Commit | Line | Data |
---|---|---|
028ba5df TY |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * move_extents.c | |
5 | * | |
6 | * Copyright (C) 2011 Oracle. All rights reserved. | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public | |
10 | * License version 2 as published by the Free Software Foundation. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | * General Public License for more details. | |
16 | */ | |
17 | #include <linux/fs.h> | |
18 | #include <linux/types.h> | |
19 | #include <linux/mount.h> | |
20 | #include <linux/swap.h> | |
21 | ||
22 | #include <cluster/masklog.h> | |
23 | ||
24 | #include "ocfs2.h" | |
25 | #include "ocfs2_ioctl.h" | |
26 | ||
27 | #include "alloc.h" | |
28 | #include "aops.h" | |
29 | #include "dlmglue.h" | |
30 | #include "extent_map.h" | |
31 | #include "inode.h" | |
32 | #include "journal.h" | |
33 | #include "suballoc.h" | |
34 | #include "uptodate.h" | |
35 | #include "super.h" | |
36 | #include "dir.h" | |
37 | #include "buffer_head_io.h" | |
38 | #include "sysfile.h" | |
028ba5df TY |
39 | #include "refcounttree.h" |
40 | #include "move_extents.h" | |
41 | ||
42 | struct ocfs2_move_extents_context { | |
43 | struct inode *inode; | |
44 | struct file *file; | |
45 | int auto_defrag; | |
4dfa66bd | 46 | int partial; |
028ba5df TY |
47 | int credits; |
48 | u32 new_phys_cpos; | |
49 | u32 clusters_moved; | |
50 | u64 refcount_loc; | |
51 | struct ocfs2_move_extents *range; | |
52 | struct ocfs2_extent_tree et; | |
53 | struct ocfs2_alloc_context *meta_ac; | |
54 | struct ocfs2_alloc_context *data_ac; | |
55 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
56 | }; | |
de474ee8 | 57 | |
8f603e56 TY |
58 | static int __ocfs2_move_extent(handle_t *handle, |
59 | struct ocfs2_move_extents_context *context, | |
60 | u32 cpos, u32 len, u32 p_cpos, u32 new_p_cpos, | |
61 | int ext_flags) | |
62 | { | |
63 | int ret = 0, index; | |
64 | struct inode *inode = context->inode; | |
65 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
66 | struct ocfs2_extent_rec *rec, replace_rec; | |
67 | struct ocfs2_path *path = NULL; | |
68 | struct ocfs2_extent_list *el; | |
69 | u64 ino = ocfs2_metadata_cache_owner(context->et.et_ci); | |
70 | u64 old_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cpos); | |
71 | ||
c7dd3392 | 72 | ret = ocfs2_duplicate_clusters_by_page(handle, inode, cpos, |
8f603e56 TY |
73 | p_cpos, new_p_cpos, len); |
74 | if (ret) { | |
75 | mlog_errno(ret); | |
76 | goto out; | |
77 | } | |
78 | ||
79 | memset(&replace_rec, 0, sizeof(replace_rec)); | |
80 | replace_rec.e_cpos = cpu_to_le32(cpos); | |
81 | replace_rec.e_leaf_clusters = cpu_to_le16(len); | |
82 | replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(inode->i_sb, | |
83 | new_p_cpos)); | |
84 | ||
85 | path = ocfs2_new_path_from_et(&context->et); | |
86 | if (!path) { | |
87 | ret = -ENOMEM; | |
88 | mlog_errno(ret); | |
89 | goto out; | |
90 | } | |
91 | ||
92 | ret = ocfs2_find_path(INODE_CACHE(inode), path, cpos); | |
93 | if (ret) { | |
94 | mlog_errno(ret); | |
95 | goto out; | |
96 | } | |
97 | ||
98 | el = path_leaf_el(path); | |
99 | ||
100 | index = ocfs2_search_extent_list(el, cpos); | |
981035b4 | 101 | if (index == -1) { |
17a5b9ab | 102 | ret = ocfs2_error(inode->i_sb, |
8f603e56 TY |
103 | "Inode %llu has an extent at cpos %u which can no " |
104 | "longer be found.\n", | |
105 | (unsigned long long)ino, cpos); | |
8f603e56 TY |
106 | goto out; |
107 | } | |
108 | ||
109 | rec = &el->l_recs[index]; | |
110 | ||
111 | BUG_ON(ext_flags != rec->e_flags); | |
112 | /* | |
113 | * after moving/defraging to new location, the extent is not going | |
114 | * to be refcounted anymore. | |
115 | */ | |
116 | replace_rec.e_flags = ext_flags & ~OCFS2_EXT_REFCOUNTED; | |
117 | ||
118 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), | |
119 | context->et.et_root_bh, | |
120 | OCFS2_JOURNAL_ACCESS_WRITE); | |
121 | if (ret) { | |
122 | mlog_errno(ret); | |
123 | goto out; | |
124 | } | |
125 | ||
126 | ret = ocfs2_split_extent(handle, &context->et, path, index, | |
127 | &replace_rec, context->meta_ac, | |
128 | &context->dealloc); | |
129 | if (ret) { | |
130 | mlog_errno(ret); | |
131 | goto out; | |
132 | } | |
133 | ||
134 | ocfs2_journal_dirty(handle, context->et.et_root_bh); | |
135 | ||
136 | context->new_phys_cpos = new_p_cpos; | |
137 | ||
138 | /* | |
139 | * need I to append truncate log for old clusters? | |
140 | */ | |
141 | if (old_blkno) { | |
142 | if (ext_flags & OCFS2_EXT_REFCOUNTED) | |
143 | ret = ocfs2_decrease_refcount(inode, handle, | |
144 | ocfs2_blocks_to_clusters(osb->sb, | |
145 | old_blkno), | |
146 | len, context->meta_ac, | |
147 | &context->dealloc, 1); | |
148 | else | |
149 | ret = ocfs2_truncate_log_append(osb, handle, | |
150 | old_blkno, len); | |
151 | } | |
152 | ||
6fdb702d | 153 | ocfs2_update_inode_fsync_trans(handle, inode, 0); |
8f603e56 | 154 | out: |
4704aa30 | 155 | ocfs2_free_path(path); |
8f603e56 TY |
156 | return ret; |
157 | } | |
158 | ||
de474ee8 TY |
159 | /* |
160 | * lock allocators, and reserving appropriate number of bits for | |
161 | * meta blocks and data clusters. | |
162 | * | |
163 | * in some cases, we don't need to reserve clusters, just let data_ac | |
164 | * be NULL. | |
165 | */ | |
166 | static int ocfs2_lock_allocators_move_extents(struct inode *inode, | |
167 | struct ocfs2_extent_tree *et, | |
168 | u32 clusters_to_move, | |
169 | u32 extents_to_split, | |
170 | struct ocfs2_alloc_context **meta_ac, | |
171 | struct ocfs2_alloc_context **data_ac, | |
172 | int extra_blocks, | |
173 | int *credits) | |
174 | { | |
175 | int ret, num_free_extents; | |
176 | unsigned int max_recs_needed = 2 * extents_to_split + clusters_to_move; | |
177 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
178 | ||
179 | num_free_extents = ocfs2_num_free_extents(osb, et); | |
180 | if (num_free_extents < 0) { | |
181 | ret = num_free_extents; | |
182 | mlog_errno(ret); | |
183 | goto out; | |
184 | } | |
185 | ||
186 | if (!num_free_extents || | |
187 | (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) | |
188 | extra_blocks += ocfs2_extend_meta_needed(et->et_root_el); | |
189 | ||
190 | ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, meta_ac); | |
191 | if (ret) { | |
192 | mlog_errno(ret); | |
193 | goto out; | |
194 | } | |
195 | ||
196 | if (data_ac) { | |
197 | ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac); | |
198 | if (ret) { | |
199 | mlog_errno(ret); | |
200 | goto out; | |
201 | } | |
202 | } | |
203 | ||
06f9da6e | 204 | *credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el); |
de474ee8 TY |
205 | |
206 | mlog(0, "reserve metadata_blocks: %d, data_clusters: %u, credits: %d\n", | |
207 | extra_blocks, clusters_to_move, *credits); | |
208 | out: | |
209 | if (ret) { | |
210 | if (*meta_ac) { | |
211 | ocfs2_free_alloc_context(*meta_ac); | |
212 | *meta_ac = NULL; | |
213 | } | |
214 | } | |
215 | ||
216 | return ret; | |
217 | } | |
202ee5fa TY |
218 | |
219 | /* | |
220 | * Using one journal handle to guarantee the data consistency in case | |
221 | * crash happens anywhere. | |
dda54e76 TY |
222 | * |
223 | * XXX: defrag can end up with finishing partial extent as requested, | |
224 | * due to not enough contiguous clusters can be found in allocator. | |
202ee5fa TY |
225 | */ |
226 | static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context, | |
4dfa66bd | 227 | u32 cpos, u32 phys_cpos, u32 *len, int ext_flags) |
202ee5fa | 228 | { |
4dfa66bd | 229 | int ret, credits = 0, extra_blocks = 0, partial = context->partial; |
202ee5fa TY |
230 | handle_t *handle; |
231 | struct inode *inode = context->inode; | |
232 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
233 | struct inode *tl_inode = osb->osb_tl_inode; | |
234 | struct ocfs2_refcount_tree *ref_tree = NULL; | |
235 | u32 new_phys_cpos, new_len; | |
236 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
237 | ||
4dfa66bd | 238 | if ((ext_flags & OCFS2_EXT_REFCOUNTED) && *len) { |
202ee5fa TY |
239 | |
240 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & | |
241 | OCFS2_HAS_REFCOUNT_FL)); | |
242 | ||
243 | BUG_ON(!context->refcount_loc); | |
244 | ||
245 | ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1, | |
246 | &ref_tree, NULL); | |
247 | if (ret) { | |
248 | mlog_errno(ret); | |
249 | return ret; | |
250 | } | |
251 | ||
252 | ret = ocfs2_prepare_refcount_change_for_del(inode, | |
253 | context->refcount_loc, | |
254 | phys_blkno, | |
4dfa66bd | 255 | *len, |
202ee5fa TY |
256 | &credits, |
257 | &extra_blocks); | |
258 | if (ret) { | |
259 | mlog_errno(ret); | |
260 | goto out; | |
261 | } | |
262 | } | |
263 | ||
4dfa66bd | 264 | ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1, |
202ee5fa TY |
265 | &context->meta_ac, |
266 | &context->data_ac, | |
267 | extra_blocks, &credits); | |
268 | if (ret) { | |
269 | mlog_errno(ret); | |
270 | goto out; | |
271 | } | |
272 | ||
273 | /* | |
274 | * should be using allocation reservation strategy there? | |
275 | * | |
276 | * if (context->data_ac) | |
277 | * context->data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
278 | */ | |
279 | ||
280 | mutex_lock(&tl_inode->i_mutex); | |
281 | ||
282 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
283 | ret = __ocfs2_flush_truncate_log(osb); | |
284 | if (ret < 0) { | |
285 | mlog_errno(ret); | |
286 | goto out_unlock_mutex; | |
287 | } | |
288 | } | |
289 | ||
290 | handle = ocfs2_start_trans(osb, credits); | |
291 | if (IS_ERR(handle)) { | |
292 | ret = PTR_ERR(handle); | |
293 | mlog_errno(ret); | |
294 | goto out_unlock_mutex; | |
295 | } | |
296 | ||
4dfa66bd | 297 | ret = __ocfs2_claim_clusters(handle, context->data_ac, 1, *len, |
202ee5fa TY |
298 | &new_phys_cpos, &new_len); |
299 | if (ret) { | |
300 | mlog_errno(ret); | |
301 | goto out_commit; | |
302 | } | |
303 | ||
304 | /* | |
4dfa66bd TY |
305 | * allowing partial extent moving is kind of 'pros and cons', it makes |
306 | * whole defragmentation less likely to fail, on the contrary, the bad | |
307 | * thing is it may make the fs even more fragmented after moving, let | |
308 | * userspace make a good decision here. | |
202ee5fa | 309 | */ |
4dfa66bd TY |
310 | if (new_len != *len) { |
311 | mlog(0, "len_claimed: %u, len: %u\n", new_len, *len); | |
312 | if (!partial) { | |
313 | context->range->me_flags &= ~OCFS2_MOVE_EXT_FL_COMPLETE; | |
314 | ret = -ENOSPC; | |
315 | goto out_commit; | |
316 | } | |
202ee5fa TY |
317 | } |
318 | ||
319 | mlog(0, "cpos: %u, phys_cpos: %u, new_phys_cpos: %u\n", cpos, | |
320 | phys_cpos, new_phys_cpos); | |
321 | ||
4dfa66bd | 322 | ret = __ocfs2_move_extent(handle, context, cpos, new_len, phys_cpos, |
202ee5fa TY |
323 | new_phys_cpos, ext_flags); |
324 | if (ret) | |
325 | mlog_errno(ret); | |
326 | ||
4dfa66bd TY |
327 | if (partial && (new_len != *len)) |
328 | *len = new_len; | |
329 | ||
202ee5fa TY |
330 | /* |
331 | * Here we should write the new page out first if we are | |
332 | * in write-back mode. | |
333 | */ | |
4dfa66bd | 334 | ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, *len); |
202ee5fa TY |
335 | if (ret) |
336 | mlog_errno(ret); | |
337 | ||
338 | out_commit: | |
339 | ocfs2_commit_trans(osb, handle); | |
340 | ||
341 | out_unlock_mutex: | |
342 | mutex_unlock(&tl_inode->i_mutex); | |
343 | ||
344 | if (context->data_ac) { | |
345 | ocfs2_free_alloc_context(context->data_ac); | |
346 | context->data_ac = NULL; | |
347 | } | |
348 | ||
349 | if (context->meta_ac) { | |
350 | ocfs2_free_alloc_context(context->meta_ac); | |
351 | context->meta_ac = NULL; | |
352 | } | |
353 | ||
354 | out: | |
355 | if (ref_tree) | |
356 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
357 | ||
358 | return ret; | |
359 | } | |
1c06b912 TY |
360 | |
361 | /* | |
362 | * find the victim alloc group, where #blkno fits. | |
363 | */ | |
364 | static int ocfs2_find_victim_alloc_group(struct inode *inode, | |
365 | u64 vict_blkno, | |
366 | int type, int slot, | |
367 | int *vict_bit, | |
368 | struct buffer_head **ret_bh) | |
369 | { | |
6aea6f50 | 370 | int ret, i, bits_per_unit = 0; |
1c06b912 TY |
371 | u64 blkno; |
372 | char namebuf[40]; | |
373 | ||
374 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
375 | struct buffer_head *ac_bh = NULL, *gd_bh = NULL; | |
376 | struct ocfs2_chain_list *cl; | |
377 | struct ocfs2_chain_rec *rec; | |
378 | struct ocfs2_dinode *ac_dinode; | |
379 | struct ocfs2_group_desc *bg; | |
380 | ||
381 | ocfs2_sprintf_system_inode_name(namebuf, sizeof(namebuf), type, slot); | |
382 | ret = ocfs2_lookup_ino_from_name(osb->sys_root_inode, namebuf, | |
383 | strlen(namebuf), &blkno); | |
384 | if (ret) { | |
385 | ret = -ENOENT; | |
386 | goto out; | |
387 | } | |
388 | ||
389 | ret = ocfs2_read_blocks_sync(osb, blkno, 1, &ac_bh); | |
390 | if (ret) { | |
391 | mlog_errno(ret); | |
392 | goto out; | |
393 | } | |
394 | ||
395 | ac_dinode = (struct ocfs2_dinode *)ac_bh->b_data; | |
396 | cl = &(ac_dinode->id2.i_chain); | |
397 | rec = &(cl->cl_recs[0]); | |
398 | ||
399 | if (type == GLOBAL_BITMAP_SYSTEM_INODE) | |
6aea6f50 TY |
400 | bits_per_unit = osb->s_clustersize_bits - |
401 | inode->i_sb->s_blocksize_bits; | |
1c06b912 TY |
402 | /* |
403 | * 'vict_blkno' was out of the valid range. | |
404 | */ | |
405 | if ((vict_blkno < le64_to_cpu(rec->c_blkno)) || | |
7fa05c6e | 406 | (vict_blkno >= ((u64)le32_to_cpu(ac_dinode->id1.bitmap1.i_total) << |
6aea6f50 | 407 | bits_per_unit))) { |
1c06b912 TY |
408 | ret = -EINVAL; |
409 | goto out; | |
410 | } | |
411 | ||
412 | for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i++) { | |
413 | ||
414 | rec = &(cl->cl_recs[i]); | |
415 | if (!rec) | |
416 | continue; | |
417 | ||
418 | bg = NULL; | |
419 | ||
420 | do { | |
421 | if (!bg) | |
422 | blkno = le64_to_cpu(rec->c_blkno); | |
423 | else | |
424 | blkno = le64_to_cpu(bg->bg_next_group); | |
425 | ||
426 | if (gd_bh) { | |
427 | brelse(gd_bh); | |
428 | gd_bh = NULL; | |
429 | } | |
430 | ||
431 | ret = ocfs2_read_blocks_sync(osb, blkno, 1, &gd_bh); | |
432 | if (ret) { | |
433 | mlog_errno(ret); | |
434 | goto out; | |
435 | } | |
436 | ||
437 | bg = (struct ocfs2_group_desc *)gd_bh->b_data; | |
438 | ||
439 | if (vict_blkno < (le64_to_cpu(bg->bg_blkno) + | |
440 | le16_to_cpu(bg->bg_bits))) { | |
441 | ||
442 | *ret_bh = gd_bh; | |
6aea6f50 TY |
443 | *vict_bit = (vict_blkno - blkno) >> |
444 | bits_per_unit; | |
1c06b912 TY |
445 | mlog(0, "find the victim group: #%llu, " |
446 | "total_bits: %u, vict_bit: %u\n", | |
447 | blkno, le16_to_cpu(bg->bg_bits), | |
448 | *vict_bit); | |
449 | goto out; | |
450 | } | |
451 | ||
452 | } while (le64_to_cpu(bg->bg_next_group)); | |
453 | } | |
454 | ||
455 | ret = -EINVAL; | |
456 | out: | |
457 | brelse(ac_bh); | |
458 | ||
459 | /* | |
460 | * caller has to release the gd_bh properly. | |
461 | */ | |
462 | return ret; | |
463 | } | |
99e4c750 TY |
464 | |
465 | /* | |
466 | * XXX: helper to validate and adjust moving goal. | |
467 | */ | |
468 | static int ocfs2_validate_and_adjust_move_goal(struct inode *inode, | |
469 | struct ocfs2_move_extents *range) | |
470 | { | |
471 | int ret, goal_bit = 0; | |
472 | ||
473 | struct buffer_head *gd_bh = NULL; | |
7f4804d4 | 474 | struct ocfs2_group_desc *bg; |
99e4c750 TY |
475 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
476 | int c_to_b = 1 << (osb->s_clustersize_bits - | |
477 | inode->i_sb->s_blocksize_bits); | |
478 | ||
ea5e1675 TY |
479 | /* |
480 | * make goal become cluster aligned. | |
481 | */ | |
482 | range->me_goal = ocfs2_block_to_cluster_start(inode->i_sb, | |
483 | range->me_goal); | |
99e4c750 TY |
484 | /* |
485 | * validate goal sits within global_bitmap, and return the victim | |
486 | * group desc | |
487 | */ | |
488 | ret = ocfs2_find_victim_alloc_group(inode, range->me_goal, | |
489 | GLOBAL_BITMAP_SYSTEM_INODE, | |
490 | OCFS2_INVALID_SLOT, | |
491 | &goal_bit, &gd_bh); | |
492 | if (ret) | |
493 | goto out; | |
494 | ||
495 | bg = (struct ocfs2_group_desc *)gd_bh->b_data; | |
496 | ||
7f4804d4 DC |
497 | /* |
498 | * moving goal is not allowd to start with a group desc blok(#0 blk) | |
499 | * let's compromise to the latter cluster. | |
500 | */ | |
501 | if (range->me_goal == le64_to_cpu(bg->bg_blkno)) | |
502 | range->me_goal += c_to_b; | |
503 | ||
99e4c750 TY |
504 | /* |
505 | * movement is not gonna cross two groups. | |
506 | */ | |
507 | if ((le16_to_cpu(bg->bg_bits) - goal_bit) * osb->s_clustersize < | |
508 | range->me_len) { | |
509 | ret = -EINVAL; | |
510 | goto out; | |
511 | } | |
512 | /* | |
513 | * more exact validations/adjustments will be performed later during | |
514 | * moving operation for each extent range. | |
515 | */ | |
516 | mlog(0, "extents get ready to be moved to #%llu block\n", | |
517 | range->me_goal); | |
518 | ||
519 | out: | |
520 | brelse(gd_bh); | |
521 | ||
522 | return ret; | |
523 | } | |
e6b5859c TY |
524 | |
525 | static void ocfs2_probe_alloc_group(struct inode *inode, struct buffer_head *bh, | |
526 | int *goal_bit, u32 move_len, u32 max_hop, | |
527 | u32 *phys_cpos) | |
528 | { | |
529 | int i, used, last_free_bits = 0, base_bit = *goal_bit; | |
530 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data; | |
531 | u32 base_cpos = ocfs2_blocks_to_clusters(inode->i_sb, | |
532 | le64_to_cpu(gd->bg_blkno)); | |
533 | ||
534 | for (i = base_bit; i < le16_to_cpu(gd->bg_bits); i++) { | |
535 | ||
536 | used = ocfs2_test_bit(i, (unsigned long *)gd->bg_bitmap); | |
537 | if (used) { | |
538 | /* | |
539 | * we even tried searching the free chunk by jumping | |
540 | * a 'max_hop' distance, but still failed. | |
541 | */ | |
542 | if ((i - base_bit) > max_hop) { | |
543 | *phys_cpos = 0; | |
544 | break; | |
545 | } | |
546 | ||
547 | if (last_free_bits) | |
548 | last_free_bits = 0; | |
549 | ||
550 | continue; | |
551 | } else | |
552 | last_free_bits++; | |
553 | ||
554 | if (last_free_bits == move_len) { | |
555 | *goal_bit = i; | |
556 | *phys_cpos = base_cpos + i; | |
557 | break; | |
558 | } | |
559 | } | |
560 | ||
561 | mlog(0, "found phys_cpos: %u to fit the wanted moving.\n", *phys_cpos); | |
562 | } | |
8473aa8a | 563 | |
e0847717 TY |
564 | static int ocfs2_move_extent(struct ocfs2_move_extents_context *context, |
565 | u32 cpos, u32 phys_cpos, u32 *new_phys_cpos, | |
566 | u32 len, int ext_flags) | |
567 | { | |
568 | int ret, credits = 0, extra_blocks = 0, goal_bit = 0; | |
569 | handle_t *handle; | |
570 | struct inode *inode = context->inode; | |
571 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
572 | struct inode *tl_inode = osb->osb_tl_inode; | |
573 | struct inode *gb_inode = NULL; | |
574 | struct buffer_head *gb_bh = NULL; | |
575 | struct buffer_head *gd_bh = NULL; | |
576 | struct ocfs2_group_desc *gd; | |
577 | struct ocfs2_refcount_tree *ref_tree = NULL; | |
578 | u32 move_max_hop = ocfs2_blocks_to_clusters(inode->i_sb, | |
579 | context->range->me_threshold); | |
580 | u64 phys_blkno, new_phys_blkno; | |
581 | ||
582 | phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
583 | ||
584 | if ((ext_flags & OCFS2_EXT_REFCOUNTED) && len) { | |
585 | ||
586 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & | |
587 | OCFS2_HAS_REFCOUNT_FL)); | |
588 | ||
589 | BUG_ON(!context->refcount_loc); | |
590 | ||
591 | ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1, | |
592 | &ref_tree, NULL); | |
593 | if (ret) { | |
594 | mlog_errno(ret); | |
595 | return ret; | |
596 | } | |
597 | ||
598 | ret = ocfs2_prepare_refcount_change_for_del(inode, | |
599 | context->refcount_loc, | |
600 | phys_blkno, | |
601 | len, | |
602 | &credits, | |
603 | &extra_blocks); | |
604 | if (ret) { | |
605 | mlog_errno(ret); | |
606 | goto out; | |
607 | } | |
608 | } | |
609 | ||
610 | ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1, | |
611 | &context->meta_ac, | |
612 | NULL, extra_blocks, &credits); | |
613 | if (ret) { | |
614 | mlog_errno(ret); | |
615 | goto out; | |
616 | } | |
617 | ||
618 | /* | |
619 | * need to count 2 extra credits for global_bitmap inode and | |
620 | * group descriptor. | |
621 | */ | |
622 | credits += OCFS2_INODE_UPDATE_CREDITS + 1; | |
623 | ||
624 | /* | |
625 | * ocfs2_move_extent() didn't reserve any clusters in lock_allocators() | |
626 | * logic, while we still need to lock the global_bitmap. | |
627 | */ | |
628 | gb_inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE, | |
629 | OCFS2_INVALID_SLOT); | |
630 | if (!gb_inode) { | |
631 | mlog(ML_ERROR, "unable to get global_bitmap inode\n"); | |
632 | ret = -EIO; | |
633 | goto out; | |
634 | } | |
635 | ||
636 | mutex_lock(&gb_inode->i_mutex); | |
637 | ||
638 | ret = ocfs2_inode_lock(gb_inode, &gb_bh, 1); | |
639 | if (ret) { | |
640 | mlog_errno(ret); | |
641 | goto out_unlock_gb_mutex; | |
642 | } | |
643 | ||
644 | mutex_lock(&tl_inode->i_mutex); | |
645 | ||
646 | handle = ocfs2_start_trans(osb, credits); | |
647 | if (IS_ERR(handle)) { | |
648 | ret = PTR_ERR(handle); | |
649 | mlog_errno(ret); | |
650 | goto out_unlock_tl_inode; | |
651 | } | |
652 | ||
653 | new_phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *new_phys_cpos); | |
654 | ret = ocfs2_find_victim_alloc_group(inode, new_phys_blkno, | |
655 | GLOBAL_BITMAP_SYSTEM_INODE, | |
656 | OCFS2_INVALID_SLOT, | |
657 | &goal_bit, &gd_bh); | |
658 | if (ret) { | |
659 | mlog_errno(ret); | |
660 | goto out_commit; | |
661 | } | |
662 | ||
663 | /* | |
664 | * probe the victim cluster group to find a proper | |
665 | * region to fit wanted movement, it even will perfrom | |
666 | * a best-effort attempt by compromising to a threshold | |
667 | * around the goal. | |
668 | */ | |
669 | ocfs2_probe_alloc_group(inode, gd_bh, &goal_bit, len, move_max_hop, | |
670 | new_phys_cpos); | |
3d75be7c | 671 | if (!*new_phys_cpos) { |
e0847717 TY |
672 | ret = -ENOSPC; |
673 | goto out_commit; | |
674 | } | |
675 | ||
676 | ret = __ocfs2_move_extent(handle, context, cpos, len, phys_cpos, | |
677 | *new_phys_cpos, ext_flags); | |
678 | if (ret) { | |
679 | mlog_errno(ret); | |
680 | goto out_commit; | |
681 | } | |
682 | ||
683 | gd = (struct ocfs2_group_desc *)gd_bh->b_data; | |
684 | ret = ocfs2_alloc_dinode_update_counts(gb_inode, handle, gb_bh, len, | |
685 | le16_to_cpu(gd->bg_chain)); | |
686 | if (ret) { | |
687 | mlog_errno(ret); | |
688 | goto out_commit; | |
689 | } | |
690 | ||
691 | ret = ocfs2_block_group_set_bits(handle, gb_inode, gd, gd_bh, | |
692 | goal_bit, len); | |
db66c715 YL |
693 | if (ret) { |
694 | ocfs2_rollback_alloc_dinode_counts(gb_inode, gb_bh, len, | |
695 | le16_to_cpu(gd->bg_chain)); | |
e0847717 | 696 | mlog_errno(ret); |
db66c715 | 697 | } |
e0847717 TY |
698 | |
699 | /* | |
700 | * Here we should write the new page out first if we are | |
701 | * in write-back mode. | |
702 | */ | |
703 | ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, len); | |
704 | if (ret) | |
705 | mlog_errno(ret); | |
706 | ||
707 | out_commit: | |
708 | ocfs2_commit_trans(osb, handle); | |
709 | brelse(gd_bh); | |
710 | ||
711 | out_unlock_tl_inode: | |
712 | mutex_unlock(&tl_inode->i_mutex); | |
713 | ||
714 | ocfs2_inode_unlock(gb_inode, 1); | |
715 | out_unlock_gb_mutex: | |
716 | mutex_unlock(&gb_inode->i_mutex); | |
717 | brelse(gb_bh); | |
718 | iput(gb_inode); | |
719 | ||
720 | out: | |
721 | if (context->meta_ac) { | |
722 | ocfs2_free_alloc_context(context->meta_ac); | |
723 | context->meta_ac = NULL; | |
724 | } | |
725 | ||
726 | if (ref_tree) | |
727 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
728 | ||
729 | return ret; | |
730 | } | |
ee16cc03 TY |
731 | |
732 | /* | |
733 | * Helper to calculate the defraging length in one run according to threshold. | |
734 | */ | |
735 | static void ocfs2_calc_extent_defrag_len(u32 *alloc_size, u32 *len_defraged, | |
736 | u32 threshold, int *skip) | |
737 | { | |
738 | if ((*alloc_size + *len_defraged) < threshold) { | |
739 | /* | |
740 | * proceed defragmentation until we meet the thresh | |
741 | */ | |
742 | *len_defraged += *alloc_size; | |
743 | } else if (*len_defraged == 0) { | |
744 | /* | |
745 | * XXX: skip a large extent. | |
746 | */ | |
747 | *skip = 1; | |
748 | } else { | |
749 | /* | |
750 | * split this extent to coalesce with former pieces as | |
751 | * to reach the threshold. | |
752 | * | |
753 | * we're done here with one cycle of defragmentation | |
754 | * in a size of 'thresh', resetting 'len_defraged' | |
755 | * forces a new defragmentation. | |
756 | */ | |
757 | *alloc_size = threshold - *len_defraged; | |
758 | *len_defraged = 0; | |
759 | } | |
760 | } | |
53069d4e TY |
761 | |
762 | static int __ocfs2_move_extents_range(struct buffer_head *di_bh, | |
763 | struct ocfs2_move_extents_context *context) | |
764 | { | |
765 | int ret = 0, flags, do_defrag, skip = 0; | |
766 | u32 cpos, phys_cpos, move_start, len_to_move, alloc_size; | |
767 | u32 len_defraged = 0, defrag_thresh = 0, new_phys_cpos = 0; | |
768 | ||
769 | struct inode *inode = context->inode; | |
770 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
771 | struct ocfs2_move_extents *range = context->range; | |
772 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
773 | ||
f17c20dd | 774 | if ((i_size_read(inode) == 0) || (range->me_len == 0)) |
53069d4e TY |
775 | return 0; |
776 | ||
777 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
778 | return 0; | |
779 | ||
780 | context->refcount_loc = le64_to_cpu(di->i_refcount_loc); | |
781 | ||
782 | ocfs2_init_dinode_extent_tree(&context->et, INODE_CACHE(inode), di_bh); | |
783 | ocfs2_init_dealloc_ctxt(&context->dealloc); | |
784 | ||
785 | /* | |
786 | * TO-DO XXX: | |
787 | * | |
788 | * - xattr extents. | |
789 | */ | |
790 | ||
791 | do_defrag = context->auto_defrag; | |
792 | ||
793 | /* | |
794 | * extents moving happens in unit of clusters, for the sake | |
795 | * of simplicity, we may ignore two clusters where 'byte_start' | |
796 | * and 'byte_start + len' were within. | |
797 | */ | |
798 | move_start = ocfs2_clusters_for_bytes(osb->sb, range->me_start); | |
799 | len_to_move = (range->me_start + range->me_len) >> | |
800 | osb->s_clustersize_bits; | |
801 | if (len_to_move >= move_start) | |
802 | len_to_move -= move_start; | |
803 | else | |
804 | len_to_move = 0; | |
805 | ||
dda54e76 | 806 | if (do_defrag) { |
53069d4e | 807 | defrag_thresh = range->me_threshold >> osb->s_clustersize_bits; |
dda54e76 TY |
808 | if (defrag_thresh <= 1) |
809 | goto done; | |
810 | } else | |
53069d4e TY |
811 | new_phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, |
812 | range->me_goal); | |
813 | ||
814 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u, " | |
815 | "thresh: %u\n", | |
816 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
817 | (unsigned long long)range->me_start, | |
818 | (unsigned long long)range->me_len, | |
819 | move_start, len_to_move, defrag_thresh); | |
820 | ||
821 | cpos = move_start; | |
822 | while (len_to_move) { | |
823 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &alloc_size, | |
824 | &flags); | |
825 | if (ret) { | |
826 | mlog_errno(ret); | |
827 | goto out; | |
828 | } | |
829 | ||
830 | if (alloc_size > len_to_move) | |
831 | alloc_size = len_to_move; | |
832 | ||
833 | /* | |
834 | * XXX: how to deal with a hole: | |
835 | * | |
836 | * - skip the hole of course | |
837 | * - force a new defragmentation | |
838 | */ | |
839 | if (!phys_cpos) { | |
840 | if (do_defrag) | |
841 | len_defraged = 0; | |
842 | ||
843 | goto next; | |
844 | } | |
845 | ||
846 | if (do_defrag) { | |
847 | ocfs2_calc_extent_defrag_len(&alloc_size, &len_defraged, | |
848 | defrag_thresh, &skip); | |
849 | /* | |
850 | * skip large extents | |
851 | */ | |
852 | if (skip) { | |
853 | skip = 0; | |
854 | goto next; | |
855 | } | |
856 | ||
857 | mlog(0, "#Defrag: cpos: %u, phys_cpos: %u, " | |
858 | "alloc_size: %u, len_defraged: %u\n", | |
859 | cpos, phys_cpos, alloc_size, len_defraged); | |
860 | ||
861 | ret = ocfs2_defrag_extent(context, cpos, phys_cpos, | |
4dfa66bd | 862 | &alloc_size, flags); |
53069d4e TY |
863 | } else { |
864 | ret = ocfs2_move_extent(context, cpos, phys_cpos, | |
865 | &new_phys_cpos, alloc_size, | |
866 | flags); | |
867 | ||
868 | new_phys_cpos += alloc_size; | |
869 | } | |
870 | ||
871 | if (ret < 0) { | |
872 | mlog_errno(ret); | |
873 | goto out; | |
874 | } | |
875 | ||
876 | context->clusters_moved += alloc_size; | |
877 | next: | |
878 | cpos += alloc_size; | |
879 | len_to_move -= alloc_size; | |
880 | } | |
881 | ||
dda54e76 | 882 | done: |
53069d4e TY |
883 | range->me_flags |= OCFS2_MOVE_EXT_FL_COMPLETE; |
884 | ||
885 | out: | |
886 | range->me_moved_len = ocfs2_clusters_to_bytes(osb->sb, | |
887 | context->clusters_moved); | |
888 | range->me_new_offset = ocfs2_clusters_to_bytes(osb->sb, | |
889 | context->new_phys_cpos); | |
890 | ||
891 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
892 | ocfs2_run_deallocs(osb, &context->dealloc); | |
893 | ||
894 | return ret; | |
895 | } | |
896 | ||
897 | static int ocfs2_move_extents(struct ocfs2_move_extents_context *context) | |
898 | { | |
899 | int status; | |
900 | handle_t *handle; | |
901 | struct inode *inode = context->inode; | |
902 | struct ocfs2_dinode *di; | |
903 | struct buffer_head *di_bh = NULL; | |
904 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
905 | ||
53069d4e TY |
906 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
907 | return -EROFS; | |
908 | ||
909 | mutex_lock(&inode->i_mutex); | |
910 | ||
911 | /* | |
912 | * This prevents concurrent writes from other nodes | |
913 | */ | |
914 | status = ocfs2_rw_lock(inode, 1); | |
915 | if (status) { | |
916 | mlog_errno(status); | |
917 | goto out; | |
918 | } | |
919 | ||
920 | status = ocfs2_inode_lock(inode, &di_bh, 1); | |
921 | if (status) { | |
922 | mlog_errno(status); | |
923 | goto out_rw_unlock; | |
924 | } | |
925 | ||
926 | /* | |
927 | * rememer ip_xattr_sem also needs to be held if necessary | |
928 | */ | |
929 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
930 | ||
931 | status = __ocfs2_move_extents_range(di_bh, context); | |
932 | ||
933 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
934 | if (status) { | |
935 | mlog_errno(status); | |
936 | goto out_inode_unlock; | |
937 | } | |
938 | ||
939 | /* | |
940 | * We update ctime for these changes | |
941 | */ | |
942 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
943 | if (IS_ERR(handle)) { | |
944 | status = PTR_ERR(handle); | |
945 | mlog_errno(status); | |
946 | goto out_inode_unlock; | |
947 | } | |
948 | ||
949 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, | |
950 | OCFS2_JOURNAL_ACCESS_WRITE); | |
951 | if (status) { | |
952 | mlog_errno(status); | |
953 | goto out_commit; | |
954 | } | |
955 | ||
956 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
957 | inode->i_ctime = CURRENT_TIME; | |
958 | di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
959 | di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
6fdb702d | 960 | ocfs2_update_inode_fsync_trans(handle, inode, 0); |
53069d4e TY |
961 | |
962 | ocfs2_journal_dirty(handle, di_bh); | |
963 | ||
964 | out_commit: | |
965 | ocfs2_commit_trans(osb, handle); | |
966 | ||
967 | out_inode_unlock: | |
968 | brelse(di_bh); | |
969 | ocfs2_inode_unlock(inode, 1); | |
970 | out_rw_unlock: | |
971 | ocfs2_rw_unlock(inode, 1); | |
972 | out: | |
973 | mutex_unlock(&inode->i_mutex); | |
974 | ||
975 | return status; | |
976 | } | |
977 | ||
978 | int ocfs2_ioctl_move_extents(struct file *filp, void __user *argp) | |
979 | { | |
980 | int status; | |
981 | ||
496ad9aa | 982 | struct inode *inode = file_inode(filp); |
53069d4e | 983 | struct ocfs2_move_extents range; |
85a258b7 DC |
984 | struct ocfs2_move_extents_context *context; |
985 | ||
986 | if (!argp) | |
987 | return -EINVAL; | |
53069d4e | 988 | |
a561be71 | 989 | status = mnt_want_write_file(filp); |
53069d4e TY |
990 | if (status) |
991 | return status; | |
992 | ||
bfbca926 YL |
993 | if ((!S_ISREG(inode->i_mode)) || !(filp->f_mode & FMODE_WRITE)) { |
994 | status = -EPERM; | |
85a258b7 | 995 | goto out_drop; |
bfbca926 | 996 | } |
53069d4e TY |
997 | |
998 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { | |
999 | status = -EPERM; | |
85a258b7 | 1000 | goto out_drop; |
53069d4e TY |
1001 | } |
1002 | ||
1003 | context = kzalloc(sizeof(struct ocfs2_move_extents_context), GFP_NOFS); | |
1004 | if (!context) { | |
1005 | status = -ENOMEM; | |
1006 | mlog_errno(status); | |
85a258b7 | 1007 | goto out_drop; |
53069d4e TY |
1008 | } |
1009 | ||
1010 | context->inode = inode; | |
1011 | context->file = filp; | |
1012 | ||
85a258b7 DC |
1013 | if (copy_from_user(&range, argp, sizeof(range))) { |
1014 | status = -EFAULT; | |
1015 | goto out_free; | |
53069d4e TY |
1016 | } |
1017 | ||
bfbca926 YL |
1018 | if (range.me_start > i_size_read(inode)) { |
1019 | status = -EINVAL; | |
85a258b7 | 1020 | goto out_free; |
bfbca926 | 1021 | } |
53069d4e TY |
1022 | |
1023 | if (range.me_start + range.me_len > i_size_read(inode)) | |
1024 | range.me_len = i_size_read(inode) - range.me_start; | |
1025 | ||
1026 | context->range = ⦥ | |
1027 | ||
1028 | if (range.me_flags & OCFS2_MOVE_EXT_FL_AUTO_DEFRAG) { | |
1029 | context->auto_defrag = 1; | |
dda54e76 TY |
1030 | /* |
1031 | * ok, the default theshold for the defragmentation | |
1032 | * is 1M, since our maximum clustersize was 1M also. | |
1033 | * any thought? | |
1034 | */ | |
53069d4e | 1035 | if (!range.me_threshold) |
53069d4e | 1036 | range.me_threshold = 1024 * 1024; |
dda54e76 TY |
1037 | |
1038 | if (range.me_threshold > i_size_read(inode)) | |
1039 | range.me_threshold = i_size_read(inode); | |
1040 | ||
4dfa66bd TY |
1041 | if (range.me_flags & OCFS2_MOVE_EXT_FL_PART_DEFRAG) |
1042 | context->partial = 1; | |
53069d4e TY |
1043 | } else { |
1044 | /* | |
1045 | * first best-effort attempt to validate and adjust the goal | |
1046 | * (physical address in block), while it can't guarantee later | |
1047 | * operation can succeed all the time since global_bitmap may | |
1048 | * change a bit over time. | |
1049 | */ | |
1050 | ||
1051 | status = ocfs2_validate_and_adjust_move_goal(inode, &range); | |
1052 | if (status) | |
85a258b7 | 1053 | goto out_copy; |
53069d4e TY |
1054 | } |
1055 | ||
1056 | status = ocfs2_move_extents(context); | |
1057 | if (status) | |
1058 | mlog_errno(status); | |
85a258b7 | 1059 | out_copy: |
53069d4e TY |
1060 | /* |
1061 | * movement/defragmentation may end up being partially completed, | |
1062 | * that's the reason why we need to return userspace the finished | |
1063 | * length and new_offset even if failure happens somewhere. | |
1064 | */ | |
85a258b7 DC |
1065 | if (copy_to_user(argp, &range, sizeof(range))) |
1066 | status = -EFAULT; | |
53069d4e | 1067 | |
85a258b7 | 1068 | out_free: |
53069d4e | 1069 | kfree(context); |
85a258b7 | 1070 | out_drop: |
2a79f17e | 1071 | mnt_drop_write_file(filp); |
53069d4e TY |
1072 | |
1073 | return status; | |
1074 | } |