1 /* -*- 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
32 #include <cluster/masklog.h>
33
34 #include "ocfs2.h"
35
36 #include "alloc.h"
37 #include "blockcheck.h"
38 #include "dlmglue.h"
39 #include "inode.h"
40 #include "journal.h"
41 #include "localalloc.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "sysfile.h"
45 #include "ocfs2_trace.h"
46
47 #include "buffer_head_io.h"
48
49 #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab))
50
51 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
52
53 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
54 struct ocfs2_dinode *alloc,
55 u32 *numbits,
56 struct ocfs2_alloc_reservation *resv);
57
58 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
59
60 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
61 handle_t *handle,
62 struct ocfs2_dinode *alloc,
63 struct inode *main_bm_inode,
64 struct buffer_head *main_bm_bh);
65
66 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
67 struct ocfs2_alloc_context **ac,
68 struct inode **bitmap_inode,
69 struct buffer_head **bitmap_bh);
70
71 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
72 handle_t *handle,
73 struct ocfs2_alloc_context *ac);
74
75 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
76 struct inode *local_alloc_inode);
77
78 /*
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
ocfs2_la_default_mb(struct ocfs2_super * osb)117 unsigned int ocfs2_la_default_mb(struct ocfs2_super *osb)
118 {
119 unsigned int la_mb;
120 unsigned int gd_mb;
121 unsigned int la_max_mb;
122 unsigned int megs_per_slot;
123 struct super_block *sb = osb->sb;
124
125 gd_mb = ocfs2_clusters_to_megabytes(osb->sb,
126 8 * ocfs2_group_bitmap_size(sb, 0, osb->s_feature_incompat));
127
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
186 /* 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
192 return la_mb;
193 }
194
ocfs2_la_set_sizes(struct ocfs2_super * osb,int requested_mb)195 void ocfs2_la_set_sizes(struct ocfs2_super *osb, int requested_mb)
196 {
197 struct super_block *sb = osb->sb;
198 unsigned int la_default_mb = ocfs2_la_default_mb(osb);
199 unsigned int la_max_mb;
200
201 la_max_mb = ocfs2_clusters_to_megabytes(sb,
202 ocfs2_local_alloc_size(sb) * 8);
203
204 trace_ocfs2_la_set_sizes(requested_mb, la_max_mb, la_default_mb);
205
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
ocfs2_la_state_enabled(struct ocfs2_super * osb)222 static 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
ocfs2_local_alloc_seen_free_bits(struct ocfs2_super * osb,unsigned int num_clusters)228 void 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
ocfs2_la_enable_worker(struct work_struct * work)241 void 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
251 /*
252 * Tell us whether a given allocation should use the local alloc
253 * file. Otherwise, it has to go to the main bitmap.
254 *
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.
257 */
ocfs2_alloc_should_use_local(struct ocfs2_super * osb,u64 bits)258 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
259 {
260 int ret = 0;
261 int la_bits;
262
263 spin_lock(&osb->osb_lock);
264 la_bits = osb->local_alloc_bits;
265
266 if (!ocfs2_la_state_enabled(osb))
267 goto bail;
268
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))
274 goto bail;
275
276 ret = 1;
277 bail:
278 trace_ocfs2_alloc_should_use_local(
279 (unsigned long long)bits, osb->local_alloc_state, la_bits, ret);
280 spin_unlock(&osb->osb_lock);
281 return ret;
282 }
283
ocfs2_load_local_alloc(struct ocfs2_super * osb)284 int 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
293 if (osb->local_alloc_bits == 0)
294 goto bail;
295
296 if (osb->local_alloc_bits >= osb->bitmap_cpg) {
297 mlog(ML_NOTICE, "Requested local alloc window %d is larger "
298 "than max possible %u. Using defaults.\n",
299 osb->local_alloc_bits, (osb->bitmap_cpg - 1));
300 osb->local_alloc_bits =
301 ocfs2_megabytes_to_clusters(osb->sb,
302 ocfs2_la_default_mb(osb));
303 }
304
305 /* 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
314 status = ocfs2_read_inode_block_full(inode, &alloc_bh,
315 OCFS2_BH_IGNORE_CACHE);
316 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))) {
326 mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
327 (unsigned long long)OCFS2_I(inode)->ip_blkno);
328 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
358 bail:
359 if (status < 0)
360 brelse(alloc_bh);
361 if (inode)
362 iput(inode);
363
364 trace_ocfs2_load_local_alloc(osb->local_alloc_bits);
365
366 if (status)
367 mlog_errno(status);
368 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.*/
ocfs2_shutdown_local_alloc(struct ocfs2_super * osb)378 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
379 {
380 int status;
381 handle_t *handle;
382 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
389 cancel_delayed_work(&osb->la_enable_wq);
390 flush_workqueue(ocfs2_wq);
391
392 if (osb->local_alloc_state == OCFS2_LA_UNUSED)
393 goto out;
394
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);
402 goto out;
403 }
404
405 osb->local_alloc_state = OCFS2_LA_DISABLED;
406
407 ocfs2_resmap_uninit(&osb->osb_la_resmap);
408
409 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);
415 goto out;
416 }
417
418 mutex_lock(&main_bm_inode->i_mutex);
419
420 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
421 if (status < 0) {
422 mlog_errno(status);
423 goto out_mutex;
424 }
425
426 /* WINDOW_MOVE_CREDITS is a bit heavy... */
427 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
428 if (IS_ERR(handle)) {
429 mlog_errno(PTR_ERR(handle));
430 handle = NULL;
431 goto out_unlock;
432 }
433
434 bh = osb->local_alloc_bh;
435 alloc = (struct ocfs2_dinode *) bh->b_data;
436
437 alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
438 if (!alloc_copy) {
439 status = -ENOMEM;
440 goto out_commit;
441 }
442 memcpy(alloc_copy, alloc, bh->b_size);
443
444 status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode),
445 bh, OCFS2_JOURNAL_ACCESS_WRITE);
446 if (status < 0) {
447 mlog_errno(status);
448 goto out_commit;
449 }
450
451 ocfs2_clear_local_alloc(alloc);
452 ocfs2_journal_dirty(handle, bh);
453
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
463 out_commit:
464 ocfs2_commit_trans(osb, handle);
465
466 out_unlock:
467 brelse(main_bm_bh);
468
469 ocfs2_inode_unlock(main_bm_inode, 1);
470
471 out_mutex:
472 mutex_unlock(&main_bm_inode->i_mutex);
473 iput(main_bm_inode);
474
475 out:
476 if (local_alloc_inode)
477 iput(local_alloc_inode);
478
479 if (alloc_copy)
480 kfree(alloc_copy);
481 }
482
483 /*
484 * We want to free the bitmap bits outside of any recovery context as
485 * we'll need a cluster lock to do so, but we must clear the local
486 * alloc before giving up the recovered nodes journal. To solve this,
487 * we kmalloc a copy of the local alloc before it's change for the
488 * caller to process with ocfs2_complete_local_alloc_recovery
489 */
ocfs2_begin_local_alloc_recovery(struct ocfs2_super * osb,int slot_num,struct ocfs2_dinode ** alloc_copy)490 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
491 int slot_num,
492 struct ocfs2_dinode **alloc_copy)
493 {
494 int status = 0;
495 struct buffer_head *alloc_bh = NULL;
496 struct inode *inode = NULL;
497 struct ocfs2_dinode *alloc;
498
499 trace_ocfs2_begin_local_alloc_recovery(slot_num);
500
501 *alloc_copy = NULL;
502
503 inode = ocfs2_get_system_file_inode(osb,
504 LOCAL_ALLOC_SYSTEM_INODE,
505 slot_num);
506 if (!inode) {
507 status = -EINVAL;
508 mlog_errno(status);
509 goto bail;
510 }
511
512 mutex_lock(&inode->i_mutex);
513
514 status = ocfs2_read_inode_block_full(inode, &alloc_bh,
515 OCFS2_BH_IGNORE_CACHE);
516 if (status < 0) {
517 mlog_errno(status);
518 goto bail;
519 }
520
521 *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
522 if (!(*alloc_copy)) {
523 status = -ENOMEM;
524 goto bail;
525 }
526 memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
527
528 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
529 ocfs2_clear_local_alloc(alloc);
530
531 ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check);
532 status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode));
533 if (status < 0)
534 mlog_errno(status);
535
536 bail:
537 if ((status < 0) && (*alloc_copy)) {
538 kfree(*alloc_copy);
539 *alloc_copy = NULL;
540 }
541
542 brelse(alloc_bh);
543
544 if (inode) {
545 mutex_unlock(&inode->i_mutex);
546 iput(inode);
547 }
548
549 if (status)
550 mlog_errno(status);
551 return status;
552 }
553
554 /*
555 * Step 2: By now, we've completed the journal recovery, we've stamped
556 * a clean local alloc on disk and dropped the node out of the
557 * recovery map. Dlm locks will no longer stall, so lets clear out the
558 * main bitmap.
559 */
ocfs2_complete_local_alloc_recovery(struct ocfs2_super * osb,struct ocfs2_dinode * alloc)560 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
561 struct ocfs2_dinode *alloc)
562 {
563 int status;
564 handle_t *handle;
565 struct buffer_head *main_bm_bh = NULL;
566 struct inode *main_bm_inode;
567
568 main_bm_inode = ocfs2_get_system_file_inode(osb,
569 GLOBAL_BITMAP_SYSTEM_INODE,
570 OCFS2_INVALID_SLOT);
571 if (!main_bm_inode) {
572 status = -EINVAL;
573 mlog_errno(status);
574 goto out;
575 }
576
577 mutex_lock(&main_bm_inode->i_mutex);
578
579 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
580 if (status < 0) {
581 mlog_errno(status);
582 goto out_mutex;
583 }
584
585 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
586 if (IS_ERR(handle)) {
587 status = PTR_ERR(handle);
588 handle = NULL;
589 mlog_errno(status);
590 goto out_unlock;
591 }
592
593 /* we want the bitmap change to be recorded on disk asap */
594 handle->h_sync = 1;
595
596 status = ocfs2_sync_local_to_main(osb, handle, alloc,
597 main_bm_inode, main_bm_bh);
598 if (status < 0)
599 mlog_errno(status);
600
601 ocfs2_commit_trans(osb, handle);
602
603 out_unlock:
604 ocfs2_inode_unlock(main_bm_inode, 1);
605
606 out_mutex:
607 mutex_unlock(&main_bm_inode->i_mutex);
608
609 brelse(main_bm_bh);
610
611 iput(main_bm_inode);
612
613 out:
614 if (!status)
615 ocfs2_init_steal_slots(osb);
616 if (status)
617 mlog_errno(status);
618 return status;
619 }
620
621 /*
622 * make sure we've got at least bits_wanted contiguous bits in the
623 * local alloc. You lose them when you drop i_mutex.
624 *
625 * We will add ourselves to the transaction passed in, but may start
626 * our own in order to shift windows.
627 */
ocfs2_reserve_local_alloc_bits(struct ocfs2_super * osb,u32 bits_wanted,struct ocfs2_alloc_context * ac)628 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
629 u32 bits_wanted,
630 struct ocfs2_alloc_context *ac)
631 {
632 int status;
633 struct ocfs2_dinode *alloc;
634 struct inode *local_alloc_inode;
635 unsigned int free_bits;
636
637 BUG_ON(!ac);
638
639 local_alloc_inode =
640 ocfs2_get_system_file_inode(osb,
641 LOCAL_ALLOC_SYSTEM_INODE,
642 osb->slot_num);
643 if (!local_alloc_inode) {
644 status = -ENOENT;
645 mlog_errno(status);
646 goto bail;
647 }
648
649 mutex_lock(&local_alloc_inode->i_mutex);
650
651 /*
652 * We must double check state and allocator bits because
653 * another process may have changed them while holding i_mutex.
654 */
655 spin_lock(&osb->osb_lock);
656 if (!ocfs2_la_state_enabled(osb) ||
657 (bits_wanted > osb->local_alloc_bits)) {
658 spin_unlock(&osb->osb_lock);
659 status = -ENOSPC;
660 goto bail;
661 }
662 spin_unlock(&osb->osb_lock);
663
664 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
665
666 #ifdef CONFIG_OCFS2_DEBUG_FS
667 if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
668 ocfs2_local_alloc_count_bits(alloc)) {
669 ocfs2_error(osb->sb, "local alloc inode %llu says it has "
670 "%u free bits, but a count shows %u",
671 (unsigned long long)le64_to_cpu(alloc->i_blkno),
672 le32_to_cpu(alloc->id1.bitmap1.i_used),
673 ocfs2_local_alloc_count_bits(alloc));
674 status = -EIO;
675 goto bail;
676 }
677 #endif
678
679 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
680 le32_to_cpu(alloc->id1.bitmap1.i_used);
681 if (bits_wanted > free_bits) {
682 /* uhoh, window change time. */
683 status =
684 ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
685 if (status < 0) {
686 if (status != -ENOSPC)
687 mlog_errno(status);
688 goto bail;
689 }
690
691 /*
692 * Under certain conditions, the window slide code
693 * might have reduced the number of bits available or
694 * disabled the the local alloc entirely. Re-check
695 * here and return -ENOSPC if necessary.
696 */
697 status = -ENOSPC;
698 if (!ocfs2_la_state_enabled(osb))
699 goto bail;
700
701 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
702 le32_to_cpu(alloc->id1.bitmap1.i_used);
703 if (bits_wanted > free_bits)
704 goto bail;
705 }
706
707 ac->ac_inode = local_alloc_inode;
708 /* We should never use localalloc from another slot */
709 ac->ac_alloc_slot = osb->slot_num;
710 ac->ac_which = OCFS2_AC_USE_LOCAL;
711 get_bh(osb->local_alloc_bh);
712 ac->ac_bh = osb->local_alloc_bh;
713 status = 0;
714 bail:
715 if (status < 0 && local_alloc_inode) {
716 mutex_unlock(&local_alloc_inode->i_mutex);
717 iput(local_alloc_inode);
718 }
719
720 trace_ocfs2_reserve_local_alloc_bits(
721 (unsigned long long)ac->ac_max_block,
722 bits_wanted, osb->slot_num, status);
723
724 if (status)
725 mlog_errno(status);
726 return status;
727 }
728
ocfs2_claim_local_alloc_bits(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac,u32 bits_wanted,u32 * bit_off,u32 * num_bits)729 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
730 handle_t *handle,
731 struct ocfs2_alloc_context *ac,
732 u32 bits_wanted,
733 u32 *bit_off,
734 u32 *num_bits)
735 {
736 int status, start;
737 struct inode *local_alloc_inode;
738 void *bitmap;
739 struct ocfs2_dinode *alloc;
740 struct ocfs2_local_alloc *la;
741
742 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
743
744 local_alloc_inode = ac->ac_inode;
745 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
746 la = OCFS2_LOCAL_ALLOC(alloc);
747
748 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted,
749 ac->ac_resv);
750 if (start == -1) {
751 /* TODO: Shouldn't we just BUG here? */
752 status = -ENOSPC;
753 mlog_errno(status);
754 goto bail;
755 }
756
757 bitmap = la->la_bitmap;
758 *bit_off = le32_to_cpu(la->la_bm_off) + start;
759 *num_bits = bits_wanted;
760
761 status = ocfs2_journal_access_di(handle,
762 INODE_CACHE(local_alloc_inode),
763 osb->local_alloc_bh,
764 OCFS2_JOURNAL_ACCESS_WRITE);
765 if (status < 0) {
766 mlog_errno(status);
767 goto bail;
768 }
769
770 ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start,
771 bits_wanted);
772
773 while(bits_wanted--)
774 ocfs2_set_bit(start++, bitmap);
775
776 le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
777 ocfs2_journal_dirty(handle, osb->local_alloc_bh);
778
779 bail:
780 if (status)
781 mlog_errno(status);
782 return status;
783 }
784
ocfs2_local_alloc_count_bits(struct ocfs2_dinode * alloc)785 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
786 {
787 int i;
788 u8 *buffer;
789 u32 count = 0;
790 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
791
792 buffer = la->la_bitmap;
793 for (i = 0; i < le16_to_cpu(la->la_size); i++)
794 count += hweight8(buffer[i]);
795
796 trace_ocfs2_local_alloc_count_bits(count);
797 return count;
798 }
799
ocfs2_local_alloc_find_clear_bits(struct ocfs2_super * osb,struct ocfs2_dinode * alloc,u32 * numbits,struct ocfs2_alloc_reservation * resv)800 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
801 struct ocfs2_dinode *alloc,
802 u32 *numbits,
803 struct ocfs2_alloc_reservation *resv)
804 {
805 int numfound, bitoff, left, startoff, lastzero;
806 int local_resv = 0;
807 struct ocfs2_alloc_reservation r;
808 void *bitmap = NULL;
809 struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap;
810
811 if (!alloc->id1.bitmap1.i_total) {
812 bitoff = -1;
813 goto bail;
814 }
815
816 if (!resv) {
817 local_resv = 1;
818 ocfs2_resv_init_once(&r);
819 ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP);
820 resv = &r;
821 }
822
823 numfound = *numbits;
824 if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) {
825 if (numfound < *numbits)
826 *numbits = numfound;
827 goto bail;
828 }
829
830 /*
831 * Code error. While reservations are enabled, local
832 * allocation should _always_ go through them.
833 */
834 BUG_ON(osb->osb_resv_level != 0);
835
836 /*
837 * Reservations are disabled. Handle this the old way.
838 */
839
840 bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
841
842 numfound = bitoff = startoff = 0;
843 lastzero = -1;
844 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
845 while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
846 if (bitoff == left) {
847 /* mlog(0, "bitoff (%d) == left", bitoff); */
848 break;
849 }
850 /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
851 "numfound = %d\n", bitoff, startoff, numfound);*/
852
853 /* Ok, we found a zero bit... is it contig. or do we
854 * start over?*/
855 if (bitoff == startoff) {
856 /* we found a zero */
857 numfound++;
858 startoff++;
859 } else {
860 /* got a zero after some ones */
861 numfound = 1;
862 startoff = bitoff+1;
863 }
864 /* we got everything we needed */
865 if (numfound == *numbits) {
866 /* mlog(0, "Found it all!\n"); */
867 break;
868 }
869 }
870
871 trace_ocfs2_local_alloc_find_clear_bits_search_bitmap(bitoff, numfound);
872
873 if (numfound == *numbits)
874 bitoff = startoff - numfound;
875 else
876 bitoff = -1;
877
878 bail:
879 if (local_resv)
880 ocfs2_resv_discard(resmap, resv);
881
882 trace_ocfs2_local_alloc_find_clear_bits(*numbits,
883 le32_to_cpu(alloc->id1.bitmap1.i_total),
884 bitoff, numfound);
885
886 return bitoff;
887 }
888
ocfs2_clear_local_alloc(struct ocfs2_dinode * alloc)889 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
890 {
891 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
892 int i;
893
894 alloc->id1.bitmap1.i_total = 0;
895 alloc->id1.bitmap1.i_used = 0;
896 la->la_bm_off = 0;
897 for(i = 0; i < le16_to_cpu(la->la_size); i++)
898 la->la_bitmap[i] = 0;
899 }
900
901 #if 0
902 /* turn this on and uncomment below to aid debugging window shifts. */
903 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
904 unsigned int start,
905 unsigned int count)
906 {
907 unsigned int tmp = count;
908 while(tmp--) {
909 if (ocfs2_test_bit(start + tmp, bitmap)) {
910 printk("ocfs2_verify_zero_bits: start = %u, count = "
911 "%u\n", start, count);
912 printk("ocfs2_verify_zero_bits: bit %u is set!",
913 start + tmp);
914 BUG();
915 }
916 }
917 }
918 #endif
919
920 /*
921 * sync the local alloc to main bitmap.
922 *
923 * assumes you've already locked the main bitmap -- the bitmap inode
924 * passed is used for caching.
925 */
ocfs2_sync_local_to_main(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_dinode * alloc,struct inode * main_bm_inode,struct buffer_head * main_bm_bh)926 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
927 handle_t *handle,
928 struct ocfs2_dinode *alloc,
929 struct inode *main_bm_inode,
930 struct buffer_head *main_bm_bh)
931 {
932 int status = 0;
933 int bit_off, left, count, start;
934 u64 la_start_blk;
935 u64 blkno;
936 void *bitmap;
937 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
938
939 trace_ocfs2_sync_local_to_main(
940 le32_to_cpu(alloc->id1.bitmap1.i_total),
941 le32_to_cpu(alloc->id1.bitmap1.i_used));
942
943 if (!alloc->id1.bitmap1.i_total) {
944 goto bail;
945 }
946
947 if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
948 le32_to_cpu(alloc->id1.bitmap1.i_total)) {
949 goto bail;
950 }
951
952 la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
953 le32_to_cpu(la->la_bm_off));
954 bitmap = la->la_bitmap;
955 start = count = bit_off = 0;
956 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
957
958 while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
959 != -1) {
960 if ((bit_off < left) && (bit_off == start)) {
961 count++;
962 start++;
963 continue;
964 }
965 if (count) {
966 blkno = la_start_blk +
967 ocfs2_clusters_to_blocks(osb->sb,
968 start - count);
969
970 trace_ocfs2_sync_local_to_main_free(
971 count, start - count,
972 (unsigned long long)la_start_blk,
973 (unsigned long long)blkno);
974
975 status = ocfs2_release_clusters(handle,
976 main_bm_inode,
977 main_bm_bh, blkno,
978 count);
979 if (status < 0) {
980 mlog_errno(status);
981 goto bail;
982 }
983 }
984 if (bit_off >= left)
985 break;
986 count = 1;
987 start = bit_off + 1;
988 }
989
990 bail:
991 if (status)
992 mlog_errno(status);
993 return status;
994 }
995
996 enum ocfs2_la_event {
997 OCFS2_LA_EVENT_SLIDE, /* Normal window slide. */
998 OCFS2_LA_EVENT_FRAGMENTED, /* The global bitmap has
999 * enough bits theoretically
1000 * free, but a contiguous
1001 * allocation could not be
1002 * found. */
1003 OCFS2_LA_EVENT_ENOSPC, /* Global bitmap doesn't have
1004 * enough bits free to satisfy
1005 * our request. */
1006 };
1007 #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
1008 /*
1009 * Given an event, calculate the size of our next local alloc window.
1010 *
1011 * This should always be called under i_mutex of the local alloc inode
1012 * so that local alloc disabling doesn't race with processes trying to
1013 * use the allocator.
1014 *
1015 * Returns the state which the local alloc was left in. This value can
1016 * be ignored by some paths.
1017 */
ocfs2_recalc_la_window(struct ocfs2_super * osb,enum ocfs2_la_event event)1018 static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
1019 enum ocfs2_la_event event)
1020 {
1021 unsigned int bits;
1022 int state;
1023
1024 spin_lock(&osb->osb_lock);
1025 if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
1026 WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
1027 goto out_unlock;
1028 }
1029
1030 /*
1031 * ENOSPC and fragmentation are treated similarly for now.
1032 */
1033 if (event == OCFS2_LA_EVENT_ENOSPC ||
1034 event == OCFS2_LA_EVENT_FRAGMENTED) {
1035 /*
1036 * We ran out of contiguous space in the primary
1037 * bitmap. Drastically reduce the number of bits used
1038 * by local alloc until we have to disable it.
1039 */
1040 bits = osb->local_alloc_bits >> 1;
1041 if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
1042 /*
1043 * By setting state to THROTTLED, we'll keep
1044 * the number of local alloc bits used down
1045 * until an event occurs which would give us
1046 * reason to assume the bitmap situation might
1047 * have changed.
1048 */
1049 osb->local_alloc_state = OCFS2_LA_THROTTLED;
1050 osb->local_alloc_bits = bits;
1051 } else {
1052 osb->local_alloc_state = OCFS2_LA_DISABLED;
1053 }
1054 queue_delayed_work(ocfs2_wq, &osb->la_enable_wq,
1055 OCFS2_LA_ENABLE_INTERVAL);
1056 goto out_unlock;
1057 }
1058
1059 /*
1060 * Don't increase the size of the local alloc window until we
1061 * know we might be able to fulfill the request. Otherwise, we
1062 * risk bouncing around the global bitmap during periods of
1063 * low space.
1064 */
1065 if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
1066 osb->local_alloc_bits = osb->local_alloc_default_bits;
1067
1068 out_unlock:
1069 state = osb->local_alloc_state;
1070 spin_unlock(&osb->osb_lock);
1071
1072 return state;
1073 }
1074
ocfs2_local_alloc_reserve_for_window(struct ocfs2_super * osb,struct ocfs2_alloc_context ** ac,struct inode ** bitmap_inode,struct buffer_head ** bitmap_bh)1075 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
1076 struct ocfs2_alloc_context **ac,
1077 struct inode **bitmap_inode,
1078 struct buffer_head **bitmap_bh)
1079 {
1080 int status;
1081
1082 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1083 if (!(*ac)) {
1084 status = -ENOMEM;
1085 mlog_errno(status);
1086 goto bail;
1087 }
1088
1089 retry_enospc:
1090 (*ac)->ac_bits_wanted = osb->local_alloc_default_bits;
1091 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1092 if (status == -ENOSPC) {
1093 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
1094 OCFS2_LA_DISABLED)
1095 goto bail;
1096
1097 ocfs2_free_ac_resource(*ac);
1098 memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
1099 goto retry_enospc;
1100 }
1101 if (status < 0) {
1102 mlog_errno(status);
1103 goto bail;
1104 }
1105
1106 *bitmap_inode = (*ac)->ac_inode;
1107 igrab(*bitmap_inode);
1108 *bitmap_bh = (*ac)->ac_bh;
1109 get_bh(*bitmap_bh);
1110 status = 0;
1111 bail:
1112 if ((status < 0) && *ac) {
1113 ocfs2_free_alloc_context(*ac);
1114 *ac = NULL;
1115 }
1116
1117 if (status)
1118 mlog_errno(status);
1119 return status;
1120 }
1121
1122 /*
1123 * pass it the bitmap lock in lock_bh if you have it.
1124 */
ocfs2_local_alloc_new_window(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac)1125 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
1126 handle_t *handle,
1127 struct ocfs2_alloc_context *ac)
1128 {
1129 int status = 0;
1130 u32 cluster_off, cluster_count;
1131 struct ocfs2_dinode *alloc = NULL;
1132 struct ocfs2_local_alloc *la;
1133
1134 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1135 la = OCFS2_LOCAL_ALLOC(alloc);
1136
1137 trace_ocfs2_local_alloc_new_window(
1138 le32_to_cpu(alloc->id1.bitmap1.i_total),
1139 osb->local_alloc_bits);
1140
1141 /* Instruct the allocation code to try the most recently used
1142 * cluster group. We'll re-record the group used this pass
1143 * below. */
1144 ac->ac_last_group = osb->la_last_gd;
1145
1146 /* we used the generic suballoc reserve function, but we set
1147 * everything up nicely, so there's no reason why we can't use
1148 * the more specific cluster api to claim bits. */
1149 status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits,
1150 &cluster_off, &cluster_count);
1151 if (status == -ENOSPC) {
1152 retry_enospc:
1153 /*
1154 * Note: We could also try syncing the journal here to
1155 * allow use of any free bits which the current
1156 * transaction can't give us access to. --Mark
1157 */
1158 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
1159 OCFS2_LA_DISABLED)
1160 goto bail;
1161
1162 ac->ac_bits_wanted = osb->local_alloc_default_bits;
1163 status = ocfs2_claim_clusters(handle, ac,
1164 osb->local_alloc_bits,
1165 &cluster_off,
1166 &cluster_count);
1167 if (status == -ENOSPC)
1168 goto retry_enospc;
1169 /*
1170 * We only shrunk the *minimum* number of in our
1171 * request - it's entirely possible that the allocator
1172 * might give us more than we asked for.
1173 */
1174 if (status == 0) {
1175 spin_lock(&osb->osb_lock);
1176 osb->local_alloc_bits = cluster_count;
1177 spin_unlock(&osb->osb_lock);
1178 }
1179 }
1180 if (status < 0) {
1181 if (status != -ENOSPC)
1182 mlog_errno(status);
1183 goto bail;
1184 }
1185
1186 osb->la_last_gd = ac->ac_last_group;
1187
1188 la->la_bm_off = cpu_to_le32(cluster_off);
1189 alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
1190 /* just in case... In the future when we find space ourselves,
1191 * we don't have to get all contiguous -- but we'll have to
1192 * set all previously used bits in bitmap and update
1193 * la_bits_set before setting the bits in the main bitmap. */
1194 alloc->id1.bitmap1.i_used = 0;
1195 memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
1196 le16_to_cpu(la->la_size));
1197
1198 ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count,
1199 OCFS2_LOCAL_ALLOC(alloc)->la_bitmap);
1200
1201 trace_ocfs2_local_alloc_new_window_result(
1202 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off,
1203 le32_to_cpu(alloc->id1.bitmap1.i_total));
1204
1205 bail:
1206 if (status)
1207 mlog_errno(status);
1208 return status;
1209 }
1210
1211 /* Note that we do *NOT* lock the local alloc inode here as
1212 * it's been locked already for us. */
ocfs2_local_alloc_slide_window(struct ocfs2_super * osb,struct inode * local_alloc_inode)1213 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
1214 struct inode *local_alloc_inode)
1215 {
1216 int status = 0;
1217 struct buffer_head *main_bm_bh = NULL;
1218 struct inode *main_bm_inode = NULL;
1219 handle_t *handle = NULL;
1220 struct ocfs2_dinode *alloc;
1221 struct ocfs2_dinode *alloc_copy = NULL;
1222 struct ocfs2_alloc_context *ac = NULL;
1223
1224 ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
1225
1226 /* This will lock the main bitmap for us. */
1227 status = ocfs2_local_alloc_reserve_for_window(osb,
1228 &ac,
1229 &main_bm_inode,
1230 &main_bm_bh);
1231 if (status < 0) {
1232 if (status != -ENOSPC)
1233 mlog_errno(status);
1234 goto bail;
1235 }
1236
1237 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
1238 if (IS_ERR(handle)) {
1239 status = PTR_ERR(handle);
1240 handle = NULL;
1241 mlog_errno(status);
1242 goto bail;
1243 }
1244
1245 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1246
1247 /* We want to clear the local alloc before doing anything
1248 * else, so that if we error later during this operation,
1249 * local alloc shutdown won't try to double free main bitmap
1250 * bits. Make a copy so the sync function knows which bits to
1251 * free. */
1252 alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
1253 if (!alloc_copy) {
1254 status = -ENOMEM;
1255 mlog_errno(status);
1256 goto bail;
1257 }
1258 memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
1259
1260 status = ocfs2_journal_access_di(handle,
1261 INODE_CACHE(local_alloc_inode),
1262 osb->local_alloc_bh,
1263 OCFS2_JOURNAL_ACCESS_WRITE);
1264 if (status < 0) {
1265 mlog_errno(status);
1266 goto bail;
1267 }
1268
1269 ocfs2_clear_local_alloc(alloc);
1270 ocfs2_journal_dirty(handle, osb->local_alloc_bh);
1271
1272 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
1273 main_bm_inode, main_bm_bh);
1274 if (status < 0) {
1275 mlog_errno(status);
1276 goto bail;
1277 }
1278
1279 status = ocfs2_local_alloc_new_window(osb, handle, ac);
1280 if (status < 0) {
1281 if (status != -ENOSPC)
1282 mlog_errno(status);
1283 goto bail;
1284 }
1285
1286 atomic_inc(&osb->alloc_stats.moves);
1287
1288 bail:
1289 if (handle)
1290 ocfs2_commit_trans(osb, handle);
1291
1292 brelse(main_bm_bh);
1293
1294 if (main_bm_inode)
1295 iput(main_bm_inode);
1296
1297 if (alloc_copy)
1298 kfree(alloc_copy);
1299
1300 if (ac)
1301 ocfs2_free_alloc_context(ac);
1302
1303 if (status)
1304 mlog_errno(status);
1305 return status;
1306 }
1307
1308