blob: d7a6f928c31726731ee0d58a839f5b6daf427676 [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 * suballoc.c
5 *
6 * metadata alloc and free
7 * Inspired by ext3 block groups.
8 *
9 * Copyright (C) 2002, 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 as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public
22 * License along with this program; if not, write to the
23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24 * Boston, MA 021110-1307, USA.
25 */
26
27#include <linux/fs.h>
28#include <linux/types.h>
29#include <linux/slab.h>
30#include <linux/highmem.h>
31
32#define MLOG_MASK_PREFIX ML_DISK_ALLOC
33#include <cluster/masklog.h>
34
35#include "ocfs2.h"
36
37#include "alloc.h"
38#include "dlmglue.h"
39#include "inode.h"
40#include "journal.h"
41#include "localalloc.h"
42#include "suballoc.h"
43#include "super.h"
44#include "sysfile.h"
45#include "uptodate.h"
46
47#include "buffer_head_io.h"
48
Tao Maffda89a2008-03-03 17:12:09 +080049#define NOT_ALLOC_NEW_GROUP 0
50#define ALLOC_NEW_GROUP 1
51
Tao Ma4d0ddb22008-03-05 16:11:46 +080052#define OCFS2_MAX_INODES_TO_STEAL 1024
53
Mark Fashehccd979b2005-12-15 14:31:24 -080054static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
55static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
56static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
Mark Fasheh1fabe142006-10-09 18:11:45 -070057static int ocfs2_block_group_fill(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -080058 struct inode *alloc_inode,
59 struct buffer_head *bg_bh,
60 u64 group_blkno,
61 u16 my_chain,
62 struct ocfs2_chain_list *cl);
63static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
64 struct inode *alloc_inode,
Joel Becker1187c962008-09-03 20:03:39 -070065 struct buffer_head *bh,
66 u64 max_block);
Mark Fashehccd979b2005-12-15 14:31:24 -080067
Mark Fashehccd979b2005-12-15 14:31:24 -080068static int ocfs2_cluster_group_search(struct inode *inode,
69 struct buffer_head *group_bh,
70 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -070071 u64 max_block,
Mark Fashehccd979b2005-12-15 14:31:24 -080072 u16 *bit_off, u16 *bits_found);
73static int ocfs2_block_group_search(struct inode *inode,
74 struct buffer_head *group_bh,
75 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -070076 u64 max_block,
Mark Fashehccd979b2005-12-15 14:31:24 -080077 u16 *bit_off, u16 *bits_found);
Mark Fashehccd979b2005-12-15 14:31:24 -080078static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
79 struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -070080 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -080081 u32 bits_wanted,
82 u32 min_bits,
83 u16 *bit_off,
84 unsigned int *num_bits,
85 u64 *bg_blkno);
86static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
87 int nr);
Mark Fasheh1fabe142006-10-09 18:11:45 -070088static inline int ocfs2_block_group_set_bits(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -080089 struct inode *alloc_inode,
90 struct ocfs2_group_desc *bg,
91 struct buffer_head *group_bh,
92 unsigned int bit_off,
93 unsigned int num_bits);
Mark Fasheh1fabe142006-10-09 18:11:45 -070094static inline int ocfs2_block_group_clear_bits(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -080095 struct inode *alloc_inode,
96 struct ocfs2_group_desc *bg,
97 struct buffer_head *group_bh,
98 unsigned int bit_off,
99 unsigned int num_bits);
100
Mark Fasheh1fabe142006-10-09 18:11:45 -0700101static int ocfs2_relink_block_group(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800102 struct inode *alloc_inode,
103 struct buffer_head *fe_bh,
104 struct buffer_head *bg_bh,
105 struct buffer_head *prev_bg_bh,
106 u16 chain);
107static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
108 u32 wanted);
Mark Fashehccd979b2005-12-15 14:31:24 -0800109static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
110 u64 bg_blkno,
111 u16 bg_bit_off);
Mark Fashehccd979b2005-12-15 14:31:24 -0800112static inline void ocfs2_block_to_cluster_group(struct inode *inode,
113 u64 data_blkno,
114 u64 *bg_blkno,
115 u16 *bg_bit_off);
Joel Becker1187c962008-09-03 20:03:39 -0700116static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
117 u32 bits_wanted, u64 max_block,
118 struct ocfs2_alloc_context **ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800119
Mark Fasheh9c7af402008-07-28 18:02:53 -0700120void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
Mark Fashehccd979b2005-12-15 14:31:24 -0800121{
Mark Fashehda5cbf22006-10-06 18:34:35 -0700122 struct inode *inode = ac->ac_inode;
123
124 if (inode) {
125 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
Mark Fashehe63aecb62007-10-18 15:30:42 -0700126 ocfs2_inode_unlock(inode, 1);
Mark Fashehda5cbf22006-10-06 18:34:35 -0700127
128 mutex_unlock(&inode->i_mutex);
129
130 iput(inode);
Tao Ma4d0ddb22008-03-05 16:11:46 +0800131 ac->ac_inode = NULL;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700132 }
Tao Ma4d0ddb22008-03-05 16:11:46 +0800133 if (ac->ac_bh) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800134 brelse(ac->ac_bh);
Tao Ma4d0ddb22008-03-05 16:11:46 +0800135 ac->ac_bh = NULL;
136 }
137}
138
139void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
140{
141 ocfs2_free_ac_resource(ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800142 kfree(ac);
143}
144
145static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
146{
147 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
148}
149
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700150/* somewhat more expensive than our other checks, so use sparingly. */
Tao Mad6590722007-12-18 15:47:03 +0800151int ocfs2_check_group_descriptor(struct super_block *sb,
152 struct ocfs2_dinode *di,
153 struct ocfs2_group_desc *gd)
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700154{
155 unsigned int max_bits;
156
157 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
158 OCFS2_RO_ON_INVALID_GROUP_DESC(sb, gd);
159 return -EIO;
160 }
161
162 if (di->i_blkno != gd->bg_parent_dinode) {
163 ocfs2_error(sb, "Group descriptor # %llu has bad parent "
164 "pointer (%llu, expected %llu)",
165 (unsigned long long)le64_to_cpu(gd->bg_blkno),
166 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
167 (unsigned long long)le64_to_cpu(di->i_blkno));
168 return -EIO;
169 }
170
171 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
172 if (le16_to_cpu(gd->bg_bits) > max_bits) {
173 ocfs2_error(sb, "Group descriptor # %llu has bit count of %u",
174 (unsigned long long)le64_to_cpu(gd->bg_blkno),
175 le16_to_cpu(gd->bg_bits));
176 return -EIO;
177 }
178
179 if (le16_to_cpu(gd->bg_chain) >=
180 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
181 ocfs2_error(sb, "Group descriptor # %llu has bad chain %u",
182 (unsigned long long)le64_to_cpu(gd->bg_blkno),
183 le16_to_cpu(gd->bg_chain));
184 return -EIO;
185 }
186
187 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
188 ocfs2_error(sb, "Group descriptor # %llu has bit count %u but "
189 "claims that %u are free",
190 (unsigned long long)le64_to_cpu(gd->bg_blkno),
191 le16_to_cpu(gd->bg_bits),
192 le16_to_cpu(gd->bg_free_bits_count));
193 return -EIO;
194 }
195
196 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
197 ocfs2_error(sb, "Group descriptor # %llu has bit count %u but "
198 "max bitmap bits of %u",
199 (unsigned long long)le64_to_cpu(gd->bg_blkno),
200 le16_to_cpu(gd->bg_bits),
201 8 * le16_to_cpu(gd->bg_size));
202 return -EIO;
203 }
204
205 return 0;
206}
207
Mark Fasheh1fabe142006-10-09 18:11:45 -0700208static int ocfs2_block_group_fill(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800209 struct inode *alloc_inode,
210 struct buffer_head *bg_bh,
211 u64 group_blkno,
212 u16 my_chain,
213 struct ocfs2_chain_list *cl)
214{
215 int status = 0;
216 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
217 struct super_block * sb = alloc_inode->i_sb;
218
219 mlog_entry_void();
220
221 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
Mark Fashehb06970532006-03-03 10:24:33 -0800222 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
223 "b_blocknr (%llu)",
224 (unsigned long long)group_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800225 (unsigned long long) bg_bh->b_blocknr);
226 status = -EIO;
227 goto bail;
228 }
229
230 status = ocfs2_journal_access(handle,
231 alloc_inode,
232 bg_bh,
233 OCFS2_JOURNAL_ACCESS_CREATE);
234 if (status < 0) {
235 mlog_errno(status);
236 goto bail;
237 }
238
239 memset(bg, 0, sb->s_blocksize);
240 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
241 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
242 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
243 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
244 bg->bg_chain = cpu_to_le16(my_chain);
245 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
246 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
247 bg->bg_blkno = cpu_to_le64(group_blkno);
248 /* set the 1st bit in the bitmap to account for the descriptor block */
249 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
250 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
251
252 status = ocfs2_journal_dirty(handle, bg_bh);
253 if (status < 0)
254 mlog_errno(status);
255
256 /* There is no need to zero out or otherwise initialize the
257 * other blocks in a group - All valid FS metadata in a block
258 * group stores the superblock fs_generation value at
259 * allocation time. */
260
261bail:
262 mlog_exit(status);
263 return status;
264}
265
266static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
267{
268 u16 curr, best;
269
270 best = curr = 0;
271 while (curr < le16_to_cpu(cl->cl_count)) {
272 if (le32_to_cpu(cl->cl_recs[best].c_total) >
273 le32_to_cpu(cl->cl_recs[curr].c_total))
274 best = curr;
275 curr++;
276 }
277 return best;
278}
279
280/*
281 * We expect the block group allocator to already be locked.
282 */
283static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
284 struct inode *alloc_inode,
Joel Becker1187c962008-09-03 20:03:39 -0700285 struct buffer_head *bh,
286 u64 max_block)
Mark Fashehccd979b2005-12-15 14:31:24 -0800287{
288 int status, credits;
289 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
290 struct ocfs2_chain_list *cl;
291 struct ocfs2_alloc_context *ac = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700292 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800293 u32 bit_off, num_bits;
294 u16 alloc_rec;
295 u64 bg_blkno;
296 struct buffer_head *bg_bh = NULL;
297 struct ocfs2_group_desc *bg;
298
299 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
300
301 mlog_entry_void();
302
Mark Fashehccd979b2005-12-15 14:31:24 -0800303 cl = &fe->id2.i_chain;
Joel Becker1187c962008-09-03 20:03:39 -0700304 status = ocfs2_reserve_clusters_with_limit(osb,
305 le16_to_cpu(cl->cl_cpg),
306 max_block, &ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800307 if (status < 0) {
308 if (status != -ENOSPC)
309 mlog_errno(status);
310 goto bail;
311 }
312
313 credits = ocfs2_calc_group_alloc_credits(osb->sb,
314 le16_to_cpu(cl->cl_cpg));
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700315 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800316 if (IS_ERR(handle)) {
317 status = PTR_ERR(handle);
318 handle = NULL;
319 mlog_errno(status);
320 goto bail;
321 }
322
323 status = ocfs2_claim_clusters(osb,
324 handle,
325 ac,
326 le16_to_cpu(cl->cl_cpg),
327 &bit_off,
328 &num_bits);
329 if (status < 0) {
330 if (status != -ENOSPC)
331 mlog_errno(status);
332 goto bail;
333 }
334
335 alloc_rec = ocfs2_find_smallest_chain(cl);
336
337 /* setup the group */
338 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Mark Fashehb06970532006-03-03 10:24:33 -0800339 mlog(0, "new descriptor, record %u, at block %llu\n",
340 alloc_rec, (unsigned long long)bg_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800341
342 bg_bh = sb_getblk(osb->sb, bg_blkno);
343 if (!bg_bh) {
344 status = -EIO;
345 mlog_errno(status);
346 goto bail;
347 }
348 ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh);
349
350 status = ocfs2_block_group_fill(handle,
351 alloc_inode,
352 bg_bh,
353 bg_blkno,
354 alloc_rec,
355 cl);
356 if (status < 0) {
357 mlog_errno(status);
358 goto bail;
359 }
360
361 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
362
363 status = ocfs2_journal_access(handle, alloc_inode,
364 bh, OCFS2_JOURNAL_ACCESS_WRITE);
365 if (status < 0) {
366 mlog_errno(status);
367 goto bail;
368 }
369
370 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
371 le16_to_cpu(bg->bg_free_bits_count));
372 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
373 cl->cl_recs[alloc_rec].c_blkno = cpu_to_le64(bg_blkno);
374 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
375 le16_add_cpu(&cl->cl_next_free_rec, 1);
376
377 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
378 le16_to_cpu(bg->bg_free_bits_count));
379 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
380 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
381
382 status = ocfs2_journal_dirty(handle, bh);
383 if (status < 0) {
384 mlog_errno(status);
385 goto bail;
386 }
387
388 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
389 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
390 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
391 le32_to_cpu(fe->i_clusters)));
392 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
393 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
Mark Fasheh8110b072007-03-22 16:53:23 -0700394 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800395
396 status = 0;
397bail:
398 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700399 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800400
401 if (ac)
402 ocfs2_free_alloc_context(ac);
403
404 if (bg_bh)
405 brelse(bg_bh);
406
407 mlog_exit(status);
408 return status;
409}
410
411static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
Mark Fashehda5cbf22006-10-06 18:34:35 -0700412 struct ocfs2_alloc_context *ac,
413 int type,
Tao Maffda89a2008-03-03 17:12:09 +0800414 u32 slot,
415 int alloc_new_group)
Mark Fashehccd979b2005-12-15 14:31:24 -0800416{
417 int status;
418 u32 bits_wanted = ac->ac_bits_wanted;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700419 struct inode *alloc_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -0800420 struct buffer_head *bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800421 struct ocfs2_dinode *fe;
422 u32 free_bits;
423
424 mlog_entry_void();
425
Mark Fashehda5cbf22006-10-06 18:34:35 -0700426 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
427 if (!alloc_inode) {
428 mlog_errno(-EINVAL);
429 return -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800430 }
431
Mark Fashehda5cbf22006-10-06 18:34:35 -0700432 mutex_lock(&alloc_inode->i_mutex);
433
Mark Fashehe63aecb62007-10-18 15:30:42 -0700434 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
Mark Fashehda5cbf22006-10-06 18:34:35 -0700435 if (status < 0) {
436 mutex_unlock(&alloc_inode->i_mutex);
437 iput(alloc_inode);
438
439 mlog_errno(status);
440 return status;
441 }
442
443 ac->ac_inode = alloc_inode;
Tao Maa4a48912008-03-03 17:12:30 +0800444 ac->ac_alloc_slot = slot;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700445
Mark Fashehccd979b2005-12-15 14:31:24 -0800446 fe = (struct ocfs2_dinode *) bh->b_data;
447 if (!OCFS2_IS_VALID_DINODE(fe)) {
448 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
449 status = -EIO;
450 goto bail;
451 }
452 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
Mark Fashehb06970532006-03-03 10:24:33 -0800453 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
454 (unsigned long long)le64_to_cpu(fe->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800455 status = -EIO;
456 goto bail;
457 }
458
459 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
460 le32_to_cpu(fe->id1.bitmap1.i_used);
461
462 if (bits_wanted > free_bits) {
463 /* cluster bitmap never grows */
464 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
465 mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
466 bits_wanted, free_bits);
467 status = -ENOSPC;
468 goto bail;
469 }
470
Tao Maffda89a2008-03-03 17:12:09 +0800471 if (alloc_new_group != ALLOC_NEW_GROUP) {
472 mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
473 "and we don't alloc a new group for it.\n",
474 slot, bits_wanted, free_bits);
475 status = -ENOSPC;
476 goto bail;
477 }
478
Joel Becker1187c962008-09-03 20:03:39 -0700479 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
480 ac->ac_max_block);
Mark Fashehccd979b2005-12-15 14:31:24 -0800481 if (status < 0) {
482 if (status != -ENOSPC)
483 mlog_errno(status);
484 goto bail;
485 }
486 atomic_inc(&osb->alloc_stats.bg_extends);
487
488 /* You should never ask for this much metadata */
489 BUG_ON(bits_wanted >
490 (le32_to_cpu(fe->id1.bitmap1.i_total)
491 - le32_to_cpu(fe->id1.bitmap1.i_used)));
492 }
493
494 get_bh(bh);
495 ac->ac_bh = bh;
496bail:
497 if (bh)
498 brelse(bh);
499
500 mlog_exit(status);
501 return status;
502}
503
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800504int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
505 int blocks,
506 struct ocfs2_alloc_context **ac)
Mark Fashehccd979b2005-12-15 14:31:24 -0800507{
508 int status;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700509 u32 slot;
Mark Fashehccd979b2005-12-15 14:31:24 -0800510
Robert P. J. Daycd861282006-12-13 00:34:52 -0800511 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800512 if (!(*ac)) {
513 status = -ENOMEM;
514 mlog_errno(status);
515 goto bail;
516 }
517
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800518 (*ac)->ac_bits_wanted = blocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800519 (*ac)->ac_which = OCFS2_AC_USE_META;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700520 slot = osb->slot_num;
Mark Fashehccd979b2005-12-15 14:31:24 -0800521 (*ac)->ac_group_search = ocfs2_block_group_search;
522
Mark Fashehda5cbf22006-10-06 18:34:35 -0700523 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
Tao Maffda89a2008-03-03 17:12:09 +0800524 EXTENT_ALLOC_SYSTEM_INODE,
525 slot, ALLOC_NEW_GROUP);
Mark Fashehccd979b2005-12-15 14:31:24 -0800526 if (status < 0) {
527 if (status != -ENOSPC)
528 mlog_errno(status);
529 goto bail;
530 }
531
532 status = 0;
533bail:
534 if ((status < 0) && *ac) {
535 ocfs2_free_alloc_context(*ac);
536 *ac = NULL;
537 }
538
Mark Fashehccd979b2005-12-15 14:31:24 -0800539 mlog_exit(status);
540 return status;
541}
542
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800543int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
544 struct ocfs2_extent_list *root_el,
545 struct ocfs2_alloc_context **ac)
546{
547 return ocfs2_reserve_new_metadata_blocks(osb,
548 ocfs2_extend_meta_needed(root_el),
549 ac);
550}
551
Tao Ma4d0ddb22008-03-05 16:11:46 +0800552static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
553 struct ocfs2_alloc_context *ac)
554{
555 int i, status = -ENOSPC;
556 s16 slot = ocfs2_get_inode_steal_slot(osb);
557
558 /* Start to steal inodes from the first slot after ours. */
559 if (slot == OCFS2_INVALID_SLOT)
560 slot = osb->slot_num + 1;
561
562 for (i = 0; i < osb->max_slots; i++, slot++) {
563 if (slot == osb->max_slots)
564 slot = 0;
565
566 if (slot == osb->slot_num)
567 continue;
568
569 status = ocfs2_reserve_suballoc_bits(osb, ac,
570 INODE_ALLOC_SYSTEM_INODE,
571 slot, NOT_ALLOC_NEW_GROUP);
572 if (status >= 0) {
573 ocfs2_set_inode_steal_slot(osb, slot);
574 break;
575 }
576
577 ocfs2_free_ac_resource(ac);
578 }
579
580 return status;
581}
582
Mark Fashehccd979b2005-12-15 14:31:24 -0800583int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -0800584 struct ocfs2_alloc_context **ac)
585{
586 int status;
Tao Ma4d0ddb22008-03-05 16:11:46 +0800587 s16 slot = ocfs2_get_inode_steal_slot(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800588
Robert P. J. Daycd861282006-12-13 00:34:52 -0800589 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800590 if (!(*ac)) {
591 status = -ENOMEM;
592 mlog_errno(status);
593 goto bail;
594 }
595
596 (*ac)->ac_bits_wanted = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800597 (*ac)->ac_which = OCFS2_AC_USE_INODE;
598
Mark Fashehccd979b2005-12-15 14:31:24 -0800599 (*ac)->ac_group_search = ocfs2_block_group_search;
600
Tao Ma4d0ddb22008-03-05 16:11:46 +0800601 /*
Joel Becker1187c962008-09-03 20:03:39 -0700602 * stat(2) can't handle i_ino > 32bits, so we tell the
603 * lower levels not to allocate us a block group past that
Joel Becker12462f12008-09-03 20:03:40 -0700604 * limit. The 'inode64' mount option avoids this behavior.
Joel Becker1187c962008-09-03 20:03:39 -0700605 */
Joel Becker12462f12008-09-03 20:03:40 -0700606 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
607 (*ac)->ac_max_block = (u32)~0U;
Joel Becker1187c962008-09-03 20:03:39 -0700608
609 /*
Tao Ma4d0ddb22008-03-05 16:11:46 +0800610 * slot is set when we successfully steal inode from other nodes.
611 * It is reset in 3 places:
612 * 1. when we flush the truncate log
613 * 2. when we complete local alloc recovery.
614 * 3. when we successfully allocate from our own slot.
615 * After it is set, we will go on stealing inodes until we find the
616 * need to check our slots to see whether there is some space for us.
617 */
618 if (slot != OCFS2_INVALID_SLOT &&
619 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL)
620 goto inode_steal;
621
622 atomic_set(&osb->s_num_inodes_stolen, 0);
Mark Fashehda5cbf22006-10-06 18:34:35 -0700623 status = ocfs2_reserve_suballoc_bits(osb, *ac,
624 INODE_ALLOC_SYSTEM_INODE,
Tao Maffda89a2008-03-03 17:12:09 +0800625 osb->slot_num, ALLOC_NEW_GROUP);
Tao Ma4d0ddb22008-03-05 16:11:46 +0800626 if (status >= 0) {
627 status = 0;
628
629 /*
630 * Some inodes must be freed by us, so try to allocate
631 * from our own next time.
632 */
633 if (slot != OCFS2_INVALID_SLOT)
634 ocfs2_init_inode_steal_slot(osb);
635 goto bail;
636 } else if (status < 0 && status != -ENOSPC) {
637 mlog_errno(status);
638 goto bail;
639 }
640
641 ocfs2_free_ac_resource(*ac);
642
643inode_steal:
644 status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
645 atomic_inc(&osb->s_num_inodes_stolen);
Mark Fashehccd979b2005-12-15 14:31:24 -0800646 if (status < 0) {
647 if (status != -ENOSPC)
648 mlog_errno(status);
649 goto bail;
650 }
651
652 status = 0;
653bail:
654 if ((status < 0) && *ac) {
655 ocfs2_free_alloc_context(*ac);
656 *ac = NULL;
657 }
658
Mark Fashehccd979b2005-12-15 14:31:24 -0800659 mlog_exit(status);
660 return status;
661}
662
663/* local alloc code has to do the same thing, so rather than do this
664 * twice.. */
665int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
666 struct ocfs2_alloc_context *ac)
667{
668 int status;
669
Mark Fashehccd979b2005-12-15 14:31:24 -0800670 ac->ac_which = OCFS2_AC_USE_MAIN;
671 ac->ac_group_search = ocfs2_cluster_group_search;
672
Mark Fashehda5cbf22006-10-06 18:34:35 -0700673 status = ocfs2_reserve_suballoc_bits(osb, ac,
674 GLOBAL_BITMAP_SYSTEM_INODE,
Tao Maffda89a2008-03-03 17:12:09 +0800675 OCFS2_INVALID_SLOT,
676 ALLOC_NEW_GROUP);
Mark Fashehda5cbf22006-10-06 18:34:35 -0700677 if (status < 0 && status != -ENOSPC) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800678 mlog_errno(status);
Mark Fashehda5cbf22006-10-06 18:34:35 -0700679 goto bail;
680 }
681
Mark Fashehccd979b2005-12-15 14:31:24 -0800682bail:
683 return status;
684}
685
686/* Callers don't need to care which bitmap (local alloc or main) to
687 * use so we figure it out for them, but unfortunately this clutters
688 * things a bit. */
Joel Becker1187c962008-09-03 20:03:39 -0700689static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
690 u32 bits_wanted, u64 max_block,
691 struct ocfs2_alloc_context **ac)
Mark Fashehccd979b2005-12-15 14:31:24 -0800692{
693 int status;
694
695 mlog_entry_void();
696
Robert P. J. Daycd861282006-12-13 00:34:52 -0800697 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800698 if (!(*ac)) {
699 status = -ENOMEM;
700 mlog_errno(status);
701 goto bail;
702 }
703
704 (*ac)->ac_bits_wanted = bits_wanted;
Joel Becker1187c962008-09-03 20:03:39 -0700705 (*ac)->ac_max_block = max_block;
Mark Fashehccd979b2005-12-15 14:31:24 -0800706
707 status = -ENOSPC;
708 if (ocfs2_alloc_should_use_local(osb, bits_wanted)) {
709 status = ocfs2_reserve_local_alloc_bits(osb,
Mark Fashehccd979b2005-12-15 14:31:24 -0800710 bits_wanted,
711 *ac);
Joel Becker1187c962008-09-03 20:03:39 -0700712 if (status == -EFBIG) {
713 /* The local alloc window is outside ac_max_block.
714 * use the main bitmap. */
715 status = -ENOSPC;
716 } else if ((status < 0) && (status != -ENOSPC)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800717 mlog_errno(status);
718 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800719 }
720 }
721
722 if (status == -ENOSPC) {
723 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
724 if (status < 0) {
725 if (status != -ENOSPC)
726 mlog_errno(status);
727 goto bail;
728 }
729 }
730
731 status = 0;
732bail:
733 if ((status < 0) && *ac) {
734 ocfs2_free_alloc_context(*ac);
735 *ac = NULL;
736 }
737
738 mlog_exit(status);
739 return status;
740}
741
Joel Becker1187c962008-09-03 20:03:39 -0700742int ocfs2_reserve_clusters(struct ocfs2_super *osb,
743 u32 bits_wanted,
744 struct ocfs2_alloc_context **ac)
745{
746 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0, ac);
747}
748
Mark Fashehccd979b2005-12-15 14:31:24 -0800749/*
750 * More or less lifted from ext3. I'll leave their description below:
751 *
752 * "For ext3 allocations, we must not reuse any blocks which are
753 * allocated in the bitmap buffer's "last committed data" copy. This
754 * prevents deletes from freeing up the page for reuse until we have
755 * committed the delete transaction.
756 *
757 * If we didn't do this, then deleting something and reallocating it as
758 * data would allow the old block to be overwritten before the
759 * transaction committed (because we force data to disk before commit).
760 * This would lead to corruption if we crashed between overwriting the
761 * data and committing the delete.
762 *
763 * @@@ We may want to make this allocation behaviour conditional on
764 * data-writes at some point, and disable it for metadata allocations or
765 * sync-data inodes."
766 *
767 * Note: OCFS2 already does this differently for metadata vs data
Joe Perchesc78bad12008-02-03 17:33:42 +0200768 * allocations, as those bitmaps are separate and undo access is never
Mark Fashehccd979b2005-12-15 14:31:24 -0800769 * called on a metadata group descriptor.
770 */
771static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
772 int nr)
773{
774 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
775
776 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
777 return 0;
778 if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data)
779 return 1;
780
781 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
782 return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
783}
784
785static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
786 struct buffer_head *bg_bh,
787 unsigned int bits_wanted,
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700788 unsigned int total_bits,
Mark Fashehccd979b2005-12-15 14:31:24 -0800789 u16 *bit_off,
790 u16 *bits_found)
791{
792 void *bitmap;
793 u16 best_offset, best_size;
794 int offset, start, found, status = 0;
795 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
796
797 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
798 OCFS2_RO_ON_INVALID_GROUP_DESC(osb->sb, bg);
799 return -EIO;
800 }
801
802 found = start = best_offset = best_size = 0;
803 bitmap = bg->bg_bitmap;
804
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700805 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
806 if (offset == total_bits)
Mark Fashehccd979b2005-12-15 14:31:24 -0800807 break;
808
809 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
810 /* We found a zero, but we can't use it as it
811 * hasn't been put to disk yet! */
812 found = 0;
813 start = offset + 1;
814 } else if (offset == start) {
815 /* we found a zero */
816 found++;
817 /* move start to the next bit to test */
818 start++;
819 } else {
820 /* got a zero after some ones */
821 found = 1;
822 start = offset + 1;
823 }
824 if (found > best_size) {
825 best_size = found;
826 best_offset = start - found;
827 }
828 /* we got everything we needed */
829 if (found == bits_wanted) {
830 /* mlog(0, "Found it all!\n"); */
831 break;
832 }
833 }
834
835 /* XXX: I think the first clause is equivalent to the second
836 * - jlbec */
837 if (found == bits_wanted) {
838 *bit_off = start - found;
839 *bits_found = found;
840 } else if (best_size) {
841 *bit_off = best_offset;
842 *bits_found = best_size;
843 } else {
844 status = -ENOSPC;
845 /* No error log here -- see the comment above
846 * ocfs2_test_bg_bit_allocatable */
847 }
848
849 return status;
850}
851
Mark Fasheh1fabe142006-10-09 18:11:45 -0700852static inline int ocfs2_block_group_set_bits(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800853 struct inode *alloc_inode,
854 struct ocfs2_group_desc *bg,
855 struct buffer_head *group_bh,
856 unsigned int bit_off,
857 unsigned int num_bits)
858{
859 int status;
860 void *bitmap = bg->bg_bitmap;
861 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
862
863 mlog_entry_void();
864
865 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
866 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
867 status = -EIO;
868 goto bail;
869 }
870 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
871
872 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
873 num_bits);
874
875 if (ocfs2_is_cluster_bitmap(alloc_inode))
876 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
877
878 status = ocfs2_journal_access(handle,
879 alloc_inode,
880 group_bh,
881 journal_type);
882 if (status < 0) {
883 mlog_errno(status);
884 goto bail;
885 }
886
887 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
888
889 while(num_bits--)
890 ocfs2_set_bit(bit_off++, bitmap);
891
892 status = ocfs2_journal_dirty(handle,
893 group_bh);
894 if (status < 0) {
895 mlog_errno(status);
896 goto bail;
897 }
898
899bail:
900 mlog_exit(status);
901 return status;
902}
903
904/* find the one with the most empty bits */
905static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
906{
907 u16 curr, best;
908
909 BUG_ON(!cl->cl_next_free_rec);
910
911 best = curr = 0;
912 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
913 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
914 le32_to_cpu(cl->cl_recs[best].c_free))
915 best = curr;
916 curr++;
917 }
918
919 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
920 return best;
921}
922
Mark Fasheh1fabe142006-10-09 18:11:45 -0700923static int ocfs2_relink_block_group(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800924 struct inode *alloc_inode,
925 struct buffer_head *fe_bh,
926 struct buffer_head *bg_bh,
927 struct buffer_head *prev_bg_bh,
928 u16 chain)
929{
930 int status;
931 /* there is a really tiny chance the journal calls could fail,
932 * but we wouldn't want inconsistent blocks in *any* case. */
933 u64 fe_ptr, bg_ptr, prev_bg_ptr;
934 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
935 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
936 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
937
938 if (!OCFS2_IS_VALID_DINODE(fe)) {
939 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
940 status = -EIO;
941 goto out;
942 }
943 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
944 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
945 status = -EIO;
946 goto out;
947 }
948 if (!OCFS2_IS_VALID_GROUP_DESC(prev_bg)) {
949 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, prev_bg);
950 status = -EIO;
951 goto out;
952 }
953
Mark Fashehb06970532006-03-03 10:24:33 -0800954 mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700955 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
956 (unsigned long long)le64_to_cpu(bg->bg_blkno),
957 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800958
959 fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
960 bg_ptr = le64_to_cpu(bg->bg_next_group);
961 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
962
963 status = ocfs2_journal_access(handle, alloc_inode, prev_bg_bh,
964 OCFS2_JOURNAL_ACCESS_WRITE);
965 if (status < 0) {
966 mlog_errno(status);
967 goto out_rollback;
968 }
969
970 prev_bg->bg_next_group = bg->bg_next_group;
971
972 status = ocfs2_journal_dirty(handle, prev_bg_bh);
973 if (status < 0) {
974 mlog_errno(status);
975 goto out_rollback;
976 }
977
978 status = ocfs2_journal_access(handle, alloc_inode, bg_bh,
979 OCFS2_JOURNAL_ACCESS_WRITE);
980 if (status < 0) {
981 mlog_errno(status);
982 goto out_rollback;
983 }
984
985 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
986
987 status = ocfs2_journal_dirty(handle, bg_bh);
988 if (status < 0) {
989 mlog_errno(status);
990 goto out_rollback;
991 }
992
993 status = ocfs2_journal_access(handle, alloc_inode, fe_bh,
994 OCFS2_JOURNAL_ACCESS_WRITE);
995 if (status < 0) {
996 mlog_errno(status);
997 goto out_rollback;
998 }
999
1000 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1001
1002 status = ocfs2_journal_dirty(handle, fe_bh);
1003 if (status < 0) {
1004 mlog_errno(status);
1005 goto out_rollback;
1006 }
1007
1008 status = 0;
1009out_rollback:
1010 if (status < 0) {
1011 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1012 bg->bg_next_group = cpu_to_le64(bg_ptr);
1013 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1014 }
1015out:
1016 mlog_exit(status);
1017 return status;
1018}
1019
1020static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1021 u32 wanted)
1022{
1023 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1024}
1025
1026/* return 0 on success, -ENOSPC to keep searching and any other < 0
1027 * value on error. */
1028static int ocfs2_cluster_group_search(struct inode *inode,
1029 struct buffer_head *group_bh,
1030 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -07001031 u64 max_block,
Mark Fashehccd979b2005-12-15 14:31:24 -08001032 u16 *bit_off, u16 *bits_found)
1033{
1034 int search = -ENOSPC;
1035 int ret;
Joel Becker1187c962008-09-03 20:03:39 -07001036 u64 blkoff;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001037 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fasheh9c7af402008-07-28 18:02:53 -07001038 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001039 u16 tmp_off, tmp_found;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001040 unsigned int max_bits, gd_cluster_off;
Mark Fashehccd979b2005-12-15 14:31:24 -08001041
1042 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1043
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001044 if (gd->bg_free_bits_count) {
1045 max_bits = le16_to_cpu(gd->bg_bits);
1046
1047 /* Tail groups in cluster bitmaps which aren't cpg
1048 * aligned are prone to partial extention by a failed
1049 * fs resize. If the file system resize never got to
1050 * update the dinode cluster count, then we don't want
1051 * to trust any clusters past it, regardless of what
1052 * the group descriptor says. */
1053 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1054 le64_to_cpu(gd->bg_blkno));
1055 if ((gd_cluster_off + max_bits) >
1056 OCFS2_I(inode)->ip_clusters) {
1057 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1058 mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1059 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1060 le16_to_cpu(gd->bg_bits),
1061 OCFS2_I(inode)->ip_clusters, max_bits);
1062 }
1063
Mark Fashehccd979b2005-12-15 14:31:24 -08001064 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1065 group_bh, bits_wanted,
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001066 max_bits,
Mark Fashehccd979b2005-12-15 14:31:24 -08001067 &tmp_off, &tmp_found);
1068 if (ret)
1069 return ret;
1070
Joel Becker1187c962008-09-03 20:03:39 -07001071 if (max_block) {
1072 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1073 gd_cluster_off +
1074 tmp_off + tmp_found);
1075 mlog(0, "Checking %llu against %llu\n",
1076 (unsigned long long)blkoff,
1077 (unsigned long long)max_block);
1078 if (blkoff > max_block)
1079 return -ENOSPC;
1080 }
1081
Mark Fashehccd979b2005-12-15 14:31:24 -08001082 /* ocfs2_block_group_find_clear_bits() might
1083 * return success, but we still want to return
1084 * -ENOSPC unless it found the minimum number
1085 * of bits. */
1086 if (min_bits <= tmp_found) {
1087 *bit_off = tmp_off;
1088 *bits_found = tmp_found;
1089 search = 0; /* success */
Mark Fasheh9c7af402008-07-28 18:02:53 -07001090 } else if (tmp_found) {
1091 /*
1092 * Don't show bits which we'll be returning
1093 * for allocation to the local alloc bitmap.
1094 */
1095 ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
Mark Fashehccd979b2005-12-15 14:31:24 -08001096 }
1097 }
1098
1099 return search;
1100}
1101
1102static int ocfs2_block_group_search(struct inode *inode,
1103 struct buffer_head *group_bh,
1104 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -07001105 u64 max_block,
Mark Fashehccd979b2005-12-15 14:31:24 -08001106 u16 *bit_off, u16 *bits_found)
1107{
1108 int ret = -ENOSPC;
Joel Becker1187c962008-09-03 20:03:39 -07001109 u64 blkoff;
Mark Fashehccd979b2005-12-15 14:31:24 -08001110 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1111
1112 BUG_ON(min_bits != 1);
1113 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1114
Joel Becker1187c962008-09-03 20:03:39 -07001115 if (bg->bg_free_bits_count) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001116 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1117 group_bh, bits_wanted,
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001118 le16_to_cpu(bg->bg_bits),
Mark Fashehccd979b2005-12-15 14:31:24 -08001119 bit_off, bits_found);
Joel Becker1187c962008-09-03 20:03:39 -07001120 if (!ret && max_block) {
1121 blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1122 *bits_found;
1123 mlog(0, "Checking %llu against %llu\n",
1124 (unsigned long long)blkoff,
1125 (unsigned long long)max_block);
1126 if (blkoff > max_block)
1127 ret = -ENOSPC;
1128 }
1129 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001130
1131 return ret;
1132}
1133
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001134static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001135 handle_t *handle,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001136 struct buffer_head *di_bh,
1137 u32 num_bits,
1138 u16 chain)
1139{
1140 int ret;
1141 u32 tmp_used;
1142 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1143 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1144
1145 ret = ocfs2_journal_access(handle, inode, di_bh,
1146 OCFS2_JOURNAL_ACCESS_WRITE);
1147 if (ret < 0) {
1148 mlog_errno(ret);
1149 goto out;
1150 }
1151
1152 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1153 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1154 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1155
1156 ret = ocfs2_journal_dirty(handle, di_bh);
1157 if (ret < 0)
1158 mlog_errno(ret);
1159
1160out:
1161 return ret;
1162}
1163
1164static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001165 handle_t *handle,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001166 u32 bits_wanted,
1167 u32 min_bits,
1168 u16 *bit_off,
1169 unsigned int *num_bits,
1170 u64 gd_blkno,
1171 u16 *bits_left)
1172{
1173 int ret;
1174 u16 found;
1175 struct buffer_head *group_bh = NULL;
1176 struct ocfs2_group_desc *gd;
1177 struct inode *alloc_inode = ac->ac_inode;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001178
1179 ret = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb), gd_blkno,
1180 &group_bh, OCFS2_BH_CACHED, alloc_inode);
1181 if (ret < 0) {
1182 mlog_errno(ret);
1183 return ret;
1184 }
1185
1186 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1187 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
1188 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, gd);
1189 ret = -EIO;
1190 goto out;
1191 }
1192
1193 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
Joel Becker1187c962008-09-03 20:03:39 -07001194 ac->ac_max_block, bit_off, &found);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001195 if (ret < 0) {
1196 if (ret != -ENOSPC)
1197 mlog_errno(ret);
1198 goto out;
1199 }
1200
1201 *num_bits = found;
1202
1203 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1204 *num_bits,
1205 le16_to_cpu(gd->bg_chain));
1206 if (ret < 0) {
1207 mlog_errno(ret);
1208 goto out;
1209 }
1210
1211 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1212 *bit_off, *num_bits);
1213 if (ret < 0)
1214 mlog_errno(ret);
1215
1216 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1217
1218out:
1219 brelse(group_bh);
1220
1221 return ret;
1222}
1223
Mark Fashehccd979b2005-12-15 14:31:24 -08001224static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001225 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001226 u32 bits_wanted,
1227 u32 min_bits,
1228 u16 *bit_off,
1229 unsigned int *num_bits,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001230 u64 *bg_blkno,
1231 u16 *bits_left)
Mark Fashehccd979b2005-12-15 14:31:24 -08001232{
1233 int status;
1234 u16 chain, tmp_bits;
1235 u32 tmp_used;
1236 u64 next_group;
Mark Fashehccd979b2005-12-15 14:31:24 -08001237 struct inode *alloc_inode = ac->ac_inode;
1238 struct buffer_head *group_bh = NULL;
1239 struct buffer_head *prev_group_bh = NULL;
1240 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1241 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1242 struct ocfs2_group_desc *bg;
1243
1244 chain = ac->ac_chain;
Mark Fashehb06970532006-03-03 10:24:33 -08001245 mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1246 bits_wanted, chain,
1247 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001248
1249 status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb),
1250 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1251 &group_bh, OCFS2_BH_CACHED, alloc_inode);
1252 if (status < 0) {
1253 mlog_errno(status);
1254 goto bail;
1255 }
1256 bg = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001257 status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
1258 if (status) {
1259 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001260 goto bail;
1261 }
1262
1263 status = -ENOSPC;
1264 /* for now, the chain search is a bit simplistic. We just use
1265 * the 1st group with any empty bits. */
Joel Becker1187c962008-09-03 20:03:39 -07001266 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1267 bits_wanted, min_bits,
1268 ac->ac_max_block, bit_off,
1269 &tmp_bits)) == -ENOSPC) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001270 if (!bg->bg_next_group)
1271 break;
Mark Fashehccd979b2005-12-15 14:31:24 -08001272 if (prev_group_bh) {
1273 brelse(prev_group_bh);
1274 prev_group_bh = NULL;
1275 }
1276 next_group = le64_to_cpu(bg->bg_next_group);
1277 prev_group_bh = group_bh;
1278 group_bh = NULL;
1279 status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb),
1280 next_group, &group_bh,
1281 OCFS2_BH_CACHED, alloc_inode);
1282 if (status < 0) {
1283 mlog_errno(status);
1284 goto bail;
1285 }
1286 bg = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001287 status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
1288 if (status) {
1289 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001290 goto bail;
1291 }
1292 }
1293 if (status < 0) {
1294 if (status != -ENOSPC)
1295 mlog_errno(status);
1296 goto bail;
1297 }
1298
Mark Fashehb06970532006-03-03 10:24:33 -08001299 mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -07001300 tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001301
1302 *num_bits = tmp_bits;
1303
1304 BUG_ON(*num_bits == 0);
1305
1306 /*
1307 * Keep track of previous block descriptor read. When
1308 * we find a target, if we have read more than X
1309 * number of descriptors, and the target is reasonably
1310 * empty, relink him to top of his chain.
1311 *
1312 * We've read 0 extra blocks and only send one more to
1313 * the transaction, yet the next guy to search has a
1314 * much easier time.
1315 *
1316 * Do this *after* figuring out how many bits we're taking out
1317 * of our target group.
1318 */
1319 if (ac->ac_allow_chain_relink &&
1320 (prev_group_bh) &&
1321 (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1322 status = ocfs2_relink_block_group(handle, alloc_inode,
1323 ac->ac_bh, group_bh,
1324 prev_group_bh, chain);
1325 if (status < 0) {
1326 mlog_errno(status);
1327 goto bail;
1328 }
1329 }
1330
1331 /* Ok, claim our bits now: set the info on dinode, chainlist
1332 * and then the group */
1333 status = ocfs2_journal_access(handle,
1334 alloc_inode,
1335 ac->ac_bh,
1336 OCFS2_JOURNAL_ACCESS_WRITE);
1337 if (status < 0) {
1338 mlog_errno(status);
1339 goto bail;
1340 }
1341
1342 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1343 fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1344 le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1345
1346 status = ocfs2_journal_dirty(handle,
1347 ac->ac_bh);
1348 if (status < 0) {
1349 mlog_errno(status);
1350 goto bail;
1351 }
1352
1353 status = ocfs2_block_group_set_bits(handle,
1354 alloc_inode,
1355 bg,
1356 group_bh,
1357 *bit_off,
1358 *num_bits);
1359 if (status < 0) {
1360 mlog_errno(status);
1361 goto bail;
1362 }
1363
Mark Fashehb06970532006-03-03 10:24:33 -08001364 mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
Mark Fasheh1ca1a112007-04-27 16:01:25 -07001365 (unsigned long long)le64_to_cpu(fe->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001366
1367 *bg_blkno = le64_to_cpu(bg->bg_blkno);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001368 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001369bail:
1370 if (group_bh)
1371 brelse(group_bh);
1372 if (prev_group_bh)
1373 brelse(prev_group_bh);
1374
1375 mlog_exit(status);
1376 return status;
1377}
1378
1379/* will give out up to bits_wanted contiguous bits. */
1380static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1381 struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001382 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001383 u32 bits_wanted,
1384 u32 min_bits,
1385 u16 *bit_off,
1386 unsigned int *num_bits,
1387 u64 *bg_blkno)
1388{
1389 int status;
1390 u16 victim, i;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001391 u16 bits_left = 0;
1392 u64 hint_blkno = ac->ac_last_group;
Mark Fashehccd979b2005-12-15 14:31:24 -08001393 struct ocfs2_chain_list *cl;
1394 struct ocfs2_dinode *fe;
1395
1396 mlog_entry_void();
1397
1398 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1399 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1400 BUG_ON(!ac->ac_bh);
1401
1402 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1403 if (!OCFS2_IS_VALID_DINODE(fe)) {
1404 OCFS2_RO_ON_INVALID_DINODE(osb->sb, fe);
1405 status = -EIO;
1406 goto bail;
1407 }
1408 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1409 le32_to_cpu(fe->id1.bitmap1.i_total)) {
Mark Fashehb06970532006-03-03 10:24:33 -08001410 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1411 "bits but only %u total.",
1412 (unsigned long long)le64_to_cpu(fe->i_blkno),
Mark Fashehccd979b2005-12-15 14:31:24 -08001413 le32_to_cpu(fe->id1.bitmap1.i_used),
1414 le32_to_cpu(fe->id1.bitmap1.i_total));
1415 status = -EIO;
1416 goto bail;
1417 }
1418
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001419 if (hint_blkno) {
1420 /* Attempt to short-circuit the usual search mechanism
1421 * by jumping straight to the most recently used
1422 * allocation group. This helps us mantain some
1423 * contiguousness across allocations. */
Mark Fashehda5cbf22006-10-06 18:34:35 -07001424 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1425 min_bits, bit_off, num_bits,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001426 hint_blkno, &bits_left);
1427 if (!status) {
1428 /* Be careful to update *bg_blkno here as the
1429 * caller is expecting it to be filled in, and
1430 * ocfs2_search_one_group() won't do that for
1431 * us. */
1432 *bg_blkno = hint_blkno;
1433 goto set_hint;
1434 }
1435 if (status < 0 && status != -ENOSPC) {
1436 mlog_errno(status);
1437 goto bail;
1438 }
1439 }
1440
Mark Fashehccd979b2005-12-15 14:31:24 -08001441 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1442
1443 victim = ocfs2_find_victim_chain(cl);
1444 ac->ac_chain = victim;
1445 ac->ac_allow_chain_relink = 1;
1446
Mark Fashehda5cbf22006-10-06 18:34:35 -07001447 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001448 num_bits, bg_blkno, &bits_left);
Mark Fashehccd979b2005-12-15 14:31:24 -08001449 if (!status)
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001450 goto set_hint;
Mark Fashehccd979b2005-12-15 14:31:24 -08001451 if (status < 0 && status != -ENOSPC) {
1452 mlog_errno(status);
1453 goto bail;
1454 }
1455
1456 mlog(0, "Search of victim chain %u came up with nothing, "
1457 "trying all chains now.\n", victim);
1458
1459 /* If we didn't pick a good victim, then just default to
1460 * searching each chain in order. Don't allow chain relinking
1461 * because we only calculate enough journal credits for one
1462 * relink per alloc. */
1463 ac->ac_allow_chain_relink = 0;
1464 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1465 if (i == victim)
1466 continue;
1467 if (!cl->cl_recs[i].c_free)
1468 continue;
1469
1470 ac->ac_chain = i;
Mark Fashehda5cbf22006-10-06 18:34:35 -07001471 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001472 bit_off, num_bits, bg_blkno,
1473 &bits_left);
Mark Fashehccd979b2005-12-15 14:31:24 -08001474 if (!status)
1475 break;
1476 if (status < 0 && status != -ENOSPC) {
1477 mlog_errno(status);
1478 goto bail;
1479 }
1480 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001481
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001482set_hint:
1483 if (status != -ENOSPC) {
1484 /* If the next search of this group is not likely to
1485 * yield a suitable extent, then we reset the last
1486 * group hint so as to not waste a disk read */
1487 if (bits_left < min_bits)
1488 ac->ac_last_group = 0;
1489 else
1490 ac->ac_last_group = *bg_blkno;
1491 }
1492
1493bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001494 mlog_exit(status);
1495 return status;
1496}
1497
1498int ocfs2_claim_metadata(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001499 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001500 struct ocfs2_alloc_context *ac,
1501 u32 bits_wanted,
1502 u16 *suballoc_bit_start,
1503 unsigned int *num_bits,
1504 u64 *blkno_start)
1505{
1506 int status;
1507 u64 bg_blkno;
1508
1509 BUG_ON(!ac);
1510 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1511 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
Mark Fashehccd979b2005-12-15 14:31:24 -08001512
1513 status = ocfs2_claim_suballoc_bits(osb,
1514 ac,
Mark Fashehda5cbf22006-10-06 18:34:35 -07001515 handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001516 bits_wanted,
1517 1,
1518 suballoc_bit_start,
1519 num_bits,
1520 &bg_blkno);
1521 if (status < 0) {
1522 mlog_errno(status);
1523 goto bail;
1524 }
1525 atomic_inc(&osb->alloc_stats.bg_allocs);
1526
1527 *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1528 ac->ac_bits_given += (*num_bits);
1529 status = 0;
1530bail:
1531 mlog_exit(status);
1532 return status;
1533}
1534
1535int ocfs2_claim_new_inode(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001536 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001537 struct ocfs2_alloc_context *ac,
1538 u16 *suballoc_bit,
1539 u64 *fe_blkno)
1540{
1541 int status;
1542 unsigned int num_bits;
1543 u64 bg_blkno;
1544
1545 mlog_entry_void();
1546
1547 BUG_ON(!ac);
1548 BUG_ON(ac->ac_bits_given != 0);
1549 BUG_ON(ac->ac_bits_wanted != 1);
1550 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001551
1552 status = ocfs2_claim_suballoc_bits(osb,
1553 ac,
Mark Fashehda5cbf22006-10-06 18:34:35 -07001554 handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001555 1,
1556 1,
1557 suballoc_bit,
1558 &num_bits,
1559 &bg_blkno);
1560 if (status < 0) {
1561 mlog_errno(status);
1562 goto bail;
1563 }
1564 atomic_inc(&osb->alloc_stats.bg_allocs);
1565
1566 BUG_ON(num_bits != 1);
1567
1568 *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1569 ac->ac_bits_given++;
1570 status = 0;
1571bail:
1572 mlog_exit(status);
1573 return status;
1574}
1575
1576/* translate a group desc. blkno and it's bitmap offset into
1577 * disk cluster offset. */
1578static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1579 u64 bg_blkno,
1580 u16 bg_bit_off)
1581{
1582 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1583 u32 cluster = 0;
1584
1585 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1586
1587 if (bg_blkno != osb->first_cluster_group_blkno)
1588 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1589 cluster += (u32) bg_bit_off;
1590 return cluster;
1591}
1592
1593/* given a cluster offset, calculate which block group it belongs to
1594 * and return that block offset. */
Tao Mad6590722007-12-18 15:47:03 +08001595u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
Mark Fashehccd979b2005-12-15 14:31:24 -08001596{
1597 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1598 u32 group_no;
1599
1600 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1601
1602 group_no = cluster / osb->bitmap_cpg;
1603 if (!group_no)
1604 return osb->first_cluster_group_blkno;
1605 return ocfs2_clusters_to_blocks(inode->i_sb,
1606 group_no * osb->bitmap_cpg);
1607}
1608
1609/* given the block number of a cluster start, calculate which cluster
1610 * group and descriptor bitmap offset that corresponds to. */
1611static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1612 u64 data_blkno,
1613 u64 *bg_blkno,
1614 u16 *bg_bit_off)
1615{
1616 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1617 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1618
1619 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1620
1621 *bg_blkno = ocfs2_which_cluster_group(inode,
1622 data_cluster);
1623
1624 if (*bg_blkno == osb->first_cluster_group_blkno)
1625 *bg_bit_off = (u16) data_cluster;
1626 else
1627 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1628 data_blkno - *bg_blkno);
1629}
1630
1631/*
1632 * min_bits - minimum contiguous chunk from this total allocation we
1633 * can handle. set to what we asked for originally for a full
1634 * contig. allocation, set to '1' to indicate we can deal with extents
1635 * of any size.
1636 */
Mark Fasheh415cb802007-09-16 20:10:16 -07001637int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1638 handle_t *handle,
1639 struct ocfs2_alloc_context *ac,
1640 u32 min_clusters,
1641 u32 max_clusters,
1642 u32 *cluster_start,
1643 u32 *num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08001644{
1645 int status;
Mark Fasheh415cb802007-09-16 20:10:16 -07001646 unsigned int bits_wanted = max_clusters;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001647 u64 bg_blkno = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001648 u16 bg_bit_off;
1649
1650 mlog_entry_void();
1651
Mark Fashehccd979b2005-12-15 14:31:24 -08001652 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1653
1654 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1655 && ac->ac_which != OCFS2_AC_USE_MAIN);
Mark Fashehccd979b2005-12-15 14:31:24 -08001656
1657 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1658 status = ocfs2_claim_local_alloc_bits(osb,
1659 handle,
1660 ac,
1661 bits_wanted,
1662 cluster_start,
1663 num_clusters);
1664 if (!status)
1665 atomic_inc(&osb->alloc_stats.local_data);
1666 } else {
1667 if (min_clusters > (osb->bitmap_cpg - 1)) {
1668 /* The only paths asking for contiguousness
1669 * should know about this already. */
Sunil Mushran2fbe8d12007-12-20 14:58:11 -08001670 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1671 "group bitmap size %u!\n", min_clusters,
1672 osb->bitmap_cpg);
Mark Fashehccd979b2005-12-15 14:31:24 -08001673 status = -ENOSPC;
1674 goto bail;
1675 }
1676 /* clamp the current request down to a realistic size. */
1677 if (bits_wanted > (osb->bitmap_cpg - 1))
1678 bits_wanted = osb->bitmap_cpg - 1;
1679
1680 status = ocfs2_claim_suballoc_bits(osb,
1681 ac,
Mark Fashehda5cbf22006-10-06 18:34:35 -07001682 handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001683 bits_wanted,
1684 min_clusters,
1685 &bg_bit_off,
1686 num_clusters,
1687 &bg_blkno);
1688 if (!status) {
1689 *cluster_start =
1690 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1691 bg_blkno,
1692 bg_bit_off);
1693 atomic_inc(&osb->alloc_stats.bitmap_data);
1694 }
1695 }
1696 if (status < 0) {
1697 if (status != -ENOSPC)
1698 mlog_errno(status);
1699 goto bail;
1700 }
1701
1702 ac->ac_bits_given += *num_clusters;
1703
1704bail:
1705 mlog_exit(status);
1706 return status;
1707}
1708
Mark Fasheh415cb802007-09-16 20:10:16 -07001709int ocfs2_claim_clusters(struct ocfs2_super *osb,
1710 handle_t *handle,
1711 struct ocfs2_alloc_context *ac,
1712 u32 min_clusters,
1713 u32 *cluster_start,
1714 u32 *num_clusters)
1715{
1716 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1717
1718 return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1719 bits_wanted, cluster_start, num_clusters);
1720}
1721
Mark Fasheh1fabe142006-10-09 18:11:45 -07001722static inline int ocfs2_block_group_clear_bits(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001723 struct inode *alloc_inode,
1724 struct ocfs2_group_desc *bg,
1725 struct buffer_head *group_bh,
1726 unsigned int bit_off,
1727 unsigned int num_bits)
1728{
1729 int status;
1730 unsigned int tmp;
1731 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1732 struct ocfs2_group_desc *undo_bg = NULL;
1733
1734 mlog_entry_void();
1735
1736 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
1737 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
1738 status = -EIO;
1739 goto bail;
1740 }
1741
1742 mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1743
1744 if (ocfs2_is_cluster_bitmap(alloc_inode))
1745 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1746
1747 status = ocfs2_journal_access(handle, alloc_inode, group_bh,
1748 journal_type);
1749 if (status < 0) {
1750 mlog_errno(status);
1751 goto bail;
1752 }
1753
1754 if (ocfs2_is_cluster_bitmap(alloc_inode))
1755 undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data;
1756
1757 tmp = num_bits;
1758 while(tmp--) {
1759 ocfs2_clear_bit((bit_off + tmp),
1760 (unsigned long *) bg->bg_bitmap);
1761 if (ocfs2_is_cluster_bitmap(alloc_inode))
1762 ocfs2_set_bit(bit_off + tmp,
1763 (unsigned long *) undo_bg->bg_bitmap);
1764 }
1765 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1766
1767 status = ocfs2_journal_dirty(handle, group_bh);
1768 if (status < 0)
1769 mlog_errno(status);
1770bail:
1771 return status;
1772}
1773
1774/*
1775 * expects the suballoc inode to already be locked.
1776 */
Mark Fasheh2b604352007-06-22 15:45:27 -07001777int ocfs2_free_suballoc_bits(handle_t *handle,
1778 struct inode *alloc_inode,
1779 struct buffer_head *alloc_bh,
1780 unsigned int start_bit,
1781 u64 bg_blkno,
1782 unsigned int count)
Mark Fashehccd979b2005-12-15 14:31:24 -08001783{
1784 int status = 0;
1785 u32 tmp_used;
1786 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
1787 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1788 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1789 struct buffer_head *group_bh = NULL;
1790 struct ocfs2_group_desc *group;
1791
1792 mlog_entry_void();
1793
1794 if (!OCFS2_IS_VALID_DINODE(fe)) {
1795 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
1796 status = -EIO;
1797 goto bail;
1798 }
1799 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1800
Mark Fashehb06970532006-03-03 10:24:33 -08001801 mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1802 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1803 (unsigned long long)bg_blkno, start_bit);
Mark Fashehccd979b2005-12-15 14:31:24 -08001804
1805 status = ocfs2_read_block(osb, bg_blkno, &group_bh, OCFS2_BH_CACHED,
1806 alloc_inode);
1807 if (status < 0) {
1808 mlog_errno(status);
1809 goto bail;
1810 }
1811
1812 group = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001813 status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, group);
1814 if (status) {
1815 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001816 goto bail;
1817 }
1818 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1819
1820 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1821 group, group_bh,
1822 start_bit, count);
1823 if (status < 0) {
1824 mlog_errno(status);
1825 goto bail;
1826 }
1827
1828 status = ocfs2_journal_access(handle, alloc_inode, alloc_bh,
1829 OCFS2_JOURNAL_ACCESS_WRITE);
1830 if (status < 0) {
1831 mlog_errno(status);
1832 goto bail;
1833 }
1834
1835 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
1836 count);
1837 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1838 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
1839
1840 status = ocfs2_journal_dirty(handle, alloc_bh);
1841 if (status < 0) {
1842 mlog_errno(status);
1843 goto bail;
1844 }
1845
1846bail:
1847 if (group_bh)
1848 brelse(group_bh);
1849
1850 mlog_exit(status);
1851 return status;
1852}
1853
Mark Fasheh1fabe142006-10-09 18:11:45 -07001854int ocfs2_free_dinode(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001855 struct inode *inode_alloc_inode,
1856 struct buffer_head *inode_alloc_bh,
1857 struct ocfs2_dinode *di)
1858{
1859 u64 blk = le64_to_cpu(di->i_blkno);
1860 u16 bit = le16_to_cpu(di->i_suballoc_bit);
1861 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1862
1863 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
1864 inode_alloc_bh, bit, bg_blkno, 1);
1865}
1866
Mark Fasheh1fabe142006-10-09 18:11:45 -07001867int ocfs2_free_clusters(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001868 struct inode *bitmap_inode,
1869 struct buffer_head *bitmap_bh,
1870 u64 start_blk,
1871 unsigned int num_clusters)
1872{
1873 int status;
1874 u16 bg_start_bit;
1875 u64 bg_blkno;
1876 struct ocfs2_dinode *fe;
1877
1878 /* You can't ever have a contiguous set of clusters
1879 * bigger than a block group bitmap so we never have to worry
1880 * about looping on them. */
1881
1882 mlog_entry_void();
1883
1884 /* This is expensive. We can safely remove once this stuff has
1885 * gotten tested really well. */
1886 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
1887
1888 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
1889
1890 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
1891 &bg_start_bit);
1892
Mark Fashehb06970532006-03-03 10:24:33 -08001893 mlog(0, "want to free %u clusters starting at block %llu\n",
1894 num_clusters, (unsigned long long)start_blk);
1895 mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
1896 (unsigned long long)bg_blkno, bg_start_bit);
Mark Fashehccd979b2005-12-15 14:31:24 -08001897
1898 status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
1899 bg_start_bit, bg_blkno,
1900 num_clusters);
Mark Fasheh9c7af402008-07-28 18:02:53 -07001901 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001902 mlog_errno(status);
Mark Fasheh9c7af402008-07-28 18:02:53 -07001903 goto out;
1904 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001905
Mark Fasheh9c7af402008-07-28 18:02:53 -07001906 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
1907 num_clusters);
1908
1909out:
Mark Fashehccd979b2005-12-15 14:31:24 -08001910 mlog_exit(status);
1911 return status;
1912}
1913
1914static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
1915{
1916 printk("Block Group:\n");
1917 printk("bg_signature: %s\n", bg->bg_signature);
1918 printk("bg_size: %u\n", bg->bg_size);
1919 printk("bg_bits: %u\n", bg->bg_bits);
1920 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
1921 printk("bg_chain: %u\n", bg->bg_chain);
1922 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
Mark Fashehb06970532006-03-03 10:24:33 -08001923 printk("bg_next_group: %llu\n",
1924 (unsigned long long)bg->bg_next_group);
1925 printk("bg_parent_dinode: %llu\n",
1926 (unsigned long long)bg->bg_parent_dinode);
1927 printk("bg_blkno: %llu\n",
1928 (unsigned long long)bg->bg_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001929}
1930
1931static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
1932{
1933 int i;
1934
Mark Fashehb06970532006-03-03 10:24:33 -08001935 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001936 printk("i_signature: %s\n", fe->i_signature);
Mark Fashehb06970532006-03-03 10:24:33 -08001937 printk("i_size: %llu\n",
1938 (unsigned long long)fe->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08001939 printk("i_clusters: %u\n", fe->i_clusters);
1940 printk("i_generation: %u\n",
1941 le32_to_cpu(fe->i_generation));
1942 printk("id1.bitmap1.i_used: %u\n",
1943 le32_to_cpu(fe->id1.bitmap1.i_used));
1944 printk("id1.bitmap1.i_total: %u\n",
1945 le32_to_cpu(fe->id1.bitmap1.i_total));
1946 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
1947 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
1948 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
1949 printk("id2.i_chain.cl_next_free_rec: %u\n",
1950 fe->id2.i_chain.cl_next_free_rec);
1951 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
1952 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
1953 fe->id2.i_chain.cl_recs[i].c_free);
1954 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
1955 fe->id2.i_chain.cl_recs[i].c_total);
Mark Fashehb06970532006-03-03 10:24:33 -08001956 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
1957 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001958 }
1959}
Tao Mae7d4cb62008-08-18 17:38:44 +08001960
1961/*
1962 * For a given allocation, determine which allocators will need to be
1963 * accessed, and lock them, reserving the appropriate number of bits.
1964 *
1965 * Sparse file systems call this from ocfs2_write_begin_nolock()
1966 * and ocfs2_allocate_unwritten_extents().
1967 *
1968 * File systems which don't support holes call this from
1969 * ocfs2_extend_allocation().
1970 */
Joel Beckerf99b9b72008-08-20 19:36:33 -07001971int ocfs2_lock_allocators(struct inode *inode,
1972 struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +08001973 u32 clusters_to_add, u32 extents_to_split,
1974 struct ocfs2_alloc_context **data_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07001975 struct ocfs2_alloc_context **meta_ac)
Tao Mae7d4cb62008-08-18 17:38:44 +08001976{
1977 int ret = 0, num_free_extents;
1978 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
1979 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1980
1981 *meta_ac = NULL;
1982 if (data_ac)
1983 *data_ac = NULL;
1984
1985 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
1986
Joel Beckerf99b9b72008-08-20 19:36:33 -07001987 num_free_extents = ocfs2_num_free_extents(osb, inode, et);
Tao Mae7d4cb62008-08-18 17:38:44 +08001988 if (num_free_extents < 0) {
1989 ret = num_free_extents;
1990 mlog_errno(ret);
1991 goto out;
1992 }
1993
1994 /*
1995 * Sparse allocation file systems need to be more conservative
1996 * with reserving room for expansion - the actual allocation
1997 * happens while we've got a journal handle open so re-taking
1998 * a cluster lock (because we ran out of room for another
1999 * extent) will violate ordering rules.
2000 *
2001 * Most of the time we'll only be seeing this 1 cluster at a time
2002 * anyway.
2003 *
2004 * Always lock for any unwritten extents - we might want to
2005 * add blocks during a split.
2006 */
2007 if (!num_free_extents ||
2008 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
Joel Beckerf99b9b72008-08-20 19:36:33 -07002009 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
Tao Mae7d4cb62008-08-18 17:38:44 +08002010 if (ret < 0) {
2011 if (ret != -ENOSPC)
2012 mlog_errno(ret);
2013 goto out;
2014 }
2015 }
2016
2017 if (clusters_to_add == 0)
2018 goto out;
2019
2020 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2021 if (ret < 0) {
2022 if (ret != -ENOSPC)
2023 mlog_errno(ret);
2024 goto out;
2025 }
2026
2027out:
2028 if (ret) {
2029 if (*meta_ac) {
2030 ocfs2_free_alloc_context(*meta_ac);
2031 *meta_ac = NULL;
2032 }
2033
2034 /*
2035 * We cannot have an error and a non null *data_ac.
2036 */
2037 }
2038
2039 return ret;
2040}