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 | ||
ccd979bd MF |
32 | #include <cluster/masklog.h> |
33 | ||
34 | #include "ocfs2.h" | |
35 | ||
36 | #include "alloc.h" | |
13723d00 | 37 | #include "blockcheck.h" |
ccd979bd MF |
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" | |
a04733d8 | 45 | #include "ocfs2_trace.h" |
ccd979bd MF |
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 | ||
a04733d8 | 204 | trace_ocfs2_la_set_sizes(requested_mb, la_max_mb, la_default_mb); |
73c8a800 MF |
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 | ||
9c7af40b MF |
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 | ||
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 | ||
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 | ||
ccd979bd MF |
251 | /* |
252 | * Tell us whether a given allocation should use the local alloc | |
253 | * file. Otherwise, it has to go to the main bitmap. | |
9c7af40b MF |
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. | |
ccd979bd MF |
257 | */ |
258 | int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits) | |
259 | { | |
2fbe8d1e | 260 | int ret = 0; |
9c7af40b MF |
261 | int la_bits; |
262 | ||
263 | spin_lock(&osb->osb_lock); | |
264 | la_bits = osb->local_alloc_bits; | |
ccd979bd | 265 | |
9c7af40b | 266 | if (!ocfs2_la_state_enabled(osb)) |
2fbe8d1e | 267 | goto bail; |
ccd979bd MF |
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)) | |
2fbe8d1e | 274 | goto bail; |
ccd979bd | 275 | |
2fbe8d1e SM |
276 | ret = 1; |
277 | bail: | |
a04733d8 TM |
278 | trace_ocfs2_alloc_should_use_local( |
279 | (unsigned long long)bits, osb->local_alloc_state, la_bits, ret); | |
9c7af40b | 280 | spin_unlock(&osb->osb_lock); |
2fbe8d1e | 281 | return ret; |
ccd979bd MF |
282 | } |
283 | ||
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 | ||
ebcee4b5 | 293 | if (osb->local_alloc_bits == 0) |
2fbe8d1e SM |
294 | goto bail; |
295 | ||
ebcee4b5 | 296 | if (osb->local_alloc_bits >= osb->bitmap_cpg) { |
2fbe8d1e SM |
297 | mlog(ML_NOTICE, "Requested local alloc window %d is larger " |
298 | "than max possible %u. Using defaults.\n", | |
ebcee4b5 MF |
299 | osb->local_alloc_bits, (osb->bitmap_cpg - 1)); |
300 | osb->local_alloc_bits = | |
301 | ocfs2_megabytes_to_clusters(osb->sb, | |
6b82021b | 302 | ocfs2_la_default_mb(osb)); |
2fbe8d1e SM |
303 | } |
304 | ||
ccd979bd MF |
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 | ||
b657c95c JB |
314 | status = ocfs2_read_inode_block_full(inode, &alloc_bh, |
315 | OCFS2_BH_IGNORE_CACHE); | |
ccd979bd MF |
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))) { | |
b0697053 MF |
326 | mlog(ML_ERROR, "Invalid local alloc inode, %llu\n", |
327 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
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) | |
a81cb88b | 360 | brelse(alloc_bh); |
72865d92 | 361 | iput(inode); |
ccd979bd | 362 | |
a04733d8 | 363 | trace_ocfs2_load_local_alloc(osb->local_alloc_bits); |
2fbe8d1e | 364 | |
c1e8d35e TM |
365 | if (status) |
366 | mlog_errno(status); | |
ccd979bd MF |
367 | return status; |
368 | } | |
369 | ||
370 | /* | |
371 | * return any unused bits to the bitmap and write out a clean | |
372 | * local_alloc. | |
373 | * | |
374 | * local_alloc_bh is optional. If not passed, we will simply use the | |
375 | * one off osb. If you do pass it however, be warned that it *will* be | |
376 | * returned brelse'd and NULL'd out.*/ | |
377 | void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb) | |
378 | { | |
379 | int status; | |
1fabe148 | 380 | handle_t *handle; |
ccd979bd MF |
381 | struct inode *local_alloc_inode = NULL; |
382 | struct buffer_head *bh = NULL; | |
383 | struct buffer_head *main_bm_bh = NULL; | |
384 | struct inode *main_bm_inode = NULL; | |
385 | struct ocfs2_dinode *alloc_copy = NULL; | |
386 | struct ocfs2_dinode *alloc = NULL; | |
387 | ||
9c7af40b | 388 | cancel_delayed_work(&osb->la_enable_wq); |
35ddf78e | 389 | flush_workqueue(osb->ocfs2_wq); |
9c7af40b | 390 | |
ccd979bd | 391 | if (osb->local_alloc_state == OCFS2_LA_UNUSED) |
8898a5a5 | 392 | goto out; |
ccd979bd MF |
393 | |
394 | local_alloc_inode = | |
395 | ocfs2_get_system_file_inode(osb, | |
396 | LOCAL_ALLOC_SYSTEM_INODE, | |
397 | osb->slot_num); | |
398 | if (!local_alloc_inode) { | |
399 | status = -ENOENT; | |
400 | mlog_errno(status); | |
8898a5a5 | 401 | goto out; |
ccd979bd MF |
402 | } |
403 | ||
404 | osb->local_alloc_state = OCFS2_LA_DISABLED; | |
405 | ||
d02f00cc MF |
406 | ocfs2_resmap_uninit(&osb->osb_la_resmap); |
407 | ||
ccd979bd MF |
408 | main_bm_inode = ocfs2_get_system_file_inode(osb, |
409 | GLOBAL_BITMAP_SYSTEM_INODE, | |
410 | OCFS2_INVALID_SLOT); | |
411 | if (!main_bm_inode) { | |
412 | status = -EINVAL; | |
413 | mlog_errno(status); | |
8898a5a5 | 414 | goto out; |
ccd979bd MF |
415 | } |
416 | ||
5955102c | 417 | inode_lock(main_bm_inode); |
8898a5a5 | 418 | |
e63aecb6 | 419 | status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); |
ccd979bd MF |
420 | if (status < 0) { |
421 | mlog_errno(status); | |
8898a5a5 | 422 | goto out_mutex; |
ccd979bd MF |
423 | } |
424 | ||
425 | /* WINDOW_MOVE_CREDITS is a bit heavy... */ | |
65eff9cc | 426 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
427 | if (IS_ERR(handle)) { |
428 | mlog_errno(PTR_ERR(handle)); | |
429 | handle = NULL; | |
8898a5a5 | 430 | goto out_unlock; |
ccd979bd MF |
431 | } |
432 | ||
433 | bh = osb->local_alloc_bh; | |
434 | alloc = (struct ocfs2_dinode *) bh->b_data; | |
435 | ||
4ba1c5bf | 436 | alloc_copy = kmalloc(bh->b_size, GFP_NOFS); |
ccd979bd MF |
437 | if (!alloc_copy) { |
438 | status = -ENOMEM; | |
8898a5a5 | 439 | goto out_commit; |
ccd979bd MF |
440 | } |
441 | memcpy(alloc_copy, alloc, bh->b_size); | |
442 | ||
0cf2f763 JB |
443 | status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode), |
444 | bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
445 | if (status < 0) { |
446 | mlog_errno(status); | |
8898a5a5 | 447 | goto out_commit; |
ccd979bd MF |
448 | } |
449 | ||
450 | ocfs2_clear_local_alloc(alloc); | |
ec20cec7 | 451 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
452 | |
453 | brelse(bh); | |
454 | osb->local_alloc_bh = NULL; | |
455 | osb->local_alloc_state = OCFS2_LA_UNUSED; | |
456 | ||
457 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | |
458 | main_bm_inode, main_bm_bh); | |
459 | if (status < 0) | |
460 | mlog_errno(status); | |
461 | ||
8898a5a5 | 462 | out_commit: |
02dc1af4 | 463 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 464 | |
8898a5a5 | 465 | out_unlock: |
a81cb88b | 466 | brelse(main_bm_bh); |
ccd979bd | 467 | |
e63aecb6 | 468 | ocfs2_inode_unlock(main_bm_inode, 1); |
ccd979bd | 469 | |
8898a5a5 | 470 | out_mutex: |
5955102c | 471 | inode_unlock(main_bm_inode); |
8898a5a5 MF |
472 | iput(main_bm_inode); |
473 | ||
474 | out: | |
72865d92 | 475 | iput(local_alloc_inode); |
ccd979bd | 476 | |
d787ab09 | 477 | kfree(alloc_copy); |
ccd979bd MF |
478 | } |
479 | ||
480 | /* | |
481 | * We want to free the bitmap bits outside of any recovery context as | |
482 | * we'll need a cluster lock to do so, but we must clear the local | |
483 | * alloc before giving up the recovered nodes journal. To solve this, | |
484 | * we kmalloc a copy of the local alloc before it's change for the | |
485 | * caller to process with ocfs2_complete_local_alloc_recovery | |
486 | */ | |
487 | int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb, | |
488 | int slot_num, | |
489 | struct ocfs2_dinode **alloc_copy) | |
490 | { | |
491 | int status = 0; | |
492 | struct buffer_head *alloc_bh = NULL; | |
493 | struct inode *inode = NULL; | |
494 | struct ocfs2_dinode *alloc; | |
495 | ||
a04733d8 | 496 | trace_ocfs2_begin_local_alloc_recovery(slot_num); |
ccd979bd MF |
497 | |
498 | *alloc_copy = NULL; | |
499 | ||
500 | inode = ocfs2_get_system_file_inode(osb, | |
501 | LOCAL_ALLOC_SYSTEM_INODE, | |
502 | slot_num); | |
503 | if (!inode) { | |
504 | status = -EINVAL; | |
505 | mlog_errno(status); | |
506 | goto bail; | |
507 | } | |
508 | ||
5955102c | 509 | inode_lock(inode); |
ccd979bd | 510 | |
b657c95c JB |
511 | status = ocfs2_read_inode_block_full(inode, &alloc_bh, |
512 | OCFS2_BH_IGNORE_CACHE); | |
ccd979bd MF |
513 | if (status < 0) { |
514 | mlog_errno(status); | |
515 | goto bail; | |
516 | } | |
517 | ||
518 | *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL); | |
519 | if (!(*alloc_copy)) { | |
520 | status = -ENOMEM; | |
521 | goto bail; | |
522 | } | |
523 | memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size); | |
524 | ||
525 | alloc = (struct ocfs2_dinode *) alloc_bh->b_data; | |
526 | ocfs2_clear_local_alloc(alloc); | |
527 | ||
13723d00 | 528 | ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check); |
8cb471e8 | 529 | status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode)); |
ccd979bd MF |
530 | if (status < 0) |
531 | mlog_errno(status); | |
532 | ||
533 | bail: | |
d787ab09 | 534 | if (status < 0) { |
ccd979bd MF |
535 | kfree(*alloc_copy); |
536 | *alloc_copy = NULL; | |
537 | } | |
538 | ||
a81cb88b | 539 | brelse(alloc_bh); |
ccd979bd MF |
540 | |
541 | if (inode) { | |
5955102c | 542 | inode_unlock(inode); |
ccd979bd MF |
543 | iput(inode); |
544 | } | |
545 | ||
c1e8d35e TM |
546 | if (status) |
547 | mlog_errno(status); | |
ccd979bd MF |
548 | return status; |
549 | } | |
550 | ||
551 | /* | |
552 | * Step 2: By now, we've completed the journal recovery, we've stamped | |
553 | * a clean local alloc on disk and dropped the node out of the | |
554 | * recovery map. Dlm locks will no longer stall, so lets clear out the | |
555 | * main bitmap. | |
556 | */ | |
557 | int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb, | |
558 | struct ocfs2_dinode *alloc) | |
559 | { | |
560 | int status; | |
1fabe148 | 561 | handle_t *handle; |
ccd979bd | 562 | struct buffer_head *main_bm_bh = NULL; |
8898a5a5 | 563 | struct inode *main_bm_inode; |
ccd979bd | 564 | |
ccd979bd MF |
565 | main_bm_inode = ocfs2_get_system_file_inode(osb, |
566 | GLOBAL_BITMAP_SYSTEM_INODE, | |
567 | OCFS2_INVALID_SLOT); | |
568 | if (!main_bm_inode) { | |
569 | status = -EINVAL; | |
570 | mlog_errno(status); | |
8898a5a5 | 571 | goto out; |
ccd979bd MF |
572 | } |
573 | ||
5955102c | 574 | inode_lock(main_bm_inode); |
8898a5a5 | 575 | |
e63aecb6 | 576 | status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); |
ccd979bd MF |
577 | if (status < 0) { |
578 | mlog_errno(status); | |
8898a5a5 | 579 | goto out_mutex; |
ccd979bd MF |
580 | } |
581 | ||
65eff9cc | 582 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
583 | if (IS_ERR(handle)) { |
584 | status = PTR_ERR(handle); | |
585 | handle = NULL; | |
586 | mlog_errno(status); | |
8898a5a5 | 587 | goto out_unlock; |
ccd979bd MF |
588 | } |
589 | ||
590 | /* we want the bitmap change to be recorded on disk asap */ | |
1fabe148 | 591 | handle->h_sync = 1; |
ccd979bd MF |
592 | |
593 | status = ocfs2_sync_local_to_main(osb, handle, alloc, | |
594 | main_bm_inode, main_bm_bh); | |
595 | if (status < 0) | |
596 | mlog_errno(status); | |
597 | ||
02dc1af4 | 598 | ocfs2_commit_trans(osb, handle); |
8898a5a5 MF |
599 | |
600 | out_unlock: | |
e63aecb6 | 601 | ocfs2_inode_unlock(main_bm_inode, 1); |
8898a5a5 MF |
602 | |
603 | out_mutex: | |
5955102c | 604 | inode_unlock(main_bm_inode); |
ccd979bd | 605 | |
a81cb88b | 606 | brelse(main_bm_bh); |
ccd979bd | 607 | |
8898a5a5 | 608 | iput(main_bm_inode); |
ccd979bd | 609 | |
8898a5a5 | 610 | out: |
4d0ddb2c | 611 | if (!status) |
b89c5428 | 612 | ocfs2_init_steal_slots(osb); |
c1e8d35e TM |
613 | if (status) |
614 | mlog_errno(status); | |
ccd979bd MF |
615 | return status; |
616 | } | |
617 | ||
618 | /* | |
9c7af40b | 619 | * make sure we've got at least bits_wanted contiguous bits in the |
1b1dcc1b | 620 | * local alloc. You lose them when you drop i_mutex. |
ccd979bd MF |
621 | * |
622 | * We will add ourselves to the transaction passed in, but may start | |
623 | * our own in order to shift windows. | |
624 | */ | |
625 | int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb, | |
ccd979bd MF |
626 | u32 bits_wanted, |
627 | struct ocfs2_alloc_context *ac) | |
628 | { | |
629 | int status; | |
630 | struct ocfs2_dinode *alloc; | |
631 | struct inode *local_alloc_inode; | |
632 | unsigned int free_bits; | |
633 | ||
ccd979bd | 634 | BUG_ON(!ac); |
ccd979bd MF |
635 | |
636 | local_alloc_inode = | |
637 | ocfs2_get_system_file_inode(osb, | |
638 | LOCAL_ALLOC_SYSTEM_INODE, | |
639 | osb->slot_num); | |
640 | if (!local_alloc_inode) { | |
641 | status = -ENOENT; | |
642 | mlog_errno(status); | |
643 | goto bail; | |
644 | } | |
da5cbf2f | 645 | |
5955102c | 646 | inode_lock(local_alloc_inode); |
da5cbf2f | 647 | |
9c7af40b MF |
648 | /* |
649 | * We must double check state and allocator bits because | |
650 | * another process may have changed them while holding i_mutex. | |
651 | */ | |
652 | spin_lock(&osb->osb_lock); | |
653 | if (!ocfs2_la_state_enabled(osb) || | |
654 | (bits_wanted > osb->local_alloc_bits)) { | |
655 | spin_unlock(&osb->osb_lock); | |
ccd979bd MF |
656 | status = -ENOSPC; |
657 | goto bail; | |
658 | } | |
9c7af40b | 659 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
660 | |
661 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
662 | ||
e407e397 | 663 | #ifdef CONFIG_OCFS2_DEBUG_FS |
ccd979bd MF |
664 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) != |
665 | ocfs2_local_alloc_count_bits(alloc)) { | |
7ecef14a | 666 | ocfs2_error(osb->sb, "local alloc inode %llu says it has %u used bits, but a count shows %u\n", |
b0697053 | 667 | (unsigned long long)le64_to_cpu(alloc->i_blkno), |
ccd979bd MF |
668 | le32_to_cpu(alloc->id1.bitmap1.i_used), |
669 | ocfs2_local_alloc_count_bits(alloc)); | |
670 | status = -EIO; | |
671 | goto bail; | |
672 | } | |
5a58c3ef | 673 | #endif |
ccd979bd MF |
674 | |
675 | free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) - | |
676 | le32_to_cpu(alloc->id1.bitmap1.i_used); | |
677 | if (bits_wanted > free_bits) { | |
678 | /* uhoh, window change time. */ | |
679 | status = | |
680 | ocfs2_local_alloc_slide_window(osb, local_alloc_inode); | |
681 | if (status < 0) { | |
682 | if (status != -ENOSPC) | |
683 | mlog_errno(status); | |
684 | goto bail; | |
685 | } | |
9c7af40b MF |
686 | |
687 | /* | |
688 | * Under certain conditions, the window slide code | |
689 | * might have reduced the number of bits available or | |
690 | * disabled the the local alloc entirely. Re-check | |
691 | * here and return -ENOSPC if necessary. | |
692 | */ | |
693 | status = -ENOSPC; | |
694 | if (!ocfs2_la_state_enabled(osb)) | |
695 | goto bail; | |
696 | ||
697 | free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) - | |
698 | le32_to_cpu(alloc->id1.bitmap1.i_used); | |
699 | if (bits_wanted > free_bits) | |
700 | goto bail; | |
ccd979bd MF |
701 | } |
702 | ||
8fccfc82 | 703 | ac->ac_inode = local_alloc_inode; |
a4a48911 TM |
704 | /* We should never use localalloc from another slot */ |
705 | ac->ac_alloc_slot = osb->slot_num; | |
8fccfc82 | 706 | ac->ac_which = OCFS2_AC_USE_LOCAL; |
ccd979bd MF |
707 | get_bh(osb->local_alloc_bh); |
708 | ac->ac_bh = osb->local_alloc_bh; | |
ccd979bd MF |
709 | status = 0; |
710 | bail: | |
bda0233b | 711 | if (status < 0 && local_alloc_inode) { |
5955102c | 712 | inode_unlock(local_alloc_inode); |
8fccfc82 | 713 | iput(local_alloc_inode); |
bda0233b | 714 | } |
ccd979bd | 715 | |
a04733d8 TM |
716 | trace_ocfs2_reserve_local_alloc_bits( |
717 | (unsigned long long)ac->ac_max_block, | |
718 | bits_wanted, osb->slot_num, status); | |
2fbe8d1e | 719 | |
c1e8d35e TM |
720 | if (status) |
721 | mlog_errno(status); | |
ccd979bd MF |
722 | return status; |
723 | } | |
724 | ||
725 | int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb, | |
1fabe148 | 726 | handle_t *handle, |
ccd979bd | 727 | struct ocfs2_alloc_context *ac, |
415cb800 | 728 | u32 bits_wanted, |
ccd979bd MF |
729 | u32 *bit_off, |
730 | u32 *num_bits) | |
731 | { | |
732 | int status, start; | |
733 | struct inode *local_alloc_inode; | |
ccd979bd MF |
734 | void *bitmap; |
735 | struct ocfs2_dinode *alloc; | |
736 | struct ocfs2_local_alloc *la; | |
737 | ||
ccd979bd MF |
738 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL); |
739 | ||
ccd979bd MF |
740 | local_alloc_inode = ac->ac_inode; |
741 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
742 | la = OCFS2_LOCAL_ALLOC(alloc); | |
743 | ||
d02f00cc MF |
744 | start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted, |
745 | ac->ac_resv); | |
ccd979bd MF |
746 | if (start == -1) { |
747 | /* TODO: Shouldn't we just BUG here? */ | |
748 | status = -ENOSPC; | |
749 | mlog_errno(status); | |
750 | goto bail; | |
751 | } | |
752 | ||
753 | bitmap = la->la_bitmap; | |
754 | *bit_off = le32_to_cpu(la->la_bm_off) + start; | |
ccd979bd MF |
755 | *num_bits = bits_wanted; |
756 | ||
0cf2f763 JB |
757 | status = ocfs2_journal_access_di(handle, |
758 | INODE_CACHE(local_alloc_inode), | |
13723d00 JB |
759 | osb->local_alloc_bh, |
760 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
761 | if (status < 0) { |
762 | mlog_errno(status); | |
763 | goto bail; | |
764 | } | |
765 | ||
d02f00cc MF |
766 | ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start, |
767 | bits_wanted); | |
768 | ||
ccd979bd MF |
769 | while(bits_wanted--) |
770 | ocfs2_set_bit(start++, bitmap); | |
771 | ||
0dd3256e | 772 | le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits); |
ec20cec7 | 773 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); |
ccd979bd | 774 | |
ccd979bd | 775 | bail: |
c1e8d35e TM |
776 | if (status) |
777 | mlog_errno(status); | |
ccd979bd MF |
778 | return status; |
779 | } | |
780 | ||
fb951eb5 ZW |
781 | int ocfs2_free_local_alloc_bits(struct ocfs2_super *osb, |
782 | handle_t *handle, | |
783 | struct ocfs2_alloc_context *ac, | |
784 | u32 bit_off, | |
785 | u32 num_bits) | |
786 | { | |
787 | int status, start; | |
788 | u32 clear_bits; | |
789 | struct inode *local_alloc_inode; | |
790 | void *bitmap; | |
791 | struct ocfs2_dinode *alloc; | |
792 | struct ocfs2_local_alloc *la; | |
793 | ||
794 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL); | |
795 | ||
796 | local_alloc_inode = ac->ac_inode; | |
797 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
798 | la = OCFS2_LOCAL_ALLOC(alloc); | |
799 | ||
800 | bitmap = la->la_bitmap; | |
801 | start = bit_off - le32_to_cpu(la->la_bm_off); | |
802 | clear_bits = num_bits; | |
803 | ||
804 | status = ocfs2_journal_access_di(handle, | |
805 | INODE_CACHE(local_alloc_inode), | |
806 | osb->local_alloc_bh, | |
807 | OCFS2_JOURNAL_ACCESS_WRITE); | |
808 | if (status < 0) { | |
809 | mlog_errno(status); | |
810 | goto bail; | |
811 | } | |
812 | ||
813 | while (clear_bits--) | |
814 | ocfs2_clear_bit(start++, bitmap); | |
815 | ||
816 | le32_add_cpu(&alloc->id1.bitmap1.i_used, -num_bits); | |
817 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); | |
818 | ||
819 | bail: | |
820 | return status; | |
821 | } | |
822 | ||
ccd979bd MF |
823 | static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc) |
824 | { | |
a75613ec | 825 | u32 count; |
ccd979bd MF |
826 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); |
827 | ||
a75613ec | 828 | count = memweight(la->la_bitmap, le16_to_cpu(la->la_size)); |
ccd979bd | 829 | |
a04733d8 | 830 | trace_ocfs2_local_alloc_count_bits(count); |
ccd979bd MF |
831 | return count; |
832 | } | |
833 | ||
834 | static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb, | |
d02f00cc MF |
835 | struct ocfs2_dinode *alloc, |
836 | u32 *numbits, | |
837 | struct ocfs2_alloc_reservation *resv) | |
ccd979bd | 838 | { |
023d4ea3 | 839 | int numfound = 0, bitoff, left, startoff, lastzero; |
d02f00cc MF |
840 | int local_resv = 0; |
841 | struct ocfs2_alloc_reservation r; | |
ccd979bd | 842 | void *bitmap = NULL; |
d02f00cc | 843 | struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap; |
ccd979bd | 844 | |
ccd979bd | 845 | if (!alloc->id1.bitmap1.i_total) { |
ccd979bd MF |
846 | bitoff = -1; |
847 | goto bail; | |
848 | } | |
849 | ||
d02f00cc MF |
850 | if (!resv) { |
851 | local_resv = 1; | |
852 | ocfs2_resv_init_once(&r); | |
853 | ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP); | |
854 | resv = &r; | |
855 | } | |
856 | ||
857 | numfound = *numbits; | |
858 | if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) { | |
859 | if (numfound < *numbits) | |
860 | *numbits = numfound; | |
861 | goto bail; | |
862 | } | |
863 | ||
864 | /* | |
865 | * Code error. While reservations are enabled, local | |
866 | * allocation should _always_ go through them. | |
867 | */ | |
868 | BUG_ON(osb->osb_resv_level != 0); | |
869 | ||
870 | /* | |
871 | * Reservations are disabled. Handle this the old way. | |
872 | */ | |
873 | ||
ccd979bd MF |
874 | bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap; |
875 | ||
876 | numfound = bitoff = startoff = 0; | |
877 | lastzero = -1; | |
878 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); | |
879 | while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) { | |
880 | if (bitoff == left) { | |
881 | /* mlog(0, "bitoff (%d) == left", bitoff); */ | |
882 | break; | |
883 | } | |
884 | /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, " | |
885 | "numfound = %d\n", bitoff, startoff, numfound);*/ | |
886 | ||
887 | /* Ok, we found a zero bit... is it contig. or do we | |
888 | * start over?*/ | |
889 | if (bitoff == startoff) { | |
890 | /* we found a zero */ | |
891 | numfound++; | |
892 | startoff++; | |
893 | } else { | |
894 | /* got a zero after some ones */ | |
895 | numfound = 1; | |
896 | startoff = bitoff+1; | |
897 | } | |
898 | /* we got everything we needed */ | |
d02f00cc | 899 | if (numfound == *numbits) { |
ccd979bd MF |
900 | /* mlog(0, "Found it all!\n"); */ |
901 | break; | |
902 | } | |
903 | } | |
904 | ||
a04733d8 | 905 | trace_ocfs2_local_alloc_find_clear_bits_search_bitmap(bitoff, numfound); |
ccd979bd | 906 | |
3e4218df | 907 | if (numfound == *numbits) |
ccd979bd | 908 | bitoff = startoff - numfound; |
3e4218df | 909 | else |
ccd979bd MF |
910 | bitoff = -1; |
911 | ||
912 | bail: | |
d02f00cc MF |
913 | if (local_resv) |
914 | ocfs2_resv_discard(resmap, resv); | |
915 | ||
a04733d8 TM |
916 | trace_ocfs2_local_alloc_find_clear_bits(*numbits, |
917 | le32_to_cpu(alloc->id1.bitmap1.i_total), | |
918 | bitoff, numfound); | |
919 | ||
ccd979bd MF |
920 | return bitoff; |
921 | } | |
922 | ||
923 | static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc) | |
924 | { | |
925 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | |
926 | int i; | |
ccd979bd MF |
927 | |
928 | alloc->id1.bitmap1.i_total = 0; | |
929 | alloc->id1.bitmap1.i_used = 0; | |
930 | la->la_bm_off = 0; | |
931 | for(i = 0; i < le16_to_cpu(la->la_size); i++) | |
932 | la->la_bitmap[i] = 0; | |
ccd979bd MF |
933 | } |
934 | ||
935 | #if 0 | |
936 | /* turn this on and uncomment below to aid debugging window shifts. */ | |
937 | static void ocfs2_verify_zero_bits(unsigned long *bitmap, | |
938 | unsigned int start, | |
939 | unsigned int count) | |
940 | { | |
941 | unsigned int tmp = count; | |
942 | while(tmp--) { | |
943 | if (ocfs2_test_bit(start + tmp, bitmap)) { | |
944 | printk("ocfs2_verify_zero_bits: start = %u, count = " | |
945 | "%u\n", start, count); | |
946 | printk("ocfs2_verify_zero_bits: bit %u is set!", | |
947 | start + tmp); | |
948 | BUG(); | |
949 | } | |
950 | } | |
951 | } | |
952 | #endif | |
953 | ||
954 | /* | |
955 | * sync the local alloc to main bitmap. | |
956 | * | |
957 | * assumes you've already locked the main bitmap -- the bitmap inode | |
958 | * passed is used for caching. | |
959 | */ | |
960 | static int ocfs2_sync_local_to_main(struct ocfs2_super *osb, | |
1fabe148 | 961 | handle_t *handle, |
ccd979bd MF |
962 | struct ocfs2_dinode *alloc, |
963 | struct inode *main_bm_inode, | |
964 | struct buffer_head *main_bm_bh) | |
965 | { | |
966 | int status = 0; | |
967 | int bit_off, left, count, start; | |
968 | u64 la_start_blk; | |
969 | u64 blkno; | |
970 | void *bitmap; | |
971 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | |
972 | ||
a04733d8 | 973 | trace_ocfs2_sync_local_to_main( |
ef6b689b TM |
974 | le32_to_cpu(alloc->id1.bitmap1.i_total), |
975 | le32_to_cpu(alloc->id1.bitmap1.i_used)); | |
ccd979bd MF |
976 | |
977 | if (!alloc->id1.bitmap1.i_total) { | |
ccd979bd MF |
978 | goto bail; |
979 | } | |
980 | ||
981 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) == | |
982 | le32_to_cpu(alloc->id1.bitmap1.i_total)) { | |
ccd979bd MF |
983 | goto bail; |
984 | } | |
985 | ||
986 | la_start_blk = ocfs2_clusters_to_blocks(osb->sb, | |
987 | le32_to_cpu(la->la_bm_off)); | |
988 | bitmap = la->la_bitmap; | |
989 | start = count = bit_off = 0; | |
990 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); | |
991 | ||
992 | while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start)) | |
993 | != -1) { | |
994 | if ((bit_off < left) && (bit_off == start)) { | |
995 | count++; | |
996 | start++; | |
997 | continue; | |
998 | } | |
999 | if (count) { | |
1000 | blkno = la_start_blk + | |
1001 | ocfs2_clusters_to_blocks(osb->sb, | |
1002 | start - count); | |
1003 | ||
a04733d8 | 1004 | trace_ocfs2_sync_local_to_main_free( |
b0697053 MF |
1005 | count, start - count, |
1006 | (unsigned long long)la_start_blk, | |
1007 | (unsigned long long)blkno); | |
ccd979bd | 1008 | |
b4414eea MF |
1009 | status = ocfs2_release_clusters(handle, |
1010 | main_bm_inode, | |
1011 | main_bm_bh, blkno, | |
1012 | count); | |
ccd979bd MF |
1013 | if (status < 0) { |
1014 | mlog_errno(status); | |
1015 | goto bail; | |
1016 | } | |
1017 | } | |
1018 | if (bit_off >= left) | |
1019 | break; | |
1020 | count = 1; | |
1021 | start = bit_off + 1; | |
1022 | } | |
1023 | ||
1024 | bail: | |
c1e8d35e TM |
1025 | if (status) |
1026 | mlog_errno(status); | |
ccd979bd MF |
1027 | return status; |
1028 | } | |
1029 | ||
9c7af40b MF |
1030 | enum ocfs2_la_event { |
1031 | OCFS2_LA_EVENT_SLIDE, /* Normal window slide. */ | |
1032 | OCFS2_LA_EVENT_FRAGMENTED, /* The global bitmap has | |
1033 | * enough bits theoretically | |
1034 | * free, but a contiguous | |
1035 | * allocation could not be | |
1036 | * found. */ | |
1037 | OCFS2_LA_EVENT_ENOSPC, /* Global bitmap doesn't have | |
1038 | * enough bits free to satisfy | |
1039 | * our request. */ | |
1040 | }; | |
1041 | #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ) | |
1042 | /* | |
1043 | * Given an event, calculate the size of our next local alloc window. | |
1044 | * | |
1045 | * This should always be called under i_mutex of the local alloc inode | |
1046 | * so that local alloc disabling doesn't race with processes trying to | |
1047 | * use the allocator. | |
1048 | * | |
1049 | * Returns the state which the local alloc was left in. This value can | |
1050 | * be ignored by some paths. | |
1051 | */ | |
1052 | static int ocfs2_recalc_la_window(struct ocfs2_super *osb, | |
1053 | enum ocfs2_la_event event) | |
1054 | { | |
1055 | unsigned int bits; | |
1056 | int state; | |
1057 | ||
1058 | spin_lock(&osb->osb_lock); | |
1059 | if (osb->local_alloc_state == OCFS2_LA_DISABLED) { | |
1060 | WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED); | |
1061 | goto out_unlock; | |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * ENOSPC and fragmentation are treated similarly for now. | |
1066 | */ | |
1067 | if (event == OCFS2_LA_EVENT_ENOSPC || | |
1068 | event == OCFS2_LA_EVENT_FRAGMENTED) { | |
1069 | /* | |
1070 | * We ran out of contiguous space in the primary | |
1071 | * bitmap. Drastically reduce the number of bits used | |
1072 | * by local alloc until we have to disable it. | |
1073 | */ | |
1074 | bits = osb->local_alloc_bits >> 1; | |
1075 | if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) { | |
1076 | /* | |
1077 | * By setting state to THROTTLED, we'll keep | |
1078 | * the number of local alloc bits used down | |
1079 | * until an event occurs which would give us | |
1080 | * reason to assume the bitmap situation might | |
1081 | * have changed. | |
1082 | */ | |
1083 | osb->local_alloc_state = OCFS2_LA_THROTTLED; | |
1084 | osb->local_alloc_bits = bits; | |
1085 | } else { | |
1086 | osb->local_alloc_state = OCFS2_LA_DISABLED; | |
1087 | } | |
35ddf78e | 1088 | queue_delayed_work(osb->ocfs2_wq, &osb->la_enable_wq, |
9c7af40b MF |
1089 | OCFS2_LA_ENABLE_INTERVAL); |
1090 | goto out_unlock; | |
1091 | } | |
1092 | ||
1093 | /* | |
1094 | * Don't increase the size of the local alloc window until we | |
1095 | * know we might be able to fulfill the request. Otherwise, we | |
1096 | * risk bouncing around the global bitmap during periods of | |
1097 | * low space. | |
1098 | */ | |
1099 | if (osb->local_alloc_state != OCFS2_LA_THROTTLED) | |
1100 | osb->local_alloc_bits = osb->local_alloc_default_bits; | |
1101 | ||
1102 | out_unlock: | |
1103 | state = osb->local_alloc_state; | |
1104 | spin_unlock(&osb->osb_lock); | |
1105 | ||
1106 | return state; | |
1107 | } | |
1108 | ||
ccd979bd | 1109 | static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb, |
ccd979bd MF |
1110 | struct ocfs2_alloc_context **ac, |
1111 | struct inode **bitmap_inode, | |
1112 | struct buffer_head **bitmap_bh) | |
1113 | { | |
1114 | int status; | |
1115 | ||
cd861280 | 1116 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
1117 | if (!(*ac)) { |
1118 | status = -ENOMEM; | |
1119 | mlog_errno(status); | |
1120 | goto bail; | |
1121 | } | |
1122 | ||
9c7af40b | 1123 | retry_enospc: |
7e9b7937 | 1124 | (*ac)->ac_bits_wanted = osb->local_alloc_bits; |
ccd979bd | 1125 | status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac); |
9c7af40b MF |
1126 | if (status == -ENOSPC) { |
1127 | if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) == | |
1128 | OCFS2_LA_DISABLED) | |
1129 | goto bail; | |
1130 | ||
1131 | ocfs2_free_ac_resource(*ac); | |
1132 | memset(*ac, 0, sizeof(struct ocfs2_alloc_context)); | |
1133 | goto retry_enospc; | |
1134 | } | |
ccd979bd | 1135 | if (status < 0) { |
9c7af40b | 1136 | mlog_errno(status); |
ccd979bd MF |
1137 | goto bail; |
1138 | } | |
1139 | ||
1140 | *bitmap_inode = (*ac)->ac_inode; | |
1141 | igrab(*bitmap_inode); | |
1142 | *bitmap_bh = (*ac)->ac_bh; | |
1143 | get_bh(*bitmap_bh); | |
1144 | status = 0; | |
1145 | bail: | |
1146 | if ((status < 0) && *ac) { | |
1147 | ocfs2_free_alloc_context(*ac); | |
1148 | *ac = NULL; | |
1149 | } | |
1150 | ||
c1e8d35e TM |
1151 | if (status) |
1152 | mlog_errno(status); | |
ccd979bd MF |
1153 | return status; |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * pass it the bitmap lock in lock_bh if you have it. | |
1158 | */ | |
1159 | static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb, | |
1fabe148 | 1160 | handle_t *handle, |
ccd979bd MF |
1161 | struct ocfs2_alloc_context *ac) |
1162 | { | |
1163 | int status = 0; | |
1164 | u32 cluster_off, cluster_count; | |
1165 | struct ocfs2_dinode *alloc = NULL; | |
1166 | struct ocfs2_local_alloc *la; | |
1167 | ||
ccd979bd MF |
1168 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; |
1169 | la = OCFS2_LOCAL_ALLOC(alloc); | |
1170 | ||
a04733d8 TM |
1171 | trace_ocfs2_local_alloc_new_window( |
1172 | le32_to_cpu(alloc->id1.bitmap1.i_total), | |
1173 | osb->local_alloc_bits); | |
883d4cae MF |
1174 | |
1175 | /* Instruct the allocation code to try the most recently used | |
1176 | * cluster group. We'll re-record the group used this pass | |
1177 | * below. */ | |
1178 | ac->ac_last_group = osb->la_last_gd; | |
1179 | ||
ccd979bd MF |
1180 | /* we used the generic suballoc reserve function, but we set |
1181 | * everything up nicely, so there's no reason why we can't use | |
1182 | * the more specific cluster api to claim bits. */ | |
1ed9b777 | 1183 | status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits, |
ccd979bd | 1184 | &cluster_off, &cluster_count); |
9c7af40b MF |
1185 | if (status == -ENOSPC) { |
1186 | retry_enospc: | |
1187 | /* | |
1188 | * Note: We could also try syncing the journal here to | |
1189 | * allow use of any free bits which the current | |
1190 | * transaction can't give us access to. --Mark | |
1191 | */ | |
1192 | if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) == | |
1193 | OCFS2_LA_DISABLED) | |
1194 | goto bail; | |
1195 | ||
7e9b7937 | 1196 | ac->ac_bits_wanted = osb->local_alloc_bits; |
1ed9b777 | 1197 | status = ocfs2_claim_clusters(handle, ac, |
9c7af40b MF |
1198 | osb->local_alloc_bits, |
1199 | &cluster_off, | |
1200 | &cluster_count); | |
1201 | if (status == -ENOSPC) | |
1202 | goto retry_enospc; | |
1203 | /* | |
1204 | * We only shrunk the *minimum* number of in our | |
1205 | * request - it's entirely possible that the allocator | |
1206 | * might give us more than we asked for. | |
1207 | */ | |
1208 | if (status == 0) { | |
1209 | spin_lock(&osb->osb_lock); | |
1210 | osb->local_alloc_bits = cluster_count; | |
1211 | spin_unlock(&osb->osb_lock); | |
1212 | } | |
1213 | } | |
ccd979bd MF |
1214 | if (status < 0) { |
1215 | if (status != -ENOSPC) | |
1216 | mlog_errno(status); | |
1217 | goto bail; | |
1218 | } | |
1219 | ||
883d4cae MF |
1220 | osb->la_last_gd = ac->ac_last_group; |
1221 | ||
ccd979bd MF |
1222 | la->la_bm_off = cpu_to_le32(cluster_off); |
1223 | alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count); | |
1224 | /* just in case... In the future when we find space ourselves, | |
1225 | * we don't have to get all contiguous -- but we'll have to | |
1226 | * set all previously used bits in bitmap and update | |
1227 | * la_bits_set before setting the bits in the main bitmap. */ | |
1228 | alloc->id1.bitmap1.i_used = 0; | |
1229 | memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0, | |
1230 | le16_to_cpu(la->la_size)); | |
1231 | ||
d02f00cc MF |
1232 | ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count, |
1233 | OCFS2_LOCAL_ALLOC(alloc)->la_bitmap); | |
1234 | ||
a04733d8 TM |
1235 | trace_ocfs2_local_alloc_new_window_result( |
1236 | OCFS2_LOCAL_ALLOC(alloc)->la_bm_off, | |
1237 | le32_to_cpu(alloc->id1.bitmap1.i_total)); | |
ccd979bd MF |
1238 | |
1239 | bail: | |
c1e8d35e TM |
1240 | if (status) |
1241 | mlog_errno(status); | |
ccd979bd MF |
1242 | return status; |
1243 | } | |
1244 | ||
1245 | /* Note that we do *NOT* lock the local alloc inode here as | |
1246 | * it's been locked already for us. */ | |
1247 | static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb, | |
1248 | struct inode *local_alloc_inode) | |
1249 | { | |
1250 | int status = 0; | |
1251 | struct buffer_head *main_bm_bh = NULL; | |
1252 | struct inode *main_bm_inode = NULL; | |
1fabe148 | 1253 | handle_t *handle = NULL; |
ccd979bd MF |
1254 | struct ocfs2_dinode *alloc; |
1255 | struct ocfs2_dinode *alloc_copy = NULL; | |
1256 | struct ocfs2_alloc_context *ac = NULL; | |
1257 | ||
9c7af40b MF |
1258 | ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE); |
1259 | ||
ccd979bd MF |
1260 | /* This will lock the main bitmap for us. */ |
1261 | status = ocfs2_local_alloc_reserve_for_window(osb, | |
ccd979bd MF |
1262 | &ac, |
1263 | &main_bm_inode, | |
1264 | &main_bm_bh); | |
1265 | if (status < 0) { | |
1266 | if (status != -ENOSPC) | |
1267 | mlog_errno(status); | |
1268 | goto bail; | |
1269 | } | |
1270 | ||
65eff9cc | 1271 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
1272 | if (IS_ERR(handle)) { |
1273 | status = PTR_ERR(handle); | |
1274 | handle = NULL; | |
1275 | mlog_errno(status); | |
1276 | goto bail; | |
1277 | } | |
1278 | ||
1279 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
1280 | ||
1281 | /* We want to clear the local alloc before doing anything | |
1282 | * else, so that if we error later during this operation, | |
1283 | * local alloc shutdown won't try to double free main bitmap | |
1284 | * bits. Make a copy so the sync function knows which bits to | |
1285 | * free. */ | |
4ba1c5bf | 1286 | alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS); |
ccd979bd MF |
1287 | if (!alloc_copy) { |
1288 | status = -ENOMEM; | |
1289 | mlog_errno(status); | |
1290 | goto bail; | |
1291 | } | |
1292 | memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size); | |
1293 | ||
0cf2f763 JB |
1294 | status = ocfs2_journal_access_di(handle, |
1295 | INODE_CACHE(local_alloc_inode), | |
13723d00 JB |
1296 | osb->local_alloc_bh, |
1297 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
1298 | if (status < 0) { |
1299 | mlog_errno(status); | |
1300 | goto bail; | |
1301 | } | |
1302 | ||
1303 | ocfs2_clear_local_alloc(alloc); | |
ec20cec7 | 1304 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); |
ccd979bd MF |
1305 | |
1306 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | |
1307 | main_bm_inode, main_bm_bh); | |
1308 | if (status < 0) { | |
1309 | mlog_errno(status); | |
1310 | goto bail; | |
1311 | } | |
1312 | ||
1313 | status = ocfs2_local_alloc_new_window(osb, handle, ac); | |
1314 | if (status < 0) { | |
1315 | if (status != -ENOSPC) | |
1316 | mlog_errno(status); | |
1317 | goto bail; | |
1318 | } | |
1319 | ||
1320 | atomic_inc(&osb->alloc_stats.moves); | |
1321 | ||
ccd979bd MF |
1322 | bail: |
1323 | if (handle) | |
02dc1af4 | 1324 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1325 | |
a81cb88b | 1326 | brelse(main_bm_bh); |
ccd979bd | 1327 | |
72865d92 | 1328 | iput(main_bm_inode); |
d787ab09 | 1329 | kfree(alloc_copy); |
ccd979bd MF |
1330 | |
1331 | if (ac) | |
1332 | ocfs2_free_alloc_context(ac); | |
1333 | ||
c1e8d35e TM |
1334 | if (status) |
1335 | mlog_errno(status); | |
ccd979bd MF |
1336 | return status; |
1337 | } | |
1338 |