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 @@ -1620,6 +1620,7 @@ 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 @@ -2033,6 +2034,10 @@ struct mmpd_data {
20 # define ATTRIB_NORET __attribute__((noreturn))
21 # define NORET_AND noreturn,
23 +struct ext4_xattr_ino_array {
24 + unsigned int xia_count; /* # of used item in the array */
25 + unsigned int xia_inodes[0];
28 extern unsigned int ext4_count_free(char *bitmap, unsigned numchars);
29 void ext4_inode_bitmap_csum_set(struct super_block *sb, ext4_group_t group,
30 @@ -2238,6 +2243,7 @@ extern void ext4_set_inode_flags(struct
31 extern void ext4_get_inode_flags(struct ext4_inode_info *);
32 extern int ext4_alloc_da_blocks(struct inode *inode);
33 extern void ext4_set_aops(struct inode *inode);
34 +extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk);
35 extern int ext4_writepage_trans_blocks(struct inode *);
36 extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
37 extern int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
38 Index: linux-stage/fs/ext4/inode.c
39 ===================================================================
40 --- linux-stage.orig/fs/ext4/inode.c
41 +++ linux-stage/fs/ext4/inode.c
42 @@ -136,8 +136,6 @@ static void ext4_invalidatepage(struct p
44 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
45 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
46 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
50 * Test whether an inode is a fast symlink.
51 @@ -186,6 +184,8 @@ void ext4_evict_inode(struct inode *inod
55 + int extra_credits = 3;
56 + struct ext4_xattr_ino_array *lea_ino_array = NULL;
58 trace_ext4_evict_inode(inode);
60 @@ -235,8 +235,8 @@ void ext4_evict_inode(struct inode *inod
61 * protection against it
63 sb_start_intwrite(inode->i_sb);
64 - handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
65 - ext4_blocks_for_truncate(inode)+3);
67 + handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits);
69 ext4_std_error(inode->i_sb, PTR_ERR(handle));
71 @@ -251,6 +251,32 @@ void ext4_evict_inode(struct inode *inod
74 ext4_handle_sync(handle);
76 + /* Delete xattr inode before deleting the main inode. */
77 + err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
79 + ext4_warning(inode->i_sb,
80 + "couldn't delete inode's xattr (err %d)", err);
84 + if (!IS_NOQUOTA(inode))
85 + extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
87 + if (!ext4_handle_has_enough_credits(handle,
88 + ext4_blocks_for_truncate(inode) + extra_credits)) {
89 + err = ext4_journal_extend(handle,
90 + ext4_blocks_for_truncate(inode) + extra_credits);
92 + err = ext4_journal_restart(handle,
93 + ext4_blocks_for_truncate(inode) + extra_credits);
95 + ext4_warning(inode->i_sb,
96 + "couldn't extend journal (err %d)", err);
102 err = ext4_mark_inode_dirty(handle, inode);
104 @@ -267,10 +293,10 @@ void ext4_evict_inode(struct inode *inod
105 * enough credits left in the handle to remove the inode from
106 * the orphan list and set the dtime field.
108 - if (!ext4_handle_has_enough_credits(handle, 3)) {
109 - err = ext4_journal_extend(handle, 3);
110 + if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
111 + err = ext4_journal_extend(handle, extra_credits);
113 - err = ext4_journal_restart(handle, 3);
114 + err = ext4_journal_restart(handle, extra_credits);
116 ext4_warning(inode->i_sb,
117 "couldn't extend journal (err %d)", err);
118 @@ -307,6 +333,9 @@ void ext4_evict_inode(struct inode *inod
119 ext4_free_inode(handle, inode);
120 ext4_journal_stop(handle);
121 sb_end_intwrite(inode->i_sb);
123 + if (lea_ino_array != NULL)
124 + ext4_xattr_inode_array_free(inode, lea_ino_array);
127 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
128 @@ -5140,7 +5169,7 @@ static int ext4_index_trans_blocks(struc
130 * Also account for superblock, inode, quota and xattr blocks
132 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
133 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
136 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
137 Index: linux-stage/fs/ext4/xattr.c
138 ===================================================================
139 --- linux-stage.orig/fs/ext4/xattr.c
140 +++ linux-stage/fs/ext4/xattr.c
141 @@ -195,9 +195,11 @@ ext4_xattr_check_names(struct ext4_xattr
142 /* Find the end of the names list */
143 while (!IS_LAST_ENTRY(e)) {
144 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
145 - if ((void *)next >= end)
146 + if ((void *)next >= end &&
147 + entry->e_value_inum == 0)
148 return -EFSCORRUPTED;
149 - if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
150 + if (strnlen(e->e_name, e->e_name_len) != e->e_name_len &&
151 + entry->e_value_inum == 0)
152 return -EFSCORRUPTED;
155 @@ -205,13 +207,12 @@ ext4_xattr_check_names(struct ext4_xattr
156 /* Check the values */
157 while (!IS_LAST_ENTRY(entry)) {
158 u32 size = le32_to_cpu(entry->e_value_size);
159 - if (entry->e_value_block != 0)
160 - return -EFSCORRUPTED;
162 - if (size > INT_MAX)
163 + if (size > INT_MAX && entry->e_value_inum == 0)
164 return -EFSCORRUPTED;
168 + entry->e_value_inum == 0) {
169 u16 offs = le16_to_cpu(entry->e_value_offs);
172 @@ -256,19 +257,26 @@ ext4_xattr_check_block(struct inode *ino
176 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
177 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
178 + struct inode *inode)
180 size_t value_size = le32_to_cpu(entry->e_value_size);
182 - if (entry->e_value_block != 0 || value_size > size ||
183 + if (!entry->e_value_inum &&
184 le16_to_cpu(entry->e_value_offs) + value_size > size)
186 + if (entry->e_value_inum &&
187 + (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
188 + le32_to_cpu(entry->e_value_inum) >
189 + le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
195 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
196 - const char *name, size_t size, int sorted)
197 + const char *name, size_t size, int sorted,
198 + struct inode *inode)
200 struct ext4_xattr_entry *entry;
202 @@ -288,11 +296,109 @@ ext4_xattr_find_entry(struct ext4_xattr_
206 - if (!cmp && ext4_xattr_check_entry(entry, size))
207 + if (!cmp && ext4_xattr_check_entry(entry, size, inode))
209 return cmp ? -ENODATA : 0;
213 + * Read the EA value from an inode.
215 +static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
217 + unsigned long block = 0;
218 + struct buffer_head *bh = NULL;
219 + int err, blocksize;
220 + size_t csize, ret_size = 0;
225 + blocksize = ea_inode->i_sb->s_blocksize;
227 + while (ret_size < *size) {
228 + csize = (*size - ret_size) > blocksize ? blocksize :
230 + bh = ext4_bread(NULL, ea_inode, block, 0, &err);
235 + memcpy(buf, bh->b_data, csize);
249 + * Fetch the xattr inode from disk.
251 + * The xattr inode stores the parent inode number and generation so that
252 + * the kernel and e2fsck can verify the xattr inode is valid upon access.
254 +struct inode *ext4_xattr_inode_iget(struct inode *parent,
255 + unsigned long ea_ino, int *err)
257 + struct inode *ea_inode = NULL;
259 + ea_inode = ext4_iget(parent->i_sb, ea_ino);
260 + if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
261 + int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
262 + ext4_error(parent->i_sb, "error while reading EA inode %lu "
263 + "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
264 + *err = rc != 0 ? rc : -EIO;
268 + if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
269 + ea_inode->i_generation != parent->i_generation) {
270 + ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
271 + "to parent invalid.", ea_ino);
276 + if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
277 + ext4_error(parent->i_sb, "EA inode %lu does not have "
278 + "EXT4_EA_INODE_FL flag set.\n", ea_ino);
292 + * Read the value from the EA inode.
294 +static int ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino,
295 + void *buffer, size_t *size)
297 + struct inode *ea_inode = NULL;
300 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
304 + err = ext4_xattr_inode_read(ea_inode, buffer, size);
311 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
312 void *buffer, size_t buffer_size)
313 @@ -324,7 +430,8 @@ bad_block:
315 ext4_xattr_cache_insert(bh);
317 - error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
318 + error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
323 @@ -334,8 +441,16 @@ bad_block:
325 if (size > buffer_size)
327 - memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
329 + if (entry->e_value_inum) {
330 + error = ext4_xattr_inode_get(inode,
331 + le32_to_cpu(entry->e_value_inum),
336 + memcpy(buffer, bh->b_data +
337 + le16_to_cpu(entry->e_value_offs), size);
342 @@ -369,7 +484,7 @@ ext4_xattr_ibody_get(struct inode *inode
345 error = ext4_xattr_find_entry(&entry, name_index, name,
346 - end - (void *)entry, 0);
347 + end - (void *)entry, 0, inode);
350 size = le32_to_cpu(entry->e_value_size);
351 @@ -377,8 +492,16 @@ ext4_xattr_ibody_get(struct inode *inode
353 if (size > buffer_size)
355 - memcpy(buffer, (void *)IFIRST(header) +
356 - le16_to_cpu(entry->e_value_offs), size);
357 + if (entry->e_value_inum) {
358 + error = ext4_xattr_inode_get(inode,
359 + le32_to_cpu(entry->e_value_inum),
364 + memcpy(buffer, (void *)IFIRST(header) +
365 + le16_to_cpu(entry->e_value_offs), size);
370 @@ -623,7 +746,7 @@ static size_t ext4_xattr_free_space(stru
371 size_t *min_offs, void *base, int *total)
373 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
374 - if (!last->e_value_block && last->e_value_size) {
375 + if (!last->e_value_inum && last->e_value_size) {
376 size_t offs = le16_to_cpu(last->e_value_offs);
377 if (offs < *min_offs)
379 @@ -634,11 +757,193 @@ static size_t ext4_xattr_free_space(stru
380 return (*min_offs - ((void *)last - base) - sizeof(__u32));
384 + * Write the value of the EA in an inode.
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, &ret);
429 + ret = ext4_journal_get_write_access(handle, bh);
433 + memcpy(bh->b_data, buf, csize);
434 + set_buffer_uptodate(bh);
435 + ext4_handle_dirty_metadata(handle, ea_inode, bh);
442 + mutex_lock(&ea_inode->i_mutex);
443 + i_size_write(ea_inode, wsize);
444 + ext4_update_i_disksize(ea_inode, wsize);
445 + mutex_unlock(&ea_inode->i_mutex);
447 + ext4_mark_inode_dirty(handle, ea_inode);
455 +static void ext4_xattr_inode_set_ref(struct inode *ea_inode, __u64 ref_count)
457 + ea_inode->i_ctime.tv_sec = (__u32)(ref_count >> 32);
458 + ea_inode->i_version = (__u32)ref_count;
461 +static void ext4_xattr_inode_set_hash(struct inode *ea_inode, __u32 hash)
463 + ea_inode->i_atime.tv_sec = hash;
467 + * Create an inode to store the value of a large EA.
469 +static struct inode *
470 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode, __u32 hash)
472 + struct inode *ea_inode = NULL;
475 + * Let the next inode be the goal, so we try and allocate the EA inode
476 + * in the same group, or nearby one.
478 + ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
479 + S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
481 + if (!IS_ERR(ea_inode)) {
482 + ea_inode->i_op = &ext4_file_inode_operations;
483 + ea_inode->i_fop = &ext4_file_operations.kabi_fops;
484 + ext4_set_aops(ea_inode);
485 + ea_inode->i_generation = inode->i_generation;
486 + EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
489 + * A back-pointer from EA inode to parent inode will be useful
492 + EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
493 + unlock_new_inode(ea_inode);
495 + ext4_xattr_inode_set_ref(ea_inode, 1);
496 + ext4_xattr_inode_set_hash(ea_inode, hash);
503 + * Unlink the inode storing the value of the EA.
506 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
508 + struct inode *ea_inode = NULL;
511 + ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
515 + clear_nlink(ea_inode);
522 +ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
524 + if (ext4_has_metadata_csum(sbi->s_sb))
525 + return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
530 + * Add value of the EA in an inode.
533 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
534 + const void *value, size_t value_len)
536 + struct inode *ea_inode = NULL;
540 + /* Create an inode for the EA value */
541 + hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
542 + ea_inode = ext4_xattr_inode_create(handle, inode, hash);
543 + if (IS_ERR(ea_inode))
546 + err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
548 + clear_nlink(ea_inode);
550 + *ea_ino = ea_inode->i_ino;
558 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
559 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
560 + handle_t *handle, struct inode *inode)
562 struct ext4_xattr_entry *last, *next;
563 size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
564 + int in_inode = i->in_inode;
566 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
567 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
568 + (EXT4_XATTR_SIZE(i->value_len) >
569 + EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
572 /* Compute min_offs and last. */
574 @@ -647,7 +952,7 @@ ext4_xattr_set_entry(struct ext4_xattr_i
575 if ((void *)next >= s->end) {
578 - if (!last->e_value_block && last->e_value_size) {
579 + if (!last->e_value_inum && last->e_value_size) {
580 size_t offs = le16_to_cpu(last->e_value_offs);
583 @@ -655,15 +960,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
585 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
587 - if (!s->here->e_value_block && s->here->e_value_size) {
589 + !s->here->e_value_inum && s->here->e_value_size) {
590 size_t size = le32_to_cpu(s->here->e_value_size);
591 free += EXT4_XATTR_SIZE(size);
593 free += EXT4_XATTR_LEN(name_len);
596 - if (free < EXT4_XATTR_LEN(name_len) +
597 - EXT4_XATTR_SIZE(i->value_len))
598 + size_t value_len = EXT4_XATTR_SIZE(i->value_len);
603 + if (free < value_len ||
604 + free < EXT4_XATTR_LEN(name_len) + value_len)
608 @@ -677,7 +988,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
609 s->here->e_name_len = name_len;
610 memcpy(s->here->e_name, i->name, name_len);
612 - if (!s->here->e_value_block && s->here->e_value_size) {
613 + if (!s->here->e_value_inum && s->here->e_value_size &&
614 + s->here->e_value_offs > 0) {
615 void *first_val = s->base + min_offs;
616 size_t offs = le16_to_cpu(s->here->e_value_offs);
617 void *val = s->base + offs;
618 @@ -711,13 +1023,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
620 while (!IS_LAST_ENTRY(last)) {
621 size_t o = le16_to_cpu(last->e_value_offs);
622 - if (!last->e_value_block &&
623 + if (!last->e_value_inum &&
624 last->e_value_size && o < offs)
626 cpu_to_le16(o + size);
627 last = EXT4_XATTR_NEXT(last);
630 + if (s->here->e_value_inum) {
631 + ext4_xattr_inode_unlink(inode,
632 + le32_to_cpu(s->here->e_value_inum));
633 + s->here->e_value_inum = 0;
636 /* Remove the old name. */
637 size_t size = EXT4_XATTR_LEN(name_len);
638 @@ -731,10 +1048,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
640 /* Insert the new value. */
641 s->here->e_value_size = cpu_to_le32(i->value_len);
642 - if (i->value_len) {
644 + unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
645 + ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
647 + s->here->e_value_inum = cpu_to_le32(ea_ino);
648 + s->here->e_value_offs = 0;
649 + } else if (i->value_len) {
650 size_t size = EXT4_XATTR_SIZE(i->value_len);
651 void *val = s->base + min_offs - size;
652 s->here->e_value_offs = cpu_to_le16(min_offs - size);
653 + s->here->e_value_inum = 0;
654 if (i->value == EXT4_ZERO_XATTR_VALUE) {
655 memset(val, 0, size);
657 @@ -784,7 +1108,7 @@ ext4_xattr_block_find(struct inode *inod
658 bs->s.end = bs->bh->b_data + bs->bh->b_size;
659 bs->s.here = bs->s.first;
660 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
661 - i->name, bs->bh->b_size, 1);
662 + i->name, bs->bh->b_size, 1, inode);
663 if (error && error != -ENODATA)
665 bs->s.not_found = error;
666 @@ -808,8 +1132,6 @@ ext4_xattr_block_set(handle_t *handle, s
668 #define header(x) ((struct ext4_xattr_header *)(x))
670 - if (i->value && i->value_len > sb->s_blocksize)
673 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
675 @@ -825,7 +1147,7 @@ ext4_xattr_block_set(handle_t *handle, s
678 ea_bdebug(bs->bh, "modifying in-place");
679 - error = ext4_xattr_set_entry(i, s);
680 + error = ext4_xattr_set_entry(i, s, handle, inode);
682 if (!IS_LAST_ENTRY(s->first))
683 ext4_xattr_rehash(header(s->base),
684 @@ -876,7 +1198,7 @@ ext4_xattr_block_set(handle_t *handle, s
685 s->end = s->base + sb->s_blocksize;
688 - error = ext4_xattr_set_entry(i, s);
689 + error = ext4_xattr_set_entry(i, s, handle, inode);
693 @@ -1020,7 +1342,7 @@ int ext4_xattr_ibody_find(struct inode *
694 /* Find the named attribute. */
695 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
697 - (void *)is->s.base, 0);
698 + (void *)is->s.base, 0, inode);
699 if (error && error != -ENODATA)
701 is->s.not_found = error;
702 @@ -1038,7 +1360,7 @@ int ext4_xattr_ibody_inline_set(handle_t
704 if (EXT4_I(inode)->i_extra_isize == 0)
706 - error = ext4_xattr_set_entry(i, s);
707 + error = ext4_xattr_set_entry(i, s, handle, inode);
710 header = IHDR(inode, ext4_raw_inode(&is->iloc));
711 @@ -1062,7 +1384,7 @@ static int ext4_xattr_ibody_set(handle_t
713 if (EXT4_I(inode)->i_extra_isize == 0)
715 - error = ext4_xattr_set_entry(i, s);
716 + error = ext4_xattr_set_entry(i, s, handle, inode);
719 header = IHDR(inode, ext4_raw_inode(&is->iloc));
720 @@ -1098,7 +1420,7 @@ ext4_xattr_set_handle(handle_t *handle,
723 .value_len = value_len,
727 struct ext4_xattr_ibody_find is = {
728 .s = { .not_found = -ENODATA, },
729 @@ -1165,6 +1487,15 @@ ext4_xattr_set_handle(handle_t *handle,
732 error = ext4_xattr_block_set(handle, inode, &i, &bs);
733 + if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
734 + EXT4_FEATURE_INCOMPAT_EA_INODE) &&
735 + error == -ENOSPC) {
736 + /* xattr not fit to block, store at external
739 + error = ext4_xattr_ibody_set(handle, inode,
744 if (!is.s.not_found) {
745 @@ -1211,9 +1542,22 @@ ext4_xattr_set(struct inode *inode, int
746 const void *value, size_t value_len, int flags)
749 + struct super_block *sb = inode->i_sb;
750 int error, retries = 0;
751 int credits = ext4_jbd2_credits_xattr(inode);
753 + if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
754 + EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
755 + int nrblocks = (value_len + sb->s_blocksize - 1) >>
756 + sb->s_blocksize_bits;
758 + /* For new inode */
759 + credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
761 + /* For data blocks of EA inode */
762 + credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
766 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
767 if (IS_ERR(handle)) {
768 @@ -1225,7 +1569,7 @@ retry:
769 value, value_len, flags);
770 error2 = ext4_journal_stop(handle);
771 if (error == -ENOSPC &&
772 - ext4_should_retry_alloc(inode->i_sb, &retries))
773 + ext4_should_retry_alloc(sb, &retries))
777 @@ -1247,7 +1591,7 @@ static void ext4_xattr_shift_entries(str
779 /* Adjust the value offsets of the entries */
780 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
781 - if (!last->e_value_block && last->e_value_size) {
782 + if (!last->e_value_inum && last->e_value_size) {
783 new_offs = le16_to_cpu(last->e_value_offs) +
785 BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
786 @@ -1494,21 +1838,135 @@ cleanup:
790 +#define EIA_INCR 16 /* must be 2^n */
791 +#define EIA_MASK (EIA_INCR - 1)
792 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
793 + * If @lea_ino_array is new or full it will be grown and the old
794 + * contents copied over.
797 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
799 + if (*lea_ino_array == NULL) {
801 + * Start with 15 inodes, so it fits into a power-of-two size.
802 + * If *lea_ino_array is NULL, this is essentially offsetof()
805 + kmalloc(offsetof(struct ext4_xattr_ino_array,
806 + xia_inodes[EIA_MASK]),
808 + if (*lea_ino_array == NULL)
810 + (*lea_ino_array)->xia_count = 0;
811 + } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
812 + /* expand the array once all 15 + n * 16 slots are full */
813 + struct ext4_xattr_ino_array *new_array = NULL;
814 + int count = (*lea_ino_array)->xia_count;
816 + /* if new_array is NULL, this is essentially offsetof() */
817 + new_array = kmalloc(
818 + offsetof(struct ext4_xattr_ino_array,
819 + xia_inodes[count + EIA_INCR]),
821 + if (new_array == NULL)
823 + memcpy(new_array, *lea_ino_array,
824 + offsetof(struct ext4_xattr_ino_array,
825 + xia_inodes[count]));
826 + kfree(*lea_ino_array);
827 + *lea_ino_array = new_array;
829 + (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
834 + * Add xattr inode to orphan list
837 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
838 + int credits, struct ext4_xattr_ino_array *lea_ino_array)
840 + struct inode *ea_inode = NULL;
841 + int idx = 0, error = 0;
843 + if (lea_ino_array == NULL)
846 + for (; idx < lea_ino_array->xia_count; ++idx) {
847 + if (!ext4_handle_has_enough_credits(handle, credits)) {
848 + error = ext4_journal_extend(handle, credits);
850 + error = ext4_journal_restart(handle, credits);
853 + ext4_warning(inode->i_sb,
854 + "couldn't extend journal "
855 + "(err %d)", error);
859 + ea_inode = ext4_xattr_inode_iget(inode,
860 + lea_ino_array->xia_inodes[idx], &error);
863 + ext4_orphan_add(handle, ea_inode);
864 + /* the inode's i_count will be released by caller */
871 * ext4_xattr_delete_inode()
873 - * Free extended attribute resources associated with this inode. This
874 + * Free extended attribute resources associated with this inode. Traverse
875 + * all entries and unlink any xattr inodes associated with this inode. This
876 * is called immediately before an inode is freed. We have exclusive
877 - * access to the inode.
878 + * access to the inode. If an orphan inode is deleted it will also delete any
879 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
880 + * to ensure they belong to the parent inode and were not deleted already.
883 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
885 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
886 + struct ext4_xattr_ino_array **lea_ino_array)
888 struct buffer_head *bh = NULL;
889 + struct ext4_xattr_ibody_header *header;
890 + struct ext4_inode *raw_inode;
891 + struct ext4_iloc iloc;
892 + struct ext4_xattr_entry *entry;
893 + int credits = 3, error = 0;
895 - if (!EXT4_I(inode)->i_file_acl)
896 + if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
897 + goto delete_external_ea;
899 + error = ext4_get_inode_loc(inode, &iloc);
902 + raw_inode = ext4_raw_inode(&iloc);
903 + header = IHDR(inode, raw_inode);
904 + for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
905 + entry = EXT4_XATTR_NEXT(entry)) {
906 + if (!entry->e_value_inum)
908 + if (ext4_expand_ino_array(lea_ino_array,
909 + entry->e_value_inum) != 0) {
913 + entry->e_value_inum = 0;
918 + if (!EXT4_I(inode)->i_file_acl) {
919 + /* add xattr inode to orphan list */
920 + ext4_xattr_inode_orphan_add(handle, inode, credits,
924 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
926 EXT4_ERROR_INODE(inode, "block %llu read error",
927 @@ -1521,11 +1979,69 @@ ext4_xattr_delete_inode(handle_t *handle
928 EXT4_I(inode)->i_file_acl);
932 + for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
933 + entry = EXT4_XATTR_NEXT(entry)) {
934 + if (!entry->e_value_inum)
936 + if (ext4_expand_ino_array(lea_ino_array,
937 + entry->e_value_inum) != 0)
939 + entry->e_value_inum = 0;
942 + /* add xattr inode to orphan list */
943 + error = ext4_xattr_inode_orphan_add(handle, inode, credits,
948 + if (!IS_NOQUOTA(inode))
949 + credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
951 + if (!ext4_handle_has_enough_credits(handle, credits)) {
952 + error = ext4_journal_extend(handle, credits);
954 + error = ext4_journal_restart(handle, credits);
956 + ext4_warning(inode->i_sb,
957 + "couldn't extend journal (err %d)", error);
962 ext4_xattr_release_block(handle, inode, bh);
963 EXT4_I(inode)->i_file_acl = 0;
972 +ext4_xattr_inode_array_free(struct inode *inode,
973 + struct ext4_xattr_ino_array *lea_ino_array)
975 + struct inode *ea_inode = NULL;
979 + if (lea_ino_array == NULL)
982 + for (; idx < lea_ino_array->xia_count; ++idx) {
983 + ea_inode = ext4_xattr_inode_iget(inode,
984 + lea_ino_array->xia_inodes[idx], &err);
987 + /* for inode's i_count get from ext4_xattr_delete_inode */
988 + if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
990 + clear_nlink(ea_inode);
993 + kfree(lea_ino_array);
997 @@ -1595,10 +2111,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
998 entry1->e_name_index != entry2->e_name_index ||
999 entry1->e_name_len != entry2->e_name_len ||
1000 entry1->e_value_size != entry2->e_value_size ||
1001 + entry1->e_value_inum != entry2->e_value_inum ||
1002 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1004 - if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1006 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1007 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1008 le32_to_cpu(entry1->e_value_size)))
1009 @@ -1682,7 +2197,7 @@ static inline void ext4_xattr_hash_entry
1013 - if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1014 + if (!entry->e_value_inum && entry->e_value_size) {
1015 __le32 *value = (__le32 *)((char *)header +
1016 le16_to_cpu(entry->e_value_offs));
1017 for (n = (le32_to_cpu(entry->e_value_size) +
1018 Index: linux-stage/fs/ext4/xattr.h
1019 ===================================================================
1020 --- linux-stage.orig/fs/ext4/xattr.h
1021 +++ linux-stage/fs/ext4/xattr.h
1022 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
1023 __u8 e_name_len; /* length of name */
1024 __u8 e_name_index; /* attribute name index */
1025 __le16 e_value_offs; /* offset in disk block of value */
1026 - __le32 e_value_block; /* disk block attribute is stored on (n/i) */
1027 + __le32 e_value_inum; /* inode in which the value is stored */
1028 __le32 e_value_size; /* size of attribute value */
1029 __le32 e_hash; /* hash value of name and value */
1030 char e_name[0]; /* attribute name */
1031 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
1032 EXT4_I(inode)->i_extra_isize))
1033 #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1036 + * Link EA inode back to parent one using i_mtime field.
1037 + * Extra integer type conversion added to ignore higher
1038 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1040 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum) \
1042 + (inode)->i_mtime.tv_sec = inum; \
1045 +#define EXT4_XATTR_INODE_GET_PARENT(inode) \
1046 + ((__u32)(inode)->i_mtime.tv_sec)
1049 + * The minimum size of EA value when you start storing it in an external inode
1050 + * size of block - size of header - size of 1 entry - 4 null bytes
1052 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b) \
1053 + ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1055 #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1056 #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1057 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1058 @@ -75,10 +95,11 @@ struct ext4_xattr_entry {
1059 #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1061 struct ext4_xattr_info {
1070 struct ext4_xattr_search {
1071 @@ -106,7 +127,13 @@ extern int ext4_xattr_get(struct inode *
1072 extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1073 extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1075 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1076 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1078 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1079 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1080 + struct ext4_xattr_ino_array **array);
1081 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1082 + struct ext4_xattr_ino_array *array);
1083 extern void ext4_xattr_put_super(struct super_block *);
1085 extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1086 Index: linux-stage/fs/ext4/ialloc.c
1087 ===================================================================
1088 --- linux-stage.orig/fs/ext4/ialloc.c
1089 +++ linux-stage/fs/ext4/ialloc.c
1090 @@ -247,7 +247,6 @@ void ext4_free_inode(handle_t *handle, s
1091 * as writing the quota to disk may need the lock as well.
1093 dquot_initialize(inode);
1094 - ext4_xattr_delete_inode(handle, inode);
1095 dquot_free_inode(inode);
1098 Index: linux-stage/fs/ext4/inline.c
1099 ===================================================================
1100 --- linux-stage.orig/fs/ext4/inline.c
1101 +++ linux-stage/fs/ext4/inline.c
1102 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1104 /* Compute min_offs. */
1105 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1106 - if (!entry->e_value_block && entry->e_value_size) {
1107 + if (!entry->e_value_inum && entry->e_value_size) {
1108 size_t offs = le16_to_cpu(entry->e_value_offs);
1109 if (offs < min_offs)
1111 Index: linux-stage/fs/ext4/extents.c
1112 ===================================================================
1113 --- linux-stage.orig/fs/ext4/extents.c
1114 +++ linux-stage/fs/ext4/extents.c
1115 @@ -2474,7 +2474,8 @@ int ext4_ext_index_trans_blocks(struct i
1117 static inline int get_default_free_blocks_flags(struct inode *inode)
1119 - if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1120 + if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
1121 + ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
1122 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
1123 else if (ext4_should_journal_data(inode))
1124 return EXT4_FREE_BLOCKS_FORGET;
1125 Index: linux-stage/fs/ext4/indirect.c
1126 ===================================================================
1127 --- linux-stage.orig/fs/ext4/indirect.c
1128 +++ linux-stage/fs/ext4/indirect.c
1129 @@ -967,7 +967,8 @@ static int ext4_clear_blocks(handle_t *h
1130 int flags = EXT4_FREE_BLOCKS_VALIDATED;
1133 - if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1134 + if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
1135 + ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
1136 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
1137 else if (ext4_should_journal_data(inode))
1138 flags |= EXT4_FREE_BLOCKS_FORGET;