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 @@ -248,9 +248,36 @@ void ext4_evict_inode(struct inode *inod
85 sb_end_intwrite(inode->i_sb);
90 ext4_handle_sync(handle);
93 + * Delete xattr inode before deleting the main inode.
95 + err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
97 + ext4_warning(inode->i_sb,
98 + "couldn't delete inode's xattr (err %d)", err);
102 + if (!IS_NOQUOTA(inode))
103 + extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
105 + if (!ext4_handle_has_enough_credits(handle,
106 + ext4_blocks_for_truncate(inode) + extra_credits)) {
107 + err = ext4_journal_extend(handle,
108 + ext4_blocks_for_truncate(inode) + extra_credits);
110 + err = ext4_journal_restart(handle,
111 + ext4_blocks_for_truncate(inode) + extra_credits);
113 + ext4_warning(inode->i_sb,
114 + "couldn't extend journal (err %d)", err);
120 err = ext4_mark_inode_dirty(handle, inode);
122 @@ -305,8 +332,12 @@ void ext4_evict_inode(struct inode *inod
123 ext4_clear_inode(inode);
125 ext4_free_inode(handle, inode);
127 ext4_journal_stop(handle);
128 sb_end_intwrite(inode->i_sb);
130 + if (lea_ino_array != NULL)
131 + ext4_xattr_inode_array_free(inode, lea_ino_array);
134 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
135 @@ -4777,7 +4808,7 @@ static int ext4_index_trans_blocks(struc
137 * Also account for superblock, inode, quota and xattr blocks
139 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
140 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
143 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
144 Index: linux-stage/fs/ext4/xattr.c
145 ===================================================================
146 --- linux-stage.orig/fs/ext4/xattr.c
147 +++ linux-stage/fs/ext4/xattr.c
148 @@ -233,19 +233,26 @@ ext4_xattr_check_block(struct inode *ino
152 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
153 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
154 + struct inode *inode)
156 size_t value_size = le32_to_cpu(entry->e_value_size);
158 - if (entry->e_value_block != 0 || value_size > size ||
159 - le16_to_cpu(entry->e_value_offs) + value_size > size)
160 + if ((entry->e_value_inum == 0) &&
161 + (le16_to_cpu(entry->e_value_offs) + value_size > size))
163 + if (entry->e_value_inum != 0 &&
164 + (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
165 + le32_to_cpu(entry->e_value_inum) >
166 + le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
172 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
173 - const char *name, size_t size, int sorted)
174 + const char *name, size_t size, int sorted,
175 + struct inode *inode)
177 struct ext4_xattr_entry *entry;
179 @@ -265,11 +272,104 @@ ext4_xattr_find_entry(struct ext4_xattr_
183 - if (!cmp && ext4_xattr_check_entry(entry, size))
184 + if (!cmp && ext4_xattr_check_entry(entry, size, inode))
186 return cmp ? -ENODATA : 0;
190 + * Read the EA value from an inode.
193 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
195 + unsigned long block = 0;
196 + struct buffer_head *bh = NULL;
197 + int err, blocksize;
198 + size_t csize, ret_size = 0;
203 + blocksize = ea_inode->i_sb->s_blocksize;
205 + while (ret_size < *size) {
206 + csize = (*size - ret_size) > blocksize ? blocksize :
208 + bh = ext4_bread(NULL, ea_inode, block, 0, &err);
213 + memcpy(buf, bh->b_data, csize);
226 +struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
228 + struct inode *ea_inode = NULL;
230 + ea_inode = ext4_iget(parent->i_sb, ea_ino);
231 + if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
232 + int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
233 + ext4_error(parent->i_sb, "error while reading EA inode %lu "
234 + "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
235 + *err = rc != 0 ? rc : -EIO;
239 + if (ea_inode->i_xattr_inode_parent != parent->i_ino ||
240 + ea_inode->i_generation != parent->i_generation) {
241 + ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
242 + "to parent invalid.", ea_ino);
247 + if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
248 + ext4_error(parent->i_sb, "EA inode %lu does not have "
249 + "EXT4_EA_INODE_FL flag set.\n", ea_ino);
263 + * Read the value from the EA inode.
266 +ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
269 + struct inode *ea_inode = NULL;
272 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
276 + err = ext4_xattr_inode_read(ea_inode, buffer, size);
283 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
284 void *buffer, size_t buffer_size)
285 @@ -301,7 +400,8 @@ bad_block:
287 ext4_xattr_cache_insert(bh);
289 - error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
290 + error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
295 @@ -311,8 +411,16 @@ bad_block:
297 if (size > buffer_size)
299 - memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
301 + if (entry->e_value_inum != 0) {
302 + error = ext4_xattr_inode_get(inode,
303 + le32_to_cpu(entry->e_value_inum),
308 + memcpy(buffer, bh->b_data +
309 + le16_to_cpu(entry->e_value_offs), size);
314 @@ -346,7 +454,7 @@ ext4_xattr_ibody_get(struct inode *inode
317 error = ext4_xattr_find_entry(&entry, name_index, name,
318 - end - (void *)entry, 0);
319 + end - (void *)entry, 0, inode);
322 size = le32_to_cpu(entry->e_value_size);
323 @@ -354,8 +462,16 @@ ext4_xattr_ibody_get(struct inode *inode
325 if (size > buffer_size)
327 - memcpy(buffer, (void *)IFIRST(header) +
328 - le16_to_cpu(entry->e_value_offs), size);
329 + if (entry->e_value_inum != 0) {
330 + error = ext4_xattr_inode_get(inode,
331 + le32_to_cpu(entry->e_value_inum),
336 + memcpy(buffer, (void *)IFIRST(header) +
337 + le16_to_cpu(entry->e_value_offs), size);
342 @@ -597,7 +713,7 @@ static size_t ext4_xattr_free_space(stru
344 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
345 *total += EXT4_XATTR_LEN(last->e_name_len);
346 - if (!last->e_value_block && last->e_value_size) {
347 + if (last->e_value_inum == 0 && last->e_value_size > 0) {
348 size_t offs = le16_to_cpu(last->e_value_offs);
349 if (offs < *min_offs)
351 @@ -606,16 +722,172 @@ static size_t ext4_xattr_free_space(stru
352 return (*min_offs - ((void *)last - base) - sizeof(__u32));
356 + * Write the value of the EA in an inode.
359 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
360 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
361 + const void *buf, int bufsize)
363 + struct buffer_head *bh = NULL;
364 + unsigned long block = 0;
365 + unsigned blocksize = ea_inode->i_sb->s_blocksize;
366 + unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
367 + int csize, wsize = 0;
372 + while (ret >= 0 && ret < max_blocks) {
373 + struct ext4_map_blocks map;
374 + map.m_lblk = block += ret;
375 + map.m_len = max_blocks -= ret;
377 + ret = ext4_map_blocks(handle, ea_inode, &map,
378 + EXT4_GET_BLOCKS_CREATE);
380 + ext4_mark_inode_dirty(handle, ea_inode);
381 + if (ret == -ENOSPC &&
382 + ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
394 + while (wsize < bufsize) {
397 + csize = (bufsize - wsize) > blocksize ? blocksize :
399 + bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
402 + ret = ext4_journal_get_write_access(handle, bh);
406 + memcpy(bh->b_data, buf, csize);
407 + set_buffer_uptodate(bh);
408 + ext4_handle_dirty_metadata(handle, ea_inode, bh);
415 + i_size_write(ea_inode, wsize);
416 + ext4_update_i_disksize(ea_inode, wsize);
418 + ext4_mark_inode_dirty(handle, ea_inode);
427 + * Create an inode to store the value of a large EA.
429 +static struct inode *
430 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
432 + struct inode *ea_inode = NULL;
435 + * Let the next inode be the goal, so we try and allocate the EA inode
436 + * in the same group, or nearby one.
438 + ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
439 + S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
441 + if (!IS_ERR(ea_inode)) {
442 + ea_inode->i_op = &ext4_file_inode_operations;
443 + ea_inode->i_fop = &ext4_file_operations;
444 + ext4_set_aops(ea_inode);
445 + ea_inode->i_generation = inode->i_generation;
446 + EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
449 + * A back-pointer from EA inode to parent inode will be useful
452 + ea_inode->i_xattr_inode_parent = inode->i_ino;
453 + unlock_new_inode(ea_inode);
460 + * Unlink the inode storing the value of the EA.
463 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
465 + struct inode *ea_inode = NULL;
468 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
472 + clear_nlink(ea_inode);
479 + * Add value of the EA in an inode.
482 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
483 + const void *value, size_t value_len)
485 + struct inode *ea_inode = NULL;
488 + /* Create an inode for the EA value */
489 + ea_inode = ext4_xattr_inode_create(handle, inode);
490 + if (IS_ERR(ea_inode))
493 + err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
495 + clear_nlink(ea_inode);
497 + *ea_ino = ea_inode->i_ino;
505 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
506 + handle_t *handle, struct inode *inode)
508 struct ext4_xattr_entry *last;
509 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
510 + int in_inode = i->in_inode;
512 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
513 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
514 + (EXT4_XATTR_SIZE(i->value_len) >
515 + EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
518 /* Compute min_offs and last. */
520 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
521 - if (!last->e_value_block && last->e_value_size) {
522 + if (last->e_value_inum == 0 && last->e_value_size > 0) {
523 size_t offs = le16_to_cpu(last->e_value_offs);
526 @@ -623,16 +895,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
528 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
530 - if (!s->here->e_value_block && s->here->e_value_size) {
531 + if (!in_inode && s->here->e_value_inum == 0 &&
532 + s->here->e_value_size > 0) {
533 size_t size = le32_to_cpu(s->here->e_value_size);
534 free += EXT4_XATTR_SIZE(size);
536 free += EXT4_XATTR_LEN(name_len);
539 - if (free < EXT4_XATTR_SIZE(i->value_len) ||
540 - free < EXT4_XATTR_LEN(name_len) +
541 - EXT4_XATTR_SIZE(i->value_len))
542 + size_t value_len = EXT4_XATTR_SIZE(i->value_len);
547 + if (free < value_len ||
548 + free < EXT4_XATTR_LEN(name_len) + value_len)
552 @@ -646,7 +923,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
553 s->here->e_name_len = name_len;
554 memcpy(s->here->e_name, i->name, name_len);
556 - if (!s->here->e_value_block && s->here->e_value_size) {
557 + if (s->here->e_value_offs > 0 && s->here->e_value_inum == 0 &&
558 + s->here->e_value_size > 0) {
559 void *first_val = s->base + min_offs;
560 size_t offs = le16_to_cpu(s->here->e_value_offs);
561 void *val = s->base + offs;
562 @@ -680,13 +958,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
564 while (!IS_LAST_ENTRY(last)) {
565 size_t o = le16_to_cpu(last->e_value_offs);
566 - if (!last->e_value_block &&
567 + if (!last->e_value_inum &&
568 last->e_value_size && o < offs)
570 cpu_to_le16(o + size);
571 last = EXT4_XATTR_NEXT(last);
574 + if (s->here->e_value_inum != 0) {
575 + ext4_xattr_inode_unlink(inode,
576 + le32_to_cpu(s->here->e_value_inum));
577 + s->here->e_value_inum = 0;
580 /* Remove the old name. */
581 size_t size = EXT4_XATTR_LEN(name_len);
582 @@ -700,10 +982,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
584 /* Insert the new value. */
585 s->here->e_value_size = cpu_to_le32(i->value_len);
586 - if (i->value_len) {
588 + unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
589 + ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
591 + s->here->e_value_inum = cpu_to_le32(ea_ino);
592 + s->here->e_value_offs = 0;
593 + } else if (i->value_len) {
594 size_t size = EXT4_XATTR_SIZE(i->value_len);
595 void *val = s->base + min_offs - size;
596 s->here->e_value_offs = cpu_to_le16(min_offs - size);
597 + s->here->e_value_inum = 0;
598 if (i->value == EXT4_ZERO_XATTR_VALUE) {
599 memset(val, 0, size);
601 @@ -753,7 +1042,7 @@ ext4_xattr_block_find(struct inode *inod
602 bs->s.end = bs->bh->b_data + bs->bh->b_size;
603 bs->s.here = bs->s.first;
604 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
605 - i->name, bs->bh->b_size, 1);
606 + i->name, bs->bh->b_size, 1, inode);
607 if (error && error != -ENODATA)
609 bs->s.not_found = error;
610 @@ -777,8 +1066,6 @@ ext4_xattr_block_set(handle_t *handle, s
612 #define header(x) ((struct ext4_xattr_header *)(x))
614 - if (i->value && i->value_len > sb->s_blocksize)
617 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
619 @@ -794,7 +1081,7 @@ ext4_xattr_block_set(handle_t *handle, s
622 ea_bdebug(bs->bh, "modifying in-place");
623 - error = ext4_xattr_set_entry(i, s);
624 + error = ext4_xattr_set_entry(i, s, handle, inode);
626 if (!IS_LAST_ENTRY(s->first))
627 ext4_xattr_rehash(header(s->base),
628 @@ -845,7 +1132,7 @@ ext4_xattr_block_set(handle_t *handle, s
629 s->end = s->base + sb->s_blocksize;
632 - error = ext4_xattr_set_entry(i, s);
633 + error = ext4_xattr_set_entry(i, s, handle, inode);
637 @@ -994,7 +1281,7 @@ int ext4_xattr_ibody_find(struct inode *
638 /* Find the named attribute. */
639 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
641 - (void *)is->s.base, 0);
642 + (void *)is->s.base, 0, inode);
643 if (error && error != -ENODATA)
645 is->s.not_found = error;
646 @@ -1012,7 +1299,7 @@ int ext4_xattr_ibody_inline_set(handle_t
648 if (EXT4_I(inode)->i_extra_isize == 0)
650 - error = ext4_xattr_set_entry(i, s);
651 + error = ext4_xattr_set_entry(i, s, handle, inode);
653 if (error == -ENOSPC &&
654 ext4_has_inline_data(inode)) {
655 @@ -1024,7 +1311,7 @@ int ext4_xattr_ibody_inline_set(handle_t
656 error = ext4_xattr_ibody_find(inode, i, is);
659 - error = ext4_xattr_set_entry(i, s);
660 + error = ext4_xattr_set_entry(i, s, handle, inode);
664 @@ -1050,7 +1337,7 @@ static int ext4_xattr_ibody_set(handle_t
666 if (EXT4_I(inode)->i_extra_isize == 0)
668 - error = ext4_xattr_set_entry(i, s);
669 + error = ext4_xattr_set_entry(i, s, handle, inode);
672 header = IHDR(inode, ext4_raw_inode(&is->iloc));
673 @@ -1086,7 +1373,7 @@ ext4_xattr_set_handle(handle_t *handle,
676 .value_len = value_len,
680 struct ext4_xattr_ibody_find is = {
681 .s = { .not_found = -ENODATA, },
682 @@ -1151,6 +1438,15 @@ ext4_xattr_set_handle(handle_t *handle,
685 error = ext4_xattr_block_set(handle, inode, &i, &bs);
686 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
687 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
688 + error == -ENOSPC) {
689 + /* xattr not fit to block, store at external
692 + error = ext4_xattr_ibody_set(handle, inode,
697 if (!is.s.not_found) {
698 @@ -1197,9 +1493,22 @@ ext4_xattr_set(struct inode *inode, int
699 const void *value, size_t value_len, int flags)
702 + struct super_block *sb = inode->i_sb;
703 int error, retries = 0;
704 int credits = ext4_jbd2_credits_xattr(inode);
706 + if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
707 + EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
708 + int nrblocks = (value_len + sb->s_blocksize - 1) >>
709 + sb->s_blocksize_bits;
711 + /* For new inode */
712 + credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
714 + /* For data blocks of EA inode */
715 + credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
719 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
720 if (IS_ERR(handle)) {
721 @@ -1211,7 +1520,7 @@ retry:
722 value, value_len, flags);
723 error2 = ext4_journal_stop(handle);
724 if (error == -ENOSPC &&
725 - ext4_should_retry_alloc(inode->i_sb, &retries))
726 + ext4_should_retry_alloc(sb, &retries))
730 @@ -1233,7 +1542,7 @@ static void ext4_xattr_shift_entries(str
732 /* Adjust the value offsets of the entries */
733 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
734 - if (!last->e_value_block && last->e_value_size) {
735 + if (last->e_value_inum == 0 && last->e_value_size > 0) {
736 new_offs = le16_to_cpu(last->e_value_offs) +
738 BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
739 @@ -1472,21 +1781,135 @@ cleanup:
743 +#define EIA_INCR 16 /* must be 2^n */
744 +#define EIA_MASK (EIA_INCR - 1)
745 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
746 + * If @lea_ino_array is new or full it will be grown and the old
747 + * contents copied over.
750 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
752 + if (*lea_ino_array == NULL) {
754 + * Start with 15 inodes, so it fits into a power-of-two size.
755 + * If *lea_ino_array is NULL, this is essentially offsetof()
758 + kmalloc(offsetof(struct ext4_xattr_ino_array,
759 + xia_inodes[EIA_MASK]),
761 + if (*lea_ino_array == NULL)
763 + (*lea_ino_array)->xia_count = 0;
764 + } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
765 + /* expand the array once all 15 + n * 16 slots are full */
766 + struct ext4_xattr_ino_array *new_array = NULL;
767 + int count = (*lea_ino_array)->xia_count;
769 + /* if new_array is NULL, this is essentially offsetof() */
770 + new_array = kmalloc(
771 + offsetof(struct ext4_xattr_ino_array,
772 + xia_inodes[count + EIA_INCR]),
774 + if (new_array == NULL)
776 + memcpy(new_array, *lea_ino_array,
777 + offsetof(struct ext4_xattr_ino_array,
778 + xia_inodes[count]));
779 + kfree(*lea_ino_array);
780 + *lea_ino_array = new_array;
782 + (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
787 + * Add xattr inode to orphan list
790 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
791 + int credits, struct ext4_xattr_ino_array *lea_ino_array)
793 + struct inode *ea_inode = NULL;
794 + int idx = 0, error = 0;
796 + if (lea_ino_array == NULL)
799 + for (; idx < lea_ino_array->xia_count; ++idx) {
800 + if (!ext4_handle_has_enough_credits(handle, credits)) {
801 + error = ext4_journal_extend(handle, credits);
803 + error = ext4_journal_restart(handle, credits);
806 + ext4_warning(inode->i_sb,
807 + "couldn't extend journal "
808 + "(err %d)", error);
812 + ea_inode = ext4_xattr_inode_iget(inode,
813 + lea_ino_array->xia_inodes[idx], &error);
816 + ext4_orphan_add(handle, ea_inode);
817 + /* the inode's i_count will be released by caller */
824 * ext4_xattr_delete_inode()
826 - * Free extended attribute resources associated with this inode. This
827 + * Free extended attribute resources associated with this inode. Traverse
828 + * all entries and unlink any xattr inodes associated with this inode. This
829 * is called immediately before an inode is freed. We have exclusive
830 - * access to the inode.
831 + * access to the inode. If an orphan inode is deleted it will also delete any
832 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
833 + * to ensure they belong to the parent inode and were not deleted already.
836 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
838 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
839 + struct ext4_xattr_ino_array **lea_ino_array)
841 struct buffer_head *bh = NULL;
842 + struct ext4_xattr_ibody_header *header;
843 + struct ext4_inode *raw_inode;
844 + struct ext4_iloc iloc;
845 + struct ext4_xattr_entry *entry;
846 + int credits = 3, error = 0;
848 - if (!EXT4_I(inode)->i_file_acl)
849 + if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
850 + goto delete_external_ea;
852 + error = ext4_get_inode_loc(inode, &iloc);
855 + raw_inode = ext4_raw_inode(&iloc);
856 + header = IHDR(inode, raw_inode);
857 + entry = IFIRST(header);
858 + for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
859 + if (entry->e_value_inum == 0)
861 + if (ext4_expand_ino_array(lea_ino_array,
862 + entry->e_value_inum) != 0) {
866 + entry->e_value_inum = 0;
871 + if (!EXT4_I(inode)->i_file_acl) {
872 + /* add xattr inode to orphan list */
873 + ext4_xattr_inode_orphan_add(handle, inode, credits,
877 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
879 EXT4_ERROR_INODE(inode, "block %llu read error",
880 @@ -1499,11 +1922,69 @@ ext4_xattr_delete_inode(handle_t *handle
881 EXT4_I(inode)->i_file_acl);
885 + entry = BFIRST(bh);
886 + for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
887 + if (entry->e_value_inum == 0)
889 + if (ext4_expand_ino_array(lea_ino_array,
890 + entry->e_value_inum) != 0)
892 + entry->e_value_inum = 0;
895 + /* add xattr inode to orphan list */
896 + error = ext4_xattr_inode_orphan_add(handle, inode, credits,
901 + if (!IS_NOQUOTA(inode))
902 + credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
904 + if (!ext4_handle_has_enough_credits(handle, credits)) {
905 + error = ext4_journal_extend(handle, credits);
907 + error = ext4_journal_restart(handle, credits);
909 + ext4_warning(inode->i_sb,
910 + "couldn't extend journal (err %d)", error);
915 ext4_xattr_release_block(handle, inode, bh);
916 EXT4_I(inode)->i_file_acl = 0;
925 +ext4_xattr_inode_array_free(struct inode *inode,
926 + struct ext4_xattr_ino_array *lea_ino_array)
928 + struct inode *ea_inode = NULL;
932 + if (lea_ino_array == NULL)
935 + for (; idx < lea_ino_array->xia_count; ++idx) {
936 + ea_inode = ext4_xattr_inode_iget(inode,
937 + lea_ino_array->xia_inodes[idx], &err);
940 + /* for inode's i_count get from ext4_xattr_delete_inode */
941 + if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
943 + clear_nlink(ea_inode);
946 + kfree(lea_ino_array);
950 @@ -1573,10 +2054,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
951 entry1->e_name_index != entry2->e_name_index ||
952 entry1->e_name_len != entry2->e_name_len ||
953 entry1->e_value_size != entry2->e_value_size ||
954 + entry1->e_value_inum != entry2->e_value_inum ||
955 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
957 - if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
959 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
960 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
961 le32_to_cpu(entry1->e_value_size)))
962 @@ -1660,7 +2140,7 @@ static inline void ext4_xattr_hash_entry
966 - if (entry->e_value_block == 0 && entry->e_value_size != 0) {
967 + if (entry->e_value_inum == 0 && entry->e_value_size != 0) {
968 __le32 *value = (__le32 *)((char *)header +
969 le16_to_cpu(entry->e_value_offs));
970 for (n = (le32_to_cpu(entry->e_value_size) +
971 Index: linux-stage/fs/ext4/xattr.h
972 ===================================================================
973 --- linux-stage.orig/fs/ext4/xattr.h
974 +++ linux-stage/fs/ext4/xattr.h
975 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
976 __u8 e_name_len; /* length of name */
977 __u8 e_name_index; /* attribute name index */
978 __le16 e_value_offs; /* offset in disk block of value */
979 - __le32 e_value_block; /* disk block attribute is stored on (n/i) */
980 + __le32 e_value_inum; /* inode in which the value is stored */
981 __le32 e_value_size; /* size of attribute value */
982 __le32 e_hash; /* hash value of name and value */
983 char e_name[0]; /* attribute name */
984 @@ -67,6 +67,15 @@ struct ext4_xattr_entry {
985 EXT4_I(inode)->i_extra_isize))
986 #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
988 +#define i_xattr_inode_parent i_mtime.tv_sec
991 + * The minimum size of EA value when you start storing it in an external inode
992 + * size of block - size of header - size of 1 entry - 4 null bytes
994 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b) \
995 + ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
997 #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
998 #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
999 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1000 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1001 #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1003 struct ext4_xattr_info {
1012 struct ext4_xattr_search {
1013 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1014 extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1015 extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1017 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1018 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1020 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1021 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1022 + struct ext4_xattr_ino_array **array);
1023 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1024 + struct ext4_xattr_ino_array *array);
1025 extern void ext4_xattr_put_super(struct super_block *);
1027 extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1028 Index: linux-stage/fs/ext4/ialloc.c
1029 ===================================================================
1030 --- linux-stage.orig/fs/ext4/ialloc.c
1031 +++ linux-stage/fs/ext4/ialloc.c
1032 @@ -252,7 +252,6 @@ void ext4_free_inode(handle_t *handle, s
1033 * as writing the quota to disk may need the lock as well.
1035 dquot_initialize(inode);
1036 - ext4_xattr_delete_inode(handle, inode);
1037 dquot_free_inode(inode);
1040 Index: linux-stage/fs/ext4/inline.c
1041 ===================================================================
1042 --- linux-stage.orig/fs/ext4/inline.c
1043 +++ linux-stage/fs/ext4/inline.c
1044 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1046 /* Compute min_offs. */
1047 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1048 - if (!entry->e_value_block && entry->e_value_size) {
1049 + if (!entry->e_value_inum && entry->e_value_size) {
1050 size_t offs = le16_to_cpu(entry->e_value_offs);
1051 if (offs < min_offs)