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 @@ -1579,6 +1579,7 @@ 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 @@ -1979,6 +1980,12 @@ 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 @@ -1990,6 +1997,10 @@ 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 @@ -2194,6 +2205,7 @@ 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_zero_partial_blocks(handle_t *handle, struct inode *inode,
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 @@ -134,8 +134,6 @@ 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 @@ -184,6 +182,8 @@ 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 @@ -236,8 +236,8 @@ 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 @@ -249,9 +249,36 @@ 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 @@ -306,8 +333,12 @@ 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 @@ -4681,7 +4712,7 @@ 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 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
150 while (!IS_LAST_ENTRY(entry)) {
151 if (entry->e_value_size != 0 &&
152 + entry->e_value_inum == 0 &&
153 (value_start + le16_to_cpu(entry->e_value_offs) <
154 (void *)e + sizeof(__u32) ||
155 value_start + le16_to_cpu(entry->e_value_offs) +
156 @@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
160 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
161 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
162 + struct inode *inode)
164 size_t value_size = le32_to_cpu(entry->e_value_size);
166 - if (entry->e_value_block != 0 || value_size > size ||
167 + if (!entry->e_value_inum &&
168 le16_to_cpu(entry->e_value_offs) + value_size > size)
170 + if (entry->e_value_inum &&
171 + (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
172 + le32_to_cpu(entry->e_value_inum) >
173 + le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
179 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
180 - const char *name, size_t size, int sorted)
181 + const char *name, size_t size, int sorted,
182 + struct inode *inode)
184 struct ext4_xattr_entry *entry;
186 @@ -265,11 +273,104 @@ ext4_xattr_find_entry(struct ext4_xattr_
190 - if (!cmp && ext4_xattr_check_entry(entry, size))
191 + if (!cmp && ext4_xattr_check_entry(entry, size, inode))
193 return cmp ? -ENODATA : 0;
197 + * Read the EA value from an inode.
200 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
202 + unsigned long block = 0;
203 + struct buffer_head *bh = NULL;
204 + int err, blocksize;
205 + size_t csize, ret_size = 0;
210 + blocksize = ea_inode->i_sb->s_blocksize;
212 + while (ret_size < *size) {
213 + csize = (*size - ret_size) > blocksize ? blocksize :
215 + bh = ext4_bread(NULL, ea_inode, block, 0, &err);
220 + memcpy(buf, bh->b_data, csize);
233 +struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
235 + struct inode *ea_inode = NULL;
237 + ea_inode = ext4_iget(parent->i_sb, ea_ino);
238 + if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
239 + int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
240 + ext4_error(parent->i_sb, "error while reading EA inode %lu "
241 + "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
242 + *err = rc != 0 ? rc : -EIO;
246 + if (ea_inode->i_xattr_inode_parent != parent->i_ino ||
247 + ea_inode->i_generation != parent->i_generation) {
248 + ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
249 + "to parent invalid.", ea_ino);
254 + if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
255 + ext4_error(parent->i_sb, "EA inode %lu does not have "
256 + "EXT4_EA_INODE_FL flag set.\n", ea_ino);
270 + * Read the value from the EA inode.
273 +ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
276 + struct inode *ea_inode = NULL;
279 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
283 + err = ext4_xattr_inode_read(ea_inode, buffer, size);
290 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
291 void *buffer, size_t buffer_size)
292 @@ -301,7 +401,8 @@ bad_block:
294 ext4_xattr_cache_insert(bh);
296 - error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
297 + error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
302 @@ -311,8 +412,16 @@ bad_block:
304 if (size > buffer_size)
306 - memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
308 + if (entry->e_value_inum) {
309 + error = ext4_xattr_inode_get(inode,
310 + le32_to_cpu(entry->e_value_inum),
315 + memcpy(buffer, bh->b_data +
316 + le16_to_cpu(entry->e_value_offs), size);
321 @@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
324 error = ext4_xattr_find_entry(&entry, name_index, name,
325 - end - (void *)entry, 0);
326 + end - (void *)entry, 0, inode);
329 size = le32_to_cpu(entry->e_value_size);
330 @@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
332 if (size > buffer_size)
334 - memcpy(buffer, (void *)IFIRST(header) +
335 - le16_to_cpu(entry->e_value_offs), size);
336 + if (entry->e_value_inum) {
337 + error = ext4_xattr_inode_get(inode,
338 + le32_to_cpu(entry->e_value_inum),
343 + memcpy(buffer, (void *)IFIRST(header) +
344 + le16_to_cpu(entry->e_value_offs), size);
349 @@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
350 size_t *min_offs, void *base, int *total)
352 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
353 - if (!last->e_value_block && last->e_value_size) {
354 + if (!last->e_value_inum && last->e_value_size) {
355 size_t offs = le16_to_cpu(last->e_value_offs);
356 if (offs < *min_offs)
358 @@ -611,16 +728,174 @@ static size_t ext4_xattr_free_space(stru
359 return (*min_offs - ((void *)last - base) - sizeof(__u32));
363 + * Write the value of the EA in an inode.
366 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
367 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
368 + const void *buf, int bufsize)
370 + struct buffer_head *bh = NULL;
371 + unsigned long block = 0;
372 + unsigned blocksize = ea_inode->i_sb->s_blocksize;
373 + unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
374 + int csize, wsize = 0;
379 + while (ret >= 0 && ret < max_blocks) {
380 + struct ext4_map_blocks map;
381 + map.m_lblk = block += ret;
382 + map.m_len = max_blocks -= ret;
384 + ret = ext4_map_blocks(handle, ea_inode, &map,
385 + EXT4_GET_BLOCKS_CREATE);
387 + ext4_mark_inode_dirty(handle, ea_inode);
388 + if (ret == -ENOSPC &&
389 + ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
401 + while (wsize < bufsize) {
404 + csize = (bufsize - wsize) > blocksize ? blocksize :
406 + bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
409 + ret = ext4_journal_get_write_access(handle, bh);
413 + memcpy(bh->b_data, buf, csize);
414 + set_buffer_uptodate(bh);
415 + ext4_handle_dirty_metadata(handle, ea_inode, bh);
422 + mutex_lock(&ea_inode->i_mutex);
423 + i_size_write(ea_inode, wsize);
424 + ext4_update_i_disksize(ea_inode, wsize);
425 + mutex_unlock(&ea_inode->i_mutex);
427 + ext4_mark_inode_dirty(handle, ea_inode);
436 + * Create an inode to store the value of a large EA.
438 +static struct inode *
439 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
441 + struct inode *ea_inode = NULL;
444 + * Let the next inode be the goal, so we try and allocate the EA inode
445 + * in the same group, or nearby one.
447 + ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
448 + S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
450 + if (!IS_ERR(ea_inode)) {
451 + ea_inode->i_op = &ext4_file_inode_operations;
452 + ea_inode->i_fop = &ext4_file_operations;
453 + ext4_set_aops(ea_inode);
454 + ea_inode->i_generation = inode->i_generation;
455 + EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
458 + * A back-pointer from EA inode to parent inode will be useful
461 + ea_inode->i_xattr_inode_parent = inode->i_ino;
462 + unlock_new_inode(ea_inode);
469 + * Unlink the inode storing the value of the EA.
472 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
474 + struct inode *ea_inode = NULL;
477 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
481 + clear_nlink(ea_inode);
488 + * Add value of the EA in an inode.
491 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
492 + const void *value, size_t value_len)
494 + struct inode *ea_inode = NULL;
497 + /* Create an inode for the EA value */
498 + ea_inode = ext4_xattr_inode_create(handle, inode);
499 + if (IS_ERR(ea_inode))
502 + err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
504 + clear_nlink(ea_inode);
506 + *ea_ino = ea_inode->i_ino;
514 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
515 + handle_t *handle, struct inode *inode)
517 struct ext4_xattr_entry *last;
518 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
519 + int in_inode = i->in_inode;
521 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
522 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
523 + (EXT4_XATTR_SIZE(i->value_len) >
524 + EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
527 /* Compute min_offs and last. */
529 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
530 - if (!last->e_value_block && last->e_value_size) {
531 + if (!last->e_value_inum && last->e_value_size) {
532 size_t offs = le16_to_cpu(last->e_value_offs);
535 @@ -628,15 +903,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
537 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
539 - if (!s->here->e_value_block && s->here->e_value_size) {
541 + !s->here->e_value_inum && s->here->e_value_size) {
542 size_t size = le32_to_cpu(s->here->e_value_size);
543 free += EXT4_XATTR_SIZE(size);
545 free += EXT4_XATTR_LEN(name_len);
548 - if (free < EXT4_XATTR_LEN(name_len) +
549 - EXT4_XATTR_SIZE(i->value_len))
550 + size_t value_len = EXT4_XATTR_SIZE(i->value_len);
555 + if (free < value_len ||
556 + free < EXT4_XATTR_LEN(name_len) + value_len)
560 @@ -651,7 +931,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
561 s->here->e_name_len = name_len;
562 memcpy(s->here->e_name, i->name, name_len);
564 - if (!s->here->e_value_block && s->here->e_value_size) {
565 + if (!s->here->e_value_inum && s->here->e_value_size &&
566 + s->here->e_value_offs > 0) {
567 void *first_val = s->base + min_offs;
568 size_t offs = le16_to_cpu(s->here->e_value_offs);
569 void *val = s->base + offs;
570 @@ -685,13 +966,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
572 while (!IS_LAST_ENTRY(last)) {
573 size_t o = le16_to_cpu(last->e_value_offs);
574 - if (!last->e_value_block &&
575 + if (!last->e_value_inum &&
576 last->e_value_size && o < offs)
578 cpu_to_le16(o + size);
579 last = EXT4_XATTR_NEXT(last);
582 + if (s->here->e_value_inum) {
583 + ext4_xattr_inode_unlink(inode,
584 + le32_to_cpu(s->here->e_value_inum));
585 + s->here->e_value_inum = 0;
588 /* Remove the old name. */
589 size_t size = EXT4_XATTR_LEN(name_len);
590 @@ -705,10 +990,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
592 /* Insert the new value. */
593 s->here->e_value_size = cpu_to_le32(i->value_len);
594 - if (i->value_len) {
596 + unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
597 + ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
599 + s->here->e_value_inum = cpu_to_le32(ea_ino);
600 + s->here->e_value_offs = 0;
601 + } else if (i->value_len) {
602 size_t size = EXT4_XATTR_SIZE(i->value_len);
603 void *val = s->base + min_offs - size;
604 s->here->e_value_offs = cpu_to_le16(min_offs - size);
605 + s->here->e_value_inum = 0;
606 if (i->value == EXT4_ZERO_XATTR_VALUE) {
607 memset(val, 0, size);
609 @@ -758,7 +1050,7 @@ ext4_xattr_block_find(struct inode *inod
610 bs->s.end = bs->bh->b_data + bs->bh->b_size;
611 bs->s.here = bs->s.first;
612 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
613 - i->name, bs->bh->b_size, 1);
614 + i->name, bs->bh->b_size, 1, inode);
615 if (error && error != -ENODATA)
617 bs->s.not_found = error;
618 @@ -782,8 +1074,6 @@ ext4_xattr_block_set(handle_t *handle, s
620 #define header(x) ((struct ext4_xattr_header *)(x))
622 - if (i->value && i->value_len > sb->s_blocksize)
625 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
627 @@ -799,7 +1089,7 @@ ext4_xattr_block_set(handle_t *handle, s
630 ea_bdebug(bs->bh, "modifying in-place");
631 - error = ext4_xattr_set_entry(i, s);
632 + error = ext4_xattr_set_entry(i, s, handle, inode);
634 if (!IS_LAST_ENTRY(s->first))
635 ext4_xattr_rehash(header(s->base),
636 @@ -850,7 +1140,7 @@ ext4_xattr_block_set(handle_t *handle, s
637 s->end = s->base + sb->s_blocksize;
640 - error = ext4_xattr_set_entry(i, s);
641 + error = ext4_xattr_set_entry(i, s, handle, inode);
645 @@ -1000,7 +1290,7 @@ int ext4_xattr_ibody_find(struct inode *
646 /* Find the named attribute. */
647 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
649 - (void *)is->s.base, 0);
650 + (void *)is->s.base, 0, inode);
651 if (error && error != -ENODATA)
653 is->s.not_found = error;
654 @@ -1018,7 +1308,7 @@ int ext4_xattr_ibody_inline_set(handle_t
656 if (EXT4_I(inode)->i_extra_isize == 0)
658 - error = ext4_xattr_set_entry(i, s);
659 + error = ext4_xattr_set_entry(i, s, handle, inode);
661 if (error == -ENOSPC &&
662 ext4_has_inline_data(inode)) {
663 @@ -1030,7 +1320,7 @@ int ext4_xattr_ibody_inline_set(handle_t
664 error = ext4_xattr_ibody_find(inode, i, is);
667 - error = ext4_xattr_set_entry(i, s);
668 + error = ext4_xattr_set_entry(i, s, handle, inode);
672 @@ -1056,7 +1346,7 @@ static int ext4_xattr_ibody_set(handle_t
674 if (EXT4_I(inode)->i_extra_isize == 0)
676 - error = ext4_xattr_set_entry(i, s);
677 + error = ext4_xattr_set_entry(i, s, handle, inode);
680 header = IHDR(inode, ext4_raw_inode(&is->iloc));
681 @@ -1092,7 +1382,7 @@ ext4_xattr_set_handle(handle_t *handle,
684 .value_len = value_len,
688 struct ext4_xattr_ibody_find is = {
689 .s = { .not_found = -ENODATA, },
690 @@ -1157,6 +1447,15 @@ ext4_xattr_set_handle(handle_t *handle,
693 error = ext4_xattr_block_set(handle, inode, &i, &bs);
694 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
695 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
696 + error == -ENOSPC) {
697 + /* xattr not fit to block, store at external
700 + error = ext4_xattr_ibody_set(handle, inode,
705 if (!is.s.not_found) {
706 @@ -1203,9 +1502,22 @@ ext4_xattr_set(struct inode *inode, int
707 const void *value, size_t value_len, int flags)
710 + struct super_block *sb = inode->i_sb;
711 int error, retries = 0;
712 int credits = ext4_jbd2_credits_xattr(inode);
714 + if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
715 + EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
716 + int nrblocks = (value_len + sb->s_blocksize - 1) >>
717 + sb->s_blocksize_bits;
719 + /* For new inode */
720 + credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
722 + /* For data blocks of EA inode */
723 + credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
727 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
728 if (IS_ERR(handle)) {
729 @@ -1217,7 +1529,7 @@ retry:
730 value, value_len, flags);
731 error2 = ext4_journal_stop(handle);
732 if (error == -ENOSPC &&
733 - ext4_should_retry_alloc(inode->i_sb, &retries))
734 + ext4_should_retry_alloc(sb, &retries))
738 @@ -1239,7 +1551,7 @@ static void ext4_xattr_shift_entries(str
740 /* Adjust the value offsets of the entries */
741 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
742 - if (!last->e_value_block && last->e_value_size) {
743 + if (!last->e_value_inum && last->e_value_size) {
744 new_offs = le16_to_cpu(last->e_value_offs) +
746 BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
747 @@ -1477,21 +1789,135 @@ cleanup:
751 +#define EIA_INCR 16 /* must be 2^n */
752 +#define EIA_MASK (EIA_INCR - 1)
753 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
754 + * If @lea_ino_array is new or full it will be grown and the old
755 + * contents copied over.
758 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
760 + if (*lea_ino_array == NULL) {
762 + * Start with 15 inodes, so it fits into a power-of-two size.
763 + * If *lea_ino_array is NULL, this is essentially offsetof()
766 + kmalloc(offsetof(struct ext4_xattr_ino_array,
767 + xia_inodes[EIA_MASK]),
769 + if (*lea_ino_array == NULL)
771 + (*lea_ino_array)->xia_count = 0;
772 + } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
773 + /* expand the array once all 15 + n * 16 slots are full */
774 + struct ext4_xattr_ino_array *new_array = NULL;
775 + int count = (*lea_ino_array)->xia_count;
777 + /* if new_array is NULL, this is essentially offsetof() */
778 + new_array = kmalloc(
779 + offsetof(struct ext4_xattr_ino_array,
780 + xia_inodes[count + EIA_INCR]),
782 + if (new_array == NULL)
784 + memcpy(new_array, *lea_ino_array,
785 + offsetof(struct ext4_xattr_ino_array,
786 + xia_inodes[count]));
787 + kfree(*lea_ino_array);
788 + *lea_ino_array = new_array;
790 + (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
795 + * Add xattr inode to orphan list
798 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
799 + int credits, struct ext4_xattr_ino_array *lea_ino_array)
801 + struct inode *ea_inode = NULL;
802 + int idx = 0, error = 0;
804 + if (lea_ino_array == NULL)
807 + for (; idx < lea_ino_array->xia_count; ++idx) {
808 + if (!ext4_handle_has_enough_credits(handle, credits)) {
809 + error = ext4_journal_extend(handle, credits);
811 + error = ext4_journal_restart(handle, credits);
814 + ext4_warning(inode->i_sb,
815 + "couldn't extend journal "
816 + "(err %d)", error);
820 + ea_inode = ext4_xattr_inode_iget(inode,
821 + lea_ino_array->xia_inodes[idx], &error);
824 + ext4_orphan_add(handle, ea_inode);
825 + /* the inode's i_count will be released by caller */
832 * ext4_xattr_delete_inode()
834 - * Free extended attribute resources associated with this inode. This
835 + * Free extended attribute resources associated with this inode. Traverse
836 + * all entries and unlink any xattr inodes associated with this inode. This
837 * is called immediately before an inode is freed. We have exclusive
838 - * access to the inode.
839 + * access to the inode. If an orphan inode is deleted it will also delete any
840 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
841 + * to ensure they belong to the parent inode and were not deleted already.
844 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
846 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
847 + struct ext4_xattr_ino_array **lea_ino_array)
849 struct buffer_head *bh = NULL;
850 + struct ext4_xattr_ibody_header *header;
851 + struct ext4_inode *raw_inode;
852 + struct ext4_iloc iloc;
853 + struct ext4_xattr_entry *entry;
854 + int credits = 3, error = 0;
856 - if (!EXT4_I(inode)->i_file_acl)
857 + if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
858 + goto delete_external_ea;
860 + error = ext4_get_inode_loc(inode, &iloc);
863 + raw_inode = ext4_raw_inode(&iloc);
864 + header = IHDR(inode, raw_inode);
865 + for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
866 + entry = EXT4_XATTR_NEXT(entry)) {
867 + if (!entry->e_value_inum)
869 + if (ext4_expand_ino_array(lea_ino_array,
870 + entry->e_value_inum) != 0) {
874 + entry->e_value_inum = 0;
879 + if (!EXT4_I(inode)->i_file_acl) {
880 + /* add xattr inode to orphan list */
881 + ext4_xattr_inode_orphan_add(handle, inode, credits,
885 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
887 EXT4_ERROR_INODE(inode, "block %llu read error",
888 @@ -1504,11 +1930,69 @@ ext4_xattr_delete_inode(handle_t *handle
889 EXT4_I(inode)->i_file_acl);
893 + for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
894 + entry = EXT4_XATTR_NEXT(entry)) {
895 + if (!entry->e_value_inum)
897 + if (ext4_expand_ino_array(lea_ino_array,
898 + entry->e_value_inum) != 0)
900 + entry->e_value_inum = 0;
903 + /* add xattr inode to orphan list */
904 + error = ext4_xattr_inode_orphan_add(handle, inode, credits,
909 + if (!IS_NOQUOTA(inode))
910 + credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
912 + if (!ext4_handle_has_enough_credits(handle, credits)) {
913 + error = ext4_journal_extend(handle, credits);
915 + error = ext4_journal_restart(handle, credits);
917 + ext4_warning(inode->i_sb,
918 + "couldn't extend journal (err %d)", error);
923 ext4_xattr_release_block(handle, inode, bh);
924 EXT4_I(inode)->i_file_acl = 0;
933 +ext4_xattr_inode_array_free(struct inode *inode,
934 + struct ext4_xattr_ino_array *lea_ino_array)
936 + struct inode *ea_inode = NULL;
940 + if (lea_ino_array == NULL)
943 + for (; idx < lea_ino_array->xia_count; ++idx) {
944 + ea_inode = ext4_xattr_inode_iget(inode,
945 + lea_ino_array->xia_inodes[idx], &err);
948 + /* for inode's i_count get from ext4_xattr_delete_inode */
949 + if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
951 + clear_nlink(ea_inode);
954 + kfree(lea_ino_array);
958 @@ -1578,10 +2062,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
959 entry1->e_name_index != entry2->e_name_index ||
960 entry1->e_name_len != entry2->e_name_len ||
961 entry1->e_value_size != entry2->e_value_size ||
962 + entry1->e_value_inum != entry2->e_value_inum ||
963 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
965 - if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
967 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
968 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
969 le32_to_cpu(entry1->e_value_size)))
970 @@ -1665,7 +2148,7 @@ static inline void ext4_xattr_hash_entry
974 - if (entry->e_value_block == 0 && entry->e_value_size != 0) {
975 + if (!entry->e_value_inum && entry->e_value_size) {
976 __le32 *value = (__le32 *)((char *)header +
977 le16_to_cpu(entry->e_value_offs));
978 for (n = (le32_to_cpu(entry->e_value_size) +
979 Index: linux-stage/fs/ext4/xattr.h
980 ===================================================================
981 --- linux-stage.orig/fs/ext4/xattr.h
982 +++ linux-stage/fs/ext4/xattr.h
983 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
984 __u8 e_name_len; /* length of name */
985 __u8 e_name_index; /* attribute name index */
986 __le16 e_value_offs; /* offset in disk block of value */
987 - __le32 e_value_block; /* disk block attribute is stored on (n/i) */
988 + __le32 e_value_inum; /* inode in which the value is stored */
989 __le32 e_value_size; /* size of attribute value */
990 __le32 e_hash; /* hash value of name and value */
991 char e_name[0]; /* attribute name */
992 @@ -67,6 +67,15 @@ struct ext4_xattr_entry {
993 EXT4_I(inode)->i_extra_isize))
994 #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
996 +#define i_xattr_inode_parent i_mtime.tv_sec
999 + * The minimum size of EA value when you start storing it in an external inode
1000 + * size of block - size of header - size of 1 entry - 4 null bytes
1002 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b) \
1003 + ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1005 #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1006 #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1007 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1008 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1009 #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1011 struct ext4_xattr_info {
1020 struct ext4_xattr_search {
1021 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1022 extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1023 extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1025 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1026 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1028 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1029 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1030 + struct ext4_xattr_ino_array **array);
1031 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1032 + struct ext4_xattr_ino_array *array);
1033 extern void ext4_xattr_put_super(struct super_block *);
1035 extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1036 Index: linux-stage/fs/ext4/ialloc.c
1037 ===================================================================
1038 --- linux-stage.orig/fs/ext4/ialloc.c
1039 +++ linux-stage/fs/ext4/ialloc.c
1040 @@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
1041 * as writing the quota to disk may need the lock as well.
1043 dquot_initialize(inode);
1044 - ext4_xattr_delete_inode(handle, inode);
1045 dquot_free_inode(inode);
1048 Index: linux-stage/fs/ext4/inline.c
1049 ===================================================================
1050 --- linux-stage.orig/fs/ext4/inline.c
1051 +++ linux-stage/fs/ext4/inline.c
1052 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1054 /* Compute min_offs. */
1055 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1056 - if (!entry->e_value_block && entry->e_value_size) {
1057 + if (!entry->e_value_inum && entry->e_value_size) {
1058 size_t offs = le16_to_cpu(entry->e_value_offs);
1059 if (offs < min_offs)