blob: 0cb889a17ae14db8ffc7e0688543be132a126ea6 [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
Mark Fashehccd979b2005-12-15 14:31:24 -080032#include <cluster/masklog.h>
33
34#include "ocfs2.h"
35
36#include "alloc.h"
Joel Beckerd6b32bb2008-10-17 14:55:01 -070037#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080038#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"
Tao Ma2f73e132011-02-22 08:22:33 +080046#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080047
48#include "buffer_head_io.h"
49
Tao Maffda89a2008-03-03 17:12:09 +080050#define NOT_ALLOC_NEW_GROUP 0
Tao Ma60ca81e2009-02-25 00:53:24 +080051#define ALLOC_NEW_GROUP 0x1
52#define ALLOC_GROUPS_FROM_GLOBAL 0x2
Tao Maffda89a2008-03-03 17:12:09 +080053
Tiger Yangb89c5422010-01-25 14:11:06 +080054#define OCFS2_MAX_TO_STEAL 1024
Tao Ma4d0ddb22008-03-05 16:11:46 +080055
Joel Becker7d1fe092010-04-13 14:30:19 +080056struct ocfs2_suballoc_result {
Joel Becker2b6cb572010-03-26 10:09:15 +080057 u64 sr_bg_blkno; /* The bg we allocated from. Set
58 to 0 when a block group is
59 contiguous. */
Mark Fashehe49e2762010-08-13 15:15:17 -070060 u64 sr_bg_stable_blkno; /*
61 * Doesn't change, always
62 * set to target block
63 * group descriptor
64 * block.
65 */
Joel Beckerba206632010-03-26 10:08:59 +080066 u64 sr_blkno; /* The first allocated block */
Joel Becker7d1fe092010-04-13 14:30:19 +080067 unsigned int sr_bit_offset; /* The bit in the bg */
68 unsigned int sr_bits; /* How many bits we claimed */
69};
70
Mark Fashehb2b6ebf2010-08-26 13:06:50 -070071static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
72{
73 if (res->sr_blkno == 0)
74 return 0;
75
76 if (res->sr_bg_blkno)
77 return res->sr_bg_blkno;
78
79 return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
80}
81
Mark Fashehccd979b2005-12-15 14:31:24 -080082static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
83static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
84static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
Mark Fasheh1fabe142006-10-09 18:11:45 -070085static int ocfs2_block_group_fill(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -080086 struct inode *alloc_inode,
87 struct buffer_head *bg_bh,
88 u64 group_blkno,
Joel Becker798db352010-04-13 14:26:32 +080089 unsigned int group_clusters,
Mark Fashehccd979b2005-12-15 14:31:24 -080090 u16 my_chain,
91 struct ocfs2_chain_list *cl);
92static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
93 struct inode *alloc_inode,
Joel Becker1187c962008-09-03 20:03:39 -070094 struct buffer_head *bh,
Tao Ma60ca81e2009-02-25 00:53:24 +080095 u64 max_block,
Tao Mafeb473a2009-02-25 00:53:25 +080096 u64 *last_alloc_group,
Tao Ma60ca81e2009-02-25 00:53:24 +080097 int flags);
Mark Fashehccd979b2005-12-15 14:31:24 -080098
Mark Fashehccd979b2005-12-15 14:31:24 -080099static int ocfs2_cluster_group_search(struct inode *inode,
100 struct buffer_head *group_bh,
101 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -0700102 u64 max_block,
Joel Becker7d1fe092010-04-13 14:30:19 +0800103 struct ocfs2_suballoc_result *res);
Mark Fashehccd979b2005-12-15 14:31:24 -0800104static int ocfs2_block_group_search(struct inode *inode,
105 struct buffer_head *group_bh,
106 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -0700107 u64 max_block,
Joel Becker7d1fe092010-04-13 14:30:19 +0800108 struct ocfs2_suballoc_result *res);
Joel Beckeraa8f8e92010-03-26 10:08:07 +0800109static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -0700110 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800111 u32 bits_wanted,
112 u32 min_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +0800113 struct ocfs2_suballoc_result *res);
Mark Fashehccd979b2005-12-15 14:31:24 -0800114static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
115 int nr);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700116static int ocfs2_relink_block_group(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800117 struct inode *alloc_inode,
118 struct buffer_head *fe_bh,
119 struct buffer_head *bg_bh,
120 struct buffer_head *prev_bg_bh,
121 u16 chain);
122static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
123 u32 wanted);
Mark Fashehccd979b2005-12-15 14:31:24 -0800124static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
125 u64 bg_blkno,
126 u16 bg_bit_off);
Mark Fashehccd979b2005-12-15 14:31:24 -0800127static inline void ocfs2_block_to_cluster_group(struct inode *inode,
128 u64 data_blkno,
129 u64 *bg_blkno,
130 u16 *bg_bit_off);
Joel Becker1187c962008-09-03 20:03:39 -0700131static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
132 u32 bits_wanted, u64 max_block,
Tao Ma60ca81e2009-02-25 00:53:24 +0800133 int flags,
Joel Becker1187c962008-09-03 20:03:39 -0700134 struct ocfs2_alloc_context **ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800135
Mark Fasheh9c7af402008-07-28 18:02:53 -0700136void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
Mark Fashehccd979b2005-12-15 14:31:24 -0800137{
Mark Fashehda5cbf22006-10-06 18:34:35 -0700138 struct inode *inode = ac->ac_inode;
139
140 if (inode) {
141 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
Mark Fashehe63aecb62007-10-18 15:30:42 -0700142 ocfs2_inode_unlock(inode, 1);
Mark Fashehda5cbf22006-10-06 18:34:35 -0700143
144 mutex_unlock(&inode->i_mutex);
145
146 iput(inode);
Tao Ma4d0ddb22008-03-05 16:11:46 +0800147 ac->ac_inode = NULL;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700148 }
Mark Fasheha81cb882008-10-07 14:25:16 -0700149 brelse(ac->ac_bh);
150 ac->ac_bh = NULL;
Mark Fashehe3b4a972009-12-07 13:16:07 -0800151 ac->ac_resv = NULL;
Mark Fashehe49e2762010-08-13 15:15:17 -0700152 if (ac->ac_find_loc_priv) {
153 kfree(ac->ac_find_loc_priv);
154 ac->ac_find_loc_priv = NULL;
155 }
Tao Ma4d0ddb22008-03-05 16:11:46 +0800156}
157
158void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
159{
160 ocfs2_free_ac_resource(ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800161 kfree(ac);
162}
163
164static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
165{
166 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
167}
168
Joel Becker57e3e792008-11-13 14:49:13 -0800169#define do_error(fmt, ...) \
170 do{ \
Tao Ma78c37eb2010-03-03 11:26:27 +0800171 if (resize) \
Joel Becker57e3e792008-11-13 14:49:13 -0800172 mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__); \
173 else \
174 ocfs2_error(sb, fmt, ##__VA_ARGS__); \
175 } while (0)
176
Joel Becker970e4932008-11-13 14:49:19 -0800177static int ocfs2_validate_gd_self(struct super_block *sb,
178 struct buffer_head *bh,
Tao Ma78c37eb2010-03-03 11:26:27 +0800179 int resize)
Joel Becker970e4932008-11-13 14:49:19 -0800180{
181 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
182
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700183 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
Joel Becker68f64d42008-11-13 14:49:14 -0800184 do_error("Group descriptor #%llu has bad signature %.*s",
185 (unsigned long long)bh->b_blocknr, 7,
Joel Becker57e3e792008-11-13 14:49:13 -0800186 gd->bg_signature);
187 return -EINVAL;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700188 }
189
Joel Becker68f64d42008-11-13 14:49:14 -0800190 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
191 do_error("Group descriptor #%llu has an invalid bg_blkno "
192 "of %llu",
193 (unsigned long long)bh->b_blocknr,
194 (unsigned long long)le64_to_cpu(gd->bg_blkno));
195 return -EINVAL;
196 }
197
198 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
199 do_error("Group descriptor #%llu has an invalid "
200 "fs_generation of #%u",
201 (unsigned long long)bh->b_blocknr,
202 le32_to_cpu(gd->bg_generation));
203 return -EINVAL;
204 }
205
Joel Becker970e4932008-11-13 14:49:19 -0800206 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
207 do_error("Group descriptor #%llu has bit count %u but "
208 "claims that %u are free",
209 (unsigned long long)bh->b_blocknr,
210 le16_to_cpu(gd->bg_bits),
211 le16_to_cpu(gd->bg_free_bits_count));
212 return -EINVAL;
213 }
214
215 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
216 do_error("Group descriptor #%llu has bit count %u but "
217 "max bitmap bits of %u",
218 (unsigned long long)bh->b_blocknr,
219 le16_to_cpu(gd->bg_bits),
220 8 * le16_to_cpu(gd->bg_size));
221 return -EINVAL;
222 }
223
224 return 0;
225}
226
227static int ocfs2_validate_gd_parent(struct super_block *sb,
228 struct ocfs2_dinode *di,
229 struct buffer_head *bh,
Tao Ma78c37eb2010-03-03 11:26:27 +0800230 int resize)
Joel Becker970e4932008-11-13 14:49:19 -0800231{
232 unsigned int max_bits;
233 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
234
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700235 if (di->i_blkno != gd->bg_parent_dinode) {
Joel Becker68f64d42008-11-13 14:49:14 -0800236 do_error("Group descriptor #%llu has bad parent "
Joel Becker57e3e792008-11-13 14:49:13 -0800237 "pointer (%llu, expected %llu)",
Joel Becker68f64d42008-11-13 14:49:14 -0800238 (unsigned long long)bh->b_blocknr,
Joel Becker57e3e792008-11-13 14:49:13 -0800239 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
240 (unsigned long long)le64_to_cpu(di->i_blkno));
241 return -EINVAL;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700242 }
243
244 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
245 if (le16_to_cpu(gd->bg_bits) > max_bits) {
Joel Becker68f64d42008-11-13 14:49:14 -0800246 do_error("Group descriptor #%llu has bit count of %u",
247 (unsigned long long)bh->b_blocknr,
Joel Becker57e3e792008-11-13 14:49:13 -0800248 le16_to_cpu(gd->bg_bits));
249 return -EINVAL;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700250 }
251
Tao Ma78c37eb2010-03-03 11:26:27 +0800252 /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
253 if ((le16_to_cpu(gd->bg_chain) >
254 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
255 ((le16_to_cpu(gd->bg_chain) ==
256 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
Joel Becker68f64d42008-11-13 14:49:14 -0800257 do_error("Group descriptor #%llu has bad chain %u",
258 (unsigned long long)bh->b_blocknr,
Joel Becker57e3e792008-11-13 14:49:13 -0800259 le16_to_cpu(gd->bg_chain));
260 return -EINVAL;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700261 }
262
Joel Becker970e4932008-11-13 14:49:19 -0800263 return 0;
264}
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700265
Joel Becker57e3e792008-11-13 14:49:13 -0800266#undef do_error
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700267
Joel Becker970e4932008-11-13 14:49:19 -0800268/*
269 * This version only prints errors. It does not fail the filesystem, and
270 * exists only for resize.
271 */
272int ocfs2_check_group_descriptor(struct super_block *sb,
273 struct ocfs2_dinode *di,
274 struct buffer_head *bh)
275{
276 int rc;
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700277 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
Joel Becker970e4932008-11-13 14:49:19 -0800278
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700279 BUG_ON(!buffer_uptodate(bh));
280
281 /*
282 * If the ecc fails, we return the error but otherwise
283 * leave the filesystem running. We know any error is
284 * local to this block.
285 */
286 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
Joel Becker13723d02008-10-17 19:25:01 -0700287 if (rc) {
288 mlog(ML_ERROR,
289 "Checksum failed for group descriptor %llu\n",
290 (unsigned long long)bh->b_blocknr);
291 } else
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700292 rc = ocfs2_validate_gd_self(sb, bh, 1);
Joel Becker970e4932008-11-13 14:49:19 -0800293 if (!rc)
294 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
295
296 return rc;
297}
298
299static int ocfs2_validate_group_descriptor(struct super_block *sb,
300 struct buffer_head *bh)
301{
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700302 int rc;
303 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
304
Tao Ma2f73e132011-02-22 08:22:33 +0800305 trace_ocfs2_validate_group_descriptor(
306 (unsigned long long)bh->b_blocknr);
Joel Becker970e4932008-11-13 14:49:19 -0800307
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700308 BUG_ON(!buffer_uptodate(bh));
309
310 /*
311 * If the ecc fails, we return the error but otherwise
312 * leave the filesystem running. We know any error is
313 * local to this block.
314 */
315 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
316 if (rc)
317 return rc;
318
319 /*
320 * Errors after here are fatal.
321 */
322
Joel Becker970e4932008-11-13 14:49:19 -0800323 return ocfs2_validate_gd_self(sb, bh, 0);
Mark Fasheh7bf72ed2006-05-03 17:46:50 -0700324}
325
Joel Becker68f64d42008-11-13 14:49:14 -0800326int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
327 u64 gd_blkno, struct buffer_head **bh)
328{
329 int rc;
330 struct buffer_head *tmp = *bh;
331
Joel Becker8cb471e2009-02-10 20:00:41 -0800332 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
Joel Becker970e4932008-11-13 14:49:19 -0800333 ocfs2_validate_group_descriptor);
Joel Becker68f64d42008-11-13 14:49:14 -0800334 if (rc)
335 goto out;
336
Joel Becker970e4932008-11-13 14:49:19 -0800337 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
Joel Becker68f64d42008-11-13 14:49:14 -0800338 if (rc) {
339 brelse(tmp);
340 goto out;
341 }
342
343 /* If ocfs2_read_block() got us a new bh, pass it up. */
344 if (!*bh)
345 *bh = tmp;
346
347out:
348 return rc;
349}
350
Joel Becker798db352010-04-13 14:26:32 +0800351static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
352 struct ocfs2_group_desc *bg,
353 struct ocfs2_chain_list *cl,
Tao Ma47dea422010-09-13 15:13:50 +0800354 u64 p_blkno, unsigned int clusters)
Joel Becker798db352010-04-13 14:26:32 +0800355{
356 struct ocfs2_extent_list *el = &bg->bg_list;
357 struct ocfs2_extent_rec *rec;
358
Tao Ma47119542010-04-22 14:09:15 +0800359 BUG_ON(!ocfs2_supports_discontig_bg(osb));
Joel Becker798db352010-04-13 14:26:32 +0800360 if (!el->l_next_free_rec)
361 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
362 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
Tao Ma47119542010-04-22 14:09:15 +0800363 rec->e_blkno = cpu_to_le64(p_blkno);
Joel Becker798db352010-04-13 14:26:32 +0800364 rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
365 le16_to_cpu(cl->cl_bpc));
Tao Ma47dea422010-09-13 15:13:50 +0800366 rec->e_leaf_clusters = cpu_to_le16(clusters);
Joel Becker798db352010-04-13 14:26:32 +0800367 le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
Tao Ma47119542010-04-22 14:09:15 +0800368 le16_add_cpu(&bg->bg_free_bits_count,
369 clusters * le16_to_cpu(cl->cl_bpc));
Joel Becker798db352010-04-13 14:26:32 +0800370 le16_add_cpu(&el->l_next_free_rec, 1);
371}
372
Mark Fasheh1fabe142006-10-09 18:11:45 -0700373static int ocfs2_block_group_fill(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800374 struct inode *alloc_inode,
375 struct buffer_head *bg_bh,
376 u64 group_blkno,
Joel Becker798db352010-04-13 14:26:32 +0800377 unsigned int group_clusters,
Mark Fashehccd979b2005-12-15 14:31:24 -0800378 u16 my_chain,
379 struct ocfs2_chain_list *cl)
380{
381 int status = 0;
Joel Becker798db352010-04-13 14:26:32 +0800382 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800383 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
384 struct super_block * sb = alloc_inode->i_sb;
385
Mark Fashehccd979b2005-12-15 14:31:24 -0800386 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
Mark Fashehb06970532006-03-03 10:24:33 -0800387 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
388 "b_blocknr (%llu)",
389 (unsigned long long)group_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800390 (unsigned long long) bg_bh->b_blocknr);
391 status = -EIO;
392 goto bail;
393 }
394
Joel Becker13723d02008-10-17 19:25:01 -0700395 status = ocfs2_journal_access_gd(handle,
Joel Becker0cf2f762009-02-12 16:41:25 -0800396 INODE_CACHE(alloc_inode),
Joel Becker13723d02008-10-17 19:25:01 -0700397 bg_bh,
398 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800399 if (status < 0) {
400 mlog_errno(status);
401 goto bail;
402 }
403
404 memset(bg, 0, sb->s_blocksize);
405 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
406 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
Tao Ma85718822010-04-13 14:38:06 +0800407 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
408 osb->s_feature_incompat));
Mark Fashehccd979b2005-12-15 14:31:24 -0800409 bg->bg_chain = cpu_to_le16(my_chain);
410 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
411 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
412 bg->bg_blkno = cpu_to_le64(group_blkno);
Joel Becker798db352010-04-13 14:26:32 +0800413 if (group_clusters == le16_to_cpu(cl->cl_cpg))
414 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
415 else
Tao Ma47119542010-04-22 14:09:15 +0800416 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
Joel Becker798db352010-04-13 14:26:32 +0800417 group_clusters);
418
Mark Fashehccd979b2005-12-15 14:31:24 -0800419 /* set the 1st bit in the bitmap to account for the descriptor block */
420 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
421 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
422
Joel Beckerec20cec2010-03-19 14:13:52 -0700423 ocfs2_journal_dirty(handle, bg_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800424
425 /* There is no need to zero out or otherwise initialize the
426 * other blocks in a group - All valid FS metadata in a block
427 * group stores the superblock fs_generation value at
428 * allocation time. */
429
430bail:
Tao Mac1e8d352011-03-07 16:43:21 +0800431 if (status)
432 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800433 return status;
434}
435
436static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
437{
438 u16 curr, best;
439
440 best = curr = 0;
441 while (curr < le16_to_cpu(cl->cl_count)) {
442 if (le32_to_cpu(cl->cl_recs[best].c_total) >
443 le32_to_cpu(cl->cl_recs[curr].c_total))
444 best = curr;
445 curr++;
446 }
447 return best;
448}
449
Joel Becker798db352010-04-13 14:26:32 +0800450static struct buffer_head *
451ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
452 struct inode *alloc_inode,
453 struct ocfs2_alloc_context *ac,
454 struct ocfs2_chain_list *cl)
455{
456 int status;
457 u32 bit_off, num_bits;
458 u64 bg_blkno;
459 struct buffer_head *bg_bh;
460 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
461
Joel Becker1ed9b772010-05-06 13:59:06 +0800462 status = ocfs2_claim_clusters(handle, ac,
Joel Becker798db352010-04-13 14:26:32 +0800463 le16_to_cpu(cl->cl_cpg), &bit_off,
464 &num_bits);
465 if (status < 0) {
466 if (status != -ENOSPC)
467 mlog_errno(status);
468 goto bail;
469 }
470
471 /* setup the group */
472 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Tao Ma2f73e132011-02-22 08:22:33 +0800473 trace_ocfs2_block_group_alloc_contig(
474 (unsigned long long)bg_blkno, alloc_rec);
Joel Becker798db352010-04-13 14:26:32 +0800475
476 bg_bh = sb_getblk(osb->sb, bg_blkno);
477 if (!bg_bh) {
Rui Xiang7391a292013-11-12 15:06:54 -0800478 status = -ENOMEM;
Joel Becker798db352010-04-13 14:26:32 +0800479 mlog_errno(status);
480 goto bail;
481 }
482 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
483
484 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
485 bg_blkno, num_bits, alloc_rec, cl);
486 if (status < 0) {
487 brelse(bg_bh);
488 mlog_errno(status);
489 }
490
491bail:
492 return status ? ERR_PTR(status) : bg_bh;
493}
494
495static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
496 handle_t *handle,
497 struct ocfs2_alloc_context *ac,
498 unsigned int min_bits,
499 u32 *bit_off, u32 *num_bits)
500{
Joel Becker18d3a982010-05-18 16:47:55 -0700501 int status = 0;
Joel Becker798db352010-04-13 14:26:32 +0800502
503 while (min_bits) {
Joel Becker1ed9b772010-05-06 13:59:06 +0800504 status = ocfs2_claim_clusters(handle, ac, min_bits,
Joel Becker798db352010-04-13 14:26:32 +0800505 bit_off, num_bits);
506 if (status != -ENOSPC)
507 break;
508
509 min_bits >>= 1;
510 }
511
512 return status;
513}
514
515static int ocfs2_block_group_grow_discontig(handle_t *handle,
516 struct inode *alloc_inode,
517 struct buffer_head *bg_bh,
518 struct ocfs2_alloc_context *ac,
519 struct ocfs2_chain_list *cl,
520 unsigned int min_bits)
521{
522 int status;
523 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
524 struct ocfs2_group_desc *bg =
525 (struct ocfs2_group_desc *)bg_bh->b_data;
Tao Ma47119542010-04-22 14:09:15 +0800526 unsigned int needed = le16_to_cpu(cl->cl_cpg) -
527 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
Joel Becker798db352010-04-13 14:26:32 +0800528 u32 p_cpos, clusters;
529 u64 p_blkno;
530 struct ocfs2_extent_list *el = &bg->bg_list;
531
532 status = ocfs2_journal_access_gd(handle,
533 INODE_CACHE(alloc_inode),
534 bg_bh,
535 OCFS2_JOURNAL_ACCESS_CREATE);
536 if (status < 0) {
537 mlog_errno(status);
538 goto bail;
539 }
540
541 while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
542 le16_to_cpu(el->l_count))) {
Joel Becker798db352010-04-13 14:26:32 +0800543 if (min_bits > needed)
544 min_bits = needed;
545 status = ocfs2_block_group_claim_bits(osb, handle, ac,
546 min_bits, &p_cpos,
547 &clusters);
548 if (status < 0) {
549 if (status != -ENOSPC)
550 mlog_errno(status);
551 goto bail;
552 }
553 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
554 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
555 clusters);
556
557 min_bits = clusters;
Tao Ma47119542010-04-22 14:09:15 +0800558 needed = le16_to_cpu(cl->cl_cpg) -
559 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
Joel Becker798db352010-04-13 14:26:32 +0800560 }
561
562 if (needed > 0) {
Tao Ma47119542010-04-22 14:09:15 +0800563 /*
564 * We have used up all the extent rec but can't fill up
565 * the cpg. So bail out.
566 */
567 status = -ENOSPC;
568 goto bail;
Joel Becker798db352010-04-13 14:26:32 +0800569 }
570
571 ocfs2_journal_dirty(handle, bg_bh);
572
573bail:
574 return status;
575}
576
Joel Becker8b06bc52010-03-26 10:09:29 +0800577static void ocfs2_bg_alloc_cleanup(handle_t *handle,
578 struct ocfs2_alloc_context *cluster_ac,
579 struct inode *alloc_inode,
580 struct buffer_head *bg_bh)
Joel Becker798db352010-04-13 14:26:32 +0800581{
Joel Becker8b06bc52010-03-26 10:09:29 +0800582 int i, ret;
Joel Becker798db352010-04-13 14:26:32 +0800583 struct ocfs2_group_desc *bg;
584 struct ocfs2_extent_list *el;
585 struct ocfs2_extent_rec *rec;
586
587 if (!bg_bh)
588 return;
589
590 bg = (struct ocfs2_group_desc *)bg_bh->b_data;
591 el = &bg->bg_list;
592 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
593 rec = &el->l_recs[i];
Joel Becker8b06bc52010-03-26 10:09:29 +0800594 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
595 cluster_ac->ac_bh,
596 le64_to_cpu(rec->e_blkno),
Al Viro72094e42012-04-13 12:30:02 -0400597 le16_to_cpu(rec->e_leaf_clusters));
Joel Becker8b06bc52010-03-26 10:09:29 +0800598 if (ret)
599 mlog_errno(ret);
600 /* Try all the clusters to free */
Joel Becker798db352010-04-13 14:26:32 +0800601 }
602
603 ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
604 brelse(bg_bh);
605}
606
607static struct buffer_head *
608ocfs2_block_group_alloc_discontig(handle_t *handle,
609 struct inode *alloc_inode,
610 struct ocfs2_alloc_context *ac,
Joel Becker8b06bc52010-03-26 10:09:29 +0800611 struct ocfs2_chain_list *cl)
Joel Becker798db352010-04-13 14:26:32 +0800612{
613 int status;
614 u32 bit_off, num_bits;
615 u64 bg_blkno;
616 unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
617 struct buffer_head *bg_bh = NULL;
618 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
619 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
620
Tao Ma47119542010-04-22 14:09:15 +0800621 if (!ocfs2_supports_discontig_bg(osb)) {
Joel Becker798db352010-04-13 14:26:32 +0800622 status = -ENOSPC;
623 goto bail;
624 }
625
Joel Becker8b06bc52010-03-26 10:09:29 +0800626 status = ocfs2_extend_trans(handle,
627 ocfs2_calc_bg_discontig_credits(osb->sb));
628 if (status) {
629 mlog_errno(status);
630 goto bail;
631 }
632
Joel Becker95ec0ad2010-03-26 10:10:08 +0800633 /*
634 * We're going to be grabbing from multiple cluster groups.
635 * We don't have enough credits to relink them all, and the
636 * cluster groups will be staying in cache for the duration of
637 * this operation.
638 */
Xiaowei.Hu309a85b2013-02-27 17:02:49 -0800639 ac->ac_disable_chain_relink = 1;
Joel Becker95ec0ad2010-03-26 10:10:08 +0800640
Joel Becker798db352010-04-13 14:26:32 +0800641 /* Claim the first region */
642 status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
643 &bit_off, &num_bits);
644 if (status < 0) {
645 if (status != -ENOSPC)
646 mlog_errno(status);
647 goto bail;
648 }
649 min_bits = num_bits;
650
651 /* setup the group */
652 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Tao Ma2f73e132011-02-22 08:22:33 +0800653 trace_ocfs2_block_group_alloc_discontig(
654 (unsigned long long)bg_blkno, alloc_rec);
Joel Becker798db352010-04-13 14:26:32 +0800655
656 bg_bh = sb_getblk(osb->sb, bg_blkno);
657 if (!bg_bh) {
Rui Xiang7391a292013-11-12 15:06:54 -0800658 status = -ENOMEM;
Joel Becker798db352010-04-13 14:26:32 +0800659 mlog_errno(status);
660 goto bail;
661 }
662 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
663
664 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
665 bg_blkno, num_bits, alloc_rec, cl);
666 if (status < 0) {
667 mlog_errno(status);
668 goto bail;
669 }
670
671 status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
672 bg_bh, ac, cl, min_bits);
673 if (status)
674 mlog_errno(status);
675
676bail:
677 if (status)
Joel Becker8b06bc52010-03-26 10:09:29 +0800678 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
Joel Becker798db352010-04-13 14:26:32 +0800679 return status ? ERR_PTR(status) : bg_bh;
680}
681
Mark Fashehccd979b2005-12-15 14:31:24 -0800682/*
683 * We expect the block group allocator to already be locked.
684 */
685static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
686 struct inode *alloc_inode,
Joel Becker1187c962008-09-03 20:03:39 -0700687 struct buffer_head *bh,
Tao Ma60ca81e2009-02-25 00:53:24 +0800688 u64 max_block,
Tao Mafeb473a2009-02-25 00:53:25 +0800689 u64 *last_alloc_group,
Tao Ma60ca81e2009-02-25 00:53:24 +0800690 int flags)
Mark Fashehccd979b2005-12-15 14:31:24 -0800691{
692 int status, credits;
693 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
694 struct ocfs2_chain_list *cl;
695 struct ocfs2_alloc_context *ac = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700696 handle_t *handle = NULL;
Tao Ma47119542010-04-22 14:09:15 +0800697 u16 alloc_rec;
Mark Fashehccd979b2005-12-15 14:31:24 -0800698 struct buffer_head *bg_bh = NULL;
699 struct ocfs2_group_desc *bg;
700
701 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
702
Mark Fashehccd979b2005-12-15 14:31:24 -0800703 cl = &fe->id2.i_chain;
Joel Becker1187c962008-09-03 20:03:39 -0700704 status = ocfs2_reserve_clusters_with_limit(osb,
705 le16_to_cpu(cl->cl_cpg),
Tao Ma60ca81e2009-02-25 00:53:24 +0800706 max_block, flags, &ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800707 if (status < 0) {
708 if (status != -ENOSPC)
709 mlog_errno(status);
710 goto bail;
711 }
712
713 credits = ocfs2_calc_group_alloc_credits(osb->sb,
714 le16_to_cpu(cl->cl_cpg));
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700715 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800716 if (IS_ERR(handle)) {
717 status = PTR_ERR(handle);
718 handle = NULL;
719 mlog_errno(status);
720 goto bail;
721 }
722
Tao Mafeb473a2009-02-25 00:53:25 +0800723 if (last_alloc_group && *last_alloc_group != 0) {
Tao Ma2f73e132011-02-22 08:22:33 +0800724 trace_ocfs2_block_group_alloc(
725 (unsigned long long)*last_alloc_group);
Tao Mafeb473a2009-02-25 00:53:25 +0800726 ac->ac_last_group = *last_alloc_group;
727 }
Joel Becker798db352010-04-13 14:26:32 +0800728
729 bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
730 ac, cl);
731 if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
732 bg_bh = ocfs2_block_group_alloc_discontig(handle,
733 alloc_inode,
Joel Becker8b06bc52010-03-26 10:09:29 +0800734 ac, cl);
Joel Becker798db352010-04-13 14:26:32 +0800735 if (IS_ERR(bg_bh)) {
736 status = PTR_ERR(bg_bh);
737 bg_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800738 if (status != -ENOSPC)
739 mlog_errno(status);
740 goto bail;
741 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800742 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
743
Joel Becker0cf2f762009-02-12 16:41:25 -0800744 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
Joel Becker13723d02008-10-17 19:25:01 -0700745 bh, OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800746 if (status < 0) {
747 mlog_errno(status);
748 goto bail;
749 }
750
Tao Ma47119542010-04-22 14:09:15 +0800751 alloc_rec = le16_to_cpu(bg->bg_chain);
752 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
Mark Fashehccd979b2005-12-15 14:31:24 -0800753 le16_to_cpu(bg->bg_free_bits_count));
Tao Ma47119542010-04-22 14:09:15 +0800754 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
Joel Becker798db352010-04-13 14:26:32 +0800755 le16_to_cpu(bg->bg_bits));
Tao Ma0a463b72010-07-08 11:11:11 +0800756 cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -0800757 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
758 le16_add_cpu(&cl->cl_next_free_rec, 1);
759
760 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
761 le16_to_cpu(bg->bg_free_bits_count));
762 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
763 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
764
Joel Beckerec20cec2010-03-19 14:13:52 -0700765 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800766
767 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
768 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
769 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
770 le32_to_cpu(fe->i_clusters)));
771 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
772 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
Mark Fasheh8110b072007-03-22 16:53:23 -0700773 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -0700774 ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800775
776 status = 0;
Tao Mafeb473a2009-02-25 00:53:25 +0800777
778 /* save the new last alloc group so that the caller can cache it. */
779 if (last_alloc_group)
780 *last_alloc_group = ac->ac_last_group;
781
Mark Fashehccd979b2005-12-15 14:31:24 -0800782bail:
783 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700784 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800785
786 if (ac)
787 ocfs2_free_alloc_context(ac);
788
Mark Fasheha81cb882008-10-07 14:25:16 -0700789 brelse(bg_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800790
Tao Mac1e8d352011-03-07 16:43:21 +0800791 if (status)
792 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800793 return status;
794}
795
796static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
Mark Fashehda5cbf22006-10-06 18:34:35 -0700797 struct ocfs2_alloc_context *ac,
798 int type,
Tao Maffda89a2008-03-03 17:12:09 +0800799 u32 slot,
Tao Mafeb473a2009-02-25 00:53:25 +0800800 u64 *last_alloc_group,
Tao Ma60ca81e2009-02-25 00:53:24 +0800801 int flags)
Mark Fashehccd979b2005-12-15 14:31:24 -0800802{
803 int status;
804 u32 bits_wanted = ac->ac_bits_wanted;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700805 struct inode *alloc_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -0800806 struct buffer_head *bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800807 struct ocfs2_dinode *fe;
808 u32 free_bits;
809
Mark Fashehda5cbf22006-10-06 18:34:35 -0700810 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
811 if (!alloc_inode) {
812 mlog_errno(-EINVAL);
813 return -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800814 }
815
Mark Fashehda5cbf22006-10-06 18:34:35 -0700816 mutex_lock(&alloc_inode->i_mutex);
817
Mark Fashehe63aecb62007-10-18 15:30:42 -0700818 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
Mark Fashehda5cbf22006-10-06 18:34:35 -0700819 if (status < 0) {
820 mutex_unlock(&alloc_inode->i_mutex);
821 iput(alloc_inode);
822
823 mlog_errno(status);
824 return status;
825 }
826
827 ac->ac_inode = alloc_inode;
Tao Maa4a48912008-03-03 17:12:30 +0800828 ac->ac_alloc_slot = slot;
Mark Fashehda5cbf22006-10-06 18:34:35 -0700829
Mark Fashehccd979b2005-12-15 14:31:24 -0800830 fe = (struct ocfs2_dinode *) bh->b_data;
Joel Becker10995aa2008-11-13 14:49:12 -0800831
832 /* The bh was validated by the inode read inside
833 * ocfs2_inode_lock(). Any corruption is a code bug. */
834 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
835
Mark Fashehccd979b2005-12-15 14:31:24 -0800836 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
Mark Fashehb06970532006-03-03 10:24:33 -0800837 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
838 (unsigned long long)le64_to_cpu(fe->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800839 status = -EIO;
840 goto bail;
841 }
842
843 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
844 le32_to_cpu(fe->id1.bitmap1.i_used);
845
846 if (bits_wanted > free_bits) {
847 /* cluster bitmap never grows */
848 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
Tao Ma2f73e132011-02-22 08:22:33 +0800849 trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
850 free_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800851 status = -ENOSPC;
852 goto bail;
853 }
854
Tao Ma60ca81e2009-02-25 00:53:24 +0800855 if (!(flags & ALLOC_NEW_GROUP)) {
Tao Ma2f73e132011-02-22 08:22:33 +0800856 trace_ocfs2_reserve_suballoc_bits_no_new_group(
857 slot, bits_wanted, free_bits);
Tao Maffda89a2008-03-03 17:12:09 +0800858 status = -ENOSPC;
859 goto bail;
860 }
861
Joel Becker1187c962008-09-03 20:03:39 -0700862 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
Tao Mafeb473a2009-02-25 00:53:25 +0800863 ac->ac_max_block,
864 last_alloc_group, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800865 if (status < 0) {
866 if (status != -ENOSPC)
867 mlog_errno(status);
868 goto bail;
869 }
870 atomic_inc(&osb->alloc_stats.bg_extends);
871
872 /* You should never ask for this much metadata */
873 BUG_ON(bits_wanted >
874 (le32_to_cpu(fe->id1.bitmap1.i_total)
875 - le32_to_cpu(fe->id1.bitmap1.i_used)));
876 }
877
878 get_bh(bh);
879 ac->ac_bh = bh;
880bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700881 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800882
Tao Mac1e8d352011-03-07 16:43:21 +0800883 if (status)
884 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800885 return status;
886}
887
Tiger Yangb89c5422010-01-25 14:11:06 +0800888static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
889{
890 spin_lock(&osb->osb_lock);
891 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
892 spin_unlock(&osb->osb_lock);
893 atomic_set(&osb->s_num_inodes_stolen, 0);
894}
895
896static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
897{
898 spin_lock(&osb->osb_lock);
899 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
900 spin_unlock(&osb->osb_lock);
901 atomic_set(&osb->s_num_meta_stolen, 0);
902}
903
904void ocfs2_init_steal_slots(struct ocfs2_super *osb)
905{
906 ocfs2_init_inode_steal_slot(osb);
907 ocfs2_init_meta_steal_slot(osb);
908}
909
910static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
911{
912 spin_lock(&osb->osb_lock);
913 if (type == INODE_ALLOC_SYSTEM_INODE)
914 osb->s_inode_steal_slot = slot;
915 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
916 osb->s_meta_steal_slot = slot;
917 spin_unlock(&osb->osb_lock);
918}
919
920static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
921{
922 int slot = OCFS2_INVALID_SLOT;
923
924 spin_lock(&osb->osb_lock);
925 if (type == INODE_ALLOC_SYSTEM_INODE)
926 slot = osb->s_inode_steal_slot;
927 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
928 slot = osb->s_meta_steal_slot;
929 spin_unlock(&osb->osb_lock);
930
931 return slot;
932}
933
934static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
935{
936 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
937}
938
939static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
940{
941 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
942}
943
944static int ocfs2_steal_resource(struct ocfs2_super *osb,
945 struct ocfs2_alloc_context *ac,
946 int type)
947{
948 int i, status = -ENOSPC;
949 int slot = __ocfs2_get_steal_slot(osb, type);
950
951 /* Start to steal resource from the first slot after ours. */
952 if (slot == OCFS2_INVALID_SLOT)
953 slot = osb->slot_num + 1;
954
955 for (i = 0; i < osb->max_slots; i++, slot++) {
956 if (slot == osb->max_slots)
957 slot = 0;
958
959 if (slot == osb->slot_num)
960 continue;
961
962 status = ocfs2_reserve_suballoc_bits(osb, ac,
963 type,
964 (u32)slot, NULL,
965 NOT_ALLOC_NEW_GROUP);
966 if (status >= 0) {
967 __ocfs2_set_steal_slot(osb, slot, type);
968 break;
969 }
970
971 ocfs2_free_ac_resource(ac);
972 }
973
974 return status;
975}
976
977static int ocfs2_steal_inode(struct ocfs2_super *osb,
978 struct ocfs2_alloc_context *ac)
979{
980 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
981}
982
983static int ocfs2_steal_meta(struct ocfs2_super *osb,
984 struct ocfs2_alloc_context *ac)
985{
986 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
987}
988
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800989int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
990 int blocks,
991 struct ocfs2_alloc_context **ac)
Mark Fashehccd979b2005-12-15 14:31:24 -0800992{
993 int status;
Tiger Yangb89c5422010-01-25 14:11:06 +0800994 int slot = ocfs2_get_meta_steal_slot(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800995
Robert P. J. Daycd861282006-12-13 00:34:52 -0800996 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800997 if (!(*ac)) {
998 status = -ENOMEM;
999 mlog_errno(status);
1000 goto bail;
1001 }
1002
Tiger Yangcf1d6c72008-08-18 17:11:00 +08001003 (*ac)->ac_bits_wanted = blocks;
Mark Fashehccd979b2005-12-15 14:31:24 -08001004 (*ac)->ac_which = OCFS2_AC_USE_META;
Mark Fashehccd979b2005-12-15 14:31:24 -08001005 (*ac)->ac_group_search = ocfs2_block_group_search;
1006
Tiger Yangb89c5422010-01-25 14:11:06 +08001007 if (slot != OCFS2_INVALID_SLOT &&
1008 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
1009 goto extent_steal;
1010
1011 atomic_set(&osb->s_num_meta_stolen, 0);
Mark Fashehda5cbf22006-10-06 18:34:35 -07001012 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
Tao Maffda89a2008-03-03 17:12:09 +08001013 EXTENT_ALLOC_SYSTEM_INODE,
Tiger Yangb89c5422010-01-25 14:11:06 +08001014 (u32)osb->slot_num, NULL,
Mark Fasheh33d5d382010-02-24 13:34:09 -08001015 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
Tiger Yangb89c5422010-01-25 14:11:06 +08001016
1017
1018 if (status >= 0) {
1019 status = 0;
1020 if (slot != OCFS2_INVALID_SLOT)
1021 ocfs2_init_meta_steal_slot(osb);
1022 goto bail;
1023 } else if (status < 0 && status != -ENOSPC) {
1024 mlog_errno(status);
1025 goto bail;
1026 }
1027
1028 ocfs2_free_ac_resource(*ac);
1029
1030extent_steal:
1031 status = ocfs2_steal_meta(osb, *ac);
1032 atomic_inc(&osb->s_num_meta_stolen);
Mark Fashehccd979b2005-12-15 14:31:24 -08001033 if (status < 0) {
1034 if (status != -ENOSPC)
1035 mlog_errno(status);
1036 goto bail;
1037 }
1038
1039 status = 0;
1040bail:
1041 if ((status < 0) && *ac) {
1042 ocfs2_free_alloc_context(*ac);
1043 *ac = NULL;
1044 }
1045
Tao Mac1e8d352011-03-07 16:43:21 +08001046 if (status)
1047 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001048 return status;
1049}
1050
Tiger Yangcf1d6c72008-08-18 17:11:00 +08001051int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1052 struct ocfs2_extent_list *root_el,
1053 struct ocfs2_alloc_context **ac)
1054{
1055 return ocfs2_reserve_new_metadata_blocks(osb,
1056 ocfs2_extend_meta_needed(root_el),
1057 ac);
1058}
1059
Mark Fashehccd979b2005-12-15 14:31:24 -08001060int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -08001061 struct ocfs2_alloc_context **ac)
1062{
1063 int status;
Tiger Yangb89c5422010-01-25 14:11:06 +08001064 int slot = ocfs2_get_inode_steal_slot(osb);
Tao Mafeb473a2009-02-25 00:53:25 +08001065 u64 alloc_group;
Mark Fashehccd979b2005-12-15 14:31:24 -08001066
Robert P. J. Daycd861282006-12-13 00:34:52 -08001067 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001068 if (!(*ac)) {
1069 status = -ENOMEM;
1070 mlog_errno(status);
1071 goto bail;
1072 }
1073
1074 (*ac)->ac_bits_wanted = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08001075 (*ac)->ac_which = OCFS2_AC_USE_INODE;
1076
Mark Fashehccd979b2005-12-15 14:31:24 -08001077 (*ac)->ac_group_search = ocfs2_block_group_search;
1078
Tao Ma4d0ddb22008-03-05 16:11:46 +08001079 /*
Joel Becker1187c962008-09-03 20:03:39 -07001080 * stat(2) can't handle i_ino > 32bits, so we tell the
1081 * lower levels not to allocate us a block group past that
Joel Becker12462f12008-09-03 20:03:40 -07001082 * limit. The 'inode64' mount option avoids this behavior.
Joel Becker1187c962008-09-03 20:03:39 -07001083 */
Joel Becker12462f12008-09-03 20:03:40 -07001084 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1085 (*ac)->ac_max_block = (u32)~0U;
Joel Becker1187c962008-09-03 20:03:39 -07001086
1087 /*
Tao Ma4d0ddb22008-03-05 16:11:46 +08001088 * slot is set when we successfully steal inode from other nodes.
1089 * It is reset in 3 places:
1090 * 1. when we flush the truncate log
1091 * 2. when we complete local alloc recovery.
1092 * 3. when we successfully allocate from our own slot.
1093 * After it is set, we will go on stealing inodes until we find the
1094 * need to check our slots to see whether there is some space for us.
1095 */
1096 if (slot != OCFS2_INVALID_SLOT &&
Tiger Yangb89c5422010-01-25 14:11:06 +08001097 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
Tao Ma4d0ddb22008-03-05 16:11:46 +08001098 goto inode_steal;
1099
1100 atomic_set(&osb->s_num_inodes_stolen, 0);
Tao Mafeb473a2009-02-25 00:53:25 +08001101 alloc_group = osb->osb_inode_alloc_group;
Mark Fashehda5cbf22006-10-06 18:34:35 -07001102 status = ocfs2_reserve_suballoc_bits(osb, *ac,
1103 INODE_ALLOC_SYSTEM_INODE,
Tiger Yangb89c5422010-01-25 14:11:06 +08001104 (u32)osb->slot_num,
Tao Mafeb473a2009-02-25 00:53:25 +08001105 &alloc_group,
Tao Ma60ca81e2009-02-25 00:53:24 +08001106 ALLOC_NEW_GROUP |
1107 ALLOC_GROUPS_FROM_GLOBAL);
Tao Ma4d0ddb22008-03-05 16:11:46 +08001108 if (status >= 0) {
1109 status = 0;
1110
Tao Mafeb473a2009-02-25 00:53:25 +08001111 spin_lock(&osb->osb_lock);
1112 osb->osb_inode_alloc_group = alloc_group;
1113 spin_unlock(&osb->osb_lock);
Tao Ma2f73e132011-02-22 08:22:33 +08001114 trace_ocfs2_reserve_new_inode_new_group(
1115 (unsigned long long)alloc_group);
Tao Mafeb473a2009-02-25 00:53:25 +08001116
Tao Ma4d0ddb22008-03-05 16:11:46 +08001117 /*
1118 * Some inodes must be freed by us, so try to allocate
1119 * from our own next time.
1120 */
1121 if (slot != OCFS2_INVALID_SLOT)
1122 ocfs2_init_inode_steal_slot(osb);
1123 goto bail;
1124 } else if (status < 0 && status != -ENOSPC) {
1125 mlog_errno(status);
1126 goto bail;
1127 }
1128
1129 ocfs2_free_ac_resource(*ac);
1130
1131inode_steal:
Tiger Yangb89c5422010-01-25 14:11:06 +08001132 status = ocfs2_steal_inode(osb, *ac);
Tao Ma4d0ddb22008-03-05 16:11:46 +08001133 atomic_inc(&osb->s_num_inodes_stolen);
Mark Fashehccd979b2005-12-15 14:31:24 -08001134 if (status < 0) {
1135 if (status != -ENOSPC)
1136 mlog_errno(status);
1137 goto bail;
1138 }
1139
1140 status = 0;
1141bail:
1142 if ((status < 0) && *ac) {
1143 ocfs2_free_alloc_context(*ac);
1144 *ac = NULL;
1145 }
1146
Tao Mac1e8d352011-03-07 16:43:21 +08001147 if (status)
1148 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001149 return status;
1150}
1151
1152/* local alloc code has to do the same thing, so rather than do this
1153 * twice.. */
1154int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1155 struct ocfs2_alloc_context *ac)
1156{
1157 int status;
1158
Mark Fashehccd979b2005-12-15 14:31:24 -08001159 ac->ac_which = OCFS2_AC_USE_MAIN;
1160 ac->ac_group_search = ocfs2_cluster_group_search;
1161
Mark Fashehda5cbf22006-10-06 18:34:35 -07001162 status = ocfs2_reserve_suballoc_bits(osb, ac,
1163 GLOBAL_BITMAP_SYSTEM_INODE,
Tao Mafeb473a2009-02-25 00:53:25 +08001164 OCFS2_INVALID_SLOT, NULL,
Tao Maffda89a2008-03-03 17:12:09 +08001165 ALLOC_NEW_GROUP);
Mark Fashehda5cbf22006-10-06 18:34:35 -07001166 if (status < 0 && status != -ENOSPC) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001167 mlog_errno(status);
Mark Fashehda5cbf22006-10-06 18:34:35 -07001168 goto bail;
1169 }
1170
Mark Fashehccd979b2005-12-15 14:31:24 -08001171bail:
1172 return status;
1173}
1174
1175/* Callers don't need to care which bitmap (local alloc or main) to
1176 * use so we figure it out for them, but unfortunately this clutters
1177 * things a bit. */
Joel Becker1187c962008-09-03 20:03:39 -07001178static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1179 u32 bits_wanted, u64 max_block,
Tao Ma60ca81e2009-02-25 00:53:24 +08001180 int flags,
Joel Becker1187c962008-09-03 20:03:39 -07001181 struct ocfs2_alloc_context **ac)
Mark Fashehccd979b2005-12-15 14:31:24 -08001182{
1183 int status;
1184
Robert P. J. Daycd861282006-12-13 00:34:52 -08001185 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001186 if (!(*ac)) {
1187 status = -ENOMEM;
1188 mlog_errno(status);
1189 goto bail;
1190 }
1191
1192 (*ac)->ac_bits_wanted = bits_wanted;
Joel Becker1187c962008-09-03 20:03:39 -07001193 (*ac)->ac_max_block = max_block;
Mark Fashehccd979b2005-12-15 14:31:24 -08001194
1195 status = -ENOSPC;
Tao Ma60ca81e2009-02-25 00:53:24 +08001196 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1197 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001198 status = ocfs2_reserve_local_alloc_bits(osb,
Mark Fashehccd979b2005-12-15 14:31:24 -08001199 bits_wanted,
1200 *ac);
Mark Fasheha57c8fd2010-03-16 21:01:00 -07001201 if ((status < 0) && (status != -ENOSPC)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001202 mlog_errno(status);
1203 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08001204 }
1205 }
1206
1207 if (status == -ENOSPC) {
1208 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1209 if (status < 0) {
1210 if (status != -ENOSPC)
1211 mlog_errno(status);
1212 goto bail;
1213 }
1214 }
1215
1216 status = 0;
1217bail:
1218 if ((status < 0) && *ac) {
1219 ocfs2_free_alloc_context(*ac);
1220 *ac = NULL;
1221 }
1222
Tao Mac1e8d352011-03-07 16:43:21 +08001223 if (status)
1224 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001225 return status;
1226}
1227
Joel Becker1187c962008-09-03 20:03:39 -07001228int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1229 u32 bits_wanted,
1230 struct ocfs2_alloc_context **ac)
1231{
Tao Ma60ca81e2009-02-25 00:53:24 +08001232 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1233 ALLOC_NEW_GROUP, ac);
Joel Becker1187c962008-09-03 20:03:39 -07001234}
1235
Mark Fashehccd979b2005-12-15 14:31:24 -08001236/*
1237 * More or less lifted from ext3. I'll leave their description below:
1238 *
1239 * "For ext3 allocations, we must not reuse any blocks which are
1240 * allocated in the bitmap buffer's "last committed data" copy. This
1241 * prevents deletes from freeing up the page for reuse until we have
1242 * committed the delete transaction.
1243 *
1244 * If we didn't do this, then deleting something and reallocating it as
1245 * data would allow the old block to be overwritten before the
1246 * transaction committed (because we force data to disk before commit).
1247 * This would lead to corruption if we crashed between overwriting the
1248 * data and committing the delete.
1249 *
1250 * @@@ We may want to make this allocation behaviour conditional on
1251 * data-writes at some point, and disable it for metadata allocations or
1252 * sync-data inodes."
1253 *
1254 * Note: OCFS2 already does this differently for metadata vs data
Joe Perchesc78bad12008-02-03 17:33:42 +02001255 * allocations, as those bitmaps are separate and undo access is never
Mark Fashehccd979b2005-12-15 14:31:24 -08001256 * called on a metadata group descriptor.
1257 */
1258static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1259 int nr)
1260{
1261 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
Sunil Mushran94e41ec2009-06-19 14:45:54 -07001262 int ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001263
1264 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1265 return 0;
Sunil Mushran94e41ec2009-06-19 14:45:54 -07001266
1267 if (!buffer_jbd(bg_bh))
Mark Fashehccd979b2005-12-15 14:31:24 -08001268 return 1;
1269
Sunil Mushran94e41ec2009-06-19 14:45:54 -07001270 jbd_lock_bh_state(bg_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001271 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
Sunil Mushran94e41ec2009-06-19 14:45:54 -07001272 if (bg)
1273 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1274 else
1275 ret = 1;
1276 jbd_unlock_bh_state(bg_bh);
1277
1278 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001279}
1280
1281static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1282 struct buffer_head *bg_bh,
1283 unsigned int bits_wanted,
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001284 unsigned int total_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +08001285 struct ocfs2_suballoc_result *res)
Mark Fashehccd979b2005-12-15 14:31:24 -08001286{
1287 void *bitmap;
1288 u16 best_offset, best_size;
1289 int offset, start, found, status = 0;
1290 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1291
Joel Becker42035302008-11-13 14:49:15 -08001292 /* Callers got this descriptor from
1293 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1294 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
Mark Fashehccd979b2005-12-15 14:31:24 -08001295
1296 found = start = best_offset = best_size = 0;
1297 bitmap = bg->bg_bitmap;
1298
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001299 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1300 if (offset == total_bits)
Mark Fashehccd979b2005-12-15 14:31:24 -08001301 break;
1302
1303 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1304 /* We found a zero, but we can't use it as it
1305 * hasn't been put to disk yet! */
1306 found = 0;
1307 start = offset + 1;
1308 } else if (offset == start) {
1309 /* we found a zero */
1310 found++;
1311 /* move start to the next bit to test */
1312 start++;
1313 } else {
1314 /* got a zero after some ones */
1315 found = 1;
1316 start = offset + 1;
1317 }
1318 if (found > best_size) {
1319 best_size = found;
1320 best_offset = start - found;
1321 }
1322 /* we got everything we needed */
1323 if (found == bits_wanted) {
1324 /* mlog(0, "Found it all!\n"); */
1325 break;
1326 }
1327 }
1328
Joel Becker7d1fe092010-04-13 14:30:19 +08001329 if (best_size) {
1330 res->sr_bit_offset = best_offset;
1331 res->sr_bits = best_size;
Mark Fashehccd979b2005-12-15 14:31:24 -08001332 } else {
1333 status = -ENOSPC;
1334 /* No error log here -- see the comment above
1335 * ocfs2_test_bg_bit_allocatable */
1336 }
1337
1338 return status;
1339}
1340
Younger Liu0a2fcd892014-01-21 15:48:33 -08001341int ocfs2_block_group_set_bits(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001342 struct inode *alloc_inode,
1343 struct ocfs2_group_desc *bg,
1344 struct buffer_head *group_bh,
1345 unsigned int bit_off,
1346 unsigned int num_bits)
1347{
1348 int status;
1349 void *bitmap = bg->bg_bitmap;
1350 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1351
Joel Becker42035302008-11-13 14:49:15 -08001352 /* All callers get the descriptor via
1353 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1354 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
Mark Fashehccd979b2005-12-15 14:31:24 -08001355 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1356
Tao Ma2f73e132011-02-22 08:22:33 +08001357 trace_ocfs2_block_group_set_bits(bit_off, num_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08001358
1359 if (ocfs2_is_cluster_bitmap(alloc_inode))
1360 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1361
Joel Becker13723d02008-10-17 19:25:01 -07001362 status = ocfs2_journal_access_gd(handle,
Joel Becker0cf2f762009-02-12 16:41:25 -08001363 INODE_CACHE(alloc_inode),
Joel Becker13723d02008-10-17 19:25:01 -07001364 group_bh,
1365 journal_type);
Mark Fashehccd979b2005-12-15 14:31:24 -08001366 if (status < 0) {
1367 mlog_errno(status);
1368 goto bail;
1369 }
1370
1371 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
Srinivas Eeda9b5cd102010-10-05 15:53:06 -07001372 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1373 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
1374 " count %u but claims %u are freed. num_bits %d",
1375 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1376 le16_to_cpu(bg->bg_bits),
1377 le16_to_cpu(bg->bg_free_bits_count), num_bits);
1378 return -EROFS;
1379 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001380 while(num_bits--)
1381 ocfs2_set_bit(bit_off++, bitmap);
1382
Joel Beckerec20cec2010-03-19 14:13:52 -07001383 ocfs2_journal_dirty(handle, group_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001384
1385bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001386 return status;
1387}
1388
1389/* find the one with the most empty bits */
1390static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1391{
1392 u16 curr, best;
1393
1394 BUG_ON(!cl->cl_next_free_rec);
1395
1396 best = curr = 0;
1397 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1398 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1399 le32_to_cpu(cl->cl_recs[best].c_free))
1400 best = curr;
1401 curr++;
1402 }
1403
1404 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1405 return best;
1406}
1407
Mark Fasheh1fabe142006-10-09 18:11:45 -07001408static int ocfs2_relink_block_group(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001409 struct inode *alloc_inode,
1410 struct buffer_head *fe_bh,
1411 struct buffer_head *bg_bh,
1412 struct buffer_head *prev_bg_bh,
1413 u16 chain)
1414{
1415 int status;
1416 /* there is a really tiny chance the journal calls could fail,
1417 * but we wouldn't want inconsistent blocks in *any* case. */
Jie Liu49309842013-07-03 15:00:59 -07001418 u64 bg_ptr, prev_bg_ptr;
Mark Fashehccd979b2005-12-15 14:31:24 -08001419 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1420 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1421 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1422
Joel Becker42035302008-11-13 14:49:15 -08001423 /* The caller got these descriptors from
1424 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1425 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1426 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
Mark Fashehccd979b2005-12-15 14:31:24 -08001427
Tao Ma2f73e132011-02-22 08:22:33 +08001428 trace_ocfs2_relink_block_group(
1429 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1430 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1431 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001432
Mark Fashehccd979b2005-12-15 14:31:24 -08001433 bg_ptr = le64_to_cpu(bg->bg_next_group);
1434 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1435
Joel Becker0cf2f762009-02-12 16:41:25 -08001436 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1437 prev_bg_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001438 OCFS2_JOURNAL_ACCESS_WRITE);
Andrew Morton25e28922013-07-03 15:01:00 -07001439 if (status < 0)
Jie Liu49309842013-07-03 15:00:59 -07001440 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001441
1442 prev_bg->bg_next_group = bg->bg_next_group;
Joel Beckerec20cec2010-03-19 14:13:52 -07001443 ocfs2_journal_dirty(handle, prev_bg_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001444
Joel Becker0cf2f762009-02-12 16:41:25 -08001445 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1446 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
Andrew Morton25e28922013-07-03 15:01:00 -07001447 if (status < 0)
Jie Liu49309842013-07-03 15:00:59 -07001448 goto out_rollback_prev_bg;
Mark Fashehccd979b2005-12-15 14:31:24 -08001449
1450 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
Joel Beckerec20cec2010-03-19 14:13:52 -07001451 ocfs2_journal_dirty(handle, bg_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001452
Joel Becker0cf2f762009-02-12 16:41:25 -08001453 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1454 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
Andrew Morton25e28922013-07-03 15:01:00 -07001455 if (status < 0)
Jie Liu49309842013-07-03 15:00:59 -07001456 goto out_rollback_bg;
Mark Fashehccd979b2005-12-15 14:31:24 -08001457
1458 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
Joel Beckerec20cec2010-03-19 14:13:52 -07001459 ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001460
Jie Liu49309842013-07-03 15:00:59 -07001461out:
Andrew Morton25e28922013-07-03 15:01:00 -07001462 if (status < 0)
1463 mlog_errno(status);
Jie Liu49309842013-07-03 15:00:59 -07001464 return status;
Joel Becker42035302008-11-13 14:49:15 -08001465
Jie Liu49309842013-07-03 15:00:59 -07001466out_rollback_bg:
1467 bg->bg_next_group = cpu_to_le64(bg_ptr);
1468out_rollback_prev_bg:
1469 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
Andrew Morton25e28922013-07-03 15:01:00 -07001470 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001471}
1472
1473static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1474 u32 wanted)
1475{
1476 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1477}
1478
1479/* return 0 on success, -ENOSPC to keep searching and any other < 0
1480 * value on error. */
1481static int ocfs2_cluster_group_search(struct inode *inode,
1482 struct buffer_head *group_bh,
1483 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -07001484 u64 max_block,
Joel Becker7d1fe092010-04-13 14:30:19 +08001485 struct ocfs2_suballoc_result *res)
Mark Fashehccd979b2005-12-15 14:31:24 -08001486{
1487 int search = -ENOSPC;
1488 int ret;
Joel Becker1187c962008-09-03 20:03:39 -07001489 u64 blkoff;
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001490 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fasheh9c7af402008-07-28 18:02:53 -07001491 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001492 unsigned int max_bits, gd_cluster_off;
Mark Fashehccd979b2005-12-15 14:31:24 -08001493
1494 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1495
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001496 if (gd->bg_free_bits_count) {
1497 max_bits = le16_to_cpu(gd->bg_bits);
1498
1499 /* Tail groups in cluster bitmaps which aren't cpg
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001500 * aligned are prone to partial extension by a failed
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001501 * fs resize. If the file system resize never got to
1502 * update the dinode cluster count, then we don't want
1503 * to trust any clusters past it, regardless of what
1504 * the group descriptor says. */
1505 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1506 le64_to_cpu(gd->bg_blkno));
1507 if ((gd_cluster_off + max_bits) >
1508 OCFS2_I(inode)->ip_clusters) {
1509 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
Tao Ma2f73e132011-02-22 08:22:33 +08001510 trace_ocfs2_cluster_group_search_wrong_max_bits(
1511 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1512 le16_to_cpu(gd->bg_bits),
1513 OCFS2_I(inode)->ip_clusters, max_bits);
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001514 }
1515
Mark Fashehccd979b2005-12-15 14:31:24 -08001516 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1517 group_bh, bits_wanted,
Joel Becker7d1fe092010-04-13 14:30:19 +08001518 max_bits, res);
Mark Fashehccd979b2005-12-15 14:31:24 -08001519 if (ret)
1520 return ret;
1521
Joel Becker1187c962008-09-03 20:03:39 -07001522 if (max_block) {
1523 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1524 gd_cluster_off +
Joel Becker7d1fe092010-04-13 14:30:19 +08001525 res->sr_bit_offset +
1526 res->sr_bits);
Tao Ma2f73e132011-02-22 08:22:33 +08001527 trace_ocfs2_cluster_group_search_max_block(
1528 (unsigned long long)blkoff,
1529 (unsigned long long)max_block);
Joel Becker1187c962008-09-03 20:03:39 -07001530 if (blkoff > max_block)
1531 return -ENOSPC;
1532 }
1533
Mark Fashehccd979b2005-12-15 14:31:24 -08001534 /* ocfs2_block_group_find_clear_bits() might
1535 * return success, but we still want to return
1536 * -ENOSPC unless it found the minimum number
1537 * of bits. */
Joel Becker7d1fe092010-04-13 14:30:19 +08001538 if (min_bits <= res->sr_bits)
Mark Fashehccd979b2005-12-15 14:31:24 -08001539 search = 0; /* success */
Joel Becker7d1fe092010-04-13 14:30:19 +08001540 else if (res->sr_bits) {
Mark Fasheh9c7af402008-07-28 18:02:53 -07001541 /*
1542 * Don't show bits which we'll be returning
1543 * for allocation to the local alloc bitmap.
1544 */
Joel Becker7d1fe092010-04-13 14:30:19 +08001545 ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08001546 }
1547 }
1548
1549 return search;
1550}
1551
1552static int ocfs2_block_group_search(struct inode *inode,
1553 struct buffer_head *group_bh,
1554 u32 bits_wanted, u32 min_bits,
Joel Becker1187c962008-09-03 20:03:39 -07001555 u64 max_block,
Joel Becker7d1fe092010-04-13 14:30:19 +08001556 struct ocfs2_suballoc_result *res)
Mark Fashehccd979b2005-12-15 14:31:24 -08001557{
1558 int ret = -ENOSPC;
Joel Becker1187c962008-09-03 20:03:39 -07001559 u64 blkoff;
Mark Fashehccd979b2005-12-15 14:31:24 -08001560 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1561
1562 BUG_ON(min_bits != 1);
1563 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1564
Joel Becker1187c962008-09-03 20:03:39 -07001565 if (bg->bg_free_bits_count) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001566 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1567 group_bh, bits_wanted,
Mark Fasheh7bf72ed2006-05-03 17:46:50 -07001568 le16_to_cpu(bg->bg_bits),
Joel Becker7d1fe092010-04-13 14:30:19 +08001569 res);
Joel Becker1187c962008-09-03 20:03:39 -07001570 if (!ret && max_block) {
Joel Becker7d1fe092010-04-13 14:30:19 +08001571 blkoff = le64_to_cpu(bg->bg_blkno) +
1572 res->sr_bit_offset + res->sr_bits;
Tao Ma2f73e132011-02-22 08:22:33 +08001573 trace_ocfs2_block_group_search_max_block(
1574 (unsigned long long)blkoff,
1575 (unsigned long long)max_block);
Joel Becker1187c962008-09-03 20:03:39 -07001576 if (blkoff > max_block)
1577 ret = -ENOSPC;
1578 }
1579 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001580
1581 return ret;
1582}
1583
Younger Liu0a2fcd892014-01-21 15:48:33 -08001584int ocfs2_alloc_dinode_update_counts(struct inode *inode,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001585 handle_t *handle,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001586 struct buffer_head *di_bh,
1587 u32 num_bits,
1588 u16 chain)
1589{
1590 int ret;
1591 u32 tmp_used;
1592 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1593 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1594
Joel Becker0cf2f762009-02-12 16:41:25 -08001595 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001596 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001597 if (ret < 0) {
1598 mlog_errno(ret);
1599 goto out;
1600 }
1601
1602 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1603 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1604 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
Joel Beckerec20cec2010-03-19 14:13:52 -07001605 ocfs2_journal_dirty(handle, di_bh);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001606
1607out:
1608 return ret;
1609}
1610
Younger Liudb66c712014-04-03 14:47:10 -07001611void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1612 struct buffer_head *di_bh,
1613 u32 num_bits,
1614 u16 chain)
1615{
1616 u32 tmp_used;
1617 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1618 struct ocfs2_chain_list *cl;
1619
1620 cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1621 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1622 di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1623 le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1624}
1625
Joel Beckerba206632010-03-26 10:08:59 +08001626static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1627 struct ocfs2_extent_rec *rec,
1628 struct ocfs2_chain_list *cl)
Joel Becker13e434c2010-03-26 10:08:27 +08001629{
1630 unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1631 unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
Al Viro72094e42012-04-13 12:30:02 -04001632 unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
Joel Becker13e434c2010-03-26 10:08:27 +08001633
1634 if (res->sr_bit_offset < bitoff)
1635 return 0;
1636 if (res->sr_bit_offset >= (bitoff + bitcount))
1637 return 0;
Joel Beckerba206632010-03-26 10:08:59 +08001638 res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1639 (res->sr_bit_offset - bitoff);
Joel Becker13e434c2010-03-26 10:08:27 +08001640 if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1641 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1642 return 1;
1643}
1644
Joel Beckerba206632010-03-26 10:08:59 +08001645static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1646 struct ocfs2_group_desc *bg,
1647 struct ocfs2_suballoc_result *res)
Joel Becker13e434c2010-03-26 10:08:27 +08001648{
1649 int i;
Joel Becker2b6cb572010-03-26 10:09:15 +08001650 u64 bg_blkno = res->sr_bg_blkno; /* Save off */
Joel Becker13e434c2010-03-26 10:08:27 +08001651 struct ocfs2_extent_rec *rec;
1652 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1653 struct ocfs2_chain_list *cl = &di->id2.i_chain;
1654
Joel Beckerba206632010-03-26 10:08:59 +08001655 if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1656 res->sr_blkno = 0;
Joel Becker13e434c2010-03-26 10:08:27 +08001657 return;
Joel Beckerba206632010-03-26 10:08:59 +08001658 }
Joel Becker13e434c2010-03-26 10:08:27 +08001659
Joel Beckerba206632010-03-26 10:08:59 +08001660 res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
Joel Becker2b6cb572010-03-26 10:09:15 +08001661 res->sr_bg_blkno = 0; /* Clear it for contig block groups */
Tao Ma47119542010-04-22 14:09:15 +08001662 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
Joel Beckerba206632010-03-26 10:08:59 +08001663 !bg->bg_list.l_next_free_rec)
Joel Becker13e434c2010-03-26 10:08:27 +08001664 return;
1665
1666 for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1667 rec = &bg->bg_list.l_recs[i];
Joel Becker2b6cb572010-03-26 10:09:15 +08001668 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1669 res->sr_bg_blkno = bg_blkno; /* Restore */
Joel Becker13e434c2010-03-26 10:08:27 +08001670 break;
Joel Becker2b6cb572010-03-26 10:09:15 +08001671 }
Joel Becker13e434c2010-03-26 10:08:27 +08001672 }
1673}
1674
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001675static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001676 handle_t *handle,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001677 u32 bits_wanted,
1678 u32 min_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +08001679 struct ocfs2_suballoc_result *res,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001680 u16 *bits_left)
1681{
1682 int ret;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001683 struct buffer_head *group_bh = NULL;
1684 struct ocfs2_group_desc *gd;
Joel Becker68f64d42008-11-13 14:49:14 -08001685 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001686 struct inode *alloc_inode = ac->ac_inode;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001687
Joel Becker7d1fe092010-04-13 14:30:19 +08001688 ret = ocfs2_read_group_descriptor(alloc_inode, di,
1689 res->sr_bg_blkno, &group_bh);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001690 if (ret < 0) {
1691 mlog_errno(ret);
1692 return ret;
1693 }
1694
1695 gd = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001696 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +08001697 ac->ac_max_block, res);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001698 if (ret < 0) {
1699 if (ret != -ENOSPC)
1700 mlog_errno(ret);
1701 goto out;
1702 }
1703
Joel Becker13e434c2010-03-26 10:08:27 +08001704 if (!ret)
Joel Beckerba206632010-03-26 10:08:59 +08001705 ocfs2_bg_discontig_fix_result(ac, gd, res);
Joel Becker13e434c2010-03-26 10:08:27 +08001706
Mark Fashehe49e2762010-08-13 15:15:17 -07001707 /*
1708 * sr_bg_blkno might have been changed by
1709 * ocfs2_bg_discontig_fix_result
1710 */
1711 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1712
1713 if (ac->ac_find_loc_only)
1714 goto out_loc_only;
1715
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001716 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
Joel Becker7d1fe092010-04-13 14:30:19 +08001717 res->sr_bits,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001718 le16_to_cpu(gd->bg_chain));
1719 if (ret < 0) {
1720 mlog_errno(ret);
1721 goto out;
1722 }
1723
1724 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
Joel Becker7d1fe092010-04-13 14:30:19 +08001725 res->sr_bit_offset, res->sr_bits);
Younger Liudb66c712014-04-03 14:47:10 -07001726 if (ret < 0) {
1727 ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1728 res->sr_bits,
1729 le16_to_cpu(gd->bg_chain));
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001730 mlog_errno(ret);
Younger Liudb66c712014-04-03 14:47:10 -07001731 }
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001732
Mark Fashehe49e2762010-08-13 15:15:17 -07001733out_loc_only:
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001734 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1735
1736out:
1737 brelse(group_bh);
1738
1739 return ret;
1740}
1741
Mark Fashehccd979b2005-12-15 14:31:24 -08001742static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001743 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001744 u32 bits_wanted,
1745 u32 min_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +08001746 struct ocfs2_suballoc_result *res,
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001747 u16 *bits_left)
Mark Fashehccd979b2005-12-15 14:31:24 -08001748{
1749 int status;
Joel Becker7d1fe092010-04-13 14:30:19 +08001750 u16 chain;
Mark Fashehccd979b2005-12-15 14:31:24 -08001751 u64 next_group;
Mark Fashehccd979b2005-12-15 14:31:24 -08001752 struct inode *alloc_inode = ac->ac_inode;
1753 struct buffer_head *group_bh = NULL;
1754 struct buffer_head *prev_group_bh = NULL;
1755 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1756 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1757 struct ocfs2_group_desc *bg;
1758
1759 chain = ac->ac_chain;
Tao Ma2f73e132011-02-22 08:22:33 +08001760 trace_ocfs2_search_chain_begin(
1761 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1762 bits_wanted, chain);
Mark Fashehccd979b2005-12-15 14:31:24 -08001763
Joel Becker68f64d42008-11-13 14:49:14 -08001764 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1765 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1766 &group_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001767 if (status < 0) {
1768 mlog_errno(status);
1769 goto bail;
1770 }
1771 bg = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001772
1773 status = -ENOSPC;
1774 /* for now, the chain search is a bit simplistic. We just use
1775 * the 1st group with any empty bits. */
Joel Becker1187c962008-09-03 20:03:39 -07001776 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1777 bits_wanted, min_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +08001778 ac->ac_max_block,
1779 res)) == -ENOSPC) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001780 if (!bg->bg_next_group)
1781 break;
Mark Fasheha81cb882008-10-07 14:25:16 -07001782
1783 brelse(prev_group_bh);
1784 prev_group_bh = NULL;
1785
Mark Fashehccd979b2005-12-15 14:31:24 -08001786 next_group = le64_to_cpu(bg->bg_next_group);
1787 prev_group_bh = group_bh;
1788 group_bh = NULL;
Joel Becker68f64d42008-11-13 14:49:14 -08001789 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1790 next_group, &group_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001791 if (status < 0) {
1792 mlog_errno(status);
1793 goto bail;
1794 }
1795 bg = (struct ocfs2_group_desc *) group_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001796 }
1797 if (status < 0) {
1798 if (status != -ENOSPC)
1799 mlog_errno(status);
1800 goto bail;
1801 }
1802
Tao Ma2f73e132011-02-22 08:22:33 +08001803 trace_ocfs2_search_chain_succ(
1804 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08001805
Joel Becker7d1fe092010-04-13 14:30:19 +08001806 res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001807
Joel Becker7d1fe092010-04-13 14:30:19 +08001808 BUG_ON(res->sr_bits == 0);
Joel Becker13e434c2010-03-26 10:08:27 +08001809 if (!status)
Joel Beckerba206632010-03-26 10:08:59 +08001810 ocfs2_bg_discontig_fix_result(ac, bg, res);
Joel Becker13e434c2010-03-26 10:08:27 +08001811
Mark Fashehe49e2762010-08-13 15:15:17 -07001812 /*
1813 * sr_bg_blkno might have been changed by
1814 * ocfs2_bg_discontig_fix_result
1815 */
1816 res->sr_bg_stable_blkno = group_bh->b_blocknr;
Mark Fashehccd979b2005-12-15 14:31:24 -08001817
1818 /*
1819 * Keep track of previous block descriptor read. When
1820 * we find a target, if we have read more than X
1821 * number of descriptors, and the target is reasonably
1822 * empty, relink him to top of his chain.
1823 *
1824 * We've read 0 extra blocks and only send one more to
1825 * the transaction, yet the next guy to search has a
1826 * much easier time.
1827 *
1828 * Do this *after* figuring out how many bits we're taking out
1829 * of our target group.
1830 */
Xiaowei.Hu309a85b2013-02-27 17:02:49 -08001831 if (!ac->ac_disable_chain_relink &&
Mark Fashehccd979b2005-12-15 14:31:24 -08001832 (prev_group_bh) &&
Joel Becker7d1fe092010-04-13 14:30:19 +08001833 (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001834 status = ocfs2_relink_block_group(handle, alloc_inode,
1835 ac->ac_bh, group_bh,
1836 prev_group_bh, chain);
1837 if (status < 0) {
1838 mlog_errno(status);
1839 goto bail;
1840 }
1841 }
1842
Mark Fashehe49e2762010-08-13 15:15:17 -07001843 if (ac->ac_find_loc_only)
1844 goto out_loc_only;
1845
Mark Fashehd5134982010-08-13 15:15:16 -07001846 status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1847 ac->ac_bh, res->sr_bits,
1848 chain);
1849 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001850 mlog_errno(status);
1851 goto bail;
1852 }
1853
Mark Fashehccd979b2005-12-15 14:31:24 -08001854 status = ocfs2_block_group_set_bits(handle,
1855 alloc_inode,
1856 bg,
1857 group_bh,
Joel Becker7d1fe092010-04-13 14:30:19 +08001858 res->sr_bit_offset,
1859 res->sr_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08001860 if (status < 0) {
Younger Liudb66c712014-04-03 14:47:10 -07001861 ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1862 ac->ac_bh, res->sr_bits, chain);
Mark Fashehccd979b2005-12-15 14:31:24 -08001863 mlog_errno(status);
1864 goto bail;
1865 }
1866
Tao Ma2f73e132011-02-22 08:22:33 +08001867 trace_ocfs2_search_chain_end(
1868 (unsigned long long)le64_to_cpu(fe->i_blkno),
1869 res->sr_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08001870
Mark Fashehe49e2762010-08-13 15:15:17 -07001871out_loc_only:
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001872 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001873bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001874 brelse(group_bh);
1875 brelse(prev_group_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001876
Tao Mac1e8d352011-03-07 16:43:21 +08001877 if (status)
1878 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001879 return status;
1880}
1881
1882/* will give out up to bits_wanted contiguous bits. */
Joel Beckeraa8f8e92010-03-26 10:08:07 +08001883static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001884 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001885 u32 bits_wanted,
1886 u32 min_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +08001887 struct ocfs2_suballoc_result *res)
Mark Fashehccd979b2005-12-15 14:31:24 -08001888{
1889 int status;
1890 u16 victim, i;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001891 u16 bits_left = 0;
Mark Fashehb2b6ebf2010-08-26 13:06:50 -07001892 u64 hint = ac->ac_last_group;
Mark Fashehccd979b2005-12-15 14:31:24 -08001893 struct ocfs2_chain_list *cl;
1894 struct ocfs2_dinode *fe;
1895
Mark Fashehccd979b2005-12-15 14:31:24 -08001896 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1897 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1898 BUG_ON(!ac->ac_bh);
1899
1900 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
Joel Becker10995aa2008-11-13 14:49:12 -08001901
1902 /* The bh was validated by the inode read during
1903 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1904 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1905
Mark Fashehccd979b2005-12-15 14:31:24 -08001906 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1907 le32_to_cpu(fe->id1.bitmap1.i_total)) {
Joel Beckeraa8f8e92010-03-26 10:08:07 +08001908 ocfs2_error(ac->ac_inode->i_sb,
1909 "Chain allocator dinode %llu has %u used "
Mark Fashehb06970532006-03-03 10:24:33 -08001910 "bits but only %u total.",
1911 (unsigned long long)le64_to_cpu(fe->i_blkno),
Mark Fashehccd979b2005-12-15 14:31:24 -08001912 le32_to_cpu(fe->id1.bitmap1.i_used),
1913 le32_to_cpu(fe->id1.bitmap1.i_total));
1914 status = -EIO;
1915 goto bail;
1916 }
1917
Mark Fashehb2b6ebf2010-08-26 13:06:50 -07001918 res->sr_bg_blkno = hint;
Joel Becker7d1fe092010-04-13 14:30:19 +08001919 if (res->sr_bg_blkno) {
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001920 /* Attempt to short-circuit the usual search mechanism
1921 * by jumping straight to the most recently used
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04001922 * allocation group. This helps us maintain some
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001923 * contiguousness across allocations. */
Mark Fashehda5cbf22006-10-06 18:34:35 -07001924 status = ocfs2_search_one_group(ac, handle, bits_wanted,
Joel Becker7d1fe092010-04-13 14:30:19 +08001925 min_bits, res, &bits_left);
1926 if (!status)
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001927 goto set_hint;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001928 if (status < 0 && status != -ENOSPC) {
1929 mlog_errno(status);
1930 goto bail;
1931 }
1932 }
1933
Mark Fashehccd979b2005-12-15 14:31:24 -08001934 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1935
1936 victim = ocfs2_find_victim_chain(cl);
1937 ac->ac_chain = victim;
Mark Fashehccd979b2005-12-15 14:31:24 -08001938
Joel Becker7d1fe092010-04-13 14:30:19 +08001939 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1940 res, &bits_left);
Mark Fashehb2b6ebf2010-08-26 13:06:50 -07001941 if (!status) {
1942 hint = ocfs2_group_from_res(res);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001943 goto set_hint;
Mark Fashehb2b6ebf2010-08-26 13:06:50 -07001944 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001945 if (status < 0 && status != -ENOSPC) {
1946 mlog_errno(status);
1947 goto bail;
1948 }
1949
Tao Ma2f73e132011-02-22 08:22:33 +08001950 trace_ocfs2_claim_suballoc_bits(victim);
Mark Fashehccd979b2005-12-15 14:31:24 -08001951
1952 /* If we didn't pick a good victim, then just default to
1953 * searching each chain in order. Don't allow chain relinking
1954 * because we only calculate enough journal credits for one
1955 * relink per alloc. */
Xiaowei.Hu309a85b2013-02-27 17:02:49 -08001956 ac->ac_disable_chain_relink = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08001957 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1958 if (i == victim)
1959 continue;
1960 if (!cl->cl_recs[i].c_free)
1961 continue;
1962
1963 ac->ac_chain = i;
Mark Fashehda5cbf22006-10-06 18:34:35 -07001964 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
Joel Becker7d1fe092010-04-13 14:30:19 +08001965 res, &bits_left);
Mark Fashehb2b6ebf2010-08-26 13:06:50 -07001966 if (!status) {
1967 hint = ocfs2_group_from_res(res);
Mark Fashehccd979b2005-12-15 14:31:24 -08001968 break;
Mark Fashehb2b6ebf2010-08-26 13:06:50 -07001969 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001970 if (status < 0 && status != -ENOSPC) {
1971 mlog_errno(status);
1972 goto bail;
1973 }
1974 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001975
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001976set_hint:
1977 if (status != -ENOSPC) {
1978 /* If the next search of this group is not likely to
1979 * yield a suitable extent, then we reset the last
1980 * group hint so as to not waste a disk read */
1981 if (bits_left < min_bits)
1982 ac->ac_last_group = 0;
1983 else
Mark Fashehb2b6ebf2010-08-26 13:06:50 -07001984 ac->ac_last_group = hint;
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001985 }
1986
1987bail:
Tao Mac1e8d352011-03-07 16:43:21 +08001988 if (status)
1989 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001990 return status;
1991}
1992
Joel Becker1ed9b772010-05-06 13:59:06 +08001993int ocfs2_claim_metadata(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001994 struct ocfs2_alloc_context *ac,
1995 u32 bits_wanted,
Joel Becker2b6cb572010-03-26 10:09:15 +08001996 u64 *suballoc_loc,
Mark Fashehccd979b2005-12-15 14:31:24 -08001997 u16 *suballoc_bit_start,
1998 unsigned int *num_bits,
1999 u64 *blkno_start)
2000{
2001 int status;
Joel Beckerba206632010-03-26 10:08:59 +08002002 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
Mark Fashehccd979b2005-12-15 14:31:24 -08002003
2004 BUG_ON(!ac);
2005 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2006 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
Mark Fashehccd979b2005-12-15 14:31:24 -08002007
Joel Beckeraa8f8e92010-03-26 10:08:07 +08002008 status = ocfs2_claim_suballoc_bits(ac,
Mark Fashehda5cbf22006-10-06 18:34:35 -07002009 handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08002010 bits_wanted,
2011 1,
Joel Becker7d1fe092010-04-13 14:30:19 +08002012 &res);
Mark Fashehccd979b2005-12-15 14:31:24 -08002013 if (status < 0) {
2014 mlog_errno(status);
2015 goto bail;
2016 }
Joel Becker1ed9b772010-05-06 13:59:06 +08002017 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
Mark Fashehccd979b2005-12-15 14:31:24 -08002018
Joel Becker2b6cb572010-03-26 10:09:15 +08002019 *suballoc_loc = res.sr_bg_blkno;
Joel Becker7d1fe092010-04-13 14:30:19 +08002020 *suballoc_bit_start = res.sr_bit_offset;
Joel Beckerba206632010-03-26 10:08:59 +08002021 *blkno_start = res.sr_blkno;
Joel Becker7d1fe092010-04-13 14:30:19 +08002022 ac->ac_bits_given += res.sr_bits;
2023 *num_bits = res.sr_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -08002024 status = 0;
2025bail:
Tao Mac1e8d352011-03-07 16:43:21 +08002026 if (status)
2027 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002028 return status;
2029}
2030
Tao Ma13821152009-02-25 00:53:23 +08002031static void ocfs2_init_inode_ac_group(struct inode *dir,
Tao Maabf1b3c2010-04-27 08:30:36 +08002032 struct buffer_head *parent_di_bh,
Tao Ma13821152009-02-25 00:53:23 +08002033 struct ocfs2_alloc_context *ac)
2034{
Tao Maabf1b3c2010-04-27 08:30:36 +08002035 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
Tao Ma13821152009-02-25 00:53:23 +08002036 /*
2037 * Try to allocate inodes from some specific group.
2038 *
2039 * If the parent dir has recorded the last group used in allocation,
2040 * cool, use it. Otherwise if we try to allocate new inode from the
2041 * same slot the parent dir belongs to, use the same chunk.
2042 *
2043 * We are very careful here to avoid the mistake of setting
2044 * ac_last_group to a group descriptor from a different (unlocked) slot.
2045 */
2046 if (OCFS2_I(dir)->ip_last_used_group &&
2047 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2048 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
Tao Maabf1b3c2010-04-27 08:30:36 +08002049 else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2050 if (di->i_suballoc_loc)
2051 ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2052 else
2053 ac->ac_last_group = ocfs2_which_suballoc_group(
2054 le64_to_cpu(di->i_blkno),
2055 le16_to_cpu(di->i_suballoc_bit));
2056 }
Tao Ma13821152009-02-25 00:53:23 +08002057}
2058
2059static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2060 struct ocfs2_alloc_context *ac)
2061{
2062 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2063 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2064}
2065
Mark Fashehe49e2762010-08-13 15:15:17 -07002066int ocfs2_find_new_inode_loc(struct inode *dir,
2067 struct buffer_head *parent_fe_bh,
2068 struct ocfs2_alloc_context *ac,
2069 u64 *fe_blkno)
2070{
2071 int ret;
2072 handle_t *handle = NULL;
2073 struct ocfs2_suballoc_result *res;
2074
2075 BUG_ON(!ac);
2076 BUG_ON(ac->ac_bits_given != 0);
2077 BUG_ON(ac->ac_bits_wanted != 1);
2078 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2079
2080 res = kzalloc(sizeof(*res), GFP_NOFS);
2081 if (res == NULL) {
2082 ret = -ENOMEM;
2083 mlog_errno(ret);
2084 goto out;
2085 }
2086
2087 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2088
2089 /*
2090 * The handle started here is for chain relink. Alternatively,
2091 * we could just disable relink for these calls.
2092 */
2093 handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2094 if (IS_ERR(handle)) {
2095 ret = PTR_ERR(handle);
2096 handle = NULL;
2097 mlog_errno(ret);
2098 goto out;
2099 }
2100
2101 /*
2102 * This will instruct ocfs2_claim_suballoc_bits and
2103 * ocfs2_search_one_group to search but save actual allocation
2104 * for later.
2105 */
2106 ac->ac_find_loc_only = 1;
2107
2108 ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2109 if (ret < 0) {
2110 mlog_errno(ret);
2111 goto out;
2112 }
2113
2114 ac->ac_find_loc_priv = res;
2115 *fe_blkno = res->sr_blkno;
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07002116 ocfs2_update_inode_fsync_trans(handle, dir, 0);
Mark Fashehe49e2762010-08-13 15:15:17 -07002117out:
2118 if (handle)
2119 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2120
2121 if (ret)
2122 kfree(res);
2123
2124 return ret;
2125}
2126
2127int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2128 struct inode *dir,
2129 struct ocfs2_alloc_context *ac,
2130 u64 *suballoc_loc,
2131 u16 *suballoc_bit,
2132 u64 di_blkno)
2133{
2134 int ret;
2135 u16 chain;
2136 struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2137 struct buffer_head *bg_bh = NULL;
2138 struct ocfs2_group_desc *bg;
2139 struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2140
2141 /*
2142 * Since di_blkno is being passed back in, we check for any
2143 * inconsistencies which may have happened between
2144 * calls. These are code bugs as di_blkno is not expected to
2145 * change once returned from ocfs2_find_new_inode_loc()
2146 */
2147 BUG_ON(res->sr_blkno != di_blkno);
2148
2149 ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2150 res->sr_bg_stable_blkno, &bg_bh);
2151 if (ret) {
2152 mlog_errno(ret);
2153 goto out;
2154 }
2155
2156 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2157 chain = le16_to_cpu(bg->bg_chain);
2158
2159 ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2160 ac->ac_bh, res->sr_bits,
2161 chain);
2162 if (ret) {
2163 mlog_errno(ret);
2164 goto out;
2165 }
2166
2167 ret = ocfs2_block_group_set_bits(handle,
2168 ac->ac_inode,
2169 bg,
2170 bg_bh,
2171 res->sr_bit_offset,
2172 res->sr_bits);
2173 if (ret < 0) {
Younger Liudb66c712014-04-03 14:47:10 -07002174 ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2175 ac->ac_bh, res->sr_bits, chain);
Mark Fashehe49e2762010-08-13 15:15:17 -07002176 mlog_errno(ret);
2177 goto out;
2178 }
2179
Tao Ma2f73e132011-02-22 08:22:33 +08002180 trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2181 res->sr_bits);
Mark Fashehe49e2762010-08-13 15:15:17 -07002182
2183 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2184
2185 BUG_ON(res->sr_bits != 1);
2186
2187 *suballoc_loc = res->sr_bg_blkno;
2188 *suballoc_bit = res->sr_bit_offset;
2189 ac->ac_bits_given++;
2190 ocfs2_save_inode_ac_group(dir, ac);
2191
2192out:
2193 brelse(bg_bh);
2194
2195 return ret;
2196}
2197
Joel Becker1ed9b772010-05-06 13:59:06 +08002198int ocfs2_claim_new_inode(handle_t *handle,
Tao Ma13821152009-02-25 00:53:23 +08002199 struct inode *dir,
2200 struct buffer_head *parent_fe_bh,
Mark Fashehccd979b2005-12-15 14:31:24 -08002201 struct ocfs2_alloc_context *ac,
Joel Becker2b6cb572010-03-26 10:09:15 +08002202 u64 *suballoc_loc,
Mark Fashehccd979b2005-12-15 14:31:24 -08002203 u16 *suballoc_bit,
2204 u64 *fe_blkno)
2205{
2206 int status;
Joel Becker2b6cb572010-03-26 10:09:15 +08002207 struct ocfs2_suballoc_result res;
Mark Fashehccd979b2005-12-15 14:31:24 -08002208
Mark Fashehccd979b2005-12-15 14:31:24 -08002209 BUG_ON(!ac);
2210 BUG_ON(ac->ac_bits_given != 0);
2211 BUG_ON(ac->ac_bits_wanted != 1);
2212 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
Mark Fashehccd979b2005-12-15 14:31:24 -08002213
Tao Ma13821152009-02-25 00:53:23 +08002214 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2215
Joel Beckeraa8f8e92010-03-26 10:08:07 +08002216 status = ocfs2_claim_suballoc_bits(ac,
Mark Fashehda5cbf22006-10-06 18:34:35 -07002217 handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08002218 1,
2219 1,
Joel Becker7d1fe092010-04-13 14:30:19 +08002220 &res);
Mark Fashehccd979b2005-12-15 14:31:24 -08002221 if (status < 0) {
2222 mlog_errno(status);
2223 goto bail;
2224 }
Joel Becker1ed9b772010-05-06 13:59:06 +08002225 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
Mark Fashehccd979b2005-12-15 14:31:24 -08002226
Joel Becker7d1fe092010-04-13 14:30:19 +08002227 BUG_ON(res.sr_bits != 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08002228
Joel Becker2b6cb572010-03-26 10:09:15 +08002229 *suballoc_loc = res.sr_bg_blkno;
Joel Becker7d1fe092010-04-13 14:30:19 +08002230 *suballoc_bit = res.sr_bit_offset;
Joel Beckerba206632010-03-26 10:08:59 +08002231 *fe_blkno = res.sr_blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -08002232 ac->ac_bits_given++;
Tao Ma13821152009-02-25 00:53:23 +08002233 ocfs2_save_inode_ac_group(dir, ac);
Mark Fashehccd979b2005-12-15 14:31:24 -08002234 status = 0;
2235bail:
Tao Mac1e8d352011-03-07 16:43:21 +08002236 if (status)
2237 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002238 return status;
2239}
2240
2241/* translate a group desc. blkno and it's bitmap offset into
2242 * disk cluster offset. */
2243static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2244 u64 bg_blkno,
2245 u16 bg_bit_off)
2246{
2247 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2248 u32 cluster = 0;
2249
2250 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2251
2252 if (bg_blkno != osb->first_cluster_group_blkno)
2253 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2254 cluster += (u32) bg_bit_off;
2255 return cluster;
2256}
2257
2258/* given a cluster offset, calculate which block group it belongs to
2259 * and return that block offset. */
Tao Mad6590722007-12-18 15:47:03 +08002260u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
Mark Fashehccd979b2005-12-15 14:31:24 -08002261{
2262 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2263 u32 group_no;
2264
2265 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2266
2267 group_no = cluster / osb->bitmap_cpg;
2268 if (!group_no)
2269 return osb->first_cluster_group_blkno;
2270 return ocfs2_clusters_to_blocks(inode->i_sb,
2271 group_no * osb->bitmap_cpg);
2272}
2273
2274/* given the block number of a cluster start, calculate which cluster
2275 * group and descriptor bitmap offset that corresponds to. */
2276static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2277 u64 data_blkno,
2278 u64 *bg_blkno,
2279 u16 *bg_bit_off)
2280{
2281 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2282 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2283
2284 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2285
2286 *bg_blkno = ocfs2_which_cluster_group(inode,
2287 data_cluster);
2288
2289 if (*bg_blkno == osb->first_cluster_group_blkno)
2290 *bg_bit_off = (u16) data_cluster;
2291 else
2292 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2293 data_blkno - *bg_blkno);
2294}
2295
2296/*
2297 * min_bits - minimum contiguous chunk from this total allocation we
2298 * can handle. set to what we asked for originally for a full
2299 * contig. allocation, set to '1' to indicate we can deal with extents
2300 * of any size.
2301 */
Joel Becker1ed9b772010-05-06 13:59:06 +08002302int __ocfs2_claim_clusters(handle_t *handle,
Mark Fasheh415cb802007-09-16 20:10:16 -07002303 struct ocfs2_alloc_context *ac,
2304 u32 min_clusters,
2305 u32 max_clusters,
2306 u32 *cluster_start,
2307 u32 *num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08002308{
2309 int status;
Mark Fasheh415cb802007-09-16 20:10:16 -07002310 unsigned int bits_wanted = max_clusters;
Joel Beckerba206632010-03-26 10:08:59 +08002311 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
Joel Becker1ed9b772010-05-06 13:59:06 +08002312 struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002313
Mark Fashehccd979b2005-12-15 14:31:24 -08002314 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2315
2316 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2317 && ac->ac_which != OCFS2_AC_USE_MAIN);
Mark Fashehccd979b2005-12-15 14:31:24 -08002318
2319 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
Mark Fasheh33d5d382010-02-24 13:34:09 -08002320 WARN_ON(min_clusters > 1);
2321
Mark Fashehccd979b2005-12-15 14:31:24 -08002322 status = ocfs2_claim_local_alloc_bits(osb,
2323 handle,
2324 ac,
2325 bits_wanted,
2326 cluster_start,
2327 num_clusters);
2328 if (!status)
2329 atomic_inc(&osb->alloc_stats.local_data);
2330 } else {
2331 if (min_clusters > (osb->bitmap_cpg - 1)) {
2332 /* The only paths asking for contiguousness
2333 * should know about this already. */
Sunil Mushran2fbe8d12007-12-20 14:58:11 -08002334 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2335 "group bitmap size %u!\n", min_clusters,
2336 osb->bitmap_cpg);
Mark Fashehccd979b2005-12-15 14:31:24 -08002337 status = -ENOSPC;
2338 goto bail;
2339 }
2340 /* clamp the current request down to a realistic size. */
2341 if (bits_wanted > (osb->bitmap_cpg - 1))
2342 bits_wanted = osb->bitmap_cpg - 1;
2343
Joel Beckeraa8f8e92010-03-26 10:08:07 +08002344 status = ocfs2_claim_suballoc_bits(ac,
Mark Fashehda5cbf22006-10-06 18:34:35 -07002345 handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08002346 bits_wanted,
2347 min_clusters,
Joel Becker7d1fe092010-04-13 14:30:19 +08002348 &res);
Mark Fashehccd979b2005-12-15 14:31:24 -08002349 if (!status) {
Joel Beckerba206632010-03-26 10:08:59 +08002350 BUG_ON(res.sr_blkno); /* cluster alloc can't set */
Mark Fashehccd979b2005-12-15 14:31:24 -08002351 *cluster_start =
2352 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
Joel Becker7d1fe092010-04-13 14:30:19 +08002353 res.sr_bg_blkno,
2354 res.sr_bit_offset);
Mark Fashehccd979b2005-12-15 14:31:24 -08002355 atomic_inc(&osb->alloc_stats.bitmap_data);
Tao Ma47119542010-04-22 14:09:15 +08002356 *num_clusters = res.sr_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -08002357 }
2358 }
2359 if (status < 0) {
2360 if (status != -ENOSPC)
2361 mlog_errno(status);
2362 goto bail;
2363 }
2364
Tao Ma47119542010-04-22 14:09:15 +08002365 ac->ac_bits_given += *num_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08002366
2367bail:
Tao Mac1e8d352011-03-07 16:43:21 +08002368 if (status)
2369 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002370 return status;
2371}
2372
Joel Becker1ed9b772010-05-06 13:59:06 +08002373int ocfs2_claim_clusters(handle_t *handle,
Mark Fasheh415cb802007-09-16 20:10:16 -07002374 struct ocfs2_alloc_context *ac,
2375 u32 min_clusters,
2376 u32 *cluster_start,
2377 u32 *num_clusters)
2378{
2379 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2380
Joel Becker1ed9b772010-05-06 13:59:06 +08002381 return __ocfs2_claim_clusters(handle, ac, min_clusters,
Mark Fasheh415cb802007-09-16 20:10:16 -07002382 bits_wanted, cluster_start, num_clusters);
2383}
2384
Mark Fashehb4414ee2010-03-11 18:31:09 -08002385static int ocfs2_block_group_clear_bits(handle_t *handle,
2386 struct inode *alloc_inode,
2387 struct ocfs2_group_desc *bg,
2388 struct buffer_head *group_bh,
2389 unsigned int bit_off,
2390 unsigned int num_bits,
2391 void (*undo_fn)(unsigned int bit,
2392 unsigned long *bmap))
Mark Fashehccd979b2005-12-15 14:31:24 -08002393{
2394 int status;
2395 unsigned int tmp;
Mark Fashehccd979b2005-12-15 14:31:24 -08002396 struct ocfs2_group_desc *undo_bg = NULL;
2397
Joel Becker42035302008-11-13 14:49:15 -08002398 /* The caller got this descriptor from
2399 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2400 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
Mark Fashehccd979b2005-12-15 14:31:24 -08002401
Tao Ma2f73e132011-02-22 08:22:33 +08002402 trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08002403
Mark Fashehb4414ee2010-03-11 18:31:09 -08002404 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
Joel Becker0cf2f762009-02-12 16:41:25 -08002405 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
Mark Fashehb4414ee2010-03-11 18:31:09 -08002406 group_bh,
2407 undo_fn ?
2408 OCFS2_JOURNAL_ACCESS_UNDO :
2409 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08002410 if (status < 0) {
2411 mlog_errno(status);
2412 goto bail;
2413 }
2414
Mark Fashehb4414ee2010-03-11 18:31:09 -08002415 if (undo_fn) {
Sunil Mushran94e41ec2009-06-19 14:45:54 -07002416 jbd_lock_bh_state(group_bh);
2417 undo_bg = (struct ocfs2_group_desc *)
2418 bh2jh(group_bh)->b_committed_data;
2419 BUG_ON(!undo_bg);
2420 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002421
2422 tmp = num_bits;
2423 while(tmp--) {
2424 ocfs2_clear_bit((bit_off + tmp),
2425 (unsigned long *) bg->bg_bitmap);
Mark Fashehb4414ee2010-03-11 18:31:09 -08002426 if (undo_fn)
2427 undo_fn(bit_off + tmp,
2428 (unsigned long *) undo_bg->bg_bitmap);
Mark Fashehccd979b2005-12-15 14:31:24 -08002429 }
2430 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
Srinivas Eeda9b5cd102010-10-05 15:53:06 -07002431 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2432 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2433 " count %u but claims %u are freed. num_bits %d",
2434 (unsigned long long)le64_to_cpu(bg->bg_blkno),
2435 le16_to_cpu(bg->bg_bits),
2436 le16_to_cpu(bg->bg_free_bits_count), num_bits);
2437 return -EROFS;
2438 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002439
Mark Fashehb4414ee2010-03-11 18:31:09 -08002440 if (undo_fn)
Sunil Mushran94e41ec2009-06-19 14:45:54 -07002441 jbd_unlock_bh_state(group_bh);
2442
Joel Beckerec20cec2010-03-19 14:13:52 -07002443 ocfs2_journal_dirty(handle, group_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002444bail:
2445 return status;
2446}
2447
2448/*
2449 * expects the suballoc inode to already be locked.
2450 */
Mark Fashehb4414ee2010-03-11 18:31:09 -08002451static int _ocfs2_free_suballoc_bits(handle_t *handle,
2452 struct inode *alloc_inode,
2453 struct buffer_head *alloc_bh,
2454 unsigned int start_bit,
2455 u64 bg_blkno,
2456 unsigned int count,
2457 void (*undo_fn)(unsigned int bit,
2458 unsigned long *bitmap))
Mark Fashehccd979b2005-12-15 14:31:24 -08002459{
2460 int status = 0;
2461 u32 tmp_used;
Mark Fashehccd979b2005-12-15 14:31:24 -08002462 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2463 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2464 struct buffer_head *group_bh = NULL;
2465 struct ocfs2_group_desc *group;
2466
Joel Becker10995aa2008-11-13 14:49:12 -08002467 /* The alloc_bh comes from ocfs2_free_dinode() or
2468 * ocfs2_free_clusters(). The callers have all locked the
2469 * allocator and gotten alloc_bh from the lock call. This
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002470 * validates the dinode buffer. Any corruption that has happened
Joel Becker10995aa2008-11-13 14:49:12 -08002471 * is a code bug. */
2472 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
Mark Fashehccd979b2005-12-15 14:31:24 -08002473 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2474
Tao Ma2f73e132011-02-22 08:22:33 +08002475 trace_ocfs2_free_suballoc_bits(
2476 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2477 (unsigned long long)bg_blkno,
2478 start_bit, count);
Mark Fashehccd979b2005-12-15 14:31:24 -08002479
Joel Becker68f64d42008-11-13 14:49:14 -08002480 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2481 &group_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002482 if (status < 0) {
2483 mlog_errno(status);
2484 goto bail;
2485 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002486 group = (struct ocfs2_group_desc *) group_bh->b_data;
Joel Becker68f64d42008-11-13 14:49:14 -08002487
Mark Fashehccd979b2005-12-15 14:31:24 -08002488 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2489
2490 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2491 group, group_bh,
Mark Fashehb4414ee2010-03-11 18:31:09 -08002492 start_bit, count, undo_fn);
Mark Fashehccd979b2005-12-15 14:31:24 -08002493 if (status < 0) {
2494 mlog_errno(status);
2495 goto bail;
2496 }
2497
Joel Becker0cf2f762009-02-12 16:41:25 -08002498 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2499 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08002500 if (status < 0) {
2501 mlog_errno(status);
2502 goto bail;
2503 }
2504
2505 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2506 count);
2507 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2508 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
Joel Beckerec20cec2010-03-19 14:13:52 -07002509 ocfs2_journal_dirty(handle, alloc_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002510
2511bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07002512 brelse(group_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002513
Tao Mac1e8d352011-03-07 16:43:21 +08002514 if (status)
2515 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002516 return status;
2517}
2518
Mark Fashehb4414ee2010-03-11 18:31:09 -08002519int ocfs2_free_suballoc_bits(handle_t *handle,
2520 struct inode *alloc_inode,
2521 struct buffer_head *alloc_bh,
2522 unsigned int start_bit,
2523 u64 bg_blkno,
2524 unsigned int count)
2525{
2526 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2527 start_bit, bg_blkno, count, NULL);
2528}
2529
Mark Fasheh1fabe142006-10-09 18:11:45 -07002530int ocfs2_free_dinode(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08002531 struct inode *inode_alloc_inode,
2532 struct buffer_head *inode_alloc_bh,
2533 struct ocfs2_dinode *di)
2534{
2535 u64 blk = le64_to_cpu(di->i_blkno);
2536 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2537 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2538
Tao Ma74380c42010-03-22 14:20:18 +08002539 if (di->i_suballoc_loc)
2540 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
Mark Fashehccd979b2005-12-15 14:31:24 -08002541 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2542 inode_alloc_bh, bit, bg_blkno, 1);
2543}
2544
Mark Fashehb4414ee2010-03-11 18:31:09 -08002545static int _ocfs2_free_clusters(handle_t *handle,
2546 struct inode *bitmap_inode,
2547 struct buffer_head *bitmap_bh,
2548 u64 start_blk,
2549 unsigned int num_clusters,
2550 void (*undo_fn)(unsigned int bit,
2551 unsigned long *bitmap))
Mark Fashehccd979b2005-12-15 14:31:24 -08002552{
2553 int status;
2554 u16 bg_start_bit;
2555 u64 bg_blkno;
2556 struct ocfs2_dinode *fe;
2557
2558 /* You can't ever have a contiguous set of clusters
2559 * bigger than a block group bitmap so we never have to worry
Tao Maef6b6892011-02-21 11:10:44 +08002560 * about looping on them.
2561 * This is expensive. We can safely remove once this stuff has
Mark Fashehccd979b2005-12-15 14:31:24 -08002562 * gotten tested really well. */
2563 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2564
2565 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2566
2567 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2568 &bg_start_bit);
2569
Tao Ma2f73e132011-02-22 08:22:33 +08002570 trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2571 (unsigned long long)start_blk,
2572 bg_start_bit, num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08002573
Mark Fashehb4414ee2010-03-11 18:31:09 -08002574 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2575 bg_start_bit, bg_blkno,
2576 num_clusters, undo_fn);
Mark Fasheh9c7af402008-07-28 18:02:53 -07002577 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002578 mlog_errno(status);
Mark Fasheh9c7af402008-07-28 18:02:53 -07002579 goto out;
2580 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002581
Mark Fasheh9c7af402008-07-28 18:02:53 -07002582 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2583 num_clusters);
2584
2585out:
Tao Mac1e8d352011-03-07 16:43:21 +08002586 if (status)
2587 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002588 return status;
2589}
2590
Mark Fashehb4414ee2010-03-11 18:31:09 -08002591int ocfs2_free_clusters(handle_t *handle,
2592 struct inode *bitmap_inode,
2593 struct buffer_head *bitmap_bh,
2594 u64 start_blk,
2595 unsigned int num_clusters)
2596{
2597 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2598 start_blk, num_clusters,
2599 _ocfs2_set_bit);
2600}
2601
2602/*
2603 * Give never-used clusters back to the global bitmap. We don't need
2604 * to protect these bits in the undo buffer.
2605 */
2606int ocfs2_release_clusters(handle_t *handle,
2607 struct inode *bitmap_inode,
2608 struct buffer_head *bitmap_bh,
2609 u64 start_blk,
2610 unsigned int num_clusters)
2611{
2612 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2613 start_blk, num_clusters,
2614 _ocfs2_clear_bit);
2615}
2616
Mark Fashehccd979b2005-12-15 14:31:24 -08002617static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2618{
2619 printk("Block Group:\n");
2620 printk("bg_signature: %s\n", bg->bg_signature);
2621 printk("bg_size: %u\n", bg->bg_size);
2622 printk("bg_bits: %u\n", bg->bg_bits);
2623 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2624 printk("bg_chain: %u\n", bg->bg_chain);
2625 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
Mark Fashehb06970532006-03-03 10:24:33 -08002626 printk("bg_next_group: %llu\n",
2627 (unsigned long long)bg->bg_next_group);
2628 printk("bg_parent_dinode: %llu\n",
2629 (unsigned long long)bg->bg_parent_dinode);
2630 printk("bg_blkno: %llu\n",
2631 (unsigned long long)bg->bg_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08002632}
2633
2634static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2635{
2636 int i;
2637
Mark Fashehb06970532006-03-03 10:24:33 -08002638 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08002639 printk("i_signature: %s\n", fe->i_signature);
Mark Fashehb06970532006-03-03 10:24:33 -08002640 printk("i_size: %llu\n",
2641 (unsigned long long)fe->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08002642 printk("i_clusters: %u\n", fe->i_clusters);
2643 printk("i_generation: %u\n",
2644 le32_to_cpu(fe->i_generation));
2645 printk("id1.bitmap1.i_used: %u\n",
2646 le32_to_cpu(fe->id1.bitmap1.i_used));
2647 printk("id1.bitmap1.i_total: %u\n",
2648 le32_to_cpu(fe->id1.bitmap1.i_total));
2649 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
2650 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
2651 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
2652 printk("id2.i_chain.cl_next_free_rec: %u\n",
2653 fe->id2.i_chain.cl_next_free_rec);
2654 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2655 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
2656 fe->id2.i_chain.cl_recs[i].c_free);
2657 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2658 fe->id2.i_chain.cl_recs[i].c_total);
Mark Fashehb06970532006-03-03 10:24:33 -08002659 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2660 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08002661 }
2662}
Tao Mae7d4cb62008-08-18 17:38:44 +08002663
2664/*
2665 * For a given allocation, determine which allocators will need to be
2666 * accessed, and lock them, reserving the appropriate number of bits.
2667 *
2668 * Sparse file systems call this from ocfs2_write_begin_nolock()
2669 * and ocfs2_allocate_unwritten_extents().
2670 *
2671 * File systems which don't support holes call this from
2672 * ocfs2_extend_allocation().
2673 */
Joel Beckerf99b9b72008-08-20 19:36:33 -07002674int ocfs2_lock_allocators(struct inode *inode,
2675 struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +08002676 u32 clusters_to_add, u32 extents_to_split,
2677 struct ocfs2_alloc_context **data_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07002678 struct ocfs2_alloc_context **meta_ac)
Tao Mae7d4cb62008-08-18 17:38:44 +08002679{
2680 int ret = 0, num_free_extents;
2681 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2682 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2683
2684 *meta_ac = NULL;
2685 if (data_ac)
2686 *data_ac = NULL;
2687
2688 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2689
Joel Becker3d03a302009-02-12 17:49:26 -08002690 num_free_extents = ocfs2_num_free_extents(osb, et);
Tao Mae7d4cb62008-08-18 17:38:44 +08002691 if (num_free_extents < 0) {
2692 ret = num_free_extents;
2693 mlog_errno(ret);
2694 goto out;
2695 }
2696
2697 /*
2698 * Sparse allocation file systems need to be more conservative
2699 * with reserving room for expansion - the actual allocation
2700 * happens while we've got a journal handle open so re-taking
2701 * a cluster lock (because we ran out of room for another
2702 * extent) will violate ordering rules.
2703 *
2704 * Most of the time we'll only be seeing this 1 cluster at a time
2705 * anyway.
2706 *
2707 * Always lock for any unwritten extents - we might want to
2708 * add blocks during a split.
2709 */
2710 if (!num_free_extents ||
2711 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
Joel Beckerf99b9b72008-08-20 19:36:33 -07002712 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
Tao Mae7d4cb62008-08-18 17:38:44 +08002713 if (ret < 0) {
2714 if (ret != -ENOSPC)
2715 mlog_errno(ret);
2716 goto out;
2717 }
2718 }
2719
2720 if (clusters_to_add == 0)
2721 goto out;
2722
2723 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2724 if (ret < 0) {
2725 if (ret != -ENOSPC)
2726 mlog_errno(ret);
2727 goto out;
2728 }
2729
2730out:
2731 if (ret) {
2732 if (*meta_ac) {
2733 ocfs2_free_alloc_context(*meta_ac);
2734 *meta_ac = NULL;
2735 }
2736
2737 /*
2738 * We cannot have an error and a non null *data_ac.
2739 */
2740 }
2741
2742 return ret;
2743}
wengang wang6ca497a2009-03-06 21:29:10 +08002744
2745/*
2746 * Read the inode specified by blkno to get suballoc_slot and
2747 * suballoc_bit.
2748 */
2749static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
Tao Ma889f0042010-09-02 13:10:10 +08002750 u16 *suballoc_slot, u64 *group_blkno,
2751 u16 *suballoc_bit)
wengang wang6ca497a2009-03-06 21:29:10 +08002752{
2753 int status;
2754 struct buffer_head *inode_bh = NULL;
2755 struct ocfs2_dinode *inode_fe;
2756
Tao Ma2f73e132011-02-22 08:22:33 +08002757 trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
wengang wang6ca497a2009-03-06 21:29:10 +08002758
2759 /* dirty read disk */
2760 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2761 if (status < 0) {
Joel Becker5b09b502009-04-21 16:31:20 -07002762 mlog(ML_ERROR, "read block %llu failed %d\n",
2763 (unsigned long long)blkno, status);
wengang wang6ca497a2009-03-06 21:29:10 +08002764 goto bail;
2765 }
2766
2767 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2768 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
Joel Becker5b09b502009-04-21 16:31:20 -07002769 mlog(ML_ERROR, "invalid inode %llu requested\n",
2770 (unsigned long long)blkno);
wengang wang6ca497a2009-03-06 21:29:10 +08002771 status = -EINVAL;
2772 goto bail;
2773 }
2774
Tao Ma0fba8132009-03-19 05:08:43 +08002775 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
wengang wang6ca497a2009-03-06 21:29:10 +08002776 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2777 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
Joel Becker5b09b502009-04-21 16:31:20 -07002778 (unsigned long long)blkno,
2779 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
wengang wang6ca497a2009-03-06 21:29:10 +08002780 status = -EINVAL;
2781 goto bail;
2782 }
2783
2784 if (suballoc_slot)
2785 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2786 if (suballoc_bit)
2787 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
Tao Ma889f0042010-09-02 13:10:10 +08002788 if (group_blkno)
2789 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
wengang wang6ca497a2009-03-06 21:29:10 +08002790
2791bail:
2792 brelse(inode_bh);
2793
Tao Mac1e8d352011-03-07 16:43:21 +08002794 if (status)
2795 mlog_errno(status);
wengang wang6ca497a2009-03-06 21:29:10 +08002796 return status;
2797}
2798
2799/*
2800 * test whether bit is SET in allocator bitmap or not. on success, 0
2801 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2802 * is returned and *res is meaningless. Call this after you have
2803 * cluster locked against suballoc, or you may get a result based on
2804 * non-up2date contents
2805 */
2806static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2807 struct inode *suballoc,
Tao Ma889f0042010-09-02 13:10:10 +08002808 struct buffer_head *alloc_bh,
2809 u64 group_blkno, u64 blkno,
wengang wang6ca497a2009-03-06 21:29:10 +08002810 u16 bit, int *res)
2811{
Tao Maabf1b3c2010-04-27 08:30:36 +08002812 struct ocfs2_dinode *alloc_di;
wengang wang6ca497a2009-03-06 21:29:10 +08002813 struct ocfs2_group_desc *group;
2814 struct buffer_head *group_bh = NULL;
2815 u64 bg_blkno;
2816 int status;
2817
Tao Ma2f73e132011-02-22 08:22:33 +08002818 trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2819 (unsigned int)bit);
wengang wang6ca497a2009-03-06 21:29:10 +08002820
Tao Maabf1b3c2010-04-27 08:30:36 +08002821 alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2822 if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
wengang wang6ca497a2009-03-06 21:29:10 +08002823 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2824 (unsigned int)bit,
Tao Maabf1b3c2010-04-27 08:30:36 +08002825 ocfs2_bits_per_group(&alloc_di->id2.i_chain));
wengang wang6ca497a2009-03-06 21:29:10 +08002826 status = -EINVAL;
2827 goto bail;
2828 }
2829
Tao Ma889f0042010-09-02 13:10:10 +08002830 bg_blkno = group_blkno ? group_blkno :
2831 ocfs2_which_suballoc_group(blkno, bit);
Tao Maabf1b3c2010-04-27 08:30:36 +08002832 status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
wengang wang6ca497a2009-03-06 21:29:10 +08002833 &group_bh);
2834 if (status < 0) {
Joel Becker5b09b502009-04-21 16:31:20 -07002835 mlog(ML_ERROR, "read group %llu failed %d\n",
2836 (unsigned long long)bg_blkno, status);
wengang wang6ca497a2009-03-06 21:29:10 +08002837 goto bail;
2838 }
2839
2840 group = (struct ocfs2_group_desc *) group_bh->b_data;
2841 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2842
2843bail:
2844 brelse(group_bh);
2845
Tao Mac1e8d352011-03-07 16:43:21 +08002846 if (status)
2847 mlog_errno(status);
wengang wang6ca497a2009-03-06 21:29:10 +08002848 return status;
2849}
2850
2851/*
2852 * Test if the bit representing this inode (blkno) is set in the
2853 * suballocator.
2854 *
2855 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2856 *
2857 * In the event of failure, a negative value is returned and *res is
2858 * meaningless.
2859 *
2860 * Callers must make sure to hold nfs_sync_lock to prevent
2861 * ocfs2_delete_inode() on another node from accessing the same
2862 * suballocator concurrently.
2863 */
2864int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2865{
2866 int status;
Tao Ma889f0042010-09-02 13:10:10 +08002867 u64 group_blkno = 0;
wengang wang6ca497a2009-03-06 21:29:10 +08002868 u16 suballoc_bit = 0, suballoc_slot = 0;
2869 struct inode *inode_alloc_inode;
2870 struct buffer_head *alloc_bh = NULL;
2871
Tao Ma2f73e132011-02-22 08:22:33 +08002872 trace_ocfs2_test_inode_bit((unsigned long long)blkno);
wengang wang6ca497a2009-03-06 21:29:10 +08002873
2874 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
Tao Ma889f0042010-09-02 13:10:10 +08002875 &group_blkno, &suballoc_bit);
wengang wang6ca497a2009-03-06 21:29:10 +08002876 if (status < 0) {
2877 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2878 goto bail;
2879 }
2880
2881 inode_alloc_inode =
2882 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2883 suballoc_slot);
2884 if (!inode_alloc_inode) {
2885 /* the error code could be inaccurate, but we are not able to
2886 * get the correct one. */
2887 status = -EINVAL;
2888 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2889 (u32)suballoc_slot);
2890 goto bail;
2891 }
2892
2893 mutex_lock(&inode_alloc_inode->i_mutex);
2894 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2895 if (status < 0) {
2896 mutex_unlock(&inode_alloc_inode->i_mutex);
jiangyiwen7dc3e832014-04-03 14:47:12 -07002897 iput(inode_alloc_inode);
wengang wang6ca497a2009-03-06 21:29:10 +08002898 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2899 (u32)suballoc_slot, status);
2900 goto bail;
2901 }
2902
2903 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
Tao Ma889f0042010-09-02 13:10:10 +08002904 group_blkno, blkno, suballoc_bit, res);
wengang wang6ca497a2009-03-06 21:29:10 +08002905 if (status < 0)
2906 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2907
2908 ocfs2_inode_unlock(inode_alloc_inode, 0);
2909 mutex_unlock(&inode_alloc_inode->i_mutex);
2910
2911 iput(inode_alloc_inode);
2912 brelse(alloc_bh);
2913bail:
Tao Mac1e8d352011-03-07 16:43:21 +08002914 if (status)
2915 mlog_errno(status);
wengang wang6ca497a2009-03-06 21:29:10 +08002916 return status;
2917}