Whamcloud - gitweb
LU-7581 ldiskfs: RHEL7.2 fix wrong EA inode backpointer check
[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 @@ -308,6 +335,9 @@ void ext4_evict_inode(struct inode *inod
118                 ext4_free_inode(handle, inode);
119         ext4_journal_stop(handle);
120         sb_end_intwrite(inode->i_sb);
121 +
122 +       if (lea_ino_array != NULL)
123 +               ext4_xattr_inode_array_free(inode, lea_ino_array);
124         return;
125  no_delete:
126         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
127 @@ -4681,7 +4711,7 @@ static int ext4_index_trans_blocks(struc
128   *
129   * Also account for superblock, inode, quota and xattr blocks
130   */
131 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
132 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
133                                   int pextents)
134  {
135         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
136 Index: linux-stage/fs/ext4/xattr.c
137 ===================================================================
138 --- linux-stage.orig/fs/ext4/xattr.c
139 +++ linux-stage/fs/ext4/xattr.c
140 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
141  
142         while (!IS_LAST_ENTRY(entry)) {
143                 if (entry->e_value_size != 0 &&
144 +                   entry->e_value_inum == 0 &&
145                     (value_start + le16_to_cpu(entry->e_value_offs) <
146                      (void *)e + sizeof(__u32) ||
147                      value_start + le16_to_cpu(entry->e_value_offs) +
148 @@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
149  }
150  
151  static inline int
152 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
153 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
154 +                      struct inode *inode)
155  {
156         size_t value_size = le32_to_cpu(entry->e_value_size);
157  
158 -       if (entry->e_value_block != 0 || value_size > size ||
159 +       if (!entry->e_value_inum &&
160             le16_to_cpu(entry->e_value_offs) + value_size > size)
161 +               return -EIO;
162 +       if (entry->e_value_inum &&
163 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
164 +            le32_to_cpu(entry->e_value_inum) >
165 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
166                 return -EIO;
167         return 0;
168  }
169  
170  static int
171  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
172 -                     const char *name, size_t size, int sorted)
173 +                     const char *name, size_t size, int sorted,
174 +                     struct inode *inode)
175  {
176         struct ext4_xattr_entry *entry;
177         size_t name_len;
178 @@ -265,11 +273,109 @@ ext4_xattr_find_entry(struct ext4_xattr_
179                         break;
180         }
181         *pentry = entry;
182 -       if (!cmp && ext4_xattr_check_entry(entry, size))
183 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
184                         return -EIO;
185         return cmp ? -ENODATA : 0;
186  }
187  
188 +/*
189 + * Read the EA value from an inode.
190 + */
191 +static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
192 +{
193 +       unsigned long block = 0;
194 +       struct buffer_head *bh = NULL;
195 +       int err, blocksize;
196 +       size_t csize, ret_size = 0;
197 +
198 +       if (*size == 0)
199 +               return 0;
200 +
201 +       blocksize = ea_inode->i_sb->s_blocksize;
202 +
203 +       while (ret_size < *size) {
204 +               csize = (*size - ret_size) > blocksize ? blocksize :
205 +                                                       *size - ret_size;
206 +               bh = ext4_bread(NULL, ea_inode, block, 0, &err);
207 +               if (!bh) {
208 +                       *size = ret_size;
209 +                       return err;
210 +               }
211 +               memcpy(buf, bh->b_data, csize);
212 +               brelse(bh);
213 +
214 +               buf += csize;
215 +               block += 1;
216 +               ret_size += csize;
217 +       }
218 +
219 +       *size = ret_size;
220 +
221 +       return err;
222 +}
223 +
224 +/*
225 + * Fetch the xattr inode from disk.
226 + *
227 + * The xattr inode stores the parent inode number and generation so that
228 + * the kernel and e2fsck can verify the xattr inode is valid upon access.
229 + */
230 +struct inode *ext4_xattr_inode_iget(struct inode *parent,
231 +                                   unsigned long ea_ino, int *err)
232 +{
233 +       struct inode *ea_inode = NULL;
234 +
235 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
236 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
237 +               int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
238 +               ext4_error(parent->i_sb, "error while reading EA inode %lu "
239 +                          "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
240 +               *err = rc != 0 ? rc : -EIO;
241 +               return NULL;
242 +       }
243 +
244 +       if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
245 +           ea_inode->i_generation != parent->i_generation) {
246 +               ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
247 +                          "to parent invalid.", ea_ino);
248 +               *err = -EINVAL;
249 +               goto error;
250 +       }
251 +
252 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
253 +               ext4_error(parent->i_sb, "EA inode %lu does not have "
254 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
255 +               *err = -EINVAL;
256 +               goto error;
257 +       }
258 +
259 +       *err = 0;
260 +       return ea_inode;
261 +
262 +error:
263 +       iput(ea_inode);
264 +       return NULL;
265 +}
266 +
267 +/*
268 + * Read the value from the EA inode.
269 + */
270 +static int ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino,
271 +                               void *buffer, size_t *size)
272 +{
273 +       struct inode *ea_inode = NULL;
274 +       int err;
275 +
276 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
277 +       if (err)
278 +               return err;
279 +
280 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
281 +       iput(ea_inode);
282 +
283 +       return err;
284 +}
285 +
286  static int
287  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
288                      void *buffer, size_t buffer_size)
289 @@ -301,7 +401,8 @@ bad_block:
290         }
291         ext4_xattr_cache_insert(bh);
292         entry = BFIRST(bh);
293 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
294 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
295 +                                     inode);
296         if (error == -EIO)
297                 goto bad_block;
298         if (error)
299 @@ -311,8 +412,16 @@ bad_block:
300                 error = -ERANGE;
301                 if (size > buffer_size)
302                         goto cleanup;
303 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
304 -                      size);
305 +               if (entry->e_value_inum) {
306 +                       error = ext4_xattr_inode_get(inode,
307 +                                            le32_to_cpu(entry->e_value_inum),
308 +                                            buffer, &size);
309 +                       if (error)
310 +                               goto cleanup;
311 +               } else {
312 +                       memcpy(buffer, bh->b_data +
313 +                              le16_to_cpu(entry->e_value_offs), size);
314 +               }
315         }
316         error = size;
317  
318 @@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
319         if (error)
320                 goto cleanup;
321         error = ext4_xattr_find_entry(&entry, name_index, name,
322 -                                     end - (void *)entry, 0);
323 +                                     end - (void *)entry, 0, inode);
324         if (error)
325                 goto cleanup;
326         size = le32_to_cpu(entry->e_value_size);
327 @@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
328                 error = -ERANGE;
329                 if (size > buffer_size)
330                         goto cleanup;
331 -               memcpy(buffer, (void *)IFIRST(header) +
332 -                      le16_to_cpu(entry->e_value_offs), size);
333 +               if (entry->e_value_inum) {
334 +                       error = ext4_xattr_inode_get(inode,
335 +                                            le32_to_cpu(entry->e_value_inum),
336 +                                            buffer, &size);
337 +                       if (error)
338 +                               goto cleanup;
339 +               } else {
340 +                       memcpy(buffer, (void *)IFIRST(header) +
341 +                              le16_to_cpu(entry->e_value_offs), size);
342 +               }
343         }
344         error = size;
345  
346 @@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
347                                     size_t *min_offs, void *base, int *total)
348  {
349         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
350 -               if (!last->e_value_block && last->e_value_size) {
351 +               if (!last->e_value_inum && last->e_value_size) {
352                         size_t offs = le16_to_cpu(last->e_value_offs);
353                         if (offs < *min_offs)
354                                 *min_offs = offs;
355 @@ -611,16 +728,174 @@ static size_t ext4_xattr_free_space(stru
356         return (*min_offs - ((void *)last - base) - sizeof(__u32));
357  }
358  
359 +/*
360 + * Write the value of the EA in an inode.
361 + */
362  static int
363 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
364 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
365 +                      const void *buf, int bufsize)
366 +{
367 +       struct buffer_head *bh = NULL;
368 +       unsigned long block = 0;
369 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
370 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
371 +       int csize, wsize = 0;
372 +       int ret = 0;
373 +       int retries = 0;
374 +
375 +retry:
376 +       while (ret >= 0 && ret < max_blocks) {
377 +               struct ext4_map_blocks map;
378 +               map.m_lblk = block += ret;
379 +               map.m_len = max_blocks -= ret;
380 +
381 +               ret = ext4_map_blocks(handle, ea_inode, &map,
382 +                                     EXT4_GET_BLOCKS_CREATE);
383 +               if (ret <= 0) {
384 +                       ext4_mark_inode_dirty(handle, ea_inode);
385 +                       if (ret == -ENOSPC &&
386 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
387 +                               ret = 0;
388 +                               goto retry;
389 +                       }
390 +                       break;
391 +               }
392 +       }
393 +
394 +       if (ret < 0)
395 +               return ret;
396 +
397 +       block = 0;
398 +       while (wsize < bufsize) {
399 +               if (bh != NULL)
400 +                       brelse(bh);
401 +               csize = (bufsize - wsize) > blocksize ? blocksize :
402 +                                                               bufsize - wsize;
403 +               bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
404 +               if (!bh)
405 +                       goto out;
406 +               ret = ext4_journal_get_write_access(handle, bh);
407 +               if (ret)
408 +                       goto out;
409 +
410 +               memcpy(bh->b_data, buf, csize);
411 +               set_buffer_uptodate(bh);
412 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
413 +
414 +               buf += csize;
415 +               wsize += csize;
416 +               block += 1;
417 +       }
418 +
419 +       mutex_lock(&ea_inode->i_mutex);
420 +       i_size_write(ea_inode, wsize);
421 +       ext4_update_i_disksize(ea_inode, wsize);
422 +       mutex_unlock(&ea_inode->i_mutex);
423 +
424 +       ext4_mark_inode_dirty(handle, ea_inode);
425 +
426 +out:
427 +       brelse(bh);
428 +
429 +       return ret;
430 +}
431 +
432 +/*
433 + * Create an inode to store the value of a large EA.
434 + */
435 +static struct inode *
436 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
437 +{
438 +       struct inode *ea_inode = NULL;
439 +
440 +       /*
441 +        * Let the next inode be the goal, so we try and allocate the EA inode
442 +        * in the same group, or nearby one.
443 +        */
444 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
445 +                                 S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
446 +
447 +       if (!IS_ERR(ea_inode)) {
448 +               ea_inode->i_op = &ext4_file_inode_operations;
449 +               ea_inode->i_fop = &ext4_file_operations;
450 +               ext4_set_aops(ea_inode);
451 +               ea_inode->i_generation = inode->i_generation;
452 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
453 +
454 +               /*
455 +                * A back-pointer from EA inode to parent inode will be useful
456 +                * for e2fsck.
457 +                */
458 +               EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
459 +               unlock_new_inode(ea_inode);
460 +       }
461 +
462 +       return ea_inode;
463 +}
464 +
465 +/*
466 + * Unlink the inode storing the value of the EA.
467 + */
468 +int
469 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
470 +{
471 +       struct inode *ea_inode = NULL;
472 +       int err;
473 +
474 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
475 +       if (err)
476 +               return err;
477 +
478 +       clear_nlink(ea_inode);
479 +       iput(ea_inode);
480 +
481 +       return 0;
482 +}
483 +
484 +/*
485 + * Add value of the EA in an inode.
486 + */
487 +static int
488 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
489 +                    const void *value, size_t value_len)
490 +{
491 +       struct inode *ea_inode = NULL;
492 +       int err;
493 +
494 +       /* Create an inode for the EA value */
495 +       ea_inode = ext4_xattr_inode_create(handle, inode);
496 +       if (IS_ERR(ea_inode))
497 +               return -1;
498 +
499 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
500 +       if (err)
501 +               clear_nlink(ea_inode);
502 +       else
503 +               *ea_ino = ea_inode->i_ino;
504 +
505 +       iput(ea_inode);
506 +
507 +       return err;
508 +}
509 +
510 +static int
511 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
512 +                    handle_t *handle, struct inode *inode)
513  {
514         struct ext4_xattr_entry *last;
515         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
516 +       int in_inode = i->in_inode;
517 +
518 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
519 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
520 +           (EXT4_XATTR_SIZE(i->value_len) >
521 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
522 +               in_inode = 1;
523  
524         /* Compute min_offs and last. */
525         last = s->first;
526         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
527 -               if (!last->e_value_block && last->e_value_size) {
528 +               if (!last->e_value_inum && last->e_value_size) {
529                         size_t offs = le16_to_cpu(last->e_value_offs);
530                         if (offs < min_offs)
531                                 min_offs = offs;
532 @@ -628,15 +903,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
533         }
534         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
535         if (!s->not_found) {
536 -               if (!s->here->e_value_block && s->here->e_value_size) {
537 +               if (!in_inode &&
538 +                   !s->here->e_value_inum && s->here->e_value_size) {
539                         size_t size = le32_to_cpu(s->here->e_value_size);
540                         free += EXT4_XATTR_SIZE(size);
541                 }
542                 free += EXT4_XATTR_LEN(name_len);
543         }
544         if (i->value) {
545 -               if (free < EXT4_XATTR_LEN(name_len) +
546 -                          EXT4_XATTR_SIZE(i->value_len))
547 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
548 +
549 +               if (in_inode)
550 +                       value_len = 0;
551 +
552 +               if (free < value_len ||
553 +                   free < EXT4_XATTR_LEN(name_len) + value_len)
554                         return -ENOSPC;
555         }
556  
557 @@ -651,7 +931,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
558                 s->here->e_name_len = name_len;
559                 memcpy(s->here->e_name, i->name, name_len);
560         } else {
561 -               if (!s->here->e_value_block && s->here->e_value_size) {
562 +               if (!s->here->e_value_inum && s->here->e_value_size &&
563 +                   s->here->e_value_offs > 0) {
564                         void *first_val = s->base + min_offs;
565                         size_t offs = le16_to_cpu(s->here->e_value_offs);
566                         void *val = s->base + offs;
567 @@ -685,13 +966,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
568                         last = s->first;
569                         while (!IS_LAST_ENTRY(last)) {
570                                 size_t o = le16_to_cpu(last->e_value_offs);
571 -                               if (!last->e_value_block &&
572 +                               if (!last->e_value_inum &&
573                                     last->e_value_size && o < offs)
574                                         last->e_value_offs =
575                                                 cpu_to_le16(o + size);
576                                 last = EXT4_XATTR_NEXT(last);
577                         }
578                 }
579 +               if (s->here->e_value_inum) {
580 +                       ext4_xattr_inode_unlink(inode,
581 +                                       le32_to_cpu(s->here->e_value_inum));
582 +                       s->here->e_value_inum = 0;
583 +               }
584                 if (!i->value) {
585                         /* Remove the old name. */
586                         size_t size = EXT4_XATTR_LEN(name_len);
587 @@ -705,10 +990,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
588         if (i->value) {
589                 /* Insert the new value. */
590                 s->here->e_value_size = cpu_to_le32(i->value_len);
591 -               if (i->value_len) {
592 +               if (in_inode) {
593 +                       unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
594 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
595 +                                            i->value_len);
596 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
597 +                       s->here->e_value_offs = 0;
598 +               } else if (i->value_len) {
599                         size_t size = EXT4_XATTR_SIZE(i->value_len);
600                         void *val = s->base + min_offs - size;
601                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
602 +                       s->here->e_value_inum = 0;
603                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
604                                 memset(val, 0, size);
605                         } else {
606 @@ -758,7 +1050,7 @@ ext4_xattr_block_find(struct inode *inod
607                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
608                 bs->s.here = bs->s.first;
609                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
610 -                                             i->name, bs->bh->b_size, 1);
611 +                                             i->name, bs->bh->b_size, 1, inode);
612                 if (error && error != -ENODATA)
613                         goto cleanup;
614                 bs->s.not_found = error;
615 @@ -782,8 +1074,6 @@ ext4_xattr_block_set(handle_t *handle, s
616  
617  #define header(x) ((struct ext4_xattr_header *)(x))
618  
619 -       if (i->value && i->value_len > sb->s_blocksize)
620 -               return -ENOSPC;
621         if (s->base) {
622                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
623                                         bs->bh->b_blocknr);
624 @@ -799,7 +1089,7 @@ ext4_xattr_block_set(handle_t *handle, s
625                                 ce = NULL;
626                         }
627                         ea_bdebug(bs->bh, "modifying in-place");
628 -                       error = ext4_xattr_set_entry(i, s);
629 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
630                         if (!error) {
631                                 if (!IS_LAST_ENTRY(s->first))
632                                         ext4_xattr_rehash(header(s->base),
633 @@ -850,7 +1140,7 @@ ext4_xattr_block_set(handle_t *handle, s
634                 s->end = s->base + sb->s_blocksize;
635         }
636  
637 -       error = ext4_xattr_set_entry(i, s);
638 +       error = ext4_xattr_set_entry(i, s, handle, inode);
639         if (error == -EIO)
640                 goto bad_block;
641         if (error)
642 @@ -1000,7 +1290,7 @@ int ext4_xattr_ibody_find(struct inode *
643                 /* Find the named attribute. */
644                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
645                                               i->name, is->s.end -
646 -                                             (void *)is->s.base, 0);
647 +                                             (void *)is->s.base, 0, inode);
648                 if (error && error != -ENODATA)
649                         return error;
650                 is->s.not_found = error;
651 @@ -1018,7 +1308,7 @@ int ext4_xattr_ibody_inline_set(handle_t
652  
653         if (EXT4_I(inode)->i_extra_isize == 0)
654                 return -ENOSPC;
655 -       error = ext4_xattr_set_entry(i, s);
656 +       error = ext4_xattr_set_entry(i, s, handle, inode);
657         if (error) {
658                 if (error == -ENOSPC &&
659                     ext4_has_inline_data(inode)) {
660 @@ -1030,7 +1320,7 @@ int ext4_xattr_ibody_inline_set(handle_t
661                         error = ext4_xattr_ibody_find(inode, i, is);
662                         if (error)
663                                 return error;
664 -                       error = ext4_xattr_set_entry(i, s);
665 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
666                 }
667                 if (error)
668                         return error;
669 @@ -1056,7 +1346,7 @@ static int ext4_xattr_ibody_set(handle_t
670  
671         if (EXT4_I(inode)->i_extra_isize == 0)
672                 return -ENOSPC;
673 -       error = ext4_xattr_set_entry(i, s);
674 +       error = ext4_xattr_set_entry(i, s, handle, inode);
675         if (error)
676                 return error;
677         header = IHDR(inode, ext4_raw_inode(&is->iloc));
678 @@ -1092,7 +1382,7 @@ ext4_xattr_set_handle(handle_t *handle, 
679                 .name = name,
680                 .value = value,
681                 .value_len = value_len,
682 -
683 +               .in_inode = 0,
684         };
685         struct ext4_xattr_ibody_find is = {
686                 .s = { .not_found = -ENODATA, },
687 @@ -1157,6 +1447,15 @@ ext4_xattr_set_handle(handle_t *handle, 
688                                         goto cleanup;
689                         }
690                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
691 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
692 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
693 +                           error == -ENOSPC) {
694 +                               /* xattr not fit to block, store at external
695 +                                * inode */
696 +                               i.in_inode = 1;
697 +                               error = ext4_xattr_ibody_set(handle, inode,
698 +                                                            &i, &is);
699 +                       }
700                         if (error)
701                                 goto cleanup;
702                         if (!is.s.not_found) {
703 @@ -1203,9 +1502,22 @@ ext4_xattr_set(struct inode *inode, int 
704                const void *value, size_t value_len, int flags)
705  {
706         handle_t *handle;
707 +       struct super_block *sb = inode->i_sb;
708         int error, retries = 0;
709         int credits = ext4_jbd2_credits_xattr(inode);
710  
711 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
712 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
713 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
714 +                                       sb->s_blocksize_bits;
715 +
716 +               /* For new inode */
717 +               credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
718 +
719 +               /* For data blocks of EA inode */
720 +               credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
721 +       }
722 +
723  retry:
724         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
725         if (IS_ERR(handle)) {
726 @@ -1217,7 +1529,7 @@ retry:
727                                               value, value_len, flags);
728                 error2 = ext4_journal_stop(handle);
729                 if (error == -ENOSPC &&
730 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
731 +                   ext4_should_retry_alloc(sb, &retries))
732                         goto retry;
733                 if (error == 0)
734                         error = error2;
735 @@ -1239,7 +1551,7 @@ static void ext4_xattr_shift_entries(str
736  
737         /* Adjust the value offsets of the entries */
738         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
739 -               if (!last->e_value_block && last->e_value_size) {
740 +               if (!last->e_value_inum && last->e_value_size) {
741                         new_offs = le16_to_cpu(last->e_value_offs) +
742                                                         value_offs_shift;
743                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
744 @@ -1477,21 +1789,135 @@ cleanup:
745  }
746  
747  
748 +#define EIA_INCR 16 /* must be 2^n */
749 +#define EIA_MASK (EIA_INCR - 1)
750 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
751 + * If @lea_ino_array is new or full it will be grown and the old
752 + * contents copied over.
753 + */
754 +static int
755 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
756 +{
757 +       if (*lea_ino_array == NULL) {
758 +               /*
759 +                * Start with 15 inodes, so it fits into a power-of-two size.
760 +                * If *lea_ino_array is NULL, this is essentially offsetof()
761 +                */
762 +               (*lea_ino_array) =
763 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
764 +                                        xia_inodes[EIA_MASK]),
765 +                               GFP_NOFS);
766 +               if (*lea_ino_array == NULL)
767 +                       return -ENOMEM;
768 +               (*lea_ino_array)->xia_count = 0;
769 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
770 +               /* expand the array once all 15 + n * 16 slots are full */
771 +               struct ext4_xattr_ino_array *new_array = NULL;
772 +               int count = (*lea_ino_array)->xia_count;
773 +
774 +               /* if new_array is NULL, this is essentially offsetof() */
775 +               new_array = kmalloc(
776 +                               offsetof(struct ext4_xattr_ino_array,
777 +                                        xia_inodes[count + EIA_INCR]),
778 +                               GFP_NOFS);
779 +               if (new_array == NULL)
780 +                       return -ENOMEM;
781 +               memcpy(new_array, *lea_ino_array,
782 +                      offsetof(struct ext4_xattr_ino_array,
783 +                               xia_inodes[count]));
784 +               kfree(*lea_ino_array);
785 +               *lea_ino_array = new_array;
786 +       }
787 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
788 +       return 0;
789 +}
790 +
791 +/**
792 + * Add xattr inode to orphan list
793 + */
794 +static int
795 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
796 +                       int credits, struct ext4_xattr_ino_array *lea_ino_array)
797 +{
798 +       struct inode *ea_inode = NULL;
799 +       int idx = 0, error = 0;
800 +
801 +       if (lea_ino_array == NULL)
802 +               return 0;
803 +
804 +       for (; idx < lea_ino_array->xia_count; ++idx) {
805 +               if (!ext4_handle_has_enough_credits(handle, credits)) {
806 +                       error = ext4_journal_extend(handle, credits);
807 +                       if (error > 0)
808 +                               error = ext4_journal_restart(handle, credits);
809 +
810 +                       if (error != 0) {
811 +                               ext4_warning(inode->i_sb,
812 +                                       "couldn't extend journal "
813 +                                       "(err %d)", error);
814 +                               return error;
815 +                       }
816 +               }
817 +               ea_inode = ext4_xattr_inode_iget(inode,
818 +                               lea_ino_array->xia_inodes[idx], &error);
819 +               if (error)
820 +                       continue;
821 +               ext4_orphan_add(handle, ea_inode);
822 +               /* the inode's i_count will be released by caller */
823 +       }
824 +
825 +       return 0;
826 +}
827  
828  /*
829   * ext4_xattr_delete_inode()
830   *
831 - * Free extended attribute resources associated with this inode. This
832 + * Free extended attribute resources associated with this inode. Traverse
833 + * all entries and unlink any xattr inodes associated with this inode. This
834   * is called immediately before an inode is freed. We have exclusive
835 - * access to the inode.
836 + * access to the inode. If an orphan inode is deleted it will also delete any
837 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
838 + * to ensure they belong to the parent inode and were not deleted already.
839   */
840 -void
841 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
842 +int
843 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
844 +                       struct ext4_xattr_ino_array **lea_ino_array)
845  {
846         struct buffer_head *bh = NULL;
847 +       struct ext4_xattr_ibody_header *header;
848 +       struct ext4_inode *raw_inode;
849 +       struct ext4_iloc iloc;
850 +       struct ext4_xattr_entry *entry;
851 +       int credits = 3, error = 0;
852  
853 -       if (!EXT4_I(inode)->i_file_acl)
854 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
855 +               goto delete_external_ea;
856 +
857 +       error = ext4_get_inode_loc(inode, &iloc);
858 +       if (error)
859                 goto cleanup;
860 +       raw_inode = ext4_raw_inode(&iloc);
861 +       header = IHDR(inode, raw_inode);
862 +       for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
863 +            entry = EXT4_XATTR_NEXT(entry)) {
864 +               if (!entry->e_value_inum)
865 +                       continue;
866 +               if (ext4_expand_ino_array(lea_ino_array,
867 +                                         entry->e_value_inum) != 0) {
868 +                       brelse(iloc.bh);
869 +                       goto cleanup;
870 +               }
871 +               entry->e_value_inum = 0;
872 +       }
873 +       brelse(iloc.bh);
874 +
875 +delete_external_ea:
876 +       if (!EXT4_I(inode)->i_file_acl) {
877 +               /* add xattr inode to orphan list */
878 +               ext4_xattr_inode_orphan_add(handle, inode, credits,
879 +                                               *lea_ino_array);
880 +               goto cleanup;
881 +       }
882         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
883         if (!bh) {
884                 EXT4_ERROR_INODE(inode, "block %llu read error",
885 @@ -1504,11 +1930,69 @@ ext4_xattr_delete_inode(handle_t *handle
886                                  EXT4_I(inode)->i_file_acl);
887                 goto cleanup;
888         }
889 +
890 +       for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
891 +            entry = EXT4_XATTR_NEXT(entry)) {
892 +               if (!entry->e_value_inum)
893 +                       continue;
894 +               if (ext4_expand_ino_array(lea_ino_array,
895 +                                         entry->e_value_inum) != 0)
896 +                       goto cleanup;
897 +               entry->e_value_inum = 0;
898 +       }
899 +
900 +       /* add xattr inode to orphan list */
901 +       error = ext4_xattr_inode_orphan_add(handle, inode, credits,
902 +                                       *lea_ino_array);
903 +       if (error != 0)
904 +               goto cleanup;
905 +
906 +       if (!IS_NOQUOTA(inode))
907 +               credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
908 +
909 +       if (!ext4_handle_has_enough_credits(handle, credits)) {
910 +               error = ext4_journal_extend(handle, credits);
911 +               if (error > 0)
912 +                       error = ext4_journal_restart(handle, credits);
913 +               if (error != 0) {
914 +                       ext4_warning(inode->i_sb,
915 +                               "couldn't extend journal (err %d)", error);
916 +                       goto cleanup;
917 +               }
918 +       }
919 +
920         ext4_xattr_release_block(handle, inode, bh);
921         EXT4_I(inode)->i_file_acl = 0;
922  
923  cleanup:
924         brelse(bh);
925 +
926 +       return error;
927 +}
928 +
929 +void
930 +ext4_xattr_inode_array_free(struct inode *inode,
931 +                           struct ext4_xattr_ino_array *lea_ino_array)
932 +{
933 +       struct inode    *ea_inode = NULL;
934 +       int             idx = 0;
935 +       int             err;
936 +
937 +       if (lea_ino_array == NULL)
938 +               return;
939 +
940 +       for (; idx < lea_ino_array->xia_count; ++idx) {
941 +               ea_inode = ext4_xattr_inode_iget(inode,
942 +                               lea_ino_array->xia_inodes[idx], &err);
943 +               if (err)
944 +                       continue;
945 +               /* for inode's i_count get from ext4_xattr_delete_inode */
946 +               if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
947 +                       iput(ea_inode);
948 +               clear_nlink(ea_inode);
949 +               iput(ea_inode);
950 +       }
951 +       kfree(lea_ino_array);
952  }
953  
954  /*
955 @@ -1578,10 +2062,9 @@ ext4_xattr_cmp(struct ext4_xattr_header 
956                     entry1->e_name_index != entry2->e_name_index ||
957                     entry1->e_name_len != entry2->e_name_len ||
958                     entry1->e_value_size != entry2->e_value_size ||
959 +                   entry1->e_value_inum != entry2->e_value_inum ||
960                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
961                         return 1;
962 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
963 -                       return -EIO;
964                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
965                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
966                            le32_to_cpu(entry1->e_value_size)))
967 @@ -1665,7 +2148,7 @@ static inline void ext4_xattr_hash_entry
968                        *name++;
969         }
970  
971 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
972 +       if (!entry->e_value_inum && entry->e_value_size) {
973                 __le32 *value = (__le32 *)((char *)header +
974                         le16_to_cpu(entry->e_value_offs));
975                 for (n = (le32_to_cpu(entry->e_value_size) +
976 Index: linux-stage/fs/ext4/xattr.h
977 ===================================================================
978 --- linux-stage.orig/fs/ext4/xattr.h
979 +++ linux-stage/fs/ext4/xattr.h
980 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
981         __u8    e_name_len;     /* length of name */
982         __u8    e_name_index;   /* attribute name index */
983         __le16  e_value_offs;   /* offset in disk block of value */
984 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
985 +       __le32  e_value_inum;   /* inode in which the value is stored */
986         __le32  e_value_size;   /* size of attribute value */
987         __le32  e_hash;         /* hash value of name and value */
988         char    e_name[0];      /* attribute name */
989 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
990                 EXT4_I(inode)->i_extra_isize))
991  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
992  
993 +/*
994 + * Link EA inode back to parent one using i_mtime field.
995 + * Extra integer type conversion added to ignore higher
996 + * bits in i_mtime.tv_sec which might be set by ext4_get()
997 + */
998 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum)       \
999 +do {                                                   \
1000 +       (inode)->i_mtime.tv_sec = inum;                 \
1001 +} while(0)
1002 +
1003 +#define EXT4_XATTR_INODE_GET_PARENT(inode)             \
1004 +       ((__u32)(inode)->i_mtime.tv_sec)
1005 +
1006 +/*
1007 + * The minimum size of EA value when you start storing it in an external inode
1008 + * size of block - size of header - size of 1 entry - 4 null bytes
1009 +*/
1010 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
1011 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1012 +
1013  #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1014  #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1015  #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1016 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1017  #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1018  
1019  struct ext4_xattr_info {
1020 -       int name_index;
1021         const char *name;
1022         const void *value;
1023         size_t value_len;
1024 +       int name_index;
1025 +       int in_inode;
1026  };
1027  
1028  struct ext4_xattr_search {
1029 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1030  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1031  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1032  
1033 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1034 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1035 +                                          int *err);
1036 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1037 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1038 +                                  struct ext4_xattr_ino_array **array);
1039 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1040 +                                       struct ext4_xattr_ino_array *array);
1041  extern void ext4_xattr_put_super(struct super_block *);
1042  
1043  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1044 Index: linux-stage/fs/ext4/ialloc.c
1045 ===================================================================
1046 --- linux-stage.orig/fs/ext4/ialloc.c
1047 +++ linux-stage/fs/ext4/ialloc.c
1048 @@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
1049          * as writing the quota to disk may need the lock as well.
1050          */
1051         dquot_initialize(inode);
1052 -       ext4_xattr_delete_inode(handle, inode);
1053         dquot_free_inode(inode);
1054         dquot_drop(inode);
1055  
1056 Index: linux-stage/fs/ext4/inline.c
1057 ===================================================================
1058 --- linux-stage.orig/fs/ext4/inline.c
1059 +++ linux-stage/fs/ext4/inline.c
1060 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1061  
1062         /* Compute min_offs. */
1063         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1064 -               if (!entry->e_value_block && entry->e_value_size) {
1065 +               if (!entry->e_value_inum && entry->e_value_size) {
1066                         size_t offs = le16_to_cpu(entry->e_value_offs);
1067                         if (offs < min_offs)
1068                                 min_offs = offs;