Whamcloud - gitweb
LU-8364 ldiskfs: fixes for failover mode
[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 @@ -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 @@ -249,9 +249,36 @@ void ext4_evict_inode(struct inode *inod
85                 sb_end_intwrite(inode->i_sb);
86                 goto no_delete;
87         }
88 -
89         if (IS_SYNC(inode))
90                 ext4_handle_sync(handle);
91 +
92 +       /*
93 +        * Delete xattr inode before deleting the main inode.
94 +        */
95 +       err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
96 +       if (err) {
97 +               ext4_warning(inode->i_sb,
98 +                            "couldn't delete inode's xattr (err %d)", err);
99 +               goto stop_handle;
100 +       }
101 +
102 +       if (!IS_NOQUOTA(inode))
103 +               extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
104 +
105 +       if (!ext4_handle_has_enough_credits(handle,
106 +                       ext4_blocks_for_truncate(inode) + extra_credits)) {
107 +               err = ext4_journal_extend(handle,
108 +                       ext4_blocks_for_truncate(inode) + extra_credits);
109 +               if (err > 0)
110 +                       err = ext4_journal_restart(handle,
111 +                       ext4_blocks_for_truncate(inode) + extra_credits);
112 +               if (err != 0) {
113 +                       ext4_warning(inode->i_sb,
114 +                                    "couldn't extend journal (err %d)", err);
115 +                       goto stop_handle;
116 +               }
117 +       }
118 +
119         inode->i_size = 0;
120         err = ext4_mark_inode_dirty(handle, inode);
121         if (err) {
122 @@ -269,10 +296,10 @@ void ext4_evict_inode(struct inode *inod
123          * enough credits left in the handle to remove the inode from
124          * the orphan list and set the dtime field.
125          */
126 -       if (!ext4_handle_has_enough_credits(handle, 3)) {
127 -               err = ext4_journal_extend(handle, 3);
128 +       if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
129 +               err = ext4_journal_extend(handle, extra_credits);
130                 if (err > 0)
131 -                       err = ext4_journal_restart(handle, 3);
132 +                       err = ext4_journal_restart(handle, extra_credits);
133                 if (err != 0) {
134                         ext4_warning(inode->i_sb,
135                                      "couldn't extend journal (err %d)", err);
136 @@ -306,8 +333,12 @@ void ext4_evict_inode(struct inode *inod
137                 ext4_clear_inode(inode);
138         else
139                 ext4_free_inode(handle, inode);
140 +
141         ext4_journal_stop(handle);
142         sb_end_intwrite(inode->i_sb);
143 +
144 +       if (lea_ino_array != NULL)
145 +               ext4_xattr_inode_array_free(inode, lea_ino_array);
146         return;
147  no_delete:
148         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
149 @@ -4681,7 +4712,7 @@ static int ext4_index_trans_blocks(struc
150   *
151   * Also account for superblock, inode, quota and xattr blocks
152   */
153 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
154 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
155                                   int pextents)
156  {
157         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
158 Index: linux-stage/fs/ext4/xattr.c
159 ===================================================================
160 --- linux-stage.orig/fs/ext4/xattr.c
161 +++ linux-stage/fs/ext4/xattr.c
162 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
163  
164         while (!IS_LAST_ENTRY(entry)) {
165                 if (entry->e_value_size != 0 &&
166 +                   entry->e_value_inum == 0 &&
167                     (value_start + le16_to_cpu(entry->e_value_offs) <
168                      (void *)e + sizeof(__u32) ||
169                      value_start + le16_to_cpu(entry->e_value_offs) +
170 @@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
171  }
172  
173  static inline int
174 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
175 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
176 +                      struct inode *inode)
177  {
178         size_t value_size = le32_to_cpu(entry->e_value_size);
179  
180 -       if (entry->e_value_block != 0 || value_size > size ||
181 +       if (!entry->e_value_inum &&
182             le16_to_cpu(entry->e_value_offs) + value_size > size)
183                 return -EFSCORRUPTED;
184 +       if (entry->e_value_inum &&
185 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
186 +            le32_to_cpu(entry->e_value_inum) >
187 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
188 +               return -EFSCORRUPTED;
189         return 0;
190  }
191  
192  static int
193  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
194 -                     const char *name, size_t size, int sorted)
195 +                     const char *name, size_t size, int sorted,
196 +                     struct inode *inode)
197  {
198         struct ext4_xattr_entry *entry;
199         size_t name_len;
200 @@ -265,11 +273,104 @@ ext4_xattr_find_entry(struct ext4_xattr_
201                         break;
202         }
203         *pentry = entry;
204 -       if (!cmp && ext4_xattr_check_entry(entry, size))
205 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
206                 return -EFSCORRUPTED;
207         return cmp ? -ENODATA : 0;
208  }
209  
210 +/*
211 + * Read the EA value from an inode.
212 + */
213 +static int
214 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
215 +{
216 +       unsigned long block = 0;
217 +       struct buffer_head *bh = NULL;
218 +       int blocksize;
219 +       size_t csize, ret_size = 0;
220 +
221 +       if (*size == 0)
222 +               return 0;
223 +
224 +       blocksize = ea_inode->i_sb->s_blocksize;
225 +
226 +       while (ret_size < *size) {
227 +               csize = (*size - ret_size) > blocksize ? blocksize :
228 +                                                       *size - ret_size;
229 +               bh = ext4_bread(NULL, ea_inode, block, 0);
230 +               if (IS_ERR(bh)) {
231 +                       *size = ret_size;
232 +                       return PTR_ERR(bh);
233 +               }
234 +               memcpy(buf, bh->b_data, csize);
235 +               brelse(bh);
236 +
237 +               buf += csize;
238 +               block += 1;
239 +               ret_size += csize;
240 +       }
241 +
242 +       *size = ret_size;
243 +
244 +       return 0;
245 +}
246 +
247 +struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
248 +{
249 +       struct inode *ea_inode = NULL;
250 +
251 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
252 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
253 +               int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
254 +               ext4_error(parent->i_sb, "error while reading EA inode %lu "
255 +                          "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
256 +               *err = rc != 0 ? rc : -EIO;
257 +               return NULL;
258 +       }
259 +
260 +       if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
261 +           ea_inode->i_generation != parent->i_generation) {
262 +               ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
263 +                          "to parent invalid.", ea_ino);
264 +               *err = -EINVAL;
265 +               goto error;
266 +       }
267 +
268 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
269 +               ext4_error(parent->i_sb, "EA inode %lu does not have "
270 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
271 +               *err = -EINVAL;
272 +               goto error;
273 +       }
274 +
275 +       *err = 0;
276 +       return ea_inode;
277 +
278 +error:
279 +       iput(ea_inode);
280 +       return NULL;
281 +}
282 +
283 +/*
284 + * Read the value from the EA inode.
285 + */
286 +static int
287 +ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
288 +                    size_t *size)
289 +{
290 +       struct inode *ea_inode = NULL;
291 +       int err;
292 +
293 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
294 +       if (err)
295 +               return err;
296 +
297 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
298 +       iput(ea_inode);
299 +
300 +       return err;
301 +}
302 +
303  static int
304  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
305                      void *buffer, size_t buffer_size)
306 @@ -301,7 +401,8 @@ bad_block:
307         }
308         ext4_xattr_cache_insert(bh);
309         entry = BFIRST(bh);
310 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
311 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
312 +                                     inode);
313         if (error == -EFSCORRUPTED)
314                 goto bad_block;
315         if (error)
316 @@ -311,8 +412,16 @@ bad_block:
317                 error = -ERANGE;
318                 if (size > buffer_size)
319                         goto cleanup;
320 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
321 -                      size);
322 +               if (entry->e_value_inum) {
323 +                       error = ext4_xattr_inode_get(inode,
324 +                                            le32_to_cpu(entry->e_value_inum),
325 +                                            buffer, &size);
326 +                       if (error)
327 +                               goto cleanup;
328 +               } else {
329 +                       memcpy(buffer, bh->b_data +
330 +                              le16_to_cpu(entry->e_value_offs), size);
331 +               }
332         }
333         error = size;
334  
335 @@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
336         if (error)
337                 goto cleanup;
338         error = ext4_xattr_find_entry(&entry, name_index, name,
339 -                                     end - (void *)entry, 0);
340 +                                     end - (void *)entry, 0, inode);
341         if (error)
342                 goto cleanup;
343         size = le32_to_cpu(entry->e_value_size);
344 @@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
345                 error = -ERANGE;
346                 if (size > buffer_size)
347                         goto cleanup;
348 -               memcpy(buffer, (void *)IFIRST(header) +
349 -                      le16_to_cpu(entry->e_value_offs), size);
350 +               if (entry->e_value_inum) {
351 +                       error = ext4_xattr_inode_get(inode,
352 +                                            le32_to_cpu(entry->e_value_inum),
353 +                                            buffer, &size);
354 +                       if (error)
355 +                               goto cleanup;
356 +               } else {
357 +                       memcpy(buffer, (void *)IFIRST(header) +
358 +                              le16_to_cpu(entry->e_value_offs), size);
359 +               }
360         }
361         error = size;
362  
363 @@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
364                                     size_t *min_offs, void *base, int *total)
365  {
366         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
367 -               if (!last->e_value_block && last->e_value_size) {
368 +               if (!last->e_value_inum && last->e_value_size) {
369                         size_t offs = le16_to_cpu(last->e_value_offs);
370                         if (offs < *min_offs)
371                                 *min_offs = offs;
372 @@ -611,16 +728,176 @@ static size_t ext4_xattr_free_space(stru
373         return (*min_offs - ((void *)last - base) - sizeof(__u32));
374  }
375  
376 +/*
377 + * Write the value of the EA in an inode.
378 + */
379  static int
380 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
381 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
382 +                      const void *buf, int bufsize)
383 +{
384 +       struct buffer_head *bh = NULL;
385 +       unsigned long block = 0;
386 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
387 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
388 +       int csize, wsize = 0;
389 +       int ret = 0;
390 +       int retries = 0;
391 +
392 +retry:
393 +       while (ret >= 0 && ret < max_blocks) {
394 +               struct ext4_map_blocks map;
395 +               map.m_lblk = block += ret;
396 +               map.m_len = max_blocks -= ret;
397 +
398 +               ret = ext4_map_blocks(handle, ea_inode, &map,
399 +                                     EXT4_GET_BLOCKS_CREATE);
400 +               if (ret <= 0) {
401 +                       ext4_mark_inode_dirty(handle, ea_inode);
402 +                       if (ret == -ENOSPC &&
403 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
404 +                               ret = 0;
405 +                               goto retry;
406 +                       }
407 +                       break;
408 +               }
409 +       }
410 +
411 +       if (ret < 0)
412 +               return ret;
413 +
414 +       block = 0;
415 +       while (wsize < bufsize) {
416 +               if (bh != NULL)
417 +                       brelse(bh);
418 +               csize = (bufsize - wsize) > blocksize ? blocksize :
419 +                                                               bufsize - wsize;
420 +               bh = ext4_getblk(handle, ea_inode, block, 0);
421 +               if (IS_ERR(bh)) {
422 +                       ret = PTR_ERR(bh);
423 +                       goto out;
424 +               }
425 +               ret = ext4_journal_get_write_access(handle, bh);
426 +               if (ret)
427 +                       goto out;
428 +
429 +               memcpy(bh->b_data, buf, csize);
430 +               set_buffer_uptodate(bh);
431 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
432 +
433 +               buf += csize;
434 +               wsize += csize;
435 +               block += 1;
436 +       }
437 +
438 +       mutex_lock(&ea_inode->i_mutex);
439 +       i_size_write(ea_inode, wsize);
440 +       ext4_update_i_disksize(ea_inode, wsize);
441 +       mutex_unlock(&ea_inode->i_mutex);
442 +
443 +       ext4_mark_inode_dirty(handle, ea_inode);
444 +
445 +out:
446 +       brelse(bh);
447 +
448 +       return ret;
449 +}
450 +
451 +/*
452 + * Create an inode to store the value of a large EA.
453 + */
454 +static struct inode *
455 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
456 +{
457 +       struct inode *ea_inode = NULL;
458 +
459 +       /*
460 +        * Let the next inode be the goal, so we try and allocate the EA inode
461 +        * in the same group, or nearby one.
462 +        */
463 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
464 +                                 S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
465 +
466 +       if (!IS_ERR(ea_inode)) {
467 +               ea_inode->i_op = &ext4_file_inode_operations;
468 +               ea_inode->i_fop = &ext4_file_operations;
469 +               ext4_set_aops(ea_inode);
470 +               ea_inode->i_generation = inode->i_generation;
471 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
472 +
473 +               /*
474 +                * A back-pointer from EA inode to parent inode will be useful
475 +                * for e2fsck.
476 +                */
477 +               EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
478 +               unlock_new_inode(ea_inode);
479 +       }
480 +
481 +       return ea_inode;
482 +}
483 +
484 +/*
485 + * Unlink the inode storing the value of the EA.
486 + */
487 +int
488 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
489 +{
490 +       struct inode *ea_inode = NULL;
491 +       int err;
492 +
493 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
494 +       if (err)
495 +               return err;
496 +
497 +       clear_nlink(ea_inode);
498 +       iput(ea_inode);
499 +
500 +       return 0;
501 +}
502 +
503 +/*
504 + * Add value of the EA in an inode.
505 + */
506 +static int
507 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
508 +                    const void *value, size_t value_len)
509 +{
510 +       struct inode *ea_inode = NULL;
511 +       int err;
512 +
513 +       /* Create an inode for the EA value */
514 +       ea_inode = ext4_xattr_inode_create(handle, inode);
515 +       if (IS_ERR(ea_inode))
516 +               return -1;
517 +
518 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
519 +       if (err)
520 +               clear_nlink(ea_inode);
521 +       else
522 +               *ea_ino = ea_inode->i_ino;
523 +
524 +       iput(ea_inode);
525 +
526 +       return err;
527 +}
528 +
529 +static int
530 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
531 +                    handle_t *handle, struct inode *inode)
532  {
533         struct ext4_xattr_entry *last;
534         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
535 +       int in_inode = i->in_inode;
536 +
537 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
538 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
539 +           (EXT4_XATTR_SIZE(i->value_len) >
540 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
541 +               in_inode = 1;
542  
543         /* Compute min_offs and last. */
544         last = s->first;
545         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
546 -               if (!last->e_value_block && last->e_value_size) {
547 +               if (!last->e_value_inum && last->e_value_size) {
548                         size_t offs = le16_to_cpu(last->e_value_offs);
549                         if (offs < min_offs)
550                                 min_offs = offs;
551 @@ -628,15 +903,20 @@ ext4_xattr_set_entry(struct ext4_xattr_i
552         }
553         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
554         if (!s->not_found) {
555 -               if (!s->here->e_value_block && s->here->e_value_size) {
556 +               if (!in_inode &&
557 +                   !s->here->e_value_inum && s->here->e_value_size) {
558                         size_t size = le32_to_cpu(s->here->e_value_size);
559                         free += EXT4_XATTR_SIZE(size);
560                 }
561                 free += EXT4_XATTR_LEN(name_len);
562         }
563         if (i->value) {
564 -               if (free < EXT4_XATTR_LEN(name_len) +
565 -                          EXT4_XATTR_SIZE(i->value_len))
566 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
567 +
568 +               if (in_inode)
569 +                       value_len = 0;
570 +
571 +               if (free < EXT4_XATTR_LEN(name_len) + value_len)
572                         return -ENOSPC;
573         }
574  
575 @@ -651,7 +931,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
576                 s->here->e_name_len = name_len;
577                 memcpy(s->here->e_name, i->name, name_len);
578         } else {
579 -               if (!s->here->e_value_block && s->here->e_value_size) {
580 +               if (!s->here->e_value_inum && s->here->e_value_size &&
581 +                   s->here->e_value_offs > 0) {
582                         void *first_val = s->base + min_offs;
583                         size_t offs = le16_to_cpu(s->here->e_value_offs);
584                         void *val = s->base + offs;
585 @@ -685,13 +966,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
586                         last = s->first;
587                         while (!IS_LAST_ENTRY(last)) {
588                                 size_t o = le16_to_cpu(last->e_value_offs);
589 -                               if (!last->e_value_block &&
590 +                               if (!last->e_value_inum &&
591                                     last->e_value_size && o < offs)
592                                         last->e_value_offs =
593                                                 cpu_to_le16(o + size);
594                                 last = EXT4_XATTR_NEXT(last);
595                         }
596                 }
597 +               if (s->here->e_value_inum) {
598 +                       ext4_xattr_inode_unlink(inode,
599 +                                       le32_to_cpu(s->here->e_value_inum));
600 +                       s->here->e_value_inum = 0;
601 +               }
602                 if (!i->value) {
603                         /* Remove the old name. */
604                         size_t size = EXT4_XATTR_LEN(name_len);
605 @@ -705,10 +990,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
606         if (i->value) {
607                 /* Insert the new value. */
608                 s->here->e_value_size = cpu_to_le32(i->value_len);
609 -               if (i->value_len) {
610 +               if (in_inode) {
611 +                       unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
612 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
613 +                                            i->value_len);
614 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
615 +                       s->here->e_value_offs = 0;
616 +               } else if (i->value_len) {
617                         size_t size = EXT4_XATTR_SIZE(i->value_len);
618                         void *val = s->base + min_offs - size;
619                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
620 +                       s->here->e_value_inum = 0;
621                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
622                                 memset(val, 0, size);
623                         } else {
624 @@ -758,7 +1050,7 @@ ext4_xattr_block_find(struct inode *inod
625                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
626                 bs->s.here = bs->s.first;
627                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
628 -                                             i->name, bs->bh->b_size, 1);
629 +                                            i->name, bs->bh->b_size, 1, inode);
630                 if (error && error != -ENODATA)
631                         goto cleanup;
632                 bs->s.not_found = error;
633 @@ -782,8 +1074,6 @@ ext4_xattr_block_set(handle_t *handle, s
634  
635  #define header(x) ((struct ext4_xattr_header *)(x))
636  
637 -       if (i->value && i->value_len > sb->s_blocksize)
638 -               return -ENOSPC;
639         if (s->base) {
640                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
641                                         bs->bh->b_blocknr);
642 @@ -799,7 +1089,7 @@ ext4_xattr_block_set(handle_t *handle, s
643                                 ce = NULL;
644                         }
645                         ea_bdebug(bs->bh, "modifying in-place");
646 -                       error = ext4_xattr_set_entry(i, s);
647 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
648                         if (!error) {
649                                 if (!IS_LAST_ENTRY(s->first))
650                                         ext4_xattr_rehash(header(s->base),
651 @@ -850,7 +1140,7 @@ ext4_xattr_block_set(handle_t *handle, s
652                 s->end = s->base + sb->s_blocksize;
653         }
654  
655 -       error = ext4_xattr_set_entry(i, s);
656 +       error = ext4_xattr_set_entry(i, s, handle, inode);
657         if (error == -EFSCORRUPTED)
658                 goto bad_block;
659         if (error)
660 @@ -1000,7 +1290,7 @@ int ext4_xattr_ibody_find(struct inode *
661                 /* Find the named attribute. */
662                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
663                                               i->name, is->s.end -
664 -                                             (void *)is->s.base, 0);
665 +                                             (void *)is->s.base, 0, inode);
666                 if (error && error != -ENODATA)
667                         return error;
668                 is->s.not_found = error;
669 @@ -1018,7 +1308,7 @@ int ext4_xattr_ibody_inline_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                 if (error == -ENOSPC &&
677                     ext4_has_inline_data(inode)) {
678 @@ -1030,7 +1320,7 @@ int ext4_xattr_ibody_inline_set(handle_t
679                         error = ext4_xattr_ibody_find(inode, i, is);
680                         if (error)
681                                 return error;
682 -                       error = ext4_xattr_set_entry(i, s);
683 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
684                 }
685                 if (error)
686                         return error;
687 @@ -1056,7 +1346,7 @@ static int ext4_xattr_ibody_set(handle_t
688  
689         if (EXT4_I(inode)->i_extra_isize == 0)
690                 return -ENOSPC;
691 -       error = ext4_xattr_set_entry(i, s);
692 +       error = ext4_xattr_set_entry(i, s, handle, inode);
693         if (error)
694                 return error;
695         header = IHDR(inode, ext4_raw_inode(&is->iloc));
696 @@ -1092,7 +1382,7 @@ ext4_xattr_set_handle(handle_t *handle, 
697                 .name = name,
698                 .value = value,
699                 .value_len = value_len,
700 -
701 +               .in_inode = 0,
702         };
703         struct ext4_xattr_ibody_find is = {
704                 .s = { .not_found = -ENODATA, },
705 @@ -1157,6 +1447,15 @@ ext4_xattr_set_handle(handle_t *handle, 
706                                         goto cleanup;
707                         }
708                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
709 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
710 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
711 +                           error == -ENOSPC) {
712 +                               /* xattr not fit to block, store at external
713 +                                * inode */
714 +                               i.in_inode = 1;
715 +                               error = ext4_xattr_ibody_set(handle, inode,
716 +                                                            &i, &is);
717 +                       }
718                         if (error)
719                                 goto cleanup;
720                         if (!is.s.not_found) {
721 @@ -1203,9 +1502,22 @@ ext4_xattr_set(struct inode *inode, int 
722                const void *value, size_t value_len, int flags)
723  {
724         handle_t *handle;
725 +       struct super_block *sb = inode->i_sb;
726         int error, retries = 0;
727         int credits = ext4_jbd2_credits_xattr(inode);
728  
729 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
730 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
731 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
732 +                                       sb->s_blocksize_bits;
733 +
734 +               /* For new inode */
735 +               credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
736 +
737 +               /* For data blocks of EA inode */
738 +               credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
739 +       }
740 +
741  retry:
742         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
743         if (IS_ERR(handle)) {
744 @@ -1217,7 +1529,7 @@ retry:
745                                               value, value_len, flags);
746                 error2 = ext4_journal_stop(handle);
747                 if (error == -ENOSPC &&
748 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
749 +                   ext4_should_retry_alloc(sb, &retries))
750                         goto retry;
751                 if (error == 0)
752                         error = error2;
753 @@ -1239,7 +1551,7 @@ static void ext4_xattr_shift_entries(str
754  
755         /* Adjust the value offsets of the entries */
756         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
757 -               if (!last->e_value_block && last->e_value_size) {
758 +               if (!last->e_value_inum && last->e_value_size) {
759                         new_offs = le16_to_cpu(last->e_value_offs) +
760                                                         value_offs_shift;
761                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
762 @@ -1477,21 +1789,135 @@ cleanup:
763  }
764  
765  
766 +#define EIA_INCR 16 /* must be 2^n */
767 +#define EIA_MASK (EIA_INCR - 1)
768 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
769 + * If @lea_ino_array is new or full it will be grown and the old
770 + * contents copied over.
771 + */
772 +static int
773 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
774 +{
775 +       if (*lea_ino_array == NULL) {
776 +               /*
777 +                * Start with 15 inodes, so it fits into a power-of-two size.
778 +                * If *lea_ino_array is NULL, this is essentially offsetof()
779 +                */
780 +               (*lea_ino_array) =
781 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
782 +                                        xia_inodes[EIA_MASK]),
783 +                               GFP_NOFS);
784 +               if (*lea_ino_array == NULL)
785 +                       return -ENOMEM;
786 +               (*lea_ino_array)->xia_count = 0;
787 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
788 +               /* expand the array once all 15 + n * 16 slots are full */
789 +               struct ext4_xattr_ino_array *new_array = NULL;
790 +               int count = (*lea_ino_array)->xia_count;
791 +
792 +               /* if new_array is NULL, this is essentially offsetof() */
793 +               new_array = kmalloc(
794 +                               offsetof(struct ext4_xattr_ino_array,
795 +                                        xia_inodes[count + EIA_INCR]),
796 +                               GFP_NOFS);
797 +               if (new_array == NULL)
798 +                       return -ENOMEM;
799 +               memcpy(new_array, *lea_ino_array,
800 +                      offsetof(struct ext4_xattr_ino_array,
801 +                               xia_inodes[count]));
802 +               kfree(*lea_ino_array);
803 +               *lea_ino_array = new_array;
804 +       }
805 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
806 +       return 0;
807 +}
808 +
809 +/**
810 + * Add xattr inode to orphan list
811 + */
812 +static int
813 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
814 +                       int credits, struct ext4_xattr_ino_array *lea_ino_array)
815 +{
816 +       struct inode *ea_inode = NULL;
817 +       int idx = 0, error = 0;
818 +
819 +       if (lea_ino_array == NULL)
820 +               return 0;
821 +
822 +       for (; idx < lea_ino_array->xia_count; ++idx) {
823 +               if (!ext4_handle_has_enough_credits(handle, credits)) {
824 +                       error = ext4_journal_extend(handle, credits);
825 +                       if (error > 0)
826 +                               error = ext4_journal_restart(handle, credits);
827 +
828 +                       if (error != 0) {
829 +                               ext4_warning(inode->i_sb,
830 +                                       "couldn't extend journal "
831 +                                       "(err %d)", error);
832 +                               return error;
833 +                       }
834 +               }
835 +               ea_inode = ext4_xattr_inode_iget(inode,
836 +                               lea_ino_array->xia_inodes[idx], &error);
837 +               if (error)
838 +                       continue;
839 +               ext4_orphan_add(handle, ea_inode);
840 +               /* the inode's i_count will be released by caller */
841 +       }
842 +
843 +       return 0;
844 +}
845  
846  /*
847   * ext4_xattr_delete_inode()
848   *
849 - * Free extended attribute resources associated with this inode. This
850 + * Free extended attribute resources associated with this inode. Traverse
851 + * all entries and unlink any xattr inodes associated with this inode. This
852   * is called immediately before an inode is freed. We have exclusive
853 - * access to the inode.
854 + * access to the inode. If an orphan inode is deleted it will also delete any
855 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
856 + * to ensure they belong to the parent inode and were not deleted already.
857   */
858 -void
859 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
860 +int
861 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
862 +                       struct ext4_xattr_ino_array **lea_ino_array)
863  {
864         struct buffer_head *bh = NULL;
865 +       struct ext4_xattr_ibody_header *header;
866 +       struct ext4_inode *raw_inode;
867 +       struct ext4_iloc iloc;
868 +       struct ext4_xattr_entry *entry;
869 +       int credits = 3, error = 0;
870  
871 -       if (!EXT4_I(inode)->i_file_acl)
872 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
873 +               goto delete_external_ea;
874 +
875 +       error = ext4_get_inode_loc(inode, &iloc);
876 +       if (error)
877                 goto cleanup;
878 +       raw_inode = ext4_raw_inode(&iloc);
879 +       header = IHDR(inode, raw_inode);
880 +       for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
881 +            entry = EXT4_XATTR_NEXT(entry)) {
882 +               if (!entry->e_value_inum)
883 +                       continue;
884 +               if (ext4_expand_ino_array(lea_ino_array,
885 +                                         entry->e_value_inum) != 0) {
886 +                       brelse(iloc.bh);
887 +                       goto cleanup;
888 +               }
889 +               entry->e_value_inum = 0;
890 +       }
891 +       brelse(iloc.bh);
892 +
893 +delete_external_ea:
894 +       if (!EXT4_I(inode)->i_file_acl) {
895 +               /* add xattr inode to orphan list */
896 +               ext4_xattr_inode_orphan_add(handle, inode, credits,
897 +                                               *lea_ino_array);
898 +               goto cleanup;
899 +       }
900         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
901         if (!bh) {
902                 EXT4_ERROR_INODE(inode, "block %llu read error",
903 @@ -1504,11 +1930,69 @@ ext4_xattr_delete_inode(handle_t *handle
904                                  EXT4_I(inode)->i_file_acl);
905                 goto cleanup;
906         }
907 +
908 +       for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
909 +            entry = EXT4_XATTR_NEXT(entry)) {
910 +               if (!entry->e_value_inum)
911 +                       continue;
912 +               if (ext4_expand_ino_array(lea_ino_array,
913 +                                         entry->e_value_inum) != 0)
914 +                       goto cleanup;
915 +               entry->e_value_inum = 0;
916 +       }
917 +
918 +       /* add xattr inode to orphan list */
919 +       error = ext4_xattr_inode_orphan_add(handle, inode, credits,
920 +                                       *lea_ino_array);
921 +       if (error != 0)
922 +               goto cleanup;
923 +
924 +       if (!IS_NOQUOTA(inode))
925 +               credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
926 +
927 +       if (!ext4_handle_has_enough_credits(handle, credits)) {
928 +               error = ext4_journal_extend(handle, credits);
929 +               if (error > 0)
930 +                       error = ext4_journal_restart(handle, credits);
931 +               if (error != 0) {
932 +                       ext4_warning(inode->i_sb,
933 +                               "couldn't extend journal (err %d)", error);
934 +                       goto cleanup;
935 +               }
936 +       }
937 +
938         ext4_xattr_release_block(handle, inode, bh);
939         EXT4_I(inode)->i_file_acl = 0;
940  
941  cleanup:
942         brelse(bh);
943 +
944 +       return error;
945 +}
946 +
947 +void
948 +ext4_xattr_inode_array_free(struct inode *inode,
949 +                           struct ext4_xattr_ino_array *lea_ino_array)
950 +{
951 +       struct inode    *ea_inode = NULL;
952 +       int             idx = 0;
953 +       int             err;
954 +
955 +       if (lea_ino_array == NULL)
956 +               return;
957 +
958 +       for (; idx < lea_ino_array->xia_count; ++idx) {
959 +               ea_inode = ext4_xattr_inode_iget(inode,
960 +                               lea_ino_array->xia_inodes[idx], &err);
961 +               if (err)
962 +                       continue;
963 +               /* for inode's i_count get from ext4_xattr_delete_inode */
964 +               if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
965 +                       iput(ea_inode);
966 +               clear_nlink(ea_inode);
967 +               iput(ea_inode);
968 +       }
969 +       kfree(lea_ino_array);
970  }
971  
972  /*
973 @@ -1578,10 +2062,9 @@ ext4_xattr_cmp(struct ext4_xattr_header 
974                     entry1->e_name_index != entry2->e_name_index ||
975                     entry1->e_name_len != entry2->e_name_len ||
976                     entry1->e_value_size != entry2->e_value_size ||
977 +                   entry1->e_value_inum != entry2->e_value_inum ||
978                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
979                         return 1;
980 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
981 -                       return -EFSCORRUPTED;
982                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
983                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
984                            le32_to_cpu(entry1->e_value_size)))
985 @@ -1665,7 +2148,7 @@ static inline void ext4_xattr_hash_entry
986                        *name++;
987         }
988  
989 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
990 +       if (!entry->e_value_inum && entry->e_value_size) {
991                 __le32 *value = (__le32 *)((char *)header +
992                         le16_to_cpu(entry->e_value_offs));
993                 for (n = (le32_to_cpu(entry->e_value_size) +
994 Index: linux-stage/fs/ext4/xattr.h
995 ===================================================================
996 --- linux-stage.orig/fs/ext4/xattr.h
997 +++ linux-stage/fs/ext4/xattr.h
998 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
999         __u8    e_name_len;     /* length of name */
1000         __u8    e_name_index;   /* attribute name index */
1001         __le16  e_value_offs;   /* offset in disk block of value */
1002 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
1003 +       __le32  e_value_inum;   /* inode in which the value is stored */
1004         __le32  e_value_size;   /* size of attribute value */
1005         __le32  e_hash;         /* hash value of name and value */
1006         char    e_name[0];      /* attribute name */
1007 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
1008                 EXT4_I(inode)->i_extra_isize))
1009  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1010  
1011 +/*
1012 + * Link EA inode back to parent one using i_mtime field.
1013 + * Extra integer type conversion added to ignore higher
1014 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1015 + */
1016 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum)      \
1017 +do {                                                  \
1018 +      (inode)->i_mtime.tv_sec = inum;                 \
1019 +} while(0)
1020 +
1021 +#define EXT4_XATTR_INODE_GET_PARENT(inode)            \
1022 +((__u32)(inode)->i_mtime.tv_sec)
1023 +
1024 +/*
1025 + * The minimum size of EA value when you start storing it in an external inode
1026 + * size of block - size of header - size of 1 entry - 4 null bytes
1027 +*/
1028 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
1029 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1030 +
1031  #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1032  #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1033  #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1034 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1035  #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1036  
1037  struct ext4_xattr_info {
1038 -       int name_index;
1039         const char *name;
1040         const void *value;
1041         size_t value_len;
1042 +       int name_index;
1043 +       int in_inode;
1044  };
1045  
1046  struct ext4_xattr_search {
1047 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1048  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1049  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1050  
1051 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1052 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1053 +                                          int *err);
1054 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1055 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1056 +                                  struct ext4_xattr_ino_array **array);
1057 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1058 +                                       struct ext4_xattr_ino_array *array);
1059  
1060  extern void ext4_xattr_put_super(struct super_block *);
1061  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1062 Index: linux-stage/fs/ext4/ialloc.c
1063 ===================================================================
1064 --- linux-stage.orig/fs/ext4/ialloc.c
1065 +++ linux-stage/fs/ext4/ialloc.c
1066 @@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
1067          * as writing the quota to disk may need the lock as well.
1068          */
1069         dquot_initialize(inode);
1070 -       ext4_xattr_delete_inode(handle, inode);
1071         dquot_free_inode(inode);
1072         dquot_drop(inode);
1073  
1074 Index: linux-stage/fs/ext4/inline.c
1075 ===================================================================
1076 --- linux-stage.orig/fs/ext4/inline.c
1077 +++ linux-stage/fs/ext4/inline.c
1078 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1079  
1080         /* Compute min_offs. */
1081         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1082 -               if (!entry->e_value_block && entry->e_value_size) {
1083 +               if (!entry->e_value_inum && entry->e_value_size) {
1084                         size_t offs = le16_to_cpu(entry->e_value_offs);
1085                         if (offs < min_offs)
1086                                 min_offs = offs;