blob: fcd4475d1f899754bb5ed61fa069b149a5e48c20 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * extent_map.c
5 *
6 * In-memory extent map for OCFS2. Man, this code was prettier in
7 * the library.
8 *
9 * Copyright (C) 2004 Oracle. All rights reserved.
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License, version 2, as published by the Free Software Foundation.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/fs.h>
27#include <linux/init.h>
28#include <linux/types.h>
29#include <linux/slab.h>
30#include <linux/rbtree.h>
31
32#define MLOG_MASK_PREFIX ML_EXTENT_MAP
33#include <cluster/masklog.h>
34
35#include "ocfs2.h"
36
37#include "extent_map.h"
38#include "inode.h"
39#include "super.h"
40
41#include "buffer_head_io.h"
42
43
44/*
45 * SUCK SUCK SUCK
46 * Our headers are so bad that struct ocfs2_extent_map is in ocfs.h
47 */
48
49struct ocfs2_extent_map_entry {
50 struct rb_node e_node;
51 int e_tree_depth;
52 struct ocfs2_extent_rec e_rec;
53};
54
55struct ocfs2_em_insert_context {
56 int need_left;
57 int need_right;
58 struct ocfs2_extent_map_entry *new_ent;
59 struct ocfs2_extent_map_entry *old_ent;
60 struct ocfs2_extent_map_entry *left_ent;
61 struct ocfs2_extent_map_entry *right_ent;
62};
63
64static kmem_cache_t *ocfs2_em_ent_cachep = NULL;
65
66
67static struct ocfs2_extent_map_entry *
68ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
69 u32 cpos, u32 clusters,
70 struct rb_node ***ret_p,
71 struct rb_node **ret_parent);
72static int ocfs2_extent_map_insert(struct inode *inode,
73 struct ocfs2_extent_rec *rec,
74 int tree_depth);
75static int ocfs2_extent_map_insert_entry(struct ocfs2_extent_map *em,
76 struct ocfs2_extent_map_entry *ent);
77static int ocfs2_extent_map_find_leaf(struct inode *inode,
78 u32 cpos, u32 clusters,
79 struct ocfs2_extent_list *el);
80static int ocfs2_extent_map_lookup_read(struct inode *inode,
81 u32 cpos, u32 clusters,
82 struct ocfs2_extent_map_entry **ret_ent);
83static int ocfs2_extent_map_try_insert(struct inode *inode,
84 struct ocfs2_extent_rec *rec,
85 int tree_depth,
86 struct ocfs2_em_insert_context *ctxt);
87
88/* returns 1 only if the rec contains all the given clusters -- that is that
89 * rec's cpos is <= the cluster cpos and that the rec endpoint (cpos +
90 * clusters) is >= the argument's endpoint */
91static int ocfs2_extent_rec_contains_clusters(struct ocfs2_extent_rec *rec,
92 u32 cpos, u32 clusters)
93{
94 if (le32_to_cpu(rec->e_cpos) > cpos)
95 return 0;
96 if (cpos + clusters > le32_to_cpu(rec->e_cpos) +
97 le32_to_cpu(rec->e_clusters))
98 return 0;
99 return 1;
100}
101
102
103/*
104 * Find an entry in the tree that intersects the region passed in.
105 * Note that this will find straddled intervals, it is up to the
106 * callers to enforce any boundary conditions.
107 *
108 * Callers must hold ip_lock. This lookup is not guaranteed to return
109 * a tree_depth 0 match, and as such can race inserts if the lock
110 * were not held.
111 *
112 * The rb_node garbage lets insertion share the search. Trivial
113 * callers pass NULL.
114 */
115static struct ocfs2_extent_map_entry *
116ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
117 u32 cpos, u32 clusters,
118 struct rb_node ***ret_p,
119 struct rb_node **ret_parent)
120{
121 struct rb_node **p = &em->em_extents.rb_node;
122 struct rb_node *parent = NULL;
123 struct ocfs2_extent_map_entry *ent = NULL;
124
125 while (*p)
126 {
127 parent = *p;
128 ent = rb_entry(parent, struct ocfs2_extent_map_entry,
129 e_node);
130 if ((cpos + clusters) <= le32_to_cpu(ent->e_rec.e_cpos)) {
131 p = &(*p)->rb_left;
132 ent = NULL;
133 } else if (cpos >= (le32_to_cpu(ent->e_rec.e_cpos) +
134 le32_to_cpu(ent->e_rec.e_clusters))) {
135 p = &(*p)->rb_right;
136 ent = NULL;
137 } else
138 break;
139 }
140
141 if (ret_p != NULL)
142 *ret_p = p;
143 if (ret_parent != NULL)
144 *ret_parent = parent;
145 return ent;
146}
147
148/*
149 * Find the leaf containing the interval we want. While we're on our
150 * way down the tree, fill in every record we see at any depth, because
151 * we might want it later.
152 *
153 * Note that this code is run without ip_lock. That's because it
154 * sleeps while reading. If someone is also filling the extent list at
155 * the same time we are, we might have to restart.
156 */
157static int ocfs2_extent_map_find_leaf(struct inode *inode,
158 u32 cpos, u32 clusters,
159 struct ocfs2_extent_list *el)
160{
161 int i, ret;
162 struct buffer_head *eb_bh = NULL;
163 u64 blkno;
164 u32 rec_end;
165 struct ocfs2_extent_block *eb;
166 struct ocfs2_extent_rec *rec;
167
168 /*
169 * The bh data containing the el cannot change here, because
170 * we hold alloc_sem. So we can do this without other
171 * locks.
172 */
173 while (el->l_tree_depth)
174 {
175 blkno = 0;
176 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
177 rec = &el->l_recs[i];
178 rec_end = (le32_to_cpu(rec->e_cpos) +
179 le32_to_cpu(rec->e_clusters));
180
181 ret = -EBADR;
182 if (rec_end > OCFS2_I(inode)->ip_clusters) {
183 mlog_errno(ret);
Joel Becker110ba902006-02-28 17:58:36 -0800184 ocfs2_error(inode->i_sb,
Mark Fashehb06970532006-03-03 10:24:33 -0800185 "Extent %d at e_blkno %llu of inode %llu goes past ip_clusters of %u\n",
Joel Becker110ba902006-02-28 17:58:36 -0800186 i,
Mark Fashehb06970532006-03-03 10:24:33 -0800187 (unsigned long long)le64_to_cpu(rec->e_blkno),
188 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Joel Becker110ba902006-02-28 17:58:36 -0800189 OCFS2_I(inode)->ip_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -0800190 goto out_free;
191 }
192
193 if (rec_end <= cpos) {
194 ret = ocfs2_extent_map_insert(inode, rec,
195 le16_to_cpu(el->l_tree_depth));
196 if (ret && (ret != -EEXIST)) {
197 mlog_errno(ret);
198 goto out_free;
199 }
200 continue;
201 }
202 if ((cpos + clusters) <= le32_to_cpu(rec->e_cpos)) {
203 ret = ocfs2_extent_map_insert(inode, rec,
204 le16_to_cpu(el->l_tree_depth));
205 if (ret && (ret != -EEXIST)) {
206 mlog_errno(ret);
207 goto out_free;
208 }
209 continue;
210 }
211
212 /*
213 * We've found a record that matches our
214 * interval. We don't insert it because we're
215 * about to traverse it.
216 */
217
218 /* Check to see if we're stradling */
219 ret = -ESRCH;
220 if (!ocfs2_extent_rec_contains_clusters(rec,
221 cpos,
222 clusters)) {
223 mlog_errno(ret);
224 goto out_free;
225 }
226
227 /*
228 * If we've already found a record, the el has
229 * two records covering the same interval.
230 * EEEK!
231 */
232 ret = -EBADR;
233 if (blkno) {
234 mlog_errno(ret);
Joel Becker110ba902006-02-28 17:58:36 -0800235 ocfs2_error(inode->i_sb,
Mark Fashehb06970532006-03-03 10:24:33 -0800236 "Multiple extents for (cpos = %u, clusters = %u) on inode %llu; e_blkno %llu and rec %d at e_blkno %llu\n",
Joel Becker110ba902006-02-28 17:58:36 -0800237 cpos, clusters,
Mark Fashehb06970532006-03-03 10:24:33 -0800238 (unsigned long long)OCFS2_I(inode)->ip_blkno,
239 (unsigned long long)blkno, i,
240 (unsigned long long)le64_to_cpu(rec->e_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800241 goto out_free;
242 }
243
244 blkno = le64_to_cpu(rec->e_blkno);
245 }
246
247 /*
248 * We don't support holes, and we're still up
249 * in the branches, so we'd better have found someone
250 */
251 ret = -EBADR;
252 if (!blkno) {
Joel Becker110ba902006-02-28 17:58:36 -0800253 ocfs2_error(inode->i_sb,
Mark Fashehb06970532006-03-03 10:24:33 -0800254 "No record found for (cpos = %u, clusters = %u) on inode %llu\n",
Joel Becker110ba902006-02-28 17:58:36 -0800255 cpos, clusters,
Mark Fashehb06970532006-03-03 10:24:33 -0800256 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800257 mlog_errno(ret);
258 goto out_free;
259 }
260
261 if (eb_bh) {
262 brelse(eb_bh);
263 eb_bh = NULL;
264 }
265 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
266 blkno, &eb_bh, OCFS2_BH_CACHED,
267 inode);
268 if (ret) {
269 mlog_errno(ret);
270 goto out_free;
271 }
272 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
273 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
274 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
275 ret = -EIO;
276 goto out_free;
277 }
278 el = &eb->h_list;
279 }
280
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100281 BUG_ON(el->l_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -0800282
283 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
284 rec = &el->l_recs[i];
Joel Becker110ba902006-02-28 17:58:36 -0800285
286 if ((le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)) >
287 OCFS2_I(inode)->ip_clusters) {
288 ret = -EBADR;
289 mlog_errno(ret);
290 ocfs2_error(inode->i_sb,
Mark Fashehb06970532006-03-03 10:24:33 -0800291 "Extent %d at e_blkno %llu of inode %llu goes past ip_clusters of %u\n",
Joel Becker110ba902006-02-28 17:58:36 -0800292 i,
Mark Fashehb06970532006-03-03 10:24:33 -0800293 (unsigned long long)le64_to_cpu(rec->e_blkno),
294 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Joel Becker110ba902006-02-28 17:58:36 -0800295 OCFS2_I(inode)->ip_clusters);
296 return ret;
297 }
298
Mark Fashehccd979b2005-12-15 14:31:24 -0800299 ret = ocfs2_extent_map_insert(inode, rec,
300 le16_to_cpu(el->l_tree_depth));
Joel Beckera43db302006-04-27 16:36:14 -0700301 if (ret && (ret != -EEXIST)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800302 mlog_errno(ret);
303 goto out_free;
304 }
305 }
306
307 ret = 0;
308
309out_free:
310 if (eb_bh)
311 brelse(eb_bh);
312
313 return ret;
314}
315
316/*
317 * This lookup actually will read from disk. It has one invariant:
318 * It will never re-traverse blocks. This means that all inserts should
319 * be new regions or more granular regions (both allowed by insert).
320 */
321static int ocfs2_extent_map_lookup_read(struct inode *inode,
322 u32 cpos,
323 u32 clusters,
324 struct ocfs2_extent_map_entry **ret_ent)
325{
326 int ret;
327 u64 blkno;
328 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
329 struct ocfs2_extent_map_entry *ent;
330 struct buffer_head *bh = NULL;
331 struct ocfs2_extent_block *eb;
332 struct ocfs2_dinode *di;
333 struct ocfs2_extent_list *el;
334
335 spin_lock(&OCFS2_I(inode)->ip_lock);
336 ent = ocfs2_extent_map_lookup(em, cpos, clusters, NULL, NULL);
337 if (ent) {
338 if (!ent->e_tree_depth) {
339 spin_unlock(&OCFS2_I(inode)->ip_lock);
340 *ret_ent = ent;
341 return 0;
342 }
343 blkno = le64_to_cpu(ent->e_rec.e_blkno);
344 spin_unlock(&OCFS2_I(inode)->ip_lock);
345
346 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno, &bh,
347 OCFS2_BH_CACHED, inode);
348 if (ret) {
349 mlog_errno(ret);
350 if (bh)
351 brelse(bh);
352 return ret;
353 }
354 eb = (struct ocfs2_extent_block *)bh->b_data;
355 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
356 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
357 brelse(bh);
358 return -EIO;
359 }
360 el = &eb->h_list;
361 } else {
362 spin_unlock(&OCFS2_I(inode)->ip_lock);
363
364 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
365 OCFS2_I(inode)->ip_blkno, &bh,
366 OCFS2_BH_CACHED, inode);
367 if (ret) {
368 mlog_errno(ret);
369 if (bh)
370 brelse(bh);
371 return ret;
372 }
373 di = (struct ocfs2_dinode *)bh->b_data;
374 if (!OCFS2_IS_VALID_DINODE(di)) {
375 brelse(bh);
376 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, di);
377 return -EIO;
378 }
379 el = &di->id2.i_list;
380 }
381
382 ret = ocfs2_extent_map_find_leaf(inode, cpos, clusters, el);
383 brelse(bh);
384 if (ret) {
385 mlog_errno(ret);
386 return ret;
387 }
388
389 ent = ocfs2_extent_map_lookup(em, cpos, clusters, NULL, NULL);
390 if (!ent) {
391 ret = -ESRCH;
392 mlog_errno(ret);
393 return ret;
394 }
395
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100396 /* FIXME: Make sure this isn't a corruption */
397 BUG_ON(ent->e_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -0800398
399 *ret_ent = ent;
400
401 return 0;
402}
403
404/*
405 * Callers must hold ip_lock. This can insert pieces of the tree,
406 * thus racing lookup if the lock weren't held.
407 */
408static int ocfs2_extent_map_insert_entry(struct ocfs2_extent_map *em,
409 struct ocfs2_extent_map_entry *ent)
410{
411 struct rb_node **p, *parent;
412 struct ocfs2_extent_map_entry *old_ent;
413
414 old_ent = ocfs2_extent_map_lookup(em, le32_to_cpu(ent->e_rec.e_cpos),
415 le32_to_cpu(ent->e_rec.e_clusters),
416 &p, &parent);
417 if (old_ent)
418 return -EEXIST;
419
420 rb_link_node(&ent->e_node, parent, p);
421 rb_insert_color(&ent->e_node, &em->em_extents);
422
423 return 0;
424}
425
426
427/*
428 * Simple rule: on any return code other than -EAGAIN, anything left
429 * in the insert_context will be freed.
Joel Beckera43db302006-04-27 16:36:14 -0700430 *
431 * Simple rule #2: A return code of -EEXIST from this function or
432 * its calls to ocfs2_extent_map_insert_entry() signifies that another
433 * thread beat us to the insert. It is not an actual error, but it
434 * tells the caller we have no more work to do.
Mark Fashehccd979b2005-12-15 14:31:24 -0800435 */
436static int ocfs2_extent_map_try_insert(struct inode *inode,
437 struct ocfs2_extent_rec *rec,
438 int tree_depth,
439 struct ocfs2_em_insert_context *ctxt)
440{
441 int ret;
442 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
443 struct ocfs2_extent_map_entry *old_ent;
444
445 ctxt->need_left = 0;
446 ctxt->need_right = 0;
447 ctxt->old_ent = NULL;
448
449 spin_lock(&OCFS2_I(inode)->ip_lock);
450 ret = ocfs2_extent_map_insert_entry(em, ctxt->new_ent);
451 if (!ret) {
452 ctxt->new_ent = NULL;
453 goto out_unlock;
454 }
455
Joel Beckera43db302006-04-27 16:36:14 -0700456 /* Since insert_entry failed, the map MUST have old_ent */
Mark Fashehccd979b2005-12-15 14:31:24 -0800457 old_ent = ocfs2_extent_map_lookup(em, le32_to_cpu(rec->e_cpos),
Joel Beckera43db302006-04-27 16:36:14 -0700458 le32_to_cpu(rec->e_clusters),
459 NULL, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800460
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100461 BUG_ON(!old_ent);
Mark Fashehccd979b2005-12-15 14:31:24 -0800462
Joel Beckera43db302006-04-27 16:36:14 -0700463 if (old_ent->e_tree_depth < tree_depth) {
464 /* Another thread beat us to the lower tree_depth */
465 ret = -EEXIST;
Mark Fashehccd979b2005-12-15 14:31:24 -0800466 goto out_unlock;
Joel Beckera43db302006-04-27 16:36:14 -0700467 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800468
469 if (old_ent->e_tree_depth == tree_depth) {
Joel Beckera43db302006-04-27 16:36:14 -0700470 /*
471 * Another thread beat us to this tree_depth.
472 * Let's make sure we agree with that thread (the
473 * extent_rec should be identical).
474 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800475 if (!memcmp(rec, &old_ent->e_rec,
476 sizeof(struct ocfs2_extent_rec)))
477 ret = 0;
Joel Beckera43db302006-04-27 16:36:14 -0700478 else
479 /* FIXME: Should this be ESRCH/EBADR??? */
480 ret = -EEXIST;
Mark Fashehccd979b2005-12-15 14:31:24 -0800481
Mark Fashehccd979b2005-12-15 14:31:24 -0800482 goto out_unlock;
483 }
484
485 /*
486 * We do it in this order specifically so that no actual tree
487 * changes occur until we have all the pieces we need. We
488 * don't want malloc failures to leave an inconsistent tree.
489 * Whenever we drop the lock, another process could be
490 * inserting. Also note that, if another process just beat us
491 * to an insert, we might not need the same pieces we needed
492 * the first go round. In the end, the pieces we need will
493 * be used, and the pieces we don't will be freed.
494 */
495 ctxt->need_left = !!(le32_to_cpu(rec->e_cpos) >
496 le32_to_cpu(old_ent->e_rec.e_cpos));
497 ctxt->need_right = !!((le32_to_cpu(old_ent->e_rec.e_cpos) +
498 le32_to_cpu(old_ent->e_rec.e_clusters)) >
499 (le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)));
500 ret = -EAGAIN;
501 if (ctxt->need_left) {
502 if (!ctxt->left_ent)
503 goto out_unlock;
504 *(ctxt->left_ent) = *old_ent;
505 ctxt->left_ent->e_rec.e_clusters =
506 cpu_to_le32(le32_to_cpu(rec->e_cpos) -
507 le32_to_cpu(ctxt->left_ent->e_rec.e_cpos));
508 }
509 if (ctxt->need_right) {
510 if (!ctxt->right_ent)
511 goto out_unlock;
512 *(ctxt->right_ent) = *old_ent;
513 ctxt->right_ent->e_rec.e_cpos =
514 cpu_to_le32(le32_to_cpu(rec->e_cpos) +
515 le32_to_cpu(rec->e_clusters));
516 ctxt->right_ent->e_rec.e_clusters =
517 cpu_to_le32((le32_to_cpu(old_ent->e_rec.e_cpos) +
518 le32_to_cpu(old_ent->e_rec.e_clusters)) -
519 le32_to_cpu(ctxt->right_ent->e_rec.e_cpos));
520 }
521
522 rb_erase(&old_ent->e_node, &em->em_extents);
523 /* Now that he's erased, set him up for deletion */
524 ctxt->old_ent = old_ent;
525
526 if (ctxt->need_left) {
527 ret = ocfs2_extent_map_insert_entry(em,
528 ctxt->left_ent);
529 if (ret)
530 goto out_unlock;
531 ctxt->left_ent = NULL;
532 }
533
534 if (ctxt->need_right) {
535 ret = ocfs2_extent_map_insert_entry(em,
536 ctxt->right_ent);
537 if (ret)
538 goto out_unlock;
539 ctxt->right_ent = NULL;
540 }
541
542 ret = ocfs2_extent_map_insert_entry(em, ctxt->new_ent);
543
544 if (!ret)
545 ctxt->new_ent = NULL;
546
547out_unlock:
548 spin_unlock(&OCFS2_I(inode)->ip_lock);
549
550 return ret;
551}
552
553
554static int ocfs2_extent_map_insert(struct inode *inode,
555 struct ocfs2_extent_rec *rec,
556 int tree_depth)
557{
558 int ret;
559 struct ocfs2_em_insert_context ctxt = {0, };
560
561 if ((le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)) >
562 OCFS2_I(inode)->ip_map.em_clusters) {
563 ret = -EBADR;
564 mlog_errno(ret);
565 return ret;
566 }
567
568 /* Zero e_clusters means a truncated tail record. It better be EOF */
569 if (!rec->e_clusters) {
570 if ((le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)) !=
571 OCFS2_I(inode)->ip_map.em_clusters) {
572 ret = -EBADR;
573 mlog_errno(ret);
Joel Becker110ba902006-02-28 17:58:36 -0800574 ocfs2_error(inode->i_sb,
Mark Fashehb06970532006-03-03 10:24:33 -0800575 "Zero e_clusters on non-tail extent record at e_blkno %llu on inode %llu\n",
576 (unsigned long long)le64_to_cpu(rec->e_blkno),
577 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800578 return ret;
579 }
580
581 /* Ignore the truncated tail */
582 return 0;
583 }
584
585 ret = -ENOMEM;
586 ctxt.new_ent = kmem_cache_alloc(ocfs2_em_ent_cachep,
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700587 GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800588 if (!ctxt.new_ent) {
589 mlog_errno(ret);
590 return ret;
591 }
592
593 ctxt.new_ent->e_rec = *rec;
594 ctxt.new_ent->e_tree_depth = tree_depth;
595
596 do {
597 ret = -ENOMEM;
598 if (ctxt.need_left && !ctxt.left_ent) {
599 ctxt.left_ent =
600 kmem_cache_alloc(ocfs2_em_ent_cachep,
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700601 GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800602 if (!ctxt.left_ent)
603 break;
604 }
605 if (ctxt.need_right && !ctxt.right_ent) {
606 ctxt.right_ent =
607 kmem_cache_alloc(ocfs2_em_ent_cachep,
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700608 GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800609 if (!ctxt.right_ent)
610 break;
611 }
612
613 ret = ocfs2_extent_map_try_insert(inode, rec,
614 tree_depth, &ctxt);
615 } while (ret == -EAGAIN);
616
Joel Beckera43db302006-04-27 16:36:14 -0700617 if ((ret < 0) && (ret != -EEXIST))
Mark Fashehccd979b2005-12-15 14:31:24 -0800618 mlog_errno(ret);
619
620 if (ctxt.left_ent)
621 kmem_cache_free(ocfs2_em_ent_cachep, ctxt.left_ent);
622 if (ctxt.right_ent)
623 kmem_cache_free(ocfs2_em_ent_cachep, ctxt.right_ent);
624 if (ctxt.old_ent)
625 kmem_cache_free(ocfs2_em_ent_cachep, ctxt.old_ent);
626 if (ctxt.new_ent)
627 kmem_cache_free(ocfs2_em_ent_cachep, ctxt.new_ent);
628
629 return ret;
630}
631
632/*
633 * Append this record to the tail of the extent map. It must be
634 * tree_depth 0. The record might be an extension of an existing
635 * record, and as such that needs to be handled. eg:
636 *
637 * Existing record in the extent map:
638 *
639 * cpos = 10, len = 10
Joel Becker110ba902006-02-28 17:58:36 -0800640 * |---------|
Mark Fashehccd979b2005-12-15 14:31:24 -0800641 *
642 * New Record:
643 *
644 * cpos = 10, len = 20
Joel Becker110ba902006-02-28 17:58:36 -0800645 * |------------------|
Mark Fashehccd979b2005-12-15 14:31:24 -0800646 *
647 * The passed record is the new on-disk record. The new_clusters value
648 * is how many clusters were added to the file. If the append is a
649 * contiguous append, the new_clusters has been added to
650 * rec->e_clusters. If the append is an entirely new extent, then
651 * rec->e_clusters is == new_clusters.
652 */
653int ocfs2_extent_map_append(struct inode *inode,
654 struct ocfs2_extent_rec *rec,
655 u32 new_clusters)
656{
657 int ret;
658 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
659 struct ocfs2_extent_map_entry *ent;
660 struct ocfs2_extent_rec *old;
661
662 BUG_ON(!new_clusters);
663 BUG_ON(le32_to_cpu(rec->e_clusters) < new_clusters);
664
665 if (em->em_clusters < OCFS2_I(inode)->ip_clusters) {
666 /*
667 * Size changed underneath us on disk. Drop any
668 * straddling records and update our idea of
669 * i_clusters
670 */
671 ocfs2_extent_map_drop(inode, em->em_clusters - 1);
672 em->em_clusters = OCFS2_I(inode)->ip_clusters;
673 }
674
675 mlog_bug_on_msg((le32_to_cpu(rec->e_cpos) +
676 le32_to_cpu(rec->e_clusters)) !=
677 (em->em_clusters + new_clusters),
Mark Fashehb06970532006-03-03 10:24:33 -0800678 "Inode %llu:\n"
Mark Fashehccd979b2005-12-15 14:31:24 -0800679 "rec->e_cpos = %u + rec->e_clusters = %u = %u\n"
680 "em->em_clusters = %u + new_clusters = %u = %u\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800681 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800682 le32_to_cpu(rec->e_cpos), le32_to_cpu(rec->e_clusters),
683 le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters),
684 em->em_clusters, new_clusters,
685 em->em_clusters + new_clusters);
686
687 em->em_clusters += new_clusters;
688
689 ret = -ENOENT;
690 if (le32_to_cpu(rec->e_clusters) > new_clusters) {
691 /* This is a contiguous append */
692 ent = ocfs2_extent_map_lookup(em, le32_to_cpu(rec->e_cpos), 1,
693 NULL, NULL);
694 if (ent) {
695 old = &ent->e_rec;
696 BUG_ON((le32_to_cpu(rec->e_cpos) +
697 le32_to_cpu(rec->e_clusters)) !=
698 (le32_to_cpu(old->e_cpos) +
699 le32_to_cpu(old->e_clusters) +
700 new_clusters));
701 if (ent->e_tree_depth == 0) {
702 BUG_ON(le32_to_cpu(old->e_cpos) !=
703 le32_to_cpu(rec->e_cpos));
704 BUG_ON(le64_to_cpu(old->e_blkno) !=
705 le64_to_cpu(rec->e_blkno));
706 ret = 0;
707 }
708 /*
709 * Let non-leafs fall through as -ENOENT to
710 * force insertion of the new leaf.
711 */
712 le32_add_cpu(&old->e_clusters, new_clusters);
713 }
714 }
715
716 if (ret == -ENOENT)
717 ret = ocfs2_extent_map_insert(inode, rec, 0);
718 if (ret < 0)
719 mlog_errno(ret);
720 return ret;
721}
722
723#if 0
724/* Code here is included but defined out as it completes the extent
725 * map api and may be used in the future. */
726
727/*
728 * Look up the record containing this cluster offset. This record is
729 * part of the extent map. Do not free it. Any changes you make to
730 * it will reflect in the extent map. So, if your last extent
731 * is (cpos = 10, clusters = 10) and you truncate the file by 5
732 * clusters, you can do:
733 *
734 * ret = ocfs2_extent_map_get_rec(em, orig_size - 5, &rec);
735 * rec->e_clusters -= 5;
736 *
737 * The lookup does not read from disk. If the map isn't filled in for
738 * an entry, you won't find it.
739 *
740 * Also note that the returned record is valid until alloc_sem is
741 * dropped. After that, truncate and extend can happen. Caveat Emptor.
742 */
743int ocfs2_extent_map_get_rec(struct inode *inode, u32 cpos,
744 struct ocfs2_extent_rec **rec,
745 int *tree_depth)
746{
747 int ret = -ENOENT;
748 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
749 struct ocfs2_extent_map_entry *ent;
750
751 *rec = NULL;
752
753 if (cpos >= OCFS2_I(inode)->ip_clusters)
754 return -EINVAL;
755
756 if (cpos >= em->em_clusters) {
757 /*
758 * Size changed underneath us on disk. Drop any
759 * straddling records and update our idea of
760 * i_clusters
761 */
762 ocfs2_extent_map_drop(inode, em->em_clusters - 1);
763 em->em_clusters = OCFS2_I(inode)->ip_clusters ;
764 }
765
766 ent = ocfs2_extent_map_lookup(&OCFS2_I(inode)->ip_map, cpos, 1,
767 NULL, NULL);
768
769 if (ent) {
770 *rec = &ent->e_rec;
771 if (tree_depth)
772 *tree_depth = ent->e_tree_depth;
773 ret = 0;
774 }
775
776 return ret;
777}
778
779int ocfs2_extent_map_get_clusters(struct inode *inode,
780 u32 v_cpos, int count,
781 u32 *p_cpos, int *ret_count)
782{
783 int ret;
784 u32 coff, ccount;
785 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
786 struct ocfs2_extent_map_entry *ent = NULL;
787
788 *p_cpos = ccount = 0;
789
790 if ((v_cpos + count) > OCFS2_I(inode)->ip_clusters)
791 return -EINVAL;
792
793 if ((v_cpos + count) > em->em_clusters) {
794 /*
795 * Size changed underneath us on disk. Drop any
796 * straddling records and update our idea of
797 * i_clusters
798 */
799 ocfs2_extent_map_drop(inode, em->em_clusters - 1);
800 em->em_clusters = OCFS2_I(inode)->ip_clusters;
801 }
802
803
804 ret = ocfs2_extent_map_lookup_read(inode, v_cpos, count, &ent);
805 if (ret)
806 return ret;
807
808 if (ent) {
809 /* We should never find ourselves straddling an interval */
810 if (!ocfs2_extent_rec_contains_clusters(&ent->e_rec,
811 v_cpos,
812 count))
813 return -ESRCH;
814
815 coff = v_cpos - le32_to_cpu(ent->e_rec.e_cpos);
816 *p_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
817 le64_to_cpu(ent->e_rec.e_blkno)) +
818 coff;
819
820 if (ret_count)
821 *ret_count = le32_to_cpu(ent->e_rec.e_clusters) - coff;
822
823 return 0;
824 }
825
826
827 return -ENOENT;
828}
829
830#endif /* 0 */
831
832int ocfs2_extent_map_get_blocks(struct inode *inode,
833 u64 v_blkno, int count,
834 u64 *p_blkno, int *ret_count)
835{
836 int ret;
837 u64 boff;
838 u32 cpos, clusters;
839 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
840 struct ocfs2_extent_map_entry *ent = NULL;
841 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
842 struct ocfs2_extent_rec *rec;
843
844 *p_blkno = 0;
845
846 cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
847 clusters = ocfs2_blocks_to_clusters(inode->i_sb,
848 (u64)count + bpc - 1);
849 if ((cpos + clusters) > OCFS2_I(inode)->ip_clusters) {
850 ret = -EINVAL;
851 mlog_errno(ret);
852 return ret;
853 }
854
855 if ((cpos + clusters) > em->em_clusters) {
856 /*
857 * Size changed underneath us on disk. Drop any
858 * straddling records and update our idea of
859 * i_clusters
860 */
861 ocfs2_extent_map_drop(inode, em->em_clusters - 1);
862 em->em_clusters = OCFS2_I(inode)->ip_clusters;
863 }
864
865 ret = ocfs2_extent_map_lookup_read(inode, cpos, clusters, &ent);
866 if (ret) {
867 mlog_errno(ret);
868 return ret;
869 }
870
871 if (ent)
872 {
873 rec = &ent->e_rec;
874
875 /* We should never find ourselves straddling an interval */
876 if (!ocfs2_extent_rec_contains_clusters(rec, cpos, clusters)) {
877 ret = -ESRCH;
878 mlog_errno(ret);
879 return ret;
880 }
881
882 boff = ocfs2_clusters_to_blocks(inode->i_sb, cpos -
883 le32_to_cpu(rec->e_cpos));
884 boff += (v_blkno & (u64)(bpc - 1));
885 *p_blkno = le64_to_cpu(rec->e_blkno) + boff;
886
887 if (ret_count) {
888 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb,
889 le32_to_cpu(rec->e_clusters)) - boff;
890 }
891
892 return 0;
893 }
894
895 return -ENOENT;
896}
897
898int ocfs2_extent_map_init(struct inode *inode)
899{
900 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
901
902 em->em_extents = RB_ROOT;
903 em->em_clusters = 0;
904
905 return 0;
906}
907
908/* Needs the lock */
909static void __ocfs2_extent_map_drop(struct inode *inode,
910 u32 new_clusters,
911 struct rb_node **free_head,
912 struct ocfs2_extent_map_entry **tail_ent)
913{
914 struct rb_node *node, *next;
915 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
916 struct ocfs2_extent_map_entry *ent;
917
918 *free_head = NULL;
919
920 ent = NULL;
921 node = rb_last(&em->em_extents);
922 while (node)
923 {
924 next = rb_prev(node);
925
926 ent = rb_entry(node, struct ocfs2_extent_map_entry,
927 e_node);
928 if (le32_to_cpu(ent->e_rec.e_cpos) < new_clusters)
929 break;
930
931 rb_erase(&ent->e_node, &em->em_extents);
932
933 node->rb_right = *free_head;
934 *free_head = node;
935
936 ent = NULL;
937 node = next;
938 }
939
940 /* Do we have an entry straddling new_clusters? */
941 if (tail_ent) {
942 if (ent &&
943 ((le32_to_cpu(ent->e_rec.e_cpos) +
944 le32_to_cpu(ent->e_rec.e_clusters)) > new_clusters))
945 *tail_ent = ent;
946 else
947 *tail_ent = NULL;
948 }
949}
950
951static void __ocfs2_extent_map_drop_cleanup(struct rb_node *free_head)
952{
953 struct rb_node *node;
954 struct ocfs2_extent_map_entry *ent;
955
956 while (free_head) {
957 node = free_head;
958 free_head = node->rb_right;
959
960 ent = rb_entry(node, struct ocfs2_extent_map_entry,
961 e_node);
962 kmem_cache_free(ocfs2_em_ent_cachep, ent);
963 }
964}
965
966/*
967 * Remove all entries past new_clusters, inclusive of an entry that
968 * contains new_clusters. This is effectively a cache forget.
969 *
970 * If you want to also clip the last extent by some number of clusters,
971 * you need to call ocfs2_extent_map_trunc().
972 * This code does not check or modify ip_clusters.
973 */
974int ocfs2_extent_map_drop(struct inode *inode, u32 new_clusters)
975{
976 struct rb_node *free_head = NULL;
977 struct ocfs2_extent_map *em = &OCFS2_I(inode)->ip_map;
978 struct ocfs2_extent_map_entry *ent;
979
980 spin_lock(&OCFS2_I(inode)->ip_lock);
981
982 __ocfs2_extent_map_drop(inode, new_clusters, &free_head, &ent);
983
984 if (ent) {
985 rb_erase(&ent->e_node, &em->em_extents);
986 ent->e_node.rb_right = free_head;
987 free_head = &ent->e_node;
988 }
989
990 spin_unlock(&OCFS2_I(inode)->ip_lock);
991
992 if (free_head)
993 __ocfs2_extent_map_drop_cleanup(free_head);
994
995 return 0;
996}
997
998/*
999 * Remove all entries past new_clusters and also clip any extent
1000 * straddling new_clusters, if there is one. This does not check
1001 * or modify ip_clusters
1002 */
1003int ocfs2_extent_map_trunc(struct inode *inode, u32 new_clusters)
1004{
1005 struct rb_node *free_head = NULL;
1006 struct ocfs2_extent_map_entry *ent = NULL;
1007
1008 spin_lock(&OCFS2_I(inode)->ip_lock);
1009
1010 __ocfs2_extent_map_drop(inode, new_clusters, &free_head, &ent);
1011
1012 if (ent)
1013 ent->e_rec.e_clusters = cpu_to_le32(new_clusters -
1014 le32_to_cpu(ent->e_rec.e_cpos));
1015
1016 OCFS2_I(inode)->ip_map.em_clusters = new_clusters;
1017
1018 spin_unlock(&OCFS2_I(inode)->ip_lock);
1019
1020 if (free_head)
1021 __ocfs2_extent_map_drop_cleanup(free_head);
1022
1023 return 0;
1024}
1025
1026int __init init_ocfs2_extent_maps(void)
1027{
1028 ocfs2_em_ent_cachep =
1029 kmem_cache_create("ocfs2_em_ent",
1030 sizeof(struct ocfs2_extent_map_entry),
1031 0, SLAB_HWCACHE_ALIGN, NULL, NULL);
1032 if (!ocfs2_em_ent_cachep)
1033 return -ENOMEM;
1034
1035 return 0;
1036}
1037
Adrian Bunk0c6c98f2006-01-07 20:07:02 +01001038void exit_ocfs2_extent_maps(void)
Mark Fashehccd979b2005-12-15 14:31:24 -08001039{
1040 kmem_cache_destroy(ocfs2_em_ent_cachep);
1041}