1 This patch implements the large EA support in ext4. If the size of
2 an EA value is larger than the blocksize, then the EA value would
3 not be saved in the external EA block, instead it would be saved
4 in an external EA inode. So, the patch also helps support a larger
7 Index: linux-stage/fs/ext4/ext4.h
8 ===================================================================
9 --- linux-stage.orig/fs/ext4/ext4.h
10 +++ linux-stage/fs/ext4/ext4.h
11 @@ -1545,6 +1545,7 @@ static inline void ext4_clear_state_flag
12 EXT4_FEATURE_INCOMPAT_EXTENTS| \
13 EXT4_FEATURE_INCOMPAT_64BIT| \
14 EXT4_FEATURE_INCOMPAT_FLEX_BG| \
15 + EXT4_FEATURE_INCOMPAT_EA_INODE| \
16 EXT4_FEATURE_INCOMPAT_MMP | \
17 EXT4_FEATURE_INCOMPAT_DIRDATA| \
18 EXT4_FEATURE_INCOMPAT_INLINE_DATA)
19 @@ -1945,6 +1946,12 @@ struct mmpd_data {
20 #define EXT4_MMP_MAX_CHECK_INTERVAL 300UL
23 + * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1MB
24 + * This limit is arbitrary, but is reasonable for the xattr API.
26 +#define EXT4_XATTR_MAX_LARGE_EA_SIZE (1024 * 1024)
32 @@ -1956,6 +1963,10 @@ struct mmpd_data {
33 # define ATTRIB_NORET __attribute__((noreturn))
34 # define NORET_AND noreturn,
36 +struct ext4_xattr_ino_array {
37 + unsigned int xia_count; /* # of used item in the array */
38 + unsigned int xia_inodes[0];
41 extern unsigned int ext4_count_free(char *bitmap, unsigned numchars);
42 void ext4_inode_bitmap_csum_set(struct super_block *sb, ext4_group_t group,
43 @@ -2169,6 +2180,7 @@ extern void ext4_set_inode_flags(struct
44 extern void ext4_get_inode_flags(struct ext4_inode_info *);
45 extern int ext4_alloc_da_blocks(struct inode *inode);
46 extern void ext4_set_aops(struct inode *inode);
47 +extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk);
48 extern int ext4_writepage_trans_blocks(struct inode *);
49 extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
50 extern int ext4_block_truncate_page(handle_t *handle,
51 Index: linux-stage/fs/ext4/inode.c
52 ===================================================================
53 --- linux-stage.orig/fs/ext4/inode.c
54 +++ linux-stage/fs/ext4/inode.c
55 @@ -135,8 +135,6 @@ static void ext4_invalidatepage(struct p
57 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
58 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
59 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
63 * Test whether an inode is a fast symlink.
64 @@ -182,6 +180,8 @@ void ext4_evict_inode(struct inode *inod
68 + int extra_credits = 3;
69 + struct ext4_xattr_ino_array *lea_ino_array = NULL;
71 trace_ext4_evict_inode(inode);
73 @@ -235,8 +235,8 @@ void ext4_evict_inode(struct inode *inod
74 * protection against it
76 sb_start_intwrite(inode->i_sb);
77 - handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
78 - ext4_blocks_for_truncate(inode)+3);
80 + handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits);
82 ext4_std_error(inode->i_sb, PTR_ERR(handle));
84 @@ -251,6 +251,32 @@ void ext4_evict_inode(struct inode *inod
87 ext4_handle_sync(handle);
89 + /* Delete xattr inode before deleting the main inode. */
90 + err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
92 + ext4_warning(inode->i_sb,
93 + "couldn't delete inode's xattr (err %d)", err);
97 + if (!IS_NOQUOTA(inode))
98 + extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
100 + if (!ext4_handle_has_enough_credits(handle,
101 + ext4_blocks_for_truncate(inode) + extra_credits)) {
102 + err = ext4_journal_extend(handle,
103 + ext4_blocks_for_truncate(inode) + extra_credits);
105 + err = ext4_journal_restart(handle,
106 + ext4_blocks_for_truncate(inode) + extra_credits);
108 + ext4_warning(inode->i_sb,
109 + "couldn't extend journal (err %d)", err);
115 err = ext4_mark_inode_dirty(handle, inode);
117 @@ -269,10 +296,10 @@ void ext4_evict_inode(struct inode *inod
118 * enough credits left in the handle to remove the inode from
119 * the orphan list and set the dtime field.
121 - if (!ext4_handle_has_enough_credits(handle, 3)) {
122 - err = ext4_journal_extend(handle, 3);
123 + if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
124 + err = ext4_journal_extend(handle, extra_credits);
126 - err = ext4_journal_restart(handle, 3);
127 + err = ext4_journal_restart(handle, extra_credits);
129 ext4_warning(inode->i_sb,
130 "couldn't extend journal (err %d)", err);
131 @@ -307,6 +334,9 @@ void ext4_evict_inode(struct inode *inod
132 ext4_free_inode(handle, inode);
133 ext4_journal_stop(handle);
134 sb_end_intwrite(inode->i_sb);
136 + if (lea_ino_array != NULL)
137 + ext4_xattr_inode_array_free(inode, lea_ino_array);
140 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
141 @@ -4777,7 +4807,7 @@ static int ext4_index_trans_blocks(struc
143 * Also account for superblock, inode, quota and xattr blocks
145 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
146 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
149 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
150 Index: linux-stage/fs/ext4/xattr.c
151 ===================================================================
152 --- linux-stage.orig/fs/ext4/xattr.c
153 +++ linux-stage/fs/ext4/xattr.c
154 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
156 while (!IS_LAST_ENTRY(entry)) {
157 if (entry->e_value_size != 0 &&
158 + entry->e_value_inum == 0 &&
159 (value_start + le16_to_cpu(entry->e_value_offs) <
160 (void *)e + sizeof(__u32) ||
161 value_start + le16_to_cpu(entry->e_value_offs) +
162 @@ -233,19 +233,26 @@ ext4_xattr_check_block(struct inode *ino
166 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
167 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
168 + struct inode *inode)
170 size_t value_size = le32_to_cpu(entry->e_value_size);
172 - if (entry->e_value_block != 0 || value_size > size ||
173 + if (!entry->e_value_inum &&
174 le16_to_cpu(entry->e_value_offs) + value_size > size)
176 + if (entry->e_value_inum &&
177 + (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
178 + le32_to_cpu(entry->e_value_inum) >
179 + le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
185 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
186 - const char *name, size_t size, int sorted)
187 + const char *name, size_t size, int sorted,
188 + struct inode *inode)
190 struct ext4_xattr_entry *entry;
192 @@ -265,11 +272,109 @@ ext4_xattr_find_entry(struct ext4_xattr_
196 - if (!cmp && ext4_xattr_check_entry(entry, size))
197 + if (!cmp && ext4_xattr_check_entry(entry, size, inode))
199 return cmp ? -ENODATA : 0;
203 + * Read the EA value from an inode.
205 +static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
207 + unsigned long block = 0;
208 + struct buffer_head *bh = NULL;
209 + int err, blocksize;
210 + size_t csize, ret_size = 0;
215 + blocksize = ea_inode->i_sb->s_blocksize;
217 + while (ret_size < *size) {
218 + csize = (*size - ret_size) > blocksize ? blocksize :
220 + bh = ext4_bread(NULL, ea_inode, block, 0, &err);
225 + memcpy(buf, bh->b_data, csize);
239 + * Fetch the xattr inode from disk.
241 + * The xattr inode stores the parent inode number and generation so that
242 + * the kernel and e2fsck can verify the xattr inode is valid upon access.
244 +struct inode *ext4_xattr_inode_iget(struct inode *parent,
245 + unsigned long ea_ino, int *err)
247 + struct inode *ea_inode = NULL;
249 + ea_inode = ext4_iget(parent->i_sb, ea_ino);
250 + if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
251 + int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
252 + ext4_error(parent->i_sb, "error while reading EA inode %lu "
253 + "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
254 + *err = rc != 0 ? rc : -EIO;
258 + if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
259 + ea_inode->i_generation != parent->i_generation) {
260 + ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
261 + "to parent invalid.", ea_ino);
266 + if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
267 + ext4_error(parent->i_sb, "EA inode %lu does not have "
268 + "EXT4_EA_INODE_FL flag set.\n", ea_ino);
282 + * Read the value from the EA inode.
284 +static int ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino,
285 + void *buffer, size_t *size)
287 + struct inode *ea_inode = NULL;
290 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
294 + err = ext4_xattr_inode_read(ea_inode, buffer, size);
301 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
302 void *buffer, size_t buffer_size)
303 @@ -301,7 +400,8 @@ bad_block:
305 ext4_xattr_cache_insert(bh);
307 - error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
308 + error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
313 @@ -311,8 +411,16 @@ bad_block:
315 if (size > buffer_size)
317 - memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
319 + if (entry->e_value_inum) {
320 + error = ext4_xattr_inode_get(inode,
321 + le32_to_cpu(entry->e_value_inum),
326 + memcpy(buffer, bh->b_data +
327 + le16_to_cpu(entry->e_value_offs), size);
332 @@ -346,7 +454,7 @@ ext4_xattr_ibody_get(struct inode *inode
335 error = ext4_xattr_find_entry(&entry, name_index, name,
336 - end - (void *)entry, 0);
337 + end - (void *)entry, 0, inode);
340 size = le32_to_cpu(entry->e_value_size);
341 @@ -354,8 +462,16 @@ ext4_xattr_ibody_get(struct inode *inode
343 if (size > buffer_size)
345 - memcpy(buffer, (void *)IFIRST(header) +
346 - le16_to_cpu(entry->e_value_offs), size);
347 + if (entry->e_value_inum) {
348 + error = ext4_xattr_inode_get(inode,
349 + le32_to_cpu(entry->e_value_inum),
354 + memcpy(buffer, (void *)IFIRST(header) +
355 + le16_to_cpu(entry->e_value_offs), size);
360 @@ -597,7 +713,7 @@ static size_t ext4_xattr_free_space(stru
362 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
363 *total += EXT4_XATTR_LEN(last->e_name_len);
364 - if (!last->e_value_block && last->e_value_size) {
365 + if (!last->e_value_inum && last->e_value_size) {
366 size_t offs = le16_to_cpu(last->e_value_offs);
367 if (offs < *min_offs)
369 @@ -606,16 +722,196 @@ static size_t ext4_xattr_free_space(stru
370 return (*min_offs - ((void *)last - base) - sizeof(__u32));
374 + * Write the value of the EA in an inode.
377 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
378 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
379 + const void *buf, int bufsize)
381 + struct buffer_head *bh = NULL;
382 + unsigned long block = 0;
383 + unsigned blocksize = ea_inode->i_sb->s_blocksize;
384 + unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
385 + int csize, wsize = 0;
390 + while (ret >= 0 && ret < max_blocks) {
391 + struct ext4_map_blocks map;
392 + map.m_lblk = block += ret;
393 + map.m_len = max_blocks -= ret;
395 + ret = ext4_map_blocks(handle, ea_inode, &map,
396 + EXT4_GET_BLOCKS_CREATE);
398 + ext4_mark_inode_dirty(handle, ea_inode);
399 + if (ret == -ENOSPC &&
400 + ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
412 + while (wsize < bufsize) {
415 + csize = (bufsize - wsize) > blocksize ? blocksize :
417 + bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
420 + ret = ext4_journal_get_write_access(handle, bh);
424 + memcpy(bh->b_data, buf, csize);
425 + set_buffer_uptodate(bh);
426 + ext4_handle_dirty_metadata(handle, ea_inode, bh);
433 + i_size_write(ea_inode, wsize);
434 + ext4_update_i_disksize(ea_inode, wsize);
436 + ext4_mark_inode_dirty(handle, ea_inode);
444 +static void ext4_xattr_inode_set_ref(struct inode *ea_inode, __u64 ref_count)
446 + ea_inode->i_ctime.tv_sec = (__u32)(ref_count >> 32);
447 + ea_inode->i_version = (__u32)ref_count;
450 +static void ext4_xattr_inode_set_hash(struct inode *ea_inode, __u32 hash)
452 + ea_inode->i_atime.tv_sec = hash;
456 + * Create an inode to store the value of a large EA.
458 +static struct inode *
459 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode, __u32 hash)
461 + struct inode *ea_inode = NULL;
464 + * Let the next inode be the goal, so we try and allocate the EA inode
465 + * in the same group, or nearby one.
467 + ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
468 + S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
470 + if (!IS_ERR(ea_inode)) {
471 + ea_inode->i_op = &ext4_file_inode_operations;
472 + ea_inode->i_fop = &ext4_file_operations;
473 + ext4_set_aops(ea_inode);
474 + ea_inode->i_generation = inode->i_generation;
475 + EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
478 + * A back-pointer from EA inode to parent inode will be useful
481 + EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
482 + unlock_new_inode(ea_inode);
484 + ext4_xattr_inode_set_ref(ea_inode, 1);
485 + ext4_xattr_inode_set_hash(ea_inode, hash);
492 + * Unlink the inode storing the value of the EA.
495 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
497 + struct inode *ea_inode = NULL;
500 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
504 + clear_nlink(ea_inode);
511 +ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
513 + if (ext4_has_metadata_csum(sbi->s_sb))
514 + return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
519 + * Add value of the EA in an inode.
522 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
523 + const void *value, size_t value_len)
525 + struct inode *ea_inode = NULL;
529 + /* Create an inode for the EA value */
530 + hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
531 + ea_inode = ext4_xattr_inode_create(handle, inode, hash);
532 + if (IS_ERR(ea_inode))
535 + err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
537 + clear_nlink(ea_inode);
539 + *ea_ino = ea_inode->i_ino;
547 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
548 + handle_t *handle, struct inode *inode)
550 struct ext4_xattr_entry *last;
551 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
552 + int in_inode = i->in_inode;
554 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
555 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
556 + (EXT4_XATTR_SIZE(i->value_len) >
557 + EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
560 /* Compute min_offs and last. */
562 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
563 - if (!last->e_value_block && last->e_value_size) {
564 + if (!last->e_value_inum && last->e_value_size) {
565 size_t offs = le16_to_cpu(last->e_value_offs);
568 @@ -623,16 +919,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
570 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
572 - if (!s->here->e_value_block && s->here->e_value_size) {
574 + !s->here->e_value_inum && s->here->e_value_size) {
575 size_t size = le32_to_cpu(s->here->e_value_size);
576 free += EXT4_XATTR_SIZE(size);
578 free += EXT4_XATTR_LEN(name_len);
581 - if (free < EXT4_XATTR_SIZE(i->value_len) ||
582 - free < EXT4_XATTR_LEN(name_len) +
583 - EXT4_XATTR_SIZE(i->value_len))
584 + size_t value_len = EXT4_XATTR_SIZE(i->value_len);
589 + if (free < value_len ||
590 + free < EXT4_XATTR_LEN(name_len) + value_len)
594 @@ -646,7 +947,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
595 s->here->e_name_len = name_len;
596 memcpy(s->here->e_name, i->name, name_len);
598 - if (!s->here->e_value_block && s->here->e_value_size) {
599 + if (!s->here->e_value_inum && s->here->e_value_size &&
600 + s->here->e_value_offs > 0) {
601 void *first_val = s->base + min_offs;
602 size_t offs = le16_to_cpu(s->here->e_value_offs);
603 void *val = s->base + offs;
604 @@ -680,13 +982,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
606 while (!IS_LAST_ENTRY(last)) {
607 size_t o = le16_to_cpu(last->e_value_offs);
608 - if (!last->e_value_block &&
609 + if (!last->e_value_inum &&
610 last->e_value_size && o < offs)
612 cpu_to_le16(o + size);
613 last = EXT4_XATTR_NEXT(last);
616 + if (s->here->e_value_inum) {
617 + ext4_xattr_inode_unlink(inode,
618 + le32_to_cpu(s->here->e_value_inum));
619 + s->here->e_value_inum = 0;
622 /* Remove the old name. */
623 size_t size = EXT4_XATTR_LEN(name_len);
624 @@ -700,10 +1006,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
626 /* Insert the new value. */
627 s->here->e_value_size = cpu_to_le32(i->value_len);
628 - if (i->value_len) {
630 + unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
631 + ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
633 + s->here->e_value_inum = cpu_to_le32(ea_ino);
634 + s->here->e_value_offs = 0;
635 + } else if (i->value_len) {
636 size_t size = EXT4_XATTR_SIZE(i->value_len);
637 void *val = s->base + min_offs - size;
638 s->here->e_value_offs = cpu_to_le16(min_offs - size);
639 + s->here->e_value_inum = 0;
640 if (i->value == EXT4_ZERO_XATTR_VALUE) {
641 memset(val, 0, size);
643 @@ -753,7 +1066,7 @@ ext4_xattr_block_find(struct inode *inod
644 bs->s.end = bs->bh->b_data + bs->bh->b_size;
645 bs->s.here = bs->s.first;
646 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
647 - i->name, bs->bh->b_size, 1);
648 + i->name, bs->bh->b_size, 1, inode);
649 if (error && error != -ENODATA)
651 bs->s.not_found = error;
652 @@ -777,8 +1090,6 @@ ext4_xattr_block_set(handle_t *handle, s
654 #define header(x) ((struct ext4_xattr_header *)(x))
656 - if (i->value && i->value_len > sb->s_blocksize)
659 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
661 @@ -794,7 +1105,7 @@ ext4_xattr_block_set(handle_t *handle, s
664 ea_bdebug(bs->bh, "modifying in-place");
665 - error = ext4_xattr_set_entry(i, s);
666 + error = ext4_xattr_set_entry(i, s, handle, inode);
668 if (!IS_LAST_ENTRY(s->first))
669 ext4_xattr_rehash(header(s->base),
670 @@ -845,7 +1156,7 @@ ext4_xattr_block_set(handle_t *handle, s
671 s->end = s->base + sb->s_blocksize;
674 - error = ext4_xattr_set_entry(i, s);
675 + error = ext4_xattr_set_entry(i, s, handle, inode);
679 @@ -994,7 +1305,7 @@ int ext4_xattr_ibody_find(struct inode *
680 /* Find the named attribute. */
681 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
683 - (void *)is->s.base, 0);
684 + (void *)is->s.base, 0, inode);
685 if (error && error != -ENODATA)
687 is->s.not_found = error;
688 @@ -1012,7 +1323,7 @@ int ext4_xattr_ibody_inline_set(handle_t
690 if (EXT4_I(inode)->i_extra_isize == 0)
692 - error = ext4_xattr_set_entry(i, s);
693 + error = ext4_xattr_set_entry(i, s, handle, inode);
695 if (error == -ENOSPC &&
696 ext4_has_inline_data(inode)) {
697 @@ -1024,7 +1335,7 @@ int ext4_xattr_ibody_inline_set(handle_t
698 error = ext4_xattr_ibody_find(inode, i, is);
701 - error = ext4_xattr_set_entry(i, s);
702 + error = ext4_xattr_set_entry(i, s, handle, inode);
706 @@ -1050,7 +1361,7 @@ static int ext4_xattr_ibody_set(handle_t
708 if (EXT4_I(inode)->i_extra_isize == 0)
710 - error = ext4_xattr_set_entry(i, s);
711 + error = ext4_xattr_set_entry(i, s, handle, inode);
714 header = IHDR(inode, ext4_raw_inode(&is->iloc));
715 @@ -1086,7 +1397,7 @@ ext4_xattr_set_handle(handle_t *handle,
718 .value_len = value_len,
722 struct ext4_xattr_ibody_find is = {
723 .s = { .not_found = -ENODATA, },
724 @@ -1151,6 +1462,15 @@ ext4_xattr_set_handle(handle_t *handle,
727 error = ext4_xattr_block_set(handle, inode, &i, &bs);
728 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
729 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
730 + error == -ENOSPC) {
731 + /* xattr not fit to block, store at external
734 + error = ext4_xattr_ibody_set(handle, inode,
739 if (!is.s.not_found) {
740 @@ -1197,9 +1517,22 @@ ext4_xattr_set(struct inode *inode, int
741 const void *value, size_t value_len, int flags)
744 + struct super_block *sb = inode->i_sb;
745 int error, retries = 0;
746 int credits = ext4_jbd2_credits_xattr(inode);
748 + if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
749 + EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
750 + int nrblocks = (value_len + sb->s_blocksize - 1) >>
751 + sb->s_blocksize_bits;
753 + /* For new inode */
754 + credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
756 + /* For data blocks of EA inode */
757 + credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
761 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
762 if (IS_ERR(handle)) {
763 @@ -1211,7 +1544,7 @@ retry:
764 value, value_len, flags);
765 error2 = ext4_journal_stop(handle);
766 if (error == -ENOSPC &&
767 - ext4_should_retry_alloc(inode->i_sb, &retries))
768 + ext4_should_retry_alloc(sb, &retries))
772 @@ -1233,7 +1566,7 @@ static void ext4_xattr_shift_entries(str
774 /* Adjust the value offsets of the entries */
775 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
776 - if (!last->e_value_block && last->e_value_size) {
777 + if (!last->e_value_inum && last->e_value_size) {
778 new_offs = le16_to_cpu(last->e_value_offs) +
780 BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
781 @@ -1472,21 +1805,135 @@ cleanup:
785 +#define EIA_INCR 16 /* must be 2^n */
786 +#define EIA_MASK (EIA_INCR - 1)
787 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
788 + * If @lea_ino_array is new or full it will be grown and the old
789 + * contents copied over.
792 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
794 + if (*lea_ino_array == NULL) {
796 + * Start with 15 inodes, so it fits into a power-of-two size.
797 + * If *lea_ino_array is NULL, this is essentially offsetof()
800 + kmalloc(offsetof(struct ext4_xattr_ino_array,
801 + xia_inodes[EIA_MASK]),
803 + if (*lea_ino_array == NULL)
805 + (*lea_ino_array)->xia_count = 0;
806 + } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
807 + /* expand the array once all 15 + n * 16 slots are full */
808 + struct ext4_xattr_ino_array *new_array = NULL;
809 + int count = (*lea_ino_array)->xia_count;
811 + /* if new_array is NULL, this is essentially offsetof() */
812 + new_array = kmalloc(
813 + offsetof(struct ext4_xattr_ino_array,
814 + xia_inodes[count + EIA_INCR]),
816 + if (new_array == NULL)
818 + memcpy(new_array, *lea_ino_array,
819 + offsetof(struct ext4_xattr_ino_array,
820 + xia_inodes[count]));
821 + kfree(*lea_ino_array);
822 + *lea_ino_array = new_array;
824 + (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
829 + * Add xattr inode to orphan list
832 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
833 + int credits, struct ext4_xattr_ino_array *lea_ino_array)
835 + struct inode *ea_inode = NULL;
836 + int idx = 0, error = 0;
838 + if (lea_ino_array == NULL)
841 + for (; idx < lea_ino_array->xia_count; ++idx) {
842 + if (!ext4_handle_has_enough_credits(handle, credits)) {
843 + error = ext4_journal_extend(handle, credits);
845 + error = ext4_journal_restart(handle, credits);
848 + ext4_warning(inode->i_sb,
849 + "couldn't extend journal "
850 + "(err %d)", error);
854 + ea_inode = ext4_xattr_inode_iget(inode,
855 + lea_ino_array->xia_inodes[idx], &error);
858 + ext4_orphan_add(handle, ea_inode);
859 + /* the inode's i_count will be released by caller */
866 * ext4_xattr_delete_inode()
868 - * Free extended attribute resources associated with this inode. This
869 + * Free extended attribute resources associated with this inode. Traverse
870 + * all entries and unlink any xattr inodes associated with this inode. This
871 * is called immediately before an inode is freed. We have exclusive
872 - * access to the inode.
873 + * access to the inode. If an orphan inode is deleted it will also delete any
874 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
875 + * to ensure they belong to the parent inode and were not deleted already.
878 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
880 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
881 + struct ext4_xattr_ino_array **lea_ino_array)
883 struct buffer_head *bh = NULL;
884 + struct ext4_xattr_ibody_header *header;
885 + struct ext4_inode *raw_inode;
886 + struct ext4_iloc iloc;
887 + struct ext4_xattr_entry *entry;
888 + int credits = 3, error = 0;
890 - if (!EXT4_I(inode)->i_file_acl)
891 + if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
892 + goto delete_external_ea;
894 + error = ext4_get_inode_loc(inode, &iloc);
897 + raw_inode = ext4_raw_inode(&iloc);
898 + header = IHDR(inode, raw_inode);
899 + for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
900 + entry = EXT4_XATTR_NEXT(entry)) {
901 + if (!entry->e_value_inum)
903 + if (ext4_expand_ino_array(lea_ino_array,
904 + entry->e_value_inum) != 0) {
908 + entry->e_value_inum = 0;
913 + if (!EXT4_I(inode)->i_file_acl) {
914 + /* add xattr inode to orphan list */
915 + ext4_xattr_inode_orphan_add(handle, inode, credits,
919 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
921 EXT4_ERROR_INODE(inode, "block %llu read error",
922 @@ -1499,11 +1946,69 @@ ext4_xattr_delete_inode(handle_t *handle
923 EXT4_I(inode)->i_file_acl);
927 + for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
928 + entry = EXT4_XATTR_NEXT(entry)) {
929 + if (!entry->e_value_inum)
931 + if (ext4_expand_ino_array(lea_ino_array,
932 + entry->e_value_inum) != 0)
934 + entry->e_value_inum = 0;
937 + /* add xattr inode to orphan list */
938 + error = ext4_xattr_inode_orphan_add(handle, inode, credits,
943 + if (!IS_NOQUOTA(inode))
944 + credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
946 + if (!ext4_handle_has_enough_credits(handle, credits)) {
947 + error = ext4_journal_extend(handle, credits);
949 + error = ext4_journal_restart(handle, credits);
951 + ext4_warning(inode->i_sb,
952 + "couldn't extend journal (err %d)", error);
957 ext4_xattr_release_block(handle, inode, bh);
958 EXT4_I(inode)->i_file_acl = 0;
967 +ext4_xattr_inode_array_free(struct inode *inode,
968 + struct ext4_xattr_ino_array *lea_ino_array)
970 + struct inode *ea_inode = NULL;
974 + if (lea_ino_array == NULL)
977 + for (; idx < lea_ino_array->xia_count; ++idx) {
978 + ea_inode = ext4_xattr_inode_iget(inode,
979 + lea_ino_array->xia_inodes[idx], &err);
982 + /* for inode's i_count get from ext4_xattr_delete_inode */
983 + if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
985 + clear_nlink(ea_inode);
988 + kfree(lea_ino_array);
992 @@ -1573,10 +2078,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
993 entry1->e_name_index != entry2->e_name_index ||
994 entry1->e_name_len != entry2->e_name_len ||
995 entry1->e_value_size != entry2->e_value_size ||
996 + entry1->e_value_inum != entry2->e_value_inum ||
997 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
999 - if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1001 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1002 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1003 le32_to_cpu(entry1->e_value_size)))
1004 @@ -1660,7 +2164,7 @@ static inline void ext4_xattr_hash_entry
1008 - if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1009 + if (!entry->e_value_inum && entry->e_value_size) {
1010 __le32 *value = (__le32 *)((char *)header +
1011 le16_to_cpu(entry->e_value_offs));
1012 for (n = (le32_to_cpu(entry->e_value_size) +
1013 Index: linux-stage/fs/ext4/xattr.h
1014 ===================================================================
1015 --- linux-stage.orig/fs/ext4/xattr.h
1016 +++ linux-stage/fs/ext4/xattr.h
1017 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
1018 __u8 e_name_len; /* length of name */
1019 __u8 e_name_index; /* attribute name index */
1020 __le16 e_value_offs; /* offset in disk block of value */
1021 - __le32 e_value_block; /* disk block attribute is stored on (n/i) */
1022 + __le32 e_value_inum; /* inode in which the value is stored */
1023 __le32 e_value_size; /* size of attribute value */
1024 __le32 e_hash; /* hash value of name and value */
1025 char e_name[0]; /* attribute name */
1026 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
1027 EXT4_I(inode)->i_extra_isize))
1028 #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1031 + * Link EA inode back to parent one using i_mtime field.
1032 + * Extra integer type conversion added to ignore higher
1033 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1035 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum) \
1037 + (inode)->i_mtime.tv_sec = inum; \
1040 +#define EXT4_XATTR_INODE_GET_PARENT(inode) \
1041 + ((__u32)(inode)->i_mtime.tv_sec)
1044 + * The minimum size of EA value when you start storing it in an external inode
1045 + * size of block - size of header - size of 1 entry - 4 null bytes
1047 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b) \
1048 + ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1050 #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1051 #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1052 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1053 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1054 #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1056 struct ext4_xattr_info {
1065 struct ext4_xattr_search {
1066 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1067 extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1068 extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1070 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1071 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1073 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1074 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1075 + struct ext4_xattr_ino_array **array);
1076 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1077 + struct ext4_xattr_ino_array *array);
1078 extern void ext4_xattr_put_super(struct super_block *);
1080 extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1081 Index: linux-stage/fs/ext4/ialloc.c
1082 ===================================================================
1083 --- linux-stage.orig/fs/ext4/ialloc.c
1084 +++ linux-stage/fs/ext4/ialloc.c
1085 @@ -252,7 +252,6 @@ void ext4_free_inode(handle_t *handle, s
1086 * as writing the quota to disk may need the lock as well.
1088 dquot_initialize(inode);
1089 - ext4_xattr_delete_inode(handle, inode);
1090 dquot_free_inode(inode);
1093 Index: linux-stage/fs/ext4/inline.c
1094 ===================================================================
1095 --- linux-stage.orig/fs/ext4/inline.c
1096 +++ linux-stage/fs/ext4/inline.c
1097 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1099 /* Compute min_offs. */
1100 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1101 - if (!entry->e_value_block && entry->e_value_size) {
1102 + if (!entry->e_value_inum && entry->e_value_size) {
1103 size_t offs = le16_to_cpu(entry->e_value_offs);
1104 if (offs < min_offs)