]>
Commit | Line | Data |
---|---|---|
2cb7cef9 BS |
1 | From: Tao Ma <tao.ma@oracle.com> |
2 | Subject: [PATCH 14/16] ocfs2: Enable xattr set in index btree | |
3 | Patch-mainline: 2.6.28? | |
4 | References: FATE302067 | |
5 | ||
6 | Where the previous patches added the ability of list/get xattr in buckets | |
7 | for ocfs2, this patch enables ocfs2 to store large numbers of EAs. | |
8 | ||
9 | The original design doc is written by Mark Fasheh, and it can be found in | |
10 | http://oss.oracle.com/osswiki/OCFS2/DesignDocs/IndexedEATrees. I only had to | |
11 | make small modifications to it. | |
12 | ||
13 | First, because the bucket size is 4K, a new field named xh_free_start is added | |
14 | in ocfs2_xattr_header to indicate the next valid name/value offset in a bucket. | |
15 | It is used when we store new EA name/value. With this field, we can find the | |
16 | place more quickly and what's more, we don't need to sort the name/value every | |
17 | time to let the last entry indicate the next unused space. This makes the | |
18 | insert operation more efficient for blocksizes smaller than 4k. | |
19 | ||
20 | Because of the new xh_free_start, another field named as xh_name_value_len is | |
21 | also added in ocfs2_xattr_header. It records the total length of all the | |
22 | name/values in the bucket. We need this so that we can check it and defragment | |
23 | the bucket if there is not enough contiguous free space. | |
24 | ||
25 | An xattr insertion looks like this: | |
26 | 1. xattr_index_block_find: find the right bucket by the name_hash, say bucketA. | |
27 | 2. check whether there is enough space in bucketA. If yes, insert it directly | |
28 | and modify xh_free_start and xh_name_value_len accordingly. If not, check | |
29 | xh_name_value_len to see whether we can store this by defragment the bucket. | |
30 | If yes, defragment it and go on insertion. | |
31 | 3. If defragement doesn't work, check whether there is new empty bucket in | |
32 | the clusters within this extent record. If yes, init the new bucket and move | |
33 | all the buckets after bucketA one by one to the next bucket. Move half of the | |
34 | entries in bucketA to the next bucket and go on insertion. | |
35 | 4. If there is no new bucket, grow the extent tree. | |
36 | ||
37 | As for xattr deletion, we will delete an xattr bucket when all it's xattrs | |
38 | are removed and move all the buckets after it to the previous one. When all | |
39 | the xattr buckets in an extend record are freed, free this extend records | |
40 | from ocfs2_xattr_tree. | |
41 | ||
42 | Signed-off-by: Tao Ma <tao.ma@oracle.com> | |
43 | Signed-off-by: Mark Fasheh <mfasheh@suse.com> | |
44 | --- | |
45 | fs/ocfs2/xattr.c | 2267 +++++++++++++++++++++++++++++++++++++++++++++++++++++- | |
46 | fs/ocfs2/xattr.h | 8 + | |
47 | 2 files changed, 2273 insertions(+), 2 deletions(-) | |
48 | ||
49 | diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c | |
50 | index a5ca066..408553c 100644 | |
51 | --- a/fs/ocfs2/xattr.c | |
52 | +++ b/fs/ocfs2/xattr.c | |
53 | @@ -36,6 +36,7 @@ | |
54 | #include <linux/mount.h> | |
55 | #include <linux/writeback.h> | |
56 | #include <linux/falloc.h> | |
57 | +#include <linux/sort.h> | |
58 | ||
59 | #define MLOG_MASK_PREFIX ML_XATTR | |
60 | #include <cluster/masklog.h> | |
61 | @@ -139,6 +140,13 @@ static int ocfs2_xattr_tree_list_index_block(struct inode *inode, | |
62 | char *buffer, | |
63 | size_t buffer_size); | |
64 | ||
65 | +static int ocfs2_xattr_create_index_block(struct inode *inode, | |
66 | + struct ocfs2_xattr_search *xs); | |
67 | + | |
68 | +static int ocfs2_xattr_set_entry_index_block(struct inode *inode, | |
69 | + struct ocfs2_xattr_info *xi, | |
70 | + struct ocfs2_xattr_search *xs); | |
71 | + | |
72 | static inline struct xattr_handler *ocfs2_xattr_handler(int name_index) | |
73 | { | |
74 | struct xattr_handler *handler = NULL; | |
75 | @@ -1784,6 +1792,52 @@ cleanup: | |
76 | } | |
77 | ||
78 | /* | |
79 | + * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree | |
80 | + * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header | |
81 | + * re-initialized. | |
82 | + */ | |
83 | +static int ocfs2_restore_xattr_block(struct inode *inode, | |
84 | + struct ocfs2_xattr_search *xs) | |
85 | +{ | |
86 | + int ret; | |
87 | + handle_t *handle; | |
88 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
89 | + struct ocfs2_xattr_block *xb = | |
90 | + (struct ocfs2_xattr_block *)xs->xattr_bh->b_data; | |
91 | + struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list; | |
92 | + u16 xb_flags = le16_to_cpu(xb->xb_flags); | |
93 | + | |
94 | + BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) || | |
95 | + le16_to_cpu(el->l_next_free_rec) != 0); | |
96 | + | |
97 | + handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS); | |
98 | + if (IS_ERR(handle)) { | |
99 | + ret = PTR_ERR(handle); | |
100 | + handle = NULL; | |
101 | + goto out; | |
102 | + } | |
103 | + | |
104 | + ret = ocfs2_journal_access(handle, inode, xs->xattr_bh, | |
105 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
106 | + if (ret < 0) { | |
107 | + mlog_errno(ret); | |
108 | + goto out_commit; | |
109 | + } | |
110 | + | |
111 | + memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize - | |
112 | + offsetof(struct ocfs2_xattr_block, xb_attrs)); | |
113 | + | |
114 | + xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED); | |
115 | + | |
116 | + ocfs2_journal_dirty(handle, xs->xattr_bh); | |
117 | + | |
118 | +out_commit: | |
119 | + ocfs2_commit_trans(osb, handle); | |
120 | +out: | |
121 | + return ret; | |
122 | +} | |
123 | + | |
124 | +/* | |
125 | * ocfs2_xattr_block_set() | |
126 | * | |
127 | * Set, replace or remove an extended attribute into external block. | |
128 | @@ -1878,10 +1932,25 @@ out: | |
129 | ocfs2_free_alloc_context(meta_ac); | |
130 | if (ret < 0) | |
131 | return ret; | |
132 | + } else | |
133 | + xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data; | |
134 | + | |
135 | + if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) { | |
136 | + /* Set extended attribute into external block */ | |
137 | + ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL); | |
138 | + if (!ret || ret != -ENOSPC) | |
139 | + goto end; | |
140 | + | |
141 | + ret = ocfs2_xattr_create_index_block(inode, xs); | |
142 | + if (ret) | |
143 | + goto end; | |
144 | } | |
145 | ||
146 | - /* Set extended attribute into external block */ | |
147 | - ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL); | |
148 | + ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs); | |
149 | + if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0) | |
150 | + ret = ocfs2_restore_xattr_block(inode, xs); | |
151 | + | |
152 | +end: | |
153 | ||
154 | return ret; | |
155 | } | |
156 | @@ -1903,6 +1972,7 @@ int ocfs2_xattr_set(struct inode *inode, | |
157 | struct buffer_head *di_bh = NULL; | |
158 | struct ocfs2_dinode *di; | |
159 | int ret; | |
160 | + u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
161 | ||
162 | struct ocfs2_xattr_info xi = { | |
163 | .name_index = name_index, | |
164 | @@ -2001,6 +2071,8 @@ cleanup: | |
165 | ocfs2_inode_unlock(inode, 1); | |
166 | brelse(di_bh); | |
167 | brelse(xbs.xattr_bh); | |
168 | + for (i = 0; i < blk_per_bucket; i++) | |
169 | + brelse(xbs.bucket.bhs[i]); | |
170 | ||
171 | return ret; | |
172 | } | |
173 | @@ -2491,3 +2563,2194 @@ static int ocfs2_xattr_tree_list_index_block(struct inode *inode, | |
174 | out: | |
175 | return ret; | |
176 | } | |
177 | + | |
178 | +static int cmp_xe(const void *a, const void *b) | |
179 | +{ | |
180 | + const struct ocfs2_xattr_entry *l = a, *r = b; | |
181 | + u32 l_hash = le32_to_cpu(l->xe_name_hash); | |
182 | + u32 r_hash = le32_to_cpu(r->xe_name_hash); | |
183 | + | |
184 | + if (l_hash > r_hash) | |
185 | + return 1; | |
186 | + if (l_hash < r_hash) | |
187 | + return -1; | |
188 | + return 0; | |
189 | +} | |
190 | + | |
191 | +static void swap_xe(void *a, void *b, int size) | |
192 | +{ | |
193 | + struct ocfs2_xattr_entry *l = a, *r = b, tmp; | |
194 | + | |
195 | + tmp = *l; | |
196 | + memcpy(l, r, sizeof(struct ocfs2_xattr_entry)); | |
197 | + memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry)); | |
198 | +} | |
199 | + | |
200 | +/* | |
201 | + * When the ocfs2_xattr_block is filled up, new bucket will be created | |
202 | + * and all the xattr entries will be moved to the new bucket. | |
203 | + * Note: we need to sort the entries since they are not saved in order | |
204 | + * in the ocfs2_xattr_block. | |
205 | + */ | |
206 | +static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode, | |
207 | + struct buffer_head *xb_bh, | |
208 | + struct buffer_head *xh_bh, | |
209 | + struct buffer_head *data_bh) | |
210 | +{ | |
211 | + int i, blocksize = inode->i_sb->s_blocksize; | |
212 | + u16 offset, size, off_change; | |
213 | + struct ocfs2_xattr_entry *xe; | |
214 | + struct ocfs2_xattr_block *xb = | |
215 | + (struct ocfs2_xattr_block *)xb_bh->b_data; | |
216 | + struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header; | |
217 | + struct ocfs2_xattr_header *xh = | |
218 | + (struct ocfs2_xattr_header *)xh_bh->b_data; | |
219 | + u16 count = le16_to_cpu(xb_xh->xh_count); | |
220 | + char *target = xh_bh->b_data, *src = xb_bh->b_data; | |
221 | + | |
222 | + mlog(0, "cp xattr from block %llu to bucket %llu\n", | |
223 | + (unsigned long long)xb_bh->b_blocknr, | |
224 | + (unsigned long long)xh_bh->b_blocknr); | |
225 | + | |
226 | + memset(xh_bh->b_data, 0, blocksize); | |
227 | + if (data_bh) | |
228 | + memset(data_bh->b_data, 0, blocksize); | |
229 | + /* | |
230 | + * Since the xe_name_offset is based on ocfs2_xattr_header, | |
231 | + * there is a offset change corresponding to the change of | |
232 | + * ocfs2_xattr_header's position. | |
233 | + */ | |
234 | + off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header); | |
235 | + xe = &xb_xh->xh_entries[count - 1]; | |
236 | + offset = le16_to_cpu(xe->xe_name_offset) + off_change; | |
237 | + size = blocksize - offset; | |
238 | + | |
239 | + /* copy all the names and values. */ | |
240 | + if (data_bh) | |
241 | + target = data_bh->b_data; | |
242 | + memcpy(target + offset, src + offset, size); | |
243 | + | |
244 | + /* Init new header now. */ | |
245 | + xh->xh_count = xb_xh->xh_count; | |
246 | + xh->xh_num_buckets = cpu_to_le16(1); | |
247 | + xh->xh_name_value_len = cpu_to_le16(size); | |
248 | + xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size); | |
249 | + | |
250 | + /* copy all the entries. */ | |
251 | + target = xh_bh->b_data; | |
252 | + offset = offsetof(struct ocfs2_xattr_header, xh_entries); | |
253 | + size = count * sizeof(struct ocfs2_xattr_entry); | |
254 | + memcpy(target + offset, (char *)xb_xh + offset, size); | |
255 | + | |
256 | + /* Change the xe offset for all the xe because of the move. */ | |
257 | + off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize + | |
258 | + offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header); | |
259 | + for (i = 0; i < count; i++) | |
260 | + le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change); | |
261 | + | |
262 | + mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n", | |
263 | + offset, size, off_change); | |
264 | + | |
265 | + sort(target + offset, count, sizeof(struct ocfs2_xattr_entry), | |
266 | + cmp_xe, swap_xe); | |
267 | +} | |
268 | + | |
269 | +/* | |
270 | + * After we move xattr from block to index btree, we have to | |
271 | + * update ocfs2_xattr_search to the new xe and base. | |
272 | + * | |
273 | + * When the entry is in xattr block, xattr_bh indicates the storage place. | |
274 | + * While if the entry is in index b-tree, "bucket" indicates the | |
275 | + * real place of the xattr. | |
276 | + */ | |
277 | +static int ocfs2_xattr_update_xattr_search(struct inode *inode, | |
278 | + struct ocfs2_xattr_search *xs, | |
279 | + struct buffer_head *old_bh, | |
280 | + struct buffer_head *new_bh) | |
281 | +{ | |
282 | + int ret = 0; | |
283 | + char *buf = old_bh->b_data; | |
284 | + struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf; | |
285 | + struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header; | |
286 | + int i, blocksize = inode->i_sb->s_blocksize; | |
287 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
288 | + | |
289 | + xs->bucket.bhs[0] = new_bh; | |
290 | + get_bh(new_bh); | |
291 | + xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data; | |
292 | + xs->header = xs->bucket.xh; | |
293 | + | |
294 | + xs->base = new_bh->b_data; | |
295 | + xs->end = xs->base + inode->i_sb->s_blocksize; | |
296 | + | |
297 | + if (!xs->not_found) { | |
298 | + if (OCFS2_XATTR_BUCKET_SIZE != blocksize) { | |
299 | + ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), | |
300 | + xs->bucket.bhs[0]->b_blocknr + 1, | |
301 | + blk_per_bucket - 1, &xs->bucket.bhs[1], | |
302 | + OCFS2_BH_CACHED, inode); | |
303 | + if (ret) { | |
304 | + mlog_errno(ret); | |
305 | + return ret; | |
306 | + } | |
307 | + | |
308 | + i = xs->here - old_xh->xh_entries; | |
309 | + xs->here = &xs->header->xh_entries[i]; | |
310 | + } | |
311 | + } | |
312 | + | |
313 | + return ret; | |
314 | +} | |
315 | + | |
316 | +static int ocfs2_xattr_create_index_block(struct inode *inode, | |
317 | + struct ocfs2_xattr_search *xs) | |
318 | +{ | |
319 | + int ret, credits = OCFS2_SUBALLOC_ALLOC; | |
320 | + u32 bit_off, len; | |
321 | + u64 blkno; | |
322 | + handle_t *handle; | |
323 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
324 | + struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
325 | + struct ocfs2_alloc_context *data_ac; | |
326 | + struct buffer_head *xh_bh = NULL, *data_bh = NULL; | |
327 | + struct buffer_head *xb_bh = xs->xattr_bh; | |
328 | + struct ocfs2_xattr_block *xb = | |
329 | + (struct ocfs2_xattr_block *)xb_bh->b_data; | |
330 | + struct ocfs2_xattr_tree_root *xr; | |
331 | + u16 xb_flags = le16_to_cpu(xb->xb_flags); | |
332 | + u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
333 | + | |
334 | + mlog(0, "create xattr index block for %llu\n", | |
335 | + (unsigned long long)xb_bh->b_blocknr); | |
336 | + | |
337 | + BUG_ON(xb_flags & OCFS2_XATTR_INDEXED); | |
338 | + | |
339 | + ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
340 | + if (ret) { | |
341 | + mlog_errno(ret); | |
342 | + goto out; | |
343 | + } | |
344 | + | |
345 | + /* | |
346 | + * XXX: | |
347 | + * We can use this lock for now, and maybe move to a dedicated mutex | |
348 | + * if performance becomes a problem later. | |
349 | + */ | |
350 | + down_write(&oi->ip_alloc_sem); | |
351 | + | |
352 | + /* | |
353 | + * 3 more credits, one for xattr block update, one for the 1st block | |
354 | + * of the new xattr bucket and one for the value/data. | |
355 | + */ | |
356 | + credits += 3; | |
357 | + handle = ocfs2_start_trans(osb, credits); | |
358 | + if (IS_ERR(handle)) { | |
359 | + ret = PTR_ERR(handle); | |
360 | + mlog_errno(ret); | |
361 | + goto out_sem; | |
362 | + } | |
363 | + | |
364 | + ret = ocfs2_journal_access(handle, inode, xb_bh, | |
365 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
366 | + if (ret) { | |
367 | + mlog_errno(ret); | |
368 | + goto out_commit; | |
369 | + } | |
370 | + | |
371 | + ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len); | |
372 | + if (ret) { | |
373 | + mlog_errno(ret); | |
374 | + goto out_commit; | |
375 | + } | |
376 | + | |
377 | + /* | |
378 | + * The bucket may spread in many blocks, and | |
379 | + * we will only touch the 1st block and the last block | |
380 | + * in the whole bucket(one for entry and one for data). | |
381 | + */ | |
382 | + blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
383 | + | |
384 | + mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno); | |
385 | + | |
386 | + xh_bh = sb_getblk(inode->i_sb, blkno); | |
387 | + if (!xh_bh) { | |
388 | + ret = -EIO; | |
389 | + mlog_errno(ret); | |
390 | + goto out_commit; | |
391 | + } | |
392 | + | |
393 | + ocfs2_set_new_buffer_uptodate(inode, xh_bh); | |
394 | + | |
395 | + ret = ocfs2_journal_access(handle, inode, xh_bh, | |
396 | + OCFS2_JOURNAL_ACCESS_CREATE); | |
397 | + if (ret) { | |
398 | + mlog_errno(ret); | |
399 | + goto out_commit; | |
400 | + } | |
401 | + | |
402 | + if (bpb > 1) { | |
403 | + data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1); | |
404 | + if (!data_bh) { | |
405 | + ret = -EIO; | |
406 | + mlog_errno(ret); | |
407 | + goto out_commit; | |
408 | + } | |
409 | + | |
410 | + ocfs2_set_new_buffer_uptodate(inode, data_bh); | |
411 | + | |
412 | + ret = ocfs2_journal_access(handle, inode, data_bh, | |
413 | + OCFS2_JOURNAL_ACCESS_CREATE); | |
414 | + if (ret) { | |
415 | + mlog_errno(ret); | |
416 | + goto out_commit; | |
417 | + } | |
418 | + } | |
419 | + | |
420 | + ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh); | |
421 | + | |
422 | + ocfs2_journal_dirty(handle, xh_bh); | |
423 | + if (data_bh) | |
424 | + ocfs2_journal_dirty(handle, data_bh); | |
425 | + | |
426 | + ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh); | |
427 | + | |
428 | + /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */ | |
429 | + memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize - | |
430 | + offsetof(struct ocfs2_xattr_block, xb_attrs)); | |
431 | + | |
432 | + xr = &xb->xb_attrs.xb_root; | |
433 | + xr->xt_clusters = cpu_to_le32(1); | |
434 | + xr->xt_last_eb_blk = 0; | |
435 | + xr->xt_list.l_tree_depth = 0; | |
436 | + xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb)); | |
437 | + xr->xt_list.l_next_free_rec = cpu_to_le16(1); | |
438 | + | |
439 | + xr->xt_list.l_recs[0].e_cpos = 0; | |
440 | + xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno); | |
441 | + xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1); | |
442 | + | |
443 | + xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED); | |
444 | + | |
445 | + ret = ocfs2_journal_dirty(handle, xb_bh); | |
446 | + if (ret) { | |
447 | + mlog_errno(ret); | |
448 | + goto out_commit; | |
449 | + } | |
450 | + | |
451 | +out_commit: | |
452 | + ocfs2_commit_trans(osb, handle); | |
453 | + | |
454 | +out_sem: | |
455 | + up_write(&oi->ip_alloc_sem); | |
456 | + | |
457 | +out: | |
458 | + if (data_ac) | |
459 | + ocfs2_free_alloc_context(data_ac); | |
460 | + | |
461 | + brelse(xh_bh); | |
462 | + brelse(data_bh); | |
463 | + | |
464 | + return ret; | |
465 | +} | |
466 | + | |
467 | +static int cmp_xe_offset(const void *a, const void *b) | |
468 | +{ | |
469 | + const struct ocfs2_xattr_entry *l = a, *r = b; | |
470 | + u32 l_name_offset = le16_to_cpu(l->xe_name_offset); | |
471 | + u32 r_name_offset = le16_to_cpu(r->xe_name_offset); | |
472 | + | |
473 | + if (l_name_offset < r_name_offset) | |
474 | + return 1; | |
475 | + if (l_name_offset > r_name_offset) | |
476 | + return -1; | |
477 | + return 0; | |
478 | +} | |
479 | + | |
480 | +/* | |
481 | + * defrag a xattr bucket if we find that the bucket has some | |
482 | + * holes beteen name/value pairs. | |
483 | + * We will move all the name/value pairs to the end of the bucket | |
484 | + * so that we can spare some space for insertion. | |
485 | + */ | |
486 | +static int ocfs2_defrag_xattr_bucket(struct inode *inode, | |
487 | + struct ocfs2_xattr_bucket *bucket) | |
488 | +{ | |
489 | + int ret, i; | |
490 | + size_t end, offset, len, value_len; | |
491 | + struct ocfs2_xattr_header *xh; | |
492 | + char *entries, *buf, *bucket_buf = NULL; | |
493 | + u64 blkno = bucket->bhs[0]->b_blocknr; | |
494 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
495 | + u16 xh_free_start; | |
496 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
497 | + size_t blocksize = inode->i_sb->s_blocksize; | |
498 | + handle_t *handle; | |
499 | + struct buffer_head **bhs; | |
500 | + struct ocfs2_xattr_entry *xe; | |
501 | + | |
502 | + bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket, | |
503 | + GFP_NOFS); | |
504 | + if (!bhs) | |
505 | + return -ENOMEM; | |
506 | + | |
507 | + ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs, | |
508 | + OCFS2_BH_CACHED, inode); | |
509 | + if (ret) | |
510 | + goto out; | |
511 | + | |
512 | + /* | |
513 | + * In order to make the operation more efficient and generic, | |
514 | + * we copy all the blocks into a contiguous memory and do the | |
515 | + * defragment there, so if anything is error, we will not touch | |
516 | + * the real block. | |
517 | + */ | |
518 | + bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS); | |
519 | + if (!bucket_buf) { | |
520 | + ret = -EIO; | |
521 | + goto out; | |
522 | + } | |
523 | + | |
524 | + buf = bucket_buf; | |
525 | + for (i = 0; i < blk_per_bucket; i++, buf += blocksize) | |
526 | + memcpy(buf, bhs[i]->b_data, blocksize); | |
527 | + | |
528 | + handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket); | |
529 | + if (IS_ERR(handle)) { | |
530 | + ret = PTR_ERR(handle); | |
531 | + handle = NULL; | |
532 | + mlog_errno(ret); | |
533 | + goto out; | |
534 | + } | |
535 | + | |
536 | + for (i = 0; i < blk_per_bucket; i++) { | |
537 | + ret = ocfs2_journal_access(handle, inode, bhs[i], | |
538 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
539 | + if (ret < 0) { | |
540 | + mlog_errno(ret); | |
541 | + goto commit; | |
542 | + } | |
543 | + } | |
544 | + | |
545 | + xh = (struct ocfs2_xattr_header *)bucket_buf; | |
546 | + entries = (char *)xh->xh_entries; | |
547 | + xh_free_start = le16_to_cpu(xh->xh_free_start); | |
548 | + | |
549 | + mlog(0, "adjust xattr bucket in %llu, count = %u, " | |
550 | + "xh_free_start = %u, xh_name_value_len = %u.\n", | |
551 | + blkno, le16_to_cpu(xh->xh_count), xh_free_start, | |
552 | + le16_to_cpu(xh->xh_name_value_len)); | |
553 | + | |
554 | + /* | |
555 | + * sort all the entries by their offset. | |
556 | + * the largest will be the first, so that we can | |
557 | + * move them to the end one by one. | |
558 | + */ | |
559 | + sort(entries, le16_to_cpu(xh->xh_count), | |
560 | + sizeof(struct ocfs2_xattr_entry), | |
561 | + cmp_xe_offset, swap_xe); | |
562 | + | |
563 | + /* Move all name/values to the end of the bucket. */ | |
564 | + xe = xh->xh_entries; | |
565 | + end = OCFS2_XATTR_BUCKET_SIZE; | |
566 | + for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) { | |
567 | + offset = le16_to_cpu(xe->xe_name_offset); | |
568 | + if (ocfs2_xattr_is_local(xe)) | |
569 | + value_len = OCFS2_XATTR_SIZE( | |
570 | + le64_to_cpu(xe->xe_value_size)); | |
571 | + else | |
572 | + value_len = OCFS2_XATTR_ROOT_SIZE; | |
573 | + len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len; | |
574 | + | |
575 | + /* | |
576 | + * We must make sure that the name/value pair | |
577 | + * exist in the same block. So adjust end to | |
578 | + * the previous block end if needed. | |
579 | + */ | |
580 | + if (((end - len) / blocksize != | |
581 | + (end - 1) / blocksize)) | |
582 | + end = end - end % blocksize; | |
583 | + | |
584 | + if (end > offset + len) { | |
585 | + memmove(bucket_buf + end - len, | |
586 | + bucket_buf + offset, len); | |
587 | + xe->xe_name_offset = cpu_to_le16(end - len); | |
588 | + } | |
589 | + | |
590 | + mlog_bug_on_msg(end < offset + len, "Defrag check failed for " | |
591 | + "bucket %llu\n", (unsigned long long)blkno); | |
592 | + | |
593 | + end -= len; | |
594 | + } | |
595 | + | |
596 | + mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for " | |
597 | + "bucket %llu\n", (unsigned long long)blkno); | |
598 | + | |
599 | + if (xh_free_start == end) | |
600 | + goto commit; | |
601 | + | |
602 | + memset(bucket_buf + xh_free_start, 0, end - xh_free_start); | |
603 | + xh->xh_free_start = cpu_to_le16(end); | |
604 | + | |
605 | + /* sort the entries by their name_hash. */ | |
606 | + sort(entries, le16_to_cpu(xh->xh_count), | |
607 | + sizeof(struct ocfs2_xattr_entry), | |
608 | + cmp_xe, swap_xe); | |
609 | + | |
610 | + buf = bucket_buf; | |
611 | + for (i = 0; i < blk_per_bucket; i++, buf += blocksize) { | |
612 | + memcpy(bhs[i]->b_data, buf, blocksize); | |
613 | + ocfs2_journal_dirty(handle, bhs[i]); | |
614 | + } | |
615 | + | |
616 | +commit: | |
617 | + ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); | |
618 | +out: | |
619 | + | |
620 | + if (bhs) { | |
621 | + for (i = 0; i < blk_per_bucket; i++) | |
622 | + brelse(bhs[i]); | |
623 | + } | |
624 | + kfree(bhs); | |
625 | + | |
626 | + kfree(bucket_buf); | |
627 | + return ret; | |
628 | +} | |
629 | + | |
630 | +/* | |
631 | + * Move half nums of the xattr bucket in the previous cluster to this new | |
632 | + * cluster. We only touch the last cluster of the previous extend record. | |
633 | + * | |
634 | + * first_bh is the first buffer_head of a series of bucket in the same | |
635 | + * extent rec and header_bh is the header of one bucket in this cluster. | |
636 | + * They will be updated if we move the data header_bh contains to the new | |
637 | + * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster. | |
638 | + */ | |
639 | +static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode, | |
640 | + handle_t *handle, | |
641 | + struct buffer_head **first_bh, | |
642 | + struct buffer_head **header_bh, | |
643 | + u64 new_blkno, | |
644 | + u64 prev_blkno, | |
645 | + u32 num_clusters, | |
646 | + u32 *first_hash) | |
647 | +{ | |
648 | + int i, ret, credits; | |
649 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
650 | + int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); | |
651 | + int num_buckets = ocfs2_xattr_buckets_per_cluster(osb); | |
652 | + int blocksize = inode->i_sb->s_blocksize; | |
653 | + struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL; | |
654 | + struct ocfs2_xattr_header *new_xh; | |
655 | + struct ocfs2_xattr_header *xh = | |
656 | + (struct ocfs2_xattr_header *)((*first_bh)->b_data); | |
657 | + | |
658 | + BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets); | |
659 | + BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize); | |
660 | + | |
661 | + prev_bh = *first_bh; | |
662 | + get_bh(prev_bh); | |
663 | + xh = (struct ocfs2_xattr_header *)prev_bh->b_data; | |
664 | + | |
665 | + prev_blkno += (num_clusters - 1) * bpc + bpc / 2; | |
666 | + | |
667 | + mlog(0, "move half of xattrs in cluster %llu to %llu\n", | |
668 | + prev_blkno, new_blkno); | |
669 | + | |
670 | + /* | |
671 | + * We need to update the 1st half of the new cluster and | |
672 | + * 1 more for the update of the 1st bucket of the previous | |
673 | + * extent record. | |
674 | + */ | |
675 | + credits = bpc / 2 + 1; | |
676 | + ret = ocfs2_extend_trans(handle, credits); | |
677 | + if (ret) { | |
678 | + mlog_errno(ret); | |
679 | + goto out; | |
680 | + } | |
681 | + | |
682 | + ret = ocfs2_journal_access(handle, inode, prev_bh, | |
683 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
684 | + if (ret) { | |
685 | + mlog_errno(ret); | |
686 | + goto out; | |
687 | + } | |
688 | + | |
689 | + for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) { | |
690 | + old_bh = new_bh = NULL; | |
691 | + new_bh = sb_getblk(inode->i_sb, new_blkno); | |
692 | + if (!new_bh) { | |
693 | + ret = -EIO; | |
694 | + mlog_errno(ret); | |
695 | + goto out; | |
696 | + } | |
697 | + | |
698 | + ocfs2_set_new_buffer_uptodate(inode, new_bh); | |
699 | + | |
700 | + ret = ocfs2_journal_access(handle, inode, new_bh, | |
701 | + OCFS2_JOURNAL_ACCESS_CREATE); | |
702 | + if (ret < 0) { | |
703 | + mlog_errno(ret); | |
704 | + brelse(new_bh); | |
705 | + goto out; | |
706 | + } | |
707 | + | |
708 | + ret = ocfs2_read_block(osb, prev_blkno, | |
709 | + &old_bh, OCFS2_BH_CACHED, inode); | |
710 | + if (ret < 0) { | |
711 | + mlog_errno(ret); | |
712 | + brelse(new_bh); | |
713 | + goto out; | |
714 | + } | |
715 | + | |
716 | + memcpy(new_bh->b_data, old_bh->b_data, blocksize); | |
717 | + | |
718 | + if (i == 0) { | |
719 | + new_xh = (struct ocfs2_xattr_header *)new_bh->b_data; | |
720 | + new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2); | |
721 | + | |
722 | + if (first_hash) | |
723 | + *first_hash = le32_to_cpu( | |
724 | + new_xh->xh_entries[0].xe_name_hash); | |
725 | + new_first_bh = new_bh; | |
726 | + get_bh(new_first_bh); | |
727 | + } | |
728 | + | |
729 | + ocfs2_journal_dirty(handle, new_bh); | |
730 | + | |
731 | + if (*header_bh == old_bh) { | |
732 | + brelse(*header_bh); | |
733 | + *header_bh = new_bh; | |
734 | + get_bh(*header_bh); | |
735 | + | |
736 | + brelse(*first_bh); | |
737 | + *first_bh = new_first_bh; | |
738 | + get_bh(*first_bh); | |
739 | + } | |
740 | + brelse(new_bh); | |
741 | + brelse(old_bh); | |
742 | + } | |
743 | + | |
744 | + le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2)); | |
745 | + | |
746 | + ocfs2_journal_dirty(handle, prev_bh); | |
747 | +out: | |
748 | + brelse(prev_bh); | |
749 | + brelse(new_first_bh); | |
750 | + return ret; | |
751 | +} | |
752 | + | |
753 | +static int ocfs2_read_xattr_bucket(struct inode *inode, | |
754 | + u64 blkno, | |
755 | + struct buffer_head **bhs, | |
756 | + int new) | |
757 | +{ | |
758 | + int ret = 0; | |
759 | + u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
760 | + | |
761 | + if (!new) | |
762 | + return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno, | |
763 | + blk_per_bucket, bhs, | |
764 | + OCFS2_BH_CACHED, inode); | |
765 | + | |
766 | + for (i = 0; i < blk_per_bucket; i++) { | |
767 | + bhs[i] = sb_getblk(inode->i_sb, blkno + i); | |
768 | + if (bhs[i] == NULL) { | |
769 | + ret = -EIO; | |
770 | + mlog_errno(ret); | |
771 | + break; | |
772 | + } | |
773 | + ocfs2_set_new_buffer_uptodate(inode, bhs[i]); | |
774 | + } | |
775 | + | |
776 | + return ret; | |
777 | +} | |
778 | + | |
779 | +/* | |
780 | + * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk). | |
781 | + * first_hash will record the 1st hash of the new bucket. | |
782 | + */ | |
783 | +static int ocfs2_half_xattr_bucket(struct inode *inode, | |
784 | + handle_t *handle, | |
785 | + u64 blk, | |
786 | + u64 new_blk, | |
787 | + u32 *first_hash, | |
788 | + int new_bucket_head) | |
789 | +{ | |
790 | + int ret, i; | |
791 | + u16 count, start, len, name_value_len, xe_len, name_offset; | |
792 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
793 | + struct buffer_head **s_bhs, **t_bhs = NULL; | |
794 | + struct ocfs2_xattr_header *xh; | |
795 | + struct ocfs2_xattr_entry *xe; | |
796 | + int blocksize = inode->i_sb->s_blocksize; | |
797 | + | |
798 | + mlog(0, "move half of xattrs from bucket %llu to %llu\n", | |
799 | + blk, new_blk); | |
800 | + | |
801 | + s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS); | |
802 | + if (!s_bhs) | |
803 | + return -ENOMEM; | |
804 | + | |
805 | + ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0); | |
806 | + if (ret) { | |
807 | + mlog_errno(ret); | |
808 | + goto out; | |
809 | + } | |
810 | + | |
811 | + ret = ocfs2_journal_access(handle, inode, s_bhs[0], | |
812 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
813 | + if (ret) { | |
814 | + mlog_errno(ret); | |
815 | + goto out; | |
816 | + } | |
817 | + | |
818 | + t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS); | |
819 | + if (!t_bhs) { | |
820 | + ret = -ENOMEM; | |
821 | + goto out; | |
822 | + } | |
823 | + | |
824 | + ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head); | |
825 | + if (ret) { | |
826 | + mlog_errno(ret); | |
827 | + goto out; | |
828 | + } | |
829 | + | |
830 | + for (i = 0; i < blk_per_bucket; i++) { | |
831 | + ret = ocfs2_journal_access(handle, inode, t_bhs[i], | |
832 | + OCFS2_JOURNAL_ACCESS_CREATE); | |
833 | + if (ret) { | |
834 | + mlog_errno(ret); | |
835 | + goto out; | |
836 | + } | |
837 | + } | |
838 | + | |
839 | + /* copy the whole bucket to the new first. */ | |
840 | + for (i = 0; i < blk_per_bucket; i++) | |
841 | + memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize); | |
842 | + | |
843 | + /* update the new bucket. */ | |
844 | + xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data; | |
845 | + count = le16_to_cpu(xh->xh_count); | |
846 | + start = count / 2; | |
847 | + | |
848 | + /* | |
849 | + * Calculate the total name/value len and xh_free_start for | |
850 | + * the old bucket first. | |
851 | + */ | |
852 | + name_offset = OCFS2_XATTR_BUCKET_SIZE; | |
853 | + name_value_len = 0; | |
854 | + for (i = 0; i < start; i++) { | |
855 | + xe = &xh->xh_entries[i]; | |
856 | + xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len); | |
857 | + if (ocfs2_xattr_is_local(xe)) | |
858 | + xe_len += | |
859 | + OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); | |
860 | + else | |
861 | + xe_len += OCFS2_XATTR_ROOT_SIZE; | |
862 | + name_value_len += xe_len; | |
863 | + if (le16_to_cpu(xe->xe_name_offset) < name_offset) | |
864 | + name_offset = le16_to_cpu(xe->xe_name_offset); | |
865 | + } | |
866 | + | |
867 | + /* | |
868 | + * Now begin the modification to the new bucket. | |
869 | + * | |
870 | + * In the new bucket, We just move the xattr entry to the beginning | |
871 | + * and don't touch the name/value. So there will be some holes in the | |
872 | + * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is | |
873 | + * called. | |
874 | + */ | |
875 | + xe = &xh->xh_entries[start]; | |
876 | + len = sizeof(struct ocfs2_xattr_entry) * (count - start); | |
877 | + mlog(0, "mv xattr entry len %d from %d to %d\n", len, | |
878 | + (char *)xe - (char *)xh, (char *)xh->xh_entries - (char *)xh); | |
879 | + memmove((char *)xh->xh_entries, (char *)xe, len); | |
880 | + xe = &xh->xh_entries[count - start]; | |
881 | + len = sizeof(struct ocfs2_xattr_entry) * start; | |
882 | + memset((char *)xe, 0, len); | |
883 | + | |
884 | + le16_add_cpu(&xh->xh_count, -start); | |
885 | + le16_add_cpu(&xh->xh_name_value_len, -name_value_len); | |
886 | + | |
887 | + /* Calculate xh_free_start for the new bucket. */ | |
888 | + xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE); | |
889 | + for (i = 0; i < le16_to_cpu(xh->xh_count); i++) { | |
890 | + xe = &xh->xh_entries[i]; | |
891 | + xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len); | |
892 | + if (ocfs2_xattr_is_local(xe)) | |
893 | + xe_len += | |
894 | + OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); | |
895 | + else | |
896 | + xe_len += OCFS2_XATTR_ROOT_SIZE; | |
897 | + if (le16_to_cpu(xe->xe_name_offset) < | |
898 | + le16_to_cpu(xh->xh_free_start)) | |
899 | + xh->xh_free_start = xe->xe_name_offset; | |
900 | + } | |
901 | + | |
902 | + /* set xh->xh_num_buckets for the new xh. */ | |
903 | + if (new_bucket_head) | |
904 | + xh->xh_num_buckets = cpu_to_le16(1); | |
905 | + else | |
906 | + xh->xh_num_buckets = 0; | |
907 | + | |
908 | + for (i = 0; i < blk_per_bucket; i++) { | |
909 | + ocfs2_journal_dirty(handle, t_bhs[i]); | |
910 | + if (ret) | |
911 | + mlog_errno(ret); | |
912 | + } | |
913 | + | |
914 | + /* store the first_hash of the new bucket. */ | |
915 | + if (first_hash) | |
916 | + *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash); | |
917 | + | |
918 | + /* | |
919 | + * Now only update the 1st block of the old bucket. | |
920 | + * Please note that the entry has been sorted already above. | |
921 | + */ | |
922 | + xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data; | |
923 | + memset(&xh->xh_entries[start], 0, | |
924 | + sizeof(struct ocfs2_xattr_entry) * (count - start)); | |
925 | + xh->xh_count = cpu_to_le16(start); | |
926 | + xh->xh_free_start = cpu_to_le16(name_offset); | |
927 | + xh->xh_name_value_len = cpu_to_le16(name_value_len); | |
928 | + | |
929 | + ocfs2_journal_dirty(handle, s_bhs[0]); | |
930 | + if (ret) | |
931 | + mlog_errno(ret); | |
932 | + | |
933 | +out: | |
934 | + if (s_bhs) { | |
935 | + for (i = 0; i < blk_per_bucket; i++) | |
936 | + brelse(s_bhs[i]); | |
937 | + } | |
938 | + kfree(s_bhs); | |
939 | + | |
940 | + if (t_bhs) { | |
941 | + for (i = 0; i < blk_per_bucket; i++) | |
942 | + brelse(t_bhs[i]); | |
943 | + } | |
944 | + kfree(t_bhs); | |
945 | + | |
946 | + return ret; | |
947 | +} | |
948 | + | |
949 | +/* | |
950 | + * Copy xattr from one bucket to another bucket. | |
951 | + * | |
952 | + * The caller must make sure that the journal transaction | |
953 | + * has enough space for journaling. | |
954 | + */ | |
955 | +static int ocfs2_cp_xattr_bucket(struct inode *inode, | |
956 | + handle_t *handle, | |
957 | + u64 s_blkno, | |
958 | + u64 t_blkno, | |
959 | + int t_is_new) | |
960 | +{ | |
961 | + int ret, i; | |
962 | + int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
963 | + int blocksize = inode->i_sb->s_blocksize; | |
964 | + struct buffer_head **s_bhs, **t_bhs = NULL; | |
965 | + | |
966 | + BUG_ON(s_blkno == t_blkno); | |
967 | + | |
968 | + mlog(0, "cp bucket %llu to %llu, target is %d\n", | |
969 | + s_blkno, t_blkno, t_is_new); | |
970 | + | |
971 | + s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket, | |
972 | + GFP_NOFS); | |
973 | + if (!s_bhs) | |
974 | + return -ENOMEM; | |
975 | + | |
976 | + ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0); | |
977 | + if (ret) | |
978 | + goto out; | |
979 | + | |
980 | + t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket, | |
981 | + GFP_NOFS); | |
982 | + if (!t_bhs) { | |
983 | + ret = -ENOMEM; | |
984 | + goto out; | |
985 | + } | |
986 | + | |
987 | + ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new); | |
988 | + if (ret) | |
989 | + goto out; | |
990 | + | |
991 | + for (i = 0; i < blk_per_bucket; i++) { | |
992 | + ret = ocfs2_journal_access(handle, inode, t_bhs[i], | |
993 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
994 | + if (ret) | |
995 | + goto out; | |
996 | + } | |
997 | + | |
998 | + for (i = 0; i < blk_per_bucket; i++) { | |
999 | + memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize); | |
1000 | + ocfs2_journal_dirty(handle, t_bhs[i]); | |
1001 | + } | |
1002 | + | |
1003 | +out: | |
1004 | + if (s_bhs) { | |
1005 | + for (i = 0; i < blk_per_bucket; i++) | |
1006 | + brelse(s_bhs[i]); | |
1007 | + } | |
1008 | + kfree(s_bhs); | |
1009 | + | |
1010 | + if (t_bhs) { | |
1011 | + for (i = 0; i < blk_per_bucket; i++) | |
1012 | + brelse(t_bhs[i]); | |
1013 | + } | |
1014 | + kfree(t_bhs); | |
1015 | + | |
1016 | + return ret; | |
1017 | +} | |
1018 | + | |
1019 | +/* | |
1020 | + * Copy one xattr cluster from src_blk to to_blk. | |
1021 | + * The to_blk will become the first bucket header of the cluster, so its | |
1022 | + * xh_num_buckets will be initialized as the bucket num in the cluster. | |
1023 | + */ | |
1024 | +static int ocfs2_cp_xattr_cluster(struct inode *inode, | |
1025 | + handle_t *handle, | |
1026 | + struct buffer_head *first_bh, | |
1027 | + u64 src_blk, | |
1028 | + u64 to_blk, | |
1029 | + u32 *first_hash) | |
1030 | +{ | |
1031 | + int i, ret, credits; | |
1032 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1033 | + int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); | |
1034 | + int num_buckets = ocfs2_xattr_buckets_per_cluster(osb); | |
1035 | + struct buffer_head *bh = NULL; | |
1036 | + struct ocfs2_xattr_header *xh; | |
1037 | + u64 to_blk_start = to_blk; | |
1038 | + | |
1039 | + mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk); | |
1040 | + | |
1041 | + /* | |
1042 | + * We need to update the new cluster and 1 more for the update of | |
1043 | + * the 1st bucket of the previous extent rec. | |
1044 | + */ | |
1045 | + credits = bpc + 1; | |
1046 | + ret = ocfs2_extend_trans(handle, credits); | |
1047 | + if (ret) { | |
1048 | + mlog_errno(ret); | |
1049 | + goto out; | |
1050 | + } | |
1051 | + | |
1052 | + ret = ocfs2_journal_access(handle, inode, first_bh, | |
1053 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1054 | + if (ret) { | |
1055 | + mlog_errno(ret); | |
1056 | + goto out; | |
1057 | + } | |
1058 | + | |
1059 | + for (i = 0; i < num_buckets; i++) { | |
1060 | + ret = ocfs2_cp_xattr_bucket(inode, handle, | |
1061 | + src_blk, to_blk, 1); | |
1062 | + if (ret) { | |
1063 | + mlog_errno(ret); | |
1064 | + goto out; | |
1065 | + } | |
1066 | + | |
1067 | + src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
1068 | + to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
1069 | + } | |
1070 | + | |
1071 | + /* update the old bucket header. */ | |
1072 | + xh = (struct ocfs2_xattr_header *)first_bh->b_data; | |
1073 | + le16_add_cpu(&xh->xh_num_buckets, -num_buckets); | |
1074 | + | |
1075 | + ocfs2_journal_dirty(handle, first_bh); | |
1076 | + | |
1077 | + /* update the new bucket header. */ | |
1078 | + ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode); | |
1079 | + if (ret < 0) { | |
1080 | + mlog_errno(ret); | |
1081 | + goto out; | |
1082 | + } | |
1083 | + | |
1084 | + ret = ocfs2_journal_access(handle, inode, bh, | |
1085 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1086 | + if (ret) { | |
1087 | + mlog_errno(ret); | |
1088 | + goto out; | |
1089 | + } | |
1090 | + | |
1091 | + xh = (struct ocfs2_xattr_header *)bh->b_data; | |
1092 | + xh->xh_num_buckets = cpu_to_le16(num_buckets); | |
1093 | + | |
1094 | + ocfs2_journal_dirty(handle, bh); | |
1095 | + | |
1096 | + if (first_hash) | |
1097 | + *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash); | |
1098 | +out: | |
1099 | + brelse(bh); | |
1100 | + return ret; | |
1101 | +} | |
1102 | + | |
1103 | +/* | |
1104 | + * Move half of the xattrs in this cluster to the new cluster. | |
1105 | + * This function should only be called when bucket size == cluster size. | |
1106 | + * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead. | |
1107 | + */ | |
1108 | +static int ocfs2_half_xattr_cluster(struct inode *inode, | |
1109 | + handle_t *handle, | |
1110 | + u64 prev_blk, | |
1111 | + u64 new_blk, | |
1112 | + u32 *first_hash) | |
1113 | +{ | |
1114 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
1115 | + int ret, credits = 2 * blk_per_bucket; | |
1116 | + | |
1117 | + BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize); | |
1118 | + | |
1119 | + ret = ocfs2_extend_trans(handle, credits); | |
1120 | + if (ret) { | |
1121 | + mlog_errno(ret); | |
1122 | + return ret; | |
1123 | + } | |
1124 | + | |
1125 | + /* Move half of the xattr in start_blk to the next bucket. */ | |
1126 | + return ocfs2_half_xattr_bucket(inode, handle, prev_blk, | |
1127 | + new_blk, first_hash, 1); | |
1128 | +} | |
1129 | + | |
1130 | +/* | |
1131 | + * Move some xattrs from the old cluster to the new one since they are not | |
1132 | + * contiguous in ocfs2 xattr tree. | |
1133 | + * | |
1134 | + * new_blk starts a new separate cluster, and we will move some xattrs from | |
1135 | + * prev_blk to it. v_start will be set as the first name hash value in this | |
1136 | + * new cluster so that it can be used as e_cpos during tree insertion and | |
1137 | + * don't collide with our original b-tree operations. first_bh and header_bh | |
1138 | + * will also be updated since they will be used in ocfs2_extend_xattr_bucket | |
1139 | + * to extend the insert bucket. | |
1140 | + * | |
1141 | + * The problem is how much xattr should we move to the new one and when should | |
1142 | + * we update first_bh and header_bh? | |
1143 | + * 1. If cluster size > bucket size, that means the previous cluster has more | |
1144 | + * than 1 bucket, so just move half nums of bucket into the new cluster and | |
1145 | + * update the first_bh and header_bh if the insert bucket has been moved | |
1146 | + * to the new cluster. | |
1147 | + * 2. If cluster_size == bucket_size: | |
1148 | + * a) If the previous extent rec has more than one cluster and the insert | |
1149 | + * place isn't in the last cluster, copy the entire last cluster to the | |
1150 | + * new one. This time, we don't need to upate the first_bh and header_bh | |
1151 | + * since they will not be moved into the new cluster. | |
1152 | + * b) Otherwise, move the bottom half of the xattrs in the last cluster into | |
1153 | + * the new one. And we set the extend flag to zero if the insert place is | |
1154 | + * moved into the new allocated cluster since no extend is needed. | |
1155 | + */ | |
1156 | +static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode, | |
1157 | + handle_t *handle, | |
1158 | + struct buffer_head **first_bh, | |
1159 | + struct buffer_head **header_bh, | |
1160 | + u64 new_blk, | |
1161 | + u64 prev_blk, | |
1162 | + u32 prev_clusters, | |
1163 | + u32 *v_start, | |
1164 | + int *extend) | |
1165 | +{ | |
1166 | + int ret = 0; | |
1167 | + int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); | |
1168 | + | |
1169 | + mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n", | |
1170 | + prev_blk, prev_clusters, new_blk); | |
1171 | + | |
1172 | + if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) | |
1173 | + ret = ocfs2_mv_xattr_bucket_cross_cluster(inode, | |
1174 | + handle, | |
1175 | + first_bh, | |
1176 | + header_bh, | |
1177 | + new_blk, | |
1178 | + prev_blk, | |
1179 | + prev_clusters, | |
1180 | + v_start); | |
1181 | + else { | |
1182 | + u64 last_blk = prev_blk + bpc * (prev_clusters - 1); | |
1183 | + | |
1184 | + if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk) | |
1185 | + ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh, | |
1186 | + last_blk, new_blk, | |
1187 | + v_start); | |
1188 | + else { | |
1189 | + ret = ocfs2_half_xattr_cluster(inode, handle, | |
1190 | + last_blk, new_blk, | |
1191 | + v_start); | |
1192 | + | |
1193 | + if ((*header_bh)->b_blocknr == last_blk && extend) | |
1194 | + *extend = 0; | |
1195 | + } | |
1196 | + } | |
1197 | + | |
1198 | + return ret; | |
1199 | +} | |
1200 | + | |
1201 | +/* | |
1202 | + * Add a new cluster for xattr storage. | |
1203 | + * | |
1204 | + * If the new cluster is contiguous with the previous one, it will be | |
1205 | + * appended to the same extent record, and num_clusters will be updated. | |
1206 | + * If not, we will insert a new extent for it and move some xattrs in | |
1207 | + * the last cluster into the new allocated one. | |
1208 | + * We also need to limit the maximum size of a btree leaf, otherwise we'll | |
1209 | + * lose the benefits of hashing because we'll have to search large leaves. | |
1210 | + * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize, | |
1211 | + * if it's bigger). | |
1212 | + * | |
1213 | + * first_bh is the first block of the previous extent rec and header_bh | |
1214 | + * indicates the bucket we will insert the new xattrs. They will be updated | |
1215 | + * when the header_bh is moved into the new cluster. | |
1216 | + */ | |
1217 | +static int ocfs2_add_new_xattr_cluster(struct inode *inode, | |
1218 | + struct buffer_head *root_bh, | |
1219 | + struct buffer_head **first_bh, | |
1220 | + struct buffer_head **header_bh, | |
1221 | + u32 *num_clusters, | |
1222 | + u32 prev_cpos, | |
1223 | + u64 prev_blkno, | |
1224 | + int *extend) | |
1225 | +{ | |
1226 | + int ret, credits; | |
1227 | + u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); | |
1228 | + u32 prev_clusters = *num_clusters; | |
1229 | + u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0; | |
1230 | + u64 block; | |
1231 | + handle_t *handle = NULL; | |
1232 | + struct ocfs2_alloc_context *data_ac = NULL; | |
1233 | + struct ocfs2_alloc_context *meta_ac = NULL; | |
1234 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1235 | + struct ocfs2_xattr_block *xb = | |
1236 | + (struct ocfs2_xattr_block *)root_bh->b_data; | |
1237 | + struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root; | |
1238 | + struct ocfs2_extent_list *root_el = &xb_root->xt_list; | |
1239 | + enum ocfs2_extent_tree_type type = OCFS2_XATTR_TREE_EXTENT; | |
1240 | + | |
1241 | + mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, " | |
1242 | + "previous xattr blkno = %llu\n", | |
1243 | + (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1244 | + prev_cpos, prev_blkno); | |
1245 | + | |
1246 | + ret = ocfs2_lock_allocators(inode, root_bh, root_el, | |
1247 | + clusters_to_add, 0, &data_ac, | |
1248 | + &meta_ac, type, NULL); | |
1249 | + if (ret) { | |
1250 | + mlog_errno(ret); | |
1251 | + goto leave; | |
1252 | + } | |
1253 | + | |
1254 | + credits = ocfs2_calc_extend_credits(osb->sb, root_el, clusters_to_add); | |
1255 | + handle = ocfs2_start_trans(osb, credits); | |
1256 | + if (IS_ERR(handle)) { | |
1257 | + ret = PTR_ERR(handle); | |
1258 | + handle = NULL; | |
1259 | + mlog_errno(ret); | |
1260 | + goto leave; | |
1261 | + } | |
1262 | + | |
1263 | + ret = ocfs2_journal_access(handle, inode, root_bh, | |
1264 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1265 | + if (ret < 0) { | |
1266 | + mlog_errno(ret); | |
1267 | + goto leave; | |
1268 | + } | |
1269 | + | |
1270 | + ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
1271 | + clusters_to_add, &bit_off, &num_bits); | |
1272 | + if (ret < 0) { | |
1273 | + if (ret != -ENOSPC) | |
1274 | + mlog_errno(ret); | |
1275 | + goto leave; | |
1276 | + } | |
1277 | + | |
1278 | + BUG_ON(num_bits > clusters_to_add); | |
1279 | + | |
1280 | + block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
1281 | + mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n", | |
1282 | + num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1283 | + | |
1284 | + if (prev_blkno + prev_clusters * bpc == block && | |
1285 | + (prev_clusters + num_bits) << osb->s_clustersize_bits <= | |
1286 | + OCFS2_MAX_XATTR_TREE_LEAF_SIZE) { | |
1287 | + /* | |
1288 | + * If this cluster is contiguous with the old one and | |
1289 | + * adding this new cluster, we don't surpass the limit of | |
1290 | + * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be | |
1291 | + * initialized and used like other buckets in the previous | |
1292 | + * cluster. | |
1293 | + * So add it as a contiguous one. The caller will handle | |
1294 | + * its init process. | |
1295 | + */ | |
1296 | + v_start = prev_cpos + prev_clusters; | |
1297 | + *num_clusters = prev_clusters + num_bits; | |
1298 | + mlog(0, "Add contiguous %u clusters to previous extent rec.\n", | |
1299 | + num_bits); | |
1300 | + } else { | |
1301 | + ret = ocfs2_adjust_xattr_cross_cluster(inode, | |
1302 | + handle, | |
1303 | + first_bh, | |
1304 | + header_bh, | |
1305 | + block, | |
1306 | + prev_blkno, | |
1307 | + prev_clusters, | |
1308 | + &v_start, | |
1309 | + extend); | |
1310 | + if (ret) { | |
1311 | + mlog_errno(ret); | |
1312 | + goto leave; | |
1313 | + } | |
1314 | + } | |
1315 | + | |
1316 | + mlog(0, "Insert %u clusters at block %llu for xattr at %u\n", | |
1317 | + num_bits, block, v_start); | |
1318 | + ret = ocfs2_xattr_tree_insert_extent(osb, handle, inode, root_bh, | |
1319 | + v_start, block, num_bits, | |
1320 | + 0, meta_ac); | |
1321 | + if (ret < 0) { | |
1322 | + mlog_errno(ret); | |
1323 | + goto leave; | |
1324 | + } | |
1325 | + | |
1326 | + ret = ocfs2_journal_dirty(handle, root_bh); | |
1327 | + if (ret < 0) { | |
1328 | + mlog_errno(ret); | |
1329 | + goto leave; | |
1330 | + } | |
1331 | + | |
1332 | +leave: | |
1333 | + if (handle) | |
1334 | + ocfs2_commit_trans(osb, handle); | |
1335 | + if (data_ac) | |
1336 | + ocfs2_free_alloc_context(data_ac); | |
1337 | + if (meta_ac) | |
1338 | + ocfs2_free_alloc_context(meta_ac); | |
1339 | + | |
1340 | + return ret; | |
1341 | +} | |
1342 | + | |
1343 | +/* | |
1344 | + * Extend a new xattr bucket and move xattrs to the end one by one until | |
1345 | + * We meet with start_bh. Only move half of the xattrs to the bucket after it. | |
1346 | + */ | |
1347 | +static int ocfs2_extend_xattr_bucket(struct inode *inode, | |
1348 | + struct buffer_head *first_bh, | |
1349 | + struct buffer_head *start_bh, | |
1350 | + u32 num_clusters) | |
1351 | +{ | |
1352 | + int ret, credits; | |
1353 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1354 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
1355 | + u64 start_blk = start_bh->b_blocknr, end_blk; | |
1356 | + u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb); | |
1357 | + handle_t *handle; | |
1358 | + struct ocfs2_xattr_header *first_xh = | |
1359 | + (struct ocfs2_xattr_header *)first_bh->b_data; | |
1360 | + u16 bucket = le16_to_cpu(first_xh->xh_num_buckets); | |
1361 | + | |
1362 | + mlog(0, "extend xattr bucket in %llu, xattr extend rec starting " | |
1363 | + "from %llu, len = %u\n", start_blk, | |
1364 | + (unsigned long long)first_bh->b_blocknr, num_clusters); | |
1365 | + | |
1366 | + BUG_ON(bucket >= num_buckets); | |
1367 | + | |
1368 | + end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket; | |
1369 | + | |
1370 | + /* | |
1371 | + * We will touch all the buckets after the start_bh(include it). | |
1372 | + * Add one more bucket and modify the first_bh. | |
1373 | + */ | |
1374 | + credits = end_blk - start_blk + 2 * blk_per_bucket + 1; | |
1375 | + handle = ocfs2_start_trans(osb, credits); | |
1376 | + if (IS_ERR(handle)) { | |
1377 | + ret = PTR_ERR(handle); | |
1378 | + handle = NULL; | |
1379 | + mlog_errno(ret); | |
1380 | + goto out; | |
1381 | + } | |
1382 | + | |
1383 | + ret = ocfs2_journal_access(handle, inode, first_bh, | |
1384 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1385 | + if (ret) { | |
1386 | + mlog_errno(ret); | |
1387 | + goto commit; | |
1388 | + } | |
1389 | + | |
1390 | + while (end_blk != start_blk) { | |
1391 | + ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk, | |
1392 | + end_blk + blk_per_bucket, 0); | |
1393 | + if (ret) | |
1394 | + goto commit; | |
1395 | + end_blk -= blk_per_bucket; | |
1396 | + } | |
1397 | + | |
1398 | + /* Move half of the xattr in start_blk to the next bucket. */ | |
1399 | + ret = ocfs2_half_xattr_bucket(inode, handle, start_blk, | |
1400 | + start_blk + blk_per_bucket, NULL, 0); | |
1401 | + | |
1402 | + le16_add_cpu(&first_xh->xh_num_buckets, 1); | |
1403 | + ocfs2_journal_dirty(handle, first_bh); | |
1404 | + | |
1405 | +commit: | |
1406 | + ocfs2_commit_trans(osb, handle); | |
1407 | +out: | |
1408 | + return ret; | |
1409 | +} | |
1410 | + | |
1411 | +/* | |
1412 | + * Add new xattr bucket in an extent record and adjust the buckets accordingly. | |
1413 | + * xb_bh is the ocfs2_xattr_block. | |
1414 | + * We will move all the buckets starting from header_bh to the next place. As | |
1415 | + * for this one, half num of its xattrs will be moved to the next one. | |
1416 | + * | |
1417 | + * We will allocate a new cluster if current cluster is full and adjust | |
1418 | + * header_bh and first_bh if the insert place is moved to the new cluster. | |
1419 | + */ | |
1420 | +static int ocfs2_add_new_xattr_bucket(struct inode *inode, | |
1421 | + struct buffer_head *xb_bh, | |
1422 | + struct buffer_head *header_bh) | |
1423 | +{ | |
1424 | + struct ocfs2_xattr_header *first_xh = NULL; | |
1425 | + struct buffer_head *first_bh = NULL; | |
1426 | + struct ocfs2_xattr_block *xb = | |
1427 | + (struct ocfs2_xattr_block *)xb_bh->b_data; | |
1428 | + struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root; | |
1429 | + struct ocfs2_extent_list *el = &xb_root->xt_list; | |
1430 | + struct ocfs2_xattr_header *xh = | |
1431 | + (struct ocfs2_xattr_header *)header_bh->b_data; | |
1432 | + u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash); | |
1433 | + struct super_block *sb = inode->i_sb; | |
1434 | + struct ocfs2_super *osb = OCFS2_SB(sb); | |
1435 | + int ret, num_buckets, extend = 1; | |
1436 | + u64 p_blkno; | |
1437 | + u32 e_cpos, num_clusters; | |
1438 | + | |
1439 | + mlog(0, "Add new xattr bucket starting form %llu\n", | |
1440 | + (unsigned long long)header_bh->b_blocknr); | |
1441 | + | |
1442 | + /* | |
1443 | + * Add refrence for header_bh here because it may be | |
1444 | + * changed in ocfs2_add_new_xattr_cluster and we need | |
1445 | + * to free it in the end. | |
1446 | + */ | |
1447 | + get_bh(header_bh); | |
1448 | + | |
1449 | + ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos, | |
1450 | + &num_clusters, el); | |
1451 | + if (ret) { | |
1452 | + mlog_errno(ret); | |
1453 | + goto out; | |
1454 | + } | |
1455 | + | |
1456 | + ret = ocfs2_read_block(osb, p_blkno, | |
1457 | + &first_bh, OCFS2_BH_CACHED, inode); | |
1458 | + if (ret) { | |
1459 | + mlog_errno(ret); | |
1460 | + goto out; | |
1461 | + } | |
1462 | + | |
1463 | + num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters; | |
1464 | + first_xh = (struct ocfs2_xattr_header *)first_bh->b_data; | |
1465 | + | |
1466 | + if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) { | |
1467 | + ret = ocfs2_add_new_xattr_cluster(inode, | |
1468 | + xb_bh, | |
1469 | + &first_bh, | |
1470 | + &header_bh, | |
1471 | + &num_clusters, | |
1472 | + e_cpos, | |
1473 | + p_blkno, | |
1474 | + &extend); | |
1475 | + if (ret) { | |
1476 | + mlog_errno(ret); | |
1477 | + goto out; | |
1478 | + } | |
1479 | + } | |
1480 | + | |
1481 | + if (extend) | |
1482 | + ret = ocfs2_extend_xattr_bucket(inode, | |
1483 | + first_bh, | |
1484 | + header_bh, | |
1485 | + num_clusters); | |
1486 | + if (ret) | |
1487 | + mlog_errno(ret); | |
1488 | +out: | |
1489 | + brelse(first_bh); | |
1490 | + brelse(header_bh); | |
1491 | + return ret; | |
1492 | +} | |
1493 | + | |
1494 | +static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode, | |
1495 | + struct ocfs2_xattr_bucket *bucket, | |
1496 | + int offs) | |
1497 | +{ | |
1498 | + int block_off = offs >> inode->i_sb->s_blocksize_bits; | |
1499 | + | |
1500 | + offs = offs % inode->i_sb->s_blocksize; | |
1501 | + return bucket->bhs[block_off]->b_data + offs; | |
1502 | +} | |
1503 | + | |
1504 | +/* | |
1505 | + * Handle the normal xattr set, including replace, delete and new. | |
1506 | + * When the bucket is empty, "is_empty" is set and the caller can | |
1507 | + * free this bucket. | |
1508 | + * | |
1509 | + * Note: "local" indicates the real data's locality. So we can't | |
1510 | + * just its bucket locality by its length. | |
1511 | + */ | |
1512 | +static void ocfs2_xattr_set_entry_normal(struct inode *inode, | |
1513 | + struct ocfs2_xattr_info *xi, | |
1514 | + struct ocfs2_xattr_search *xs, | |
1515 | + u32 name_hash, | |
1516 | + int local, | |
1517 | + int *is_empty) | |
1518 | +{ | |
1519 | + struct ocfs2_xattr_entry *last, *xe; | |
1520 | + int name_len = strlen(xi->name); | |
1521 | + struct ocfs2_xattr_header *xh = xs->header; | |
1522 | + u16 count = le16_to_cpu(xh->xh_count), start; | |
1523 | + size_t blocksize = inode->i_sb->s_blocksize; | |
1524 | + char *val; | |
1525 | + size_t offs, size, new_size; | |
1526 | + | |
1527 | + last = &xh->xh_entries[count]; | |
1528 | + if (!xs->not_found) { | |
1529 | + xe = xs->here; | |
1530 | + offs = le16_to_cpu(xe->xe_name_offset); | |
1531 | + if (ocfs2_xattr_is_local(xe)) | |
1532 | + size = OCFS2_XATTR_SIZE(name_len) + | |
1533 | + OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); | |
1534 | + else | |
1535 | + size = OCFS2_XATTR_SIZE(name_len) + | |
1536 | + OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE); | |
1537 | + | |
1538 | + /* | |
1539 | + * If the new value will be stored outside, xi->value has been | |
1540 | + * initalized as an empty ocfs2_xattr_value_root, and the same | |
1541 | + * goes with xi->value_len, so we can set new_size safely here. | |
1542 | + * See ocfs2_xattr_set_in_bucket. | |
1543 | + */ | |
1544 | + new_size = OCFS2_XATTR_SIZE(name_len) + | |
1545 | + OCFS2_XATTR_SIZE(xi->value_len); | |
1546 | + | |
1547 | + le16_add_cpu(&xh->xh_name_value_len, -size); | |
1548 | + if (xi->value) { | |
1549 | + if (new_size > size) | |
1550 | + goto set_new_name_value; | |
1551 | + | |
1552 | + /* Now replace the old value with new one. */ | |
1553 | + if (local) | |
1554 | + xe->xe_value_size = cpu_to_le64(xi->value_len); | |
1555 | + else | |
1556 | + xe->xe_value_size = 0; | |
1557 | + | |
1558 | + val = ocfs2_xattr_bucket_get_val(inode, | |
1559 | + &xs->bucket, offs); | |
1560 | + memset(val + OCFS2_XATTR_SIZE(name_len), 0, | |
1561 | + size - OCFS2_XATTR_SIZE(name_len)); | |
1562 | + if (OCFS2_XATTR_SIZE(xi->value_len) > 0) | |
1563 | + memcpy(val + OCFS2_XATTR_SIZE(name_len), | |
1564 | + xi->value, xi->value_len); | |
1565 | + | |
1566 | + le16_add_cpu(&xh->xh_name_value_len, new_size); | |
1567 | + ocfs2_xattr_set_local(xe, local); | |
1568 | + return; | |
1569 | + } else { | |
1570 | + /* Remove the old entry. */ | |
1571 | + last -= 1; | |
1572 | + memmove(xe, xe + 1, | |
1573 | + (void *)last - (void *)xe); | |
1574 | + memset(last, 0, sizeof(struct ocfs2_xattr_entry)); | |
1575 | + le16_add_cpu(&xh->xh_count, -1); | |
1576 | + if (xh->xh_count == 0 && is_empty) | |
1577 | + *is_empty = 1; | |
1578 | + return; | |
1579 | + } | |
1580 | + } else { | |
1581 | + /* find a new entry for insert. */ | |
1582 | + int low = 0, high = count - 1, tmp; | |
1583 | + struct ocfs2_xattr_entry *tmp_xe; | |
1584 | + | |
1585 | + while (low <= high) { | |
1586 | + tmp = (low + high) / 2; | |
1587 | + tmp_xe = &xh->xh_entries[tmp]; | |
1588 | + | |
1589 | + if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash)) | |
1590 | + low = tmp + 1; | |
1591 | + else if (name_hash < | |
1592 | + le32_to_cpu(tmp_xe->xe_name_hash)) | |
1593 | + high = tmp - 1; | |
1594 | + else | |
1595 | + break; | |
1596 | + } | |
1597 | + | |
1598 | + xe = &xh->xh_entries[low]; | |
1599 | + if (low != count) | |
1600 | + memmove(xe + 1, xe, (void *)last - (void *)xe); | |
1601 | + | |
1602 | + le16_add_cpu(&xh->xh_count, 1); | |
1603 | + memset(xe, 0, sizeof(struct ocfs2_xattr_entry)); | |
1604 | + xe->xe_name_hash = cpu_to_le32(name_hash); | |
1605 | + xe->xe_name_len = name_len; | |
1606 | + ocfs2_xattr_set_type(xe, xi->name_index); | |
1607 | + } | |
1608 | + | |
1609 | +set_new_name_value: | |
1610 | + /* Insert the new name+value. */ | |
1611 | + size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len); | |
1612 | + | |
1613 | + /* | |
1614 | + * We must make sure that the name/value pair | |
1615 | + * exists in the same block. | |
1616 | + */ | |
1617 | + offs = le16_to_cpu(xh->xh_free_start); | |
1618 | + start = offs - size; | |
1619 | + | |
1620 | + if (start >> inode->i_sb->s_blocksize_bits != | |
1621 | + (offs - 1) >> inode->i_sb->s_blocksize_bits) { | |
1622 | + offs = offs - offs % blocksize; | |
1623 | + xh->xh_free_start = cpu_to_le16(offs); | |
1624 | + } | |
1625 | + | |
1626 | + val = ocfs2_xattr_bucket_get_val(inode, | |
1627 | + &xs->bucket, offs - size); | |
1628 | + xe->xe_name_offset = cpu_to_le16(offs - size); | |
1629 | + | |
1630 | + memset(val, 0, size); | |
1631 | + memcpy(val, xi->name, name_len); | |
1632 | + memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len); | |
1633 | + | |
1634 | + xe->xe_value_size = cpu_to_le64(xi->value_len); | |
1635 | + ocfs2_xattr_set_local(xe, local); | |
1636 | + xs->here = xe; | |
1637 | + le16_add_cpu(&xh->xh_free_start, -size); | |
1638 | + le16_add_cpu(&xh->xh_name_value_len, size); | |
1639 | + | |
1640 | + return; | |
1641 | +} | |
1642 | + | |
1643 | +static int ocfs2_xattr_bucket_handle_journal(struct inode *inode, | |
1644 | + handle_t *handle, | |
1645 | + struct ocfs2_xattr_search *xs, | |
1646 | + struct buffer_head **bhs, | |
1647 | + u16 bh_num) | |
1648 | +{ | |
1649 | + int ret = 0, off, block_off; | |
1650 | + struct ocfs2_xattr_entry *xe = xs->here; | |
1651 | + | |
1652 | + /* | |
1653 | + * First calculate all the blocks we should journal_access | |
1654 | + * and journal_dirty. The first block should always be touched. | |
1655 | + */ | |
1656 | + ret = ocfs2_journal_dirty(handle, bhs[0]); | |
1657 | + if (ret) | |
1658 | + mlog_errno(ret); | |
1659 | + | |
1660 | + /* calc the data. */ | |
1661 | + off = le16_to_cpu(xe->xe_name_offset); | |
1662 | + block_off = off >> inode->i_sb->s_blocksize_bits; | |
1663 | + ret = ocfs2_journal_dirty(handle, bhs[block_off]); | |
1664 | + if (ret) | |
1665 | + mlog_errno(ret); | |
1666 | + | |
1667 | + return ret; | |
1668 | +} | |
1669 | + | |
1670 | +/* | |
1671 | + * Set the xattr entry in the specified bucket. | |
1672 | + * The bucket is indicated by xs->bucket and it should have the enough | |
1673 | + * space for the xattr insertion. | |
1674 | + */ | |
1675 | +static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode, | |
1676 | + struct ocfs2_xattr_info *xi, | |
1677 | + struct ocfs2_xattr_search *xs, | |
1678 | + u32 name_hash, | |
1679 | + int local, | |
1680 | + int *bucket_empty) | |
1681 | +{ | |
1682 | + int i, ret; | |
1683 | + handle_t *handle = NULL; | |
1684 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
1685 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1686 | + | |
1687 | + mlog(0, "Set xattr entry len = %d index = %d in bucket %llu\n", | |
1688 | + xi->value_len, xi->name_index, | |
1689 | + (unsigned long long)xs->bucket.bhs[0]->b_blocknr); | |
1690 | + | |
1691 | + if (!xs->bucket.bhs[1]) { | |
1692 | + ret = ocfs2_read_blocks(osb, | |
1693 | + xs->bucket.bhs[0]->b_blocknr + 1, | |
1694 | + blk_per_bucket - 1, &xs->bucket.bhs[1], | |
1695 | + OCFS2_BH_CACHED, inode); | |
1696 | + if (ret) { | |
1697 | + mlog_errno(ret); | |
1698 | + goto out; | |
1699 | + } | |
1700 | + } | |
1701 | + | |
1702 | + handle = ocfs2_start_trans(osb, blk_per_bucket); | |
1703 | + if (IS_ERR(handle)) { | |
1704 | + ret = PTR_ERR(handle); | |
1705 | + handle = NULL; | |
1706 | + mlog_errno(ret); | |
1707 | + goto out; | |
1708 | + } | |
1709 | + | |
1710 | + for (i = 0; i < blk_per_bucket; i++) { | |
1711 | + ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i], | |
1712 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1713 | + if (ret < 0) { | |
1714 | + mlog_errno(ret); | |
1715 | + goto out; | |
1716 | + } | |
1717 | + } | |
1718 | + | |
1719 | + ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, | |
1720 | + local, bucket_empty); | |
1721 | + | |
1722 | + /*Only dirty the blocks we have touched in set xattr. */ | |
1723 | + ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs, | |
1724 | + xs->bucket.bhs, blk_per_bucket); | |
1725 | + if (ret) | |
1726 | + mlog_errno(ret); | |
1727 | +out: | |
1728 | + ocfs2_commit_trans(osb, handle); | |
1729 | + | |
1730 | + return ret; | |
1731 | +} | |
1732 | + | |
1733 | +static int ocfs2_xattr_value_update_size(struct inode *inode, | |
1734 | + struct buffer_head *xe_bh, | |
1735 | + struct ocfs2_xattr_entry *xe, | |
1736 | + u64 new_size) | |
1737 | +{ | |
1738 | + int ret; | |
1739 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1740 | + handle_t *handle = NULL; | |
1741 | + | |
1742 | + handle = ocfs2_start_trans(osb, 1); | |
1743 | + if (handle == NULL) { | |
1744 | + ret = -ENOMEM; | |
1745 | + mlog_errno(ret); | |
1746 | + goto out; | |
1747 | + } | |
1748 | + | |
1749 | + ret = ocfs2_journal_access(handle, inode, xe_bh, | |
1750 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1751 | + if (ret < 0) { | |
1752 | + mlog_errno(ret); | |
1753 | + goto out_commit; | |
1754 | + } | |
1755 | + | |
1756 | + xe->xe_value_size = cpu_to_le64(new_size); | |
1757 | + | |
1758 | + ret = ocfs2_journal_dirty(handle, xe_bh); | |
1759 | + if (ret < 0) | |
1760 | + mlog_errno(ret); | |
1761 | + | |
1762 | +out_commit: | |
1763 | + ocfs2_commit_trans(osb, handle); | |
1764 | +out: | |
1765 | + return ret; | |
1766 | +} | |
1767 | + | |
1768 | +/* | |
1769 | + * Truncate the specified xe_off entry in xattr bucket. | |
1770 | + * bucket is indicated by header_bh and len is the new length. | |
1771 | + * Both the ocfs2_xattr_value_root and the entry will be updated here. | |
1772 | + * | |
1773 | + * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed. | |
1774 | + */ | |
1775 | +static int ocfs2_xattr_bucket_value_truncate(struct inode *inode, | |
1776 | + struct buffer_head *header_bh, | |
1777 | + int xe_off, | |
1778 | + int len) | |
1779 | +{ | |
1780 | + int ret, offset; | |
1781 | + u64 value_blk; | |
1782 | + struct buffer_head *value_bh = NULL; | |
1783 | + struct ocfs2_xattr_value_root *xv; | |
1784 | + struct ocfs2_xattr_entry *xe; | |
1785 | + struct ocfs2_xattr_header *xh = | |
1786 | + (struct ocfs2_xattr_header *)header_bh->b_data; | |
1787 | + size_t blocksize = inode->i_sb->s_blocksize; | |
1788 | + | |
1789 | + xe = &xh->xh_entries[xe_off]; | |
1790 | + | |
1791 | + BUG_ON(!xe || ocfs2_xattr_is_local(xe)); | |
1792 | + | |
1793 | + offset = le16_to_cpu(xe->xe_name_offset) + | |
1794 | + OCFS2_XATTR_SIZE(xe->xe_name_len); | |
1795 | + | |
1796 | + value_blk = offset / blocksize; | |
1797 | + | |
1798 | + /* We don't allow ocfs2_xattr_value to be stored in different block. */ | |
1799 | + BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize); | |
1800 | + value_blk += header_bh->b_blocknr; | |
1801 | + | |
1802 | + ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk, | |
1803 | + &value_bh, OCFS2_BH_CACHED, inode); | |
1804 | + if (ret) { | |
1805 | + mlog_errno(ret); | |
1806 | + goto out; | |
1807 | + } | |
1808 | + | |
1809 | + xv = (struct ocfs2_xattr_value_root *) | |
1810 | + (value_bh->b_data + offset % blocksize); | |
1811 | + | |
1812 | + mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n", | |
1813 | + xe_off, (unsigned long long)header_bh->b_blocknr, len); | |
1814 | + ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len); | |
1815 | + if (ret) { | |
1816 | + mlog_errno(ret); | |
1817 | + goto out; | |
1818 | + } | |
1819 | + | |
1820 | + ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len); | |
1821 | + if (ret) { | |
1822 | + mlog_errno(ret); | |
1823 | + goto out; | |
1824 | + } | |
1825 | + | |
1826 | +out: | |
1827 | + brelse(value_bh); | |
1828 | + return ret; | |
1829 | +} | |
1830 | + | |
1831 | +static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode, | |
1832 | + struct ocfs2_xattr_search *xs, | |
1833 | + int len) | |
1834 | +{ | |
1835 | + int ret, offset; | |
1836 | + struct ocfs2_xattr_entry *xe = xs->here; | |
1837 | + struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base; | |
1838 | + | |
1839 | + BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe)); | |
1840 | + | |
1841 | + offset = xe - xh->xh_entries; | |
1842 | + ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0], | |
1843 | + offset, len); | |
1844 | + if (ret) | |
1845 | + mlog_errno(ret); | |
1846 | + | |
1847 | + return ret; | |
1848 | +} | |
1849 | + | |
1850 | +static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode, | |
1851 | + struct ocfs2_xattr_search *xs, | |
1852 | + char *val, | |
1853 | + int value_len) | |
1854 | +{ | |
1855 | + int offset; | |
1856 | + struct ocfs2_xattr_value_root *xv; | |
1857 | + struct ocfs2_xattr_entry *xe = xs->here; | |
1858 | + | |
1859 | + BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe)); | |
1860 | + | |
1861 | + offset = le16_to_cpu(xe->xe_name_offset) + | |
1862 | + OCFS2_XATTR_SIZE(xe->xe_name_len); | |
1863 | + | |
1864 | + xv = (struct ocfs2_xattr_value_root *)(xs->base + offset); | |
1865 | + | |
1866 | + return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len); | |
1867 | +} | |
1868 | + | |
1869 | +/* | |
1870 | + * Remove the xattr bucket pointed by bucket_bh. | |
1871 | + * All the buckets after it in the same xattr extent rec will be | |
1872 | + * move forward one by one. | |
1873 | + */ | |
1874 | +static int ocfs2_rm_xattr_bucket(struct inode *inode, | |
1875 | + struct buffer_head *first_bh, | |
1876 | + struct ocfs2_xattr_bucket *bucket) | |
1877 | +{ | |
1878 | + int ret = 0, credits; | |
1879 | + struct ocfs2_xattr_header *xh = | |
1880 | + (struct ocfs2_xattr_header *)first_bh->b_data; | |
1881 | + u16 bucket_num = le16_to_cpu(xh->xh_num_buckets); | |
1882 | + u64 end, start = bucket->bhs[0]->b_blocknr; | |
1883 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1884 | + handle_t *handle; | |
1885 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
1886 | + | |
1887 | + end = first_bh->b_blocknr + (bucket_num - 1) * blk_per_bucket; | |
1888 | + | |
1889 | + mlog(0, "rm xattr bucket %llu\n", start); | |
1890 | + /* | |
1891 | + * We need to update the first xattr_header and all the buckets starting | |
1892 | + * from start in this xattr rec. | |
1893 | + * | |
1894 | + * XXX: Should we empty the old last bucket here? | |
1895 | + */ | |
1896 | + credits = 1 + end - start; | |
1897 | + handle = ocfs2_start_trans(osb, credits); | |
1898 | + if (IS_ERR(handle)) { | |
1899 | + ret = PTR_ERR(handle); | |
1900 | + mlog_errno(ret); | |
1901 | + return ret; | |
1902 | + } | |
1903 | + | |
1904 | + ret = ocfs2_journal_access(handle, inode, first_bh, | |
1905 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1906 | + if (ret) { | |
1907 | + mlog_errno(ret); | |
1908 | + goto out_commit; | |
1909 | + } | |
1910 | + | |
1911 | + | |
1912 | + while (start < end) { | |
1913 | + ret = ocfs2_cp_xattr_bucket(inode, handle, | |
1914 | + start + blk_per_bucket, | |
1915 | + start, 0); | |
1916 | + if (ret) { | |
1917 | + mlog_errno(ret); | |
1918 | + goto out_commit; | |
1919 | + } | |
1920 | + start += blk_per_bucket; | |
1921 | + } | |
1922 | + | |
1923 | + /* update the first_bh. */ | |
1924 | + xh->xh_num_buckets = cpu_to_le16(bucket_num - 1); | |
1925 | + ocfs2_journal_dirty(handle, first_bh); | |
1926 | + | |
1927 | +out_commit: | |
1928 | + ocfs2_commit_trans(osb, handle); | |
1929 | + return ret; | |
1930 | +} | |
1931 | + | |
1932 | +static int ocfs2_rm_xattr_cluster(struct inode *inode, | |
1933 | + struct buffer_head *root_bh, | |
1934 | + u64 blkno, | |
1935 | + u32 cpos, | |
1936 | + u32 len) | |
1937 | +{ | |
1938 | + int ret; | |
1939 | + struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1940 | + struct inode *tl_inode = osb->osb_tl_inode; | |
1941 | + handle_t *handle; | |
1942 | + struct ocfs2_xattr_block *xb = | |
1943 | + (struct ocfs2_xattr_block *)root_bh->b_data; | |
1944 | + struct ocfs2_extent_list *root_el = &xb->xb_attrs.xb_root.xt_list; | |
1945 | + struct ocfs2_alloc_context *meta_ac = NULL; | |
1946 | + struct ocfs2_cached_dealloc_ctxt dealloc; | |
1947 | + | |
1948 | + ocfs2_init_dealloc_ctxt(&dealloc); | |
1949 | + | |
1950 | + mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n", | |
1951 | + cpos, len, (unsigned long long)blkno); | |
1952 | + | |
1953 | + ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len); | |
1954 | + | |
1955 | + ret = ocfs2_lock_allocators(inode, root_bh, root_el, | |
1956 | + 0, 1, NULL, &meta_ac, | |
1957 | + OCFS2_XATTR_TREE_EXTENT, NULL); | |
1958 | + if (ret) { | |
1959 | + mlog_errno(ret); | |
1960 | + return ret; | |
1961 | + } | |
1962 | + | |
1963 | + mutex_lock(&tl_inode->i_mutex); | |
1964 | + | |
1965 | + if (ocfs2_truncate_log_needs_flush(osb)) { | |
1966 | + ret = __ocfs2_flush_truncate_log(osb); | |
1967 | + if (ret < 0) { | |
1968 | + mlog_errno(ret); | |
1969 | + goto out; | |
1970 | + } | |
1971 | + } | |
1972 | + | |
1973 | + handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); | |
1974 | + if (handle == NULL) { | |
1975 | + ret = -ENOMEM; | |
1976 | + mlog_errno(ret); | |
1977 | + goto out; | |
1978 | + } | |
1979 | + | |
1980 | + ret = ocfs2_journal_access(handle, inode, root_bh, | |
1981 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
1982 | + if (ret) { | |
1983 | + mlog_errno(ret); | |
1984 | + goto out_commit; | |
1985 | + } | |
1986 | + | |
1987 | + ret = ocfs2_remove_extent(inode, root_bh, cpos, len, handle, meta_ac, | |
1988 | + &dealloc, OCFS2_XATTR_TREE_EXTENT, NULL); | |
1989 | + if (ret) { | |
1990 | + mlog_errno(ret); | |
1991 | + goto out_commit; | |
1992 | + } | |
1993 | + | |
1994 | + le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len); | |
1995 | + | |
1996 | + ret = ocfs2_journal_dirty(handle, root_bh); | |
1997 | + if (ret) { | |
1998 | + mlog_errno(ret); | |
1999 | + goto out_commit; | |
2000 | + } | |
2001 | + | |
2002 | + ret = ocfs2_truncate_log_append(osb, handle, blkno, len); | |
2003 | + if (ret) | |
2004 | + mlog_errno(ret); | |
2005 | + | |
2006 | +out_commit: | |
2007 | + ocfs2_commit_trans(osb, handle); | |
2008 | +out: | |
2009 | + ocfs2_schedule_truncate_log_flush(osb, 1); | |
2010 | + | |
2011 | + mutex_unlock(&tl_inode->i_mutex); | |
2012 | + | |
2013 | + if (meta_ac) | |
2014 | + ocfs2_free_alloc_context(meta_ac); | |
2015 | + | |
2016 | + ocfs2_run_deallocs(osb, &dealloc); | |
2017 | + | |
2018 | + return ret; | |
2019 | +} | |
2020 | + | |
2021 | +/* | |
2022 | + * Free the xattr bucket indicated by xs->bucket and if all the buckets | |
2023 | + * in the clusters is free, free the clusters also. | |
2024 | + */ | |
2025 | +static int ocfs2_xattr_bucket_shrink(struct inode *inode, | |
2026 | + struct ocfs2_xattr_info *xi, | |
2027 | + struct ocfs2_xattr_search *xs, | |
2028 | + u32 name_hash) | |
2029 | +{ | |
2030 | + int ret; | |
2031 | + u32 e_cpos, num_clusters; | |
2032 | + u64 p_blkno; | |
2033 | + struct buffer_head *first_bh = NULL; | |
2034 | + struct ocfs2_xattr_header *first_xh; | |
2035 | + struct ocfs2_xattr_block *xb = | |
2036 | + (struct ocfs2_xattr_block *)xs->xattr_bh->b_data; | |
2037 | + | |
2038 | + BUG_ON(xs->header->xh_count != 0); | |
2039 | + | |
2040 | + ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, | |
2041 | + &e_cpos, &num_clusters, | |
2042 | + &xb->xb_attrs.xb_root.xt_list); | |
2043 | + if (ret) { | |
2044 | + mlog_errno(ret); | |
2045 | + return ret; | |
2046 | + } | |
2047 | + | |
2048 | + ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, | |
2049 | + &first_bh, OCFS2_BH_CACHED, inode); | |
2050 | + if (ret) { | |
2051 | + mlog_errno(ret); | |
2052 | + return ret; | |
2053 | + } | |
2054 | + | |
2055 | + ret = ocfs2_rm_xattr_bucket(inode, first_bh, &xs->bucket); | |
2056 | + if (ret) { | |
2057 | + mlog_errno(ret); | |
2058 | + goto out; | |
2059 | + } | |
2060 | + | |
2061 | + first_xh = (struct ocfs2_xattr_header *)first_bh->b_data; | |
2062 | + if (first_xh->xh_num_buckets == 0) | |
2063 | + ret = ocfs2_rm_xattr_cluster(inode, xs->xattr_bh, | |
2064 | + p_blkno, e_cpos, | |
2065 | + num_clusters); | |
2066 | + | |
2067 | +out: | |
2068 | + brelse(first_bh); | |
2069 | + return ret; | |
2070 | +} | |
2071 | + | |
2072 | +static void ocfs2_xattr_bucket_remove_xs(struct inode *inode, | |
2073 | + struct ocfs2_xattr_search *xs) | |
2074 | +{ | |
2075 | + handle_t *handle = NULL; | |
2076 | + struct ocfs2_xattr_header *xh = xs->bucket.xh; | |
2077 | + struct ocfs2_xattr_entry *last = &xh->xh_entries[ | |
2078 | + le16_to_cpu(xh->xh_count) - 1]; | |
2079 | + int ret = 0; | |
2080 | + | |
2081 | + handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1); | |
2082 | + if (IS_ERR(handle)) { | |
2083 | + ret = PTR_ERR(handle); | |
2084 | + mlog_errno(ret); | |
2085 | + return; | |
2086 | + } | |
2087 | + | |
2088 | + ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0], | |
2089 | + OCFS2_JOURNAL_ACCESS_WRITE); | |
2090 | + if (ret) { | |
2091 | + mlog_errno(ret); | |
2092 | + goto out_commit; | |
2093 | + } | |
2094 | + | |
2095 | + /* Remove the old entry. */ | |
2096 | + memmove(xs->here, xs->here + 1, | |
2097 | + (void *)last - (void *)xs->here); | |
2098 | + memset(last, 0, sizeof(struct ocfs2_xattr_entry)); | |
2099 | + le16_add_cpu(&xh->xh_count, -1); | |
2100 | + | |
2101 | + ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]); | |
2102 | + if (ret < 0) | |
2103 | + mlog_errno(ret); | |
2104 | +out_commit: | |
2105 | + ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); | |
2106 | +} | |
2107 | + | |
2108 | +/* | |
2109 | + * Set the xattr name/value in the bucket specified in xs. | |
2110 | + * | |
2111 | + * As the new value in xi may be stored in the bucket or in an outside cluster, | |
2112 | + * we divide the whole process into 3 steps: | |
2113 | + * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket) | |
2114 | + * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs) | |
2115 | + * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside) | |
2116 | + * 4. If the clusters for the new outside value can't be allocated, we need | |
2117 | + * to free the xattr we allocated in set. | |
2118 | + */ | |
2119 | +static int ocfs2_xattr_set_in_bucket(struct inode *inode, | |
2120 | + struct ocfs2_xattr_info *xi, | |
2121 | + struct ocfs2_xattr_search *xs) | |
2122 | +{ | |
2123 | + int ret, local = 1, bucket_empty = 0; | |
2124 | + size_t value_len; | |
2125 | + char *val = (char *)xi->value; | |
2126 | + struct ocfs2_xattr_entry *xe = xs->here; | |
2127 | + u32 name_hash = ocfs2_xattr_hash_by_name(inode, | |
2128 | + xi->name_index, xi->name); | |
2129 | + | |
2130 | + if (!xs->not_found && !ocfs2_xattr_is_local(xe)) { | |
2131 | + /* | |
2132 | + * We need to truncate the xattr storage first. | |
2133 | + * | |
2134 | + * If both the old and new value are stored to | |
2135 | + * outside block, we only need to truncate | |
2136 | + * the storage and then set the value outside. | |
2137 | + * | |
2138 | + * If the new value should be stored within block, | |
2139 | + * we should free all the outside block first and | |
2140 | + * the modification to the xattr block will be done | |
2141 | + * by following steps. | |
2142 | + */ | |
2143 | + if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) | |
2144 | + value_len = xi->value_len; | |
2145 | + else | |
2146 | + value_len = 0; | |
2147 | + | |
2148 | + ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs, | |
2149 | + value_len); | |
2150 | + if (ret) | |
2151 | + goto out; | |
2152 | + | |
2153 | + if (value_len) | |
2154 | + goto set_value_outside; | |
2155 | + } | |
2156 | + | |
2157 | + value_len = xi->value_len; | |
2158 | + /* So we have to handle the inside block change now. */ | |
2159 | + if (value_len > OCFS2_XATTR_INLINE_SIZE) { | |
2160 | + /* | |
2161 | + * If the new value will be stored outside of block, | |
2162 | + * initalize a new empty value root and insert it first. | |
2163 | + */ | |
2164 | + local = 0; | |
2165 | + xi->value = &def_xv; | |
2166 | + xi->value_len = OCFS2_XATTR_ROOT_SIZE; | |
2167 | + } | |
2168 | + | |
2169 | + ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, | |
2170 | + local, &bucket_empty); | |
2171 | + if (ret) { | |
2172 | + mlog_errno(ret); | |
2173 | + goto out; | |
2174 | + } | |
2175 | + | |
2176 | + if (value_len > OCFS2_XATTR_INLINE_SIZE) { | |
2177 | + /* allocate the space now for the outside block storage. */ | |
2178 | + ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs, | |
2179 | + value_len); | |
2180 | + if (ret) { | |
2181 | + mlog_errno(ret); | |
2182 | + | |
2183 | + if (xs->not_found) { | |
2184 | + /* | |
2185 | + * We can't allocate enough clusters for outside | |
2186 | + * storage and we have allocated xattr already, | |
2187 | + * so need to remove it. | |
2188 | + */ | |
2189 | + ocfs2_xattr_bucket_remove_xs(inode, xs); | |
2190 | + } | |
2191 | + goto out; | |
2192 | + } | |
2193 | + } else { | |
2194 | + if (bucket_empty) | |
2195 | + ret = ocfs2_xattr_bucket_shrink(inode, xi, | |
2196 | + xs, name_hash); | |
2197 | + goto out; | |
2198 | + } | |
2199 | + | |
2200 | +set_value_outside: | |
2201 | + ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len); | |
2202 | +out: | |
2203 | + return ret; | |
2204 | +} | |
2205 | + | |
2206 | +/* check whether the xattr bucket is filled up with the same hash value. */ | |
2207 | +static int ocfs2_check_xattr_bucket_collision(struct inode *inode, | |
2208 | + struct ocfs2_xattr_bucket *bucket) | |
2209 | +{ | |
2210 | + struct ocfs2_xattr_header *xh = bucket->xh; | |
2211 | + | |
2212 | + if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash == | |
2213 | + xh->xh_entries[0].xe_name_hash) { | |
2214 | + mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, " | |
2215 | + "hash = %u\n", | |
2216 | + (unsigned long long)bucket->bhs[0]->b_blocknr, | |
2217 | + le32_to_cpu(xh->xh_entries[0].xe_name_hash)); | |
2218 | + return -ENOSPC; | |
2219 | + } | |
2220 | + | |
2221 | + return 0; | |
2222 | +} | |
2223 | + | |
2224 | +static int ocfs2_xattr_set_entry_index_block(struct inode *inode, | |
2225 | + struct ocfs2_xattr_info *xi, | |
2226 | + struct ocfs2_xattr_search *xs) | |
2227 | +{ | |
2228 | + struct ocfs2_xattr_header *xh; | |
2229 | + struct ocfs2_xattr_entry *xe; | |
2230 | + u16 count, header_size, xh_free_start; | |
2231 | + int i, free, max_free, need, old; | |
2232 | + size_t value_size = 0, name_len = strlen(xi->name); | |
2233 | + size_t blocksize = inode->i_sb->s_blocksize; | |
2234 | + int ret, allocation = 0; | |
2235 | + u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); | |
2236 | + | |
2237 | + mlog_entry("Set xattr %s in xattr index block\n", xi->name); | |
2238 | + | |
2239 | +try_again: | |
2240 | + xh = xs->header; | |
2241 | + count = le16_to_cpu(xh->xh_count); | |
2242 | + xh_free_start = le16_to_cpu(xh->xh_free_start); | |
2243 | + header_size = sizeof(struct ocfs2_xattr_header) + | |
2244 | + count * sizeof(struct ocfs2_xattr_entry); | |
2245 | + max_free = OCFS2_XATTR_BUCKET_SIZE - | |
2246 | + le16_to_cpu(xh->xh_name_value_len) - header_size; | |
2247 | + | |
2248 | + mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size " | |
2249 | + "of %u which exceed block size\n", | |
2250 | + (unsigned long long)xs->bucket.bhs[0]->b_blocknr, | |
2251 | + header_size); | |
2252 | + | |
2253 | + if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) | |
2254 | + value_size = OCFS2_XATTR_ROOT_SIZE; | |
2255 | + else if (xi->value) | |
2256 | + value_size = OCFS2_XATTR_SIZE(xi->value_len); | |
2257 | + | |
2258 | + if (xs->not_found) | |
2259 | + need = sizeof(struct ocfs2_xattr_entry) + | |
2260 | + OCFS2_XATTR_SIZE(name_len) + value_size; | |
2261 | + else { | |
2262 | + need = value_size + OCFS2_XATTR_SIZE(name_len); | |
2263 | + | |
2264 | + /* | |
2265 | + * We only replace the old value if the new length is smaller | |
2266 | + * than the old one. Otherwise we will allocate new space in the | |
2267 | + * bucket to store it. | |
2268 | + */ | |
2269 | + xe = xs->here; | |
2270 | + if (ocfs2_xattr_is_local(xe)) | |
2271 | + old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); | |
2272 | + else | |
2273 | + old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE); | |
2274 | + | |
2275 | + if (old >= value_size) | |
2276 | + need = 0; | |
2277 | + } | |
2278 | + | |
2279 | + free = xh_free_start - header_size; | |
2280 | + /* | |
2281 | + * We need to make sure the new name/value pair | |
2282 | + * can exist in the same block. | |
2283 | + */ | |
2284 | + if (xh_free_start % blocksize < need) | |
2285 | + free -= xh_free_start % blocksize; | |
2286 | + | |
2287 | + mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, " | |
2288 | + "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len =" | |
2289 | + " %u\n", xs->not_found, | |
2290 | + (unsigned long long)xs->bucket.bhs[0]->b_blocknr, | |
2291 | + free, need, max_free, le16_to_cpu(xh->xh_free_start), | |
2292 | + le16_to_cpu(xh->xh_name_value_len)); | |
2293 | + | |
2294 | + if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) { | |
2295 | + if (need <= max_free && | |
2296 | + count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) { | |
2297 | + /* | |
2298 | + * We can create the space by defragment. Since only the | |
2299 | + * name/value will be moved, the xe shouldn't be changed | |
2300 | + * in xs. | |
2301 | + */ | |
2302 | + ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket); | |
2303 | + if (ret) { | |
2304 | + mlog_errno(ret); | |
2305 | + goto out; | |
2306 | + } | |
2307 | + | |
2308 | + xh_free_start = le16_to_cpu(xh->xh_free_start); | |
2309 | + free = xh_free_start - header_size; | |
2310 | + if (xh_free_start % blocksize < need) | |
2311 | + free -= xh_free_start % blocksize; | |
2312 | + | |
2313 | + if (free >= need) | |
2314 | + goto xattr_set; | |
2315 | + | |
2316 | + mlog(0, "Can't get enough space for xattr insert by " | |
2317 | + "defragment. Need %u bytes, but we have %d, so " | |
2318 | + "allocate new bucket for it.\n", need, free); | |
2319 | + } | |
2320 | + | |
2321 | + /* | |
2322 | + * We have to add new buckets or clusters and one | |
2323 | + * allocation should leave us enough space for insert. | |
2324 | + */ | |
2325 | + BUG_ON(allocation); | |
2326 | + | |
2327 | + /* | |
2328 | + * We do not allow for overlapping ranges between buckets. And | |
2329 | + * the maximum number of collisions we will allow for then is | |
2330 | + * one bucket's worth, so check it here whether we need to | |
2331 | + * add a new bucket for the insert. | |
2332 | + */ | |
2333 | + ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket); | |
2334 | + if (ret) { | |
2335 | + mlog_errno(ret); | |
2336 | + goto out; | |
2337 | + } | |
2338 | + | |
2339 | + ret = ocfs2_add_new_xattr_bucket(inode, | |
2340 | + xs->xattr_bh, | |
2341 | + xs->bucket.bhs[0]); | |
2342 | + if (ret) { | |
2343 | + mlog_errno(ret); | |
2344 | + goto out; | |
2345 | + } | |
2346 | + | |
2347 | + for (i = 0; i < blk_per_bucket; i++) | |
2348 | + brelse(xs->bucket.bhs[i]); | |
2349 | + | |
2350 | + memset(&xs->bucket, 0, sizeof(xs->bucket)); | |
2351 | + | |
2352 | + ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh, | |
2353 | + xi->name_index, | |
2354 | + xi->name, xs); | |
2355 | + if (ret && ret != -ENODATA) | |
2356 | + goto out; | |
2357 | + xs->not_found = ret; | |
2358 | + allocation = 1; | |
2359 | + goto try_again; | |
2360 | + } | |
2361 | + | |
2362 | +xattr_set: | |
2363 | + ret = ocfs2_xattr_set_in_bucket(inode, xi, xs); | |
2364 | +out: | |
2365 | + mlog_exit(ret); | |
2366 | + return ret; | |
2367 | +} | |
2368 | diff --git a/fs/ocfs2/xattr.h b/fs/ocfs2/xattr.h | |
2369 | index a69c8aa..d33dbe5 100644 | |
2370 | --- a/fs/ocfs2/xattr.h | |
2371 | +++ b/fs/ocfs2/xattr.h | |
2372 | @@ -64,4 +64,12 @@ static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb) | |
2373 | { | |
2374 | return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits); | |
2375 | } | |
2376 | + | |
2377 | +static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb) | |
2378 | +{ | |
2379 | + u16 len = sb->s_blocksize - | |
2380 | + offsetof(struct ocfs2_xattr_header, xh_entries); | |
2381 | + | |
2382 | + return len / sizeof(struct ocfs2_xattr_entry); | |
2383 | +} | |
2384 | #endif /* OCFS2_XATTR_H */ | |
2385 | -- | |
2386 | 1.5.4.5 | |
2387 |