Whamcloud - gitweb
3b05c6e713445a680d447a9cf84f932a9b3a7fca
[fs/lustre-release.git] / ldiskfs / kernel_patches / patches / rhel7.2 / ext4-large-eas.patch
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
5 number of EAs.
6
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
21  
22  /*
23 + * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1MB
24 + * This limit is arbitrary, but is reasonable for the xattr API.
25 + */
26 +#define EXT4_XATTR_MAX_LARGE_EA_SIZE    (1024 * 1024)
27 +
28 +/*
29   * Function prototypes
30   */
31  
32 @@ -1990,6 +1997,10 @@ struct mmpd_data {
33  # define ATTRIB_NORET  __attribute__((noreturn))
34  # define NORET_AND     noreturn,
35  
36 +struct ext4_xattr_ino_array {
37 +       unsigned int xia_count;         /* # of used item in the array */
38 +       unsigned int xia_inodes[0];
39 +};
40  /* bitmap.c */
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
56                                 unsigned int length);
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,
60 -                                 int pextents);
61  
62  /*
63   * Test whether an inode is a fast symlink.
64 @@ -184,6 +182,8 @@ void ext4_evict_inode(struct inode *inod
65  {
66         handle_t *handle;
67         int err;
68 +       int extra_credits = 3;
69 +       struct ext4_xattr_ino_array *lea_ino_array = NULL;
70  
71         trace_ext4_evict_inode(inode);
72  
73 @@ -236,8 +236,8 @@ void ext4_evict_inode(struct inode *inod
74          * protection against it
75          */
76         sb_start_intwrite(inode->i_sb);
77 -       handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
78 -                                   ext4_blocks_for_truncate(inode)+3);
79 +
80 +       handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits);
81         if (IS_ERR(handle)) {
82                 ext4_std_error(inode->i_sb, PTR_ERR(handle));
83                 /*
84 @@ -252,6 +252,32 @@ void ext4_evict_inode(struct inode *inod
85  
86         if (IS_SYNC(inode))
87                 ext4_handle_sync(handle);
88 +
89 +       /* Delete xattr inode before deleting the main inode. */
90 +       err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
91 +       if (err) {
92 +               ext4_warning(inode->i_sb,
93 +                            "couldn't delete inode's xattr (err %d)", err);
94 +               goto stop_handle;
95 +       }
96 +
97 +       if (!IS_NOQUOTA(inode))
98 +               extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
99 +
100 +       if (!ext4_handle_has_enough_credits(handle,
101 +                       ext4_blocks_for_truncate(inode) + extra_credits)) {
102 +               err = ext4_journal_extend(handle,
103 +                       ext4_blocks_for_truncate(inode) + extra_credits);
104 +               if (err > 0)
105 +                       err = ext4_journal_restart(handle,
106 +                       ext4_blocks_for_truncate(inode) + extra_credits);
107 +               if (err != 0) {
108 +                       ext4_warning(inode->i_sb,
109 +                                    "couldn't extend journal (err %d)", err);
110 +                       goto stop_handle;
111 +               }
112 +       }
113 +
114         inode->i_size = 0;
115         err = ext4_mark_inode_dirty(handle, inode);
116         if (err) {
117 @@ -269,10 +296,10 @@ void ext4_evict_inode(struct inode *inod
118          * enough credits left in the handle to remove the inode from
119          * the orphan list and set the dtime field.
120          */
121 -       if (!ext4_handle_has_enough_credits(handle, 3)) {
122 -               err = ext4_journal_extend(handle, 3);
123 +       if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
124 +               err = ext4_journal_extend(handle, extra_credits);
125                 if (err > 0)
126 -                       err = ext4_journal_restart(handle, 3);
127 +                       err = ext4_journal_restart(handle, extra_credits);
128                 if (err != 0) {
129                         ext4_warning(inode->i_sb,
130                                      "couldn't extend journal (err %d)", err);
131 @@ -308,6 +335,9 @@ void ext4_evict_inode(struct inode *inod
132                 ext4_free_inode(handle, inode);
133         ext4_journal_stop(handle);
134         sb_end_intwrite(inode->i_sb);
135 +
136 +       if (lea_ino_array != NULL)
137 +               ext4_xattr_inode_array_free(inode, lea_ino_array);
138         return;
139  no_delete:
140         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
141 @@ -4681,7 +4711,7 @@ static int ext4_index_trans_blocks(struc
142   *
143   * Also account for superblock, inode, quota and xattr blocks
144   */
145 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
146 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
147                                   int pextents)
148  {
149         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
150 Index: linux-stage/fs/ext4/xattr.c
151 ===================================================================
152 --- linux-stage.orig/fs/ext4/xattr.c
153 +++ linux-stage/fs/ext4/xattr.c
154 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
155  
156         while (!IS_LAST_ENTRY(entry)) {
157                 if (entry->e_value_size != 0 &&
158 +                   entry->e_value_inum == 0 &&
159                     (value_start + le16_to_cpu(entry->e_value_offs) <
160                      (void *)e + sizeof(__u32) ||
161                      value_start + le16_to_cpu(entry->e_value_offs) +
162 @@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
163  }
164  
165  static inline int
166 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
167 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
168 +                      struct inode *inode)
169  {
170         size_t value_size = le32_to_cpu(entry->e_value_size);
171  
172 -       if (entry->e_value_block != 0 || value_size > size ||
173 +       if (!entry->e_value_inum &&
174             le16_to_cpu(entry->e_value_offs) + value_size > size)
175 +               return -EIO;
176 +       if (entry->e_value_inum &&
177 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
178 +            le32_to_cpu(entry->e_value_inum) >
179 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
180                 return -EIO;
181         return 0;
182  }
183  
184  static int
185  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
186 -                     const char *name, size_t size, int sorted)
187 +                     const char *name, size_t size, int sorted,
188 +                     struct inode *inode)
189  {
190         struct ext4_xattr_entry *entry;
191         size_t name_len;
192 @@ -265,11 +273,109 @@ ext4_xattr_find_entry(struct ext4_xattr_
193                         break;
194         }
195         *pentry = entry;
196 -       if (!cmp && ext4_xattr_check_entry(entry, size))
197 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
198                         return -EIO;
199         return cmp ? -ENODATA : 0;
200  }
201  
202 +/*
203 + * Read the EA value from an inode.
204 + */
205 +static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
206 +{
207 +       unsigned long block = 0;
208 +       struct buffer_head *bh = NULL;
209 +       int err, blocksize;
210 +       size_t csize, ret_size = 0;
211 +
212 +       if (*size == 0)
213 +               return 0;
214 +
215 +       blocksize = ea_inode->i_sb->s_blocksize;
216 +
217 +       while (ret_size < *size) {
218 +               csize = (*size - ret_size) > blocksize ? blocksize :
219 +                                                       *size - ret_size;
220 +               bh = ext4_bread(NULL, ea_inode, block, 0, &err);
221 +               if (!bh) {
222 +                       *size = ret_size;
223 +                       return err;
224 +               }
225 +               memcpy(buf, bh->b_data, csize);
226 +               brelse(bh);
227 +
228 +               buf += csize;
229 +               block += 1;
230 +               ret_size += csize;
231 +       }
232 +
233 +       *size = ret_size;
234 +
235 +       return err;
236 +}
237 +
238 +/*
239 + * Fetch the xattr inode from disk.
240 + *
241 + * The xattr inode stores the parent inode number and generation so that
242 + * the kernel and e2fsck can verify the xattr inode is valid upon access.
243 + */
244 +struct inode *ext4_xattr_inode_iget(struct inode *parent,
245 +                                   unsigned long ea_ino, int *err)
246 +{
247 +       struct inode *ea_inode = NULL;
248 +
249 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
250 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
251 +               int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
252 +               ext4_error(parent->i_sb, "error while reading EA inode %lu "
253 +                          "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
254 +               *err = rc != 0 ? rc : -EIO;
255 +               return NULL;
256 +       }
257 +
258 +       if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
259 +           ea_inode->i_generation != parent->i_generation) {
260 +               ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
261 +                          "to parent invalid.", ea_ino);
262 +               *err = -EINVAL;
263 +               goto error;
264 +       }
265 +
266 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
267 +               ext4_error(parent->i_sb, "EA inode %lu does not have "
268 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
269 +               *err = -EINVAL;
270 +               goto error;
271 +       }
272 +
273 +       *err = 0;
274 +       return ea_inode;
275 +
276 +error:
277 +       iput(ea_inode);
278 +       return NULL;
279 +}
280 +
281 +/*
282 + * Read the value from the EA inode.
283 + */
284 +static int ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino,
285 +                               void *buffer, size_t *size)
286 +{
287 +       struct inode *ea_inode = NULL;
288 +       int err;
289 +
290 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
291 +       if (err)
292 +               return err;
293 +
294 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
295 +       iput(ea_inode);
296 +
297 +       return err;
298 +}
299 +
300  static int
301  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
302                      void *buffer, size_t buffer_size)
303 @@ -301,7 +401,8 @@ bad_block:
304         }
305         ext4_xattr_cache_insert(bh);
306         entry = BFIRST(bh);
307 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
308 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
309 +                                     inode);
310         if (error == -EIO)
311                 goto bad_block;
312         if (error)
313 @@ -311,8 +412,16 @@ bad_block:
314                 error = -ERANGE;
315                 if (size > buffer_size)
316                         goto cleanup;
317 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
318 -                      size);
319 +               if (entry->e_value_inum) {
320 +                       error = ext4_xattr_inode_get(inode,
321 +                                            le32_to_cpu(entry->e_value_inum),
322 +                                            buffer, &size);
323 +                       if (error)
324 +                               goto cleanup;
325 +               } else {
326 +                       memcpy(buffer, bh->b_data +
327 +                              le16_to_cpu(entry->e_value_offs), size);
328 +               }
329         }
330         error = size;
331  
332 @@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
333         if (error)
334                 goto cleanup;
335         error = ext4_xattr_find_entry(&entry, name_index, name,
336 -                                     end - (void *)entry, 0);
337 +                                     end - (void *)entry, 0, inode);
338         if (error)
339                 goto cleanup;
340         size = le32_to_cpu(entry->e_value_size);
341 @@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
342                 error = -ERANGE;
343                 if (size > buffer_size)
344                         goto cleanup;
345 -               memcpy(buffer, (void *)IFIRST(header) +
346 -                      le16_to_cpu(entry->e_value_offs), size);
347 +               if (entry->e_value_inum) {
348 +                       error = ext4_xattr_inode_get(inode,
349 +                                            le32_to_cpu(entry->e_value_inum),
350 +                                            buffer, &size);
351 +                       if (error)
352 +                               goto cleanup;
353 +               } else {
354 +                       memcpy(buffer, (void *)IFIRST(header) +
355 +                              le16_to_cpu(entry->e_value_offs), size);
356 +               }
357         }
358         error = size;
359  
360 @@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
361                                     size_t *min_offs, void *base, int *total)
362  {
363         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
364 -               if (!last->e_value_block && last->e_value_size) {
365 +               if (!last->e_value_inum && last->e_value_size) {
366                         size_t offs = le16_to_cpu(last->e_value_offs);
367                         if (offs < *min_offs)
368                                 *min_offs = offs;
369 @@ -611,16 +728,174 @@ static size_t ext4_xattr_free_space(stru
370         return (*min_offs - ((void *)last - base) - sizeof(__u32));
371  }
372  
373 +/*
374 + * Write the value of the EA in an inode.
375 + */
376  static int
377 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
378 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
379 +                      const void *buf, int bufsize)
380 +{
381 +       struct buffer_head *bh = NULL;
382 +       unsigned long block = 0;
383 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
384 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
385 +       int csize, wsize = 0;
386 +       int ret = 0;
387 +       int retries = 0;
388 +
389 +retry:
390 +       while (ret >= 0 && ret < max_blocks) {
391 +               struct ext4_map_blocks map;
392 +               map.m_lblk = block += ret;
393 +               map.m_len = max_blocks -= ret;
394 +
395 +               ret = ext4_map_blocks(handle, ea_inode, &map,
396 +                                     EXT4_GET_BLOCKS_CREATE);
397 +               if (ret <= 0) {
398 +                       ext4_mark_inode_dirty(handle, ea_inode);
399 +                       if (ret == -ENOSPC &&
400 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
401 +                               ret = 0;
402 +                               goto retry;
403 +                       }
404 +                       break;
405 +               }
406 +       }
407 +
408 +       if (ret < 0)
409 +               return ret;
410 +
411 +       block = 0;
412 +       while (wsize < bufsize) {
413 +               if (bh != NULL)
414 +                       brelse(bh);
415 +               csize = (bufsize - wsize) > blocksize ? blocksize :
416 +                                                               bufsize - wsize;
417 +               bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
418 +               if (!bh)
419 +                       goto out;
420 +               ret = ext4_journal_get_write_access(handle, bh);
421 +               if (ret)
422 +                       goto out;
423 +
424 +               memcpy(bh->b_data, buf, csize);
425 +               set_buffer_uptodate(bh);
426 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
427 +
428 +               buf += csize;
429 +               wsize += csize;
430 +               block += 1;
431 +       }
432 +
433 +       mutex_lock(&ea_inode->i_mutex);
434 +       i_size_write(ea_inode, wsize);
435 +       ext4_update_i_disksize(ea_inode, wsize);
436 +       mutex_unlock(&ea_inode->i_mutex);
437 +
438 +       ext4_mark_inode_dirty(handle, ea_inode);
439 +
440 +out:
441 +       brelse(bh);
442 +
443 +       return ret;
444 +}
445 +
446 +/*
447 + * Create an inode to store the value of a large EA.
448 + */
449 +static struct inode *
450 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
451 +{
452 +       struct inode *ea_inode = NULL;
453 +
454 +       /*
455 +        * Let the next inode be the goal, so we try and allocate the EA inode
456 +        * in the same group, or nearby one.
457 +        */
458 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
459 +                                 S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
460 +
461 +       if (!IS_ERR(ea_inode)) {
462 +               ea_inode->i_op = &ext4_file_inode_operations;
463 +               ea_inode->i_fop = &ext4_file_operations;
464 +               ext4_set_aops(ea_inode);
465 +               ea_inode->i_generation = inode->i_generation;
466 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
467 +
468 +               /*
469 +                * A back-pointer from EA inode to parent inode will be useful
470 +                * for e2fsck.
471 +                */
472 +               EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
473 +               unlock_new_inode(ea_inode);
474 +       }
475 +
476 +       return ea_inode;
477 +}
478 +
479 +/*
480 + * Unlink the inode storing the value of the EA.
481 + */
482 +int
483 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
484 +{
485 +       struct inode *ea_inode = NULL;
486 +       int err;
487 +
488 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
489 +       if (err)
490 +               return err;
491 +
492 +       clear_nlink(ea_inode);
493 +       iput(ea_inode);
494 +
495 +       return 0;
496 +}
497 +
498 +/*
499 + * Add value of the EA in an inode.
500 + */
501 +static int
502 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
503 +                    const void *value, size_t value_len)
504 +{
505 +       struct inode *ea_inode = NULL;
506 +       int err;
507 +
508 +       /* Create an inode for the EA value */
509 +       ea_inode = ext4_xattr_inode_create(handle, inode);
510 +       if (IS_ERR(ea_inode))
511 +               return -1;
512 +
513 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
514 +       if (err)
515 +               clear_nlink(ea_inode);
516 +       else
517 +               *ea_ino = ea_inode->i_ino;
518 +
519 +       iput(ea_inode);
520 +
521 +       return err;
522 +}
523 +
524 +static int
525 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
526 +                    handle_t *handle, struct inode *inode)
527  {
528         struct ext4_xattr_entry *last;
529         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
530 +       int in_inode = i->in_inode;
531 +
532 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
533 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
534 +           (EXT4_XATTR_SIZE(i->value_len) >
535 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
536 +               in_inode = 1;
537  
538         /* Compute min_offs and last. */
539         last = s->first;
540         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
541 -               if (!last->e_value_block && last->e_value_size) {
542 +               if (!last->e_value_inum && last->e_value_size) {
543                         size_t offs = le16_to_cpu(last->e_value_offs);
544                         if (offs < min_offs)
545                                 min_offs = offs;
546 @@ -628,15 +903,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
547         }
548         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
549         if (!s->not_found) {
550 -               if (!s->here->e_value_block && s->here->e_value_size) {
551 +               if (!in_inode &&
552 +                   !s->here->e_value_inum && s->here->e_value_size) {
553                         size_t size = le32_to_cpu(s->here->e_value_size);
554                         free += EXT4_XATTR_SIZE(size);
555                 }
556                 free += EXT4_XATTR_LEN(name_len);
557         }
558         if (i->value) {
559 -               if (free < EXT4_XATTR_LEN(name_len) +
560 -                          EXT4_XATTR_SIZE(i->value_len))
561 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
562 +
563 +               if (in_inode)
564 +                       value_len = 0;
565 +
566 +               if (free < value_len ||
567 +                   free < EXT4_XATTR_LEN(name_len) + value_len)
568                         return -ENOSPC;
569         }
570  
571 @@ -651,7 +931,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
572                 s->here->e_name_len = name_len;
573                 memcpy(s->here->e_name, i->name, name_len);
574         } else {
575 -               if (!s->here->e_value_block && s->here->e_value_size) {
576 +               if (!s->here->e_value_inum && s->here->e_value_size &&
577 +                   s->here->e_value_offs > 0) {
578                         void *first_val = s->base + min_offs;
579                         size_t offs = le16_to_cpu(s->here->e_value_offs);
580                         void *val = s->base + offs;
581 @@ -685,13 +966,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
582                         last = s->first;
583                         while (!IS_LAST_ENTRY(last)) {
584                                 size_t o = le16_to_cpu(last->e_value_offs);
585 -                               if (!last->e_value_block &&
586 +                               if (!last->e_value_inum &&
587                                     last->e_value_size && o < offs)
588                                         last->e_value_offs =
589                                                 cpu_to_le16(o + size);
590                                 last = EXT4_XATTR_NEXT(last);
591                         }
592                 }
593 +               if (s->here->e_value_inum) {
594 +                       ext4_xattr_inode_unlink(inode,
595 +                                       le32_to_cpu(s->here->e_value_inum));
596 +                       s->here->e_value_inum = 0;
597 +               }
598                 if (!i->value) {
599                         /* Remove the old name. */
600                         size_t size = EXT4_XATTR_LEN(name_len);
601 @@ -705,10 +990,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
602         if (i->value) {
603                 /* Insert the new value. */
604                 s->here->e_value_size = cpu_to_le32(i->value_len);
605 -               if (i->value_len) {
606 +               if (in_inode) {
607 +                       unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
608 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
609 +                                            i->value_len);
610 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
611 +                       s->here->e_value_offs = 0;
612 +               } else if (i->value_len) {
613                         size_t size = EXT4_XATTR_SIZE(i->value_len);
614                         void *val = s->base + min_offs - size;
615                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
616 +                       s->here->e_value_inum = 0;
617                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
618                                 memset(val, 0, size);
619                         } else {
620 @@ -758,7 +1050,7 @@ ext4_xattr_block_find(struct inode *inod
621                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
622                 bs->s.here = bs->s.first;
623                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
624 -                                             i->name, bs->bh->b_size, 1);
625 +                                             i->name, bs->bh->b_size, 1, inode);
626                 if (error && error != -ENODATA)
627                         goto cleanup;
628                 bs->s.not_found = error;
629 @@ -782,8 +1074,6 @@ ext4_xattr_block_set(handle_t *handle, s
630  
631  #define header(x) ((struct ext4_xattr_header *)(x))
632  
633 -       if (i->value && i->value_len > sb->s_blocksize)
634 -               return -ENOSPC;
635         if (s->base) {
636                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
637                                         bs->bh->b_blocknr);
638 @@ -799,7 +1089,7 @@ ext4_xattr_block_set(handle_t *handle, s
639                                 ce = NULL;
640                         }
641                         ea_bdebug(bs->bh, "modifying in-place");
642 -                       error = ext4_xattr_set_entry(i, s);
643 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
644                         if (!error) {
645                                 if (!IS_LAST_ENTRY(s->first))
646                                         ext4_xattr_rehash(header(s->base),
647 @@ -850,7 +1140,7 @@ ext4_xattr_block_set(handle_t *handle, s
648                 s->end = s->base + sb->s_blocksize;
649         }
650  
651 -       error = ext4_xattr_set_entry(i, s);
652 +       error = ext4_xattr_set_entry(i, s, handle, inode);
653         if (error == -EIO)
654                 goto bad_block;
655         if (error)
656 @@ -1000,7 +1290,7 @@ int ext4_xattr_ibody_find(struct inode *
657                 /* Find the named attribute. */
658                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
659                                               i->name, is->s.end -
660 -                                             (void *)is->s.base, 0);
661 +                                             (void *)is->s.base, 0, inode);
662                 if (error && error != -ENODATA)
663                         return error;
664                 is->s.not_found = error;
665 @@ -1018,7 +1308,7 @@ int ext4_xattr_ibody_inline_set(handle_t
666  
667         if (EXT4_I(inode)->i_extra_isize == 0)
668                 return -ENOSPC;
669 -       error = ext4_xattr_set_entry(i, s);
670 +       error = ext4_xattr_set_entry(i, s, handle, inode);
671         if (error) {
672                 if (error == -ENOSPC &&
673                     ext4_has_inline_data(inode)) {
674 @@ -1030,7 +1320,7 @@ int ext4_xattr_ibody_inline_set(handle_t
675                         error = ext4_xattr_ibody_find(inode, i, is);
676                         if (error)
677                                 return error;
678 -                       error = ext4_xattr_set_entry(i, s);
679 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
680                 }
681                 if (error)
682                         return error;
683 @@ -1056,7 +1346,7 @@ static int ext4_xattr_ibody_set(handle_t
684  
685         if (EXT4_I(inode)->i_extra_isize == 0)
686                 return -ENOSPC;
687 -       error = ext4_xattr_set_entry(i, s);
688 +       error = ext4_xattr_set_entry(i, s, handle, inode);
689         if (error)
690                 return error;
691         header = IHDR(inode, ext4_raw_inode(&is->iloc));
692 @@ -1092,7 +1382,7 @@ ext4_xattr_set_handle(handle_t *handle, 
693                 .name = name,
694                 .value = value,
695                 .value_len = value_len,
696 -
697 +               .in_inode = 0,
698         };
699         struct ext4_xattr_ibody_find is = {
700                 .s = { .not_found = -ENODATA, },
701 @@ -1157,6 +1447,15 @@ ext4_xattr_set_handle(handle_t *handle, 
702                                         goto cleanup;
703                         }
704                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
705 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
706 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
707 +                           error == -ENOSPC) {
708 +                               /* xattr not fit to block, store at external
709 +                                * inode */
710 +                               i.in_inode = 1;
711 +                               error = ext4_xattr_ibody_set(handle, inode,
712 +                                                            &i, &is);
713 +                       }
714                         if (error)
715                                 goto cleanup;
716                         if (!is.s.not_found) {
717 @@ -1203,9 +1502,22 @@ ext4_xattr_set(struct inode *inode, int 
718                const void *value, size_t value_len, int flags)
719  {
720         handle_t *handle;
721 +       struct super_block *sb = inode->i_sb;
722         int error, retries = 0;
723         int credits = ext4_jbd2_credits_xattr(inode);
724  
725 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
726 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
727 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
728 +                                       sb->s_blocksize_bits;
729 +
730 +               /* For new inode */
731 +               credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
732 +
733 +               /* For data blocks of EA inode */
734 +               credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
735 +       }
736 +
737  retry:
738         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
739         if (IS_ERR(handle)) {
740 @@ -1217,7 +1529,7 @@ retry:
741                                               value, value_len, flags);
742                 error2 = ext4_journal_stop(handle);
743                 if (error == -ENOSPC &&
744 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
745 +                   ext4_should_retry_alloc(sb, &retries))
746                         goto retry;
747                 if (error == 0)
748                         error = error2;
749 @@ -1239,7 +1551,7 @@ static void ext4_xattr_shift_entries(str
750  
751         /* Adjust the value offsets of the entries */
752         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
753 -               if (!last->e_value_block && last->e_value_size) {
754 +               if (!last->e_value_inum && last->e_value_size) {
755                         new_offs = le16_to_cpu(last->e_value_offs) +
756                                                         value_offs_shift;
757                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
758 @@ -1477,21 +1789,135 @@ cleanup:
759  }
760  
761  
762 +#define EIA_INCR 16 /* must be 2^n */
763 +#define EIA_MASK (EIA_INCR - 1)
764 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
765 + * If @lea_ino_array is new or full it will be grown and the old
766 + * contents copied over.
767 + */
768 +static int
769 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
770 +{
771 +       if (*lea_ino_array == NULL) {
772 +               /*
773 +                * Start with 15 inodes, so it fits into a power-of-two size.
774 +                * If *lea_ino_array is NULL, this is essentially offsetof()
775 +                */
776 +               (*lea_ino_array) =
777 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
778 +                                        xia_inodes[EIA_MASK]),
779 +                               GFP_NOFS);
780 +               if (*lea_ino_array == NULL)
781 +                       return -ENOMEM;
782 +               (*lea_ino_array)->xia_count = 0;
783 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
784 +               /* expand the array once all 15 + n * 16 slots are full */
785 +               struct ext4_xattr_ino_array *new_array = NULL;
786 +               int count = (*lea_ino_array)->xia_count;
787 +
788 +               /* if new_array is NULL, this is essentially offsetof() */
789 +               new_array = kmalloc(
790 +                               offsetof(struct ext4_xattr_ino_array,
791 +                                        xia_inodes[count + EIA_INCR]),
792 +                               GFP_NOFS);
793 +               if (new_array == NULL)
794 +                       return -ENOMEM;
795 +               memcpy(new_array, *lea_ino_array,
796 +                      offsetof(struct ext4_xattr_ino_array,
797 +                               xia_inodes[count]));
798 +               kfree(*lea_ino_array);
799 +               *lea_ino_array = new_array;
800 +       }
801 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
802 +       return 0;
803 +}
804 +
805 +/**
806 + * Add xattr inode to orphan list
807 + */
808 +static int
809 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
810 +                       int credits, struct ext4_xattr_ino_array *lea_ino_array)
811 +{
812 +       struct inode *ea_inode = NULL;
813 +       int idx = 0, error = 0;
814 +
815 +       if (lea_ino_array == NULL)
816 +               return 0;
817 +
818 +       for (; idx < lea_ino_array->xia_count; ++idx) {
819 +               if (!ext4_handle_has_enough_credits(handle, credits)) {
820 +                       error = ext4_journal_extend(handle, credits);
821 +                       if (error > 0)
822 +                               error = ext4_journal_restart(handle, credits);
823 +
824 +                       if (error != 0) {
825 +                               ext4_warning(inode->i_sb,
826 +                                       "couldn't extend journal "
827 +                                       "(err %d)", error);
828 +                               return error;
829 +                       }
830 +               }
831 +               ea_inode = ext4_xattr_inode_iget(inode,
832 +                               lea_ino_array->xia_inodes[idx], &error);
833 +               if (error)
834 +                       continue;
835 +               ext4_orphan_add(handle, ea_inode);
836 +               /* the inode's i_count will be released by caller */
837 +       }
838 +
839 +       return 0;
840 +}
841  
842  /*
843   * ext4_xattr_delete_inode()
844   *
845 - * Free extended attribute resources associated with this inode. This
846 + * Free extended attribute resources associated with this inode. Traverse
847 + * all entries and unlink any xattr inodes associated with this inode. This
848   * is called immediately before an inode is freed. We have exclusive
849 - * access to the inode.
850 + * access to the inode. If an orphan inode is deleted it will also delete any
851 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
852 + * to ensure they belong to the parent inode and were not deleted already.
853   */
854 -void
855 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
856 +int
857 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
858 +                       struct ext4_xattr_ino_array **lea_ino_array)
859  {
860         struct buffer_head *bh = NULL;
861 +       struct ext4_xattr_ibody_header *header;
862 +       struct ext4_inode *raw_inode;
863 +       struct ext4_iloc iloc;
864 +       struct ext4_xattr_entry *entry;
865 +       int credits = 3, error = 0;
866  
867 -       if (!EXT4_I(inode)->i_file_acl)
868 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
869 +               goto delete_external_ea;
870 +
871 +       error = ext4_get_inode_loc(inode, &iloc);
872 +       if (error)
873                 goto cleanup;
874 +       raw_inode = ext4_raw_inode(&iloc);
875 +       header = IHDR(inode, raw_inode);
876 +       for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
877 +            entry = EXT4_XATTR_NEXT(entry)) {
878 +               if (!entry->e_value_inum)
879 +                       continue;
880 +               if (ext4_expand_ino_array(lea_ino_array,
881 +                                         entry->e_value_inum) != 0) {
882 +                       brelse(iloc.bh);
883 +                       goto cleanup;
884 +               }
885 +               entry->e_value_inum = 0;
886 +       }
887 +       brelse(iloc.bh);
888 +
889 +delete_external_ea:
890 +       if (!EXT4_I(inode)->i_file_acl) {
891 +               /* add xattr inode to orphan list */
892 +               ext4_xattr_inode_orphan_add(handle, inode, credits,
893 +                                               *lea_ino_array);
894 +               goto cleanup;
895 +       }
896         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
897         if (!bh) {
898                 EXT4_ERROR_INODE(inode, "block %llu read error",
899 @@ -1504,11 +1930,69 @@ ext4_xattr_delete_inode(handle_t *handle
900                                  EXT4_I(inode)->i_file_acl);
901                 goto cleanup;
902         }
903 +
904 +       for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
905 +            entry = EXT4_XATTR_NEXT(entry)) {
906 +               if (!entry->e_value_inum)
907 +                       continue;
908 +               if (ext4_expand_ino_array(lea_ino_array,
909 +                                         entry->e_value_inum) != 0)
910 +                       goto cleanup;
911 +               entry->e_value_inum = 0;
912 +       }
913 +
914 +       /* add xattr inode to orphan list */
915 +       error = ext4_xattr_inode_orphan_add(handle, inode, credits,
916 +                                       *lea_ino_array);
917 +       if (error != 0)
918 +               goto cleanup;
919 +
920 +       if (!IS_NOQUOTA(inode))
921 +               credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
922 +
923 +       if (!ext4_handle_has_enough_credits(handle, credits)) {
924 +               error = ext4_journal_extend(handle, credits);
925 +               if (error > 0)
926 +                       error = ext4_journal_restart(handle, credits);
927 +               if (error != 0) {
928 +                       ext4_warning(inode->i_sb,
929 +                               "couldn't extend journal (err %d)", error);
930 +                       goto cleanup;
931 +               }
932 +       }
933 +
934         ext4_xattr_release_block(handle, inode, bh);
935         EXT4_I(inode)->i_file_acl = 0;
936  
937  cleanup:
938         brelse(bh);
939 +
940 +       return error;
941 +}
942 +
943 +void
944 +ext4_xattr_inode_array_free(struct inode *inode,
945 +                           struct ext4_xattr_ino_array *lea_ino_array)
946 +{
947 +       struct inode    *ea_inode = NULL;
948 +       int             idx = 0;
949 +       int             err;
950 +
951 +       if (lea_ino_array == NULL)
952 +               return;
953 +
954 +       for (; idx < lea_ino_array->xia_count; ++idx) {
955 +               ea_inode = ext4_xattr_inode_iget(inode,
956 +                               lea_ino_array->xia_inodes[idx], &err);
957 +               if (err)
958 +                       continue;
959 +               /* for inode's i_count get from ext4_xattr_delete_inode */
960 +               if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
961 +                       iput(ea_inode);
962 +               clear_nlink(ea_inode);
963 +               iput(ea_inode);
964 +       }
965 +       kfree(lea_ino_array);
966  }
967  
968  /*
969 @@ -1578,10 +2062,9 @@ ext4_xattr_cmp(struct ext4_xattr_header 
970                     entry1->e_name_index != entry2->e_name_index ||
971                     entry1->e_name_len != entry2->e_name_len ||
972                     entry1->e_value_size != entry2->e_value_size ||
973 +                   entry1->e_value_inum != entry2->e_value_inum ||
974                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
975                         return 1;
976 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
977 -                       return -EIO;
978                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
979                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
980                            le32_to_cpu(entry1->e_value_size)))
981 @@ -1665,7 +2148,7 @@ static inline void ext4_xattr_hash_entry
982                        *name++;
983         }
984  
985 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
986 +       if (!entry->e_value_inum && entry->e_value_size) {
987                 __le32 *value = (__le32 *)((char *)header +
988                         le16_to_cpu(entry->e_value_offs));
989                 for (n = (le32_to_cpu(entry->e_value_size) +
990 Index: linux-stage/fs/ext4/xattr.h
991 ===================================================================
992 --- linux-stage.orig/fs/ext4/xattr.h
993 +++ linux-stage/fs/ext4/xattr.h
994 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
995         __u8    e_name_len;     /* length of name */
996         __u8    e_name_index;   /* attribute name index */
997         __le16  e_value_offs;   /* offset in disk block of value */
998 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
999 +       __le32  e_value_inum;   /* inode in which the value is stored */
1000         __le32  e_value_size;   /* size of attribute value */
1001         __le32  e_hash;         /* hash value of name and value */
1002         char    e_name[0];      /* attribute name */
1003 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
1004                 EXT4_I(inode)->i_extra_isize))
1005  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1006  
1007 +/*
1008 + * Link EA inode back to parent one using i_mtime field.
1009 + * Extra integer type conversion added to ignore higher
1010 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1011 + */
1012 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum)       \
1013 +do {                                                   \
1014 +       (inode)->i_mtime.tv_sec = inum;                 \
1015 +} while(0)
1016 +
1017 +#define EXT4_XATTR_INODE_GET_PARENT(inode)             \
1018 +       ((__u32)(inode)->i_mtime.tv_sec)
1019 +
1020 +/*
1021 + * The minimum size of EA value when you start storing it in an external inode
1022 + * size of block - size of header - size of 1 entry - 4 null bytes
1023 +*/
1024 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
1025 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1026 +
1027  #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1028  #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1029  #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1030 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1031  #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1032  
1033  struct ext4_xattr_info {
1034 -       int name_index;
1035         const char *name;
1036         const void *value;
1037         size_t value_len;
1038 +       int name_index;
1039 +       int in_inode;
1040  };
1041  
1042  struct ext4_xattr_search {
1043 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1044  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1045  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1046  
1047 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1048 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1049 +                                          int *err);
1050 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1051 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1052 +                                  struct ext4_xattr_ino_array **array);
1053 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1054 +                                       struct ext4_xattr_ino_array *array);
1055  extern void ext4_xattr_put_super(struct super_block *);
1056  
1057  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1058 Index: linux-stage/fs/ext4/ialloc.c
1059 ===================================================================
1060 --- linux-stage.orig/fs/ext4/ialloc.c
1061 +++ linux-stage/fs/ext4/ialloc.c
1062 @@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
1063          * as writing the quota to disk may need the lock as well.
1064          */
1065         dquot_initialize(inode);
1066 -       ext4_xattr_delete_inode(handle, inode);
1067         dquot_free_inode(inode);
1068         dquot_drop(inode);
1069  
1070 Index: linux-stage/fs/ext4/inline.c
1071 ===================================================================
1072 --- linux-stage.orig/fs/ext4/inline.c
1073 +++ linux-stage/fs/ext4/inline.c
1074 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1075  
1076         /* Compute min_offs. */
1077         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1078 -               if (!entry->e_value_block && entry->e_value_size) {
1079 +               if (!entry->e_value_inum && entry->e_value_size) {
1080                         size_t offs = le16_to_cpu(entry->e_value_offs);
1081                         if (offs < min_offs)
1082                                 min_offs = offs;