blob: aebeacd807c3b9505c5e9bb859008469985588e2 [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 * localalloc.c
5 *
6 * Node local data allocation
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/highmem.h>
30#include <linux/bitops.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 Becker13723d02008-10-17 19:25:01 -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"
Tao Maa04733d2011-02-22 07:56:45 +080045#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080046
47#include "buffer_head_io.h"
48
49#define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab))
50
Mark Fashehccd979b2005-12-15 14:31:24 -080051static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
52
53static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
54 struct ocfs2_dinode *alloc,
Mark Fashehd02f00c2009-12-07 13:10:48 -080055 u32 *numbits,
56 struct ocfs2_alloc_reservation *resv);
Mark Fashehccd979b2005-12-15 14:31:24 -080057
58static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
59
60static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -070061 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -080062 struct ocfs2_dinode *alloc,
63 struct inode *main_bm_inode,
64 struct buffer_head *main_bm_bh);
65
66static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -080067 struct ocfs2_alloc_context **ac,
68 struct inode **bitmap_inode,
69 struct buffer_head **bitmap_bh);
70
71static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -070072 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -080073 struct ocfs2_alloc_context *ac);
74
75static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
76 struct inode *local_alloc_inode);
77
Mark Fasheh6b820212010-04-05 18:17:14 -070078/*
79 * ocfs2_la_default_mb() - determine a default size, in megabytes of
80 * the local alloc.
81 *
82 * Generally, we'd like to pick as large a local alloc as
83 * possible. Performance on large workloads tends to scale
84 * proportionally to la size. In addition to that, the reservations
85 * code functions more efficiently as it can reserve more windows for
86 * write.
87 *
88 * Some things work against us when trying to choose a large local alloc:
89 *
90 * - We need to ensure our sizing is picked to leave enough space in
91 * group descriptors for other allocations (such as block groups,
92 * etc). Picking default sizes which are a multiple of 4 could help
93 * - block groups are allocated in 2mb and 4mb chunks.
94 *
95 * - Likewise, we don't want to starve other nodes of bits on small
96 * file systems. This can easily be taken care of by limiting our
97 * default to a reasonable size (256M) on larger cluster sizes.
98 *
99 * - Some file systems can't support very large sizes - 4k and 8k in
100 * particular are limited to less than 128 and 256 megabytes respectively.
101 *
102 * The following reference table shows group descriptor and local
103 * alloc maximums at various cluster sizes (4k blocksize)
104 *
105 * csize: 4K group: 126M la: 121M
106 * csize: 8K group: 252M la: 243M
107 * csize: 16K group: 504M la: 486M
108 * csize: 32K group: 1008M la: 972M
109 * csize: 64K group: 2016M la: 1944M
110 * csize: 128K group: 4032M la: 3888M
111 * csize: 256K group: 8064M la: 7776M
112 * csize: 512K group: 16128M la: 15552M
113 * csize: 1024K group: 32256M la: 31104M
114 */
115#define OCFS2_LA_MAX_DEFAULT_MB 256
116#define OCFS2_LA_OLD_DEFAULT 8
117unsigned int ocfs2_la_default_mb(struct ocfs2_super *osb)
118{
119 unsigned int la_mb;
120 unsigned int gd_mb;
Tao Ma1739da42010-06-09 16:43:05 +0800121 unsigned int la_max_mb;
Mark Fasheh6b820212010-04-05 18:17:14 -0700122 unsigned int megs_per_slot;
123 struct super_block *sb = osb->sb;
124
125 gd_mb = ocfs2_clusters_to_megabytes(osb->sb,
Tao Ma85718822010-04-13 14:38:06 +0800126 8 * ocfs2_group_bitmap_size(sb, 0, osb->s_feature_incompat));
Mark Fasheh6b820212010-04-05 18:17:14 -0700127
128 /*
129 * This takes care of files systems with very small group
130 * descriptors - 512 byte blocksize at cluster sizes lower
131 * than 16K and also 1k blocksize with 4k cluster size.
132 */
133 if ((sb->s_blocksize == 512 && osb->s_clustersize <= 8192)
134 || (sb->s_blocksize == 1024 && osb->s_clustersize == 4096))
135 return OCFS2_LA_OLD_DEFAULT;
136
137 /*
138 * Leave enough room for some block groups and make the final
139 * value we work from a multiple of 4.
140 */
141 gd_mb -= 16;
142 gd_mb &= 0xFFFFFFFB;
143
144 la_mb = gd_mb;
145
146 /*
147 * Keep window sizes down to a reasonable default
148 */
149 if (la_mb > OCFS2_LA_MAX_DEFAULT_MB) {
150 /*
151 * Some clustersize / blocksize combinations will have
152 * given us a larger than OCFS2_LA_MAX_DEFAULT_MB
153 * default size, but get poor distribution when
154 * limited to exactly 256 megabytes.
155 *
156 * As an example, 16K clustersize at 4K blocksize
157 * gives us a cluster group size of 504M. Paring the
158 * local alloc size down to 256 however, would give us
159 * only one window and around 200MB left in the
160 * cluster group. Instead, find the first size below
161 * 256 which would give us an even distribution.
162 *
163 * Larger cluster group sizes actually work out pretty
164 * well when pared to 256, so we don't have to do this
165 * for any group that fits more than two
166 * OCFS2_LA_MAX_DEFAULT_MB windows.
167 */
168 if (gd_mb > (2 * OCFS2_LA_MAX_DEFAULT_MB))
169 la_mb = 256;
170 else {
171 unsigned int gd_mult = gd_mb;
172
173 while (gd_mult > 256)
174 gd_mult = gd_mult >> 1;
175
176 la_mb = gd_mult;
177 }
178 }
179
180 megs_per_slot = osb->osb_clusters_at_boot / osb->max_slots;
181 megs_per_slot = ocfs2_clusters_to_megabytes(osb->sb, megs_per_slot);
182 /* Too many nodes, too few disk clusters. */
183 if (megs_per_slot < la_mb)
184 la_mb = megs_per_slot;
185
Tao Ma1739da42010-06-09 16:43:05 +0800186 /* We can't store more bits than we can in a block. */
187 la_max_mb = ocfs2_clusters_to_megabytes(osb->sb,
188 ocfs2_local_alloc_size(sb) * 8);
189 if (la_mb > la_max_mb)
190 la_mb = la_max_mb;
191
Mark Fasheh6b820212010-04-05 18:17:14 -0700192 return la_mb;
193}
194
Mark Fasheh73c8a802010-04-05 18:17:13 -0700195void ocfs2_la_set_sizes(struct ocfs2_super *osb, int requested_mb)
196{
197 struct super_block *sb = osb->sb;
Mark Fasheh6b820212010-04-05 18:17:14 -0700198 unsigned int la_default_mb = ocfs2_la_default_mb(osb);
Mark Fasheh73c8a802010-04-05 18:17:13 -0700199 unsigned int la_max_mb;
200
201 la_max_mb = ocfs2_clusters_to_megabytes(sb,
202 ocfs2_local_alloc_size(sb) * 8);
203
Tao Maa04733d2011-02-22 07:56:45 +0800204 trace_ocfs2_la_set_sizes(requested_mb, la_max_mb, la_default_mb);
Mark Fasheh73c8a802010-04-05 18:17:13 -0700205
206 if (requested_mb == -1) {
207 /* No user request - use defaults */
208 osb->local_alloc_default_bits =
209 ocfs2_megabytes_to_clusters(sb, la_default_mb);
210 } else if (requested_mb > la_max_mb) {
211 /* Request is too big, we give the maximum available */
212 osb->local_alloc_default_bits =
213 ocfs2_megabytes_to_clusters(sb, la_max_mb);
214 } else {
215 osb->local_alloc_default_bits =
216 ocfs2_megabytes_to_clusters(sb, requested_mb);
217 }
218
219 osb->local_alloc_bits = osb->local_alloc_default_bits;
220}
221
Mark Fasheh9c7af402008-07-28 18:02:53 -0700222static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb)
223{
224 return (osb->local_alloc_state == OCFS2_LA_THROTTLED ||
225 osb->local_alloc_state == OCFS2_LA_ENABLED);
226}
227
228void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
229 unsigned int num_clusters)
230{
231 spin_lock(&osb->osb_lock);
232 if (osb->local_alloc_state == OCFS2_LA_DISABLED ||
233 osb->local_alloc_state == OCFS2_LA_THROTTLED)
234 if (num_clusters >= osb->local_alloc_default_bits) {
235 cancel_delayed_work(&osb->la_enable_wq);
236 osb->local_alloc_state = OCFS2_LA_ENABLED;
237 }
238 spin_unlock(&osb->osb_lock);
239}
240
241void ocfs2_la_enable_worker(struct work_struct *work)
242{
243 struct ocfs2_super *osb =
244 container_of(work, struct ocfs2_super,
245 la_enable_wq.work);
246 spin_lock(&osb->osb_lock);
247 osb->local_alloc_state = OCFS2_LA_ENABLED;
248 spin_unlock(&osb->osb_lock);
249}
250
Mark Fashehccd979b2005-12-15 14:31:24 -0800251/*
252 * Tell us whether a given allocation should use the local alloc
253 * file. Otherwise, it has to go to the main bitmap.
Mark Fasheh9c7af402008-07-28 18:02:53 -0700254 *
255 * This function does semi-dirty reads of local alloc size and state!
256 * This is ok however, as the values are re-checked once under mutex.
Mark Fashehccd979b2005-12-15 14:31:24 -0800257 */
258int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
259{
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800260 int ret = 0;
Mark Fasheh9c7af402008-07-28 18:02:53 -0700261 int la_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -0800262
Mark Fasheh9c7af402008-07-28 18:02:53 -0700263 spin_lock(&osb->osb_lock);
264 la_bits = osb->local_alloc_bits;
265
266 if (!ocfs2_la_state_enabled(osb))
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800267 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800268
269 /* la_bits should be at least twice the size (in clusters) of
270 * a new block group. We want to be sure block group
271 * allocations go through the local alloc, so allow an
272 * allocation to take up to half the bitmap. */
273 if (bits > (la_bits / 2))
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800274 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800275
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800276 ret = 1;
277bail:
Tao Maa04733d2011-02-22 07:56:45 +0800278 trace_ocfs2_alloc_should_use_local(
279 (unsigned long long)bits, osb->local_alloc_state, la_bits, ret);
Mark Fasheh9c7af402008-07-28 18:02:53 -0700280 spin_unlock(&osb->osb_lock);
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800281 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800282}
283
284int ocfs2_load_local_alloc(struct ocfs2_super *osb)
285{
286 int status = 0;
287 struct ocfs2_dinode *alloc = NULL;
288 struct buffer_head *alloc_bh = NULL;
289 u32 num_used;
290 struct inode *inode = NULL;
291 struct ocfs2_local_alloc *la;
292
Mark Fashehebcee4b2008-07-28 14:55:20 -0700293 if (osb->local_alloc_bits == 0)
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800294 goto bail;
295
Mark Fashehebcee4b2008-07-28 14:55:20 -0700296 if (osb->local_alloc_bits >= osb->bitmap_cpg) {
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800297 mlog(ML_NOTICE, "Requested local alloc window %d is larger "
298 "than max possible %u. Using defaults.\n",
Mark Fashehebcee4b2008-07-28 14:55:20 -0700299 osb->local_alloc_bits, (osb->bitmap_cpg - 1));
300 osb->local_alloc_bits =
301 ocfs2_megabytes_to_clusters(osb->sb,
Mark Fasheh6b820212010-04-05 18:17:14 -0700302 ocfs2_la_default_mb(osb));
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800303 }
304
Mark Fashehccd979b2005-12-15 14:31:24 -0800305 /* read the alloc off disk */
306 inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
307 osb->slot_num);
308 if (!inode) {
309 status = -EINVAL;
310 mlog_errno(status);
311 goto bail;
312 }
313
Joel Beckerb657c952008-11-13 14:49:11 -0800314 status = ocfs2_read_inode_block_full(inode, &alloc_bh,
315 OCFS2_BH_IGNORE_CACHE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800316 if (status < 0) {
317 mlog_errno(status);
318 goto bail;
319 }
320
321 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
322 la = OCFS2_LOCAL_ALLOC(alloc);
323
324 if (!(le32_to_cpu(alloc->i_flags) &
325 (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
Mark Fashehb06970532006-03-03 10:24:33 -0800326 mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
327 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800328 status = -EINVAL;
329 goto bail;
330 }
331
332 if ((la->la_size == 0) ||
333 (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
334 mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
335 le16_to_cpu(la->la_size));
336 status = -EINVAL;
337 goto bail;
338 }
339
340 /* do a little verification. */
341 num_used = ocfs2_local_alloc_count_bits(alloc);
342
343 /* hopefully the local alloc has always been recovered before
344 * we load it. */
345 if (num_used
346 || alloc->id1.bitmap1.i_used
347 || alloc->id1.bitmap1.i_total
348 || la->la_bm_off)
349 mlog(ML_ERROR, "Local alloc hasn't been recovered!\n"
350 "found = %u, set = %u, taken = %u, off = %u\n",
351 num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
352 le32_to_cpu(alloc->id1.bitmap1.i_total),
353 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
354
355 osb->local_alloc_bh = alloc_bh;
356 osb->local_alloc_state = OCFS2_LA_ENABLED;
357
358bail:
359 if (status < 0)
Mark Fasheha81cb882008-10-07 14:25:16 -0700360 brelse(alloc_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800361 if (inode)
362 iput(inode);
363
Tao Maa04733d2011-02-22 07:56:45 +0800364 trace_ocfs2_load_local_alloc(osb->local_alloc_bits);
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800365
Tao Mac1e8d352011-03-07 16:43:21 +0800366 if (status)
367 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800368 return status;
369}
370
371/*
372 * return any unused bits to the bitmap and write out a clean
373 * local_alloc.
374 *
375 * local_alloc_bh is optional. If not passed, we will simply use the
376 * one off osb. If you do pass it however, be warned that it *will* be
377 * returned brelse'd and NULL'd out.*/
378void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
379{
380 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700381 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800382 struct inode *local_alloc_inode = NULL;
383 struct buffer_head *bh = NULL;
384 struct buffer_head *main_bm_bh = NULL;
385 struct inode *main_bm_inode = NULL;
386 struct ocfs2_dinode *alloc_copy = NULL;
387 struct ocfs2_dinode *alloc = NULL;
388
Mark Fasheh9c7af402008-07-28 18:02:53 -0700389 cancel_delayed_work(&osb->la_enable_wq);
390 flush_workqueue(ocfs2_wq);
391
Mark Fashehccd979b2005-12-15 14:31:24 -0800392 if (osb->local_alloc_state == OCFS2_LA_UNUSED)
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700393 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800394
395 local_alloc_inode =
396 ocfs2_get_system_file_inode(osb,
397 LOCAL_ALLOC_SYSTEM_INODE,
398 osb->slot_num);
399 if (!local_alloc_inode) {
400 status = -ENOENT;
401 mlog_errno(status);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700402 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800403 }
404
405 osb->local_alloc_state = OCFS2_LA_DISABLED;
406
Mark Fashehd02f00c2009-12-07 13:10:48 -0800407 ocfs2_resmap_uninit(&osb->osb_la_resmap);
408
Mark Fashehccd979b2005-12-15 14:31:24 -0800409 main_bm_inode = ocfs2_get_system_file_inode(osb,
410 GLOBAL_BITMAP_SYSTEM_INODE,
411 OCFS2_INVALID_SLOT);
412 if (!main_bm_inode) {
413 status = -EINVAL;
414 mlog_errno(status);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700415 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800416 }
417
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700418 mutex_lock(&main_bm_inode->i_mutex);
419
Mark Fashehe63aecb62007-10-18 15:30:42 -0700420 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800421 if (status < 0) {
422 mlog_errno(status);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700423 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -0800424 }
425
426 /* WINDOW_MOVE_CREDITS is a bit heavy... */
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700427 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800428 if (IS_ERR(handle)) {
429 mlog_errno(PTR_ERR(handle));
430 handle = NULL;
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700431 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800432 }
433
434 bh = osb->local_alloc_bh;
435 alloc = (struct ocfs2_dinode *) bh->b_data;
436
Sunil Mushran4ba1c5b2008-04-18 15:03:59 -0700437 alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800438 if (!alloc_copy) {
439 status = -ENOMEM;
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700440 goto out_commit;
Mark Fashehccd979b2005-12-15 14:31:24 -0800441 }
442 memcpy(alloc_copy, alloc, bh->b_size);
443
Joel Becker0cf2f762009-02-12 16:41:25 -0800444 status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode),
445 bh, OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800446 if (status < 0) {
447 mlog_errno(status);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700448 goto out_commit;
Mark Fashehccd979b2005-12-15 14:31:24 -0800449 }
450
451 ocfs2_clear_local_alloc(alloc);
Joel Beckerec20cec2010-03-19 14:13:52 -0700452 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800453
454 brelse(bh);
455 osb->local_alloc_bh = NULL;
456 osb->local_alloc_state = OCFS2_LA_UNUSED;
457
458 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
459 main_bm_inode, main_bm_bh);
460 if (status < 0)
461 mlog_errno(status);
462
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700463out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700464 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800465
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700466out_unlock:
Mark Fasheha81cb882008-10-07 14:25:16 -0700467 brelse(main_bm_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800468
Mark Fashehe63aecb62007-10-18 15:30:42 -0700469 ocfs2_inode_unlock(main_bm_inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800470
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700471out_mutex:
472 mutex_unlock(&main_bm_inode->i_mutex);
473 iput(main_bm_inode);
474
475out:
Mark Fashehccd979b2005-12-15 14:31:24 -0800476 if (local_alloc_inode)
477 iput(local_alloc_inode);
478
Tim Gardnerd787ab02013-02-21 16:42:44 -0800479 kfree(alloc_copy);
Mark Fashehccd979b2005-12-15 14:31:24 -0800480}
481
482/*
483 * We want to free the bitmap bits outside of any recovery context as
484 * we'll need a cluster lock to do so, but we must clear the local
485 * alloc before giving up the recovered nodes journal. To solve this,
486 * we kmalloc a copy of the local alloc before it's change for the
487 * caller to process with ocfs2_complete_local_alloc_recovery
488 */
489int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
490 int slot_num,
491 struct ocfs2_dinode **alloc_copy)
492{
493 int status = 0;
494 struct buffer_head *alloc_bh = NULL;
495 struct inode *inode = NULL;
496 struct ocfs2_dinode *alloc;
497
Tao Maa04733d2011-02-22 07:56:45 +0800498 trace_ocfs2_begin_local_alloc_recovery(slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -0800499
500 *alloc_copy = NULL;
501
502 inode = ocfs2_get_system_file_inode(osb,
503 LOCAL_ALLOC_SYSTEM_INODE,
504 slot_num);
505 if (!inode) {
506 status = -EINVAL;
507 mlog_errno(status);
508 goto bail;
509 }
510
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800511 mutex_lock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800512
Joel Beckerb657c952008-11-13 14:49:11 -0800513 status = ocfs2_read_inode_block_full(inode, &alloc_bh,
514 OCFS2_BH_IGNORE_CACHE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800515 if (status < 0) {
516 mlog_errno(status);
517 goto bail;
518 }
519
520 *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
521 if (!(*alloc_copy)) {
522 status = -ENOMEM;
523 goto bail;
524 }
525 memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
526
527 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
528 ocfs2_clear_local_alloc(alloc);
529
Joel Becker13723d02008-10-17 19:25:01 -0700530 ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -0800531 status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800532 if (status < 0)
533 mlog_errno(status);
534
535bail:
Tim Gardnerd787ab02013-02-21 16:42:44 -0800536 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800537 kfree(*alloc_copy);
538 *alloc_copy = NULL;
539 }
540
Mark Fasheha81cb882008-10-07 14:25:16 -0700541 brelse(alloc_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800542
543 if (inode) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800544 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800545 iput(inode);
546 }
547
Tao Mac1e8d352011-03-07 16:43:21 +0800548 if (status)
549 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800550 return status;
551}
552
553/*
554 * Step 2: By now, we've completed the journal recovery, we've stamped
555 * a clean local alloc on disk and dropped the node out of the
556 * recovery map. Dlm locks will no longer stall, so lets clear out the
557 * main bitmap.
558 */
559int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
560 struct ocfs2_dinode *alloc)
561{
562 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700563 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800564 struct buffer_head *main_bm_bh = NULL;
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700565 struct inode *main_bm_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -0800566
Mark Fashehccd979b2005-12-15 14:31:24 -0800567 main_bm_inode = ocfs2_get_system_file_inode(osb,
568 GLOBAL_BITMAP_SYSTEM_INODE,
569 OCFS2_INVALID_SLOT);
570 if (!main_bm_inode) {
571 status = -EINVAL;
572 mlog_errno(status);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700573 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800574 }
575
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700576 mutex_lock(&main_bm_inode->i_mutex);
577
Mark Fashehe63aecb62007-10-18 15:30:42 -0700578 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800579 if (status < 0) {
580 mlog_errno(status);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700581 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -0800582 }
583
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700584 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800585 if (IS_ERR(handle)) {
586 status = PTR_ERR(handle);
587 handle = NULL;
588 mlog_errno(status);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700589 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800590 }
591
592 /* we want the bitmap change to be recorded on disk asap */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700593 handle->h_sync = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800594
595 status = ocfs2_sync_local_to_main(osb, handle, alloc,
596 main_bm_inode, main_bm_bh);
597 if (status < 0)
598 mlog_errno(status);
599
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700600 ocfs2_commit_trans(osb, handle);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700601
602out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700603 ocfs2_inode_unlock(main_bm_inode, 1);
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700604
605out_mutex:
606 mutex_unlock(&main_bm_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800607
Mark Fasheha81cb882008-10-07 14:25:16 -0700608 brelse(main_bm_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800609
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700610 iput(main_bm_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800611
Mark Fasheh8898a5a2006-10-05 15:42:08 -0700612out:
Tao Ma4d0ddb22008-03-05 16:11:46 +0800613 if (!status)
Tiger Yangb89c5422010-01-25 14:11:06 +0800614 ocfs2_init_steal_slots(osb);
Tao Mac1e8d352011-03-07 16:43:21 +0800615 if (status)
616 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800617 return status;
618}
619
620/*
Mark Fasheh9c7af402008-07-28 18:02:53 -0700621 * make sure we've got at least bits_wanted contiguous bits in the
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800622 * local alloc. You lose them when you drop i_mutex.
Mark Fashehccd979b2005-12-15 14:31:24 -0800623 *
624 * We will add ourselves to the transaction passed in, but may start
625 * our own in order to shift windows.
626 */
627int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -0800628 u32 bits_wanted,
629 struct ocfs2_alloc_context *ac)
630{
631 int status;
632 struct ocfs2_dinode *alloc;
633 struct inode *local_alloc_inode;
634 unsigned int free_bits;
635
Mark Fashehccd979b2005-12-15 14:31:24 -0800636 BUG_ON(!ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800637
638 local_alloc_inode =
639 ocfs2_get_system_file_inode(osb,
640 LOCAL_ALLOC_SYSTEM_INODE,
641 osb->slot_num);
642 if (!local_alloc_inode) {
643 status = -ENOENT;
644 mlog_errno(status);
645 goto bail;
646 }
Mark Fashehda5cbf22006-10-06 18:34:35 -0700647
648 mutex_lock(&local_alloc_inode->i_mutex);
649
Mark Fasheh9c7af402008-07-28 18:02:53 -0700650 /*
651 * We must double check state and allocator bits because
652 * another process may have changed them while holding i_mutex.
653 */
654 spin_lock(&osb->osb_lock);
655 if (!ocfs2_la_state_enabled(osb) ||
656 (bits_wanted > osb->local_alloc_bits)) {
657 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800658 status = -ENOSPC;
659 goto bail;
660 }
Mark Fasheh9c7af402008-07-28 18:02:53 -0700661 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800662
663 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
664
Joel Beckere407e392008-06-12 22:35:39 -0700665#ifdef CONFIG_OCFS2_DEBUG_FS
Mark Fashehccd979b2005-12-15 14:31:24 -0800666 if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
667 ocfs2_local_alloc_count_bits(alloc)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800668 ocfs2_error(osb->sb, "local alloc inode %llu says it has "
Mark Fashehccd979b2005-12-15 14:31:24 -0800669 "%u free bits, but a count shows %u",
Mark Fashehb06970532006-03-03 10:24:33 -0800670 (unsigned long long)le64_to_cpu(alloc->i_blkno),
Mark Fashehccd979b2005-12-15 14:31:24 -0800671 le32_to_cpu(alloc->id1.bitmap1.i_used),
672 ocfs2_local_alloc_count_bits(alloc));
673 status = -EIO;
674 goto bail;
675 }
Jan Kara5a58c3e2007-11-13 19:59:33 +0100676#endif
Mark Fashehccd979b2005-12-15 14:31:24 -0800677
678 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
679 le32_to_cpu(alloc->id1.bitmap1.i_used);
680 if (bits_wanted > free_bits) {
681 /* uhoh, window change time. */
682 status =
683 ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
684 if (status < 0) {
685 if (status != -ENOSPC)
686 mlog_errno(status);
687 goto bail;
688 }
Mark Fasheh9c7af402008-07-28 18:02:53 -0700689
690 /*
691 * Under certain conditions, the window slide code
692 * might have reduced the number of bits available or
693 * disabled the the local alloc entirely. Re-check
694 * here and return -ENOSPC if necessary.
695 */
696 status = -ENOSPC;
697 if (!ocfs2_la_state_enabled(osb))
698 goto bail;
699
700 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
701 le32_to_cpu(alloc->id1.bitmap1.i_used);
702 if (bits_wanted > free_bits)
703 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800704 }
705
Mark Fasheh8fccfc82007-05-09 17:34:26 -0700706 ac->ac_inode = local_alloc_inode;
Tao Maa4a48912008-03-03 17:12:30 +0800707 /* We should never use localalloc from another slot */
708 ac->ac_alloc_slot = osb->slot_num;
Mark Fasheh8fccfc82007-05-09 17:34:26 -0700709 ac->ac_which = OCFS2_AC_USE_LOCAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800710 get_bh(osb->local_alloc_bh);
711 ac->ac_bh = osb->local_alloc_bh;
Mark Fashehccd979b2005-12-15 14:31:24 -0800712 status = 0;
713bail:
Sunil Mushranbda02332007-09-21 11:41:43 -0700714 if (status < 0 && local_alloc_inode) {
715 mutex_unlock(&local_alloc_inode->i_mutex);
Mark Fasheh8fccfc82007-05-09 17:34:26 -0700716 iput(local_alloc_inode);
Sunil Mushranbda02332007-09-21 11:41:43 -0700717 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800718
Tao Maa04733d2011-02-22 07:56:45 +0800719 trace_ocfs2_reserve_local_alloc_bits(
720 (unsigned long long)ac->ac_max_block,
721 bits_wanted, osb->slot_num, status);
Sunil Mushran2fbe8d12007-12-20 14:58:11 -0800722
Tao Mac1e8d352011-03-07 16:43:21 +0800723 if (status)
724 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800725 return status;
726}
727
728int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -0700729 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800730 struct ocfs2_alloc_context *ac,
Mark Fasheh415cb802007-09-16 20:10:16 -0700731 u32 bits_wanted,
Mark Fashehccd979b2005-12-15 14:31:24 -0800732 u32 *bit_off,
733 u32 *num_bits)
734{
735 int status, start;
736 struct inode *local_alloc_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -0800737 void *bitmap;
738 struct ocfs2_dinode *alloc;
739 struct ocfs2_local_alloc *la;
740
Mark Fashehccd979b2005-12-15 14:31:24 -0800741 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
742
Mark Fashehccd979b2005-12-15 14:31:24 -0800743 local_alloc_inode = ac->ac_inode;
744 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
745 la = OCFS2_LOCAL_ALLOC(alloc);
746
Mark Fashehd02f00c2009-12-07 13:10:48 -0800747 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted,
748 ac->ac_resv);
Mark Fashehccd979b2005-12-15 14:31:24 -0800749 if (start == -1) {
750 /* TODO: Shouldn't we just BUG here? */
751 status = -ENOSPC;
752 mlog_errno(status);
753 goto bail;
754 }
755
756 bitmap = la->la_bitmap;
757 *bit_off = le32_to_cpu(la->la_bm_off) + start;
Mark Fashehccd979b2005-12-15 14:31:24 -0800758 *num_bits = bits_wanted;
759
Joel Becker0cf2f762009-02-12 16:41:25 -0800760 status = ocfs2_journal_access_di(handle,
761 INODE_CACHE(local_alloc_inode),
Joel Becker13723d02008-10-17 19:25:01 -0700762 osb->local_alloc_bh,
763 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800764 if (status < 0) {
765 mlog_errno(status);
766 goto bail;
767 }
768
Mark Fashehd02f00c2009-12-07 13:10:48 -0800769 ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start,
770 bits_wanted);
771
Mark Fashehccd979b2005-12-15 14:31:24 -0800772 while(bits_wanted--)
773 ocfs2_set_bit(start++, bitmap);
774
Marcin Slusarz0dd32562008-02-13 00:06:18 +0100775 le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
Joel Beckerec20cec2010-03-19 14:13:52 -0700776 ocfs2_journal_dirty(handle, osb->local_alloc_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800777
Mark Fashehccd979b2005-12-15 14:31:24 -0800778bail:
Tao Mac1e8d352011-03-07 16:43:21 +0800779 if (status)
780 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800781 return status;
782}
783
784static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
785{
Akinobu Mitaa75613e2012-07-30 14:41:03 -0700786 u32 count;
Mark Fashehccd979b2005-12-15 14:31:24 -0800787 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
788
Akinobu Mitaa75613e2012-07-30 14:41:03 -0700789 count = memweight(la->la_bitmap, le16_to_cpu(la->la_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800790
Tao Maa04733d2011-02-22 07:56:45 +0800791 trace_ocfs2_local_alloc_count_bits(count);
Mark Fashehccd979b2005-12-15 14:31:24 -0800792 return count;
793}
794
795static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
Mark Fashehd02f00c2009-12-07 13:10:48 -0800796 struct ocfs2_dinode *alloc,
797 u32 *numbits,
798 struct ocfs2_alloc_reservation *resv)
Mark Fashehccd979b2005-12-15 14:31:24 -0800799{
800 int numfound, bitoff, left, startoff, lastzero;
Mark Fashehd02f00c2009-12-07 13:10:48 -0800801 int local_resv = 0;
802 struct ocfs2_alloc_reservation r;
Mark Fashehccd979b2005-12-15 14:31:24 -0800803 void *bitmap = NULL;
Mark Fashehd02f00c2009-12-07 13:10:48 -0800804 struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap;
Mark Fashehccd979b2005-12-15 14:31:24 -0800805
Mark Fashehccd979b2005-12-15 14:31:24 -0800806 if (!alloc->id1.bitmap1.i_total) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800807 bitoff = -1;
808 goto bail;
809 }
810
Mark Fashehd02f00c2009-12-07 13:10:48 -0800811 if (!resv) {
812 local_resv = 1;
813 ocfs2_resv_init_once(&r);
814 ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP);
815 resv = &r;
816 }
817
818 numfound = *numbits;
819 if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) {
820 if (numfound < *numbits)
821 *numbits = numfound;
822 goto bail;
823 }
824
825 /*
826 * Code error. While reservations are enabled, local
827 * allocation should _always_ go through them.
828 */
829 BUG_ON(osb->osb_resv_level != 0);
830
831 /*
832 * Reservations are disabled. Handle this the old way.
833 */
834
Mark Fashehccd979b2005-12-15 14:31:24 -0800835 bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
836
837 numfound = bitoff = startoff = 0;
838 lastzero = -1;
839 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
840 while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
841 if (bitoff == left) {
842 /* mlog(0, "bitoff (%d) == left", bitoff); */
843 break;
844 }
845 /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
846 "numfound = %d\n", bitoff, startoff, numfound);*/
847
848 /* Ok, we found a zero bit... is it contig. or do we
849 * start over?*/
850 if (bitoff == startoff) {
851 /* we found a zero */
852 numfound++;
853 startoff++;
854 } else {
855 /* got a zero after some ones */
856 numfound = 1;
857 startoff = bitoff+1;
858 }
859 /* we got everything we needed */
Mark Fashehd02f00c2009-12-07 13:10:48 -0800860 if (numfound == *numbits) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800861 /* mlog(0, "Found it all!\n"); */
862 break;
863 }
864 }
865
Tao Maa04733d2011-02-22 07:56:45 +0800866 trace_ocfs2_local_alloc_find_clear_bits_search_bitmap(bitoff, numfound);
Mark Fashehccd979b2005-12-15 14:31:24 -0800867
Tao Ma3e4218d2010-04-06 16:46:46 +0800868 if (numfound == *numbits)
Mark Fashehccd979b2005-12-15 14:31:24 -0800869 bitoff = startoff - numfound;
Tao Ma3e4218d2010-04-06 16:46:46 +0800870 else
Mark Fashehccd979b2005-12-15 14:31:24 -0800871 bitoff = -1;
872
873bail:
Mark Fashehd02f00c2009-12-07 13:10:48 -0800874 if (local_resv)
875 ocfs2_resv_discard(resmap, resv);
876
Tao Maa04733d2011-02-22 07:56:45 +0800877 trace_ocfs2_local_alloc_find_clear_bits(*numbits,
878 le32_to_cpu(alloc->id1.bitmap1.i_total),
879 bitoff, numfound);
880
Mark Fashehccd979b2005-12-15 14:31:24 -0800881 return bitoff;
882}
883
884static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
885{
886 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
887 int i;
Mark Fashehccd979b2005-12-15 14:31:24 -0800888
889 alloc->id1.bitmap1.i_total = 0;
890 alloc->id1.bitmap1.i_used = 0;
891 la->la_bm_off = 0;
892 for(i = 0; i < le16_to_cpu(la->la_size); i++)
893 la->la_bitmap[i] = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800894}
895
896#if 0
897/* turn this on and uncomment below to aid debugging window shifts. */
898static void ocfs2_verify_zero_bits(unsigned long *bitmap,
899 unsigned int start,
900 unsigned int count)
901{
902 unsigned int tmp = count;
903 while(tmp--) {
904 if (ocfs2_test_bit(start + tmp, bitmap)) {
905 printk("ocfs2_verify_zero_bits: start = %u, count = "
906 "%u\n", start, count);
907 printk("ocfs2_verify_zero_bits: bit %u is set!",
908 start + tmp);
909 BUG();
910 }
911 }
912}
913#endif
914
915/*
916 * sync the local alloc to main bitmap.
917 *
918 * assumes you've already locked the main bitmap -- the bitmap inode
919 * passed is used for caching.
920 */
921static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -0700922 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800923 struct ocfs2_dinode *alloc,
924 struct inode *main_bm_inode,
925 struct buffer_head *main_bm_bh)
926{
927 int status = 0;
928 int bit_off, left, count, start;
929 u64 la_start_blk;
930 u64 blkno;
931 void *bitmap;
932 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
933
Tao Maa04733d2011-02-22 07:56:45 +0800934 trace_ocfs2_sync_local_to_main(
Tao Maef6b6892011-02-21 11:10:44 +0800935 le32_to_cpu(alloc->id1.bitmap1.i_total),
936 le32_to_cpu(alloc->id1.bitmap1.i_used));
Mark Fashehccd979b2005-12-15 14:31:24 -0800937
938 if (!alloc->id1.bitmap1.i_total) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800939 goto bail;
940 }
941
942 if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
943 le32_to_cpu(alloc->id1.bitmap1.i_total)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800944 goto bail;
945 }
946
947 la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
948 le32_to_cpu(la->la_bm_off));
949 bitmap = la->la_bitmap;
950 start = count = bit_off = 0;
951 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
952
953 while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
954 != -1) {
955 if ((bit_off < left) && (bit_off == start)) {
956 count++;
957 start++;
958 continue;
959 }
960 if (count) {
961 blkno = la_start_blk +
962 ocfs2_clusters_to_blocks(osb->sb,
963 start - count);
964
Tao Maa04733d2011-02-22 07:56:45 +0800965 trace_ocfs2_sync_local_to_main_free(
Mark Fashehb06970532006-03-03 10:24:33 -0800966 count, start - count,
967 (unsigned long long)la_start_blk,
968 (unsigned long long)blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800969
Mark Fashehb4414ee2010-03-11 18:31:09 -0800970 status = ocfs2_release_clusters(handle,
971 main_bm_inode,
972 main_bm_bh, blkno,
973 count);
Mark Fashehccd979b2005-12-15 14:31:24 -0800974 if (status < 0) {
975 mlog_errno(status);
976 goto bail;
977 }
978 }
979 if (bit_off >= left)
980 break;
981 count = 1;
982 start = bit_off + 1;
983 }
984
985bail:
Tao Mac1e8d352011-03-07 16:43:21 +0800986 if (status)
987 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800988 return status;
989}
990
Mark Fasheh9c7af402008-07-28 18:02:53 -0700991enum ocfs2_la_event {
992 OCFS2_LA_EVENT_SLIDE, /* Normal window slide. */
993 OCFS2_LA_EVENT_FRAGMENTED, /* The global bitmap has
994 * enough bits theoretically
995 * free, but a contiguous
996 * allocation could not be
997 * found. */
998 OCFS2_LA_EVENT_ENOSPC, /* Global bitmap doesn't have
999 * enough bits free to satisfy
1000 * our request. */
1001};
1002#define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
1003/*
1004 * Given an event, calculate the size of our next local alloc window.
1005 *
1006 * This should always be called under i_mutex of the local alloc inode
1007 * so that local alloc disabling doesn't race with processes trying to
1008 * use the allocator.
1009 *
1010 * Returns the state which the local alloc was left in. This value can
1011 * be ignored by some paths.
1012 */
1013static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
1014 enum ocfs2_la_event event)
1015{
1016 unsigned int bits;
1017 int state;
1018
1019 spin_lock(&osb->osb_lock);
1020 if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
1021 WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
1022 goto out_unlock;
1023 }
1024
1025 /*
1026 * ENOSPC and fragmentation are treated similarly for now.
1027 */
1028 if (event == OCFS2_LA_EVENT_ENOSPC ||
1029 event == OCFS2_LA_EVENT_FRAGMENTED) {
1030 /*
1031 * We ran out of contiguous space in the primary
1032 * bitmap. Drastically reduce the number of bits used
1033 * by local alloc until we have to disable it.
1034 */
1035 bits = osb->local_alloc_bits >> 1;
1036 if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
1037 /*
1038 * By setting state to THROTTLED, we'll keep
1039 * the number of local alloc bits used down
1040 * until an event occurs which would give us
1041 * reason to assume the bitmap situation might
1042 * have changed.
1043 */
1044 osb->local_alloc_state = OCFS2_LA_THROTTLED;
1045 osb->local_alloc_bits = bits;
1046 } else {
1047 osb->local_alloc_state = OCFS2_LA_DISABLED;
1048 }
1049 queue_delayed_work(ocfs2_wq, &osb->la_enable_wq,
1050 OCFS2_LA_ENABLE_INTERVAL);
1051 goto out_unlock;
1052 }
1053
1054 /*
1055 * Don't increase the size of the local alloc window until we
1056 * know we might be able to fulfill the request. Otherwise, we
1057 * risk bouncing around the global bitmap during periods of
1058 * low space.
1059 */
1060 if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
1061 osb->local_alloc_bits = osb->local_alloc_default_bits;
1062
1063out_unlock:
1064 state = osb->local_alloc_state;
1065 spin_unlock(&osb->osb_lock);
1066
1067 return state;
1068}
1069
Mark Fashehccd979b2005-12-15 14:31:24 -08001070static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -08001071 struct ocfs2_alloc_context **ac,
1072 struct inode **bitmap_inode,
1073 struct buffer_head **bitmap_bh)
1074{
1075 int status;
1076
Robert P. J. Daycd861282006-12-13 00:34:52 -08001077 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001078 if (!(*ac)) {
1079 status = -ENOMEM;
1080 mlog_errno(status);
1081 goto bail;
1082 }
1083
Mark Fasheh9c7af402008-07-28 18:02:53 -07001084retry_enospc:
Mark Fashehb22b63e2010-03-11 18:43:46 -08001085 (*ac)->ac_bits_wanted = osb->local_alloc_default_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -08001086 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
Mark Fasheh9c7af402008-07-28 18:02:53 -07001087 if (status == -ENOSPC) {
1088 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
1089 OCFS2_LA_DISABLED)
1090 goto bail;
1091
1092 ocfs2_free_ac_resource(*ac);
1093 memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
1094 goto retry_enospc;
1095 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001096 if (status < 0) {
Mark Fasheh9c7af402008-07-28 18:02:53 -07001097 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001098 goto bail;
1099 }
1100
1101 *bitmap_inode = (*ac)->ac_inode;
1102 igrab(*bitmap_inode);
1103 *bitmap_bh = (*ac)->ac_bh;
1104 get_bh(*bitmap_bh);
1105 status = 0;
1106bail:
1107 if ((status < 0) && *ac) {
1108 ocfs2_free_alloc_context(*ac);
1109 *ac = NULL;
1110 }
1111
Tao Mac1e8d352011-03-07 16:43:21 +08001112 if (status)
1113 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001114 return status;
1115}
1116
1117/*
1118 * pass it the bitmap lock in lock_bh if you have it.
1119 */
1120static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07001121 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001122 struct ocfs2_alloc_context *ac)
1123{
1124 int status = 0;
1125 u32 cluster_off, cluster_count;
1126 struct ocfs2_dinode *alloc = NULL;
1127 struct ocfs2_local_alloc *la;
1128
Mark Fashehccd979b2005-12-15 14:31:24 -08001129 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1130 la = OCFS2_LOCAL_ALLOC(alloc);
1131
Tao Maa04733d2011-02-22 07:56:45 +08001132 trace_ocfs2_local_alloc_new_window(
1133 le32_to_cpu(alloc->id1.bitmap1.i_total),
1134 osb->local_alloc_bits);
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001135
1136 /* Instruct the allocation code to try the most recently used
1137 * cluster group. We'll re-record the group used this pass
1138 * below. */
1139 ac->ac_last_group = osb->la_last_gd;
1140
Mark Fashehccd979b2005-12-15 14:31:24 -08001141 /* we used the generic suballoc reserve function, but we set
1142 * everything up nicely, so there's no reason why we can't use
1143 * the more specific cluster api to claim bits. */
Joel Becker1ed9b772010-05-06 13:59:06 +08001144 status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits,
Mark Fashehccd979b2005-12-15 14:31:24 -08001145 &cluster_off, &cluster_count);
Mark Fasheh9c7af402008-07-28 18:02:53 -07001146 if (status == -ENOSPC) {
1147retry_enospc:
1148 /*
1149 * Note: We could also try syncing the journal here to
1150 * allow use of any free bits which the current
1151 * transaction can't give us access to. --Mark
1152 */
1153 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
1154 OCFS2_LA_DISABLED)
1155 goto bail;
1156
Mark Fashehb22b63e2010-03-11 18:43:46 -08001157 ac->ac_bits_wanted = osb->local_alloc_default_bits;
Joel Becker1ed9b772010-05-06 13:59:06 +08001158 status = ocfs2_claim_clusters(handle, ac,
Mark Fasheh9c7af402008-07-28 18:02:53 -07001159 osb->local_alloc_bits,
1160 &cluster_off,
1161 &cluster_count);
1162 if (status == -ENOSPC)
1163 goto retry_enospc;
1164 /*
1165 * We only shrunk the *minimum* number of in our
1166 * request - it's entirely possible that the allocator
1167 * might give us more than we asked for.
1168 */
1169 if (status == 0) {
1170 spin_lock(&osb->osb_lock);
1171 osb->local_alloc_bits = cluster_count;
1172 spin_unlock(&osb->osb_lock);
1173 }
1174 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001175 if (status < 0) {
1176 if (status != -ENOSPC)
1177 mlog_errno(status);
1178 goto bail;
1179 }
1180
Mark Fasheh883d4ca2006-06-05 16:41:00 -04001181 osb->la_last_gd = ac->ac_last_group;
1182
Mark Fashehccd979b2005-12-15 14:31:24 -08001183 la->la_bm_off = cpu_to_le32(cluster_off);
1184 alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
1185 /* just in case... In the future when we find space ourselves,
1186 * we don't have to get all contiguous -- but we'll have to
1187 * set all previously used bits in bitmap and update
1188 * la_bits_set before setting the bits in the main bitmap. */
1189 alloc->id1.bitmap1.i_used = 0;
1190 memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
1191 le16_to_cpu(la->la_size));
1192
Mark Fashehd02f00c2009-12-07 13:10:48 -08001193 ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count,
1194 OCFS2_LOCAL_ALLOC(alloc)->la_bitmap);
1195
Tao Maa04733d2011-02-22 07:56:45 +08001196 trace_ocfs2_local_alloc_new_window_result(
1197 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off,
1198 le32_to_cpu(alloc->id1.bitmap1.i_total));
Mark Fashehccd979b2005-12-15 14:31:24 -08001199
1200bail:
Tao Mac1e8d352011-03-07 16:43:21 +08001201 if (status)
1202 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001203 return status;
1204}
1205
1206/* Note that we do *NOT* lock the local alloc inode here as
1207 * it's been locked already for us. */
1208static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
1209 struct inode *local_alloc_inode)
1210{
1211 int status = 0;
1212 struct buffer_head *main_bm_bh = NULL;
1213 struct inode *main_bm_inode = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001214 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001215 struct ocfs2_dinode *alloc;
1216 struct ocfs2_dinode *alloc_copy = NULL;
1217 struct ocfs2_alloc_context *ac = NULL;
1218
Mark Fasheh9c7af402008-07-28 18:02:53 -07001219 ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
1220
Mark Fashehccd979b2005-12-15 14:31:24 -08001221 /* This will lock the main bitmap for us. */
1222 status = ocfs2_local_alloc_reserve_for_window(osb,
Mark Fashehccd979b2005-12-15 14:31:24 -08001223 &ac,
1224 &main_bm_inode,
1225 &main_bm_bh);
1226 if (status < 0) {
1227 if (status != -ENOSPC)
1228 mlog_errno(status);
1229 goto bail;
1230 }
1231
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001232 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001233 if (IS_ERR(handle)) {
1234 status = PTR_ERR(handle);
1235 handle = NULL;
1236 mlog_errno(status);
1237 goto bail;
1238 }
1239
1240 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1241
1242 /* We want to clear the local alloc before doing anything
1243 * else, so that if we error later during this operation,
1244 * local alloc shutdown won't try to double free main bitmap
1245 * bits. Make a copy so the sync function knows which bits to
1246 * free. */
Sunil Mushran4ba1c5b2008-04-18 15:03:59 -07001247 alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001248 if (!alloc_copy) {
1249 status = -ENOMEM;
1250 mlog_errno(status);
1251 goto bail;
1252 }
1253 memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
1254
Joel Becker0cf2f762009-02-12 16:41:25 -08001255 status = ocfs2_journal_access_di(handle,
1256 INODE_CACHE(local_alloc_inode),
Joel Becker13723d02008-10-17 19:25:01 -07001257 osb->local_alloc_bh,
1258 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001259 if (status < 0) {
1260 mlog_errno(status);
1261 goto bail;
1262 }
1263
1264 ocfs2_clear_local_alloc(alloc);
Joel Beckerec20cec2010-03-19 14:13:52 -07001265 ocfs2_journal_dirty(handle, osb->local_alloc_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001266
1267 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
1268 main_bm_inode, main_bm_bh);
1269 if (status < 0) {
1270 mlog_errno(status);
1271 goto bail;
1272 }
1273
1274 status = ocfs2_local_alloc_new_window(osb, handle, ac);
1275 if (status < 0) {
1276 if (status != -ENOSPC)
1277 mlog_errno(status);
1278 goto bail;
1279 }
1280
1281 atomic_inc(&osb->alloc_stats.moves);
1282
Mark Fashehccd979b2005-12-15 14:31:24 -08001283bail:
1284 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001285 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001286
Mark Fasheha81cb882008-10-07 14:25:16 -07001287 brelse(main_bm_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001288
1289 if (main_bm_inode)
1290 iput(main_bm_inode);
1291
Tim Gardnerd787ab02013-02-21 16:42:44 -08001292 kfree(alloc_copy);
Mark Fashehccd979b2005-12-15 14:31:24 -08001293
1294 if (ac)
1295 ocfs2_free_alloc_context(ac);
1296
Tao Mac1e8d352011-03-07 16:43:21 +08001297 if (status)
1298 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001299 return status;
1300}
1301