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 @@ -269,10 +296,10 @@ void ext4_evict_inode(struct inode *inod
123 * enough credits left in the handle to remove the inode from
124 * the orphan list and set the dtime field.
126 - if (!ext4_handle_has_enough_credits(handle, 3)) {
127 - err = ext4_journal_extend(handle, 3);
128 + if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
129 + err = ext4_journal_extend(handle, extra_credits);
131 - err = ext4_journal_restart(handle, 3);
132 + err = ext4_journal_restart(handle, extra_credits);
134 ext4_warning(inode->i_sb,
135 "couldn't extend journal (err %d)", err);
136 @@ -306,8 +333,12 @@ void ext4_evict_inode(struct inode *inod
137 ext4_clear_inode(inode);
139 ext4_free_inode(handle, inode);
141 ext4_journal_stop(handle);
142 sb_end_intwrite(inode->i_sb);
144 + if (lea_ino_array != NULL)
145 + ext4_xattr_inode_array_free(inode, lea_ino_array);
148 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
149 @@ -4681,7 +4712,7 @@ static int ext4_index_trans_blocks(struc
151 * Also account for superblock, inode, quota and xattr blocks
153 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
154 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
157 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
158 Index: linux-stage/fs/ext4/xattr.c
159 ===================================================================
160 --- linux-stage.orig/fs/ext4/xattr.c
161 +++ linux-stage/fs/ext4/xattr.c
162 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
164 while (!IS_LAST_ENTRY(entry)) {
165 if (entry->e_value_size != 0 &&
166 + entry->e_value_inum == 0 &&
167 (value_start + le16_to_cpu(entry->e_value_offs) <
168 (void *)e + sizeof(__u32) ||
169 value_start + le16_to_cpu(entry->e_value_offs) +
170 @@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
174 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
175 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
176 + struct inode *inode)
178 size_t value_size = le32_to_cpu(entry->e_value_size);
180 - if (entry->e_value_block != 0 || value_size > size ||
181 + if (!entry->e_value_inum &&
182 le16_to_cpu(entry->e_value_offs) + value_size > size)
184 + if (entry->e_value_inum &&
185 + (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
186 + le32_to_cpu(entry->e_value_inum) >
187 + le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
193 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
194 - const char *name, size_t size, int sorted)
195 + const char *name, size_t size, int sorted,
196 + struct inode *inode)
198 struct ext4_xattr_entry *entry;
200 @@ -265,11 +273,104 @@ ext4_xattr_find_entry(struct ext4_xattr_
204 - if (!cmp && ext4_xattr_check_entry(entry, size))
205 + if (!cmp && ext4_xattr_check_entry(entry, size, inode))
207 return cmp ? -ENODATA : 0;
211 + * Read the EA value from an inode.
214 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
216 + unsigned long block = 0;
217 + struct buffer_head *bh = NULL;
218 + int err, blocksize;
219 + size_t csize, ret_size = 0;
224 + blocksize = ea_inode->i_sb->s_blocksize;
226 + while (ret_size < *size) {
227 + csize = (*size - ret_size) > blocksize ? blocksize :
229 + bh = ext4_bread(NULL, ea_inode, block, 0, &err);
234 + memcpy(buf, bh->b_data, csize);
247 +struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
249 + struct inode *ea_inode = NULL;
251 + ea_inode = ext4_iget(parent->i_sb, ea_ino);
252 + if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
253 + int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
254 + ext4_error(parent->i_sb, "error while reading EA inode %lu "
255 + "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
256 + *err = rc != 0 ? rc : -EIO;
260 + if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
261 + ea_inode->i_generation != parent->i_generation) {
262 + ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
263 + "to parent invalid.", ea_ino);
268 + if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
269 + ext4_error(parent->i_sb, "EA inode %lu does not have "
270 + "EXT4_EA_INODE_FL flag set.\n", ea_ino);
284 + * Read the value from the EA inode.
287 +ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
290 + struct inode *ea_inode = NULL;
293 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
297 + err = ext4_xattr_inode_read(ea_inode, buffer, size);
304 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
305 void *buffer, size_t buffer_size)
306 @@ -301,7 +401,8 @@ bad_block:
308 ext4_xattr_cache_insert(bh);
310 - error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
311 + error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
316 @@ -311,8 +412,16 @@ bad_block:
318 if (size > buffer_size)
320 - memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
322 + if (entry->e_value_inum) {
323 + error = ext4_xattr_inode_get(inode,
324 + le32_to_cpu(entry->e_value_inum),
329 + memcpy(buffer, bh->b_data +
330 + le16_to_cpu(entry->e_value_offs), size);
335 @@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
338 error = ext4_xattr_find_entry(&entry, name_index, name,
339 - end - (void *)entry, 0);
340 + end - (void *)entry, 0, inode);
343 size = le32_to_cpu(entry->e_value_size);
344 @@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
346 if (size > buffer_size)
348 - memcpy(buffer, (void *)IFIRST(header) +
349 - le16_to_cpu(entry->e_value_offs), size);
350 + if (entry->e_value_inum) {
351 + error = ext4_xattr_inode_get(inode,
352 + le32_to_cpu(entry->e_value_inum),
357 + memcpy(buffer, (void *)IFIRST(header) +
358 + le16_to_cpu(entry->e_value_offs), size);
363 @@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
364 size_t *min_offs, void *base, int *total)
366 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
367 - if (!last->e_value_block && last->e_value_size) {
368 + if (!last->e_value_inum && last->e_value_size) {
369 size_t offs = le16_to_cpu(last->e_value_offs);
370 if (offs < *min_offs)
372 @@ -611,16 +728,174 @@ static size_t ext4_xattr_free_space(stru
373 return (*min_offs - ((void *)last - base) - sizeof(__u32));
377 + * Write the value of the EA in an inode.
380 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
381 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
382 + const void *buf, int bufsize)
384 + struct buffer_head *bh = NULL;
385 + unsigned long block = 0;
386 + unsigned blocksize = ea_inode->i_sb->s_blocksize;
387 + unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
388 + int csize, wsize = 0;
393 + while (ret >= 0 && ret < max_blocks) {
394 + struct ext4_map_blocks map;
395 + map.m_lblk = block += ret;
396 + map.m_len = max_blocks -= ret;
398 + ret = ext4_map_blocks(handle, ea_inode, &map,
399 + EXT4_GET_BLOCKS_CREATE);
401 + ext4_mark_inode_dirty(handle, ea_inode);
402 + if (ret == -ENOSPC &&
403 + ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
415 + while (wsize < bufsize) {
418 + csize = (bufsize - wsize) > blocksize ? blocksize :
420 + bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
423 + ret = ext4_journal_get_write_access(handle, bh);
427 + memcpy(bh->b_data, buf, csize);
428 + set_buffer_uptodate(bh);
429 + ext4_handle_dirty_metadata(handle, ea_inode, bh);
436 + mutex_lock(&ea_inode->i_mutex);
437 + i_size_write(ea_inode, wsize);
438 + ext4_update_i_disksize(ea_inode, wsize);
439 + mutex_unlock(&ea_inode->i_mutex);
441 + ext4_mark_inode_dirty(handle, ea_inode);
450 + * Create an inode to store the value of a large EA.
452 +static struct inode *
453 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
455 + struct inode *ea_inode = NULL;
458 + * Let the next inode be the goal, so we try and allocate the EA inode
459 + * in the same group, or nearby one.
461 + ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
462 + S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
464 + if (!IS_ERR(ea_inode)) {
465 + ea_inode->i_op = &ext4_file_inode_operations;
466 + ea_inode->i_fop = &ext4_file_operations;
467 + ext4_set_aops(ea_inode);
468 + ea_inode->i_generation = inode->i_generation;
469 + EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
472 + * A back-pointer from EA inode to parent inode will be useful
475 + EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
476 + unlock_new_inode(ea_inode);
483 + * Unlink the inode storing the value of the EA.
486 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
488 + struct inode *ea_inode = NULL;
491 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
495 + clear_nlink(ea_inode);
502 + * Add value of the EA in an inode.
505 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
506 + const void *value, size_t value_len)
508 + struct inode *ea_inode = NULL;
511 + /* Create an inode for the EA value */
512 + ea_inode = ext4_xattr_inode_create(handle, inode);
513 + if (IS_ERR(ea_inode))
516 + err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
518 + clear_nlink(ea_inode);
520 + *ea_ino = ea_inode->i_ino;
528 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
529 + handle_t *handle, struct inode *inode)
531 struct ext4_xattr_entry *last;
532 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
533 + int in_inode = i->in_inode;
535 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
536 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
537 + (EXT4_XATTR_SIZE(i->value_len) >
538 + EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
541 /* Compute min_offs and last. */
543 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
544 - if (!last->e_value_block && last->e_value_size) {
545 + if (!last->e_value_inum && last->e_value_size) {
546 size_t offs = le16_to_cpu(last->e_value_offs);
549 @@ -628,16 +903,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
551 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
553 - if (!s->here->e_value_block && s->here->e_value_size) {
555 + !s->here->e_value_inum && s->here->e_value_size) {
556 size_t size = le32_to_cpu(s->here->e_value_size);
557 free += EXT4_XATTR_SIZE(size);
559 free += EXT4_XATTR_LEN(name_len);
562 - if (free < EXT4_XATTR_SIZE(i->value_len) ||
563 - free < EXT4_XATTR_LEN(name_len) +
564 - EXT4_XATTR_SIZE(i->value_len))
565 + size_t value_len = EXT4_XATTR_SIZE(i->value_len);
570 + if (free < value_len ||
571 + free < EXT4_XATTR_LEN(name_len) + value_len)
575 @@ -651,7 +931,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
576 s->here->e_name_len = name_len;
577 memcpy(s->here->e_name, i->name, name_len);
579 - if (!s->here->e_value_block && s->here->e_value_size) {
580 + if (!s->here->e_value_inum && s->here->e_value_size &&
581 + s->here->e_value_offs > 0) {
582 void *first_val = s->base + min_offs;
583 size_t offs = le16_to_cpu(s->here->e_value_offs);
584 void *val = s->base + offs;
585 @@ -685,13 +966,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
587 while (!IS_LAST_ENTRY(last)) {
588 size_t o = le16_to_cpu(last->e_value_offs);
589 - if (!last->e_value_block &&
590 + if (!last->e_value_inum &&
591 last->e_value_size && o < offs)
593 cpu_to_le16(o + size);
594 last = EXT4_XATTR_NEXT(last);
597 + if (s->here->e_value_inum) {
598 + ext4_xattr_inode_unlink(inode,
599 + le32_to_cpu(s->here->e_value_inum));
600 + s->here->e_value_inum = 0;
603 /* Remove the old name. */
604 size_t size = EXT4_XATTR_LEN(name_len);
605 @@ -705,10 +990,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
607 /* Insert the new value. */
608 s->here->e_value_size = cpu_to_le32(i->value_len);
609 - if (i->value_len) {
611 + unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
612 + ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
614 + s->here->e_value_inum = cpu_to_le32(ea_ino);
615 + s->here->e_value_offs = 0;
616 + } else if (i->value_len) {
617 size_t size = EXT4_XATTR_SIZE(i->value_len);
618 void *val = s->base + min_offs - size;
619 s->here->e_value_offs = cpu_to_le16(min_offs - size);
620 + s->here->e_value_inum = 0;
621 if (i->value == EXT4_ZERO_XATTR_VALUE) {
622 memset(val, 0, size);
624 @@ -758,7 +1050,7 @@ ext4_xattr_block_find(struct inode *inod
625 bs->s.end = bs->bh->b_data + bs->bh->b_size;
626 bs->s.here = bs->s.first;
627 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
628 - i->name, bs->bh->b_size, 1);
629 + i->name, bs->bh->b_size, 1, inode);
630 if (error && error != -ENODATA)
632 bs->s.not_found = error;
633 @@ -782,8 +1074,6 @@ ext4_xattr_block_set(handle_t *handle, s
635 #define header(x) ((struct ext4_xattr_header *)(x))
637 - if (i->value && i->value_len > sb->s_blocksize)
640 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
642 @@ -799,7 +1089,7 @@ ext4_xattr_block_set(handle_t *handle, s
645 ea_bdebug(bs->bh, "modifying in-place");
646 - error = ext4_xattr_set_entry(i, s);
647 + error = ext4_xattr_set_entry(i, s, handle, inode);
649 if (!IS_LAST_ENTRY(s->first))
650 ext4_xattr_rehash(header(s->base),
651 @@ -850,7 +1140,7 @@ ext4_xattr_block_set(handle_t *handle, s
652 s->end = s->base + sb->s_blocksize;
655 - error = ext4_xattr_set_entry(i, s);
656 + error = ext4_xattr_set_entry(i, s, handle, inode);
660 @@ -1000,7 +1290,7 @@ int ext4_xattr_ibody_find(struct inode *
661 /* Find the named attribute. */
662 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
664 - (void *)is->s.base, 0);
665 + (void *)is->s.base, 0, inode);
666 if (error && error != -ENODATA)
668 is->s.not_found = error;
669 @@ -1018,7 +1308,7 @@ int ext4_xattr_ibody_inline_set(handle_t
671 if (EXT4_I(inode)->i_extra_isize == 0)
673 - error = ext4_xattr_set_entry(i, s);
674 + error = ext4_xattr_set_entry(i, s, handle, inode);
676 if (error == -ENOSPC &&
677 ext4_has_inline_data(inode)) {
678 @@ -1030,7 +1320,7 @@ int ext4_xattr_ibody_inline_set(handle_t
679 error = ext4_xattr_ibody_find(inode, i, is);
682 - error = ext4_xattr_set_entry(i, s);
683 + error = ext4_xattr_set_entry(i, s, handle, inode);
687 @@ -1056,7 +1346,7 @@ static int ext4_xattr_ibody_set(handle_t
689 if (EXT4_I(inode)->i_extra_isize == 0)
691 - error = ext4_xattr_set_entry(i, s);
692 + error = ext4_xattr_set_entry(i, s, handle, inode);
695 header = IHDR(inode, ext4_raw_inode(&is->iloc));
696 @@ -1092,7 +1382,7 @@ ext4_xattr_set_handle(handle_t *handle,
699 .value_len = value_len,
703 struct ext4_xattr_ibody_find is = {
704 .s = { .not_found = -ENODATA, },
705 @@ -1157,6 +1447,15 @@ ext4_xattr_set_handle(handle_t *handle,
708 error = ext4_xattr_block_set(handle, inode, &i, &bs);
709 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
710 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
711 + error == -ENOSPC) {
712 + /* xattr not fit to block, store at external
715 + error = ext4_xattr_ibody_set(handle, inode,
720 if (!is.s.not_found) {
721 @@ -1203,9 +1502,22 @@ ext4_xattr_set(struct inode *inode, int
722 const void *value, size_t value_len, int flags)
725 + struct super_block *sb = inode->i_sb;
726 int error, retries = 0;
727 int credits = ext4_jbd2_credits_xattr(inode);
729 + if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
730 + EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
731 + int nrblocks = (value_len + sb->s_blocksize - 1) >>
732 + sb->s_blocksize_bits;
734 + /* For new inode */
735 + credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
737 + /* For data blocks of EA inode */
738 + credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
742 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
743 if (IS_ERR(handle)) {
744 @@ -1217,7 +1529,7 @@ retry:
745 value, value_len, flags);
746 error2 = ext4_journal_stop(handle);
747 if (error == -ENOSPC &&
748 - ext4_should_retry_alloc(inode->i_sb, &retries))
749 + ext4_should_retry_alloc(sb, &retries))
753 @@ -1239,7 +1551,7 @@ static void ext4_xattr_shift_entries(str
755 /* Adjust the value offsets of the entries */
756 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
757 - if (!last->e_value_block && last->e_value_size) {
758 + if (!last->e_value_inum && last->e_value_size) {
759 new_offs = le16_to_cpu(last->e_value_offs) +
761 BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
762 @@ -1477,21 +1789,135 @@ cleanup:
766 +#define EIA_INCR 16 /* must be 2^n */
767 +#define EIA_MASK (EIA_INCR - 1)
768 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
769 + * If @lea_ino_array is new or full it will be grown and the old
770 + * contents copied over.
773 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
775 + if (*lea_ino_array == NULL) {
777 + * Start with 15 inodes, so it fits into a power-of-two size.
778 + * If *lea_ino_array is NULL, this is essentially offsetof()
781 + kmalloc(offsetof(struct ext4_xattr_ino_array,
782 + xia_inodes[EIA_MASK]),
784 + if (*lea_ino_array == NULL)
786 + (*lea_ino_array)->xia_count = 0;
787 + } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
788 + /* expand the array once all 15 + n * 16 slots are full */
789 + struct ext4_xattr_ino_array *new_array = NULL;
790 + int count = (*lea_ino_array)->xia_count;
792 + /* if new_array is NULL, this is essentially offsetof() */
793 + new_array = kmalloc(
794 + offsetof(struct ext4_xattr_ino_array,
795 + xia_inodes[count + EIA_INCR]),
797 + if (new_array == NULL)
799 + memcpy(new_array, *lea_ino_array,
800 + offsetof(struct ext4_xattr_ino_array,
801 + xia_inodes[count]));
802 + kfree(*lea_ino_array);
803 + *lea_ino_array = new_array;
805 + (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
810 + * Add xattr inode to orphan list
813 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
814 + int credits, struct ext4_xattr_ino_array *lea_ino_array)
816 + struct inode *ea_inode = NULL;
817 + int idx = 0, error = 0;
819 + if (lea_ino_array == NULL)
822 + for (; idx < lea_ino_array->xia_count; ++idx) {
823 + if (!ext4_handle_has_enough_credits(handle, credits)) {
824 + error = ext4_journal_extend(handle, credits);
826 + error = ext4_journal_restart(handle, credits);
829 + ext4_warning(inode->i_sb,
830 + "couldn't extend journal "
831 + "(err %d)", error);
835 + ea_inode = ext4_xattr_inode_iget(inode,
836 + lea_ino_array->xia_inodes[idx], &error);
839 + ext4_orphan_add(handle, ea_inode);
840 + /* the inode's i_count will be released by caller */
847 * ext4_xattr_delete_inode()
849 - * Free extended attribute resources associated with this inode. This
850 + * Free extended attribute resources associated with this inode. Traverse
851 + * all entries and unlink any xattr inodes associated with this inode. This
852 * is called immediately before an inode is freed. We have exclusive
853 - * access to the inode.
854 + * access to the inode. If an orphan inode is deleted it will also delete any
855 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
856 + * to ensure they belong to the parent inode and were not deleted already.
859 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
861 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
862 + struct ext4_xattr_ino_array **lea_ino_array)
864 struct buffer_head *bh = NULL;
865 + struct ext4_xattr_ibody_header *header;
866 + struct ext4_inode *raw_inode;
867 + struct ext4_iloc iloc;
868 + struct ext4_xattr_entry *entry;
869 + int credits = 3, error = 0;
871 - if (!EXT4_I(inode)->i_file_acl)
872 + if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
873 + goto delete_external_ea;
875 + error = ext4_get_inode_loc(inode, &iloc);
878 + raw_inode = ext4_raw_inode(&iloc);
879 + header = IHDR(inode, raw_inode);
880 + for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
881 + entry = EXT4_XATTR_NEXT(entry)) {
882 + if (!entry->e_value_inum)
884 + if (ext4_expand_ino_array(lea_ino_array,
885 + entry->e_value_inum) != 0) {
889 + entry->e_value_inum = 0;
894 + if (!EXT4_I(inode)->i_file_acl) {
895 + /* add xattr inode to orphan list */
896 + ext4_xattr_inode_orphan_add(handle, inode, credits,
900 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
902 EXT4_ERROR_INODE(inode, "block %llu read error",
903 @@ -1504,11 +1930,69 @@ ext4_xattr_delete_inode(handle_t *handle
904 EXT4_I(inode)->i_file_acl);
908 + for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
909 + entry = EXT4_XATTR_NEXT(entry)) {
910 + if (!entry->e_value_inum)
912 + if (ext4_expand_ino_array(lea_ino_array,
913 + entry->e_value_inum) != 0)
915 + entry->e_value_inum = 0;
918 + /* add xattr inode to orphan list */
919 + error = ext4_xattr_inode_orphan_add(handle, inode, credits,
924 + if (!IS_NOQUOTA(inode))
925 + credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
927 + if (!ext4_handle_has_enough_credits(handle, credits)) {
928 + error = ext4_journal_extend(handle, credits);
930 + error = ext4_journal_restart(handle, credits);
932 + ext4_warning(inode->i_sb,
933 + "couldn't extend journal (err %d)", error);
938 ext4_xattr_release_block(handle, inode, bh);
939 EXT4_I(inode)->i_file_acl = 0;
948 +ext4_xattr_inode_array_free(struct inode *inode,
949 + struct ext4_xattr_ino_array *lea_ino_array)
951 + struct inode *ea_inode = NULL;
955 + if (lea_ino_array == NULL)
958 + for (; idx < lea_ino_array->xia_count; ++idx) {
959 + ea_inode = ext4_xattr_inode_iget(inode,
960 + lea_ino_array->xia_inodes[idx], &err);
963 + /* for inode's i_count get from ext4_xattr_delete_inode */
964 + if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
966 + clear_nlink(ea_inode);
969 + kfree(lea_ino_array);
973 @@ -1578,10 +2062,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
974 entry1->e_name_index != entry2->e_name_index ||
975 entry1->e_name_len != entry2->e_name_len ||
976 entry1->e_value_size != entry2->e_value_size ||
977 + entry1->e_value_inum != entry2->e_value_inum ||
978 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
980 - if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
982 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
983 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
984 le32_to_cpu(entry1->e_value_size)))
985 @@ -1665,7 +2148,7 @@ static inline void ext4_xattr_hash_entry
989 - if (entry->e_value_block == 0 && entry->e_value_size != 0) {
990 + if (!entry->e_value_inum && entry->e_value_size) {
991 __le32 *value = (__le32 *)((char *)header +
992 le16_to_cpu(entry->e_value_offs));
993 for (n = (le32_to_cpu(entry->e_value_size) +
994 Index: linux-stage/fs/ext4/xattr.h
995 ===================================================================
996 --- linux-stage.orig/fs/ext4/xattr.h
997 +++ linux-stage/fs/ext4/xattr.h
998 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
999 __u8 e_name_len; /* length of name */
1000 __u8 e_name_index; /* attribute name index */
1001 __le16 e_value_offs; /* offset in disk block of value */
1002 - __le32 e_value_block; /* disk block attribute is stored on (n/i) */
1003 + __le32 e_value_inum; /* inode in which the value is stored */
1004 __le32 e_value_size; /* size of attribute value */
1005 __le32 e_hash; /* hash value of name and value */
1006 char e_name[0]; /* attribute name */
1007 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
1008 EXT4_I(inode)->i_extra_isize))
1009 #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1012 + * Link EA inode back to parent one using i_mtime field.
1013 + * Extra integer type conversion added to ignore higher
1014 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1016 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum) \
1018 + (inode)->i_mtime.tv_sec = inum; \
1021 +#define EXT4_XATTR_INODE_GET_PARENT(inode) \
1022 +((__u32)(inode)->i_mtime.tv_sec)
1025 + * The minimum size of EA value when you start storing it in an external inode
1026 + * size of block - size of header - size of 1 entry - 4 null bytes
1028 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b) \
1029 + ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1031 #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1032 #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1033 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1034 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1035 #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1037 struct ext4_xattr_info {
1046 struct ext4_xattr_search {
1047 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1048 extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1049 extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1051 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1052 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1054 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1055 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1056 + struct ext4_xattr_ino_array **array);
1057 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1058 + struct ext4_xattr_ino_array *array);
1059 extern void ext4_xattr_put_super(struct super_block *);
1061 extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1062 Index: linux-stage/fs/ext4/ialloc.c
1063 ===================================================================
1064 --- linux-stage.orig/fs/ext4/ialloc.c
1065 +++ linux-stage/fs/ext4/ialloc.c
1066 @@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
1067 * as writing the quota to disk may need the lock as well.
1069 dquot_initialize(inode);
1070 - ext4_xattr_delete_inode(handle, inode);
1071 dquot_free_inode(inode);
1074 Index: linux-stage/fs/ext4/inline.c
1075 ===================================================================
1076 --- linux-stage.orig/fs/ext4/inline.c
1077 +++ linux-stage/fs/ext4/inline.c
1078 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1080 /* Compute min_offs. */
1081 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1082 - if (!entry->e_value_block && entry->e_value_size) {
1083 + if (!entry->e_value_inum && entry->e_value_size) {
1084 size_t offs = le16_to_cpu(entry->e_value_offs);
1085 if (offs < min_offs)