE_SHIFT)
2053 len = size & ~PAGE_CACHE_MASK;
2054 else
2055 len = PAGE_CACHE_SIZE;
2056 if (map) {
2057 cur_logical = index << (PAGE_CACHE_SHIFT -
2058 inode->i_blkbits);
2059 pblock = map->m_pblk + (cur_logical -
2060 map->m_lblk);
2061 }
2028 index++; 2062 index++;
2029 2063
2030 BUG_ON(!PageLocked(page)); 2064 BUG_ON(!PageLocked(page));
2031 BUG_ON(PageWriteback(page)); 2065 BUG_ON(PageWriteback(page));
2032 2066
2033 pages_skipped = mpd->wbc->pages_skipped;
2034 err = mapping->a_ops->writepage(page, mpd->wbc);
2035 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2036 /*
2037 * have successfully written the page
2038 * without skipping the same
2039 */
2040 mpd->pages_written++;
2041 /* 2067 /*
2042 * In error case, we have to continue because 2068 * If the page does not have buffers (for
2043 * remaining pages are still locked 2069 * whatever reason), try to create them using
2044 * XXX: unlock and re-dirty them? 2070 * __block_write_begin. If this fails,
2071 * redirty the page and move on.
2045 */ 2072 */
2046 if (ret == 0) 2073 if (!page_has_buffers(page)) {
2047 ret = err; 2074 if (__block_write_begin(page, 0, len,
2048 } 2075 noalloc_get_block_write)) {
2049 pagevec_release(&pvec); 2076 redirty_page:
2050 } 2077 redirty_page_for_writepage(mpd->wbc,
2051 return ret; 2078 page);
2052} 2079 unlock_page(page);
2053 2080 continue;
2054/* 2081 }
2055 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers 2082 commit_write = 1;
2056 * 2083 }
2057 * the function goes through all passed space and put actual disk
2058 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2059 */
2060static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd,
2061 struct ext4_map_blocks *map)
2062{
2063 struct inode *inode = mpd->inode;
2064 struct address_space *mapping = inode->i_mapping;
2065 int blocks = map->m_len;
2066 sector_t pblock = map->m_pblk, cur_logical;
2067 struct buffer_head *head, *bh;
2068 pgoff_t index, end;
2069 struct pagevec pvec;
2070 int nr_pages, i;
2071
2072 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2073 end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2074 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2075
2076 pagevec_init(&pvec, 0);
2077
2078 while (index <= end) {
2079 /* XXX: optimize tail */
2080 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2081 if (nr_pages == 0)
2082 break;
2083 for (i = 0; i < nr_pages; i++) {
2084 struct page *page = pvec.pages[i];
2085
2086 index = page->index;
2087 if (index > end)
2088 break;
2089 index++;
2090
2091 BUG_ON(!PageLocked(page));
2092 BUG_ON(PageWriteback(page));
2093 BUG_ON(!page_has_buffers(page));
2094
2095 bh = page_buffers(page);
2096 head = bh;
2097
2098 /* skip blocks out of the range */
2099 do {
2100 if (cur_logical >= map->m_lblk)
2101 break;
2102 cur_logical++;
2103 } while ((bh = bh->b_this_page) != head);
2104 2084
2085 bh = page_bufs = page_buffers(page);
2086 block_start = 0;
2105 do { 2087 do {
2106 if (cur_logical >= map->m_lblk + blocks) 2088 if (!bh)
2107 break; 2089 goto redirty_page;
2108 2090 if (map && (cur_logical >= map->m_lblk) &&
2109 if (buffer_delay(bh) || buffer_unwritten(bh)) { 2091 (cur_logical <= (map->m_lblk +
2110 2092 (map->m_len - 1)))) {
2111 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2112
2113 if (buffer_delay(bh)) { 2093 if (buffer_delay(bh)) {
2114 clear_buffer_delay(bh); 2094 clear_buffer_delay(bh);
2115 bh->b_blocknr = pblock; 2095 bh->b_blocknr = pblock;
2116 } else {
2117 /*
2118 * unwritten already should have
2119 * blocknr assigned. Verify that
2120 */
2121 clear_buffer_unwritten(bh);
2122 BUG_ON(bh->b_blocknr != pblock);
2123 } 2096 }
2097 if (buffer_unwritten(bh) ||
2098 buffer_mapped(bh))
2099 BUG_ON(bh->b_blocknr != pblock);
2100 if (map->m_flags & EXT4_MAP_UNINIT)
2101 set_buffer_uninit(bh);
2102 clear_buffer_unwritten(bh);
2103 }
2124 2104
2125 } else if (buffer_mapped(bh)) 2105 /* redirty page if block allocation undone */
2126 BUG_ON(bh->b_blocknr != pblock); 2106 if (buffer_delay(bh) || buffer_unwritten(bh))
2127 2107 redirty_page = 1;
2128 if (map->m_flags & EXT4_MAP_UNINIT) 2108 bh = bh->b_this_page;
2129 set_buffer_uninit(bh); 2109 block_start += bh->b_size;
2130 cur_logical++; 2110 cur_logical++;
2131 pblock++; 2111 pblock++;
2132 } while ((bh = bh->b_this_page) != head); 2112 } while (bh != page_bufs);
2113
2114 if (redirty_page)
2115 goto redirty_page;
2116
2117 if (commit_write)
2118 /* mark the buffer_heads as dirty & uptodate */
2119 block_commit_write(page, 0, len);
2120
2121 /*
2122 * Delalloc doesn't support data journalling,
2123 * but eventually maybe we'll lift this
2124 * restriction.
2125 */
2126 if (unlikely(journal_data && PageChecked(page)))
2127 err = __ext4_journalled_writepage(page, len);
2128 else
2129 err = ext4_bio_write_page(&io_submit, page,
2130 len, mpd->wbc);
2131
2132 if (!err)
2133 mpd->pages_written++;
2134 /*
2135 * In error case, we have to continue because
2136 * remaining pages are still locked
2137 */
2138 if (ret == 0)
2139 ret = err;
2133 } 2140 }
2134 pagevec_release(&pvec); 2141 pagevec_release(&pvec);
2135 } 2142 }
2143 ext4_io_submit(&io_submit);
2144 return ret;
2136} 2145}
2137 2146
2138
2139static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, 2147static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2140 sector_t logical, long blk_cnt) 2148 sector_t logical, long blk_cnt)
2141{ 2149{
@@ -2187,35 +2195,32 @@ static void ext4_print_free_blocks(struct inode *inode)
2187} 2195}
2188 2196
2189/* 2197/*
2190 * mpage_da_map_blocks - go through given space 2198 * mpage_da_map_and_submit - go through given space, map them
2199 * if necessary, and then submit them for I/O
2191 * 2200 *
2192 * @mpd - bh describing space 2201 * @mpd - bh describing space
2193 * 2202 *
2194 * The function skips space we know is already mapped to disk blocks. 2203 * The function skips space we know is already mapped to disk blocks.
2195 * 2204 *
2196 */ 2205 */
2197static int mpage_da_map_blocks(struct mpage_da_data *mpd) 2206static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
2198{ 2207{
2199 int err, blks, get_blocks_flags; 2208 int err, blks, get_blocks_flags;
2200 struct ext4_map_blocks map; 2209 struct ext4_map_blocks map, *mapp = NULL;
2201 sector_t next = mpd->b_blocknr; 2210 sector_t next = mpd->b_blocknr;
2202 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; 2211 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2203 loff_t disksize = EXT4_I(mpd->inode)->i_disksize; 2212 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2204 handle_t *handle = NULL; 2213 handle_t *handle = NULL;
2205 2214
2206 /* 2215 /*
2207 * We consider only non-mapped and non-allocated blocks 2216 * If the blocks are mapped already, or we couldn't accumulate
2217 * any blocks, then proceed immediately to the submission stage.
2208 */ 2218 */
2209 if ((mpd->b_state & (1 << BH_Mapped)) && 2219 if ((mpd->b_size == 0) ||
2210 !(mpd->b_state & (1 << BH_Delay)) && 2220 ((mpd->b_state & (1 << BH_Mapped)) &&
2211 !(mpd->b_state & (1 << BH_Unwritten))) 2221 !(mpd->b_state & (1 << BH_Delay)) &&
2212 return 0; 2222 !(mpd->b_state & (1 << BH_Unwritten))))
2213 2223 goto submit_io;
2214 /*
2215 * If we didn't accumulate anything to write simply return
2216 */
2217 if (!mpd->b_size)
2218 return 0;
2219 2224
2220 handle = ext4_journal_current_handle(); 2225 handle = ext4_journal_current_handle();
2221 BUG_ON(!handle); 2226 BUG_ON(!handle);
@@ -2252,17 +2257,18 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2252 2257
2253 err = blks; 2258 err = blks;
2254 /* 2259 /*
2255 * If get block returns with error we simply 2260 * If get block returns EAGAIN or ENOSPC and there
2256 * return. Later writepage will redirty the page and 2261 * appears to be free blocks we will call
2257 * writepages will find the dirty page again 2262 * ext4_writepage() for all of the pages which will
2263 * just redirty the pages.
2258 */ 2264 */
2259 if (err == -EAGAIN) 2265 if (err == -EAGAIN)
2260 return 0; 2266 goto submit_io;
2261 2267
2262 if (err == -ENOSPC && 2268 if (err == -ENOSPC &&
2263 ext4_count_free_blocks(sb)) { 2269 ext4_count_free_blocks(sb)) {
2264 mpd->retval = err; 2270 mpd->retval = err;
2265 return 0; 2271 goto submit_io;
2266 } 2272 }
2267 2273
2268 /* 2274 /*
@@ -2287,10 +2293,11 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2287 /* invalidate all the pages */ 2293 /* invalidate all the pages */
2288 ext4_da_block_invalidatepages(mpd, next, 2294 ext4_da_block_invalidatepages(mpd, next,
2289 mpd->b_size >> mpd->inode->i_blkbits); 2295 mpd->b_size >> mpd->inode->i_blkbits);
2290 return err; 2296 return;
2291 } 2297 }
2292 BUG_ON(blks == 0); 2298 BUG_ON(blks == 0);
2293 2299
2300 mapp = &map;
2294 if (map.m_flags & EXT4_MAP_NEW) { 2301 if (map.m_flags & EXT4_MAP_NEW) {
2295 struct block_device *bdev = mpd->inode->i_sb->s_bdev; 2302 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2296 int i; 2303 int i;
@@ -2299,18 +2306,11 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2299 unmap_underlying_metadata(bdev, map.m_pblk + i); 2306 unmap_underlying_metadata(bdev, map.m_pblk + i);
2300 } 2307 }
2301 2308
2302 /*
2303 * If blocks are delayed marked, we need to
2304 * put actual blocknr and drop delayed bit
2305 */
2306 if ((mpd->b_state & (1 << BH_Delay)) ||
2307 (mpd->b_state & (1 << BH_Unwritten)))
2308 mpage_put_bnr_to_bhs(mpd, &map);
2309
2310 if (ext4_should_order_data(mpd->inode)) { 2309 if (ext4_should_order_data(mpd->inode)) {
2311 err = ext4_jbd2_file_inode(handle, mpd->inode); 2310 err = ext4_jbd2_file_inode(handle, mpd->inode);
2312 if (err) 2311 if (err)
2313 return err; 2312 /* This only happens if the journal is aborted */
2313 return;
2314 } 2314 }
2315 2315
2316 /* 2316 /*
@@ -2321,10 +2321,16 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2321 disksize = i_size_read(mpd->inode); 2321 disksize = i_size_read(mpd->inode);
2322 if (disksize > EXT4_I(mpd->inode)->i_disksize) { 2322 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2323 ext4_update_i_disksize(mpd->inode, disksize); 2323 ext4_update_i_disksize(mpd->inode, disksize);
2324 return ext4_mark_inode_dirty(handle, mpd->inode); 2324 err = ext4_mark_inode_dirty(handle, mpd->inode);
2325 if (err)
2326 ext4_error(mpd->inode->i_sb,
2327 "Failed to mark inode %lu dirty",
2328 mpd->inode->i_ino);
2325 } 2329 }
2326 2330
2327 return 0; 2331submit_io:
2332 mpage_da_submit_io(mpd, mapp);
2333 mpd->io_done = 1;
2328} 2334}
2329 2335
2330#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ 2336#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
@@ -2401,9 +2407,7 @@ flush_it:
2401 * We couldn't merge the block to our extent, so we 2407 * We couldn't merge the block to our extent, so we
2402 * need to flush current extent and start new one 2408 * need to flush current extent and start new one
2403 */ 2409 */
2404 if (mpage_da_map_blocks(mpd) == 0) 2410 mpage_da_map_and_submit(mpd);
2405 mpage_da_submit_io(mpd);
2406 mpd->io_done = 1;
2407 return; 2411 return;
2408} 2412}
2409 2413
@@ -2422,9 +2426,9 @@ static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2422 * The function finds extents of pages and scan them for all blocks. 2426 * The function finds extents of pages and scan them for all blocks.
2423 */ 2427 */
2424static int __mpage_da_writepage(struct page *page, 2428static int __mpage_da_writepage(struct page *page,
2425 struct writeback_control *wbc, void *data) 2429 struct writeback_control *wbc,
2430 struct mpage_da_data *mpd)
2426{ 2431{
2427 struct mpage_da_data *mpd = data;
2428 struct inode *inode = mpd->inode; 2432 struct inode *inode = mpd->inode;
2429 struct buffer_head *bh, *head; 2433 struct buffer_head *bh, *head;
2430 sector_t logical; 2434 sector_t logical;
@@ -2435,15 +2439,13 @@ static int __mpage_da_writepage(struct page *page,
2435 if (mpd->next_page != page->index) { 2439 if (mpd->next_page != page->index) {
2436 /* 2440 /*
2437 * Nope, we can't. So, we map non-allocated blocks 2441 * Nope, we can't. So, we map non-allocated blocks
2438 * and start IO on them using writepage() 2442 * and start IO on them
2439 */ 2443 */
2440 if (mpd->next_page != mpd->first_page) { 2444 if (mpd->next_page != mpd->first_page) {
2441 if (mpage_da_map_blocks(mpd) == 0) 2445 mpage_da_map_and_submit(mpd);
2442 mpage_da_submit_io(mpd);
2443 /* 2446 /*
2444 * skip rest of the page in the page_vec 2447 * skip rest of the page in the page_vec
2445 */ 2448 */
2446 mpd->io_done = 1;
2447 redirty_page_for_writepage(wbc, page); 2449 redirty_page_for_writepage(wbc, page);
2448 unlock_page(page); 2450 unlock_page(page);
2449 return MPAGE_DA_EXTENT_TAIL; 2451 return MPAGE_DA_EXTENT_TAIL;
@@ -2550,8 +2552,7 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2550 if (buffer_delay(bh)) 2552 if (buffer_delay(bh))
2551 return 0; /* Not sure this could or should happen */ 2553 return 0; /* Not sure this could or should happen */
2552 /* 2554 /*
2553 * XXX: __block_prepare_write() unmaps passed block, 2555 * XXX: __block_write_begin() unmaps passed block, is it OK?
2554 * is it OK?
2555 */ 2556 */
2556 ret = ext4_da_reserve_space(inode, iblock); 2557 ret = ext4_da_reserve_space(inode, iblock);
2557 if (ret) 2558 if (ret)
@@ -2583,7 +2584,7 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2583/* 2584/*
2584 * This function is used as a standard get_block_t calback function 2585 * This function is used as a standard get_block_t calback function
2585 * when there is no desire to allocate any blocks. It is used as a 2586 * when there is no desire to allocate any blocks. It is used as a
2586 * callback function for block_prepare_write() and block_write_full_page(). 2587 * callback function for block_write_begin() and block_write_full_page().
2587 * These functions should only try to map a single block at a time. 2588 * These functions should only try to map a single block at a time.
2588 * 2589 *
2589 * Since this function doesn't do block allocations even if the caller 2590 * Since this function doesn't do block allocations even if the caller
@@ -2623,6 +2624,7 @@ static int __ext4_journalled_writepage(struct page *page,
2623 int ret = 0; 2624 int ret = 0;
2624 int err; 2625 int err;
2625 2626
2627 ClearPageChecked(page);
2626 page_bufs = page_buffers(page); 2628 page_bufs = page_buffers(page);
2627 BUG_ON(!page_bufs); 2629 BUG_ON(!page_bufs);
2628 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); 2630 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
@@ -2700,7 +2702,7 @@ static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2700static int ext4_writepage(struct page *page, 2702static int ext4_writepage(struct page *page,
2701 struct writeback_control *wbc) 2703 struct writeback_control *wbc)
2702{ 2704{
2703 int ret = 0; 2705 int ret = 0, commit_write = 0;
2704 loff_t size; 2706 loff_t size;
2705 unsigned int len; 2707 unsigned int len;
2706 struct buffer_head *page_bufs = NULL; 2708 struct buffer_head *page_bufs = NULL;
@@ -2713,71 +2715,44 @@ static int ext4_writepage(struct page *page,
2713 else 2715 else
2714 len = PAGE_CACHE_SIZE; 2716 len = PAGE_CACHE_SIZE;
2715 2717
2716 if (page_has_buffers(page)) { 2718 /*
2717 page_bufs = page_buffers(page); 2719 * If the page does not have buffers (for whatever reason),
2718 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, 2720 * try to create them using __block_write_begin. If this
2719 ext4_bh_delay_or_unwritten)) { 2721 * fails, redirty the page and move on.
2720 /* 2722 */
2721 * We don't want to do block allocation 2723 if (!page_has_buffers(page)) {
2722 * So redirty the page and return 2724 if (__block_write_begin(page, 0, len,
2723 * We may reach here when we do a journal commit 2725 noalloc_get_block_write)) {
2724 * via journal_submit_inode_data_buffers. 2726 redirty_page:
2725 * If we don't have mapping block we just ignore
2726 * them. We can also reach here via shrink_page_list
2727 */
2728 redirty_page_for_writepage(wbc, page); 2727 redirty_page_for_writepage(wbc, page);
2729 unlock_page(page); 2728 unlock_page(page);
2730 return 0; 2729 return 0;
2731 } 2730 }
2732 } else { 2731 commit_write = 1;
2732 }
2733 page_bufs = page_buffers(page);
2734 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2735 ext4_bh_delay_or_unwritten)) {
2733 /* 2736 /*
2734 * The test for page_has_buffers() is subtle: 2737 * We don't want to do block allocation, so redirty
2735 * We know the page is dirty but it lost buffers. That means 2738 * the page and return. We may reach here when we do
2736 * that at some moment in time after write_begin()/write_end() 2739 * a journal commit via journal_submit_inode_data_buffers.
2737 * has been called all buffers have been clean and thus they 2740 * We can also reach here via shrink_page_list
2738 * must have been written at least once. So they are all
2739 * mapped and we can happily proceed with mapping them
2740 * and writing the page.
2741 *
2742 * Try to initialize the buffer_heads and check whether
2743 * all are mapped and non delay. We don't want to
2744 * do block allocation here.
2745 */ 2741 */
2746 ret = block_prepare_write(page, 0, len, 2742 goto redirty_page;
2747 noalloc_get_block_write); 2743 }
2748 if (!ret) { 2744 if (commit_write)
2749 page_bufs = page_buffers(page);
2750 /* check whether all are mapped and non delay */
2751 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2752 ext4_bh_delay_or_unwritten)) {
2753 redirty_page_for_writepage(wbc, page);
2754 unlock_page(page);
2755 return 0;
2756 }
2757 } else {
2758 /*
2759 * We can't do block allocation here
2760 * so just redity the page and unlock
2761 * and return
2762 */
2763 redirty_page_for_writepage(wbc, page);
2764 unlock_page(page);
2765 return 0;
2766 }
2767 /* now mark the buffer_heads as dirty and uptodate */ 2745 /* now mark the buffer_heads as dirty and uptodate */
2768 block_commit_write(page, 0, len); 2746 block_commit_write(page, 0, len);
2769 }
2770 2747
2771 if (PageChecked(page) && ext4_should_journal_data(inode)) { 2748 if (PageChecked(page) && ext4_should_journal_data(inode))
2772 /* 2749 /*
2773 * It's mmapped pagecache. Add buffers and journal it. There 2750 * It's mmapped pagecache. Add buffers and journal it. There
2774 * doesn't seem much point in redirtying the page here. 2751 * doesn't seem much point in redirtying the page here.
2775 */ 2752 */
2776 ClearPageChecked(page);
2777 return __ext4_journalled_writepage(page, len); 2753 return __ext4_journalled_writepage(page, len);
2778 }
2779 2754
2780 if (page_bufs && buffer_uninit(page_bufs)) { 2755 if (buffer_uninit(page_bufs)) {
2781 ext4_set_bh_endio(page_bufs, inode); 2756 ext4_set_bh_endio(page_bufs, inode);
2782 ret = block_write_full_page_endio(page, noalloc_get_block_write, 2757 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2783 wbc, ext4_end_io_buffer_write); 2758 wbc, ext4_end_io_buffer_write);
@@ -2824,25 +2799,32 @@ static int ext4_da_writepages_trans_blocks(struct inode *inode)
2824 */ 2799 */
2825static int write_cache_pages_da(struct address_space *mapping, 2800static int write_cache_pages_da(struct address_space *mapping,
2826 struct writeback_control *wbc, 2801 struct writeback_control *wbc,
2827 struct mpage_da_data *mpd) 2802 struct mpage_da_data *mpd,
2803 pgoff_t *done_index)
2828{ 2804{
2829 int ret = 0; 2805 int ret = 0;
2830 int done = 0; 2806 int done = 0;
2831 struct pagevec pvec; 2807 struct pagevec pvec;
2832 int nr_pages; 2808 unsigned nr_pages;
2833 pgoff_t index; 2809 pgoff_t index;
2834 pgoff_t end; /* Inclusive */ 2810 pgoff_t end; /* Inclusive */
2835 long nr_to_write = wbc->nr_to_write; 2811 long nr_to_write = wbc->nr_to_write;
2812 int tag;
2836 2813
2837 pagevec_init(&pvec, 0); 2814 pagevec_init(&pvec, 0);
2838 index = wbc->range_start >> PAGE_CACHE_SHIFT; 2815 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2839 end = wbc->range_end >> PAGE_CACHE_SHIFT; 2816 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2840 2817
2818 if (wbc->sync_mode == WB_SYNC_ALL)
2819 tag = PAGECACHE_TAG_TOWRITE;
2820 else
2821 tag = PAGECACHE_TAG_DIRTY;
2822
2823 *done_index = index;
2841 while (!done && (index <= end)) { 2824 while (!done && (index <= end)) {
2842 int i; 2825 int i;
2843 2826
2844 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 2827 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2845 PAGECACHE_TAG_DIRTY,
2846 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 2828 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2847 if (nr_pages == 0) 2829 if (nr_pages == 0)
2848 break; 2830 break;
@@ -2862,6 +2844,8 @@ static int write_cache_pages_da(struct address_space *mapping,
2862 break; 2844 break;
2863 } 2845 }
2864 2846
2847 *done_index = page->index + 1;
2848
2865 lock_page(page); 2849 lock_page(page);
2866 2850
2867 /* 2851 /*
@@ -2947,6 +2931,8 @@ static int ext4_da_writepages(struct address_space *mapping,
2947 long desired_nr_to_write, nr_to_writebump = 0; 2931 long desired_nr_to_write, nr_to_writebump = 0;
2948 loff_t range_start = wbc->range_start; 2932 loff_t range_start = wbc->range_start;
2949 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); 2933 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2934 pgoff_t done_index = 0;
2935 pgoff_t end;
2950 2936
2951 trace_ext4_da_writepages(inode, wbc); 2937 trace_ext4_da_writepages(inode, wbc);
2952 2938
@@ -2982,8 +2968,11 @@ static int ext4_da_writepages(struct address_space *mapping,
2982 wbc->range_start = index << PAGE_CACHE_SHIFT; 2968 wbc->range_start = index << PAGE_CACHE_SHIFT;
2983 wbc->range_end = LLONG_MAX; 2969 wbc->range_end = LLONG_MAX;
2984 wbc->range_cyclic = 0; 2970 wbc->range_cyclic = 0;
2985 } else 2971 end = -1;
2972 } else {
2986 index = wbc->range_start >> PAGE_CACHE_SHIFT; 2973 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2974 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2975 }
2987 2976
2988 /* 2977 /*
2989 * This works around two forms of stupidity. The first is in 2978 * This works around two forms of stupidity. The first is in
@@ -3002,9 +2991,12 @@ static int ext4_da_writepages(struct address_space *mapping,
3002 * sbi->max_writeback_mb_bump whichever is smaller. 2991 * sbi->max_writeback_mb_bump whichever is smaller.
3003 */ 2992 */
3004 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); 2993 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
3005 if (!range_cyclic && range_whole) 2994 if (!range_cyclic && range_whole) {
3006 desired_nr_to_write = wbc->nr_to_write * 8; 2995 if (wbc->nr_to_write == LONG_MAX)
3007 else 2996 desired_nr_to_write = wbc->nr_to_write;
2997 else
2998 desired_nr_to_write = wbc->nr_to_write * 8;
2999 } else
3008 desired_nr_to_write = ext4_num_dirty_pages(inode, index, 3000 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
3009 max_pages); 3001 max_pages);
3010 if (desired_nr_to_write > max_pages) 3002 if (desired_nr_to_write > max_pages)
@@ -3021,6 +3013,9 @@ static int ext4_da_writepages(struct address_space *mapping,
3021 pages_skipped = wbc->pages_skipped; 3013 pages_skipped = wbc->pages_skipped;
3022 3014
3023retry: 3015retry:
3016 if (wbc->sync_mode == WB_SYNC_ALL)
3017 tag_pages_for_writeback(mapping, index, end);
3018
3024 while (!ret && wbc->nr_to_write > 0) { 3019 while (!ret && wbc->nr_to_write > 0) {
3025 3020
3026 /* 3021 /*
@@ -3059,16 +3054,14 @@ retry:
3059 mpd.io_done = 0; 3054 mpd.io_done = 0;
3060 mpd.pages_written = 0; 3055 mpd.pages_written = 0;
3061 mpd.retval = 0; 3056 mpd.retval = 0;
3062 ret = write_cache_pages_da(mapping, wbc, &mpd); 3057 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
3063 /* 3058 /*
3064 * If we have a contiguous extent of pages and we 3059 * If we have a contiguous extent of pages and we
3065 * haven't done the I/O yet, map the blocks and submit 3060 * haven't done the I/O yet, map the blocks and submit
3066 * them for I/O. 3061 * them for I/O.
3067 */ 3062 */
3068 if (!mpd.io_done && mpd.next_page != mpd.first_page) { 3063 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3069 if (mpage_da_map_blocks(&mpd) == 0) 3064 mpage_da_map_and_submit(&mpd);
3070 mpage_da_submit_io(&mpd);
3071 mpd.io_done = 1;
3072 ret = MPAGE_DA_EXTENT_TAIL; 3065 ret = MPAGE_DA_EXTENT_TAIL;
3073 } 3066 }
3074 trace_ext4_da_write_pages(inode, &mpd); 3067 trace_ext4_da_write_pages(inode, &mpd);
@@ -3115,14 +3108,13 @@ retry:
3115 __func__, wbc->nr_to_write, ret); 3108 __func__, wbc->nr_to_write, ret);
3116 3109
3117 /* Update index */ 3110 /* Update index */
3118 index += pages_written;
3119 wbc->range_cyclic = range_cyclic; 3111 wbc->range_cyclic = range_cyclic;
3120 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 3112 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3121 /* 3113 /*
3122 * set the writeback_index so that range_cyclic 3114 * set the writeback_index so that range_cyclic
3123 * mode will write it back later 3115 * mode will write it back later
3124 */ 3116 */
3125 mapping->writeback_index = index; 3117 mapping->writeback_index = done_index;
3126 3118
3127out_writepages: 3119out_writepages:
3128 wbc->nr_to_write -= nr_to_writebump; 3120 wbc->nr_to_write -= nr_to_writebump;
@@ -3457,15 +3449,6 @@ ext4_readpages(struct file *file, struct address_space *mapping,
3457 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); 3449 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3458} 3450}
3459 3451
3460static void ext4_free_io_end(ext4_io_end_t *io)
3461{
3462 BUG_ON(!io);
3463 if (io->page)
3464 put_page(io->page);
3465 iput(io->inode);
3466 kfree(io);
3467}
3468
3469static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) 3452static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3470{ 3453{
3471 struct buffer_head *head, *bh; 3454 struct buffer_head *head, *bh;
@@ -3642,173 +3625,6 @@ static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3642 EXT4_GET_BLOCKS_IO_CREATE_EXT); 3625 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3643} 3626}
3644 3627
3645static void dump_completed_IO(struct inode * inode)
3646{
3647#ifdef EXT4_DEBUG
3648 struct list_head *cur, *before, *after;
3649 ext4_io_end_t *io, *io0, *io1;
3650 unsigned long flags;
3651
3652 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3653 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3654 return;
3655 }
3656
3657 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3658 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3659 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3660 cur = &io->list;
3661 before = cur->prev;
3662 io0 = container_of(before, ext4_io_end_t, list);
3663 after = cur->next;
3664 io1 = container_of(after, ext4_io_end_t, list);
3665
3666 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3667 io, inode->i_ino, io0, io1);
3668 }
3669 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3670#endif
3671}
3672
3673/*
3674 * check a range of space and convert unwritten extents to written.
3675 */
3676static int ext4_end_io_nolock(ext4_io_end_t *io)
3677{
3678 struct inode *inode = io->inode;
3679 loff_t offset = io->offset;
3680 ssize_t size = io->size;
3681 int ret = 0;
3682
3683 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3684 "list->prev 0x%p\n",
3685 io, inode->i_ino, io->list.next, io->list.prev);
3686
3687 if (list_empty(&io->list))
3688 return ret;
3689
3690 if (io->flag != EXT4_IO_UNWRITTEN)
3691 return ret;
3692
3693 ret = ext4_convert_unwritten_extents(inode, offset, size);
3694 if (ret < 0) {
3695 printk(KERN_EMERG "%s: failed to convert unwritten"
3696 "extents to written extents, error is %d"
3697 " io is still on inode %lu aio dio list\n",
3698 __func__, ret, inode->i_ino);
3699 return ret;
3700 }
3701
3702 if (io->iocb)
3703 aio_complete(io->iocb, io->result, 0);
3704 /* clear the DIO AIO unwritten flag */
3705 io->flag = 0;
3706 return ret;
3707}
3708
3709/*
3710 * work on completed aio dio IO, to convert unwritten extents to extents
3711 */
3712static void ext4_end_io_work(struct work_struct *work)
3713{
3714 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3715 struct inode *inode = io->inode;
3716 struct ext4_inode_info *ei = EXT4_I(inode);
3717 unsigned long flags;
3718 int ret;
3719
3720 mutex_lock(&inode->i_mutex);
3721 ret = ext4_end_io_nolock(io);
3722 if (ret < 0) {
3723 mutex_unlock(&inode->i_mutex);
3724 return;
3725 }
3726
3727 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3728 if (!list_empty(&io->list))
3729 list_del_init(&io->list);
3730 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3731 mutex_unlock(&inode->i_mutex);
3732 ext4_free_io_end(io);
3733}
3734
3735/*
3736 * This function is called from ext4_sync_file().
3737 *
3738 * When IO is completed, the work to convert unwritten extents to
3739 * written is queued on workqueue but may not get immediately
3740 * scheduled. When fsync is called, we need to ensure the
3741 * conversion is complete before fsync returns.
3742 * The inode keeps track of a list of pending/completed IO that
3743 * might needs to do the conversion. This function walks through
3744 * the list and convert the related unwritten extents for completed IO
3745 * to written.
3746 * The function return the number of pending IOs on success.
3747 */
3748int flush_completed_IO(struct inode *inode)
3749{
3750 ext4_io_end_t *io;
3751 struct ext4_inode_info *ei = EXT4_I(inode);
3752 unsigned long flags;
3753 int ret = 0;
3754 int ret2 = 0;
3755
3756 if (list_empty(&ei->i_completed_io_list))
3757 return ret;
3758
3759 dump_completed_IO(inode);
3760 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3761 while (!list_empty(&ei->i_completed_io_list)){
3762 io = list_entry(ei->i_completed_io_list.next,
3763 ext4_io_end_t, list);
3764 /*
3765 * Calling ext4_end_io_nolock() to convert completed
3766 * IO to written.
3767 *
3768 * When ext4_sync_file() is called, run_queue() may already
3769 * about to flush the work corresponding to this io structure.
3770 * It will be upset if it founds the io structure related
3771 * to the work-to-be schedule is freed.
3772 *
3773 * Thus we need to keep the io structure still valid here after
3774 * convertion finished. The io structure has a flag to
3775 * avoid double converting from both fsync and background work
3776 * queue work.
3777 */
3778 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3779 ret = ext4_end_io_nolock(io);
3780 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3781 if (ret < 0)
3782 ret2 = ret;
3783 else
3784 list_del_init(&io->list);
3785 }
3786 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3787 return (ret2 < 0) ? ret2 : 0;
3788}
3789
3790static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3791{
3792 ext4_io_end_t *io = NULL;
3793
3794 io = kmalloc(sizeof(*io), flags);
3795
3796 if (io) {
3797 igrab(inode);
3798 io->inode = inode;
3799 io->flag = 0;
3800 io->offset = 0;
3801 io->size = 0;
3802 io->page = NULL;
3803 io->iocb = NULL;
3804 io->result = 0;
3805 INIT_WORK(&io->work, ext4_end_io_work);
3806 INIT_LIST_HEAD(&io->list);
3807 }
3808
3809 return io;
3810}
3811
3812static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, 3628static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3813 ssize_t size, void *private, int ret, 3629 ssize_t size, void *private, int ret,
3814 bool is_async) 3630 bool is_async)
@@ -3828,7 +3644,7 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3828 size); 3644 size);
3829 3645
3830 /* if not aio dio with unwritten extents, just free io and return */ 3646 /* if not aio dio with unwritten extents, just free io and return */
3831 if (io_end->flag != EXT4_IO_UNWRITTEN){ 3647 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3832 ext4_free_io_end(io_end); 3648 ext4_free_io_end(io_end);
3833 iocb->private = NULL; 3649 iocb->private = NULL;
3834out: 3650out:
@@ -3845,14 +3661,14 @@ out:
3845 } 3661 }
3846 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; 3662 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3847 3663
3848 /* queue the work to convert unwritten extents to written */
3849 queue_work(wq, &io_end->work);
3850
3851 /* Add the io_end to per-inode completed aio dio list*/ 3664 /* Add the io_end to per-inode completed aio dio list*/
3852 ei = EXT4_I(io_end->inode); 3665 ei = EXT4_I(io_end->inode);
3853 spin_lock_irqsave(&ei->i_completed_io_lock, flags); 3666 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3854 list_add_tail(&io_end->list, &ei->i_completed_io_list); 3667 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3855 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 3668 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3669
3670 /* queue the work to convert unwritten extents to written */
3671 queue_work(wq, &io_end->work);
3856 iocb->private = NULL; 3672 iocb->private = NULL;
3857} 3673}
3858 3674
@@ -3873,7 +3689,7 @@ static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3873 goto out; 3689 goto out;
3874 } 3690 }
3875 3691
3876 io_end->flag = EXT4_IO_UNWRITTEN; 3692 io_end->flag = EXT4_IO_END_UNWRITTEN;
3877 inode = io_end->inode; 3693 inode = io_end->inode;
3878 3694
3879 /* Add the io_end to per-inode completed io list*/ 3695 /* Add the io_end to per-inode completed io list*/
@@ -5464,6 +5280,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5464{ 5280{
5465 struct inode *inode = dentry->d_inode; 5281 struct inode *inode = dentry->d_inode;
5466 int error, rc = 0; 5282 int error, rc = 0;
5283 int orphan = 0;
5467 const unsigned int ia_valid = attr->ia_valid; 5284 const unsigned int ia_valid = attr->ia_valid;
5468 5285
5469 error = inode_change_ok(inode, attr); 5286 error = inode_change_ok(inode, attr);
@@ -5519,8 +5336,10 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5519 error = PTR_ERR(handle); 5336 error = PTR_ERR(handle);
5520 goto err_out; 5337 goto err_out;
5521 } 5338 }
5522 5339 if (ext4_handle_valid(handle)) {
5523 error = ext4_orphan_add(handle, inode); 5340 error = ext4_orphan_add(handle, inode);
5341 orphan = 1;
5342 }
5524 EXT4_I(inode)->i_disksize = attr->ia_size; 5343 EXT4_I(inode)->i_disksize = attr->ia_size;
5525 rc = ext4_mark_inode_dirty(handle, inode); 5344 rc = ext4_mark_inode_dirty(handle, inode);
5526 if (!error) 5345 if (!error)
@@ -5538,6 +5357,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5538 goto err_out; 5357 goto err_out;
5539 } 5358 }
5540 ext4_orphan_del(handle, inode); 5359 ext4_orphan_del(handle, inode);
5360 orphan = 0;
5541 ext4_journal_stop(handle); 5361 ext4_journal_stop(handle);
5542 goto err_out; 5362 goto err_out;
5543 } 5363 }
@@ -5560,7 +5380,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5560 * If the call to ext4_truncate failed to get a transaction handle at 5380 * If the call to ext4_truncate failed to get a transaction handle at
5561 * all, we need to clean up the in-core orphan list manually. 5381 * all, we need to clean up the in-core orphan list manually.
5562 */ 5382 */
5563 if (inode->i_nlink) 5383 if (orphan && inode->i_nlink)
5564 ext4_orphan_del(NULL, inode); 5384 ext4_orphan_del(NULL, inode);
5565 5385
5566 if (!rc && (ia_valid & ATTR_MODE)) 5386 if (!rc && (ia_valid & ATTR_MODE))
@@ -5592,9 +5412,7 @@ int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5592 * will return the blocks that include the delayed allocation 5412 * will return the blocks that include the delayed allocation
5593 * blocks for this file. 5413 * blocks for this file.
5594 */ 5414 */
5595 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5596 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; 5415 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5597 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5598 5416
5599 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; 5417 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5600 return 0; 5418 return 0;
@@ -5643,7 +5461,7 @@ static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5643 * 5461 *
5644 * Also account for superblock, inode, quota and xattr blocks 5462 * Also account for superblock, inode, quota and xattr blocks
5645 */ 5463 */
5646int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) 5464static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5647{ 5465{
5648 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); 5466 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5649 int gdpblocks; 5467 int gdpblocks;
@@ -5831,6 +5649,7 @@ int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5831 int err, ret; 5649 int err, ret;
5832 5650
5833 might_sleep(); 5651 might_sleep();
5652 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5834 err = ext4_reserve_inode_write(handle, inode, &iloc); 5653 err = ext4_reserve_inode_write(handle, inode, &iloc);
5835 if (ext4_handle_valid(handle) && 5654 if (ext4_handle_valid(handle) &&
5836 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && 5655 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
index bf5ae883b1bd..eb3bc2fe647e 100644
--- a/fs/ext4/ioctl.c
+++ b/fs/ext4/ioctl.c
@@ -331,6 +331,30 @@ mext_out:
331 return err; 331 return err;
332 } 332 }
333 333
334 case FITRIM:
335 {
336 struct super_block *sb = inode->i_sb;
337 struct fstrim_range range;
338 int ret = 0;
339
340 if (!capable(CAP_SYS_ADMIN))
341 return -EPERM;
342
343 if (copy_from_user(&range, (struct fstrim_range *)arg,
344 sizeof(range)))
345 return -EFAULT;
346
347 ret = ext4_trim_fs(sb, &range);
348 if (ret < 0)
349 return ret;
350
351 if (copy_to_user((struct fstrim_range *)arg, &range,
352 sizeof(range)))
353 return -EFAULT;
354
355 return 0;
356 }
357
334 default: 358 default:
335 return -ENOTTY; 359 return -ENOTTY;
336 } 360 }
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 19aa0d44d822..5b4d4e3a4d58 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -338,6 +338,14 @@
338static struct kmem_cache *ext4_pspace_cachep; 338static struct kmem_cache *ext4_pspace_cachep;
339static struct kmem_cache *ext4_ac_cachep; 339static struct kmem_cache *ext4_ac_cachep;
340static struct kmem_cache *ext4_free_ext_cachep; 340static struct kmem_cache *ext4_free_ext_cachep;
341
342/* We create slab caches for groupinfo data structures based on the
343 * superblock block size. There will be one per mounted filesystem for
344 * each unique s_blocksize_bits */
345#define NR_GRPINFO_CACHES \
346 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE + 1)
347static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];
348
341static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap, 349static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
342 ext4_group_t group); 350 ext4_group_t group);
343static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap, 351static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
@@ -939,6 +947,85 @@ out:
939} 947}
940 948
941/* 949/*
950 * lock the group_info alloc_sem of all the groups
951 * belonging to the same buddy cache page. This
952 * make sure other parallel operation on the buddy
953 * cache doesn't happen whild holding the buddy cache
954 * lock
955 */
956static int ext4_mb_get_buddy_cache_lock(struct super_block *sb,
957 ext4_group_t group)
958{
959 int i;
960 int block, pnum;
961 int blocks_per_page;
962 int groups_per_page;
963 ext4_group_t ngroups = ext4_get_groups_count(sb);
964 ext4_group_t first_group;
965 struct ext4_group_info *grp;
966
967 blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
968 /*
969 * the buddy cache inode stores the block bitmap
970 * and buddy information in consecutive blocks.
971 * So for each group we need two blocks.
972 */
973 block = group * 2;
974 pnum = block / blocks_per_page;
975 first_group = pnum * blocks_per_page / 2;
976
977 groups_per_page = blocks_per_page >> 1;
978 if (groups_per_page == 0)
979 groups_per_page = 1;
980 /* read all groups the page covers into the cache */
981 for (i = 0; i < groups_per_page; i++) {
982
983 if ((first_group + i) >= ngroups)
984 break;
985 grp = ext4_get_group_info(sb, first_group + i);
986 /* take all groups write allocation
987 * semaphore. This make sure there is
988 * no block allocation going on in any
989 * of that groups
990 */
991 down_write_nested(&grp->alloc_sem, i);
992 }
993 return i;
994}
995
996static void ext4_mb_put_buddy_cache_lock(struct super_block *sb,
997 ext4_group_t group, int locked_group)
998{
999 int i;
1000 int block, pnum;
1001 int blocks_per_page;
1002 ext4_group_t first_group;
1003 struct ext4_group_info *grp;
1004
1005 blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1006 /*
1007 * the buddy cache inode stores the block bitmap
1008 * and buddy information in consecutive blocks.
1009 * So for each group we need two blocks.
1010 */
1011 block = group * 2;
1012 pnum = block / blocks_per_page;
1013 first_group = pnum * blocks_per_page / 2;
1014 /* release locks on all the groups */
1015 for (i = 0; i < locked_group; i++) {
1016
1017 grp = ext4_get_group_info(sb, first_group + i);
1018 /* take all groups write allocation
1019 * semaphore. This make sure there is
1020 * no block allocation going on in any
1021 * of that groups
1022 */
1023 up_write(&grp->alloc_sem);
1024 }
1025
1026}
1027
1028/*
942 * Locking note: This routine calls ext4_mb_init_cache(), which takes the 1029 * Locking note: This routine calls ext4_mb_init_cache(), which takes the
943 * block group lock of all groups for this page; do not hold the BG lock when 1030 * block group lock of all groups for this page; do not hold the BG lock when
944 * calling this routine! 1031 * calling this routine!
@@ -1915,84 +2002,6 @@ static int ext4_mb_good_group(struct ext4_allocation_context *ac,
1915 return 0; 2002 return 0;
1916} 2003}
1917 2004
1918/*
1919 * lock the group_info alloc_sem of all the groups
1920 * belonging to the same buddy cache page. This
1921 * make sure other parallel operation on the buddy
1922 * cache doesn't happen whild holding the buddy cache
1923 * lock
1924 */
1925int ext4_mb_get_buddy_cache_lock(struct super_block *sb, ext4_group_t group)
1926{
1927 int i;
1928 int block, pnum;
1929 int blocks_per_page;
1930 int groups_per_page;
1931 ext4_group_t ngroups = ext4_get_groups_count(sb);
1932 ext4_group_t first_group;
1933 struct ext4_group_info *grp;
1934
1935 blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1936 /*
1937 * the buddy cache inode stores the block bitmap
1938 * and buddy information in consecutive blocks.
1939 * So for each group we need two blocks.
1940 */
1941 block = group * 2;
1942 pnum = block / blocks_per_page;
1943 first_group = pnum * blocks_per_page / 2;
1944
1945 groups_per_page = blocks_per_page >> 1;
1946 if (groups_per_page == 0)
1947 groups_per_page = 1;
1948 /* read all groups the page covers into the cache */
1949 for (i = 0; i < groups_per_page; i++) {
1950
1951 if ((first_group + i) >= ngroups)
1952 break;
1953 grp = ext4_get_group_info(sb, first_group + i);
1954 /* take all groups write allocation
1955 * semaphore. This make sure there is
1956 * no block allocation going on in any
1957 * of that groups
1958 */
1959 down_write_nested(&grp->alloc_sem, i);
1960 }
1961 return i;
1962}
1963
1964void ext4_mb_put_buddy_cache_lock(struct super_block *sb,
1965 ext4_group_t group, int locked_group)
1966{
1967 int i;
1968 int block, pnum;
1969 int blocks_per_page;
1970 ext4_group_t first_group;
1971 struct ext4_group_info *grp;
1972
1973 blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1974 /*
1975 * the buddy cache inode stores the block bitmap
1976 * and buddy information in consecutive blocks.
1977 * So for each group we need two blocks.
1978 */
1979 block = group * 2;
1980 pnum = block / blocks_per_page;
1981 first_group = pnum * blocks_per_page / 2;
1982 /* release locks on all the groups */
1983 for (i = 0; i < locked_group; i++) {
1984
1985 grp = ext4_get_group_info(sb, first_group + i);
1986 /* take all groups write allocation
1987 * semaphore. This make sure there is
1988 * no block allocation going on in any
1989 * of that groups
1990 */
1991 up_write(&grp->alloc_sem);
1992 }
1993
1994}
1995
1996static noinline_for_stack int 2005static noinline_for_stack int
1997ext4_mb_regular_allocator(struct ext4_allocation_context *ac) 2006ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
1998{ 2007{
@@ -2233,15 +2242,24 @@ static const struct file_operations ext4_mb_seq_groups_fops = {
2233 .release = seq_release, 2242 .release = seq_release,
2234}; 2243};
2235 2244
2245static struct kmem_cache *get_groupinfo_cache(int blocksize_bits)
2246{
2247 int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
2248 struct kmem_cache *cachep = ext4_groupinfo_caches[cache_index];
2249
2250 BUG_ON(!cachep);
2251 return cachep;
2252}
2236 2253
2237/* Create and initialize ext4_group_info data for the given group. */ 2254/* Create and initialize ext4_group_info data for the given group. */
2238int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group, 2255int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2239 struct ext4_group_desc *desc) 2256 struct ext4_group_desc *desc)
2240{ 2257{
2241 int i, len; 2258 int i;
2242 int metalen = 0; 2259 int metalen = 0;
2243 struct ext4_sb_info *sbi = EXT4_SB(sb); 2260 struct ext4_sb_info *sbi = EXT4_SB(sb);
2244 struct ext4_group_info **meta_group_info; 2261 struct ext4_group_info **meta_group_info;
2262 struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2245 2263
2246 /* 2264 /*
2247 * First check if this group is the first of a reserved block. 2265 * First check if this group is the first of a reserved block.
@@ -2261,22 +2279,16 @@ int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2261 meta_group_info; 2279 meta_group_info;
2262 } 2280 }
2263 2281
2264 /*
2265 * calculate needed size. if change bb_counters size,
2266 * don't forget about ext4_mb_generate_buddy()
2267 */
2268 len = offsetof(typeof(**meta_group_info),
2269 bb_counters[sb->s_blocksize_bits + 2]);
2270
2271 meta_group_info = 2282 meta_group_info =
2272 sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)]; 2283 sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)];
2273 i = group & (EXT4_DESC_PER_BLOCK(sb) - 1); 2284 i = group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2274 2285
2275 meta_group_info[i] = kzalloc(len, GFP_KERNEL); 2286 meta_group_info[i] = kmem_cache_alloc(cachep, GFP_KERNEL);
2276 if (meta_group_info[i] == NULL) { 2287 if (meta_group_info[i] == NULL) {
2277 printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n"); 2288 printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n");
2278 goto exit_group_info; 2289 goto exit_group_info;
2279 } 2290 }
2291 memset(meta_group_info[i], 0, kmem_cache_size(cachep));
2280 set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, 2292 set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT,
2281 &(meta_group_info[i]->bb_state)); 2293 &(meta_group_info[i]->bb_state));
2282 2294
@@ -2331,6 +2343,7 @@ static int ext4_mb_init_backend(struct super_block *sb)
2331 int num_meta_group_infos_max; 2343 int num_meta_group_infos_max;
2332 int array_size; 2344 int array_size;
2333 struct ext4_group_desc *desc; 2345 struct ext4_group_desc *desc;
2346 struct kmem_cache *cachep;
2334 2347
2335 /* This is the number of blocks used by GDT */ 2348 /* This is the number of blocks used by GDT */
2336 num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) - 2349 num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) -
@@ -2373,6 +2386,7 @@ static int ext4_mb_init_backend(struct super_block *sb)
2373 printk(KERN_ERR "EXT4-fs: can't get new inode\n"); 2386 printk(KERN_ERR "EXT4-fs: can't get new inode\n");
2374 goto err_freesgi; 2387 goto err_freesgi;
2375 } 2388 }
2389 sbi->s_buddy_cache->i_ino = get_next_ino();
2376 EXT4_I(sbi->s_buddy_cache)->i_disksize = 0; 2390 EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
2377 for (i = 0; i < ngroups; i++) { 2391 for (i = 0; i < ngroups; i++) {
2378 desc = ext4_get_group_desc(sb, i, NULL); 2392 desc = ext4_get_group_desc(sb, i, NULL);
@@ -2388,8 +2402,9 @@ static int ext4_mb_init_backend(struct super_block *sb)
2388 return 0; 2402 return 0;
2389 2403
2390err_freebuddy: 2404err_freebuddy:
2405 cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2391 while (i-- > 0) 2406 while (i-- > 0)
2392 kfree(ext4_get_group_info(sb, i)); 2407 kmem_cache_free(cachep, ext4_get_group_info(sb, i));
2393 i = num_meta_group_infos; 2408 i = num_meta_group_infos;
2394 while (i-- > 0) 2409 while (i-- > 0)
2395 kfree(sbi->s_group_info[i]); 2410 kfree(sbi->s_group_info[i]);
@@ -2406,19 +2421,48 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2406 unsigned offset; 2421 unsigned offset;
2407 unsigned max; 2422 unsigned max;
2408 int ret; 2423 int ret;
2424 int cache_index;
2425 struct kmem_cache *cachep;
2426 char *namep = NULL;
2409 2427
2410 i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_offsets); 2428 i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_offsets);
2411 2429
2412 sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL); 2430 sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
2413 if (sbi->s_mb_offsets == NULL) { 2431 if (sbi->s_mb_offsets == NULL) {
2414 return -ENOMEM; 2432 ret = -ENOMEM;
2433 goto out;
2415 } 2434 }
2416 2435
2417 i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_maxs); 2436 i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_maxs);
2418 sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL); 2437 sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
2419 if (sbi->s_mb_maxs == NULL) { 2438 if (sbi->s_mb_maxs == NULL) {
2420 kfree(sbi->s_mb_offsets); 2439 ret = -ENOMEM;
2421 return -ENOMEM; 2440 goto out;
2441 }
2442
2443 cache_index = sb->s_blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
2444 cachep = ext4_groupinfo_caches[cache_index];
2445 if (!cachep) {
2446 char name[32];
2447 int len = offsetof(struct ext4_group_info,
2448 bb_counters[sb->s_blocksize_bits + 2]);
2449
2450 sprintf(name, "ext4_groupinfo_%d", sb->s_blocksize_bits);
2451 namep = kstrdup(name, GFP_KERNEL);
2452 if (!namep) {
2453 ret = -ENOMEM;
2454 goto out;
2455 }
2456
2457 /* Need to free the kmem_cache_name() when we
2458 * destroy the slab */
2459 cachep = kmem_cache_create(namep, len, 0,
2460 SLAB_RECLAIM_ACCOUNT, NULL);
2461 if (!cachep) {
2462 ret = -ENOMEM;
2463 goto out;
2464 }
2465 ext4_groupinfo_caches[cache_index] = cachep;
2422 } 2466 }
2423 2467
2424 /* order 0 is regular bitmap */ 2468 /* order 0 is regular bitmap */
@@ -2439,9 +2483,7 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2439 /* init file for buddy data */ 2483 /* init file for buddy data */
2440 ret = ext4_mb_init_backend(sb); 2484 ret = ext4_mb_init_backend(sb);
2441 if (ret != 0) { 2485 if (ret != 0) {
2442 kfree(sbi->s_mb_offsets); 2486 goto out;
2443 kfree(sbi->s_mb_maxs);
2444 return ret;
2445 } 2487 }
2446 2488
2447 spin_lock_init(&sbi->s_md_lock); 2489 spin_lock_init(&sbi->s_md_lock);
@@ -2456,9 +2498,8 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2456 2498
2457 sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group); 2499 sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
2458 if (sbi->s_locality_groups == NULL) { 2500 if (sbi->s_locality_groups == NULL) {
2459 kfree(sbi->s_mb_offsets); 2501 ret = -ENOMEM;
2460 kfree(sbi->s_mb_maxs); 2502 goto out;
2461 return -ENOMEM;
2462 } 2503 }
2463 for_each_possible_cpu(i) { 2504 for_each_possible_cpu(i) {
2464 struct ext4_locality_group *lg; 2505 struct ext4_locality_group *lg;
@@ -2475,7 +2516,13 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2475 2516
2476 if (sbi->s_journal) 2517 if (sbi->s_journal)
2477 sbi->s_journal->j_commit_callback = release_blocks_on_commit; 2518 sbi->s_journal->j_commit_callback = release_blocks_on_commit;
2478 return 0; 2519out:
2520 if (ret) {
2521 kfree(sbi->s_mb_offsets);
2522 kfree(sbi->s_mb_maxs);
2523 kfree(namep);
2524 }
2525 return ret;
2479} 2526}
2480 2527
2481/* need to called with the ext4 group lock held */ 2528/* need to called with the ext4 group lock held */
@@ -2503,6 +2550,7 @@ int ext4_mb_release(struct super_block *sb)
2503 int num_meta_group_infos; 2550 int num_meta_group_infos;
2504 struct ext4_group_info *grinfo; 2551 struct ext4_group_info *grinfo;
2505 struct ext4_sb_info *sbi = EXT4_SB(sb); 2552 struct ext4_sb_info *sbi = EXT4_SB(sb);
2553 struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2506 2554
2507 if (sbi->s_group_info) { 2555 if (sbi->s_group_info) {
2508 for (i = 0; i < ngroups; i++) { 2556 for (i = 0; i < ngroups; i++) {
@@ -2513,7 +2561,7 @@ int ext4_mb_release(struct super_block *sb)
2513 ext4_lock_group(sb, i); 2561 ext4_lock_group(sb, i);
2514 ext4_mb_cleanup_pa(grinfo); 2562 ext4_mb_cleanup_pa(grinfo);
2515 ext4_unlock_group(sb, i); 2563 ext4_unlock_group(sb, i);
2516 kfree(grinfo); 2564 kmem_cache_free(cachep, grinfo);
2517 } 2565 }
2518 num_meta_group_infos = (ngroups + 2566 num_meta_group_infos = (ngroups +
2519 EXT4_DESC_PER_BLOCK(sb) - 1) >> 2567 EXT4_DESC_PER_BLOCK(sb) - 1) >>
@@ -2557,7 +2605,7 @@ int ext4_mb_release(struct super_block *sb)
2557 return 0; 2605 return 0;
2558} 2606}
2559 2607
2560static inline void ext4_issue_discard(struct super_block *sb, 2608static inline int ext4_issue_discard(struct super_block *sb,
2561 ext4_group_t block_group, ext4_grpblk_t block, int count) 2609 ext4_group_t block_group, ext4_grpblk_t block, int count)
2562{ 2610{
2563 int ret; 2611 int ret;
@@ -2567,10 +2615,11 @@ static inline void ext4_issue_discard(struct super_block *sb,
2567 trace_ext4_discard_blocks(sb, 2615 trace_ext4_discard_blocks(sb,
2568 (unsigned long long) discard_block, count); 2616 (unsigned long long) discard_block, count);
2569 ret = sb_issue_discard(sb, discard_block, count, GFP_NOFS, 0); 2617 ret = sb_issue_discard(sb, discard_block, count, GFP_NOFS, 0);
2570 if (ret == EOPNOTSUPP) { 2618 if (ret == -EOPNOTSUPP) {
2571 ext4_warning(sb, "discard not supported, disabling"); 2619 ext4_warning(sb, "discard not supported, disabling");
2572 clear_opt(EXT4_SB(sb)->s_mount_opt, DISCARD); 2620 clear_opt(EXT4_SB(sb)->s_mount_opt, DISCARD);
2573 } 2621 }
2622 return ret;
2574} 2623}
2575 2624
2576/* 2625/*
@@ -2658,28 +2707,22 @@ static void ext4_remove_debugfs_entry(void)
2658 2707
2659#endif 2708#endif
2660 2709
2661int __init init_ext4_mballoc(void) 2710int __init ext4_init_mballoc(void)
2662{ 2711{
2663 ext4_pspace_cachep = 2712 ext4_pspace_cachep = KMEM_CACHE(ext4_prealloc_space,
2664 kmem_cache_create("ext4_prealloc_space", 2713 SLAB_RECLAIM_ACCOUNT);
2665 sizeof(struct ext4_prealloc_space),
2666 0, SLAB_RECLAIM_ACCOUNT, NULL);
2667 if (ext4_pspace_cachep == NULL) 2714 if (ext4_pspace_cachep == NULL)
2668 return -ENOMEM; 2715 return -ENOMEM;
2669 2716
2670 ext4_ac_cachep = 2717 ext4_ac_cachep = KMEM_CACHE(ext4_allocation_context,
2671 kmem_cache_create("ext4_alloc_context", 2718 SLAB_RECLAIM_ACCOUNT);
2672 sizeof(struct ext4_allocation_context),
2673 0, SLAB_RECLAIM_ACCOUNT, NULL);
2674 if (ext4_ac_cachep == NULL) { 2719 if (ext4_ac_cachep == NULL) {
2675 kmem_cache_destroy(ext4_pspace_cachep); 2720 kmem_cache_destroy(ext4_pspace_cachep);
2676 return -ENOMEM; 2721 return -ENOMEM;
2677 } 2722 }
2678 2723
2679 ext4_free_ext_cachep = 2724 ext4_free_ext_cachep = KMEM_CACHE(ext4_free_data,
2680 kmem_cache_create("ext4_free_block_extents", 2725 SLAB_RECLAIM_ACCOUNT);
2681 sizeof(struct ext4_free_data),
2682 0, SLAB_RECLAIM_ACCOUNT, NULL);
2683 if (ext4_free_ext_cachep == NULL) { 2726 if (ext4_free_ext_cachep == NULL) {
2684 kmem_cache_destroy(ext4_pspace_cachep); 2727 kmem_cache_destroy(ext4_pspace_cachep);
2685 kmem_cache_destroy(ext4_ac_cachep); 2728 kmem_cache_destroy(ext4_ac_cachep);
@@ -2689,8 +2732,9 @@ int __init init_ext4_mballoc(void)
2689 return 0; 2732 return 0;
2690} 2733}
2691 2734
2692void exit_ext4_mballoc(void) 2735void ext4_exit_mballoc(void)
2693{ 2736{
2737 int i;
2694 /* 2738 /*
2695 * Wait for completion of call_rcu()'s on ext4_pspace_cachep 2739 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
2696 * before destroying the slab cache. 2740 * before destroying the slab cache.
@@ -2699,6 +2743,15 @@ void exit_ext4_mballoc(void)
2699 kmem_cache_destroy(ext4_pspace_cachep); 2743 kmem_cache_destroy(ext4_pspace_cachep);
2700 kmem_cache_destroy(ext4_ac_cachep); 2744 kmem_cache_destroy(ext4_ac_cachep);
2701 kmem_cache_destroy(ext4_free_ext_cachep); 2745 kmem_cache_destroy(ext4_free_ext_cachep);
2746
2747 for (i = 0; i < NR_GRPINFO_CACHES; i++) {
2748 struct kmem_cache *cachep = ext4_groupinfo_caches[i];
2749 if (cachep) {
2750 char *name = (char *)kmem_cache_name(cachep);
2751 kmem_cache_destroy(cachep);
2752 kfree(name);
2753 }
2754 }
2702 ext4_remove_debugfs_entry(); 2755 ext4_remove_debugfs_entry();
2703} 2756}
2704 2757
@@ -3535,8 +3588,7 @@ static int ext4_mb_new_preallocation(struct ext4_allocation_context *ac)
3535 */ 3588 */
3536static noinline_for_stack int 3589static noinline_for_stack int
3537ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh, 3590ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3538 struct ext4_prealloc_space *pa, 3591 struct ext4_prealloc_space *pa)
3539 struct ext4_allocation_context *ac)
3540{ 3592{
3541 struct super_block *sb = e4b->bd_sb; 3593 struct super_block *sb = e4b->bd_sb;
3542 struct ext4_sb_info *sbi = EXT4_SB(sb); 3594 struct ext4_sb_info *sbi = EXT4_SB(sb);
@@ -3554,11 +3606,6 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3554 BUG_ON(group != e4b->bd_group && pa->pa_len != 0); 3606 BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
3555 end = bit + pa->pa_len; 3607 end = bit + pa->pa_len;
3556 3608
3557 if (ac) {
3558 ac->ac_sb = sb;
3559 ac->ac_inode = pa->pa_inode;
3560 }
3561
3562 while (bit < end) { 3609 while (bit < end) {
3563 bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit); 3610 bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
3564 if (bit >= end) 3611 if (bit >= end)
@@ -3569,16 +3616,9 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3569 (unsigned) next - bit, (unsigned) group); 3616 (unsigned) next - bit, (unsigned) group);
3570 free += next - bit; 3617 free += next - bit;
3571 3618
3572 if (ac) { 3619 trace_ext4_mballoc_discard(sb, NULL, group, bit, next - bit);
3573 ac->ac_b_ex.fe_group = group; 3620 trace_ext4_mb_release_inode_pa(sb, pa->pa_inode, pa,
3574 ac->ac_b_ex.fe_start = bit; 3621 grp_blk_start + bit, next - bit);
3575 ac->ac_b_ex.fe_len = next - bit;
3576 ac->ac_b_ex.fe_logical = 0;
3577 trace_ext4_mballoc_discard(ac);
3578 }
3579
3580 trace_ext4_mb_release_inode_pa(sb, ac, pa, grp_blk_start + bit,
3581 next - bit);
3582 mb_free_blocks(pa->pa_inode, e4b, bit, next - bit); 3622 mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
3583 bit = next + 1; 3623 bit = next + 1;
3584 } 3624 }
@@ -3601,29 +3641,19 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3601 3641
3602static noinline_for_stack int 3642static noinline_for_stack int
3603ext4_mb_release_group_pa(struct ext4_buddy *e4b, 3643ext4_mb_release_group_pa(struct ext4_buddy *e4b,
3604 struct ext4_prealloc_space *pa, 3644 struct ext4_prealloc_space *pa)
3605 struct ext4_allocation_context *ac)
3606{ 3645{
3607 struct super_block *sb = e4b->bd_sb; 3646 struct super_block *sb = e4b->bd_sb;
3608 ext4_group_t group; 3647 ext4_group_t group;
3609 ext4_grpblk_t bit; 3648 ext4_grpblk_t bit;
3610 3649
3611 trace_ext4_mb_release_group_pa(sb, ac, pa); 3650 trace_ext4_mb_release_group_pa(sb, pa);
3612 BUG_ON(pa->pa_deleted == 0); 3651 BUG_ON(pa->pa_deleted == 0);
3613 ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit); 3652 ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
3614 BUG_ON(group != e4b->bd_group && pa->pa_len != 0); 3653 BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
3615 mb_free_blocks(pa->pa_inode, e4b, bit, pa->pa_len); 3654 mb_free_blocks(pa->pa_inode, e4b, bit, pa->pa_len);
3616 atomic_add(pa->pa_len, &EXT4_SB(sb)->s_mb_discarded); 3655 atomic_add(pa->pa_len, &EXT4_SB(sb)->s_mb_discarded);
3617 3656 trace_ext4_mballoc_discard(sb, NULL, group, bit, pa->pa_len);
3618 if (ac) {
3619 ac->ac_sb = sb;
3620 ac->ac_inode = NULL;
3621 ac->ac_b_ex.fe_group = group;
3622 ac->ac_b_ex.fe_start = bit;
3623 ac->ac_b_ex.fe_len = pa->pa_len;
3624 ac->ac_b_ex.fe_logical = 0;
3625 trace_ext4_mballoc_discard(ac);
3626 }
3627 3657
3628 return 0; 3658 return 0;
3629} 3659}
@@ -3644,7 +3674,6 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
3644 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 3674 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
3645 struct buffer_head *bitmap_bh = NULL; 3675 struct buffer_head *bitmap_bh = NULL;
3646 struct ext4_prealloc_space *pa, *tmp; 3676 struct ext4_prealloc_space *pa, *tmp;
3647 struct ext4_allocation_context *ac;
3648 struct list_head list; 3677 struct list_head list;
3649 struct ext4_buddy e4b; 3678 struct ext4_buddy e4b;
3650 int err; 3679 int err;
@@ -3673,9 +3702,6 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
3673 needed = EXT4_BLOCKS_PER_GROUP(sb) + 1; 3702 needed = EXT4_BLOCKS_PER_GROUP(sb) + 1;
3674 3703
3675 INIT_LIST_HEAD(&list); 3704 INIT_LIST_HEAD(&list);
3676 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
3677 if (ac)
3678 ac->ac_sb = sb;
3679repeat: 3705repeat:
3680 ext4_lock_group(sb, group); 3706 ext4_lock_group(sb, group);
3681 list_for_each_entry_safe(pa, tmp, 3707 list_for_each_entry_safe(pa, tmp,
@@ -3730,9 +3756,9 @@ repeat:
3730 spin_unlock(pa->pa_obj_lock); 3756 spin_unlock(pa->pa_obj_lock);
3731 3757
3732 if (pa->pa_type == MB_GROUP_PA) 3758 if (pa->pa_type == MB_GROUP_PA)
3733 ext4_mb_release_group_pa(&e4b, pa, ac); 3759 ext4_mb_release_group_pa(&e4b, pa);
3734 else 3760 else
3735 ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac); 3761 ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3736 3762
3737 list_del(&pa->u.pa_tmp_list); 3763 list_del(&pa->u.pa_tmp_list);
3738 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback); 3764 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
@@ -3740,8 +3766,6 @@ repeat:
3740 3766
3741out: 3767out:
3742 ext4_unlock_group(sb, group); 3768 ext4_unlock_group(sb, group);
3743 if (ac)
3744 kmem_cache_free(ext4_ac_cachep, ac);
3745 ext4_mb_unload_buddy(&e4b); 3769 ext4_mb_unload_buddy(&e4b);
3746 put_bh(bitmap_bh); 3770 put_bh(bitmap_bh);
3747 return free; 3771 return free;
@@ -3762,7 +3786,6 @@ void ext4_discard_preallocations(struct inode *inode)
3762 struct super_block *sb = inode->i_sb; 3786 struct super_block *sb = inode->i_sb;
3763 struct buffer_head *bitmap_bh = NULL; 3787 struct buffer_head *bitmap_bh = NULL;
3764 struct ext4_prealloc_space *pa, *tmp; 3788 struct ext4_prealloc_space *pa, *tmp;
3765 struct ext4_allocation_context *ac;
3766 ext4_group_t group = 0; 3789 ext4_group_t group = 0;
3767 struct list_head list; 3790 struct list_head list;
3768 struct ext4_buddy e4b; 3791 struct ext4_buddy e4b;
@@ -3778,11 +3801,6 @@ void ext4_discard_preallocations(struct inode *inode)
3778 3801
3779 INIT_LIST_HEAD(&list); 3802 INIT_LIST_HEAD(&list);
3780 3803
3781 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
3782 if (ac) {
3783 ac->ac_sb = sb;
3784 ac->ac_inode = inode;
3785 }
3786repeat: 3804repeat:
3787 /* first, collect all pa's in the inode */ 3805 /* first, collect all pa's in the inode */
3788 spin_lock(&ei->i_prealloc_lock); 3806 spin_lock(&ei->i_prealloc_lock);
@@ -3852,7 +3870,7 @@ repeat:
3852 3870
3853 ext4_lock_group(sb, group); 3871 ext4_lock_group(sb, group);
3854 list_del(&pa->pa_group_list); 3872 list_del(&pa->pa_group_list);
3855 ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac); 3873 ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3856 ext4_unlock_group(sb, group); 3874 ext4_unlock_group(sb, group);
3857 3875
3858 ext4_mb_unload_buddy(&e4b); 3876 ext4_mb_unload_buddy(&e4b);
@@ -3861,8 +3879,6 @@ repeat:
3861 list_del(&pa->u.pa_tmp_list); 3879 list_del(&pa->u.pa_tmp_list);
3862 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback); 3880 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
3863 } 3881 }
3864 if (ac)
3865 kmem_cache_free(ext4_ac_cachep, ac);
3866} 3882}
3867 3883
3868/* 3884/*
@@ -4060,14 +4076,10 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
4060 struct ext4_buddy e4b; 4076 struct ext4_buddy e4b;
4061 struct list_head discard_list; 4077 struct list_head discard_list;
4062 struct ext4_prealloc_space *pa, *tmp; 4078 struct ext4_prealloc_space *pa, *tmp;
4063 struct ext4_allocation_context *ac;
4064 4079
4065 mb_debug(1, "discard locality group preallocation\n"); 4080 mb_debug(1, "discard locality group preallocation\n");
4066 4081
4067 INIT_LIST_HEAD(&discard_list); 4082 INIT_LIST_HEAD(&discard_list);
4068 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4069 if (ac)
4070 ac->ac_sb = sb;
4071 4083
4072 spin_lock(&lg->lg_prealloc_lock); 4084 spin_lock(&lg->lg_prealloc_lock);
4073 list_for_each_entry_rcu(pa, &lg->lg_prealloc_list[order], 4085 list_for_each_entry_rcu(pa, &lg->lg_prealloc_list[order],
@@ -4119,15 +4131,13 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
4119 } 4131 }
4120 ext4_lock_group(sb, group); 4132 ext4_lock_group(sb, group);
4121 list_del(&pa->pa_group_list); 4133 list_del(&pa->pa_group_list);
4122 ext4_mb_release_group_pa(&e4b, pa, ac); 4134 ext4_mb_release_group_pa(&e4b, pa);
4123 ext4_unlock_group(sb, group); 4135 ext4_unlock_group(sb, group);
4124 4136
4125 ext4_mb_unload_buddy(&e4b); 4137 ext4_mb_unload_buddy(&e4b);
4126 list_del(&pa->u.pa_tmp_list); 4138 list_del(&pa->u.pa_tmp_list);
4127 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback); 4139 call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
4128 } 4140 }
4129 if (ac)
4130 kmem_cache_free(ext4_ac_cachep, ac);
4131} 4141}
4132 4142
4133/* 4143/*
@@ -4491,7 +4501,6 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
4491{ 4501{
4492 struct buffer_head *bitmap_bh = NULL; 4502 struct buffer_head *bitmap_bh = NULL;
4493 struct super_block *sb = inode->i_sb; 4503 struct super_block *sb = inode->i_sb;
4494 struct ext4_allocation_context *ac = NULL;
4495 struct ext4_group_desc *gdp; 4504 struct ext4_group_desc *gdp;
4496 unsigned long freed = 0; 4505 unsigned long freed = 0;
4497 unsigned int overflow; 4506 unsigned int overflow;
@@ -4531,6 +4540,8 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
4531 if (!bh) 4540 if (!bh)
4532 tbh = sb_find_get_block(inode->i_sb, 4541 tbh = sb_find_get_block(inode->i_sb,
4533 block + i); 4542 block + i);
4543 if (unlikely(!tbh))
4544 continue;
4534 ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA, 4545 ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
4535 inode, tbh, block + i); 4546 inode, tbh, block + i);
4536 } 4547 }
@@ -4546,12 +4557,6 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
4546 if (!ext4_should_writeback_data(inode)) 4557 if (!ext4_should_writeback_data(inode))
4547 flags |= EXT4_FREE_BLOCKS_METADATA; 4558 flags |= EXT4_FREE_BLOCKS_METADATA;
4548 4559
4549 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4550 if (ac) {
4551 ac->ac_inode = inode;
4552 ac->ac_sb = sb;
4553 }
4554
4555do_more: 4560do_more:
4556 overflow = 0; 4561 overflow = 0;
4557 ext4_get_group_no_and_offset(sb, block, &block_group, &bit); 4562 ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
@@ -4609,12 +4614,7 @@ do_more:
4609 BUG_ON(!mb_test_bit(bit + i, bitmap_bh->b_data)); 4614 BUG_ON(!mb_test_bit(bit + i, bitmap_bh->b_data));
4610 } 4615 }
4611#endif 4616#endif
4612 if (ac) { 4617 trace_ext4_mballoc_free(sb, inode, block_group, bit, count);
4613 ac->ac_b_ex.fe_group = block_group;
4614 ac->ac_b_ex.fe_start = bit;
4615 ac->ac_b_ex.fe_len = count;
4616 trace_ext4_mballoc_free(ac);
4617 }
4618 4618
4619 err = ext4_mb_load_buddy(sb, block_group, &e4b); 4619 err = ext4_mb_load_buddy(sb, block_group, &e4b);
4620 if (err) 4620 if (err)
@@ -4644,8 +4644,6 @@ do_more:
4644 mb_clear_bits(bitmap_bh->b_data, bit, count); 4644 mb_clear_bits(bitmap_bh->b_data, bit, count);
4645 mb_free_blocks(inode, &e4b, bit, count); 4645 mb_free_blocks(inode, &e4b, bit, count);
4646 ext4_mb_return_to_preallocation(inode, &e4b, block, count); 4646 ext4_mb_return_to_preallocation(inode, &e4b, block, count);
4647 if (test_opt(sb, DISCARD))
4648 ext4_issue_discard(sb, block_group, bit, count);
4649 } 4647 }
4650 4648
4651 ret = ext4_free_blks_count(sb, gdp) + count; 4649 ret = ext4_free_blks_count(sb, gdp) + count;
@@ -4685,7 +4683,190 @@ error_return:
4685 dquot_free_block(inode, freed); 4683 dquot_free_block(inode, freed);
4686 brelse(bitmap_bh); 4684 brelse(bitmap_bh);
4687 ext4_std_error(sb, err); 4685 ext4_std_error(sb, err);
4688 if (ac)
4689 kmem_cache_free(ext4_ac_cachep, ac);
4690 return; 4686 return;
4691} 4687}
4688
4689/**
4690 * ext4_trim_extent -- function to TRIM one single free extent in the group
4691 * @sb: super block for the file system
4692 * @start: starting block of the free extent in the alloc. group
4693 * @count: number of blocks to TRIM
4694 * @group: alloc. group we are working with
4695 * @e4b: ext4 buddy for the group
4696 *
4697 * Trim "count" blocks starting at "start" in the "group". To assure that no
4698 * one will allocate those blocks, mark it as used in buddy bitmap. This must
4699 * be called with under the group lock.
4700 */
4701static int ext4_trim_extent(struct super_block *sb, int start, int count,
4702 ext4_group_t group, struct ext4_buddy *e4b)
4703{
4704 struct ext4_free_extent ex;
4705 int ret = 0;
4706
4707 assert_spin_locked(ext4_group_lock_ptr(sb, group));
4708
4709 ex.fe_start = start;
4710 ex.fe_group = group;
4711 ex.fe_len = count;
4712
4713 /*
4714 * Mark blocks used, so no one can reuse them while
4715 * being trimmed.
4716 */
4717 mb_mark_used(e4b, &ex);
4718 ext4_unlock_group(sb, group);
4719
4720 ret = ext4_issue_discard(sb, group, start, count);
4721 if (ret)
4722 ext4_std_error(sb, ret);
4723
4724 ext4_lock_group(sb, group);
4725 mb_free_blocks(NULL, e4b, start, ex.fe_len);
4726 return ret;
4727}
4728
4729/**
4730 * ext4_trim_all_free -- function to trim all free space in alloc. group
4731 * @sb: super block for file system
4732 * @e4b: ext4 buddy
4733 * @start: first group block to examine
4734 * @max: last group block to examine
4735 * @minblocks: minimum extent block count
4736 *
4737 * ext4_trim_all_free walks through group's buddy bitmap searching for free
4738 * extents. When the free block is found, ext4_trim_extent is called to TRIM
4739 * the extent.
4740 *
4741 *
4742 * ext4_trim_all_free walks through group's block bitmap searching for free
4743 * extents. When the free extent is found, mark it as used in group buddy
4744 * bitmap. Then issue a TRIM command on this extent and free the extent in
4745 * the group buddy bitmap. This is done until whole group is scanned.
4746 */
4747ext4_grpblk_t ext4_trim_all_free(struct super_block *sb, struct ext4_buddy *e4b,
4748 ext4_grpblk_t start, ext4_grpblk_t max, ext4_grpblk_t minblocks)
4749{
4750 void *bitmap;
4751 ext4_grpblk_t next, count = 0;
4752 ext4_group_t group;
4753 int ret = 0;
4754
4755 BUG_ON(e4b == NULL);
4756
4757 bitmap = e4b->bd_bitmap;
4758 group = e4b->bd_group;
4759 start = (e4b->bd_info->bb_first_free > start) ?
4760 e4b->bd_info->bb_first_free : start;
4761 ext4_lock_group(sb, group);
4762
4763 while (start < max) {
4764 start = mb_find_next_zero_bit(bitmap, max, start);
4765 if (start >= max)
4766 break;
4767 next = mb_find_next_bit(bitmap, max, start);
4768
4769 if ((next - start) >= minblocks) {
4770 ret = ext4_trim_extent(sb, start,
4771 next - start, group, e4b);
4772 if (ret < 0)
4773 break;
4774 count += next - start;
4775 }
4776 start = next + 1;
4777
4778 if (fatal_signal_pending(current)) {
4779 count = -ERESTARTSYS;
4780 break;
4781 }
4782
4783 if (need_resched()) {
4784 ext4_unlock_group(sb, group);
4785 cond_resched();
4786 ext4_lock_group(sb, group);
4787 }
4788
4789 if ((e4b->bd_info->bb_free - count) < minblocks)
4790 break;
4791 }
4792 ext4_unlock_group(sb, group);
4793
4794 ext4_debug("trimmed %d blocks in the group %d\n",
4795 count, group);
4796
4797 if (ret < 0)
4798 count = ret;
4799
4800 return count;
4801}
4802
4803/**
4804 * ext4_trim_fs() -- trim ioctl handle function
4805 * @sb: superblock for filesystem
4806 * @range: fstrim_range structure
4807 *
4808 * start: First Byte to trim
4809 * len: number of Bytes to trim from start
4810 * minlen: minimum extent length in Bytes
4811 * ext4_trim_fs goes through all allocation groups containing Bytes from
4812 * start to start+len. For each such a group ext4_trim_all_free function
4813 * is invoked to trim all free space.
4814 */
4815int ext4_trim_fs(struct super_block *sb, struct fstrim_range *range)
4816{
4817 struct ext4_buddy e4b;
4818 ext4_group_t first_group, last_group;
4819 ext4_group_t group, ngroups = ext4_get_groups_count(sb);
4820 ext4_grpblk_t cnt = 0, first_block, last_block;
4821 uint64_t start, len, minlen, trimmed;
4822 int ret = 0;
4823
4824 start = range->start >> sb->s_blocksize_bits;
4825 len = range->len >> sb->s_blocksize_bits;
4826 minlen = range->minlen >> sb->s_blocksize_bits;
4827 trimmed = 0;
4828
4829 if (unlikely(minlen > EXT4_BLOCKS_PER_GROUP(sb)))
4830 return -EINVAL;
4831
4832 /* Determine first and last group to examine based on start and len */
4833 ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) start,
4834 &first_group, &first_block);
4835 ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) (start + len),
4836 &last_group, &last_block);
4837 last_group = (last_group > ngroups - 1) ? ngroups - 1 : last_group;
4838 last_block = EXT4_BLOCKS_PER_GROUP(sb);
4839
4840 if (first_group > last_group)
4841 return -EINVAL;
4842
4843 for (group = first_group; group <= last_group; group++) {
4844 ret = ext4_mb_load_buddy(sb, group, &e4b);
4845 if (ret) {
4846 ext4_error(sb, "Error in loading buddy "
4847 "information for %u", group);
4848 break;
4849 }
4850
4851 if (len >= EXT4_BLOCKS_PER_GROUP(sb))
4852 len -= (EXT4_BLOCKS_PER_GROUP(sb) - first_block);
4853 else
4854 last_block = len;
4855
4856 if (e4b.bd_info->bb_free >= minlen) {
4857 cnt = ext4_trim_all_free(sb, &e4b, first_block,
4858 last_block, minlen);
4859 if (cnt < 0) {
4860 ret = cnt;
4861 ext4_mb_unload_buddy(&e4b);
4862 break;
4863 }
4864 }
4865 ext4_mb_unload_buddy(&e4b);
4866 trimmed += cnt;
4867 first_block = 0;
4868 }
4869 range->len = trimmed * sb->s_blocksize;
4870
4871 return ret;
4872}
diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
index 1765c2c50a9b..25f3a974b725 100644
--- a/fs/ext4/migrate.c
+++ b/fs/ext4/migrate.c
@@ -412,7 +412,7 @@ static int free_ext_idx(handle_t *handle, struct inode *inode,
412 struct buffer_head *bh; 412 struct buffer_head *bh;
413 struct ext4_extent_header *eh; 413 struct ext4_extent_header *eh;
414 414
415 block = idx_pblock(ix); 415 block = ext4_idx_pblock(ix);
416 bh = sb_bread(inode->i_sb, block); 416 bh = sb_bread(inode->i_sb, block);
417 if (!bh) 417 if (!bh)
418 return -EIO; 418 return -EIO;
diff --git a/fs/ext4/move_extent.c b/fs/ext4/move_extent.c
index 5f1ed9fc913c..b9f3e7862f13 100644
--- a/fs/ext4/move_extent.c
+++ b/fs/ext4/move_extent.c
@@ -85,7 +85,7 @@ mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
85 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) { 85 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
86 /* leaf block */ 86 /* leaf block */
87 *extent = ++path[ppos].p_ext; 87 *extent = ++path[ppos].p_ext;
88 path[ppos].p_block = ext_pblock(path[ppos].p_ext); 88 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
89 return 0; 89 return 0;
90 } 90 }
91 91
@@ -96,7 +96,7 @@ mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
96 96
97 /* index block */ 97 /* index block */
98 path[ppos].p_idx++; 98 path[ppos].p_idx++;
99 path[ppos].p_block = idx_pblock(path[ppos].p_idx); 99 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
100 if (path[ppos+1].p_bh) 100 if (path[ppos+1].p_bh)
101 brelse(path[ppos+1].p_bh); 101 brelse(path[ppos+1].p_bh);
102 path[ppos+1].p_bh = 102 path[ppos+1].p_bh =
@@ -111,7 +111,7 @@ mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
111 path[cur_ppos].p_idx = 111 path[cur_ppos].p_idx =
112 EXT_FIRST_INDEX(path[cur_ppos].p_hdr); 112 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
113 path[cur_ppos].p_block = 113 path[cur_ppos].p_block =
114 idx_pblock(path[cur_ppos].p_idx); 114 ext4_idx_pblock(path[cur_ppos].p_idx);
115 if (path[cur_ppos+1].p_bh) 115 if (path[cur_ppos+1].p_bh)
116 brelse(path[cur_ppos+1].p_bh); 116 brelse(path[cur_ppos+1].p_bh);
117 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb, 117 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
@@ -133,7 +133,7 @@ mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
133 path[leaf_ppos].p_ext = *extent = 133 path[leaf_ppos].p_ext = *extent =
134 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr); 134 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
135 path[leaf_ppos].p_block = 135 path[leaf_ppos].p_block =
136 ext_pblock(path[leaf_ppos].p_ext); 136 ext4_ext_pblock(path[leaf_ppos].p_ext);
137 return 0; 137 return 0;
138 } 138 }
139 } 139 }
@@ -249,7 +249,7 @@ mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
249 */ 249 */
250 o_end->ee_block = end_ext->ee_block; 250 o_end->ee_block = end_ext->ee_block;
251 o_end->ee_len = end_ext->ee_len; 251 o_end->ee_len = end_ext->ee_len;
252 ext4_ext_store_pblock(o_end, ext_pblock(end_ext)); 252 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
253 } 253 }
254 254
255 o_start->ee_len = start_ext->ee_len; 255 o_start->ee_len = start_ext->ee_len;
@@ -276,7 +276,7 @@ mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
276 */ 276 */
277 o_end->ee_block = end_ext->ee_block; 277 o_end->ee_block = end_ext->ee_block;
278 o_end->ee_len = end_ext->ee_len; 278 o_end->ee_len = end_ext->ee_len;
279 ext4_ext_store_pblock(o_end, ext_pblock(end_ext)); 279 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
280 280
281 /* 281 /*
282 * Set 0 to the extent block if new_ext was 282 * Set 0 to the extent block if new_ext was
@@ -361,7 +361,7 @@ mext_insert_inside_block(struct ext4_extent *o_start,
361 /* Insert new entry */ 361 /* Insert new entry */
362 if (new_ext->ee_len) { 362 if (new_ext->ee_len) {
363 o_start[i] = *new_ext; 363 o_start[i] = *new_ext;
364 ext4_ext_store_pblock(&o_start[i++], ext_pblock(new_ext)); 364 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
365 } 365 }
366 366
367 /* Insert end entry */ 367 /* Insert end entry */
@@ -488,7 +488,7 @@ mext_leaf_block(handle_t *handle, struct inode *orig_inode,
488 start_ext.ee_len = end_ext.ee_len = 0; 488 start_ext.ee_len = end_ext.ee_len = 0;
489 489
490 new_ext.ee_block = cpu_to_le32(*from); 490 new_ext.ee_block = cpu_to_le32(*from);
491 ext4_ext_store_pblock(&new_ext, ext_pblock(dext)); 491 ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
492 new_ext.ee_len = dext->ee_len; 492 new_ext.ee_len = dext->ee_len;
493 new_ext_alen = ext4_ext_get_actual_len(&new_ext); 493 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
494 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1; 494 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
@@ -553,7 +553,7 @@ mext_leaf_block(handle_t *handle, struct inode *orig_inode,
553 copy_extent_status(oext, &end_ext); 553 copy_extent_status(oext, &end_ext);
554 end_ext_alen = ext4_ext_get_actual_len(&end_ext); 554 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
555 ext4_ext_store_pblock(&end_ext, 555 ext4_ext_store_pblock(&end_ext,
556 (ext_pblock(o_end) + oext_alen - end_ext_alen)); 556 (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
557 end_ext.ee_block = 557 end_ext.ee_block =
558 cpu_to_le32(le32_to_cpu(o_end->ee_block) + 558 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
559 oext_alen - end_ext_alen); 559 oext_alen - end_ext_alen);
@@ -604,7 +604,7 @@ mext_calc_swap_extents(struct ext4_extent *tmp_dext,
604 /* When tmp_dext is too large, pick up the target range. */ 604 /* When tmp_dext is too large, pick up the target range. */
605 diff = donor_off - le32_to_cpu(tmp_dext->ee_block); 605 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
606 606
607 ext4_ext_store_pblock(tmp_dext, ext_pblock(tmp_dext) + diff); 607 ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
608 tmp_dext->ee_block = 608 tmp_dext->ee_block =
609 cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff); 609 cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff);
610 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff); 610 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff);
@@ -613,7 +613,7 @@ mext_calc_swap_extents(struct ext4_extent *tmp_dext,
613 tmp_dext->ee_len = cpu_to_le16(max_count); 613 tmp_dext->ee_len = cpu_to_le16(max_count);
614 614
615 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block); 615 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
616 ext4_ext_store_pblock(tmp_oext, ext_pblock(tmp_oext) + orig_diff); 616 ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
617 617
618 /* Adjust extent length if donor extent is larger than orig */ 618 /* Adjust extent length if donor extent is larger than orig */
619 if (ext4_ext_get_actual_len(tmp_dext) > 619 if (ext4_ext_get_actual_len(tmp_dext) >
diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index 314c0d3b3fa9..92203b8a099f 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -856,6 +856,7 @@ static struct buffer_head * ext4_find_entry (struct inode *dir,
856 struct buffer_head *bh_use[NAMEI_RA_SIZE]; 856 struct buffer_head *bh_use[NAMEI_RA_SIZE];
857 struct buffer_head *bh, *ret = NULL; 857 struct buffer_head *bh, *ret = NULL;
858 ext4_lblk_t start, block, b; 858 ext4_lblk_t start, block, b;
859 const u8 *name = d_name->name;
859 int ra_max = 0; /* Number of bh's in the readahead 860 int ra_max = 0; /* Number of bh's in the readahead
860 buffer, bh_use[] */ 861 buffer, bh_use[] */
861 int ra_ptr = 0; /* Current index into readahead 862 int ra_ptr = 0; /* Current index into readahead
@@ -870,6 +871,16 @@ static struct buffer_head * ext4_find_entry (struct inode *dir,
870 namelen = d_name->len; 871 namelen = d_name->len;
871 if (namelen > EXT4_NAME_LEN) 872 if (namelen > EXT4_NAME_LEN)
872 return NULL; 873 return NULL;
874 if ((namelen <= 2) && (name[0] == '.') &&
875 (name[1] == '.' || name[1] == '0')) {
876 /*
877 * "." or ".." will only be in the first block
878 * NFS may look up ".."; "." should be handled by the VFS
879 */
880 block = start = 0;
881 nblocks = 1;
882 goto restart;
883 }
873 if (is_dx(dir)) { 884 if (is_dx(dir)) {
874 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err); 885 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
875 /* 886 /*
@@ -960,55 +971,35 @@ cleanup_and_exit:
960static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, 971static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
961 struct ext4_dir_entry_2 **res_dir, int *err) 972 struct ext4_dir_entry_2 **res_dir, int *err)
962{ 973{
963 struct super_block * sb; 974 struct super_block * sb = dir->i_sb;
964 struct dx_hash_info hinfo; 975 struct dx_hash_info hinfo;
965 u32 hash;
966 struct dx_frame frames[2], *frame; 976 struct dx_frame frames[2], *frame;
967 struct ext4_dir_entry_2 *de, *top;
968 struct buffer_head *bh; 977 struct buffer_head *bh;
969 ext4_lblk_t block; 978 ext4_lblk_t block;
970 int retval; 979 int retval;
971 int namelen = d_name->len;
972 const u8 *name = d_name->name;
973 980
974 sb = dir->i_sb; 981 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
975 /* NFS may look up ".." - look at dx_root directory block */ 982 return NULL;
976 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
977 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
978 return NULL;
979 } else {
980 frame = frames;
981 frame->bh = NULL; /* for dx_release() */
982 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
983 dx_set_block(frame->at, 0); /* dx_root block is 0 */
984 }
985 hash = hinfo.hash;
986 do { 983 do {
987 block = dx_get_block(frame->at); 984 block = dx_get_block(frame->at);
988 if (!(bh = ext4_bread (NULL,dir, block, 0, err))) 985 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
989 goto errout; 986 goto errout;
990 de = (struct ext4_dir_entry_2 *) bh->b_data;
991 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
992 EXT4_DIR_REC_LEN(0));
993 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
994 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
995 + ((char *) de - bh->b_data);
996
997 if (!ext4_check_dir_entry(dir, de, bh, off)) {
998 brelse(bh);
999 *err = ERR_BAD_DX_DIR;
1000 goto errout;
1001 }
1002 987
1003 if (ext4_match(namelen, name, de)) { 988 retval = search_dirblock(bh, dir, d_name,
1004 *res_dir = de; 989 block << EXT4_BLOCK_SIZE_BITS(sb),
1005 dx_release(frames); 990 res_dir);
1006 return bh; 991 if (retval == 1) { /* Success! */
1007 } 992 dx_release(frames);
993 return bh;
1008 } 994 }
1009 brelse(bh); 995 brelse(bh);
996 if (retval == -1) {
997 *err = ERR_BAD_DX_DIR;
998 goto errout;
999 }
1000
1010 /* Check to see if we should continue to search */ 1001 /* Check to see if we should continue to search */
1011 retval = ext4_htree_next_block(dir, hash, frame, 1002 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1012 frames, NULL); 1003 frames, NULL);
1013 if (retval < 0) { 1004 if (retval < 0) {
1014 ext4_warning(sb, 1005 ext4_warning(sb,
@@ -2312,7 +2303,7 @@ retry:
2312 2303
2313 inode->i_ctime = ext4_current_time(inode); 2304 inode->i_ctime = ext4_current_time(inode);
2314 ext4_inc_count(handle, inode); 2305 ext4_inc_count(handle, inode);
2315 atomic_inc(&inode->i_count); 2306 ihold(inode);
2316 2307
2317 err = ext4_add_entry(handle, dentry, inode); 2308 err = ext4_add_entry(handle, dentry, inode);
2318 if (!err) { 2309 if (!err) {
diff --git a/fs/ext4/page-io.c b/fs/ext4/page-io.c
new file mode 100644
index 000000000000..beacce11ac50
--- /dev/null
+++ b/fs/ext4/page-io.c
@@ -0,0 +1,431 @@
1/*
2 * linux/fs/ext4/page-io.c
3 *
4 * This contains the new page_io functions for ext4
5 *
6 * Written by Theodore Ts'o, 2010.
7 */
8
9#include <linux/module.h>
10#include <linux/fs.h>
11#include <linux/time.h>
12#include <linux/jbd2.h>
13#include <linux/highuid.h>
14#include <linux/pagemap.h>
15#include <linux/quotaops.h>
16#include <linux/string.h>
17#include <linux/buffer_head.h>
18#include <linux/writeback.h>
19#include <linux/pagevec.h>
20#include <linux/mpage.h>
21#include <linux/namei.h>
22#include <linux/uio.h>
23#include <linux/bio.h>
24#include <linux/workqueue.h>
25#include <linux/kernel.h>
26#include <linux/slab.h>
27
28#include "ext4_jbd2.h"
29#include "xattr.h"
30#include "acl.h"
31#include "ext4_extents.h"
32
33static struct kmem_cache *io_page_cachep, *io_end_cachep;
34
35#define WQ_HASH_SZ 37
36#define to_ioend_wq(v) (&ioend_wq[((unsigned long)v) % WQ_HASH_SZ])
37static wait_queue_head_t ioend_wq[WQ_HASH_SZ];
38
39int __init ext4_init_pageio(void)
40{
41 int i;
42
43 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
44 if (io_page_cachep == NULL)
45 return -ENOMEM;
46 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
47 if (io_page_cachep == NULL) {
48 kmem_cache_destroy(io_page_cachep);
49 return -ENOMEM;
50 }
51 for (i = 0; i < WQ_HASH_SZ; i++)
52 init_waitqueue_head(&ioend_wq[i]);
53
54 return 0;
55}
56
57void ext4_exit_pageio(void)
58{
59 kmem_cache_destroy(io_end_cachep);
60 kmem_cache_destroy(io_page_cachep);
61}
62
63void ext4_ioend_wait(struct inode *inode)
64{
65 wait_queue_head_t *wq = to_ioend_wq(inode);
66
67 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
68}
69
70static void put_io_page(struct ext4_io_page *io_page)
71{
72 if (atomic_dec_and_test(&io_page->p_count)) {
73 end_page_writeback(io_page->p_page);
74 put_page(io_page->p_page);
75 kmem_cache_free(io_page_cachep, io_page);
76 }
77}
78
79void ext4_free_io_end(ext4_io_end_t *io)
80{
81 int i;
82 wait_queue_head_t *wq;
83
84 BUG_ON(!io);
85 if (io->page)
86 put_page(io->page);
87 for (i = 0; i < io->num_io_pages; i++)
88 put_io_page(io->pages[i]);
89 io->num_io_pages = 0;
90 wq = to_ioend_wq(io->inode);
91 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
92 waitqueue_active(wq))
93 wake_up_all(wq);
94 kmem_cache_free(io_end_cachep, io);
95}
96
97/*
98 * check a range of space and convert unwritten extents to written.
99 */
100int ext4_end_io_nolock(ext4_io_end_t *io)
101{
102 struct inode *inode = io->inode;
103 loff_t offset = io->offset;
104 ssize_t size = io->size;
105 int ret = 0;
106
107 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
108 "list->prev 0x%p\n",
109 io, inode->i_ino, io->list.next, io->list.prev);
110
111 if (list_empty(&io->list))
112 return ret;
113
114 if (!(io->flag & EXT4_IO_END_UNWRITTEN))
115 return ret;
116
117 ret = ext4_convert_unwritten_extents(inode, offset, size);
118 if (ret < 0) {
119 printk(KERN_EMERG "%s: failed to convert unwritten "
120 "extents to written extents, error is %d "
121 "io is still on inode %lu aio dio list\n",
122 __func__, ret, inode->i_ino);
123 return ret;
124 }
125
126 if (io->iocb)
127 aio_complete(io->iocb, io->result, 0);
128 /* clear the DIO AIO unwritten flag */
129 io->flag &= ~EXT4_IO_END_UNWRITTEN;
130 return ret;
131}
132
133/*
134 * work on completed aio dio IO, to convert unwritten extents to extents
135 */
136static void ext4_end_io_work(struct work_struct *work)
137{
138 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
139 struct inode *inode = io->inode;
140 struct ext4_inode_info *ei = EXT4_I(inode);
141 unsigned long flags;
142 int ret;
143
144 mutex_lock(&inode->i_mutex);
145 ret = ext4_end_io_nolock(io);
146 if (ret < 0) {
147 mutex_unlock(&inode->i_mutex);
148 return;
149 }
150
151 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
152 if (!list_empty(&io->list))
153 list_del_init(&io->list);
154 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
155 mutex_unlock(&inode->i_mutex);
156 ext4_free_io_end(io);
157}
158
159ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
160{
161 ext4_io_end_t *io = NULL;
162
163 io = kmem_cache_alloc(io_end_cachep, flags);
164 if (io) {
165 memset(io, 0, sizeof(*io));
166 atomic_inc(&EXT4_I(inode)->i_ioend_count);
167 io->inode = inode;
168 INIT_WORK(&io->work, ext4_end_io_work);
169 INIT_LIST_HEAD(&io->list);
170 }
171 return io;
172}
173
174/*
175 * Print an buffer I/O error compatible with the fs/buffer.c. This
176 * provides compatibility with dmesg scrapers that look for a specific
177 * buffer I/O error message. We really need a unified error reporting
178 * structure to userspace ala Digital Unix's uerf system, but it's
179 * probably not going to happen in my lifetime, due to LKML politics...
180 */
181static void buffer_io_error(struct buffer_head *bh)
182{
183 char b[BDEVNAME_SIZE];
184 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
185 bdevname(bh->b_bdev, b),
186 (unsigned long long)bh->b_blocknr);
187}
188
189static void ext4_end_bio(struct bio *bio, int error)
190{
191 ext4_io_end_t *io_end = bio->bi_private;
192 struct workqueue_struct *wq;
193 struct inode *inode;
194 unsigned long flags;
195 int i;
196
197 BUG_ON(!io_end);
198 bio->bi_private = NULL;
199 bio->bi_end_io = NULL;
200 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
201 error = 0;
202 bio_put(bio);
203
204 for (i = 0; i < io_end->num_io_pages; i++) {
205 struct page *page = io_end->pages[i]->p_page;
206 struct buffer_head *bh, *head;
207 int partial_write = 0;
208
209 head = page_buffers(page);
210 if (error)
211 SetPageError(page);
212 BUG_ON(!head);
213 if (head->b_size == PAGE_CACHE_SIZE)
214 clear_buffer_dirty(head);
215 else {
216 loff_t offset;
217 loff_t io_end_offset = io_end->offset + io_end->size;
218
219 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
220 bh = head;
221 do {
222 if ((offset >= io_end->offset) &&
223 (offset+bh->b_size <= io_end_offset)) {
224 if (error)
225 buffer_io_error(bh);
226
227 clear_buffer_dirty(bh);
228 }
229 if (buffer_delay(bh))
230 partial_write = 1;
231 else if (!buffer_mapped(bh))
232 clear_buffer_dirty(bh);
233 else if (buffer_dirty(bh))
234 partial_write = 1;
235 offset += bh->b_size;
236 bh = bh->b_this_page;
237 } while (bh != head);
238 }
239
240 /*
241 * If this is a partial write which happened to make
242 * all buffers uptodate then we can optimize away a
243 * bogus readpage() for the next read(). Here we
244 * 'discover' whether the page went uptodate as a
245 * result of this (potentially partial) write.
246 */
247 if (!partial_write)
248 SetPageUptodate(page);
249
250 put_io_page(io_end->pages[i]);
251 }
252 io_end->num_io_pages = 0;
253 inode = io_end->inode;
254
255 if (error) {
256 io_end->flag |= EXT4_IO_END_ERROR;
257 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
258 "(offset %llu size %ld starting block %llu)",
259 inode->i_ino,
260 (unsigned long long) io_end->offset,
261 (long) io_end->size,
262 (unsigned long long)
263 bio->bi_sector >> (inode->i_blkbits - 9));
264 }
265
266 /* Add the io_end to per-inode completed io list*/
267 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
268 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
269 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
270
271 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
272 /* queue the work to convert unwritten extents to written */
273 queue_work(wq, &io_end->work);
274}
275
276void ext4_io_submit(struct ext4_io_submit *io)
277{
278 struct bio *bio = io->io_bio;
279
280 if (bio) {
281 bio_get(io->io_bio);
282 submit_bio(io->io_op, io->io_bio);
283 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
284 bio_put(io->io_bio);
285 }
286 io->io_bio = 0;
287 io->io_op = 0;
288 io->io_end = 0;
289}
290
291static int io_submit_init(struct ext4_io_submit *io,
292 struct inode *inode,
293 struct writeback_control *wbc,
294 struct buffer_head *bh)
295{
296 ext4_io_end_t *io_end;
297 struct page *page = bh->b_page;
298 int nvecs = bio_get_nr_vecs(bh->b_bdev);
299 struct bio *bio;
300
301 io_end = ext4_init_io_end(inode, GFP_NOFS);
302 if (!io_end)
303 return -ENOMEM;
304 do {
305 bio = bio_alloc(GFP_NOIO, nvecs);
306 nvecs >>= 1;
307 } while (bio == NULL);
308
309 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
310 bio->bi_bdev = bh->b_bdev;
311 bio->bi_private = io->io_end = io_end;
312 bio->bi_end_io = ext4_end_bio;
313
314 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
315
316 io->io_bio = bio;
317 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?
318 WRITE_SYNC_PLUG : WRITE);
319 io->io_next_block = bh->b_blocknr;
320 return 0;
321}
322
323static int io_submit_add_bh(struct ext4_io_submit *io,
324 struct ext4_io_page *io_page,
325 struct inode *inode,
326 struct writeback_control *wbc,
327 struct buffer_head *bh)
328{
329 ext4_io_end_t *io_end;
330 int ret;
331
332 if (buffer_new(bh)) {
333 clear_buffer_new(bh);
334 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
335 }
336
337 if (!buffer_mapped(bh) || buffer_delay(bh)) {
338 if (!buffer_mapped(bh))
339 clear_buffer_dirty(bh);
340 if (io->io_bio)
341 ext4_io_submit(io);
342 return 0;
343 }
344
345 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
346submit_and_retry:
347 ext4_io_submit(io);
348 }
349 if (io->io_bio == NULL) {
350 ret = io_submit_init(io, inode, wbc, bh);
351 if (ret)
352 return ret;
353 }
354 io_end = io->io_end;
355 if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
356 (io_end->pages[io_end->num_io_pages-1] != io_page))
357 goto submit_and_retry;
358 if (buffer_uninit(bh))
359 io->io_end->flag |= EXT4_IO_END_UNWRITTEN;
360 io->io_end->size += bh->b_size;
361 io->io_next_block++;
362 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
363 if (ret != bh->b_size)
364 goto submit_and_retry;
365 if ((io_end->num_io_pages == 0) ||
366 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
367 io_end->pages[io_end->num_io_pages++] = io_page;
368 atomic_inc(&io_page->p_count);
369 }
370 return 0;
371}
372
373int ext4_bio_write_page(struct ext4_io_submit *io,
374 struct page *page,
375 int len,
376 struct writeback_control *wbc)
377{
378 struct inode *inode = page->mapping->host;
379 unsigned block_start, block_end, blocksize;
380 struct ext4_io_page *io_page;
381 struct buffer_head *bh, *head;
382 int ret = 0;
383
384 blocksize = 1 << inode->i_blkbits;
385
386 BUG_ON(PageWriteback(page));
387 set_page_writeback(page);
388 ClearPageError(page);
389
390 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
391 if (!io_page) {
392 set_page_dirty(page);
393 unlock_page(page);
394 return -ENOMEM;
395 }
396 io_page->p_page = page;
397 atomic_set(&io_page->p_count, 1);
398 get_page(page);
399
400 for (bh = head = page_buffers(page), block_start = 0;
401 bh != head || !block_start;
402 block_start = block_end, bh = bh->b_this_page) {
403 block_end = block_start + blocksize;
404 if (block_start >= len) {
405 clear_buffer_dirty(bh);
406 set_buffer_uptodate(bh);
407 continue;
408 }
409 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
410 if (ret) {
411 /*
412 * We only get here on ENOMEM. Not much else
413 * we can do but mark the page as dirty, and
414 * better luck next time.
415 */
416 set_page_dirty(page);
417 break;
418 }
419 }
420 unlock_page(page);
421 /*
422 * If the page was truncated before we could do the writeback,
423 * or we had a memory allocation error while trying to write
424 * the first buffer head, we won't have submitted any pages for
425 * I/O. In that case we need to make sure we've cleared the
426 * PageWriteback bit from the page to prevent the system from
427 * wedging later on.
428 */
429 put_io_page(io_page);
430 return ret;
431}
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
index ca5c8aa00a2f..dc963929de65 100644
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -226,23 +226,13 @@ static int setup_new_group_blocks(struct super_block *sb,
226 } 226 }
227 227
228 /* Zero out all of the reserved backup group descriptor table blocks */ 228 /* Zero out all of the reserved backup group descriptor table blocks */
229 for (i = 0, bit = gdblocks + 1, block = start + bit; 229 ext4_debug("clear inode table blocks %#04llx -> %#04llx\n",
230 i < reserved_gdb; i++, block++, bit++) { 230 block, sbi->s_itb_per_group);
231 struct buffer_head *gdb; 231 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
232 232 GFP_NOFS);
233 ext4_debug("clear reserved block %#04llx (+%d)\n", block, bit); 233 if (err)
234 234 goto exit_bh;
235 if ((err = extend_or_restart_transaction(handle, 1, bh)))
236 goto exit_bh;
237 235
238 if (IS_ERR(gdb = bclean(handle, sb, block))) {
239 err = PTR_ERR(gdb);
240 goto exit_bh;
241 }
242 ext4_handle_dirty_metadata(handle, NULL, gdb);
243 ext4_set_bit(bit, bh->b_data);
244 brelse(gdb);
245 }
246 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap, 236 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
247 input->block_bitmap - start); 237 input->block_bitmap - start);
248 ext4_set_bit(input->block_bitmap - start, bh->b_data); 238 ext4_set_bit(input->block_bitmap - start, bh->b_data);
@@ -251,28 +241,18 @@ static int setup_new_group_blocks(struct super_block *sb,
251 ext4_set_bit(input->inode_bitmap - start, bh->b_data); 241 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
252 242
253 /* Zero out all of the inode table blocks */ 243 /* Zero out all of the inode table blocks */
254 for (i = 0, block = input->inode_table, bit = block - start; 244 block = input->inode_table;
255 i < sbi->s_itb_per_group; i++, bit++, block++) { 245 ext4_debug("clear inode table blocks %#04llx -> %#04llx\n",
256 struct buffer_head *it; 246 block, sbi->s_itb_per_group);
257 247 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
258 ext4_debug("clear inode block %#04llx (+%d)\n", block, bit); 248 if (err)
259 249 goto exit_bh;
260 if ((err = extend_or_restart_transaction(handle, 1, bh)))
261 goto exit_bh;
262
263 if (IS_ERR(it = bclean(handle, sb, block))) {
264 err = PTR_ERR(it);
265 goto exit_bh;
266 }
267 ext4_handle_dirty_metadata(handle, NULL, it);
268 brelse(it);
269 ext4_set_bit(bit, bh->b_data);
270 }
271 250
272 if ((err = extend_or_restart_transaction(handle, 2, bh))) 251 if ((err = extend_or_restart_transaction(handle, 2, bh)))
273 goto exit_bh; 252 goto exit_bh;
274 253
275 mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8, bh->b_data); 254 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
255 bh->b_data);
276 ext4_handle_dirty_metadata(handle, NULL, bh); 256 ext4_handle_dirty_metadata(handle, NULL, bh);
277 brelse(bh); 257 brelse(bh);
278 /* Mark unused entries in inode bitmap used */ 258 /* Mark unused entries in inode bitmap used */
@@ -283,8 +263,8 @@ static int setup_new_group_blocks(struct super_block *sb,
283 goto exit_journal; 263 goto exit_journal;
284 } 264 }
285 265
286 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, 266 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
287 bh->b_data); 267 bh->b_data);
288 ext4_handle_dirty_metadata(handle, NULL, bh); 268 ext4_handle_dirty_metadata(handle, NULL, bh);
289exit_bh: 269exit_bh:
290 brelse(bh); 270 brelse(bh);
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index 8ecc1e590303..e32195d6aac3 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -40,6 +40,9 @@
40#include <linux/crc16.h> 40#include <linux/crc16.h>
41#include <asm/uaccess.h> 41#include <asm/uaccess.h>
42 42
43#include <linux/kthread.h>
44#include <linux/freezer.h>
45
43#include "ext4.h" 46#include "ext4.h"
44#include "ext4_jbd2.h" 47#include "ext4_jbd2.h"
45#include "xattr.h" 48#include "xattr.h"
@@ -49,8 +52,11 @@
49#define CREATE_TRACE_POINTS 52#define CREATE_TRACE_POINTS
50#include <trace/events/ext4.h> 53#include <trace/events/ext4.h>
51 54
52struct proc_dir_entry *ext4_proc_root; 55static struct proc_dir_entry *ext4_proc_root;
53static struct kset *ext4_kset; 56static struct kset *ext4_kset;
57struct ext4_lazy_init *ext4_li_info;
58struct mutex ext4_li_mtx;
59struct ext4_features *ext4_feat;
54 60
55static int ext4_load_journal(struct super_block *, struct ext4_super_block *, 61static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
56 unsigned long journal_devnum); 62 unsigned long journal_devnum);
@@ -67,14 +73,16 @@ static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
67static int ext4_unfreeze(struct super_block *sb); 73static int ext4_unfreeze(struct super_block *sb);
68static void ext4_write_super(struct super_block *sb); 74static void ext4_write_super(struct super_block *sb);
69static int ext4_freeze(struct super_block *sb); 75static int ext4_freeze(struct super_block *sb);
70static int ext4_get_sb(struct file_system_type *fs_type, int flags, 76static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
71 const char *dev_name, void *data, struct vfsmount *mnt); 77 const char *dev_name, void *data);
78static void ext4_destroy_lazyinit_thread(void);
79static void ext4_unregister_li_request(struct super_block *sb);
72 80
73#if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23) 81#if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
74static struct file_system_type ext3_fs_type = { 82static struct file_system_type ext3_fs_type = {
75 .owner = THIS_MODULE, 83 .owner = THIS_MODULE,
76 .name = "ext3", 84 .name = "ext3",
77 .get_sb = ext4_get_sb, 85 .mount = ext4_mount,
78 .kill_sb = kill_block_super, 86 .kill_sb = kill_block_super,
79 .fs_flags = FS_REQUIRES_DEV, 87 .fs_flags = FS_REQUIRES_DEV,
80}; 88};
@@ -701,6 +709,7 @@ static void ext4_put_super(struct super_block *sb)
701 struct ext4_super_block *es = sbi->s_es; 709 struct ext4_super_block *es = sbi->s_es;
702 int i, err; 710 int i, err;
703 711
712 ext4_unregister_li_request(sb);
704 dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED); 713 dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
705 714
706 flush_workqueue(sbi->dio_unwritten_wq); 715 flush_workqueue(sbi->dio_unwritten_wq);
@@ -717,6 +726,7 @@ static void ext4_put_super(struct super_block *sb)
717 ext4_abort(sb, "Couldn't clean up the journal"); 726 ext4_abort(sb, "Couldn't clean up the journal");
718 } 727 }
719 728
729 del_timer(&sbi->s_err_report);
720 ext4_release_system_zone(sb); 730 ext4_release_system_zone(sb);
721 ext4_mb_release(sb); 731 ext4_mb_release(sb);
722 ext4_ext_release(sb); 732 ext4_ext_release(sb);
@@ -818,12 +828,22 @@ static struct inode *ext4_alloc_inode(struct super_block *sb)
818 ei->cur_aio_dio = NULL; 828 ei->cur_aio_dio = NULL;
819 ei->i_sync_tid = 0; 829 ei->i_sync_tid = 0;
820 ei->i_datasync_tid = 0; 830 ei->i_datasync_tid = 0;
831 atomic_set(&ei->i_ioend_count, 0);
821 832
822 return &ei->vfs_inode; 833 return &ei->vfs_inode;
823} 834}
824 835
836static int ext4_drop_inode(struct inode *inode)
837{
838 int drop = generic_drop_inode(inode);
839
840 trace_ext4_drop_inode(inode, drop);
841 return drop;
842}
843
825static void ext4_destroy_inode(struct inode *inode) 844static void ext4_destroy_inode(struct inode *inode)
826{ 845{
846 ext4_ioend_wait(inode);
827 if (!list_empty(&(EXT4_I(inode)->i_orphan))) { 847 if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
828 ext4_msg(inode->i_sb, KERN_ERR, 848 ext4_msg(inode->i_sb, KERN_ERR,
829 "Inode %lu (%p): orphan list check failed!", 849 "Inode %lu (%p): orphan list check failed!",
@@ -1042,6 +1062,12 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1042 !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)) 1062 !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1043 seq_puts(seq, ",block_validity"); 1063 seq_puts(seq, ",block_validity");
1044 1064
1065 if (!test_opt(sb, INIT_INODE_TABLE))
1066 seq_puts(seq, ",noinit_inode_table");
1067 else if (sbi->s_li_wait_mult)
1068 seq_printf(seq, ",init_inode_table=%u",
1069 (unsigned) sbi->s_li_wait_mult);
1070
1045 ext4_show_quota_options(seq, sb); 1071 ext4_show_quota_options(seq, sb);
1046 1072
1047 return 0; 1073 return 0;
@@ -1157,6 +1183,7 @@ static const struct super_operations ext4_sops = {
1157 .destroy_inode = ext4_destroy_inode, 1183 .destroy_inode = ext4_destroy_inode,
1158 .write_inode = ext4_write_inode, 1184 .write_inode = ext4_write_inode,
1159 .dirty_inode = ext4_dirty_inode, 1185 .dirty_inode = ext4_dirty_inode,
1186 .drop_inode = ext4_drop_inode,
1160 .evict_inode = ext4_evict_inode, 1187 .evict_inode = ext4_evict_inode,
1161 .put_super = ext4_put_super, 1188 .put_super = ext4_put_super,
1162 .sync_fs = ext4_sync_fs, 1189 .sync_fs = ext4_sync_fs,
@@ -1177,6 +1204,7 @@ static const struct super_operations ext4_nojournal_sops = {
1177 .destroy_inode = ext4_destroy_inode, 1204 .destroy_inode = ext4_destroy_inode,
1178 .write_inode = ext4_write_inode, 1205 .write_inode = ext4_write_inode,
1179 .dirty_inode = ext4_dirty_inode, 1206 .dirty_inode = ext4_dirty_inode,
1207 .drop_inode = ext4_drop_inode,
1180 .evict_inode = ext4_evict_inode, 1208 .evict_inode = ext4_evict_inode,
1181 .write_super = ext4_write_super, 1209 .write_super = ext4_write_super,
1182 .put_super = ext4_put_super, 1210 .put_super = ext4_put_super,
@@ -1216,6 +1244,7 @@ enum {
1216 Opt_inode_readahead_blks, Opt_journal_ioprio, 1244 Opt_inode_readahead_blks, Opt_journal_ioprio,
1217 Opt_dioread_nolock, Opt_dioread_lock, 1245 Opt_dioread_nolock, Opt_dioread_lock,
1218 Opt_discard, Opt_nodiscard, 1246 Opt_discard, Opt_nodiscard,
1247 Opt_init_inode_table, Opt_noinit_inode_table,
1219}; 1248};
1220 1249
1221static const match_table_t tokens = { 1250static const match_table_t tokens = {
@@ -1286,6 +1315,9 @@ static const match_table_t tokens = {
1286 {Opt_dioread_lock, "dioread_lock"}, 1315 {Opt_dioread_lock, "dioread_lock"},
1287 {Opt_discard, "discard"}, 1316 {Opt_discard, "discard"},
1288 {Opt_nodiscard, "nodiscard"}, 1317 {Opt_nodiscard, "nodiscard"},
1318 {Opt_init_inode_table, "init_itable=%u"},
1319 {Opt_init_inode_table, "init_itable"},
1320 {Opt_noinit_inode_table, "noinit_itable"},
1289 {Opt_err, NULL}, 1321 {Opt_err, NULL},
1290}; 1322};
1291 1323
@@ -1756,6 +1788,20 @@ set_qf_format:
1756 case Opt_dioread_lock: 1788 case Opt_dioread_lock:
1757 clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK); 1789 clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1758 break; 1790 break;
1791 case Opt_init_inode_table:
1792 set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1793 if (args[0].from) {
1794 if (match_int(&args[0], &option))
1795 return 0;
1796 } else
1797 option = EXT4_DEF_LI_WAIT_MULT;
1798 if (option < 0)
1799 return 0;
1800 sbi->s_li_wait_mult = option;
1801 break;
1802 case Opt_noinit_inode_table:
1803 clear_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1804 break;
1759 default: 1805 default:
1760 ext4_msg(sb, KERN_ERR, 1806 ext4_msg(sb, KERN_ERR,
1761 "Unrecognized mount option \"%s\" " 1807 "Unrecognized mount option \"%s\" "
@@ -1939,7 +1985,8 @@ int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1939} 1985}
1940 1986
1941/* Called at mount-time, super-block is locked */ 1987/* Called at mount-time, super-block is locked */
1942static int ext4_check_descriptors(struct super_block *sb) 1988static int ext4_check_descriptors(struct super_block *sb,
1989 ext4_group_t *first_not_zeroed)
1943{ 1990{
1944 struct ext4_sb_info *sbi = EXT4_SB(sb); 1991 struct ext4_sb_info *sbi = EXT4_SB(sb);
1945 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block); 1992 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
@@ -1948,7 +1995,7 @@ static int ext4_check_descriptors(struct super_block *sb)
1948 ext4_fsblk_t inode_bitmap; 1995 ext4_fsblk_t inode_bitmap;
1949 ext4_fsblk_t inode_table; 1996 ext4_fsblk_t inode_table;
1950 int flexbg_flag = 0; 1997 int flexbg_flag = 0;
1951 ext4_group_t i; 1998 ext4_group_t i, grp = sbi->s_groups_count;
1952 1999
1953 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) 2000 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1954 flexbg_flag = 1; 2001 flexbg_flag = 1;
@@ -1964,6 +2011,10 @@ static int ext4_check_descriptors(struct super_block *sb)
1964 last_block = first_block + 2011 last_block = first_block +
1965 (EXT4_BLOCKS_PER_GROUP(sb) - 1); 2012 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1966 2013
2014 if ((grp == sbi->s_groups_count) &&
2015 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2016 grp = i;
2017
1967 block_bitmap = ext4_block_bitmap(sb, gdp); 2018 block_bitmap = ext4_block_bitmap(sb, gdp);
1968 if (block_bitmap < first_block || block_bitmap > last_block) { 2019 if (block_bitmap < first_block || block_bitmap > last_block) {
1969 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: " 2020 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
@@ -2001,6 +2052,8 @@ static int ext4_check_descriptors(struct super_block *sb)
2001 if (!flexbg_flag) 2052 if (!flexbg_flag)
2002 first_block += EXT4_BLOCKS_PER_GROUP(sb); 2053 first_block += EXT4_BLOCKS_PER_GROUP(sb);
2003 } 2054 }
2055 if (NULL != first_not_zeroed)
2056 *first_not_zeroed = grp;
2004 2057
2005 ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb)); 2058 ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2006 sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb)); 2059 sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
@@ -2373,6 +2426,7 @@ static struct ext4_attr ext4_attr_##_name = { \
2373#define EXT4_ATTR(name, mode, show, store) \ 2426#define EXT4_ATTR(name, mode, show, store) \
2374static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store) 2427static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2375 2428
2429#define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2376#define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL) 2430#define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2377#define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store) 2431#define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2378#define EXT4_RW_ATTR_SBI_UI(name, elname) \ 2432#define EXT4_RW_ATTR_SBI_UI(name, elname) \
@@ -2409,6 +2463,16 @@ static struct attribute *ext4_attrs[] = {
2409 NULL, 2463 NULL,
2410}; 2464};
2411 2465
2466/* Features this copy of ext4 supports */
2467EXT4_INFO_ATTR(lazy_itable_init);
2468EXT4_INFO_ATTR(batched_discard);
2469
2470static struct attribute *ext4_feat_attrs[] = {
2471 ATTR_LIST(lazy_itable_init),
2472 ATTR_LIST(batched_discard),
2473 NULL,
2474};
2475
2412static ssize_t ext4_attr_show(struct kobject *kobj, 2476static ssize_t ext4_attr_show(struct kobject *kobj,
2413 struct attribute *attr, char *buf) 2477 struct attribute *attr, char *buf)
2414{ 2478{
@@ -2437,7 +2501,6 @@ static void ext4_sb_release(struct kobject *kobj)
2437 complete(&sbi->s_kobj_unregister); 2501 complete(&sbi->s_kobj_unregister);
2438} 2502}
2439 2503
2440
2441static const struct sysfs_ops ext4_attr_ops = { 2504static const struct sysfs_ops ext4_attr_ops = {
2442 .show = ext4_attr_show, 2505 .show = ext4_attr_show,
2443 .store = ext4_attr_store, 2506 .store = ext4_attr_store,
@@ -2449,6 +2512,17 @@ static struct kobj_type ext4_ktype = {
2449 .release = ext4_sb_release, 2512 .release = ext4_sb_release,
2450}; 2513};
2451 2514
2515static void ext4_feat_release(struct kobject *kobj)
2516{
2517 complete(&ext4_feat->f_kobj_unregister);
2518}
2519
2520static struct kobj_type ext4_feat_ktype = {
2521 .default_attrs = ext4_feat_attrs,
2522 .sysfs_ops = &ext4_attr_ops,
2523 .release = ext4_feat_release,
2524};
2525
2452/* 2526/*
2453 * Check whether this filesystem can be mounted based on 2527 * Check whether this filesystem can be mounted based on
2454 * the features present and the RDONLY/RDWR mount requested. 2528 * the features present and the RDONLY/RDWR mount requested.
@@ -2539,6 +2613,368 @@ static void print_daily_error_info(unsigned long arg)
2539 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */ 2613 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */
2540} 2614}
2541 2615
2616static void ext4_lazyinode_timeout(unsigned long data)
2617{
2618 struct task_struct *p = (struct task_struct *)data;
2619 wake_up_process(p);
2620}
2621
2622/* Find next suitable group and run ext4_init_inode_table */
2623static int ext4_run_li_request(struct ext4_li_request *elr)
2624{
2625 struct ext4_group_desc *gdp = NULL;
2626 ext4_group_t group, ngroups;
2627 struct super_block *sb;
2628 unsigned long timeout = 0;
2629 int ret = 0;
2630
2631 sb = elr->lr_super;
2632 ngroups = EXT4_SB(sb)->s_groups_count;
2633
2634 for (group = elr->lr_next_group; group < ngroups; group++) {
2635 gdp = ext4_get_group_desc(sb, group, NULL);
2636 if (!gdp) {
2637 ret = 1;
2638 break;
2639 }
2640
2641 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2642 break;
2643 }
2644
2645 if (group == ngroups)
2646 ret = 1;
2647
2648 if (!ret) {
2649 timeout = jiffies;
2650 ret = ext4_init_inode_table(sb, group,
2651 elr->lr_timeout ? 0 : 1);
2652 if (elr->lr_timeout == 0) {
2653 timeout = jiffies - timeout;
2654 if (elr->lr_sbi->s_li_wait_mult)
2655 timeout *= elr->lr_sbi->s_li_wait_mult;
2656 else
2657 timeout *= 20;
2658 elr->lr_timeout = timeout;
2659 }
2660 elr->lr_next_sched = jiffies + elr->lr_timeout;
2661 elr->lr_next_group = group + 1;
2662 }
2663
2664 return ret;
2665}
2666
2667/*
2668 * Remove lr_request from the list_request and free the
2669 * request tructure. Should be called with li_list_mtx held
2670 */
2671static void ext4_remove_li_request(struct ext4_li_request *elr)
2672{
2673 struct ext4_sb_info *sbi;
2674
2675 if (!elr)
2676 return;
2677
2678 sbi = elr->lr_sbi;
2679
2680 list_del(&elr->lr_request);
2681 sbi->s_li_request = NULL;
2682 kfree(elr);
2683}
2684
2685static void ext4_unregister_li_request(struct super_block *sb)
2686{
2687 struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2688
2689 if (!ext4_li_info)
2690 return;
2691
2692 mutex_lock(&ext4_li_info->li_list_mtx);
2693 ext4_remove_li_request(elr);
2694 mutex_unlock(&ext4_li_info->li_list_mtx);
2695}
2696
2697/*
2698 * This is the function where ext4lazyinit thread lives. It walks
2699 * through the request list searching for next scheduled filesystem.
2700 * When such a fs is found, run the lazy initialization request
2701 * (ext4_rn_li_request) and keep track of the time spend in this
2702 * function. Based on that time we compute next schedule time of
2703 * the request. When walking through the list is complete, compute
2704 * next waking time and put itself into sleep.
2705 */
2706static int ext4_lazyinit_thread(void *arg)
2707{
2708 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2709 struct list_head *pos, *n;
2710 struct ext4_li_request *elr;
2711 unsigned long next_wakeup;
2712 DEFINE_WAIT(wait);
2713
2714 BUG_ON(NULL == eli);
2715
2716 eli->li_timer.data = (unsigned long)current;
2717 eli->li_timer.function = ext4_lazyinode_timeout;
2718
2719 eli->li_task = current;
2720 wake_up(&eli->li_wait_task);
2721
2722cont_thread:
2723 while (true) {
2724 next_wakeup = MAX_JIFFY_OFFSET;
2725
2726 mutex_lock(&eli->li_list_mtx);
2727 if (list_empty(&eli->li_request_list)) {
2728 mutex_unlock(&eli->li_list_mtx);
2729 goto exit_thread;
2730 }
2731
2732 list_for_each_safe(pos, n, &eli->li_request_list) {
2733 elr = list_entry(pos, struct ext4_li_request,
2734 lr_request);
2735
2736 if (time_after_eq(jiffies, elr->lr_next_sched)) {
2737 if (ext4_run_li_request(elr) != 0) {
2738 /* error, remove the lazy_init job */
2739 ext4_remove_li_request(elr);
2740 continue;
2741 }
2742 }
2743
2744 if (time_before(elr->lr_next_sched, next_wakeup))
2745 next_wakeup = elr->lr_next_sched;
2746 }
2747 mutex_unlock(&eli->li_list_mtx);
2748
2749 if (freezing(current))
2750 refrigerator();
2751
2752 if ((time_after_eq(jiffies, next_wakeup)) ||
2753 (MAX_JIFFY_OFFSET == next_wakeup)) {
2754 cond_resched();
2755 continue;
2756 }
2757
2758 eli->li_timer.expires = next_wakeup;
2759 add_timer(&eli->li_timer);
2760 prepare_to_wait(&eli->li_wait_daemon, &wait,
2761 TASK_INTERRUPTIBLE);
2762 if (time_before(jiffies, next_wakeup))
2763 schedule();
2764 finish_wait(&eli->li_wait_daemon, &wait);
2765 }
2766
2767exit_thread:
2768 /*
2769 * It looks like the request list is empty, but we need
2770 * to check it under the li_list_mtx lock, to prevent any
2771 * additions into it, and of course we should lock ext4_li_mtx
2772 * to atomically free the list and ext4_li_info, because at
2773 * this point another ext4 filesystem could be registering
2774 * new one.
2775 */
2776 mutex_lock(&ext4_li_mtx);
2777 mutex_lock(&eli->li_list_mtx);
2778 if (!list_empty(&eli->li_request_list)) {
2779 mutex_unlock(&eli->li_list_mtx);
2780 mutex_unlock(&ext4_li_mtx);
2781 goto cont_thread;
2782 }
2783 mutex_unlock(&eli->li_list_mtx);
2784 del_timer_sync(&ext4_li_info->li_timer);
2785 eli->li_task = NULL;
2786 wake_up(&eli->li_wait_task);
2787
2788 kfree(ext4_li_info);
2789 ext4_li_info = NULL;
2790 mutex_unlock(&ext4_li_mtx);
2791
2792 return 0;
2793}
2794
2795static void ext4_clear_request_list(void)
2796{
2797 struct list_head *pos, *n;
2798 struct ext4_li_request *elr;
2799
2800 mutex_lock(&ext4_li_info->li_list_mtx);
2801 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2802 elr = list_entry(pos, struct ext4_li_request,
2803 lr_request);
2804 ext4_remove_li_request(elr);
2805 }
2806 mutex_unlock(&ext4_li_info->li_list_mtx);
2807}
2808
2809static int ext4_run_lazyinit_thread(void)
2810{
2811 struct task_struct *t;
2812
2813 t = kthread_run(ext4_lazyinit_thread, ext4_li_info, "ext4lazyinit");
2814 if (IS_ERR(t)) {
2815 int err = PTR_ERR(t);
2816 ext4_clear_request_list();
2817 del_timer_sync(&ext4_li_info->li_timer);
2818 kfree(ext4_li_info);
2819 ext4_li_info = NULL;
2820 printk(KERN_CRIT "EXT4: error %d creating inode table "
2821 "initialization thread\n",
2822 err);
2823 return err;
2824 }
2825 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2826
2827 wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2828 return 0;
2829}
2830
2831/*
2832 * Check whether it make sense to run itable init. thread or not.
2833 * If there is at least one uninitialized inode table, return
2834 * corresponding group number, else the loop goes through all
2835 * groups and return total number of groups.
2836 */
2837static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2838{
2839 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2840 struct ext4_group_desc *gdp = NULL;
2841
2842 for (group = 0; group < ngroups; group++) {
2843 gdp = ext4_get_group_desc(sb, group, NULL);
2844 if (!gdp)
2845 continue;
2846
2847 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2848 break;
2849 }
2850
2851 return group;
2852}
2853
2854static int ext4_li_info_new(void)
2855{
2856 struct ext4_lazy_init *eli = NULL;
2857
2858 eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2859 if (!eli)
2860 return -ENOMEM;
2861
2862 eli->li_task = NULL;
2863 INIT_LIST_HEAD(&eli->li_request_list);
2864 mutex_init(&eli->li_list_mtx);
2865
2866 init_waitqueue_head(&eli->li_wait_daemon);
2867 init_waitqueue_head(&eli->li_wait_task);
2868 init_timer(&eli->li_timer);
2869 eli->li_state |= EXT4_LAZYINIT_QUIT;
2870
2871 ext4_li_info = eli;
2872
2873 return 0;
2874}
2875
2876static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2877 ext4_group_t start)
2878{
2879 struct ext4_sb_info *sbi = EXT4_SB(sb);
2880 struct ext4_li_request *elr;
2881 unsigned long rnd;
2882
2883 elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2884 if (!elr)
2885 return NULL;
2886
2887 elr->lr_super = sb;
2888 elr->lr_sbi = sbi;
2889 elr->lr_next_group = start;
2890
2891 /*
2892 * Randomize first schedule time of the request to
2893 * spread the inode table initialization requests
2894 * better.
2895 */
2896 get_random_bytes(&rnd, sizeof(rnd));
2897 elr->lr_next_sched = jiffies + (unsigned long)rnd %
2898 (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2899
2900 return elr;
2901}
2902
2903static int ext4_register_li_request(struct super_block *sb,
2904 ext4_group_t first_not_zeroed)
2905{
2906 struct ext4_sb_info *sbi = EXT4_SB(sb);
2907 struct ext4_li_request *elr;
2908 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2909 int ret;
2910
2911 if (sbi->s_li_request != NULL)
2912 return 0;
2913
2914 if (first_not_zeroed == ngroups ||
2915 (sb->s_flags & MS_RDONLY) ||
2916 !test_opt(sb, INIT_INODE_TABLE)) {
2917 sbi->s_li_request = NULL;
2918 return 0;
2919 }
2920
2921 if (first_not_zeroed == ngroups) {
2922 sbi->s_li_request = NULL;
2923 return 0;
2924 }
2925
2926 elr = ext4_li_request_new(sb, first_not_zeroed);
2927 if (!elr)
2928 return -ENOMEM;
2929
2930 mutex_lock(&ext4_li_mtx);
2931
2932 if (NULL == ext4_li_info) {
2933 ret = ext4_li_info_new();
2934 if (ret)
2935 goto out;
2936 }
2937
2938 mutex_lock(&ext4_li_info->li_list_mtx);
2939 list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2940 mutex_unlock(&ext4_li_info->li_list_mtx);
2941
2942 sbi->s_li_request = elr;
2943
2944 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2945 ret = ext4_run_lazyinit_thread();
2946 if (ret)
2947 goto out;
2948 }
2949out:
2950 mutex_unlock(&ext4_li_mtx);
2951 if (ret)
2952 kfree(elr);
2953 return ret;
2954}
2955
2956/*
2957 * We do not need to lock anything since this is called on
2958 * module unload.
2959 */
2960static void ext4_destroy_lazyinit_thread(void)
2961{
2962 /*
2963 * If thread exited earlier
2964 * there's nothing to be done.
2965 */
2966 if (!ext4_li_info)
2967 return;
2968
2969 ext4_clear_request_list();
2970
2971 while (ext4_li_info->li_task) {
2972 wake_up(&ext4_li_info->li_wait_daemon);
2973 wait_event(ext4_li_info->li_wait_task,
2974 ext4_li_info->li_task == NULL);
2975 }
2976}
2977
2542static int ext4_fill_super(struct super_block *sb, void *data, int silent) 2978static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2543 __releases(kernel_lock) 2979 __releases(kernel_lock)
2544 __acquires(kernel_lock) 2980 __acquires(kernel_lock)
@@ -2564,6 +3000,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2564 __u64 blocks_count; 3000 __u64 blocks_count;
2565 int err; 3001 int err;
2566 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO; 3002 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3003 ext4_group_t first_not_zeroed;
2567 3004
2568 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 3005 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2569 if (!sbi) 3006 if (!sbi)
@@ -2624,6 +3061,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2624 3061
2625 /* Set defaults before we parse the mount options */ 3062 /* Set defaults before we parse the mount options */
2626 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 3063 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3064 set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
2627 if (def_mount_opts & EXT4_DEFM_DEBUG) 3065 if (def_mount_opts & EXT4_DEFM_DEBUG)
2628 set_opt(sbi->s_mount_opt, DEBUG); 3066 set_opt(sbi->s_mount_opt, DEBUG);
2629 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) { 3067 if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
@@ -2826,13 +3264,14 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2826 * Test whether we have more sectors than will fit in sector_t, 3264 * Test whether we have more sectors than will fit in sector_t,
2827 * and whether the max offset is addressable by the page cache. 3265 * and whether the max offset is addressable by the page cache.
2828 */ 3266 */
2829 ret = generic_check_addressable(sb->s_blocksize_bits, 3267 err = generic_check_addressable(sb->s_blocksize_bits,
2830 ext4_blocks_count(es)); 3268 ext4_blocks_count(es));
2831 if (ret) { 3269 if (err) {
2832 ext4_msg(sb, KERN_ERR, "filesystem" 3270 ext4_msg(sb, KERN_ERR, "filesystem"
2833 " too large to mount safely on this system"); 3271 " too large to mount safely on this system");
2834 if (sizeof(sector_t) < 8) 3272 if (sizeof(sector_t) < 8)
2835 ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled"); 3273 ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3274 ret = err;
2836 goto failed_mount; 3275 goto failed_mount;
2837 } 3276 }
2838 3277
@@ -2901,7 +3340,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2901 goto failed_mount2; 3340 goto failed_mount2;
2902 } 3341 }
2903 } 3342 }
2904 if (!ext4_check_descriptors(sb)) { 3343 if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
2905 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!"); 3344 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2906 goto failed_mount2; 3345 goto failed_mount2;
2907 } 3346 }
@@ -2917,6 +3356,24 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2917 get_random_bytes(&sbi->s_next_generation, sizeof(u32)); 3356 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2918 spin_lock_init(&sbi->s_next_gen_lock); 3357 spin_lock_init(&sbi->s_next_gen_lock);
2919 3358
3359 err = percpu_counter_init(&sbi->s_freeblocks_counter,
3360 ext4_count_free_blocks(sb));
3361 if (!err) {
3362 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3363 ext4_count_free_inodes(sb));
3364 }
3365 if (!err) {
3366 err = percpu_counter_init(&sbi->s_dirs_counter,
3367 ext4_count_dirs(sb));
3368 }
3369 if (!err) {
3370 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3371 }
3372 if (err) {
3373 ext4_msg(sb, KERN_ERR, "insufficient memory");
3374 goto failed_mount3;
3375 }
3376
2920 sbi->s_stripe = ext4_get_stripe_size(sbi); 3377 sbi->s_stripe = ext4_get_stripe_size(sbi);
2921 sbi->s_max_writeback_mb_bump = 128; 3378 sbi->s_max_writeback_mb_bump = 128;
2922 3379
@@ -3015,22 +3472,19 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3015 } 3472 }
3016 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio); 3473 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3017 3474
3018no_journal: 3475 /*
3019 err = percpu_counter_init(&sbi->s_freeblocks_counter, 3476 * The journal may have updated the bg summary counts, so we
3020 ext4_count_free_blocks(sb)); 3477 * need to update the global counters.
3021 if (!err) 3478 */
3022 err = percpu_counter_init(&sbi->s_freeinodes_counter, 3479 percpu_counter_set(&sbi->s_freeblocks_counter,
3023 ext4_count_free_inodes(sb)); 3480 ext4_count_free_blocks(sb));
3024 if (!err) 3481 percpu_counter_set(&sbi->s_freeinodes_counter,
3025 err = percpu_counter_init(&sbi->s_dirs_counter, 3482 ext4_count_free_inodes(sb));
3026 ext4_count_dirs(sb)); 3483 percpu_counter_set(&sbi->s_dirs_counter,
3027 if (!err) 3484 ext4_count_dirs(sb));
3028 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0); 3485 percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3029 if (err) {
3030 ext4_msg(sb, KERN_ERR, "insufficient memory");
3031 goto failed_mount_wq;
3032 }
3033 3486
3487no_journal:
3034 EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten"); 3488 EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
3035 if (!EXT4_SB(sb)->dio_unwritten_wq) { 3489 if (!EXT4_SB(sb)->dio_unwritten_wq) {
3036 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n"); 3490 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
@@ -3122,6 +3576,10 @@ no_journal:
3122 goto failed_mount4; 3576 goto failed_mount4;
3123 } 3577 }
3124 3578
3579 err = ext4_register_li_request(sb, first_not_zeroed);
3580 if (err)
3581 goto failed_mount4;
3582
3125 sbi->s_kobj.kset = ext4_kset; 3583 sbi->s_kobj.kset = ext4_kset;
3126 init_completion(&sbi->s_kobj_unregister); 3584 init_completion(&sbi->s_kobj_unregister);
3127 err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL, 3585 err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
@@ -3176,10 +3634,6 @@ failed_mount_wq:
3176 jbd2_journal_destroy(sbi->s_journal); 3634 jbd2_journal_destroy(sbi->s_journal);
3177 sbi->s_journal = NULL; 3635 sbi->s_journal = NULL;
3178 } 3636 }
3179 percpu_counter_destroy(&sbi->s_freeblocks_counter);
3180 percpu_counter_destroy(&sbi->s_freeinodes_counter);
3181 percpu_counter_destroy(&sbi->s_dirs_counter);
3182 percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3183failed_mount3: 3637failed_mount3:
3184 if (sbi->s_flex_groups) { 3638 if (sbi->s_flex_groups) {
3185 if (is_vmalloc_addr(sbi->s_flex_groups)) 3639 if (is_vmalloc_addr(sbi->s_flex_groups))
@@ -3187,6 +3641,10 @@ failed_mount3:
3187 else 3641 else
3188 kfree(sbi->s_flex_groups); 3642 kfree(sbi->s_flex_groups);
3189 } 3643 }
3644 percpu_counter_destroy(&sbi->s_freeblocks_counter);
3645 percpu_counter_destroy(&sbi->s_freeinodes_counter);
3646 percpu_counter_destroy(&sbi->s_dirs_counter);
3647 percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3190failed_mount2: 3648failed_mount2:
3191 for (i = 0; i < db_count; i++) 3649 for (i = 0; i < db_count; i++)
3192 brelse(sbi->s_group_desc[i]); 3650 brelse(sbi->s_group_desc[i]);
@@ -3461,7 +3919,7 @@ static int ext4_load_journal(struct super_block *sb,
3461 EXT4_SB(sb)->s_journal = journal; 3919 EXT4_SB(sb)->s_journal = journal;
3462 ext4_clear_journal_err(sb, es); 3920 ext4_clear_journal_err(sb, es);
3463 3921
3464 if (journal_devnum && 3922 if (!really_read_only && journal_devnum &&
3465 journal_devnum != le32_to_cpu(es->s_journal_dev)) { 3923 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3466 es->s_journal_dev = cpu_to_le32(journal_devnum); 3924 es->s_journal_dev = cpu_to_le32(journal_devnum);
3467 3925
@@ -3515,9 +3973,10 @@ static int ext4_commit_super(struct super_block *sb, int sync)
3515 es->s_kbytes_written = 3973 es->s_kbytes_written =
3516 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written); 3974 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
3517 ext4_free_blocks_count_set(es, percpu_counter_sum_positive( 3975 ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3518 &EXT4_SB(sb)->s_freeblocks_counter)); 3976 &EXT4_SB(sb)->s_freeblocks_counter));
3519 es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive( 3977 es->s_free_inodes_count =
3520 &EXT4_SB(sb)->s_freeinodes_counter)); 3978 cpu_to_le32(percpu_counter_sum_positive(
3979 &EXT4_SB(sb)->s_freeinodes_counter));
3521 sb->s_dirt = 0; 3980 sb->s_dirt = 0;
3522 BUFFER_TRACE(sbh, "marking dirty"); 3981 BUFFER_TRACE(sbh, "marking dirty");
3523 mark_buffer_dirty(sbh); 3982 mark_buffer_dirty(sbh);
@@ -3835,6 +4294,19 @@ static int ext4_remount(struct super_block *sb, int *flags, char *data)
3835 enable_quota = 1; 4294 enable_quota = 1;
3836 } 4295 }
3837 } 4296 }
4297
4298 /*
4299 * Reinitialize lazy itable initialization thread based on
4300 * current settings
4301 */
4302 if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4303 ext4_unregister_li_request(sb);
4304 else {
4305 ext4_group_t first_not_zeroed;
4306 first_not_zeroed = ext4_has_uninit_itable(sb);
4307 ext4_register_li_request(sb, first_not_zeroed);
4308 }
4309
3838 ext4_setup_system_zone(sb); 4310 ext4_setup_system_zone(sb);
3839 if (sbi->s_journal == NULL) 4311 if (sbi->s_journal == NULL)
3840 ext4_commit_super(sb, 1); 4312 ext4_commit_super(sb, 1);
@@ -4105,12 +4577,10 @@ static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4105 4577
4106static int ext4_quota_off(struct super_block *sb, int type) 4578static int ext4_quota_off(struct super_block *sb, int type)
4107{ 4579{
4108 /* Force all delayed allocation blocks to be allocated */ 4580 /* Force all delayed allocation blocks to be allocated.
4109 if (test_opt(sb, DELALLOC)) { 4581 * Caller already holds s_umount sem */
4110 down_read(&sb->s_umount); 4582 if (test_opt(sb, DELALLOC))
4111 sync_filesystem(sb); 4583 sync_filesystem(sb);
4112 up_read(&sb->s_umount);
4113 }
4114 4584
4115 return dquot_quota_off(sb, type); 4585 return dquot_quota_off(sb, type);
4116} 4586}
@@ -4216,17 +4686,17 @@ out:
4216 4686
4217#endif 4687#endif
4218 4688
4219static int ext4_get_sb(struct file_system_type *fs_type, int flags, 4689static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4220 const char *dev_name, void *data, struct vfsmount *mnt) 4690 const char *dev_name, void *data)
4221{ 4691{
4222 return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt); 4692 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4223} 4693}
4224 4694
4225#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23) 4695#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4226static struct file_system_type ext2_fs_type = { 4696static struct file_system_type ext2_fs_type = {
4227 .owner = THIS_MODULE, 4697 .owner = THIS_MODULE,
4228 .name = "ext2", 4698 .name = "ext2",
4229 .get_sb = ext4_get_sb, 4699 .mount = ext4_mount,
4230 .kill_sb = kill_block_super, 4700 .kill_sb = kill_block_super,
4231 .fs_flags = FS_REQUIRES_DEV, 4701 .fs_flags = FS_REQUIRES_DEV,
4232}; 4702};
@@ -4271,28 +4741,58 @@ static inline void unregister_as_ext3(void) { }
4271static struct file_system_type ext4_fs_type = { 4741static struct file_system_type ext4_fs_type = {
4272 .owner = THIS_MODULE, 4742 .owner = THIS_MODULE,
4273 .name = "ext4", 4743 .name = "ext4",
4274 .get_sb = ext4_get_sb, 4744 .mount = ext4_mount,
4275 .kill_sb = kill_block_super, 4745 .kill_sb = kill_block_super,
4276 .fs_flags = FS_REQUIRES_DEV, 4746 .fs_flags = FS_REQUIRES_DEV,
4277}; 4747};
4278 4748
4279static int __init init_ext4_fs(void) 4749int __init ext4_init_feat_adverts(void)
4750{
4751 struct ext4_features *ef;
4752 int ret = -ENOMEM;
4753
4754 ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4755 if (!ef)
4756 goto out;
4757
4758 ef->f_kobj.kset = ext4_kset;
4759 init_completion(&ef->f_kobj_unregister);
4760 ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4761 "features");
4762 if (ret) {
4763 kfree(ef);
4764 goto out;
4765 }
4766
4767 ext4_feat = ef;
4768 ret = 0;
4769out:
4770 return ret;
4771}
4772
4773static int __init ext4_init_fs(void)
4280{ 4774{
4281 int err; 4775 int err;
4282 4776
4283 ext4_check_flag_values(); 4777 ext4_check_flag_values();
4284 err = init_ext4_system_zone(); 4778 err = ext4_init_pageio();
4285 if (err) 4779 if (err)
4286 return err; 4780 return err;
4781 err = ext4_init_system_zone();
4782 if (err)
4783 goto out5;
4287 ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj); 4784 ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4288 if (!ext4_kset) 4785 if (!ext4_kset)
4289 goto out4; 4786 goto out4;
4290 ext4_proc_root = proc_mkdir("fs/ext4", NULL); 4787 ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4291 err = init_ext4_mballoc(); 4788
4789 err = ext4_init_feat_adverts();
4790
4791 err = ext4_init_mballoc();
4292 if (err) 4792 if (err)
4293 goto out3; 4793 goto out3;
4294 4794
4295 err = init_ext4_xattr(); 4795 err = ext4_init_xattr();
4296 if (err) 4796 if (err)
4297 goto out2; 4797 goto out2;
4298 err = init_inodecache(); 4798 err = init_inodecache();
@@ -4303,38 +4803,46 @@ static int __init init_ext4_fs(void)
4303 err = register_filesystem(&ext4_fs_type); 4803 err = register_filesystem(&ext4_fs_type);
4304 if (err) 4804 if (err)
4305 goto out; 4805 goto out;
4806
4807 ext4_li_info = NULL;
4808 mutex_init(&ext4_li_mtx);
4306 return 0; 4809 return 0;
4307out: 4810out:
4308 unregister_as_ext2(); 4811 unregister_as_ext2();
4309 unregister_as_ext3(); 4812 unregister_as_ext3();
4310 destroy_inodecache(); 4813 destroy_inodecache();
4311out1: 4814out1:
4312 exit_ext4_xattr(); 4815 ext4_exit_xattr();
4313out2: 4816out2:
4314 exit_ext4_mballoc(); 4817 ext4_exit_mballoc();
4315out3: 4818out3:
4819 kfree(ext4_feat);
4316 remove_proc_entry("fs/ext4", NULL); 4820 remove_proc_entry("fs/ext4", NULL);
4317 kset_unregister(ext4_kset); 4821 kset_unregister(ext4_kset);
4318out4: 4822out4:
4319 exit_ext4_system_zone(); 4823 ext4_exit_system_zone();
4824out5:
4825 ext4_exit_pageio();
4320 return err; 4826 return err;
4321} 4827}
4322 4828
4323static void __exit exit_ext4_fs(void) 4829static void __exit ext4_exit_fs(void)
4324{ 4830{
4831 ext4_destroy_lazyinit_thread();
4325 unregister_as_ext2(); 4832 unregister_as_ext2();
4326 unregister_as_ext3(); 4833 unregister_as_ext3();
4327 unregister_filesystem(&ext4_fs_type); 4834 unregister_filesystem(&ext4_fs_type);
4328 destroy_inodecache(); 4835 destroy_inodecache();
4329 exit_ext4_xattr(); 4836 ext4_exit_xattr();
4330 exit_ext4_mballoc(); 4837 ext4_exit_mballoc();
4331 remove_proc_entry("fs/ext4", NULL); 4838 remove_proc_entry("fs/ext4", NULL);
4332 kset_unregister(ext4_kset); 4839 kset_unregister(ext4_kset);
4333 exit_ext4_system_zone(); 4840 ext4_exit_system_zone();
4841 ext4_exit_pageio();
4334} 4842}
4335 4843
4336MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others"); 4844MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4337MODULE_DESCRIPTION("Fourth Extended Filesystem"); 4845MODULE_DESCRIPTION("Fourth Extended Filesystem");
4338MODULE_LICENSE("GPL"); 4846MODULE_LICENSE("GPL");
4339module_init(init_ext4_fs) 4847module_init(ext4_init_fs)
4340module_exit(exit_ext4_fs) 4848module_exit(ext4_exit_fs)
diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
index 3a8cd8dff1ad..fa4b899da4b3 100644
--- a/fs/ext4/xattr.c
+++ b/fs/ext4/xattr.c
@@ -1588,7 +1588,7 @@ static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1588#undef BLOCK_HASH_SHIFT 1588#undef BLOCK_HASH_SHIFT
1589 1589
1590int __init 1590int __init
1591init_ext4_xattr(void) 1591ext4_init_xattr(void)
1592{ 1592{
1593 ext4_xattr_cache = mb_cache_create("ext4_xattr", 6); 1593 ext4_xattr_cache = mb_cache_create("ext4_xattr", 6);
1594 if (!ext4_xattr_cache) 1594 if (!ext4_xattr_cache)
@@ -1597,7 +1597,7 @@ init_ext4_xattr(void)
1597} 1597}
1598 1598
1599void 1599void
1600exit_ext4_xattr(void) 1600ext4_exit_xattr(void)
1601{ 1601{
1602 if (ext4_xattr_cache) 1602 if (ext4_xattr_cache)
1603 mb_cache_destroy(ext4_xattr_cache); 1603 mb_cache_destroy(ext4_xattr_cache);
diff --git a/fs/ext4/xattr.h b/fs/ext4/xattr.h
index 518e96e43905..1ef16520b950 100644
--- a/fs/ext4/xattr.h
+++ b/fs/ext4/xattr.h
@@ -83,8 +83,8 @@ extern void ext4_xattr_put_super(struct super_block *);
83extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize, 83extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
84 struct ext4_inode *raw_inode, handle_t *handle); 84 struct ext4_inode *raw_inode, handle_t *handle);
85 85
86extern int init_ext4_xattr(void); 86extern int __init ext4_init_xattr(void);
87extern void exit_ext4_xattr(void); 87extern void ext4_exit_xattr(void);
88 88
89extern const struct xattr_handler *ext4_xattr_handlers[]; 89extern const struct xattr_handler *ext4_xattr_handlers[];
90 90
@@ -121,14 +121,14 @@ ext4_xattr_put_super(struct super_block *sb)
121{ 121{
122} 122}
123 123
124static inline int 124static __init inline int
125init_ext4_xattr(void) 125ext4_init_xattr(void)
126{ 126{
127 return 0; 127 return 0;
128} 128}
129 129
130static inline void 130static inline void
131exit_ext4_xattr(void) 131ext4_exit_xattr(void)
132{ 132{
133} 133}
134 134
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-07 20:51:42 -0500