Commit | Line | Data |
---|---|---|
ccd979bd MF |
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 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
13723d00 | 38 | #include "blockcheck.h" |
ccd979bd MF |
39 | #include "dlmglue.h" |
40 | #include "inode.h" | |
41 | #include "journal.h" | |
42 | #include "localalloc.h" | |
43 | #include "suballoc.h" | |
44 | #include "super.h" | |
45 | #include "sysfile.h" | |
46 | ||
47 | #include "buffer_head_io.h" | |
48 | ||
49 | #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab)) | |
50 | ||
ccd979bd MF |
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, | |
d02f00cc MF |
55 | u32 *numbits, |
56 | struct ocfs2_alloc_reservation *resv); | |
ccd979bd MF |
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, | |
1fabe148 | 61 | handle_t *handle, |
ccd979bd MF |
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, | |
ccd979bd MF |
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, | |
1fabe148 | 72 | handle_t *handle, |
ccd979bd MF |
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 | ||
6b82021b MF |
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 | |
117 | unsigned int ocfs2_la_default_mb(struct ocfs2_super *osb) | |
118 | { | |
119 | unsigned int la_mb; | |
120 | unsigned int gd_mb; | |
1739da40 | 121 | unsigned int la_max_mb; |
6b82021b MF |
122 | unsigned int megs_per_slot; |
123 | struct super_block *sb = osb->sb; | |
124 | ||
125 | gd_mb = ocfs2_clusters_to_megabytes(osb->sb, | |
8571882c | 126 | 8 * ocfs2_group_bitmap_size(sb, 0, osb->s_feature_incompat)); |
6b82021b MF |
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 | ||
1739da40 TM |
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 | ||
6b82021b MF |
192 | return la_mb; |
193 | } | |
194 | ||
73c8a800 MF |
195 | void ocfs2_la_set_sizes(struct ocfs2_super *osb, int requested_mb) |
196 | { | |
197 | struct super_block *sb = osb->sb; | |
6b82021b | 198 | unsigned int la_default_mb = ocfs2_la_default_mb(osb); |
73c8a800 MF |
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 | mlog(0, "requested: %dM, max: %uM, default: %uM\n", | |
205 | requested_mb, la_max_mb, la_default_mb); | |
206 | ||
207 | if (requested_mb == -1) { | |
208 | /* No user request - use defaults */ | |
209 | osb->local_alloc_default_bits = | |
210 | ocfs2_megabytes_to_clusters(sb, la_default_mb); | |
211 | } else if (requested_mb > la_max_mb) { | |
212 | /* Request is too big, we give the maximum available */ | |
213 | osb->local_alloc_default_bits = | |
214 | ocfs2_megabytes_to_clusters(sb, la_max_mb); | |
215 | } else { | |
216 | osb->local_alloc_default_bits = | |
217 | ocfs2_megabytes_to_clusters(sb, requested_mb); | |
218 | } | |
219 | ||
220 | osb->local_alloc_bits = osb->local_alloc_default_bits; | |
221 | } | |
222 | ||
9c7af40b MF |
223 | static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb) |
224 | { | |
225 | return (osb->local_alloc_state == OCFS2_LA_THROTTLED || | |
226 | osb->local_alloc_state == OCFS2_LA_ENABLED); | |
227 | } | |
228 | ||
229 | void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb, | |
230 | unsigned int num_clusters) | |
231 | { | |
232 | spin_lock(&osb->osb_lock); | |
233 | if (osb->local_alloc_state == OCFS2_LA_DISABLED || | |
234 | osb->local_alloc_state == OCFS2_LA_THROTTLED) | |
235 | if (num_clusters >= osb->local_alloc_default_bits) { | |
236 | cancel_delayed_work(&osb->la_enable_wq); | |
237 | osb->local_alloc_state = OCFS2_LA_ENABLED; | |
238 | } | |
239 | spin_unlock(&osb->osb_lock); | |
240 | } | |
241 | ||
242 | void ocfs2_la_enable_worker(struct work_struct *work) | |
243 | { | |
244 | struct ocfs2_super *osb = | |
245 | container_of(work, struct ocfs2_super, | |
246 | la_enable_wq.work); | |
247 | spin_lock(&osb->osb_lock); | |
248 | osb->local_alloc_state = OCFS2_LA_ENABLED; | |
249 | spin_unlock(&osb->osb_lock); | |
250 | } | |
251 | ||
ccd979bd MF |
252 | /* |
253 | * Tell us whether a given allocation should use the local alloc | |
254 | * file. Otherwise, it has to go to the main bitmap. | |
9c7af40b MF |
255 | * |
256 | * This function does semi-dirty reads of local alloc size and state! | |
257 | * This is ok however, as the values are re-checked once under mutex. | |
ccd979bd MF |
258 | */ |
259 | int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits) | |
260 | { | |
2fbe8d1e | 261 | int ret = 0; |
9c7af40b MF |
262 | int la_bits; |
263 | ||
264 | spin_lock(&osb->osb_lock); | |
265 | la_bits = osb->local_alloc_bits; | |
ccd979bd | 266 | |
9c7af40b | 267 | if (!ocfs2_la_state_enabled(osb)) |
2fbe8d1e | 268 | goto bail; |
ccd979bd MF |
269 | |
270 | /* la_bits should be at least twice the size (in clusters) of | |
271 | * a new block group. We want to be sure block group | |
272 | * allocations go through the local alloc, so allow an | |
273 | * allocation to take up to half the bitmap. */ | |
274 | if (bits > (la_bits / 2)) | |
2fbe8d1e | 275 | goto bail; |
ccd979bd | 276 | |
2fbe8d1e SM |
277 | ret = 1; |
278 | bail: | |
279 | mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n", | |
280 | osb->local_alloc_state, (unsigned long long)bits, la_bits, ret); | |
9c7af40b | 281 | spin_unlock(&osb->osb_lock); |
2fbe8d1e | 282 | return ret; |
ccd979bd MF |
283 | } |
284 | ||
285 | int ocfs2_load_local_alloc(struct ocfs2_super *osb) | |
286 | { | |
287 | int status = 0; | |
288 | struct ocfs2_dinode *alloc = NULL; | |
289 | struct buffer_head *alloc_bh = NULL; | |
290 | u32 num_used; | |
291 | struct inode *inode = NULL; | |
292 | struct ocfs2_local_alloc *la; | |
293 | ||
ebcee4b5 | 294 | if (osb->local_alloc_bits == 0) |
2fbe8d1e SM |
295 | goto bail; |
296 | ||
ebcee4b5 | 297 | if (osb->local_alloc_bits >= osb->bitmap_cpg) { |
2fbe8d1e SM |
298 | mlog(ML_NOTICE, "Requested local alloc window %d is larger " |
299 | "than max possible %u. Using defaults.\n", | |
ebcee4b5 MF |
300 | osb->local_alloc_bits, (osb->bitmap_cpg - 1)); |
301 | osb->local_alloc_bits = | |
302 | ocfs2_megabytes_to_clusters(osb->sb, | |
6b82021b | 303 | ocfs2_la_default_mb(osb)); |
2fbe8d1e SM |
304 | } |
305 | ||
ccd979bd MF |
306 | /* read the alloc off disk */ |
307 | inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE, | |
308 | osb->slot_num); | |
309 | if (!inode) { | |
310 | status = -EINVAL; | |
311 | mlog_errno(status); | |
312 | goto bail; | |
313 | } | |
314 | ||
b657c95c JB |
315 | status = ocfs2_read_inode_block_full(inode, &alloc_bh, |
316 | OCFS2_BH_IGNORE_CACHE); | |
ccd979bd MF |
317 | if (status < 0) { |
318 | mlog_errno(status); | |
319 | goto bail; | |
320 | } | |
321 | ||
322 | alloc = (struct ocfs2_dinode *) alloc_bh->b_data; | |
323 | la = OCFS2_LOCAL_ALLOC(alloc); | |
324 | ||
325 | if (!(le32_to_cpu(alloc->i_flags) & | |
326 | (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) { | |
b0697053 MF |
327 | mlog(ML_ERROR, "Invalid local alloc inode, %llu\n", |
328 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
329 | status = -EINVAL; |
330 | goto bail; | |
331 | } | |
332 | ||
333 | if ((la->la_size == 0) || | |
334 | (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) { | |
335 | mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n", | |
336 | le16_to_cpu(la->la_size)); | |
337 | status = -EINVAL; | |
338 | goto bail; | |
339 | } | |
340 | ||
341 | /* do a little verification. */ | |
342 | num_used = ocfs2_local_alloc_count_bits(alloc); | |
343 | ||
344 | /* hopefully the local alloc has always been recovered before | |
345 | * we load it. */ | |
346 | if (num_used | |
347 | || alloc->id1.bitmap1.i_used | |
348 | || alloc->id1.bitmap1.i_total | |
349 | || la->la_bm_off) | |
350 | mlog(ML_ERROR, "Local alloc hasn't been recovered!\n" | |
351 | "found = %u, set = %u, taken = %u, off = %u\n", | |
352 | num_used, le32_to_cpu(alloc->id1.bitmap1.i_used), | |
353 | le32_to_cpu(alloc->id1.bitmap1.i_total), | |
354 | OCFS2_LOCAL_ALLOC(alloc)->la_bm_off); | |
355 | ||
356 | osb->local_alloc_bh = alloc_bh; | |
357 | osb->local_alloc_state = OCFS2_LA_ENABLED; | |
358 | ||
359 | bail: | |
360 | if (status < 0) | |
a81cb88b | 361 | brelse(alloc_bh); |
ccd979bd MF |
362 | if (inode) |
363 | iput(inode); | |
364 | ||
ebcee4b5 | 365 | mlog(0, "Local alloc window bits = %d\n", osb->local_alloc_bits); |
2fbe8d1e | 366 | |
c1e8d35e TM |
367 | if (status) |
368 | mlog_errno(status); | |
ccd979bd MF |
369 | return status; |
370 | } | |
371 | ||
372 | /* | |
373 | * return any unused bits to the bitmap and write out a clean | |
374 | * local_alloc. | |
375 | * | |
376 | * local_alloc_bh is optional. If not passed, we will simply use the | |
377 | * one off osb. If you do pass it however, be warned that it *will* be | |
378 | * returned brelse'd and NULL'd out.*/ | |
379 | void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb) | |
380 | { | |
381 | int status; | |
1fabe148 | 382 | handle_t *handle; |
ccd979bd MF |
383 | struct inode *local_alloc_inode = NULL; |
384 | struct buffer_head *bh = NULL; | |
385 | struct buffer_head *main_bm_bh = NULL; | |
386 | struct inode *main_bm_inode = NULL; | |
387 | struct ocfs2_dinode *alloc_copy = NULL; | |
388 | struct ocfs2_dinode *alloc = NULL; | |
389 | ||
9c7af40b MF |
390 | cancel_delayed_work(&osb->la_enable_wq); |
391 | flush_workqueue(ocfs2_wq); | |
392 | ||
ccd979bd | 393 | if (osb->local_alloc_state == OCFS2_LA_UNUSED) |
8898a5a5 | 394 | goto out; |
ccd979bd MF |
395 | |
396 | local_alloc_inode = | |
397 | ocfs2_get_system_file_inode(osb, | |
398 | LOCAL_ALLOC_SYSTEM_INODE, | |
399 | osb->slot_num); | |
400 | if (!local_alloc_inode) { | |
401 | status = -ENOENT; | |
402 | mlog_errno(status); | |
8898a5a5 | 403 | goto out; |
ccd979bd MF |
404 | } |
405 | ||
406 | osb->local_alloc_state = OCFS2_LA_DISABLED; | |
407 | ||
d02f00cc MF |
408 | ocfs2_resmap_uninit(&osb->osb_la_resmap); |
409 | ||
ccd979bd MF |
410 | main_bm_inode = ocfs2_get_system_file_inode(osb, |
411 | GLOBAL_BITMAP_SYSTEM_INODE, | |
412 | OCFS2_INVALID_SLOT); | |
413 | if (!main_bm_inode) { | |
414 | status = -EINVAL; | |
415 | mlog_errno(status); | |
8898a5a5 | 416 | goto out; |
ccd979bd MF |
417 | } |
418 | ||
8898a5a5 MF |
419 | mutex_lock(&main_bm_inode->i_mutex); |
420 | ||
e63aecb6 | 421 | status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); |
ccd979bd MF |
422 | if (status < 0) { |
423 | mlog_errno(status); | |
8898a5a5 | 424 | goto out_mutex; |
ccd979bd MF |
425 | } |
426 | ||
427 | /* WINDOW_MOVE_CREDITS is a bit heavy... */ | |
65eff9cc | 428 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
429 | if (IS_ERR(handle)) { |
430 | mlog_errno(PTR_ERR(handle)); | |
431 | handle = NULL; | |
8898a5a5 | 432 | goto out_unlock; |
ccd979bd MF |
433 | } |
434 | ||
435 | bh = osb->local_alloc_bh; | |
436 | alloc = (struct ocfs2_dinode *) bh->b_data; | |
437 | ||
4ba1c5bf | 438 | alloc_copy = kmalloc(bh->b_size, GFP_NOFS); |
ccd979bd MF |
439 | if (!alloc_copy) { |
440 | status = -ENOMEM; | |
8898a5a5 | 441 | goto out_commit; |
ccd979bd MF |
442 | } |
443 | memcpy(alloc_copy, alloc, bh->b_size); | |
444 | ||
0cf2f763 JB |
445 | status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode), |
446 | bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
447 | if (status < 0) { |
448 | mlog_errno(status); | |
8898a5a5 | 449 | goto out_commit; |
ccd979bd MF |
450 | } |
451 | ||
452 | ocfs2_clear_local_alloc(alloc); | |
ec20cec7 | 453 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
454 | |
455 | brelse(bh); | |
456 | osb->local_alloc_bh = NULL; | |
457 | osb->local_alloc_state = OCFS2_LA_UNUSED; | |
458 | ||
459 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | |
460 | main_bm_inode, main_bm_bh); | |
461 | if (status < 0) | |
462 | mlog_errno(status); | |
463 | ||
8898a5a5 | 464 | out_commit: |
02dc1af4 | 465 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 466 | |
8898a5a5 | 467 | out_unlock: |
a81cb88b | 468 | brelse(main_bm_bh); |
ccd979bd | 469 | |
e63aecb6 | 470 | ocfs2_inode_unlock(main_bm_inode, 1); |
ccd979bd | 471 | |
8898a5a5 MF |
472 | out_mutex: |
473 | mutex_unlock(&main_bm_inode->i_mutex); | |
474 | iput(main_bm_inode); | |
475 | ||
476 | out: | |
ccd979bd MF |
477 | if (local_alloc_inode) |
478 | iput(local_alloc_inode); | |
479 | ||
480 | if (alloc_copy) | |
481 | kfree(alloc_copy); | |
ccd979bd MF |
482 | } |
483 | ||
484 | /* | |
485 | * We want to free the bitmap bits outside of any recovery context as | |
486 | * we'll need a cluster lock to do so, but we must clear the local | |
487 | * alloc before giving up the recovered nodes journal. To solve this, | |
488 | * we kmalloc a copy of the local alloc before it's change for the | |
489 | * caller to process with ocfs2_complete_local_alloc_recovery | |
490 | */ | |
491 | int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb, | |
492 | int slot_num, | |
493 | struct ocfs2_dinode **alloc_copy) | |
494 | { | |
495 | int status = 0; | |
496 | struct buffer_head *alloc_bh = NULL; | |
497 | struct inode *inode = NULL; | |
498 | struct ocfs2_dinode *alloc; | |
499 | ||
ef6b689b | 500 | mlog(0, "(slot_num = %d)\n", slot_num); |
ccd979bd MF |
501 | |
502 | *alloc_copy = NULL; | |
503 | ||
504 | inode = ocfs2_get_system_file_inode(osb, | |
505 | LOCAL_ALLOC_SYSTEM_INODE, | |
506 | slot_num); | |
507 | if (!inode) { | |
508 | status = -EINVAL; | |
509 | mlog_errno(status); | |
510 | goto bail; | |
511 | } | |
512 | ||
1b1dcc1b | 513 | mutex_lock(&inode->i_mutex); |
ccd979bd | 514 | |
b657c95c JB |
515 | status = ocfs2_read_inode_block_full(inode, &alloc_bh, |
516 | OCFS2_BH_IGNORE_CACHE); | |
ccd979bd MF |
517 | if (status < 0) { |
518 | mlog_errno(status); | |
519 | goto bail; | |
520 | } | |
521 | ||
522 | *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL); | |
523 | if (!(*alloc_copy)) { | |
524 | status = -ENOMEM; | |
525 | goto bail; | |
526 | } | |
527 | memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size); | |
528 | ||
529 | alloc = (struct ocfs2_dinode *) alloc_bh->b_data; | |
530 | ocfs2_clear_local_alloc(alloc); | |
531 | ||
13723d00 | 532 | ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check); |
8cb471e8 | 533 | status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode)); |
ccd979bd MF |
534 | if (status < 0) |
535 | mlog_errno(status); | |
536 | ||
537 | bail: | |
538 | if ((status < 0) && (*alloc_copy)) { | |
539 | kfree(*alloc_copy); | |
540 | *alloc_copy = NULL; | |
541 | } | |
542 | ||
a81cb88b | 543 | brelse(alloc_bh); |
ccd979bd MF |
544 | |
545 | if (inode) { | |
1b1dcc1b | 546 | mutex_unlock(&inode->i_mutex); |
ccd979bd MF |
547 | iput(inode); |
548 | } | |
549 | ||
c1e8d35e TM |
550 | if (status) |
551 | mlog_errno(status); | |
ccd979bd MF |
552 | return status; |
553 | } | |
554 | ||
555 | /* | |
556 | * Step 2: By now, we've completed the journal recovery, we've stamped | |
557 | * a clean local alloc on disk and dropped the node out of the | |
558 | * recovery map. Dlm locks will no longer stall, so lets clear out the | |
559 | * main bitmap. | |
560 | */ | |
561 | int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb, | |
562 | struct ocfs2_dinode *alloc) | |
563 | { | |
564 | int status; | |
1fabe148 | 565 | handle_t *handle; |
ccd979bd | 566 | struct buffer_head *main_bm_bh = NULL; |
8898a5a5 | 567 | struct inode *main_bm_inode; |
ccd979bd | 568 | |
ccd979bd MF |
569 | main_bm_inode = ocfs2_get_system_file_inode(osb, |
570 | GLOBAL_BITMAP_SYSTEM_INODE, | |
571 | OCFS2_INVALID_SLOT); | |
572 | if (!main_bm_inode) { | |
573 | status = -EINVAL; | |
574 | mlog_errno(status); | |
8898a5a5 | 575 | goto out; |
ccd979bd MF |
576 | } |
577 | ||
8898a5a5 MF |
578 | mutex_lock(&main_bm_inode->i_mutex); |
579 | ||
e63aecb6 | 580 | status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); |
ccd979bd MF |
581 | if (status < 0) { |
582 | mlog_errno(status); | |
8898a5a5 | 583 | goto out_mutex; |
ccd979bd MF |
584 | } |
585 | ||
65eff9cc | 586 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
587 | if (IS_ERR(handle)) { |
588 | status = PTR_ERR(handle); | |
589 | handle = NULL; | |
590 | mlog_errno(status); | |
8898a5a5 | 591 | goto out_unlock; |
ccd979bd MF |
592 | } |
593 | ||
594 | /* we want the bitmap change to be recorded on disk asap */ | |
1fabe148 | 595 | handle->h_sync = 1; |
ccd979bd MF |
596 | |
597 | status = ocfs2_sync_local_to_main(osb, handle, alloc, | |
598 | main_bm_inode, main_bm_bh); | |
599 | if (status < 0) | |
600 | mlog_errno(status); | |
601 | ||
02dc1af4 | 602 | ocfs2_commit_trans(osb, handle); |
8898a5a5 MF |
603 | |
604 | out_unlock: | |
e63aecb6 | 605 | ocfs2_inode_unlock(main_bm_inode, 1); |
8898a5a5 MF |
606 | |
607 | out_mutex: | |
608 | mutex_unlock(&main_bm_inode->i_mutex); | |
ccd979bd | 609 | |
a81cb88b | 610 | brelse(main_bm_bh); |
ccd979bd | 611 | |
8898a5a5 | 612 | iput(main_bm_inode); |
ccd979bd | 613 | |
8898a5a5 | 614 | out: |
4d0ddb2c | 615 | if (!status) |
b89c5428 | 616 | ocfs2_init_steal_slots(osb); |
c1e8d35e TM |
617 | if (status) |
618 | mlog_errno(status); | |
ccd979bd MF |
619 | return status; |
620 | } | |
621 | ||
622 | /* | |
9c7af40b | 623 | * make sure we've got at least bits_wanted contiguous bits in the |
1b1dcc1b | 624 | * local alloc. You lose them when you drop i_mutex. |
ccd979bd MF |
625 | * |
626 | * We will add ourselves to the transaction passed in, but may start | |
627 | * our own in order to shift windows. | |
628 | */ | |
629 | int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb, | |
ccd979bd MF |
630 | u32 bits_wanted, |
631 | struct ocfs2_alloc_context *ac) | |
632 | { | |
633 | int status; | |
634 | struct ocfs2_dinode *alloc; | |
635 | struct inode *local_alloc_inode; | |
636 | unsigned int free_bits; | |
637 | ||
ccd979bd | 638 | BUG_ON(!ac); |
ccd979bd MF |
639 | |
640 | local_alloc_inode = | |
641 | ocfs2_get_system_file_inode(osb, | |
642 | LOCAL_ALLOC_SYSTEM_INODE, | |
643 | osb->slot_num); | |
644 | if (!local_alloc_inode) { | |
645 | status = -ENOENT; | |
646 | mlog_errno(status); | |
647 | goto bail; | |
648 | } | |
da5cbf2f MF |
649 | |
650 | mutex_lock(&local_alloc_inode->i_mutex); | |
651 | ||
9c7af40b MF |
652 | /* |
653 | * We must double check state and allocator bits because | |
654 | * another process may have changed them while holding i_mutex. | |
655 | */ | |
656 | spin_lock(&osb->osb_lock); | |
657 | if (!ocfs2_la_state_enabled(osb) || | |
658 | (bits_wanted > osb->local_alloc_bits)) { | |
659 | spin_unlock(&osb->osb_lock); | |
ccd979bd MF |
660 | status = -ENOSPC; |
661 | goto bail; | |
662 | } | |
9c7af40b | 663 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
664 | |
665 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
666 | ||
e407e397 | 667 | #ifdef CONFIG_OCFS2_DEBUG_FS |
ccd979bd MF |
668 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) != |
669 | ocfs2_local_alloc_count_bits(alloc)) { | |
b0697053 | 670 | ocfs2_error(osb->sb, "local alloc inode %llu says it has " |
ccd979bd | 671 | "%u free bits, but a count shows %u", |
b0697053 | 672 | (unsigned long long)le64_to_cpu(alloc->i_blkno), |
ccd979bd MF |
673 | le32_to_cpu(alloc->id1.bitmap1.i_used), |
674 | ocfs2_local_alloc_count_bits(alloc)); | |
675 | status = -EIO; | |
676 | goto bail; | |
677 | } | |
5a58c3ef | 678 | #endif |
ccd979bd MF |
679 | |
680 | free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) - | |
681 | le32_to_cpu(alloc->id1.bitmap1.i_used); | |
682 | if (bits_wanted > free_bits) { | |
683 | /* uhoh, window change time. */ | |
684 | status = | |
685 | ocfs2_local_alloc_slide_window(osb, local_alloc_inode); | |
686 | if (status < 0) { | |
687 | if (status != -ENOSPC) | |
688 | mlog_errno(status); | |
689 | goto bail; | |
690 | } | |
9c7af40b MF |
691 | |
692 | /* | |
693 | * Under certain conditions, the window slide code | |
694 | * might have reduced the number of bits available or | |
695 | * disabled the the local alloc entirely. Re-check | |
696 | * here and return -ENOSPC if necessary. | |
697 | */ | |
698 | status = -ENOSPC; | |
699 | if (!ocfs2_la_state_enabled(osb)) | |
700 | goto bail; | |
701 | ||
702 | free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) - | |
703 | le32_to_cpu(alloc->id1.bitmap1.i_used); | |
704 | if (bits_wanted > free_bits) | |
705 | goto bail; | |
ccd979bd MF |
706 | } |
707 | ||
1187c968 JB |
708 | if (ac->ac_max_block) |
709 | mlog(0, "Calling in_range for max block %llu\n", | |
710 | (unsigned long long)ac->ac_max_block); | |
711 | ||
8fccfc82 | 712 | ac->ac_inode = local_alloc_inode; |
a4a48911 TM |
713 | /* We should never use localalloc from another slot */ |
714 | ac->ac_alloc_slot = osb->slot_num; | |
8fccfc82 | 715 | ac->ac_which = OCFS2_AC_USE_LOCAL; |
ccd979bd MF |
716 | get_bh(osb->local_alloc_bh); |
717 | ac->ac_bh = osb->local_alloc_bh; | |
ccd979bd MF |
718 | status = 0; |
719 | bail: | |
bda0233b SM |
720 | if (status < 0 && local_alloc_inode) { |
721 | mutex_unlock(&local_alloc_inode->i_mutex); | |
8fccfc82 | 722 | iput(local_alloc_inode); |
bda0233b | 723 | } |
ccd979bd | 724 | |
2fbe8d1e SM |
725 | mlog(0, "bits=%d, slot=%d, ret=%d\n", bits_wanted, osb->slot_num, |
726 | status); | |
727 | ||
c1e8d35e TM |
728 | if (status) |
729 | mlog_errno(status); | |
ccd979bd MF |
730 | return status; |
731 | } | |
732 | ||
733 | int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb, | |
1fabe148 | 734 | handle_t *handle, |
ccd979bd | 735 | struct ocfs2_alloc_context *ac, |
415cb800 | 736 | u32 bits_wanted, |
ccd979bd MF |
737 | u32 *bit_off, |
738 | u32 *num_bits) | |
739 | { | |
740 | int status, start; | |
741 | struct inode *local_alloc_inode; | |
ccd979bd MF |
742 | void *bitmap; |
743 | struct ocfs2_dinode *alloc; | |
744 | struct ocfs2_local_alloc *la; | |
745 | ||
ccd979bd MF |
746 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL); |
747 | ||
ccd979bd MF |
748 | local_alloc_inode = ac->ac_inode; |
749 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
750 | la = OCFS2_LOCAL_ALLOC(alloc); | |
751 | ||
d02f00cc MF |
752 | start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted, |
753 | ac->ac_resv); | |
ccd979bd MF |
754 | if (start == -1) { |
755 | /* TODO: Shouldn't we just BUG here? */ | |
756 | status = -ENOSPC; | |
757 | mlog_errno(status); | |
758 | goto bail; | |
759 | } | |
760 | ||
761 | bitmap = la->la_bitmap; | |
762 | *bit_off = le32_to_cpu(la->la_bm_off) + start; | |
ccd979bd MF |
763 | *num_bits = bits_wanted; |
764 | ||
0cf2f763 JB |
765 | status = ocfs2_journal_access_di(handle, |
766 | INODE_CACHE(local_alloc_inode), | |
13723d00 JB |
767 | osb->local_alloc_bh, |
768 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
769 | if (status < 0) { |
770 | mlog_errno(status); | |
771 | goto bail; | |
772 | } | |
773 | ||
d02f00cc MF |
774 | ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start, |
775 | bits_wanted); | |
776 | ||
ccd979bd MF |
777 | while(bits_wanted--) |
778 | ocfs2_set_bit(start++, bitmap); | |
779 | ||
0dd3256e | 780 | le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits); |
ec20cec7 | 781 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); |
ccd979bd | 782 | |
ccd979bd | 783 | bail: |
c1e8d35e TM |
784 | if (status) |
785 | mlog_errno(status); | |
ccd979bd MF |
786 | return status; |
787 | } | |
788 | ||
789 | static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc) | |
790 | { | |
791 | int i; | |
792 | u8 *buffer; | |
793 | u32 count = 0; | |
794 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | |
795 | ||
ccd979bd MF |
796 | buffer = la->la_bitmap; |
797 | for (i = 0; i < le16_to_cpu(la->la_size); i++) | |
798 | count += hweight8(buffer[i]); | |
799 | ||
c1e8d35e | 800 | mlog(0, "count %u\n", count); |
ccd979bd MF |
801 | return count; |
802 | } | |
803 | ||
804 | static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb, | |
d02f00cc MF |
805 | struct ocfs2_dinode *alloc, |
806 | u32 *numbits, | |
807 | struct ocfs2_alloc_reservation *resv) | |
ccd979bd MF |
808 | { |
809 | int numfound, bitoff, left, startoff, lastzero; | |
d02f00cc MF |
810 | int local_resv = 0; |
811 | struct ocfs2_alloc_reservation r; | |
ccd979bd | 812 | void *bitmap = NULL; |
d02f00cc | 813 | struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap; |
ccd979bd | 814 | |
ef6b689b | 815 | mlog(0, "(numbits wanted = %u)\n", *numbits); |
ccd979bd MF |
816 | |
817 | if (!alloc->id1.bitmap1.i_total) { | |
818 | mlog(0, "No bits in my window!\n"); | |
819 | bitoff = -1; | |
820 | goto bail; | |
821 | } | |
822 | ||
d02f00cc MF |
823 | if (!resv) { |
824 | local_resv = 1; | |
825 | ocfs2_resv_init_once(&r); | |
826 | ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP); | |
827 | resv = &r; | |
828 | } | |
829 | ||
830 | numfound = *numbits; | |
831 | if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) { | |
832 | if (numfound < *numbits) | |
833 | *numbits = numfound; | |
834 | goto bail; | |
835 | } | |
836 | ||
837 | /* | |
838 | * Code error. While reservations are enabled, local | |
839 | * allocation should _always_ go through them. | |
840 | */ | |
841 | BUG_ON(osb->osb_resv_level != 0); | |
842 | ||
843 | /* | |
844 | * Reservations are disabled. Handle this the old way. | |
845 | */ | |
846 | ||
ccd979bd MF |
847 | bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap; |
848 | ||
849 | numfound = bitoff = startoff = 0; | |
850 | lastzero = -1; | |
851 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); | |
852 | while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) { | |
853 | if (bitoff == left) { | |
854 | /* mlog(0, "bitoff (%d) == left", bitoff); */ | |
855 | break; | |
856 | } | |
857 | /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, " | |
858 | "numfound = %d\n", bitoff, startoff, numfound);*/ | |
859 | ||
860 | /* Ok, we found a zero bit... is it contig. or do we | |
861 | * start over?*/ | |
862 | if (bitoff == startoff) { | |
863 | /* we found a zero */ | |
864 | numfound++; | |
865 | startoff++; | |
866 | } else { | |
867 | /* got a zero after some ones */ | |
868 | numfound = 1; | |
869 | startoff = bitoff+1; | |
870 | } | |
871 | /* we got everything we needed */ | |
d02f00cc | 872 | if (numfound == *numbits) { |
ccd979bd MF |
873 | /* mlog(0, "Found it all!\n"); */ |
874 | break; | |
875 | } | |
876 | } | |
877 | ||
878 | mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff, | |
879 | numfound); | |
880 | ||
3e4218df | 881 | if (numfound == *numbits) |
ccd979bd | 882 | bitoff = startoff - numfound; |
3e4218df | 883 | else |
ccd979bd MF |
884 | bitoff = -1; |
885 | ||
886 | bail: | |
d02f00cc MF |
887 | if (local_resv) |
888 | ocfs2_resv_discard(resmap, resv); | |
889 | ||
c1e8d35e | 890 | mlog(0, "bitoff %d\n", bitoff); |
ccd979bd MF |
891 | return bitoff; |
892 | } | |
893 | ||
894 | static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc) | |
895 | { | |
896 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | |
897 | int i; | |
ccd979bd MF |
898 | |
899 | alloc->id1.bitmap1.i_total = 0; | |
900 | alloc->id1.bitmap1.i_used = 0; | |
901 | la->la_bm_off = 0; | |
902 | for(i = 0; i < le16_to_cpu(la->la_size); i++) | |
903 | la->la_bitmap[i] = 0; | |
ccd979bd MF |
904 | } |
905 | ||
906 | #if 0 | |
907 | /* turn this on and uncomment below to aid debugging window shifts. */ | |
908 | static void ocfs2_verify_zero_bits(unsigned long *bitmap, | |
909 | unsigned int start, | |
910 | unsigned int count) | |
911 | { | |
912 | unsigned int tmp = count; | |
913 | while(tmp--) { | |
914 | if (ocfs2_test_bit(start + tmp, bitmap)) { | |
915 | printk("ocfs2_verify_zero_bits: start = %u, count = " | |
916 | "%u\n", start, count); | |
917 | printk("ocfs2_verify_zero_bits: bit %u is set!", | |
918 | start + tmp); | |
919 | BUG(); | |
920 | } | |
921 | } | |
922 | } | |
923 | #endif | |
924 | ||
925 | /* | |
926 | * sync the local alloc to main bitmap. | |
927 | * | |
928 | * assumes you've already locked the main bitmap -- the bitmap inode | |
929 | * passed is used for caching. | |
930 | */ | |
931 | static int ocfs2_sync_local_to_main(struct ocfs2_super *osb, | |
1fabe148 | 932 | handle_t *handle, |
ccd979bd MF |
933 | struct ocfs2_dinode *alloc, |
934 | struct inode *main_bm_inode, | |
935 | struct buffer_head *main_bm_bh) | |
936 | { | |
937 | int status = 0; | |
938 | int bit_off, left, count, start; | |
939 | u64 la_start_blk; | |
940 | u64 blkno; | |
941 | void *bitmap; | |
942 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | |
943 | ||
ef6b689b TM |
944 | mlog(0, "total = %u, used = %u\n", |
945 | le32_to_cpu(alloc->id1.bitmap1.i_total), | |
946 | le32_to_cpu(alloc->id1.bitmap1.i_used)); | |
ccd979bd MF |
947 | |
948 | if (!alloc->id1.bitmap1.i_total) { | |
949 | mlog(0, "nothing to sync!\n"); | |
950 | goto bail; | |
951 | } | |
952 | ||
953 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) == | |
954 | le32_to_cpu(alloc->id1.bitmap1.i_total)) { | |
955 | mlog(0, "all bits were taken!\n"); | |
956 | goto bail; | |
957 | } | |
958 | ||
959 | la_start_blk = ocfs2_clusters_to_blocks(osb->sb, | |
960 | le32_to_cpu(la->la_bm_off)); | |
961 | bitmap = la->la_bitmap; | |
962 | start = count = bit_off = 0; | |
963 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); | |
964 | ||
965 | while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start)) | |
966 | != -1) { | |
967 | if ((bit_off < left) && (bit_off == start)) { | |
968 | count++; | |
969 | start++; | |
970 | continue; | |
971 | } | |
972 | if (count) { | |
973 | blkno = la_start_blk + | |
974 | ocfs2_clusters_to_blocks(osb->sb, | |
975 | start - count); | |
976 | ||
b0697053 MF |
977 | mlog(0, "freeing %u bits starting at local alloc bit " |
978 | "%u (la_start_blk = %llu, blkno = %llu)\n", | |
979 | count, start - count, | |
980 | (unsigned long long)la_start_blk, | |
981 | (unsigned long long)blkno); | |
ccd979bd | 982 | |
b4414eea MF |
983 | status = ocfs2_release_clusters(handle, |
984 | main_bm_inode, | |
985 | main_bm_bh, blkno, | |
986 | count); | |
ccd979bd MF |
987 | if (status < 0) { |
988 | mlog_errno(status); | |
989 | goto bail; | |
990 | } | |
991 | } | |
992 | if (bit_off >= left) | |
993 | break; | |
994 | count = 1; | |
995 | start = bit_off + 1; | |
996 | } | |
997 | ||
998 | bail: | |
c1e8d35e TM |
999 | if (status) |
1000 | mlog_errno(status); | |
ccd979bd MF |
1001 | return status; |
1002 | } | |
1003 | ||
9c7af40b MF |
1004 | enum ocfs2_la_event { |
1005 | OCFS2_LA_EVENT_SLIDE, /* Normal window slide. */ | |
1006 | OCFS2_LA_EVENT_FRAGMENTED, /* The global bitmap has | |
1007 | * enough bits theoretically | |
1008 | * free, but a contiguous | |
1009 | * allocation could not be | |
1010 | * found. */ | |
1011 | OCFS2_LA_EVENT_ENOSPC, /* Global bitmap doesn't have | |
1012 | * enough bits free to satisfy | |
1013 | * our request. */ | |
1014 | }; | |
1015 | #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ) | |
1016 | /* | |
1017 | * Given an event, calculate the size of our next local alloc window. | |
1018 | * | |
1019 | * This should always be called under i_mutex of the local alloc inode | |
1020 | * so that local alloc disabling doesn't race with processes trying to | |
1021 | * use the allocator. | |
1022 | * | |
1023 | * Returns the state which the local alloc was left in. This value can | |
1024 | * be ignored by some paths. | |
1025 | */ | |
1026 | static int ocfs2_recalc_la_window(struct ocfs2_super *osb, | |
1027 | enum ocfs2_la_event event) | |
1028 | { | |
1029 | unsigned int bits; | |
1030 | int state; | |
1031 | ||
1032 | spin_lock(&osb->osb_lock); | |
1033 | if (osb->local_alloc_state == OCFS2_LA_DISABLED) { | |
1034 | WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED); | |
1035 | goto out_unlock; | |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * ENOSPC and fragmentation are treated similarly for now. | |
1040 | */ | |
1041 | if (event == OCFS2_LA_EVENT_ENOSPC || | |
1042 | event == OCFS2_LA_EVENT_FRAGMENTED) { | |
1043 | /* | |
1044 | * We ran out of contiguous space in the primary | |
1045 | * bitmap. Drastically reduce the number of bits used | |
1046 | * by local alloc until we have to disable it. | |
1047 | */ | |
1048 | bits = osb->local_alloc_bits >> 1; | |
1049 | if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) { | |
1050 | /* | |
1051 | * By setting state to THROTTLED, we'll keep | |
1052 | * the number of local alloc bits used down | |
1053 | * until an event occurs which would give us | |
1054 | * reason to assume the bitmap situation might | |
1055 | * have changed. | |
1056 | */ | |
1057 | osb->local_alloc_state = OCFS2_LA_THROTTLED; | |
1058 | osb->local_alloc_bits = bits; | |
1059 | } else { | |
1060 | osb->local_alloc_state = OCFS2_LA_DISABLED; | |
1061 | } | |
1062 | queue_delayed_work(ocfs2_wq, &osb->la_enable_wq, | |
1063 | OCFS2_LA_ENABLE_INTERVAL); | |
1064 | goto out_unlock; | |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * Don't increase the size of the local alloc window until we | |
1069 | * know we might be able to fulfill the request. Otherwise, we | |
1070 | * risk bouncing around the global bitmap during periods of | |
1071 | * low space. | |
1072 | */ | |
1073 | if (osb->local_alloc_state != OCFS2_LA_THROTTLED) | |
1074 | osb->local_alloc_bits = osb->local_alloc_default_bits; | |
1075 | ||
1076 | out_unlock: | |
1077 | state = osb->local_alloc_state; | |
1078 | spin_unlock(&osb->osb_lock); | |
1079 | ||
1080 | return state; | |
1081 | } | |
1082 | ||
ccd979bd | 1083 | static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb, |
ccd979bd MF |
1084 | struct ocfs2_alloc_context **ac, |
1085 | struct inode **bitmap_inode, | |
1086 | struct buffer_head **bitmap_bh) | |
1087 | { | |
1088 | int status; | |
1089 | ||
cd861280 | 1090 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
1091 | if (!(*ac)) { |
1092 | status = -ENOMEM; | |
1093 | mlog_errno(status); | |
1094 | goto bail; | |
1095 | } | |
1096 | ||
9c7af40b | 1097 | retry_enospc: |
b22b63eb | 1098 | (*ac)->ac_bits_wanted = osb->local_alloc_default_bits; |
ccd979bd | 1099 | status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac); |
9c7af40b MF |
1100 | if (status == -ENOSPC) { |
1101 | if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) == | |
1102 | OCFS2_LA_DISABLED) | |
1103 | goto bail; | |
1104 | ||
1105 | ocfs2_free_ac_resource(*ac); | |
1106 | memset(*ac, 0, sizeof(struct ocfs2_alloc_context)); | |
1107 | goto retry_enospc; | |
1108 | } | |
ccd979bd | 1109 | if (status < 0) { |
9c7af40b | 1110 | mlog_errno(status); |
ccd979bd MF |
1111 | goto bail; |
1112 | } | |
1113 | ||
1114 | *bitmap_inode = (*ac)->ac_inode; | |
1115 | igrab(*bitmap_inode); | |
1116 | *bitmap_bh = (*ac)->ac_bh; | |
1117 | get_bh(*bitmap_bh); | |
1118 | status = 0; | |
1119 | bail: | |
1120 | if ((status < 0) && *ac) { | |
1121 | ocfs2_free_alloc_context(*ac); | |
1122 | *ac = NULL; | |
1123 | } | |
1124 | ||
c1e8d35e TM |
1125 | if (status) |
1126 | mlog_errno(status); | |
ccd979bd MF |
1127 | return status; |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * pass it the bitmap lock in lock_bh if you have it. | |
1132 | */ | |
1133 | static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb, | |
1fabe148 | 1134 | handle_t *handle, |
ccd979bd MF |
1135 | struct ocfs2_alloc_context *ac) |
1136 | { | |
1137 | int status = 0; | |
1138 | u32 cluster_off, cluster_count; | |
1139 | struct ocfs2_dinode *alloc = NULL; | |
1140 | struct ocfs2_local_alloc *la; | |
1141 | ||
ccd979bd MF |
1142 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; |
1143 | la = OCFS2_LOCAL_ALLOC(alloc); | |
1144 | ||
1145 | if (alloc->id1.bitmap1.i_total) | |
1146 | mlog(0, "asking me to alloc a new window over a non-empty " | |
1147 | "one\n"); | |
1148 | ||
1149 | mlog(0, "Allocating %u clusters for a new window.\n", | |
ebcee4b5 | 1150 | osb->local_alloc_bits); |
883d4cae MF |
1151 | |
1152 | /* Instruct the allocation code to try the most recently used | |
1153 | * cluster group. We'll re-record the group used this pass | |
1154 | * below. */ | |
1155 | ac->ac_last_group = osb->la_last_gd; | |
1156 | ||
ccd979bd MF |
1157 | /* we used the generic suballoc reserve function, but we set |
1158 | * everything up nicely, so there's no reason why we can't use | |
1159 | * the more specific cluster api to claim bits. */ | |
1ed9b777 | 1160 | status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits, |
ccd979bd | 1161 | &cluster_off, &cluster_count); |
9c7af40b MF |
1162 | if (status == -ENOSPC) { |
1163 | retry_enospc: | |
1164 | /* | |
1165 | * Note: We could also try syncing the journal here to | |
1166 | * allow use of any free bits which the current | |
1167 | * transaction can't give us access to. --Mark | |
1168 | */ | |
1169 | if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) == | |
1170 | OCFS2_LA_DISABLED) | |
1171 | goto bail; | |
1172 | ||
b22b63eb | 1173 | ac->ac_bits_wanted = osb->local_alloc_default_bits; |
1ed9b777 | 1174 | status = ocfs2_claim_clusters(handle, ac, |
9c7af40b MF |
1175 | osb->local_alloc_bits, |
1176 | &cluster_off, | |
1177 | &cluster_count); | |
1178 | if (status == -ENOSPC) | |
1179 | goto retry_enospc; | |
1180 | /* | |
1181 | * We only shrunk the *minimum* number of in our | |
1182 | * request - it's entirely possible that the allocator | |
1183 | * might give us more than we asked for. | |
1184 | */ | |
1185 | if (status == 0) { | |
1186 | spin_lock(&osb->osb_lock); | |
1187 | osb->local_alloc_bits = cluster_count; | |
1188 | spin_unlock(&osb->osb_lock); | |
1189 | } | |
1190 | } | |
ccd979bd MF |
1191 | if (status < 0) { |
1192 | if (status != -ENOSPC) | |
1193 | mlog_errno(status); | |
1194 | goto bail; | |
1195 | } | |
1196 | ||
883d4cae MF |
1197 | osb->la_last_gd = ac->ac_last_group; |
1198 | ||
ccd979bd MF |
1199 | la->la_bm_off = cpu_to_le32(cluster_off); |
1200 | alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count); | |
1201 | /* just in case... In the future when we find space ourselves, | |
1202 | * we don't have to get all contiguous -- but we'll have to | |
1203 | * set all previously used bits in bitmap and update | |
1204 | * la_bits_set before setting the bits in the main bitmap. */ | |
1205 | alloc->id1.bitmap1.i_used = 0; | |
1206 | memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0, | |
1207 | le16_to_cpu(la->la_size)); | |
1208 | ||
d02f00cc MF |
1209 | ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count, |
1210 | OCFS2_LOCAL_ALLOC(alloc)->la_bitmap); | |
1211 | ||
ccd979bd MF |
1212 | mlog(0, "New window allocated:\n"); |
1213 | mlog(0, "window la_bm_off = %u\n", | |
1214 | OCFS2_LOCAL_ALLOC(alloc)->la_bm_off); | |
1215 | mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total)); | |
1216 | ||
1217 | bail: | |
c1e8d35e TM |
1218 | if (status) |
1219 | mlog_errno(status); | |
ccd979bd MF |
1220 | return status; |
1221 | } | |
1222 | ||
1223 | /* Note that we do *NOT* lock the local alloc inode here as | |
1224 | * it's been locked already for us. */ | |
1225 | static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb, | |
1226 | struct inode *local_alloc_inode) | |
1227 | { | |
1228 | int status = 0; | |
1229 | struct buffer_head *main_bm_bh = NULL; | |
1230 | struct inode *main_bm_inode = NULL; | |
1fabe148 | 1231 | handle_t *handle = NULL; |
ccd979bd MF |
1232 | struct ocfs2_dinode *alloc; |
1233 | struct ocfs2_dinode *alloc_copy = NULL; | |
1234 | struct ocfs2_alloc_context *ac = NULL; | |
1235 | ||
9c7af40b MF |
1236 | ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE); |
1237 | ||
ccd979bd MF |
1238 | /* This will lock the main bitmap for us. */ |
1239 | status = ocfs2_local_alloc_reserve_for_window(osb, | |
ccd979bd MF |
1240 | &ac, |
1241 | &main_bm_inode, | |
1242 | &main_bm_bh); | |
1243 | if (status < 0) { | |
1244 | if (status != -ENOSPC) | |
1245 | mlog_errno(status); | |
1246 | goto bail; | |
1247 | } | |
1248 | ||
65eff9cc | 1249 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
1250 | if (IS_ERR(handle)) { |
1251 | status = PTR_ERR(handle); | |
1252 | handle = NULL; | |
1253 | mlog_errno(status); | |
1254 | goto bail; | |
1255 | } | |
1256 | ||
1257 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
1258 | ||
1259 | /* We want to clear the local alloc before doing anything | |
1260 | * else, so that if we error later during this operation, | |
1261 | * local alloc shutdown won't try to double free main bitmap | |
1262 | * bits. Make a copy so the sync function knows which bits to | |
1263 | * free. */ | |
4ba1c5bf | 1264 | alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS); |
ccd979bd MF |
1265 | if (!alloc_copy) { |
1266 | status = -ENOMEM; | |
1267 | mlog_errno(status); | |
1268 | goto bail; | |
1269 | } | |
1270 | memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size); | |
1271 | ||
0cf2f763 JB |
1272 | status = ocfs2_journal_access_di(handle, |
1273 | INODE_CACHE(local_alloc_inode), | |
13723d00 JB |
1274 | osb->local_alloc_bh, |
1275 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
1276 | if (status < 0) { |
1277 | mlog_errno(status); | |
1278 | goto bail; | |
1279 | } | |
1280 | ||
1281 | ocfs2_clear_local_alloc(alloc); | |
ec20cec7 | 1282 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); |
ccd979bd MF |
1283 | |
1284 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | |
1285 | main_bm_inode, main_bm_bh); | |
1286 | if (status < 0) { | |
1287 | mlog_errno(status); | |
1288 | goto bail; | |
1289 | } | |
1290 | ||
1291 | status = ocfs2_local_alloc_new_window(osb, handle, ac); | |
1292 | if (status < 0) { | |
1293 | if (status != -ENOSPC) | |
1294 | mlog_errno(status); | |
1295 | goto bail; | |
1296 | } | |
1297 | ||
1298 | atomic_inc(&osb->alloc_stats.moves); | |
1299 | ||
ccd979bd MF |
1300 | bail: |
1301 | if (handle) | |
02dc1af4 | 1302 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1303 | |
a81cb88b | 1304 | brelse(main_bm_bh); |
ccd979bd MF |
1305 | |
1306 | if (main_bm_inode) | |
1307 | iput(main_bm_inode); | |
1308 | ||
1309 | if (alloc_copy) | |
1310 | kfree(alloc_copy); | |
1311 | ||
1312 | if (ac) | |
1313 | ocfs2_free_alloc_context(ac); | |
1314 | ||
c1e8d35e TM |
1315 | if (status) |
1316 | mlog_errno(status); | |
ccd979bd MF |
1317 | return status; |
1318 | } | |
1319 |