Whamcloud - gitweb
LU-13004 modules: replace lnet_kiov_t with struct bio_vec
[fs/lustre-release.git] / ldiskfs / kernel_patches / patches / rhel7.2 / ext4-large-eas.patch
1 This patch implements the large EA support in ext4. If the size of
2 an EA value is larger than the blocksize, then the EA value would
3 not be saved in the external EA block, instead it would be saved
4 in an external EA inode. So, the patch also helps support a larger
5 number of EAs.
6
7 Index: linux-stage/fs/ext4/ext4.h
8 ===================================================================
9 --- linux-stage.orig/fs/ext4/ext4.h
10 +++ linux-stage/fs/ext4/ext4.h
11 @@ -1579,6 +1579,7 @@ static inline void ext4_clear_state_flag
12                                          EXT4_FEATURE_INCOMPAT_EXTENTS| \
13                                          EXT4_FEATURE_INCOMPAT_64BIT| \
14                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
15 +                                        EXT4_FEATURE_INCOMPAT_EA_INODE| \
16                                          EXT4_FEATURE_INCOMPAT_MMP |    \
17                                          EXT4_FEATURE_INCOMPAT_DIRDATA| \
18                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA)
19 @@ -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 @@ -252,6 +252,32 @@ void ext4_evict_inode(struct inode *inod
72  
73         if (IS_SYNC(inode))
74                 ext4_handle_sync(handle);
75 +
76 +       /* Delete xattr inode before deleting the main inode. */
77 +       err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
78 +       if (err) {
79 +               ext4_warning(inode->i_sb,
80 +                            "couldn't delete inode's xattr (err %d)", err);
81 +               goto stop_handle;
82 +       }
83 +
84 +       if (!IS_NOQUOTA(inode))
85 +               extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
86 +
87 +       if (!ext4_handle_has_enough_credits(handle,
88 +                       ext4_blocks_for_truncate(inode) + extra_credits)) {
89 +               err = ext4_journal_extend(handle,
90 +                       ext4_blocks_for_truncate(inode) + extra_credits);
91 +               if (err > 0)
92 +                       err = ext4_journal_restart(handle,
93 +                       ext4_blocks_for_truncate(inode) + extra_credits);
94 +               if (err != 0) {
95 +                       ext4_warning(inode->i_sb,
96 +                                    "couldn't extend journal (err %d)", err);
97 +                       goto stop_handle;
98 +               }
99 +       }
100 +
101         inode->i_size = 0;
102         err = ext4_mark_inode_dirty(handle, inode);
103         if (err) {
104 @@ -269,10 +296,10 @@ void ext4_evict_inode(struct inode *inod
105          * enough credits left in the handle to remove the inode from
106          * the orphan list and set the dtime field.
107          */
108 -       if (!ext4_handle_has_enough_credits(handle, 3)) {
109 -               err = ext4_journal_extend(handle, 3);
110 +       if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
111 +               err = ext4_journal_extend(handle, extra_credits);
112                 if (err > 0)
113 -                       err = ext4_journal_restart(handle, 3);
114 +                       err = ext4_journal_restart(handle, extra_credits);
115                 if (err != 0) {
116                         ext4_warning(inode->i_sb,
117                                      "couldn't extend journal (err %d)", err);
118 @@ -308,6 +335,9 @@ void ext4_evict_inode(struct inode *inod
119                 ext4_free_inode(handle, inode);
120         ext4_journal_stop(handle);
121         sb_end_intwrite(inode->i_sb);
122 +
123 +       if (lea_ino_array != NULL)
124 +               ext4_xattr_inode_array_free(inode, lea_ino_array);
125         return;
126  no_delete:
127         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
128 @@ -4681,7 +4711,7 @@ static int ext4_index_trans_blocks(struc
129   *
130   * Also account for superblock, inode, quota and xattr blocks
131   */
132 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
133 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
134                                   int pextents)
135  {
136         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
137 Index: linux-stage/fs/ext4/xattr.c
138 ===================================================================
139 --- linux-stage.orig/fs/ext4/xattr.c
140 +++ linux-stage/fs/ext4/xattr.c
141 @@ -201,6 +201,7 @@ ext4_xattr_check_names(struct ext4_xattr
142  
143         while (!IS_LAST_ENTRY(entry)) {
144                 if (entry->e_value_size != 0 &&
145 +                   entry->e_value_inum == 0 &&
146                     (value_start + le16_to_cpu(entry->e_value_offs) <
147                      (void *)e + sizeof(__u32) ||
148                      value_start + le16_to_cpu(entry->e_value_offs) +
149 @@ -233,19 +234,26 @@ ext4_xattr_check_block(struct inode *ino
150  }
151  
152  static inline int
153 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
154 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
155 +                      struct inode *inode)
156  {
157         size_t value_size = le32_to_cpu(entry->e_value_size);
158  
159 -       if (entry->e_value_block != 0 || value_size > size ||
160 +       if (!entry->e_value_inum &&
161             le16_to_cpu(entry->e_value_offs) + value_size > size)
162 +               return -EIO;
163 +       if (entry->e_value_inum &&
164 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
165 +            le32_to_cpu(entry->e_value_inum) >
166 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
167                 return -EIO;
168         return 0;
169  }
170  
171  static int
172  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
173 -                     const char *name, size_t size, int sorted)
174 +                     const char *name, size_t size, int sorted,
175 +                     struct inode *inode)
176  {
177         struct ext4_xattr_entry *entry;
178         size_t name_len;
179 @@ -265,11 +273,109 @@ ext4_xattr_find_entry(struct ext4_xattr_
180                         break;
181         }
182         *pentry = entry;
183 -       if (!cmp && ext4_xattr_check_entry(entry, size))
184 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
185                         return -EIO;
186         return cmp ? -ENODATA : 0;
187  }
188  
189 +/*
190 + * Read the EA value from an inode.
191 + */
192 +static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
193 +{
194 +       unsigned long block = 0;
195 +       struct buffer_head *bh = NULL;
196 +       int err, blocksize;
197 +       size_t csize, ret_size = 0;
198 +
199 +       if (*size == 0)
200 +               return 0;
201 +
202 +       blocksize = ea_inode->i_sb->s_blocksize;
203 +
204 +       while (ret_size < *size) {
205 +               csize = (*size - ret_size) > blocksize ? blocksize :
206 +                                                       *size - ret_size;
207 +               bh = ext4_bread(NULL, ea_inode, block, 0, &err);
208 +               if (!bh) {
209 +                       *size = ret_size;
210 +                       return err;
211 +               }
212 +               memcpy(buf, bh->b_data, csize);
213 +               brelse(bh);
214 +
215 +               buf += csize;
216 +               block += 1;
217 +               ret_size += csize;
218 +       }
219 +
220 +       *size = ret_size;
221 +
222 +       return err;
223 +}
224 +
225 +/*
226 + * Fetch the xattr inode from disk.
227 + *
228 + * The xattr inode stores the parent inode number and generation so that
229 + * the kernel and e2fsck can verify the xattr inode is valid upon access.
230 + */
231 +struct inode *ext4_xattr_inode_iget(struct inode *parent,
232 +                                   unsigned long ea_ino, int *err)
233 +{
234 +       struct inode *ea_inode = NULL;
235 +
236 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
237 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
238 +               int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
239 +               ext4_error(parent->i_sb, "error while reading EA inode %lu "
240 +                          "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
241 +               *err = rc != 0 ? rc : -EIO;
242 +               return NULL;
243 +       }
244 +
245 +       if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != parent->i_ino ||
246 +           ea_inode->i_generation != parent->i_generation) {
247 +               ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
248 +                          "to parent invalid.", ea_ino);
249 +               *err = -EINVAL;
250 +               goto error;
251 +       }
252 +
253 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
254 +               ext4_error(parent->i_sb, "EA inode %lu does not have "
255 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
256 +               *err = -EINVAL;
257 +               goto error;
258 +       }
259 +
260 +       *err = 0;
261 +       return ea_inode;
262 +
263 +error:
264 +       iput(ea_inode);
265 +       return NULL;
266 +}
267 +
268 +/*
269 + * Read the value from the EA inode.
270 + */
271 +static int ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino,
272 +                               void *buffer, size_t *size)
273 +{
274 +       struct inode *ea_inode = NULL;
275 +       int err;
276 +
277 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
278 +       if (err)
279 +               return err;
280 +
281 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
282 +       iput(ea_inode);
283 +
284 +       return err;
285 +}
286 +
287  static int
288  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
289                      void *buffer, size_t buffer_size)
290 @@ -301,7 +401,8 @@ bad_block:
291         }
292         ext4_xattr_cache_insert(bh);
293         entry = BFIRST(bh);
294 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
295 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
296 +                                     inode);
297         if (error == -EIO)
298                 goto bad_block;
299         if (error)
300 @@ -311,8 +412,16 @@ bad_block:
301                 error = -ERANGE;
302                 if (size > buffer_size)
303                         goto cleanup;
304 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
305 -                      size);
306 +               if (entry->e_value_inum) {
307 +                       error = ext4_xattr_inode_get(inode,
308 +                                            le32_to_cpu(entry->e_value_inum),
309 +                                            buffer, &size);
310 +                       if (error)
311 +                               goto cleanup;
312 +               } else {
313 +                       memcpy(buffer, bh->b_data +
314 +                              le16_to_cpu(entry->e_value_offs), size);
315 +               }
316         }
317         error = size;
318  
319 @@ -346,7 +455,7 @@ ext4_xattr_ibody_get(struct inode *inode
320         if (error)
321                 goto cleanup;
322         error = ext4_xattr_find_entry(&entry, name_index, name,
323 -                                     end - (void *)entry, 0);
324 +                                     end - (void *)entry, 0, inode);
325         if (error)
326                 goto cleanup;
327         size = le32_to_cpu(entry->e_value_size);
328 @@ -354,8 +463,16 @@ ext4_xattr_ibody_get(struct inode *inode
329                 error = -ERANGE;
330                 if (size > buffer_size)
331                         goto cleanup;
332 -               memcpy(buffer, (void *)IFIRST(header) +
333 -                      le16_to_cpu(entry->e_value_offs), size);
334 +               if (entry->e_value_inum) {
335 +                       error = ext4_xattr_inode_get(inode,
336 +                                            le32_to_cpu(entry->e_value_inum),
337 +                                            buffer, &size);
338 +                       if (error)
339 +                               goto cleanup;
340 +               } else {
341 +                       memcpy(buffer, (void *)IFIRST(header) +
342 +                              le16_to_cpu(entry->e_value_offs), size);
343 +               }
344         }
345         error = size;
346  
347 @@ -600,7 +717,7 @@ static size_t ext4_xattr_free_space(stru
348                                     size_t *min_offs, void *base, int *total)
349  {
350         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
351 -               if (!last->e_value_block && last->e_value_size) {
352 +               if (!last->e_value_inum && last->e_value_size) {
353                         size_t offs = le16_to_cpu(last->e_value_offs);
354                         if (offs < *min_offs)
355                                 *min_offs = offs;
356 @@ -611,16 +728,198 @@ static size_t ext4_xattr_free_space(stru
357         return (*min_offs - ((void *)last - base) - sizeof(__u32));
358  }
359  
360 +/*
361 + * Write the value of the EA in an inode.
362 + */
363  static int
364 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
365 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
366 +                      const void *buf, int bufsize)
367 +{
368 +       struct buffer_head *bh = NULL;
369 +       unsigned long block = 0;
370 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
371 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
372 +       int csize, wsize = 0;
373 +       int ret = 0;
374 +       int retries = 0;
375 +
376 +retry:
377 +       while (ret >= 0 && ret < max_blocks) {
378 +               struct ext4_map_blocks map;
379 +               map.m_lblk = block += ret;
380 +               map.m_len = max_blocks -= ret;
381 +
382 +               ret = ext4_map_blocks(handle, ea_inode, &map,
383 +                                     EXT4_GET_BLOCKS_CREATE);
384 +               if (ret <= 0) {
385 +                       ext4_mark_inode_dirty(handle, ea_inode);
386 +                       if (ret == -ENOSPC &&
387 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
388 +                               ret = 0;
389 +                               goto retry;
390 +                       }
391 +                       break;
392 +               }
393 +       }
394 +
395 +       if (ret < 0)
396 +               return ret;
397 +
398 +       block = 0;
399 +       while (wsize < bufsize) {
400 +               if (bh != NULL)
401 +                       brelse(bh);
402 +               csize = (bufsize - wsize) > blocksize ? blocksize :
403 +                                                               bufsize - wsize;
404 +               bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
405 +               if (!bh)
406 +                       goto out;
407 +               ret = ext4_journal_get_write_access(handle, bh);
408 +               if (ret)
409 +                       goto out;
410 +
411 +               memcpy(bh->b_data, buf, csize);
412 +               set_buffer_uptodate(bh);
413 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
414 +
415 +               buf += csize;
416 +               wsize += csize;
417 +               block += 1;
418 +       }
419 +
420 +       mutex_lock(&ea_inode->i_mutex);
421 +       i_size_write(ea_inode, wsize);
422 +       ext4_update_i_disksize(ea_inode, wsize);
423 +       mutex_unlock(&ea_inode->i_mutex);
424 +
425 +       ext4_mark_inode_dirty(handle, ea_inode);
426 +
427 +out:
428 +       brelse(bh);
429 +
430 +       return ret;
431 +}
432 +
433 +static void ext4_xattr_inode_set_ref(struct inode *ea_inode, __u64 ref_count)
434 +{
435 +       ea_inode->i_ctime.tv_sec = (__u32)(ref_count >> 32);
436 +       ea_inode->i_version = (__u32)ref_count;
437 +}
438 +
439 +static void ext4_xattr_inode_set_hash(struct inode *ea_inode, __u32 hash)
440 +{
441 +       ea_inode->i_atime.tv_sec = hash;
442 +}
443 +
444 +/*
445 + * Create an inode to store the value of a large EA.
446 + */
447 +static struct inode *
448 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode, __u32 hash)
449 +{
450 +       struct inode *ea_inode = NULL;
451 +
452 +       /*
453 +        * Let the next inode be the goal, so we try and allocate the EA inode
454 +        * in the same group, or nearby one.
455 +        */
456 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
457 +                                 S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
458 +
459 +       if (!IS_ERR(ea_inode)) {
460 +               ea_inode->i_op = &ext4_file_inode_operations;
461 +               ea_inode->i_fop = &ext4_file_operations;
462 +               ext4_set_aops(ea_inode);
463 +               ea_inode->i_generation = inode->i_generation;
464 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
465 +
466 +               /*
467 +                * A back-pointer from EA inode to parent inode will be useful
468 +                * for e2fsck.
469 +                */
470 +               EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
471 +               unlock_new_inode(ea_inode);
472 +
473 +               ext4_xattr_inode_set_ref(ea_inode, 1);
474 +               ext4_xattr_inode_set_hash(ea_inode, hash);
475 +       }
476 +
477 +       return ea_inode;
478 +}
479 +
480 +/*
481 + * Unlink the inode storing the value of the EA.
482 + */
483 +int
484 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
485 +{
486 +       struct inode *ea_inode = NULL;
487 +       int err;
488 +
489 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
490 +       if (err)
491 +               return err;
492 +
493 +       clear_nlink(ea_inode);
494 +       iput(ea_inode);
495 +
496 +       return 0;
497 +}
498 +
499 +static __u32
500 +ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
501 +{
502 +       if (ext4_has_metadata_csum(sbi->s_sb))
503 +               return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
504 +       return 0;
505 +}
506 +
507 +/*
508 + * Add value of the EA in an inode.
509 + */
510 +static int
511 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
512 +                    const void *value, size_t value_len)
513 +{
514 +       struct inode *ea_inode = NULL;
515 +       __u32 hash;
516 +       int err;
517 +
518 +       /* Create an inode for the EA value */
519 +       hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
520 +       ea_inode = ext4_xattr_inode_create(handle, inode, hash);
521 +       if (IS_ERR(ea_inode))
522 +               return -1;
523 +
524 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
525 +       if (err)
526 +               clear_nlink(ea_inode);
527 +       else
528 +               *ea_ino = ea_inode->i_ino;
529 +
530 +       iput(ea_inode);
531 +
532 +       return err;
533 +}
534 +
535 +static int
536 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
537 +                    handle_t *handle, struct inode *inode)
538  {
539         struct ext4_xattr_entry *last;
540         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
541 +       int in_inode = i->in_inode;
542 +
543 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
544 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
545 +           (EXT4_XATTR_SIZE(i->value_len) >
546 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
547 +               in_inode = 1;
548  
549         /* Compute min_offs and last. */
550         last = s->first;
551         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
552 -               if (!last->e_value_block && last->e_value_size) {
553 +               if (!last->e_value_inum && last->e_value_size) {
554                         size_t offs = le16_to_cpu(last->e_value_offs);
555                         if (offs < min_offs)
556                                 min_offs = offs;
557 @@ -628,15 +927,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
558         }
559         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
560         if (!s->not_found) {
561 -               if (!s->here->e_value_block && s->here->e_value_size) {
562 +               if (!in_inode &&
563 +                   !s->here->e_value_inum && s->here->e_value_size) {
564                         size_t size = le32_to_cpu(s->here->e_value_size);
565                         free += EXT4_XATTR_SIZE(size);
566                 }
567                 free += EXT4_XATTR_LEN(name_len);
568         }
569         if (i->value) {
570 -               if (free < EXT4_XATTR_LEN(name_len) +
571 -                          EXT4_XATTR_SIZE(i->value_len))
572 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
573 +
574 +               if (in_inode)
575 +                       value_len = 0;
576 +
577 +               if (free < value_len ||
578 +                   free < EXT4_XATTR_LEN(name_len) + value_len)
579                         return -ENOSPC;
580         }
581  
582 @@ -651,7 +955,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
583                 s->here->e_name_len = name_len;
584                 memcpy(s->here->e_name, i->name, name_len);
585         } else {
586 -               if (!s->here->e_value_block && s->here->e_value_size) {
587 +               if (!s->here->e_value_inum && s->here->e_value_size &&
588 +                   s->here->e_value_offs > 0) {
589                         void *first_val = s->base + min_offs;
590                         size_t offs = le16_to_cpu(s->here->e_value_offs);
591                         void *val = s->base + offs;
592 @@ -685,13 +990,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
593                         last = s->first;
594                         while (!IS_LAST_ENTRY(last)) {
595                                 size_t o = le16_to_cpu(last->e_value_offs);
596 -                               if (!last->e_value_block &&
597 +                               if (!last->e_value_inum &&
598                                     last->e_value_size && o < offs)
599                                         last->e_value_offs =
600                                                 cpu_to_le16(o + size);
601                                 last = EXT4_XATTR_NEXT(last);
602                         }
603                 }
604 +               if (s->here->e_value_inum) {
605 +                       ext4_xattr_inode_unlink(inode,
606 +                                       le32_to_cpu(s->here->e_value_inum));
607 +                       s->here->e_value_inum = 0;
608 +               }
609                 if (!i->value) {
610                         /* Remove the old name. */
611                         size_t size = EXT4_XATTR_LEN(name_len);
612 @@ -705,10 +1014,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
613         if (i->value) {
614                 /* Insert the new value. */
615                 s->here->e_value_size = cpu_to_le32(i->value_len);
616 -               if (i->value_len) {
617 +               if (in_inode) {
618 +                       unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
619 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
620 +                                            i->value_len);
621 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
622 +                       s->here->e_value_offs = 0;
623 +               } else if (i->value_len) {
624                         size_t size = EXT4_XATTR_SIZE(i->value_len);
625                         void *val = s->base + min_offs - size;
626                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
627 +                       s->here->e_value_inum = 0;
628                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
629                                 memset(val, 0, size);
630                         } else {
631 @@ -758,7 +1074,7 @@ ext4_xattr_block_find(struct inode *inod
632                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
633                 bs->s.here = bs->s.first;
634                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
635 -                                             i->name, bs->bh->b_size, 1);
636 +                                             i->name, bs->bh->b_size, 1, inode);
637                 if (error && error != -ENODATA)
638                         goto cleanup;
639                 bs->s.not_found = error;
640 @@ -782,8 +1098,6 @@ ext4_xattr_block_set(handle_t *handle, s
641  
642  #define header(x) ((struct ext4_xattr_header *)(x))
643  
644 -       if (i->value && i->value_len > sb->s_blocksize)
645 -               return -ENOSPC;
646         if (s->base) {
647                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
648                                         bs->bh->b_blocknr);
649 @@ -799,7 +1113,7 @@ ext4_xattr_block_set(handle_t *handle, s
650                                 ce = NULL;
651                         }
652                         ea_bdebug(bs->bh, "modifying in-place");
653 -                       error = ext4_xattr_set_entry(i, s);
654 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
655                         if (!error) {
656                                 if (!IS_LAST_ENTRY(s->first))
657                                         ext4_xattr_rehash(header(s->base),
658 @@ -850,7 +1164,7 @@ ext4_xattr_block_set(handle_t *handle, s
659                 s->end = s->base + sb->s_blocksize;
660         }
661  
662 -       error = ext4_xattr_set_entry(i, s);
663 +       error = ext4_xattr_set_entry(i, s, handle, inode);
664         if (error == -EIO)
665                 goto bad_block;
666         if (error)
667 @@ -1000,7 +1314,7 @@ int ext4_xattr_ibody_find(struct inode *
668                 /* Find the named attribute. */
669                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
670                                               i->name, is->s.end -
671 -                                             (void *)is->s.base, 0);
672 +                                             (void *)is->s.base, 0, inode);
673                 if (error && error != -ENODATA)
674                         return error;
675                 is->s.not_found = error;
676 @@ -1018,7 +1332,7 @@ int ext4_xattr_ibody_inline_set(handle_t
677  
678         if (EXT4_I(inode)->i_extra_isize == 0)
679                 return -ENOSPC;
680 -       error = ext4_xattr_set_entry(i, s);
681 +       error = ext4_xattr_set_entry(i, s, handle, inode);
682         if (error) {
683                 if (error == -ENOSPC &&
684                     ext4_has_inline_data(inode)) {
685 @@ -1030,7 +1344,7 @@ int ext4_xattr_ibody_inline_set(handle_t
686                         error = ext4_xattr_ibody_find(inode, i, is);
687                         if (error)
688                                 return error;
689 -                       error = ext4_xattr_set_entry(i, s);
690 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
691                 }
692                 if (error)
693                         return error;
694 @@ -1056,7 +1370,7 @@ static int ext4_xattr_ibody_set(handle_t
695  
696         if (EXT4_I(inode)->i_extra_isize == 0)
697                 return -ENOSPC;
698 -       error = ext4_xattr_set_entry(i, s);
699 +       error = ext4_xattr_set_entry(i, s, handle, inode);
700         if (error)
701                 return error;
702         header = IHDR(inode, ext4_raw_inode(&is->iloc));
703 @@ -1092,7 +1406,7 @@ ext4_xattr_set_handle(handle_t *handle,
704                 .name = name,
705                 .value = value,
706                 .value_len = value_len,
707 -
708 +               .in_inode = 0,
709         };
710         struct ext4_xattr_ibody_find is = {
711                 .s = { .not_found = -ENODATA, },
712 @@ -1157,6 +1471,15 @@ ext4_xattr_set_handle(handle_t *handle,
713                                         goto cleanup;
714                         }
715                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
716 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
717 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
718 +                           error == -ENOSPC) {
719 +                               /* xattr not fit to block, store at external
720 +                                * inode */
721 +                               i.in_inode = 1;
722 +                               error = ext4_xattr_ibody_set(handle, inode,
723 +                                                            &i, &is);
724 +                       }
725                         if (error)
726                                 goto cleanup;
727                         if (!is.s.not_found) {
728 @@ -1203,9 +1526,22 @@ ext4_xattr_set(struct inode *inode, int
729                const void *value, size_t value_len, int flags)
730  {
731         handle_t *handle;
732 +       struct super_block *sb = inode->i_sb;
733         int error, retries = 0;
734         int credits = ext4_jbd2_credits_xattr(inode);
735  
736 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
737 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
738 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
739 +                                       sb->s_blocksize_bits;
740 +
741 +               /* For new inode */
742 +               credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
743 +
744 +               /* For data blocks of EA inode */
745 +               credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
746 +       }
747 +
748  retry:
749         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
750         if (IS_ERR(handle)) {
751 @@ -1217,7 +1553,7 @@ retry:
752                                               value, value_len, flags);
753                 error2 = ext4_journal_stop(handle);
754                 if (error == -ENOSPC &&
755 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
756 +                   ext4_should_retry_alloc(sb, &retries))
757                         goto retry;
758                 if (error == 0)
759                         error = error2;
760 @@ -1239,7 +1575,7 @@ static void ext4_xattr_shift_entries(str
761  
762         /* Adjust the value offsets of the entries */
763         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
764 -               if (!last->e_value_block && last->e_value_size) {
765 +               if (!last->e_value_inum && last->e_value_size) {
766                         new_offs = le16_to_cpu(last->e_value_offs) +
767                                                         value_offs_shift;
768                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
769 @@ -1477,21 +1813,135 @@ cleanup:
770  }
771  
772  
773 +#define EIA_INCR 16 /* must be 2^n */
774 +#define EIA_MASK (EIA_INCR - 1)
775 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
776 + * If @lea_ino_array is new or full it will be grown and the old
777 + * contents copied over.
778 + */
779 +static int
780 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
781 +{
782 +       if (*lea_ino_array == NULL) {
783 +               /*
784 +                * Start with 15 inodes, so it fits into a power-of-two size.
785 +                * If *lea_ino_array is NULL, this is essentially offsetof()
786 +                */
787 +               (*lea_ino_array) =
788 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
789 +                                        xia_inodes[EIA_MASK]),
790 +                               GFP_NOFS);
791 +               if (*lea_ino_array == NULL)
792 +                       return -ENOMEM;
793 +               (*lea_ino_array)->xia_count = 0;
794 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
795 +               /* expand the array once all 15 + n * 16 slots are full */
796 +               struct ext4_xattr_ino_array *new_array = NULL;
797 +               int count = (*lea_ino_array)->xia_count;
798 +
799 +               /* if new_array is NULL, this is essentially offsetof() */
800 +               new_array = kmalloc(
801 +                               offsetof(struct ext4_xattr_ino_array,
802 +                                        xia_inodes[count + EIA_INCR]),
803 +                               GFP_NOFS);
804 +               if (new_array == NULL)
805 +                       return -ENOMEM;
806 +               memcpy(new_array, *lea_ino_array,
807 +                      offsetof(struct ext4_xattr_ino_array,
808 +                               xia_inodes[count]));
809 +               kfree(*lea_ino_array);
810 +               *lea_ino_array = new_array;
811 +       }
812 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
813 +       return 0;
814 +}
815 +
816 +/**
817 + * Add xattr inode to orphan list
818 + */
819 +static int
820 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
821 +                       int credits, struct ext4_xattr_ino_array *lea_ino_array)
822 +{
823 +       struct inode *ea_inode = NULL;
824 +       int idx = 0, error = 0;
825 +
826 +       if (lea_ino_array == NULL)
827 +               return 0;
828 +
829 +       for (; idx < lea_ino_array->xia_count; ++idx) {
830 +               if (!ext4_handle_has_enough_credits(handle, credits)) {
831 +                       error = ext4_journal_extend(handle, credits);
832 +                       if (error > 0)
833 +                               error = ext4_journal_restart(handle, credits);
834 +
835 +                       if (error != 0) {
836 +                               ext4_warning(inode->i_sb,
837 +                                       "couldn't extend journal "
838 +                                       "(err %d)", error);
839 +                               return error;
840 +                       }
841 +               }
842 +               ea_inode = ext4_xattr_inode_iget(inode,
843 +                               lea_ino_array->xia_inodes[idx], &error);
844 +               if (error)
845 +                       continue;
846 +               ext4_orphan_add(handle, ea_inode);
847 +               /* the inode's i_count will be released by caller */
848 +       }
849 +
850 +       return 0;
851 +}
852  
853  /*
854   * ext4_xattr_delete_inode()
855   *
856 - * Free extended attribute resources associated with this inode. This
857 + * Free extended attribute resources associated with this inode. Traverse
858 + * all entries and unlink any xattr inodes associated with this inode. This
859   * is called immediately before an inode is freed. We have exclusive
860 - * access to the inode.
861 + * access to the inode. If an orphan inode is deleted it will also delete any
862 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
863 + * to ensure they belong to the parent inode and were not deleted already.
864   */
865 -void
866 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
867 +int
868 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
869 +                       struct ext4_xattr_ino_array **lea_ino_array)
870  {
871         struct buffer_head *bh = NULL;
872 +       struct ext4_xattr_ibody_header *header;
873 +       struct ext4_inode *raw_inode;
874 +       struct ext4_iloc iloc;
875 +       struct ext4_xattr_entry *entry;
876 +       int credits = 3, error = 0;
877  
878 -       if (!EXT4_I(inode)->i_file_acl)
879 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
880 +               goto delete_external_ea;
881 +
882 +       error = ext4_get_inode_loc(inode, &iloc);
883 +       if (error)
884                 goto cleanup;
885 +       raw_inode = ext4_raw_inode(&iloc);
886 +       header = IHDR(inode, raw_inode);
887 +       for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
888 +            entry = EXT4_XATTR_NEXT(entry)) {
889 +               if (!entry->e_value_inum)
890 +                       continue;
891 +               if (ext4_expand_ino_array(lea_ino_array,
892 +                                         entry->e_value_inum) != 0) {
893 +                       brelse(iloc.bh);
894 +                       goto cleanup;
895 +               }
896 +               entry->e_value_inum = 0;
897 +       }
898 +       brelse(iloc.bh);
899 +
900 +delete_external_ea:
901 +       if (!EXT4_I(inode)->i_file_acl) {
902 +               /* add xattr inode to orphan list */
903 +               ext4_xattr_inode_orphan_add(handle, inode, credits,
904 +                                               *lea_ino_array);
905 +               goto cleanup;
906 +       }
907         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
908         if (!bh) {
909                 EXT4_ERROR_INODE(inode, "block %llu read error",
910 @@ -1504,11 +1954,69 @@ ext4_xattr_delete_inode(handle_t *handle
911                                  EXT4_I(inode)->i_file_acl);
912                 goto cleanup;
913         }
914 +
915 +       for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
916 +            entry = EXT4_XATTR_NEXT(entry)) {
917 +               if (!entry->e_value_inum)
918 +                       continue;
919 +               if (ext4_expand_ino_array(lea_ino_array,
920 +                                         entry->e_value_inum) != 0)
921 +                       goto cleanup;
922 +               entry->e_value_inum = 0;
923 +       }
924 +
925 +       /* add xattr inode to orphan list */
926 +       error = ext4_xattr_inode_orphan_add(handle, inode, credits,
927 +                                       *lea_ino_array);
928 +       if (error != 0)
929 +               goto cleanup;
930 +
931 +       if (!IS_NOQUOTA(inode))
932 +               credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
933 +
934 +       if (!ext4_handle_has_enough_credits(handle, credits)) {
935 +               error = ext4_journal_extend(handle, credits);
936 +               if (error > 0)
937 +                       error = ext4_journal_restart(handle, credits);
938 +               if (error != 0) {
939 +                       ext4_warning(inode->i_sb,
940 +                               "couldn't extend journal (err %d)", error);
941 +                       goto cleanup;
942 +               }
943 +       }
944 +
945         ext4_xattr_release_block(handle, inode, bh);
946         EXT4_I(inode)->i_file_acl = 0;
947  
948  cleanup:
949         brelse(bh);
950 +
951 +       return error;
952 +}
953 +
954 +void
955 +ext4_xattr_inode_array_free(struct inode *inode,
956 +                           struct ext4_xattr_ino_array *lea_ino_array)
957 +{
958 +       struct inode    *ea_inode = NULL;
959 +       int             idx = 0;
960 +       int             err;
961 +
962 +       if (lea_ino_array == NULL)
963 +               return;
964 +
965 +       for (; idx < lea_ino_array->xia_count; ++idx) {
966 +               ea_inode = ext4_xattr_inode_iget(inode,
967 +                               lea_ino_array->xia_inodes[idx], &err);
968 +               if (err)
969 +                       continue;
970 +               /* for inode's i_count get from ext4_xattr_delete_inode */
971 +               if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
972 +                       iput(ea_inode);
973 +               clear_nlink(ea_inode);
974 +               iput(ea_inode);
975 +       }
976 +       kfree(lea_ino_array);
977  }
978  
979  /*
980 @@ -1578,10 +2086,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
981                     entry1->e_name_index != entry2->e_name_index ||
982                     entry1->e_name_len != entry2->e_name_len ||
983                     entry1->e_value_size != entry2->e_value_size ||
984 +                   entry1->e_value_inum != entry2->e_value_inum ||
985                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
986                         return 1;
987 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
988 -                       return -EIO;
989                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
990                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
991                            le32_to_cpu(entry1->e_value_size)))
992 @@ -1665,7 +2172,7 @@ static inline void ext4_xattr_hash_entry
993                        *name++;
994         }
995  
996 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
997 +       if (!entry->e_value_inum && entry->e_value_size) {
998                 __le32 *value = (__le32 *)((char *)header +
999                         le16_to_cpu(entry->e_value_offs));
1000                 for (n = (le32_to_cpu(entry->e_value_size) +
1001 Index: linux-stage/fs/ext4/xattr.h
1002 ===================================================================
1003 --- linux-stage.orig/fs/ext4/xattr.h
1004 +++ linux-stage/fs/ext4/xattr.h
1005 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
1006         __u8    e_name_len;     /* length of name */
1007         __u8    e_name_index;   /* attribute name index */
1008         __le16  e_value_offs;   /* offset in disk block of value */
1009 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
1010 +       __le32  e_value_inum;   /* inode in which the value is stored */
1011         __le32  e_value_size;   /* size of attribute value */
1012         __le32  e_hash;         /* hash value of name and value */
1013         char    e_name[0];      /* attribute name */
1014 @@ -67,6 +67,26 @@ struct ext4_xattr_entry {
1015                 EXT4_I(inode)->i_extra_isize))
1016  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
1017  
1018 +/*
1019 + * Link EA inode back to parent one using i_mtime field.
1020 + * Extra integer type conversion added to ignore higher
1021 + * bits in i_mtime.tv_sec which might be set by ext4_get()
1022 + */
1023 +#define EXT4_XATTR_INODE_SET_PARENT(inode, inum)       \
1024 +do {                                                   \
1025 +       (inode)->i_mtime.tv_sec = inum;                 \
1026 +} while(0)
1027 +
1028 +#define EXT4_XATTR_INODE_GET_PARENT(inode)             \
1029 +       ((__u32)(inode)->i_mtime.tv_sec)
1030 +
1031 +/*
1032 + * The minimum size of EA value when you start storing it in an external inode
1033 + * size of block - size of header - size of 1 entry - 4 null bytes
1034 +*/
1035 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
1036 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
1037 +
1038  #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
1039  #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
1040  #define BFIRST(bh) ENTRY(BHDR(bh)+1)
1041 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
1042  #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
1043  
1044  struct ext4_xattr_info {
1045 -       int name_index;
1046         const char *name;
1047         const void *value;
1048         size_t value_len;
1049 +       int name_index;
1050 +       int in_inode;
1051  };
1052  
1053  struct ext4_xattr_search {
1054 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
1055  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1056  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1057  
1058 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1059 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1060 +                                          int *err);
1061 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1062 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1063 +                                  struct ext4_xattr_ino_array **array);
1064 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1065 +                                       struct ext4_xattr_ino_array *array);
1066  extern void ext4_xattr_put_super(struct super_block *);
1067  
1068  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1069 Index: linux-stage/fs/ext4/ialloc.c
1070 ===================================================================
1071 --- linux-stage.orig/fs/ext4/ialloc.c
1072 +++ linux-stage/fs/ext4/ialloc.c
1073 @@ -269,7 +269,6 @@ void ext4_free_inode(handle_t *handle, s
1074          * as writing the quota to disk may need the lock as well.
1075          */
1076         dquot_initialize(inode);
1077 -       ext4_xattr_delete_inode(handle, inode);
1078         dquot_free_inode(inode);
1079         dquot_drop(inode);
1080  
1081 Index: linux-stage/fs/ext4/inline.c
1082 ===================================================================
1083 --- linux-stage.orig/fs/ext4/inline.c
1084 +++ linux-stage/fs/ext4/inline.c
1085 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1086  
1087         /* Compute min_offs. */
1088         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1089 -               if (!entry->e_value_block && entry->e_value_size) {
1090 +               if (!entry->e_value_inum && entry->e_value_size) {
1091                         size_t offs = le16_to_cpu(entry->e_value_offs);
1092                         if (offs < min_offs)
1093                                 min_offs = offs;