1 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
2 index 10a2a86..217fdcc 100644
5 @@ -1781,6 +1781,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(encrypt, ENCRYPT)
6 EXT4_FEATURE_INCOMPAT_EXTENTS| \
7 EXT4_FEATURE_INCOMPAT_64BIT| \
8 EXT4_FEATURE_INCOMPAT_FLEX_BG| \
9 + EXT4_FEATURE_INCOMPAT_EA_INODE| \
10 EXT4_FEATURE_INCOMPAT_MMP | \
11 EXT4_FEATURE_INCOMPAT_DIRDATA| \
12 EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
13 @@ -2252,6 +2259,10 @@ struct mmpd_data {
14 # define ATTRIB_NORET __attribute__((noreturn))
15 # define NORET_AND noreturn,
17 +struct ext4_xattr_ino_array {
18 + unsigned int xia_count; /* # of used item in the array */
19 + unsigned int xia_inodes[0];
22 extern unsigned int ext4_count_free(char *bitmap, unsigned numchars);
23 void ext4_inode_bitmap_csum_set(struct super_block *sb, ext4_group_t group,
24 @@ -2582,6 +2593,7 @@ extern void ext4_set_inode_flags(struct inode *);
25 extern void ext4_get_inode_flags(struct ext4_inode_info *);
26 extern int ext4_alloc_da_blocks(struct inode *inode);
27 extern void ext4_set_aops(struct inode *inode);
28 +extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk);
29 extern int ext4_writepage_trans_blocks(struct inode *);
30 extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
31 extern int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
32 diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
33 index 0cccda3..43ca376 100644
34 --- a/fs/ext4/ialloc.c
35 +++ b/fs/ext4/ialloc.c
36 @@ -293,7 +293,6 @@ void ext4_free_inode(handle_t *handle, struct inode *inode)
37 * as writing the quota to disk may need the lock as well.
39 dquot_initialize(inode);
40 - ext4_xattr_delete_inode(handle, inode);
41 dquot_free_inode(inode);
44 diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
45 index 53e8a88..5112c5a 100644
46 --- a/fs/ext4/inline.c
47 +++ b/fs/ext4/inline.c
48 @@ -61,7 +61,7 @@ static int get_max_inline_xattr_value_size(struct inode *inode,
50 /* Compute min_offs. */
51 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
52 - if (!entry->e_value_block && entry->e_value_size) {
53 + if (!entry->e_value_inum && entry->e_value_size) {
54 size_t offs = le16_to_cpu(entry->e_value_offs);
57 diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
58 index 04c5f63..3034ceb 100644
61 @@ -139,8 +139,6 @@ static void ext4_invalidatepage(struct page *page, unsigned int offset,
63 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
64 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
65 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
69 * Test whether an inode is a fast symlink.
70 @@ -189,6 +187,8 @@ void ext4_evict_inode(struct inode *inode)
74 + int extra_credits = 3;
75 + struct ext4_xattr_ino_array *lea_ino_array = NULL;
77 trace_ext4_evict_inode(inode);
79 @@ -241,8 +241,8 @@ void ext4_evict_inode(struct inode *inode)
80 * protection against it
82 sb_start_intwrite(inode->i_sb);
83 - handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
84 - ext4_blocks_for_truncate(inode)+3);
86 + handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits);
88 ext4_std_error(inode->i_sb, PTR_ERR(handle));
90 @@ -254,9 +254,36 @@ void ext4_evict_inode(struct inode *inode)
91 sb_end_intwrite(inode->i_sb);
96 ext4_handle_sync(handle);
99 + * Delete xattr inode before deleting the main inode.
101 + err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
103 + ext4_warning(inode->i_sb,
104 + "couldn't delete inode's xattr (err %d)", err);
108 + if (!IS_NOQUOTA(inode))
109 + extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
111 + if (!ext4_handle_has_enough_credits(handle,
112 + ext4_blocks_for_truncate(inode) + extra_credits)) {
113 + err = ext4_journal_extend(handle,
114 + ext4_blocks_for_truncate(inode) + extra_credits);
116 + err = ext4_journal_restart(handle,
117 + ext4_blocks_for_truncate(inode) + extra_credits);
119 + ext4_warning(inode->i_sb,
120 + "couldn't extend journal (err %d)", err);
126 err = ext4_mark_inode_dirty(handle, inode);
128 @@ -273,10 +300,10 @@ void ext4_evict_inode(struct inode *inode)
129 * enough credits left in the handle to remove the inode from
130 * the orphan list and set the dtime field.
132 - if (!ext4_handle_has_enough_credits(handle, 3)) {
133 - err = ext4_journal_extend(handle, 3);
134 + if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
135 + err = ext4_journal_extend(handle, extra_credits);
137 - err = ext4_journal_restart(handle, 3);
138 + err = ext4_journal_restart(handle, extra_credits);
140 ext4_warning(inode->i_sb,
141 "couldn't extend journal (err %d)", err);
142 @@ -311,8 +338,12 @@ void ext4_evict_inode(struct inode *inode)
143 ext4_clear_inode(inode);
145 ext4_free_inode(handle, inode);
147 ext4_journal_stop(handle);
148 sb_end_intwrite(inode->i_sb);
150 + if (lea_ino_array != NULL)
151 + ext4_xattr_inode_array_free(inode, lea_ino_array);
154 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
155 @@ -5008,7 +5039,7 @@ static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
157 * Also account for superblock, inode, quota and xattr blocks
159 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
160 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
163 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
164 diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
165 index 263002f..b69145b 100644
166 --- a/fs/ext4/xattr.c
167 +++ b/fs/ext4/xattr.c
168 @@ -202,6 +202,7 @@ ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
170 while (!IS_LAST_ENTRY(entry)) {
171 if (entry->e_value_size != 0 &&
172 + entry->e_value_inum == 0 &&
173 (value_start + le16_to_cpu(entry->e_value_offs) <
174 (void *)e + sizeof(__u32) ||
175 value_start + le16_to_cpu(entry->e_value_offs) +
176 @@ -234,19 +235,26 @@ ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
180 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
181 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
182 + struct inode *inode)
184 size_t value_size = le32_to_cpu(entry->e_value_size);
186 - if (entry->e_value_block != 0 || value_size > size ||
187 + if (!entry->e_value_inum &&
188 le16_to_cpu(entry->e_value_offs) + value_size > size)
189 return -EFSCORRUPTED;
190 + if (entry->e_value_inum &&
191 + (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
192 + le32_to_cpu(entry->e_value_inum) >
193 + le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
194 + return -EFSCORRUPTED;
199 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
200 - const char *name, size_t size, int sorted)
201 + const char *name, size_t size, int sorted,
202 + struct inode *inode)
204 struct ext4_xattr_entry *entry;
206 @@ -266,11 +274,104 @@ ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
210 - if (!cmp && ext4_xattr_check_entry(entry, size))
211 + if (!cmp && ext4_xattr_check_entry(entry, size, inode))
212 return -EFSCORRUPTED;
213 return cmp ? -ENODATA : 0;
217 + * Read the EA value from an inode.
220 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
222 + unsigned long block = 0;
223 + struct buffer_head *bh = NULL;
225 + size_t csize, ret_size = 0;
230 + blocksize = ea_inode->i_sb->s_blocksize;
232 + while (ret_size < *size) {
233 + csize = (*size - ret_size) > blocksize ? blocksize :
235 + bh = ext4_bread(NULL, ea_inode, block, 0);
238 + return PTR_ERR(bh);
240 + memcpy(buf, bh->b_data, csize);
253 +struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
255 + struct inode *ea_inode = NULL;
257 + ea_inode = ext4_iget(parent->i_sb, ea_ino);
258 + if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
259 + int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
260 + ext4_error(parent->i_sb, "error while reading EA inode %lu "
261 + "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
262 + *err = rc != 0 ? rc : -EIO;
266 + if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
267 + ea_inode->i_generation != parent->i_generation) {
268 + ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
269 + "to parent invalid.", ea_ino);
274 + if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
275 + ext4_error(parent->i_sb, "EA inode %lu does not have "
276 + "EXT4_EA_INODE_FL flag set.\n", ea_ino);
290 + * Read the value from the EA inode.
293 +ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
296 + struct inode *ea_inode = NULL;
299 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
303 + err = ext4_xattr_inode_read(ea_inode, buffer, size);
310 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
311 void *buffer, size_t buffer_size)
312 @@ -303,7 +404,8 @@ bad_block:
314 ext4_xattr_cache_insert(ext4_mb_cache, bh);
316 - error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
317 + error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
319 if (error == -EFSCORRUPTED)
322 @@ -313,8 +415,16 @@ bad_block:
324 if (size > buffer_size)
326 - memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
328 + if (entry->e_value_inum) {
329 + error = ext4_xattr_inode_get(inode,
330 + le32_to_cpu(entry->e_value_inum),
335 + memcpy(buffer, bh->b_data +
336 + le16_to_cpu(entry->e_value_offs), size);
341 @@ -348,7 +458,7 @@ ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
344 error = ext4_xattr_find_entry(&entry, name_index, name,
345 - end - (void *)entry, 0);
346 + end - (void *)entry, 0, inode);
349 size = le32_to_cpu(entry->e_value_size);
350 @@ -356,8 +466,16 @@ ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
352 if (size > buffer_size)
354 - memcpy(buffer, (void *)IFIRST(header) +
355 - le16_to_cpu(entry->e_value_offs), size);
356 + if (entry->e_value_inum) {
357 + error = ext4_xattr_inode_get(inode,
358 + le32_to_cpu(entry->e_value_inum),
363 + memcpy(buffer, (void *)IFIRST(header) +
364 + le16_to_cpu(entry->e_value_offs), size);
369 @@ -603,7 +721,7 @@ static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
370 size_t *min_offs, void *base, int *total)
372 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
373 - if (!last->e_value_block && last->e_value_size) {
374 + if (!last->e_value_inum && last->e_value_size) {
375 size_t offs = le16_to_cpu(last->e_value_offs);
376 if (offs < *min_offs)
378 @@ -614,16 +732,176 @@ static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
379 return (*min_offs - ((void *)last - base) - sizeof(__u32));
383 + * Write the value of the EA in an inode.
386 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
387 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
388 + const void *buf, int bufsize)
390 + struct buffer_head *bh = NULL;
391 + unsigned long block = 0;
392 + unsigned blocksize = ea_inode->i_sb->s_blocksize;
393 + unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
394 + int csize, wsize = 0;
399 + while (ret >= 0 && ret < max_blocks) {
400 + struct ext4_map_blocks map;
401 + map.m_lblk = block += ret;
402 + map.m_len = max_blocks -= ret;
404 + ret = ext4_map_blocks(handle, ea_inode, &map,
405 + EXT4_GET_BLOCKS_CREATE);
407 + ext4_mark_inode_dirty(handle, ea_inode);
408 + if (ret == -ENOSPC &&
409 + ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
421 + while (wsize < bufsize) {
424 + csize = (bufsize - wsize) > blocksize ? blocksize :
426 + bh = ext4_getblk(handle, ea_inode, block, 0);
431 + ret = ext4_journal_get_write_access(handle, bh);
435 + memcpy(bh->b_data, buf, csize);
436 + set_buffer_uptodate(bh);
437 + ext4_handle_dirty_metadata(handle, ea_inode, bh);
444 + mutex_lock(&ea_inode->i_mutex);
445 + i_size_write(ea_inode, wsize);
446 + ext4_update_i_disksize(ea_inode, wsize);
447 + mutex_unlock(&ea_inode->i_mutex);
449 + ext4_mark_inode_dirty(handle, ea_inode);
458 + * Create an inode to store the value of a large EA.
460 +static struct inode *
461 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
463 + struct inode *ea_inode = NULL;
466 + * Let the next inode be the goal, so we try and allocate the EA inode
467 + * in the same group, or nearby one.
469 + ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
470 + S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
472 + if (!IS_ERR(ea_inode)) {
473 + ea_inode->i_op = &ext4_file_inode_operations;
474 + ea_inode->i_fop = &ext4_file_operations;
475 + ext4_set_aops(ea_inode);
476 + ea_inode->i_generation = inode->i_generation;
477 + EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
480 + * A back-pointer from EA inode to parent inode will be useful
483 + EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
484 + unlock_new_inode(ea_inode);
491 + * Unlink the inode storing the value of the EA.
494 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
496 + struct inode *ea_inode = NULL;
499 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
503 + clear_nlink(ea_inode);
510 + * Add value of the EA in an inode.
513 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
514 + const void *value, size_t value_len)
516 + struct inode *ea_inode = NULL;
519 + /* Create an inode for the EA value */
520 + ea_inode = ext4_xattr_inode_create(handle, inode);
521 + if (IS_ERR(ea_inode))
524 + err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
526 + clear_nlink(ea_inode);
528 + *ea_ino = ea_inode->i_ino;
536 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
537 + handle_t *handle, struct inode *inode)
539 struct ext4_xattr_entry *last;
540 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
541 + int in_inode = i->in_inode;
543 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
544 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
545 + (EXT4_XATTR_SIZE(i->value_len) >
546 + EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
549 /* Compute min_offs and last. */
551 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
552 - if (!last->e_value_block && last->e_value_size) {
553 + if (!last->e_value_inum && last->e_value_size) {
554 size_t offs = le16_to_cpu(last->e_value_offs);
557 @@ -631,15 +909,20 @@ ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
559 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
561 - if (!s->here->e_value_block && s->here->e_value_size) {
563 + !s->here->e_value_inum && s->here->e_value_size) {
564 size_t size = le32_to_cpu(s->here->e_value_size);
565 free += EXT4_XATTR_SIZE(size);
567 free += EXT4_XATTR_LEN(name_len);
570 - if (free < EXT4_XATTR_LEN(name_len) +
571 - EXT4_XATTR_SIZE(i->value_len))
572 + size_t value_len = EXT4_XATTR_SIZE(i->value_len);
577 + if (free < EXT4_XATTR_LEN(name_len) + value_len)
581 @@ -653,7 +936,8 @@ ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
582 s->here->e_name_len = name_len;
583 memcpy(s->here->e_name, i->name, name_len);
585 - if (!s->here->e_value_block && s->here->e_value_size) {
586 + if (!s->here->e_value_inum && s->here->e_value_size &&
587 + s->here->e_value_offs > 0) {
588 void *first_val = s->base + min_offs;
589 size_t offs = le16_to_cpu(s->here->e_value_offs);
590 void *val = s->base + offs;
591 @@ -687,13 +971,18 @@ ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
593 while (!IS_LAST_ENTRY(last)) {
594 size_t o = le16_to_cpu(last->e_value_offs);
595 - if (!last->e_value_block &&
596 + if (!last->e_value_inum &&
597 last->e_value_size && o < offs)
599 cpu_to_le16(o + size);
600 last = EXT4_XATTR_NEXT(last);
603 + if (s->here->e_value_inum) {
604 + ext4_xattr_inode_unlink(inode,
605 + le32_to_cpu(s->here->e_value_inum));
606 + s->here->e_value_inum = 0;
609 /* Remove the old name. */
610 size_t size = EXT4_XATTR_LEN(name_len);
611 @@ -707,10 +996,17 @@ ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
613 /* Insert the new value. */
614 s->here->e_value_size = cpu_to_le32(i->value_len);
615 - if (i->value_len) {
617 + unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
618 + ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
620 + s->here->e_value_inum = cpu_to_le32(ea_ino);
621 + s->here->e_value_offs = 0;
622 + } else if (i->value_len) {
623 size_t size = EXT4_XATTR_SIZE(i->value_len);
624 void *val = s->base + min_offs - size;
625 s->here->e_value_offs = cpu_to_le16(min_offs - size);
626 + s->here->e_value_inum = 0;
627 if (i->value == EXT4_ZERO_XATTR_VALUE) {
628 memset(val, 0, size);
630 @@ -760,7 +1056,7 @@ ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
631 bs->s.end = bs->bh->b_data + bs->bh->b_size;
632 bs->s.here = bs->s.first;
633 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
634 - i->name, bs->bh->b_size, 1);
635 + i->name, bs->bh->b_size, 1, inode);
636 if (error && error != -ENODATA)
638 bs->s.not_found = error;
639 @@ -785,8 +1081,6 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
641 #define header(x) ((struct ext4_xattr_header *)(x))
643 - if (i->value && i->value_len > sb->s_blocksize)
646 ce = mb_cache_entry_get(ext4_mb_cache, bs->bh->b_bdev,
648 @@ -802,7 +1096,7 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
651 ea_bdebug(bs->bh, "modifying in-place");
652 - error = ext4_xattr_set_entry(i, s);
653 + error = ext4_xattr_set_entry(i, s, handle, inode);
655 if (!IS_LAST_ENTRY(s->first))
656 ext4_xattr_rehash(header(s->base),
657 @@ -854,7 +1148,7 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
658 s->end = s->base + sb->s_blocksize;
661 - error = ext4_xattr_set_entry(i, s);
662 + error = ext4_xattr_set_entry(i, s, handle, inode);
663 if (error == -EFSCORRUPTED)
666 @@ -998,7 +1292,7 @@ int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
667 /* Find the named attribute. */
668 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
670 - (void *)is->s.base, 0);
671 + (void *)is->s.base, 0, inode);
672 if (error && error != -ENODATA)
674 is->s.not_found = error;
675 @@ -1016,7 +1310,7 @@ int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
677 if (EXT4_I(inode)->i_extra_isize == 0)
679 - error = ext4_xattr_set_entry(i, s);
680 + error = ext4_xattr_set_entry(i, s, handle, inode);
682 if (error == -ENOSPC &&
683 ext4_has_inline_data(inode)) {
684 @@ -1028,7 +1322,7 @@ int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
685 error = ext4_xattr_ibody_find(inode, i, is);
688 - error = ext4_xattr_set_entry(i, s);
689 + error = ext4_xattr_set_entry(i, s, handle, inode);
693 @@ -1054,7 +1348,7 @@ static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
695 if (EXT4_I(inode)->i_extra_isize == 0)
697 - error = ext4_xattr_set_entry(i, s);
698 + error = ext4_xattr_set_entry(i, s, handle, inode);
701 header = IHDR(inode, ext4_raw_inode(&is->iloc));
702 @@ -1090,7 +1384,7 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
705 .value_len = value_len,
709 struct ext4_xattr_ibody_find is = {
710 .s = { .not_found = -ENODATA, },
711 @@ -1155,6 +1449,15 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
714 error = ext4_xattr_block_set(handle, inode, &i, &bs);
715 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
716 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
717 + error == -ENOSPC) {
718 + /* xattr not fit to block, store at external
721 + error = ext4_xattr_ibody_set(handle, inode,
726 if (!is.s.not_found) {
727 @@ -1201,9 +1504,22 @@ ext4_xattr_set(struct inode *inode, int name_index, const char *name,
728 const void *value, size_t value_len, int flags)
731 + struct super_block *sb = inode->i_sb;
732 int error, retries = 0;
733 int credits = ext4_jbd2_credits_xattr(inode);
735 + if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
736 + EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
737 + int nrblocks = (value_len + sb->s_blocksize - 1) >>
738 + sb->s_blocksize_bits;
740 + /* For new inode */
741 + credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
743 + /* For data blocks of EA inode */
744 + credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
748 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
749 if (IS_ERR(handle)) {
750 @@ -1215,7 +1531,7 @@ retry:
751 value, value_len, flags);
752 error2 = ext4_journal_stop(handle);
753 if (error == -ENOSPC &&
754 - ext4_should_retry_alloc(inode->i_sb, &retries))
755 + ext4_should_retry_alloc(sb, &retries))
759 @@ -1237,7 +1553,7 @@ static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
761 /* Adjust the value offsets of the entries */
762 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
763 - if (!last->e_value_block && last->e_value_size) {
764 + if (!last->e_value_inum && last->e_value_size) {
765 new_offs = le16_to_cpu(last->e_value_offs) +
767 BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
768 @@ -1484,21 +1800,135 @@ cleanup:
772 +#define EIA_INCR 16 /* must be 2^n */
773 +#define EIA_MASK (EIA_INCR - 1)
774 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
775 + * If @lea_ino_array is new or full it will be grown and the old
776 + * contents copied over.
779 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
781 + if (*lea_ino_array == NULL) {
783 + * Start with 15 inodes, so it fits into a power-of-two size.
784 + * If *lea_ino_array is NULL, this is essentially offsetof()
787 + kmalloc(offsetof(struct ext4_xattr_ino_array,
788 + xia_inodes[EIA_MASK]),
790 + if (*lea_ino_array == NULL)
792 + (*lea_ino_array)->xia_count = 0;
793 + } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
794 + /* expand the array once all 15 + n * 16 slots are full */
795 + struct ext4_xattr_ino_array *new_array = NULL;
796 + int count = (*lea_ino_array)->xia_count;
798 + /* if new_array is NULL, this is essentially offsetof() */
799 + new_array = kmalloc(
800 + offsetof(struct ext4_xattr_ino_array,
801 + xia_inodes[count + EIA_INCR]),
803 + if (new_array == NULL)
805 + memcpy(new_array, *lea_ino_array,
806 + offsetof(struct ext4_xattr_ino_array,
807 + xia_inodes[count]));
808 + kfree(*lea_ino_array);
809 + *lea_ino_array = new_array;
811 + (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
816 + * Add xattr inode to orphan list
819 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
820 + int credits, struct ext4_xattr_ino_array *lea_ino_array)
822 + struct inode *ea_inode = NULL;
823 + int idx = 0, error = 0;
825 + if (lea_ino_array == NULL)
828 + for (; idx < lea_ino_array->xia_count; ++idx) {
829 + if (!ext4_handle_has_enough_credits(handle, credits)) {
830 + error = ext4_journal_extend(handle, credits);
832 + error = ext4_journal_restart(handle, credits);
835 + ext4_warning(inode->i_sb,
836 + "couldn't extend journal "
837 + "(err %d)", error);
841 + ea_inode = ext4_xattr_inode_iget(inode,
842 + lea_ino_array->xia_inodes[idx], &error);
845 + ext4_orphan_add(handle, ea_inode);
846 + /* the inode's i_count will be released by caller */
853 * ext4_xattr_delete_inode()
855 - * Free extended attribute resources associated with this inode. This
856 + * Free extended attribute resources associated with this inode. Traverse
857 + * all entries and unlink any xattr inodes associated with this inode. This
858 * is called immediately before an inode is freed. We have exclusive
859 - * access to the inode.
860 + * access to the inode. If an orphan inode is deleted it will also delete any
861 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
862 + * to ensure they belong to the parent inode and were not deleted already.
865 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
867 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
868 + struct ext4_xattr_ino_array **lea_ino_array)
870 struct buffer_head *bh = NULL;
871 + struct ext4_xattr_ibody_header *header;
872 + struct ext4_inode *raw_inode;
873 + struct ext4_iloc iloc;
874 + struct ext4_xattr_entry *entry;
875 + int credits = 3, error = 0;
877 - if (!EXT4_I(inode)->i_file_acl)
878 + if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
879 + goto delete_external_ea;
881 + error = ext4_get_inode_loc(inode, &iloc);
884 + raw_inode = ext4_raw_inode(&iloc);
885 + header = IHDR(inode, raw_inode);
886 + for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
887 + entry = EXT4_XATTR_NEXT(entry)) {
888 + if (!entry->e_value_inum)
890 + if (ext4_expand_ino_array(lea_ino_array,
891 + entry->e_value_inum) != 0) {
895 + entry->e_value_inum = 0;
900 + if (!EXT4_I(inode)->i_file_acl) {
901 + /* add xattr inode to orphan list */
902 + ext4_xattr_inode_orphan_add(handle, inode, credits,
906 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
908 EXT4_ERROR_INODE(inode, "block %llu read error",
909 @@ -1511,11 +1941,69 @@ ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
910 EXT4_I(inode)->i_file_acl);
914 + for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
915 + entry = EXT4_XATTR_NEXT(entry)) {
916 + if (!entry->e_value_inum)
918 + if (ext4_expand_ino_array(lea_ino_array,
919 + entry->e_value_inum) != 0)
921 + entry->e_value_inum = 0;
924 + /* add xattr inode to orphan list */
925 + error = ext4_xattr_inode_orphan_add(handle, inode, credits,
930 + if (!IS_NOQUOTA(inode))
931 + credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
933 + if (!ext4_handle_has_enough_credits(handle, credits)) {
934 + error = ext4_journal_extend(handle, credits);
936 + error = ext4_journal_restart(handle, credits);
938 + ext4_warning(inode->i_sb,
939 + "couldn't extend journal (err %d)", error);
944 ext4_xattr_release_block(handle, inode, bh);
945 EXT4_I(inode)->i_file_acl = 0;
954 +ext4_xattr_inode_array_free(struct inode *inode,
955 + struct ext4_xattr_ino_array *lea_ino_array)
957 + struct inode *ea_inode = NULL;
961 + if (lea_ino_array == NULL)
964 + for (; idx < lea_ino_array->xia_count; ++idx) {
965 + ea_inode = ext4_xattr_inode_iget(inode,
966 + lea_ino_array->xia_inodes[idx], &err);
969 + /* for inode's i_count get from ext4_xattr_delete_inode */
970 + if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
972 + clear_nlink(ea_inode);
975 + kfree(lea_ino_array);
979 @@ -1585,10 +2073,9 @@ ext4_xattr_cmp(struct ext4_xattr_header *header1,
980 entry1->e_name_index != entry2->e_name_index ||
981 entry1->e_name_len != entry2->e_name_len ||
982 entry1->e_value_size != entry2->e_value_size ||
983 + entry1->e_value_inum != entry2->e_value_inum ||
984 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
986 - if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
987 - return -EFSCORRUPTED;
988 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
989 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
990 le32_to_cpu(entry1->e_value_size)))
991 @@ -1673,7 +2160,7 @@ static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
995 - if (entry->e_value_block == 0 && entry->e_value_size != 0) {
996 + if (!entry->e_value_inum && entry->e_value_size) {
997 __le32 *value = (__le32 *)((char *)header +
998 le16_to_cpu(entry->e_value_offs));
999 for (n = (le32_to_cpu(entry->e_value_size) +
1000 diff --git a/fs/ext4/xattr.h b/fs/ext4/xattr.h
1001 index ddc0957..57c7ad5 100644
1002 --- a/fs/ext4/xattr.h
1003 +++ b/fs/ext4/xattr.h
1004 @@ -43,7 +43,7 @@ struct ext4_xattr_entry {
1005 __u8 e_name_len; /* length of name */
1006 __u8 e_name_index; /* attribute name index */
1007 __le16 e_value_offs; /* offset in disk block of value */
1008 - __le32 e_value_block; /* disk block attribute is stored on (n/i) */
1009 + __le32 e_value_inum; /* inode in which the value is stored */
1010 __le32 e_value_size; /* size of attribute value */
1011 __le32 e_hash; /* hash value of name and value */
1012 char e_name[0]; /* attribute name */
1013 @@ -68,6 +68,26 @@ struct ext4_xattr_entry {
1014 EXT4_I(inode)->i_extra_isize))
1015 #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1018 + * Link EA inode back to parent one using i_mtime field.
1019 + * Extra integer type conversion added to ignore higher
1020 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1022 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum) \
1024 + (inode)->i_mtime.tv_sec = inum; \
1027 +#define EXT4_XATTR_INODE_GET_PARENT(inode) \
1028 +((__u32)(inode)->i_mtime.tv_sec)
1031 + * The minimum size of EA value when you start storing it in an external inode
1032 + * size of block - size of header - size of 1 entry - 4 null bytes
1034 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b) \
1035 + ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1037 #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1038 #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1039 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1040 @@ -76,10 +96,11 @@ struct ext4_xattr_entry {
1041 #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1043 struct ext4_xattr_info {
1052 struct ext4_xattr_search {
1053 @@ -107,7 +128,14 @@ extern int ext4_xattr_get(struct inode *, int, const char *, void *, size_t);
1054 extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1055 extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1057 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1058 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1060 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1061 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1062 + struct ext4_xattr_ino_array **array);
1063 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1064 + struct ext4_xattr_ino_array *array);
1066 extern void ext4_xattr_put_super(struct super_block *);
1068 extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,