Whamcloud - gitweb
LU-11071 build: Add server build support for Ubuntu 18.04
[fs/lustre-release.git] / ldiskfs / kernel_patches / patches / sles12sp2 / 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 @@ -1990,6 +1997,10 @@ struct mmpd_data {
20  # define ATTRIB_NORET  __attribute__((noreturn))
21  # define NORET_AND     noreturn,
22  
23 +struct ext4_xattr_ino_array {
24 +       unsigned int xia_count;         /* # of used item in the array */
25 +       unsigned int xia_inodes[0];
26 +};
27  /* bitmap.c */
28  extern unsigned int ext4_count_free(char *bitmap, unsigned numchars);
29  void ext4_inode_bitmap_csum_set(struct super_block *sb, ext4_group_t group,
30 @@ -2194,6 +2205,7 @@ extern void ext4_set_inode_flags(struct
31  extern void ext4_get_inode_flags(struct ext4_inode_info *);
32  extern int ext4_alloc_da_blocks(struct inode *inode);
33  extern void ext4_set_aops(struct inode *inode);
34 +extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk);
35  extern int ext4_writepage_trans_blocks(struct inode *);
36  extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
37  extern int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
38 Index: linux-stage/fs/ext4/inode.c
39 ===================================================================
40 --- linux-stage.orig/fs/ext4/inode.c
41 +++ linux-stage/fs/ext4/inode.c
42 @@ -134,8 +134,6 @@ static void ext4_invalidatepage(struct p
43                                 unsigned int length);
44  static int __ext4_journalled_writepage(struct page *page, unsigned int len);
45  static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
46 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
47 -                                 int pextents);
48  
49  /*
50   * Test whether an inode is a fast symlink.
51 @@ -184,6 +182,8 @@ void ext4_evict_inode(struct inode *inod
52  {
53         handle_t *handle;
54         int err;
55 +       int extra_credits = 3;
56 +       struct ext4_xattr_ino_array *lea_ino_array = NULL;
57  
58         trace_ext4_evict_inode(inode);
59  
60 @@ -236,8 +236,8 @@ void ext4_evict_inode(struct inode *inod
61          * protection against it
62          */
63         sb_start_intwrite(inode->i_sb);
64 -       handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
65 -                                   ext4_blocks_for_truncate(inode)+3);
66 +
67 +       handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits);
68         if (IS_ERR(handle)) {
69                 ext4_std_error(inode->i_sb, PTR_ERR(handle));
70                 /*
71 @@ -249,9 +249,36 @@ void ext4_evict_inode(struct inode *inod
72                 sb_end_intwrite(inode->i_sb);
73                 goto no_delete;
74         }
75 -
76         if (IS_SYNC(inode))
77                 ext4_handle_sync(handle);
78 +
79 +       /*
80 +        * Delete xattr inode before deleting the main inode.
81 +        */
82 +       err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
83 +       if (err) {
84 +               ext4_warning(inode->i_sb,
85 +                            "couldn't delete inode's xattr (err %d)", err);
86 +               goto stop_handle;
87 +       }
88 +
89 +       if (!IS_NOQUOTA(inode))
90 +               extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
91 +
92 +       if (!ext4_handle_has_enough_credits(handle,
93 +                       ext4_blocks_for_truncate(inode) + extra_credits)) {
94 +               err = ext4_journal_extend(handle,
95 +                       ext4_blocks_for_truncate(inode) + extra_credits);
96 +               if (err > 0)
97 +                       err = ext4_journal_restart(handle,
98 +                       ext4_blocks_for_truncate(inode) + extra_credits);
99 +               if (err != 0) {
100 +                       ext4_warning(inode->i_sb,
101 +                                    "couldn't extend journal (err %d)", err);
102 +                       goto stop_handle;
103 +               }
104 +       }
105 +
106         inode->i_size = 0;
107         err = ext4_mark_inode_dirty(handle, inode);
108         if (err) {
109 @@ -269,10 +296,10 @@ void ext4_evict_inode(struct inode *inod
110          * enough credits left in the handle to remove the inode from
111          * the orphan list and set the dtime field.
112          */
113 -       if (!ext4_handle_has_enough_credits(handle, 3)) {
114 -               err = ext4_journal_extend(handle, 3);
115 +       if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
116 +               err = ext4_journal_extend(handle, extra_credits);
117                 if (err > 0)
118 -                       err = ext4_journal_restart(handle, 3);
119 +                       err = ext4_journal_restart(handle, extra_credits);
120                 if (err != 0) {
121                         ext4_warning(inode->i_sb,
122                                      "couldn't extend journal (err %d)", err);
123 @@ -306,8 +333,12 @@ void ext4_evict_inode(struct inode *inod
124                 ext4_clear_inode(inode);
125         else
126                 ext4_free_inode(handle, inode);
127 +
128         ext4_journal_stop(handle);
129         sb_end_intwrite(inode->i_sb);
130 +
131 +       if (lea_ino_array != NULL)
132 +               ext4_xattr_inode_array_free(inode, lea_ino_array);
133         return;
134  no_delete:
135         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
136 @@ -4681,7 +4712,7 @@ static int ext4_index_trans_blocks(struc
137   *
138   * Also account for superblock, inode, quota and xattr blocks
139   */
140 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
141 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
142                                   int pextents)
143  {
144         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
145 Index: linux-stage/fs/ext4/xattr.c
146 ===================================================================
147 --- linux-stage.orig/fs/ext4/xattr.c
148 +++ linux-stage/fs/ext4/xattr.c
149 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
150  
151         while (!IS_LAST_ENTRY(entry)) {
152                 if (entry->e_value_size != 0 &&
153 +                   entry->e_value_inum == 0 &&
154                     (value_start + le16_to_cpu(entry->e_value_offs) <
155                      (void *)e + sizeof(__u32) ||
156                      value_start + le16_to_cpu(entry->e_value_offs) +
157 @@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
158  }
159  
160  static inline int
161 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
162 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
163 +                      struct inode *inode)
164  {
165         size_t value_size = le32_to_cpu(entry->e_value_size);
166  
167 -       if (entry->e_value_block != 0 || value_size > size ||
168 +       if (!entry->e_value_inum &&
169             le16_to_cpu(entry->e_value_offs) + value_size > size)
170                 return -EFSCORRUPTED;
171 +       if (entry->e_value_inum &&
172 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
173 +            le32_to_cpu(entry->e_value_inum) >
174 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
175 +               return -EFSCORRUPTED;
176         return 0;
177  }
178  
179  static int
180  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
181 -                     const char *name, size_t size, int sorted)
182 +                     const char *name, size_t size, int sorted,
183 +                     struct inode *inode)
184  {
185         struct ext4_xattr_entry *entry;
186         size_t name_len;
187 @@ -265,11 +273,104 @@ ext4_xattr_find_entry(struct ext4_xattr_
188                         break;
189         }
190         *pentry = entry;
191 -       if (!cmp && ext4_xattr_check_entry(entry, size))
192 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
193                 return -EFSCORRUPTED;
194         return cmp ? -ENODATA : 0;
195  }
196  
197 +/*
198 + * Read the EA value from an inode.
199 + */
200 +static int
201 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
202 +{
203 +       unsigned long block = 0;
204 +       struct buffer_head *bh = NULL;
205 +       int blocksize;
206 +       size_t csize, ret_size = 0;
207 +
208 +       if (*size == 0)
209 +               return 0;
210 +
211 +       blocksize = ea_inode->i_sb->s_blocksize;
212 +
213 +       while (ret_size < *size) {
214 +               csize = (*size - ret_size) > blocksize ? blocksize :
215 +                                                       *size - ret_size;
216 +               bh = ext4_bread(NULL, ea_inode, block, 0);
217 +               if (IS_ERR(bh)) {
218 +                       *size = ret_size;
219 +                       return PTR_ERR(bh);
220 +               }
221 +               memcpy(buf, bh->b_data, csize);
222 +               brelse(bh);
223 +
224 +               buf += csize;
225 +               block += 1;
226 +               ret_size += csize;
227 +       }
228 +
229 +       *size = ret_size;
230 +
231 +       return 0;
232 +}
233 +
234 +struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
235 +{
236 +       struct inode *ea_inode = NULL;
237 +
238 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
239 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
240 +               int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
241 +               ext4_error(parent->i_sb, "error while reading EA inode %lu "
242 +                          "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
243 +               *err = rc != 0 ? rc : -EIO;
244 +               return NULL;
245 +       }
246 +
247 +       if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
248 +           ea_inode->i_generation != parent->i_generation) {
249 +               ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
250 +                          "to parent invalid.", ea_ino);
251 +               *err = -EINVAL;
252 +               goto error;
253 +       }
254 +
255 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
256 +               ext4_error(parent->i_sb, "EA inode %lu does not have "
257 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
258 +               *err = -EINVAL;
259 +               goto error;
260 +       }
261 +
262 +       *err = 0;
263 +       return ea_inode;
264 +
265 +error:
266 +       iput(ea_inode);
267 +       return NULL;
268 +}
269 +
270 +/*
271 + * Read the value from the EA inode.
272 + */
273 +static int
274 +ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
275 +                    size_t *size)
276 +{
277 +       struct inode *ea_inode = NULL;
278 +       int err;
279 +
280 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
281 +       if (err)
282 +               return err;
283 +
284 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
285 +       iput(ea_inode);
286 +
287 +       return err;
288 +}
289 +
290  static int
291  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
292                      void *buffer, size_t buffer_size)
293 @@ -301,7 +401,8 @@ bad_block:
294         }
295         ext4_xattr_cache_insert(bh);
296         entry = BFIRST(bh);
297 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
298 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
299 +                                     inode);
300         if (error == -EFSCORRUPTED)
301                 goto bad_block;
302         if (error)
303 @@ -311,8 +412,16 @@ bad_block:
304                 error = -ERANGE;
305                 if (size > buffer_size)
306                         goto cleanup;
307 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
308 -                      size);
309 +               if (entry->e_value_inum) {
310 +                       error = ext4_xattr_inode_get(inode,
311 +                                            le32_to_cpu(entry->e_value_inum),
312 +                                            buffer, &size);
313 +                       if (error)
314 +                               goto cleanup;
315 +               } else {
316 +                       memcpy(buffer, bh->b_data +
317 +                              le16_to_cpu(entry->e_value_offs), size);
318 +               }
319         }
320         error = size;
321  
322 @@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
323         if (error)
324                 goto cleanup;
325         error = ext4_xattr_find_entry(&entry, name_index, name,
326 -                                     end - (void *)entry, 0);
327 +                                     end - (void *)entry, 0, inode);
328         if (error)
329                 goto cleanup;
330         size = le32_to_cpu(entry->e_value_size);
331 @@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
332                 error = -ERANGE;
333                 if (size > buffer_size)
334                         goto cleanup;
335 -               memcpy(buffer, (void *)IFIRST(header) +
336 -                      le16_to_cpu(entry->e_value_offs), size);
337 +               if (entry->e_value_inum) {
338 +                       error = ext4_xattr_inode_get(inode,
339 +                                            le32_to_cpu(entry->e_value_inum),
340 +                                            buffer, &size);
341 +                       if (error)
342 +                               goto cleanup;
343 +               } else {
344 +                       memcpy(buffer, (void *)IFIRST(header) +
345 +                              le16_to_cpu(entry->e_value_offs), size);
346 +               }
347         }
348         error = size;
349  
350 @@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
351                                     size_t *min_offs, void *base, int *total)
352  {
353         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
354 -               if (!last->e_value_block && last->e_value_size) {
355 +               if (!last->e_value_inum && last->e_value_size) {
356                         size_t offs = le16_to_cpu(last->e_value_offs);
357                         if (offs < *min_offs)
358                                 *min_offs = offs;
359 @@ -611,16 +728,200 @@ static size_t ext4_xattr_free_space(stru
360         return (*min_offs - ((void *)last - base) - sizeof(__u32));
361  }
362  
363 +/*
364 + * Write the value of the EA in an inode.
365 + */
366  static int
367 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
368 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
369 +                      const void *buf, int bufsize)
370 +{
371 +       struct buffer_head *bh = NULL;
372 +       unsigned long block = 0;
373 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
374 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
375 +       int csize, wsize = 0;
376 +       int ret = 0;
377 +       int retries = 0;
378 +
379 +retry:
380 +       while (ret >= 0 && ret < max_blocks) {
381 +               struct ext4_map_blocks map;
382 +               map.m_lblk = block += ret;
383 +               map.m_len = max_blocks -= ret;
384 +
385 +               ret = ext4_map_blocks(handle, ea_inode, &map,
386 +                                     EXT4_GET_BLOCKS_CREATE);
387 +               if (ret <= 0) {
388 +                       ext4_mark_inode_dirty(handle, ea_inode);
389 +                       if (ret == -ENOSPC &&
390 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
391 +                               ret = 0;
392 +                               goto retry;
393 +                       }
394 +                       break;
395 +               }
396 +       }
397 +
398 +       if (ret < 0)
399 +               return ret;
400 +
401 +       block = 0;
402 +       while (wsize < bufsize) {
403 +               if (bh != NULL)
404 +                       brelse(bh);
405 +               csize = (bufsize - wsize) > blocksize ? blocksize :
406 +                                                               bufsize - wsize;
407 +               bh = ext4_getblk(handle, ea_inode, block, 0);
408 +               if (IS_ERR(bh)) {
409 +                       ret = PTR_ERR(bh);
410 +                       goto out;
411 +               }
412 +               ret = ext4_journal_get_write_access(handle, bh);
413 +               if (ret)
414 +                       goto out;
415 +
416 +               memcpy(bh->b_data, buf, csize);
417 +               set_buffer_uptodate(bh);
418 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
419 +
420 +               buf += csize;
421 +               wsize += csize;
422 +               block += 1;
423 +       }
424 +
425 +       mutex_lock(&ea_inode->i_mutex);
426 +       i_size_write(ea_inode, wsize);
427 +       ext4_update_i_disksize(ea_inode, wsize);
428 +       mutex_unlock(&ea_inode->i_mutex);
429 +
430 +       ext4_mark_inode_dirty(handle, ea_inode);
431 +
432 +out:
433 +       brelse(bh);
434 +
435 +       return ret;
436 +}
437 +
438 +static void ext4_xattr_inode_set_ref(struct inode *ea_inode, __u64 ref_count)
439 +{
440 +       ea_inode->i_ctime.tv_sec = (__u32)(ref_count >> 32);
441 +       ea_inode->i_version = (__u32)ref_count;
442 +}
443 +
444 +static void ext4_xattr_inode_set_hash(struct inode *ea_inode, __u32 hash)
445 +{
446 +       ea_inode->i_atime.tv_sec = hash;
447 +}
448 +
449 +/*
450 + * Create an inode to store the value of a large EA.
451 + */
452 +static struct inode *
453 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode, __u32 hash)
454 +{
455 +       struct inode *ea_inode = NULL;
456 +
457 +       /*
458 +        * Let the next inode be the goal, so we try and allocate the EA inode
459 +        * in the same group, or nearby one.
460 +        */
461 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
462 +                                 S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
463 +
464 +       if (!IS_ERR(ea_inode)) {
465 +               ea_inode->i_op = &ext4_file_inode_operations;
466 +               ea_inode->i_fop = &ext4_file_operations;
467 +               ext4_set_aops(ea_inode);
468 +               ea_inode->i_generation = inode->i_generation;
469 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
470 +
471 +               /*
472 +                * A back-pointer from EA inode to parent inode will be useful
473 +                * for e2fsck.
474 +                */
475 +               EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
476 +               unlock_new_inode(ea_inode);
477 +
478 +               ext4_xattr_inode_set_ref(ea_inode, 1);
479 +               ext4_xattr_inode_set_hash(ea_inode, hash);
480 +       }
481 +
482 +       return ea_inode;
483 +}
484 +
485 +/*
486 + * Unlink the inode storing the value of the EA.
487 + */
488 +int
489 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
490 +{
491 +       struct inode *ea_inode = NULL;
492 +       int err;
493 +
494 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
495 +       if (err)
496 +               return err;
497 +
498 +       clear_nlink(ea_inode);
499 +       iput(ea_inode);
500 +
501 +       return 0;
502 +}
503 +
504 +static __u32
505 +ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
506 +{
507 +       if (ext4_has_metadata_csum(sbi->s_sb))
508 +               return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
509 +       return 0;
510 +}
511 +
512 +/*
513 + * Add value of the EA in an inode.
514 + */
515 +static int
516 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
517 +                    const void *value, size_t value_len)
518 +{
519 +       struct inode *ea_inode = NULL;
520 +       __u32 hash;
521 +       int err;
522 +
523 +       /* Create an inode for the EA value */
524 +       hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
525 +       ea_inode = ext4_xattr_inode_create(handle, inode, hash);
526 +       if (IS_ERR(ea_inode))
527 +               return -1;
528 +
529 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
530 +       if (err)
531 +               clear_nlink(ea_inode);
532 +       else
533 +               *ea_ino = ea_inode->i_ino;
534 +
535 +       iput(ea_inode);
536 +
537 +       return err;
538 +}
539 +
540 +static int
541 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
542 +                    handle_t *handle, struct inode *inode)
543  {
544         struct ext4_xattr_entry *last;
545         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
546 +       int in_inode = i->in_inode;
547 +
548 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
549 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
550 +           (EXT4_XATTR_SIZE(i->value_len) >
551 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
552 +               in_inode = 1;
553  
554         /* Compute min_offs and last. */
555         last = s->first;
556         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
557 -               if (!last->e_value_block && last->e_value_size) {
558 +               if (!last->e_value_inum && last->e_value_size) {
559                         size_t offs = le16_to_cpu(last->e_value_offs);
560                         if (offs < min_offs)
561                                 min_offs = offs;
562 @@ -628,15 +927,20 @@ ext4_xattr_set_entry(struct ext4_xattr_i
563         }
564         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
565         if (!s->not_found) {
566 -               if (!s->here->e_value_block && s->here->e_value_size) {
567 +               if (!in_inode &&
568 +                   !s->here->e_value_inum && s->here->e_value_size) {
569                         size_t size = le32_to_cpu(s->here->e_value_size);
570                         free += EXT4_XATTR_SIZE(size);
571                 }
572                 free += EXT4_XATTR_LEN(name_len);
573         }
574         if (i->value) {
575 -               if (free < EXT4_XATTR_LEN(name_len) +
576 -                          EXT4_XATTR_SIZE(i->value_len))
577 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
578 +
579 +               if (in_inode)
580 +                       value_len = 0;
581 +
582 +               if (free < EXT4_XATTR_LEN(name_len) + value_len)
583                         return -ENOSPC;
584         }
585  
586 @@ -651,7 +955,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
587                 s->here->e_name_len = name_len;
588                 memcpy(s->here->e_name, i->name, name_len);
589         } else {
590 -               if (!s->here->e_value_block && s->here->e_value_size) {
591 +               if (!s->here->e_value_inum && s->here->e_value_size &&
592 +                   s->here->e_value_offs > 0) {
593                         void *first_val = s->base + min_offs;
594                         size_t offs = le16_to_cpu(s->here->e_value_offs);
595                         void *val = s->base + offs;
596 @@ -685,13 +990,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
597                         last = s->first;
598                         while (!IS_LAST_ENTRY(last)) {
599                                 size_t o = le16_to_cpu(last->e_value_offs);
600 -                               if (!last->e_value_block &&
601 +                               if (!last->e_value_inum &&
602                                     last->e_value_size && o < offs)
603                                         last->e_value_offs =
604                                                 cpu_to_le16(o + size);
605                                 last = EXT4_XATTR_NEXT(last);
606                         }
607                 }
608 +               if (s->here->e_value_inum) {
609 +                       ext4_xattr_inode_unlink(inode,
610 +                                       le32_to_cpu(s->here->e_value_inum));
611 +                       s->here->e_value_inum = 0;
612 +               }
613                 if (!i->value) {
614                         /* Remove the old name. */
615                         size_t size = EXT4_XATTR_LEN(name_len);
616 @@ -705,10 +1014,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
617         if (i->value) {
618                 /* Insert the new value. */
619                 s->here->e_value_size = cpu_to_le32(i->value_len);
620 -               if (i->value_len) {
621 +               if (in_inode) {
622 +                       unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
623 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
624 +                                            i->value_len);
625 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
626 +                       s->here->e_value_offs = 0;
627 +               } else if (i->value_len) {
628                         size_t size = EXT4_XATTR_SIZE(i->value_len);
629                         void *val = s->base + min_offs - size;
630                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
631 +                       s->here->e_value_inum = 0;
632                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
633                                 memset(val, 0, size);
634                         } else {
635 @@ -758,7 +1074,7 @@ ext4_xattr_block_find(struct inode *inod
636                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
637                 bs->s.here = bs->s.first;
638                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
639 -                                             i->name, bs->bh->b_size, 1);
640 +                                            i->name, bs->bh->b_size, 1, inode);
641                 if (error && error != -ENODATA)
642                         goto cleanup;
643                 bs->s.not_found = error;
644 @@ -782,8 +1098,6 @@ ext4_xattr_block_set(handle_t *handle, s
645  
646  #define header(x) ((struct ext4_xattr_header *)(x))
647  
648 -       if (i->value && i->value_len > sb->s_blocksize)
649 -               return -ENOSPC;
650         if (s->base) {
651                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
652                                         bs->bh->b_blocknr);
653 @@ -799,7 +1113,7 @@ ext4_xattr_block_set(handle_t *handle, s
654                                 ce = NULL;
655                         }
656                         ea_bdebug(bs->bh, "modifying in-place");
657 -                       error = ext4_xattr_set_entry(i, s);
658 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
659                         if (!error) {
660                                 if (!IS_LAST_ENTRY(s->first))
661                                         ext4_xattr_rehash(header(s->base),
662 @@ -850,7 +1164,7 @@ ext4_xattr_block_set(handle_t *handle, s
663                 s->end = s->base + sb->s_blocksize;
664         }
665  
666 -       error = ext4_xattr_set_entry(i, s);
667 +       error = ext4_xattr_set_entry(i, s, handle, inode);
668         if (error == -EFSCORRUPTED)
669                 goto bad_block;
670         if (error)
671 @@ -1000,7 +1314,7 @@ int ext4_xattr_ibody_find(struct inode *
672                 /* Find the named attribute. */
673                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
674                                               i->name, is->s.end -
675 -                                             (void *)is->s.base, 0);
676 +                                             (void *)is->s.base, 0, inode);
677                 if (error && error != -ENODATA)
678                         return error;
679                 is->s.not_found = error;
680 @@ -1018,7 +1332,7 @@ int ext4_xattr_ibody_inline_set(handle_t
681  
682         if (EXT4_I(inode)->i_extra_isize == 0)
683                 return -ENOSPC;
684 -       error = ext4_xattr_set_entry(i, s);
685 +       error = ext4_xattr_set_entry(i, s, handle, inode);
686         if (error) {
687                 if (error == -ENOSPC &&
688                     ext4_has_inline_data(inode)) {
689 @@ -1030,7 +1344,7 @@ int ext4_xattr_ibody_inline_set(handle_t
690                         error = ext4_xattr_ibody_find(inode, i, is);
691                         if (error)
692                                 return error;
693 -                       error = ext4_xattr_set_entry(i, s);
694 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
695                 }
696                 if (error)
697                         return error;
698 @@ -1056,7 +1370,7 @@ static int ext4_xattr_ibody_set(handle_t
699  
700         if (EXT4_I(inode)->i_extra_isize == 0)
701                 return -ENOSPC;
702 -       error = ext4_xattr_set_entry(i, s);
703 +       error = ext4_xattr_set_entry(i, s, handle, inode);
704         if (error)
705                 return error;
706         header = IHDR(inode, ext4_raw_inode(&is->iloc));
707 @@ -1092,7 +1406,7 @@ ext4_xattr_set_handle(handle_t *handle,
708                 .name = name,
709                 .value = value,
710                 .value_len = value_len,
711 -
712 +               .in_inode = 0,
713         };
714         struct ext4_xattr_ibody_find is = {
715                 .s = { .not_found = -ENODATA, },
716 @@ -1157,6 +1471,15 @@ ext4_xattr_set_handle(handle_t *handle,
717                                         goto cleanup;
718                         }
719                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
720 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
721 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
722 +                           error == -ENOSPC) {
723 +                               /* xattr not fit to block, store at external
724 +                                * inode */
725 +                               i.in_inode = 1;
726 +                               error = ext4_xattr_ibody_set(handle, inode,
727 +                                                            &i, &is);
728 +                       }
729                         if (error)
730                                 goto cleanup;
731                         if (!is.s.not_found) {
732 @@ -1203,9 +1526,22 @@ ext4_xattr_set(struct inode *inode, int
733                const void *value, size_t value_len, int flags)
734  {
735         handle_t *handle;
736 +       struct super_block *sb = inode->i_sb;
737         int error, retries = 0;
738         int credits = ext4_jbd2_credits_xattr(inode);
739  
740 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
741 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
742 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
743 +                                       sb->s_blocksize_bits;
744 +
745 +               /* For new inode */
746 +               credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
747 +
748 +               /* For data blocks of EA inode */
749 +               credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
750 +       }
751 +
752  retry:
753         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
754         if (IS_ERR(handle)) {
755 @@ -1217,7 +1553,7 @@ retry:
756                                               value, value_len, flags);
757                 error2 = ext4_journal_stop(handle);
758                 if (error == -ENOSPC &&
759 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
760 +                   ext4_should_retry_alloc(sb, &retries))
761                         goto retry;
762                 if (error == 0)
763                         error = error2;
764 @@ -1239,7 +1575,7 @@ static void ext4_xattr_shift_entries(str
765  
766         /* Adjust the value offsets of the entries */
767         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
768 -               if (!last->e_value_block && last->e_value_size) {
769 +               if (!last->e_value_inum && last->e_value_size) {
770                         new_offs = le16_to_cpu(last->e_value_offs) +
771                                                         value_offs_shift;
772                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
773 @@ -1477,21 +1813,135 @@ cleanup:
774  }
775  
776  
777 +#define EIA_INCR 16 /* must be 2^n */
778 +#define EIA_MASK (EIA_INCR - 1)
779 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
780 + * If @lea_ino_array is new or full it will be grown and the old
781 + * contents copied over.
782 + */
783 +static int
784 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
785 +{
786 +       if (*lea_ino_array == NULL) {
787 +               /*
788 +                * Start with 15 inodes, so it fits into a power-of-two size.
789 +                * If *lea_ino_array is NULL, this is essentially offsetof()
790 +                */
791 +               (*lea_ino_array) =
792 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
793 +                                        xia_inodes[EIA_MASK]),
794 +                               GFP_NOFS);
795 +               if (*lea_ino_array == NULL)
796 +                       return -ENOMEM;
797 +               (*lea_ino_array)->xia_count = 0;
798 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
799 +               /* expand the array once all 15 + n * 16 slots are full */
800 +               struct ext4_xattr_ino_array *new_array = NULL;
801 +               int count = (*lea_ino_array)->xia_count;
802 +
803 +               /* if new_array is NULL, this is essentially offsetof() */
804 +               new_array = kmalloc(
805 +                               offsetof(struct ext4_xattr_ino_array,
806 +                                        xia_inodes[count + EIA_INCR]),
807 +                               GFP_NOFS);
808 +               if (new_array == NULL)
809 +                       return -ENOMEM;
810 +               memcpy(new_array, *lea_ino_array,
811 +                      offsetof(struct ext4_xattr_ino_array,
812 +                               xia_inodes[count]));
813 +               kfree(*lea_ino_array);
814 +               *lea_ino_array = new_array;
815 +       }
816 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
817 +       return 0;
818 +}
819 +
820 +/**
821 + * Add xattr inode to orphan list
822 + */
823 +static int
824 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
825 +                       int credits, struct ext4_xattr_ino_array *lea_ino_array)
826 +{
827 +       struct inode *ea_inode = NULL;
828 +       int idx = 0, error = 0;
829 +
830 +       if (lea_ino_array == NULL)
831 +               return 0;
832 +
833 +       for (; idx < lea_ino_array->xia_count; ++idx) {
834 +               if (!ext4_handle_has_enough_credits(handle, credits)) {
835 +                       error = ext4_journal_extend(handle, credits);
836 +                       if (error > 0)
837 +                               error = ext4_journal_restart(handle, credits);
838 +
839 +                       if (error != 0) {
840 +                               ext4_warning(inode->i_sb,
841 +                                       "couldn't extend journal "
842 +                                       "(err %d)", error);
843 +                               return error;
844 +                       }
845 +               }
846 +               ea_inode = ext4_xattr_inode_iget(inode,
847 +                               lea_ino_array->xia_inodes[idx], &error);
848 +               if (error)
849 +                       continue;
850 +               ext4_orphan_add(handle, ea_inode);
851 +               /* the inode's i_count will be released by caller */
852 +       }
853 +
854 +       return 0;
855 +}
856  
857  /*
858   * ext4_xattr_delete_inode()
859   *
860 - * Free extended attribute resources associated with this inode. This
861 + * Free extended attribute resources associated with this inode. Traverse
862 + * all entries and unlink any xattr inodes associated with this inode. This
863   * is called immediately before an inode is freed. We have exclusive
864 - * access to the inode.
865 + * access to the inode. If an orphan inode is deleted it will also delete any
866 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
867 + * to ensure they belong to the parent inode and were not deleted already.
868   */
869 -void
870 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
871 +int
872 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
873 +                       struct ext4_xattr_ino_array **lea_ino_array)
874  {
875         struct buffer_head *bh = NULL;
876 +       struct ext4_xattr_ibody_header *header;
877 +       struct ext4_inode *raw_inode;
878 +       struct ext4_iloc iloc;
879 +       struct ext4_xattr_entry *entry;
880 +       int credits = 3, error = 0;
881  
882 -       if (!EXT4_I(inode)->i_file_acl)
883 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
884 +               goto delete_external_ea;
885 +
886 +       error = ext4_get_inode_loc(inode, &iloc);
887 +       if (error)
888                 goto cleanup;
889 +       raw_inode = ext4_raw_inode(&iloc);
890 +       header = IHDR(inode, raw_inode);
891 +       for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
892 +            entry = EXT4_XATTR_NEXT(entry)) {
893 +               if (!entry->e_value_inum)
894 +                       continue;
895 +               if (ext4_expand_ino_array(lea_ino_array,
896 +                                         entry->e_value_inum) != 0) {
897 +                       brelse(iloc.bh);
898 +                       goto cleanup;
899 +               }
900 +               entry->e_value_inum = 0;
901 +       }
902 +       brelse(iloc.bh);
903 +
904 +delete_external_ea:
905 +       if (!EXT4_I(inode)->i_file_acl) {
906 +               /* add xattr inode to orphan list */
907 +               ext4_xattr_inode_orphan_add(handle, inode, credits,
908 +                                               *lea_ino_array);
909 +               goto cleanup;
910 +       }
911         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
912         if (!bh) {
913                 EXT4_ERROR_INODE(inode, "block %llu read error",
914 @@ -1504,11 +1954,69 @@ ext4_xattr_delete_inode(handle_t *handle
915                                  EXT4_I(inode)->i_file_acl);
916                 goto cleanup;
917         }
918 +
919 +       for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
920 +            entry = EXT4_XATTR_NEXT(entry)) {
921 +               if (!entry->e_value_inum)
922 +                       continue;
923 +               if (ext4_expand_ino_array(lea_ino_array,
924 +                                         entry->e_value_inum) != 0)
925 +                       goto cleanup;
926 +               entry->e_value_inum = 0;
927 +       }
928 +
929 +       /* add xattr inode to orphan list */
930 +       error = ext4_xattr_inode_orphan_add(handle, inode, credits,
931 +                                       *lea_ino_array);
932 +       if (error != 0)
933 +               goto cleanup;
934 +
935 +       if (!IS_NOQUOTA(inode))
936 +               credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
937 +
938 +       if (!ext4_handle_has_enough_credits(handle, credits)) {
939 +               error = ext4_journal_extend(handle, credits);
940 +               if (error > 0)
941 +                       error = ext4_journal_restart(handle, credits);
942 +               if (error != 0) {
943 +                       ext4_warning(inode->i_sb,
944 +                               "couldn't extend journal (err %d)", error);
945 +                       goto cleanup;
946 +               }
947 +       }
948 +
949         ext4_xattr_release_block(handle, inode, bh);
950         EXT4_I(inode)->i_file_acl = 0;
951  
952  cleanup:
953         brelse(bh);
954 +
955 +       return error;
956 +}
957 +
958 +void
959 +ext4_xattr_inode_array_free(struct inode *inode,
960 +                           struct ext4_xattr_ino_array *lea_ino_array)
961 +{
962 +       struct inode    *ea_inode = NULL;
963 +       int             idx = 0;
964 +       int             err;
965 +
966 +       if (lea_ino_array == NULL)
967 +               return;
968 +
969 +       for (; idx < lea_ino_array->xia_count; ++idx) {
970 +               ea_inode = ext4_xattr_inode_iget(inode,
971 +                               lea_ino_array->xia_inodes[idx], &err);
972 +               if (err)
973 +                       continue;
974 +               /* for inode's i_count get from ext4_xattr_delete_inode */
975 +               if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
976 +                       iput(ea_inode);
977 +               clear_nlink(ea_inode);
978 +               iput(ea_inode);
979 +       }
980 +       kfree(lea_ino_array);
981  }
982  
983  /*
984 @@ -1578,10 +2086,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
985                     entry1->e_name_index != entry2->e_name_index ||
986                     entry1->e_name_len != entry2->e_name_len ||
987                     entry1->e_value_size != entry2->e_value_size ||
988 +                   entry1->e_value_inum != entry2->e_value_inum ||
989                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
990                         return 1;
991 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
992 -                       return -EFSCORRUPTED;
993                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
994                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
995                            le32_to_cpu(entry1->e_value_size)))
996 @@ -1665,7 +2172,7 @@ static inline void ext4_xattr_hash_entry
997                        *name++;
998         }
999  
1000 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1001 +       if (!entry->e_value_inum && entry->e_value_size) {
1002                 __le32 *value = (__le32 *)((char *)header +
1003                         le16_to_cpu(entry->e_value_offs));
1004                 for (n = (le32_to_cpu(entry->e_value_size) +
1005 Index: linux-stage/fs/ext4/xattr.h
1006 ===================================================================
1007 --- linux-stage.orig/fs/ext4/xattr.h
1008 +++ linux-stage/fs/ext4/xattr.h
1009 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
1010         __u8    e_name_len;     /* length of name */
1011         __u8    e_name_index;   /* attribute name index */
1012         __le16  e_value_offs;   /* offset in disk block of value */
1013 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
1014 +       __le32  e_value_inum;   /* inode in which the value is stored */
1015         __le32  e_value_size;   /* size of attribute value */
1016         __le32  e_hash;         /* hash value of name and value */
1017         char    e_name[0];      /* attribute name */
1018 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
1019                 EXT4_I(inode)->i_extra_isize))
1020  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1021  
1022 +/*
1023 + * Link EA inode back to parent one using i_mtime field.
1024 + * Extra integer type conversion added to ignore higher
1025 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1026 + */
1027 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum)      \
1028 +do {                                                  \
1029 +      (inode)->i_mtime.tv_sec = inum;                 \
1030 +} while(0)
1031 +
1032 +#define EXT4_XATTR_INODE_GET_PARENT(inode)            \
1033 +((__u32)(inode)->i_mtime.tv_sec)
1034 +
1035 +/*
1036 + * The minimum size of EA value when you start storing it in an external inode
1037 + * size of block - size of header - size of 1 entry - 4 null bytes
1038 +*/
1039 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
1040 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1041 +
1042  #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1043  #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1044  #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1045 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1046  #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1047  
1048  struct ext4_xattr_info {
1049 -       int name_index;
1050         const char *name;
1051         const void *value;
1052         size_t value_len;
1053 +       int name_index;
1054 +       int in_inode;
1055  };
1056  
1057  struct ext4_xattr_search {
1058 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1059  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1060  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1061  
1062 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1063 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1064 +                                          int *err);
1065 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1066 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1067 +                                  struct ext4_xattr_ino_array **array);
1068 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1069 +                                       struct ext4_xattr_ino_array *array);
1070  
1071  extern void ext4_xattr_put_super(struct super_block *);
1072  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1073 Index: linux-stage/fs/ext4/ialloc.c
1074 ===================================================================
1075 --- linux-stage.orig/fs/ext4/ialloc.c
1076 +++ linux-stage/fs/ext4/ialloc.c
1077 @@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
1078          * as writing the quota to disk may need the lock as well.
1079          */
1080         dquot_initialize(inode);
1081 -       ext4_xattr_delete_inode(handle, inode);
1082         dquot_free_inode(inode);
1083         dquot_drop(inode);
1084  
1085 Index: linux-stage/fs/ext4/inline.c
1086 ===================================================================
1087 --- linux-stage.orig/fs/ext4/inline.c
1088 +++ linux-stage/fs/ext4/inline.c
1089 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1090  
1091         /* Compute min_offs. */
1092         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1093 -               if (!entry->e_value_block && entry->e_value_size) {
1094 +               if (!entry->e_value_inum && entry->e_value_size) {
1095                         size_t offs = le16_to_cpu(entry->e_value_offs);
1096                         if (offs < min_offs)
1097                                 min_offs = offs;