blob: 70b5863a2d64e05cde6474bd387112f3b470b5e7 [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 *
Mark Fasheh363041a2007-01-17 12:31:35 -08006 * Block/Cluster mapping functions
Mark Fashehccd979b2005-12-15 14:31:24 -08007 *
8 * Copyright (C) 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License, version 2, as published by the Free Software Foundation.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public
20 * License along with this program; if not, write to the
21 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
22 * Boston, MA 021110-1307, USA.
23 */
24
25#include <linux/fs.h>
26#include <linux/init.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090027#include <linux/slab.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080028#include <linux/types.h>
Mark Fasheh00dc4172008-10-03 17:32:11 -040029#include <linux/fiemap.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080030
Mark Fashehccd979b2005-12-15 14:31:24 -080031#include <cluster/masklog.h>
32
33#include "ocfs2.h"
34
Mark Fasheh363041a2007-01-17 12:31:35 -080035#include "alloc.h"
Mark Fasheh00dc4172008-10-03 17:32:11 -040036#include "dlmglue.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080037#include "extent_map.h"
38#include "inode.h"
39#include "super.h"
Tristan Ye86239d52009-12-22 09:11:58 +080040#include "symlink.h"
Tao Maa7163572011-02-22 22:27:33 +080041#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080042
43#include "buffer_head_io.h"
44
Mark Fashehccd979b2005-12-15 14:31:24 -080045/*
Mark Fasheh83418972007-04-23 18:53:12 -070046 * The extent caching implementation is intentionally trivial.
47 *
48 * We only cache a small number of extents stored directly on the
49 * inode, so linear order operations are acceptable. If we ever want
50 * to increase the size of the extent map, then these algorithms must
51 * get smarter.
52 */
53
54void ocfs2_extent_map_init(struct inode *inode)
55{
56 struct ocfs2_inode_info *oi = OCFS2_I(inode);
57
58 oi->ip_extent_map.em_num_items = 0;
59 INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
60}
61
62static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
63 unsigned int cpos,
64 struct ocfs2_extent_map_item **ret_emi)
65{
66 unsigned int range;
67 struct ocfs2_extent_map_item *emi;
68
69 *ret_emi = NULL;
70
71 list_for_each_entry(emi, &em->em_list, ei_list) {
72 range = emi->ei_cpos + emi->ei_clusters;
73
74 if (cpos >= emi->ei_cpos && cpos < range) {
75 list_move(&emi->ei_list, &em->em_list);
76
77 *ret_emi = emi;
78 break;
79 }
80 }
81}
82
83static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
84 unsigned int *phys, unsigned int *len,
85 unsigned int *flags)
86{
87 unsigned int coff;
88 struct ocfs2_inode_info *oi = OCFS2_I(inode);
89 struct ocfs2_extent_map_item *emi;
90
91 spin_lock(&oi->ip_lock);
92
93 __ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
94 if (emi) {
95 coff = cpos - emi->ei_cpos;
96 *phys = emi->ei_phys + coff;
97 if (len)
98 *len = emi->ei_clusters - coff;
99 if (flags)
100 *flags = emi->ei_flags;
101 }
102
103 spin_unlock(&oi->ip_lock);
104
105 if (emi == NULL)
106 return -ENOENT;
107
108 return 0;
109}
110
111/*
112 * Forget about all clusters equal to or greater than cpos.
113 */
114void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
115{
Christoph Hellwig800deef2007-05-17 16:03:13 +0200116 struct ocfs2_extent_map_item *emi, *n;
Mark Fasheh83418972007-04-23 18:53:12 -0700117 struct ocfs2_inode_info *oi = OCFS2_I(inode);
118 struct ocfs2_extent_map *em = &oi->ip_extent_map;
119 LIST_HEAD(tmp_list);
120 unsigned int range;
121
122 spin_lock(&oi->ip_lock);
Christoph Hellwig800deef2007-05-17 16:03:13 +0200123 list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
Mark Fasheh83418972007-04-23 18:53:12 -0700124 if (emi->ei_cpos >= cpos) {
125 /* Full truncate of this record. */
126 list_move(&emi->ei_list, &tmp_list);
127 BUG_ON(em->em_num_items == 0);
128 em->em_num_items--;
129 continue;
130 }
131
132 range = emi->ei_cpos + emi->ei_clusters;
133 if (range > cpos) {
134 /* Partial truncate */
135 emi->ei_clusters = cpos - emi->ei_cpos;
136 }
137 }
138 spin_unlock(&oi->ip_lock);
139
Christoph Hellwig800deef2007-05-17 16:03:13 +0200140 list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
Mark Fasheh83418972007-04-23 18:53:12 -0700141 list_del(&emi->ei_list);
142 kfree(emi);
143 }
144}
145
146/*
147 * Is any part of emi2 contained within emi1
148 */
149static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
150 struct ocfs2_extent_map_item *emi2)
151{
152 unsigned int range1, range2;
153
154 /*
155 * Check if logical start of emi2 is inside emi1
156 */
157 range1 = emi1->ei_cpos + emi1->ei_clusters;
158 if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
159 return 1;
160
161 /*
162 * Check if logical end of emi2 is inside emi1
163 */
164 range2 = emi2->ei_cpos + emi2->ei_clusters;
165 if (range2 > emi1->ei_cpos && range2 <= range1)
166 return 1;
167
168 return 0;
169}
170
171static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
172 struct ocfs2_extent_map_item *src)
173{
174 dest->ei_cpos = src->ei_cpos;
175 dest->ei_phys = src->ei_phys;
176 dest->ei_clusters = src->ei_clusters;
177 dest->ei_flags = src->ei_flags;
178}
179
180/*
181 * Try to merge emi with ins. Returns 1 if merge succeeds, zero
182 * otherwise.
183 */
184static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
185 struct ocfs2_extent_map_item *ins)
186{
187 /*
188 * Handle contiguousness
189 */
190 if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
191 ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
192 ins->ei_flags == emi->ei_flags) {
193 emi->ei_clusters += ins->ei_clusters;
194 return 1;
195 } else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
Roel Kluinbd6b0bf2010-02-05 10:26:27 +0100196 (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
Mark Fasheh83418972007-04-23 18:53:12 -0700197 ins->ei_flags == emi->ei_flags) {
198 emi->ei_phys = ins->ei_phys;
199 emi->ei_cpos = ins->ei_cpos;
200 emi->ei_clusters += ins->ei_clusters;
201 return 1;
202 }
203
204 /*
205 * Overlapping extents - this shouldn't happen unless we've
206 * split an extent to change it's flags. That is exceedingly
207 * rare, so there's no sense in trying to optimize it yet.
208 */
209 if (ocfs2_ei_is_contained(emi, ins) ||
210 ocfs2_ei_is_contained(ins, emi)) {
211 ocfs2_copy_emi_fields(emi, ins);
212 return 1;
213 }
214
215 /* No merge was possible. */
216 return 0;
217}
218
219/*
220 * In order to reduce complexity on the caller, this insert function
221 * is intentionally liberal in what it will accept.
222 *
223 * The only rule is that the truncate call *must* be used whenever
224 * records have been deleted. This avoids inserting overlapping
225 * records with different physical mappings.
226 */
227void ocfs2_extent_map_insert_rec(struct inode *inode,
228 struct ocfs2_extent_rec *rec)
229{
230 struct ocfs2_inode_info *oi = OCFS2_I(inode);
231 struct ocfs2_extent_map *em = &oi->ip_extent_map;
232 struct ocfs2_extent_map_item *emi, *new_emi = NULL;
233 struct ocfs2_extent_map_item ins;
234
235 ins.ei_cpos = le32_to_cpu(rec->e_cpos);
236 ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
237 le64_to_cpu(rec->e_blkno));
238 ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
239 ins.ei_flags = rec->e_flags;
240
241search:
242 spin_lock(&oi->ip_lock);
243
244 list_for_each_entry(emi, &em->em_list, ei_list) {
245 if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
246 list_move(&emi->ei_list, &em->em_list);
247 spin_unlock(&oi->ip_lock);
248 goto out;
249 }
250 }
251
252 /*
253 * No item could be merged.
254 *
255 * Either allocate and add a new item, or overwrite the last recently
256 * inserted.
257 */
258
259 if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
260 if (new_emi == NULL) {
261 spin_unlock(&oi->ip_lock);
262
263 new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
264 if (new_emi == NULL)
265 goto out;
266
267 goto search;
268 }
269
270 ocfs2_copy_emi_fields(new_emi, &ins);
271 list_add(&new_emi->ei_list, &em->em_list);
272 em->em_num_items++;
273 new_emi = NULL;
274 } else {
275 BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
276 emi = list_entry(em->em_list.prev,
277 struct ocfs2_extent_map_item, ei_list);
278 list_move(&emi->ei_list, &em->em_list);
279 ocfs2_copy_emi_fields(emi, &ins);
280 }
281
282 spin_unlock(&oi->ip_lock);
283
284out:
285 if (new_emi)
286 kfree(new_emi);
287}
288
Mark Fasheh00dc4172008-10-03 17:32:11 -0400289static int ocfs2_last_eb_is_empty(struct inode *inode,
290 struct ocfs2_dinode *di)
291{
292 int ret, next_free;
293 u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
294 struct buffer_head *eb_bh = NULL;
295 struct ocfs2_extent_block *eb;
296 struct ocfs2_extent_list *el;
297
Joel Becker3d03a302009-02-12 17:49:26 -0800298 ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
Mark Fasheh00dc4172008-10-03 17:32:11 -0400299 if (ret) {
300 mlog_errno(ret);
301 goto out;
302 }
303
304 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
305 el = &eb->h_list;
306
Mark Fasheh00dc4172008-10-03 17:32:11 -0400307 if (el->l_tree_depth) {
308 ocfs2_error(inode->i_sb,
309 "Inode %lu has non zero tree depth in "
310 "leaf block %llu\n", inode->i_ino,
311 (unsigned long long)eb_bh->b_blocknr);
312 ret = -EROFS;
313 goto out;
314 }
315
316 next_free = le16_to_cpu(el->l_next_free_rec);
317
318 if (next_free == 0 ||
319 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
320 ret = 1;
321
322out:
323 brelse(eb_bh);
324 return ret;
325}
326
Mark Fasheh83418972007-04-23 18:53:12 -0700327/*
Mark Fasheh4f902c32007-03-09 16:26:50 -0800328 * Return the 1st index within el which contains an extent start
329 * larger than v_cluster.
330 */
331static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
332 u32 v_cluster)
333{
334 int i;
335 struct ocfs2_extent_rec *rec;
336
337 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
338 rec = &el->l_recs[i];
339
340 if (v_cluster < le32_to_cpu(rec->e_cpos))
341 break;
342 }
343
344 return i;
345}
346
347/*
348 * Figure out the size of a hole which starts at v_cluster within the given
349 * extent list.
350 *
351 * If there is no more allocation past v_cluster, we return the maximum
352 * cluster size minus v_cluster.
353 *
354 * If we have in-inode extents, then el points to the dinode list and
355 * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
356 * containing el.
357 */
Tao Mae73a8192009-08-11 14:33:14 +0800358int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
359 struct ocfs2_extent_list *el,
360 struct buffer_head *eb_bh,
361 u32 v_cluster,
362 u32 *num_clusters)
Mark Fasheh4f902c32007-03-09 16:26:50 -0800363{
364 int ret, i;
365 struct buffer_head *next_eb_bh = NULL;
366 struct ocfs2_extent_block *eb, *next_eb;
367
368 i = ocfs2_search_for_hole_index(el, v_cluster);
369
370 if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
371 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
372
373 /*
374 * Check the next leaf for any extents.
375 */
376
377 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
378 goto no_more_extents;
379
Tao Mae73a8192009-08-11 14:33:14 +0800380 ret = ocfs2_read_extent_block(ci,
Joel Becker5e965812008-11-13 14:49:16 -0800381 le64_to_cpu(eb->h_next_leaf_blk),
382 &next_eb_bh);
Mark Fasheh4f902c32007-03-09 16:26:50 -0800383 if (ret) {
384 mlog_errno(ret);
385 goto out;
386 }
Joel Becker5e965812008-11-13 14:49:16 -0800387
Mark Fasheh4f902c32007-03-09 16:26:50 -0800388 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800389 el = &next_eb->h_list;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800390 i = ocfs2_search_for_hole_index(el, v_cluster);
391 }
392
393no_more_extents:
394 if (i == le16_to_cpu(el->l_next_free_rec)) {
395 /*
396 * We're at the end of our existing allocation. Just
397 * return the maximum number of clusters we could
398 * possibly allocate.
399 */
400 *num_clusters = UINT_MAX - v_cluster;
401 } else {
402 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
403 }
404
405 ret = 0;
406out:
407 brelse(next_eb_bh);
408 return ret;
409}
410
Mark Fasheh00dc4172008-10-03 17:32:11 -0400411static int ocfs2_get_clusters_nocache(struct inode *inode,
412 struct buffer_head *di_bh,
413 u32 v_cluster, unsigned int *hole_len,
414 struct ocfs2_extent_rec *ret_rec,
415 unsigned int *is_last)
Mark Fashehccd979b2005-12-15 14:31:24 -0800416{
Mark Fasheh00dc4172008-10-03 17:32:11 -0400417 int i, ret, tree_height, len;
Mark Fashehccd979b2005-12-15 14:31:24 -0800418 struct ocfs2_dinode *di;
Mark Fasheh00dc4172008-10-03 17:32:11 -0400419 struct ocfs2_extent_block *uninitialized_var(eb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800420 struct ocfs2_extent_list *el;
Mark Fasheh363041a2007-01-17 12:31:35 -0800421 struct ocfs2_extent_rec *rec;
Mark Fasheh00dc4172008-10-03 17:32:11 -0400422 struct buffer_head *eb_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800423
Mark Fasheh00dc4172008-10-03 17:32:11 -0400424 memset(ret_rec, 0, sizeof(*ret_rec));
425 if (is_last)
426 *is_last = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800427
Mark Fasheh363041a2007-01-17 12:31:35 -0800428 di = (struct ocfs2_dinode *) di_bh->b_data;
429 el = &di->id2.i_list;
Mark Fasheh00dc4172008-10-03 17:32:11 -0400430 tree_height = le16_to_cpu(el->l_tree_depth);
Mark Fasheh363041a2007-01-17 12:31:35 -0800431
Mark Fasheh00dc4172008-10-03 17:32:11 -0400432 if (tree_height > 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -0800433 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
434 &eb_bh);
Mark Fasheh363041a2007-01-17 12:31:35 -0800435 if (ret) {
436 mlog_errno(ret);
437 goto out;
438 }
439
440 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
441 el = &eb->h_list;
Mark Fashehe48edee2007-03-07 16:46:57 -0800442
443 if (el->l_tree_depth) {
444 ocfs2_error(inode->i_sb,
445 "Inode %lu has non zero tree depth in "
446 "leaf block %llu\n", inode->i_ino,
447 (unsigned long long)eb_bh->b_blocknr);
448 ret = -EROFS;
449 goto out;
450 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800451 }
452
Mark Fasheh363041a2007-01-17 12:31:35 -0800453 i = ocfs2_search_extent_list(el, v_cluster);
454 if (i == -1) {
455 /*
Mark Fasheh00dc4172008-10-03 17:32:11 -0400456 * Holes can be larger than the maximum size of an
Daniel Mack3ad2f3f2010-02-03 08:01:28 +0800457 * extent, so we return their lengths in a separate
Mark Fasheh00dc4172008-10-03 17:32:11 -0400458 * field.
459 */
460 if (hole_len) {
Tao Mae73a8192009-08-11 14:33:14 +0800461 ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
462 el, eb_bh,
Mark Fasheh00dc4172008-10-03 17:32:11 -0400463 v_cluster, &len);
464 if (ret) {
465 mlog_errno(ret);
466 goto out;
467 }
468
469 *hole_len = len;
470 }
471 goto out_hole;
472 }
473
474 rec = &el->l_recs[i];
475
476 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
477
478 if (!rec->e_blkno) {
479 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
480 "record (%u, %u, 0)", inode->i_ino,
481 le32_to_cpu(rec->e_cpos),
482 ocfs2_rec_clusters(el, rec));
483 ret = -EROFS;
484 goto out;
485 }
486
487 *ret_rec = *rec;
488
489 /*
490 * Checking for last extent is potentially expensive - we
491 * might have to look at the next leaf over to see if it's
492 * empty.
493 *
494 * The first two checks are to see whether the caller even
495 * cares for this information, and if the extent is at least
496 * the last in it's list.
497 *
498 * If those hold true, then the extent is last if any of the
499 * additional conditions hold true:
500 * - Extent list is in-inode
501 * - Extent list is right-most
502 * - Extent list is 2nd to rightmost, with empty right-most
503 */
504 if (is_last) {
505 if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
506 if (tree_height == 0)
507 *is_last = 1;
508 else if (eb->h_blkno == di->i_last_eb_blk)
509 *is_last = 1;
510 else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
511 ret = ocfs2_last_eb_is_empty(inode, di);
512 if (ret < 0) {
513 mlog_errno(ret);
514 goto out;
515 }
516 if (ret == 1)
517 *is_last = 1;
518 }
519 }
520 }
521
522out_hole:
523 ret = 0;
524out:
525 brelse(eb_bh);
526 return ret;
527}
528
529static void ocfs2_relative_extent_offsets(struct super_block *sb,
530 u32 v_cluster,
531 struct ocfs2_extent_rec *rec,
532 u32 *p_cluster, u32 *num_clusters)
533
534{
535 u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
536
537 *p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
538 *p_cluster = *p_cluster + coff;
539
540 if (num_clusters)
541 *num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
542}
543
Tao Maf56654c2008-08-18 17:38:48 +0800544int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
545 u32 *p_cluster, u32 *num_clusters,
Tao Ma1061f9c2009-08-18 11:41:57 +0800546 struct ocfs2_extent_list *el,
547 unsigned int *extent_flags)
Tao Maf56654c2008-08-18 17:38:48 +0800548{
549 int ret = 0, i;
550 struct buffer_head *eb_bh = NULL;
551 struct ocfs2_extent_block *eb;
552 struct ocfs2_extent_rec *rec;
553 u32 coff;
554
555 if (el->l_tree_depth) {
Joel Beckerfacdb772009-02-12 18:08:48 -0800556 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
557 &eb_bh);
Tao Maf56654c2008-08-18 17:38:48 +0800558 if (ret) {
559 mlog_errno(ret);
560 goto out;
561 }
562
563 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
564 el = &eb->h_list;
565
566 if (el->l_tree_depth) {
567 ocfs2_error(inode->i_sb,
568 "Inode %lu has non zero tree depth in "
569 "xattr leaf block %llu\n", inode->i_ino,
570 (unsigned long long)eb_bh->b_blocknr);
571 ret = -EROFS;
572 goto out;
573 }
574 }
575
576 i = ocfs2_search_extent_list(el, v_cluster);
577 if (i == -1) {
578 ret = -EROFS;
579 mlog_errno(ret);
580 goto out;
581 } else {
582 rec = &el->l_recs[i];
583 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
584
585 if (!rec->e_blkno) {
586 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
587 "record (%u, %u, 0) in xattr", inode->i_ino,
588 le32_to_cpu(rec->e_cpos),
589 ocfs2_rec_clusters(el, rec));
590 ret = -EROFS;
591 goto out;
592 }
593 coff = v_cluster - le32_to_cpu(rec->e_cpos);
594 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
595 le64_to_cpu(rec->e_blkno));
596 *p_cluster = *p_cluster + coff;
597 if (num_clusters)
598 *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
Tao Ma1061f9c2009-08-18 11:41:57 +0800599
600 if (extent_flags)
601 *extent_flags = rec->e_flags;
Tao Maf56654c2008-08-18 17:38:48 +0800602 }
603out:
604 if (eb_bh)
605 brelse(eb_bh);
606 return ret;
607}
608
Mark Fasheh00dc4172008-10-03 17:32:11 -0400609int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
610 u32 *p_cluster, u32 *num_clusters,
611 unsigned int *extent_flags)
612{
613 int ret;
614 unsigned int uninitialized_var(hole_len), flags = 0;
615 struct buffer_head *di_bh = NULL;
616 struct ocfs2_extent_rec rec;
617
618 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
619 ret = -ERANGE;
620 mlog_errno(ret);
621 goto out;
622 }
623
624 ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
625 num_clusters, extent_flags);
626 if (ret == 0)
627 goto out;
628
Joel Beckerb657c952008-11-13 14:49:11 -0800629 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh00dc4172008-10-03 17:32:11 -0400630 if (ret) {
631 mlog_errno(ret);
632 goto out;
633 }
634
635 ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
636 &rec, NULL);
637 if (ret) {
638 mlog_errno(ret);
639 goto out;
640 }
641
642 if (rec.e_blkno == 0ULL) {
643 /*
Mark Fasheh363041a2007-01-17 12:31:35 -0800644 * A hole was found. Return some canned values that
Mark Fasheh4f902c32007-03-09 16:26:50 -0800645 * callers can key on. If asked for, num_clusters will
646 * be populated with the size of the hole.
Mark Fasheh363041a2007-01-17 12:31:35 -0800647 */
648 *p_cluster = 0;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800649 if (num_clusters) {
Mark Fasheh00dc4172008-10-03 17:32:11 -0400650 *num_clusters = hole_len;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800651 }
Mark Fasheh363041a2007-01-17 12:31:35 -0800652 } else {
Mark Fasheh00dc4172008-10-03 17:32:11 -0400653 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
654 p_cluster, num_clusters);
655 flags = rec.e_flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800656
Mark Fasheh00dc4172008-10-03 17:32:11 -0400657 ocfs2_extent_map_insert_rec(inode, &rec);
Mark Fasheh363041a2007-01-17 12:31:35 -0800658 }
659
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800660 if (extent_flags)
661 *extent_flags = flags;
662
Mark Fasheh363041a2007-01-17 12:31:35 -0800663out:
664 brelse(di_bh);
Mark Fasheh363041a2007-01-17 12:31:35 -0800665 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800666}
667
668/*
Mark Fasheh363041a2007-01-17 12:31:35 -0800669 * This expects alloc_sem to be held. The allocation cannot change at
670 * all while the map is in the process of being updated.
Mark Fashehccd979b2005-12-15 14:31:24 -0800671 */
Mark Fasheh363041a2007-01-17 12:31:35 -0800672int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
Mark Fasheh4f902c32007-03-09 16:26:50 -0800673 u64 *ret_count, unsigned int *extent_flags)
Mark Fashehccd979b2005-12-15 14:31:24 -0800674{
675 int ret;
Mark Fasheh363041a2007-01-17 12:31:35 -0800676 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
677 u32 cpos, num_clusters, p_cluster;
678 u64 boff = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800679
Mark Fasheh363041a2007-01-17 12:31:35 -0800680 cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800681
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800682 ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
683 extent_flags);
Mark Fasheh363041a2007-01-17 12:31:35 -0800684 if (ret) {
685 mlog_errno(ret);
686 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800687 }
688
689 /*
Mark Fasheh363041a2007-01-17 12:31:35 -0800690 * p_cluster == 0 indicates a hole.
Mark Fashehccd979b2005-12-15 14:31:24 -0800691 */
Mark Fasheh363041a2007-01-17 12:31:35 -0800692 if (p_cluster) {
693 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
Mark Fashehccd979b2005-12-15 14:31:24 -0800694 boff += (v_blkno & (u64)(bpc - 1));
Mark Fashehccd979b2005-12-15 14:31:24 -0800695 }
696
Mark Fasheh363041a2007-01-17 12:31:35 -0800697 *p_blkno = boff;
Mark Fashehccd979b2005-12-15 14:31:24 -0800698
Mark Fasheh363041a2007-01-17 12:31:35 -0800699 if (ret_count) {
700 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
701 *ret_count -= v_blkno & (u64)(bpc - 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800702 }
703
Mark Fasheh363041a2007-01-17 12:31:35 -0800704out:
705 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800706}
Mark Fasheh00dc4172008-10-03 17:32:11 -0400707
Tristan Ye86239d52009-12-22 09:11:58 +0800708/*
709 * The ocfs2_fiemap_inline() may be a little bit misleading, since
710 * it not only handles the fiemap for inlined files, but also deals
711 * with the fast symlink, cause they have no difference for extent
712 * mapping per se.
713 */
Mark Fasheh00dc4172008-10-03 17:32:11 -0400714static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
715 struct fiemap_extent_info *fieinfo,
716 u64 map_start)
717{
718 int ret;
719 unsigned int id_count;
720 struct ocfs2_dinode *di;
721 u64 phys;
722 u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
723 struct ocfs2_inode_info *oi = OCFS2_I(inode);
724
725 di = (struct ocfs2_dinode *)di_bh->b_data;
Tristan Ye86239d52009-12-22 09:11:58 +0800726 if (ocfs2_inode_is_fast_symlink(inode))
727 id_count = ocfs2_fast_symlink_chars(inode->i_sb);
728 else
729 id_count = le16_to_cpu(di->id2.i_data.id_count);
Mark Fasheh00dc4172008-10-03 17:32:11 -0400730
731 if (map_start < id_count) {
732 phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
Tristan Ye86239d52009-12-22 09:11:58 +0800733 if (ocfs2_inode_is_fast_symlink(inode))
734 phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
735 else
736 phys += offsetof(struct ocfs2_dinode,
737 id2.i_data.id_data);
Mark Fasheh00dc4172008-10-03 17:32:11 -0400738
739 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
740 flags);
741 if (ret < 0)
742 return ret;
743 }
744
745 return 0;
746}
747
748#define OCFS2_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
749
750int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
751 u64 map_start, u64 map_len)
752{
753 int ret, is_last;
754 u32 mapping_end, cpos;
755 unsigned int hole_size;
756 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
757 u64 len_bytes, phys_bytes, virt_bytes;
758 struct buffer_head *di_bh = NULL;
759 struct ocfs2_extent_rec rec;
760
761 ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
762 if (ret)
763 return ret;
764
765 ret = ocfs2_inode_lock(inode, &di_bh, 0);
766 if (ret) {
767 mlog_errno(ret);
768 goto out;
769 }
770
771 down_read(&OCFS2_I(inode)->ip_alloc_sem);
772
773 /*
Tristan Ye86239d52009-12-22 09:11:58 +0800774 * Handle inline-data and fast symlink separately.
Mark Fasheh00dc4172008-10-03 17:32:11 -0400775 */
Tristan Ye86239d52009-12-22 09:11:58 +0800776 if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
777 ocfs2_inode_is_fast_symlink(inode)) {
Mark Fasheh00dc4172008-10-03 17:32:11 -0400778 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
779 goto out_unlock;
780 }
781
782 cpos = map_start >> osb->s_clustersize_bits;
783 mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
784 map_start + map_len);
785 mapping_end -= cpos;
786 is_last = 0;
787 while (cpos < mapping_end && !is_last) {
788 u32 fe_flags;
789
790 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
791 &hole_size, &rec, &is_last);
792 if (ret) {
793 mlog_errno(ret);
794 goto out;
795 }
796
797 if (rec.e_blkno == 0ULL) {
798 cpos += hole_size;
799 continue;
800 }
801
802 fe_flags = 0;
803 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
804 fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
Sunil Mushranfaf8b702009-12-03 12:46:52 -0800805 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
806 fe_flags |= FIEMAP_EXTENT_SHARED;
Mark Fasheh00dc4172008-10-03 17:32:11 -0400807 if (is_last)
808 fe_flags |= FIEMAP_EXTENT_LAST;
809 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
810 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
811 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
812
813 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
814 len_bytes, fe_flags);
815 if (ret)
816 break;
817
818 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
819 }
820
821 if (ret > 0)
822 ret = 0;
823
824out_unlock:
825 brelse(di_bh);
826
827 up_read(&OCFS2_I(inode)->ip_alloc_sem);
828
829 ocfs2_inode_unlock(inode, 0);
830out:
831
832 return ret;
833}
Joel Beckera8549fb2008-11-13 14:49:20 -0800834
Sunil Mushran93862d52011-07-25 14:58:15 -0700835int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int origin)
836{
837 struct inode *inode = file->f_mapping->host;
838 int ret;
839 unsigned int is_last = 0, is_data = 0;
840 u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
841 u32 cpos, cend, clen, hole_size;
842 u64 extoff, extlen;
843 struct buffer_head *di_bh = NULL;
844 struct ocfs2_extent_rec rec;
845
846 BUG_ON(origin != SEEK_DATA && origin != SEEK_HOLE);
847
848 ret = ocfs2_inode_lock(inode, &di_bh, 0);
849 if (ret) {
850 mlog_errno(ret);
851 goto out;
852 }
853
854 down_read(&OCFS2_I(inode)->ip_alloc_sem);
855
856 if (*offset >= inode->i_size) {
857 ret = -ENXIO;
858 goto out_unlock;
859 }
860
861 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
862 if (origin == SEEK_HOLE)
863 *offset = inode->i_size;
864 goto out_unlock;
865 }
866
867 clen = 0;
868 cpos = *offset >> cs_bits;
869 cend = ocfs2_clusters_for_bytes(inode->i_sb, inode->i_size);
870
871 while (cpos < cend && !is_last) {
872 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
873 &rec, &is_last);
874 if (ret) {
875 mlog_errno(ret);
876 goto out_unlock;
877 }
878
879 extoff = cpos;
880 extoff <<= cs_bits;
881
882 if (rec.e_blkno == 0ULL) {
883 clen = hole_size;
884 is_data = 0;
885 } else {
886 clen = le16_to_cpu(rec.e_leaf_clusters) -
887 (cpos - le32_to_cpu(rec.e_cpos));
888 is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ? 0 : 1;
889 }
890
891 if ((!is_data && origin == SEEK_HOLE) ||
892 (is_data && origin == SEEK_DATA)) {
893 if (extoff > *offset)
894 *offset = extoff;
895 goto out_unlock;
896 }
897
898 if (!is_last)
899 cpos += clen;
900 }
901
902 if (origin == SEEK_HOLE) {
903 extoff = cpos;
904 extoff <<= cs_bits;
905 extlen = clen;
906 extlen <<= cs_bits;
907
908 if ((extoff + extlen) > inode->i_size)
909 extlen = inode->i_size - extoff;
910 extoff += extlen;
911 if (extoff > *offset)
912 *offset = extoff;
913 goto out_unlock;
914 }
915
916 ret = -ENXIO;
917
918out_unlock:
919
920 brelse(di_bh);
921
922 up_read(&OCFS2_I(inode)->ip_alloc_sem);
923
924 ocfs2_inode_unlock(inode, 0);
925out:
Sunil Mushran93862d52011-07-25 14:58:15 -0700926 return ret;
927}
928
Joel Beckera8549fb2008-11-13 14:49:20 -0800929int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
930 struct buffer_head *bhs[], int flags,
931 int (*validate)(struct super_block *sb,
932 struct buffer_head *bh))
933{
934 int rc = 0;
935 u64 p_block, p_count;
936 int i, count, done = 0;
937
Tao Maa7163572011-02-22 22:27:33 +0800938 trace_ocfs2_read_virt_blocks(
Tao Maef6b6892011-02-21 11:10:44 +0800939 inode, (unsigned long long)v_block, nr, bhs, flags,
940 validate);
Joel Beckera8549fb2008-11-13 14:49:20 -0800941
942 if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
943 i_size_read(inode)) {
944 BUG_ON(!(flags & OCFS2_BH_READAHEAD));
945 goto out;
946 }
947
948 while (done < nr) {
949 down_read(&OCFS2_I(inode)->ip_alloc_sem);
950 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
951 &p_block, &p_count, NULL);
952 up_read(&OCFS2_I(inode)->ip_alloc_sem);
953 if (rc) {
954 mlog_errno(rc);
955 break;
956 }
957
958 if (!p_block) {
959 rc = -EIO;
960 mlog(ML_ERROR,
961 "Inode #%llu contains a hole at offset %llu\n",
962 (unsigned long long)OCFS2_I(inode)->ip_blkno,
963 (unsigned long long)(v_block + done) <<
964 inode->i_sb->s_blocksize_bits);
965 break;
966 }
967
968 count = nr - done;
969 if (p_count < count)
970 count = p_count;
971
972 /*
973 * If the caller passed us bhs, they should have come
974 * from a previous readahead call to this function. Thus,
975 * they should have the right b_blocknr.
976 */
977 for (i = 0; i < count; i++) {
978 if (!bhs[done + i])
979 continue;
980 BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
981 }
982
Joel Becker8cb471e2009-02-10 20:00:41 -0800983 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
984 bhs + done, flags, validate);
Joel Beckera8549fb2008-11-13 14:49:20 -0800985 if (rc) {
986 mlog_errno(rc);
987 break;
988 }
989 done += count;
990 }
991
992out:
Joel Beckera8549fb2008-11-13 14:49:20 -0800993 return rc;
994}
995
996