Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
7 | * of the GNU General Public License v.2. | |
8 | */ | |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
f42faf4f | 15 | #include <linux/fs.h> |
5c676f6d | 16 | #include <linux/gfs2_ondisk.h> |
b3b94faa DT |
17 | |
18 | #include "gfs2.h" | |
5c676f6d SW |
19 | #include "lm_interface.h" |
20 | #include "incore.h" | |
b3b94faa DT |
21 | #include "glock.h" |
22 | #include "glops.h" | |
b3b94faa DT |
23 | #include "lops.h" |
24 | #include "meta_io.h" | |
25 | #include "quota.h" | |
26 | #include "rgrp.h" | |
27 | #include "super.h" | |
28 | #include "trans.h" | |
f42faf4f | 29 | #include "ops_file.h" |
5c676f6d | 30 | #include "util.h" |
b3b94faa | 31 | |
88c8ab1f SW |
32 | #define BFITNOENT 0xFFFFFFFF |
33 | ||
34 | /* | |
35 | * These routines are used by the resource group routines (rgrp.c) | |
36 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
37 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
38 | * | |
39 | * 0 = Free | |
40 | * 1 = Used (not metadata) | |
41 | * 2 = Unlinked (still in use) inode | |
42 | * 3 = Used (metadata) | |
88c8ab1f SW |
43 | */ |
44 | ||
45 | static const char valid_change[16] = { | |
46 | /* current */ | |
feaa7bba | 47 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 48 | /* e */ 1, 0, 0, 0, |
feaa7bba | 49 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
50 | 1, 0, 0, 0 |
51 | }; | |
52 | ||
53 | /** | |
54 | * gfs2_setbit - Set a bit in the bitmaps | |
55 | * @buffer: the buffer that holds the bitmaps | |
56 | * @buflen: the length (in bytes) of the buffer | |
57 | * @block: the block to set | |
58 | * @new_state: the new state of the block | |
59 | * | |
60 | */ | |
61 | ||
3efd7534 SW |
62 | static void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buffer, |
63 | unsigned int buflen, uint32_t block, | |
64 | unsigned char new_state) | |
88c8ab1f SW |
65 | { |
66 | unsigned char *byte, *end, cur_state; | |
67 | unsigned int bit; | |
68 | ||
69 | byte = buffer + (block / GFS2_NBBY); | |
70 | bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; | |
71 | end = buffer + buflen; | |
72 | ||
73 | gfs2_assert(rgd->rd_sbd, byte < end); | |
74 | ||
75 | cur_state = (*byte >> bit) & GFS2_BIT_MASK; | |
76 | ||
77 | if (valid_change[new_state * 4 + cur_state]) { | |
78 | *byte ^= cur_state << bit; | |
79 | *byte |= new_state << bit; | |
80 | } else | |
81 | gfs2_consist_rgrpd(rgd); | |
82 | } | |
83 | ||
84 | /** | |
85 | * gfs2_testbit - test a bit in the bitmaps | |
86 | * @buffer: the buffer that holds the bitmaps | |
87 | * @buflen: the length (in bytes) of the buffer | |
88 | * @block: the block to read | |
89 | * | |
90 | */ | |
91 | ||
3efd7534 SW |
92 | static unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd, unsigned char *buffer, |
93 | unsigned int buflen, uint32_t block) | |
88c8ab1f SW |
94 | { |
95 | unsigned char *byte, *end, cur_state; | |
96 | unsigned int bit; | |
97 | ||
98 | byte = buffer + (block / GFS2_NBBY); | |
99 | bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; | |
100 | end = buffer + buflen; | |
101 | ||
102 | gfs2_assert(rgd->rd_sbd, byte < end); | |
103 | ||
104 | cur_state = (*byte >> bit) & GFS2_BIT_MASK; | |
105 | ||
106 | return cur_state; | |
107 | } | |
108 | ||
109 | /** | |
110 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
111 | * a block in a given allocation state. | |
112 | * @buffer: the buffer that holds the bitmaps | |
113 | * @buflen: the length (in bytes) of the buffer | |
114 | * @goal: start search at this block's bit-pair (within @buffer) | |
115 | * @old_state: GFS2_BLKST_XXX the state of the block we're looking for; | |
116 | * bit 0 = alloc(1)/free(0), bit 1 = meta(1)/data(0) | |
117 | * | |
118 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
119 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
120 | * beginning of a bitmap block buffer, skipping any header structures. | |
121 | * | |
122 | * Return: the block number (bitmap buffer scope) that was found | |
123 | */ | |
124 | ||
3efd7534 SW |
125 | static uint32_t gfs2_bitfit(struct gfs2_rgrpd *rgd, unsigned char *buffer, |
126 | unsigned int buflen, uint32_t goal, | |
127 | unsigned char old_state) | |
88c8ab1f SW |
128 | { |
129 | unsigned char *byte, *end, alloc; | |
130 | uint32_t blk = goal; | |
131 | unsigned int bit; | |
132 | ||
133 | byte = buffer + (goal / GFS2_NBBY); | |
134 | bit = (goal % GFS2_NBBY) * GFS2_BIT_SIZE; | |
135 | end = buffer + buflen; | |
136 | alloc = (old_state & 1) ? 0 : 0x55; | |
137 | ||
138 | while (byte < end) { | |
139 | if ((*byte & 0x55) == alloc) { | |
140 | blk += (8 - bit) >> 1; | |
141 | ||
142 | bit = 0; | |
143 | byte++; | |
144 | ||
145 | continue; | |
146 | } | |
147 | ||
148 | if (((*byte >> bit) & GFS2_BIT_MASK) == old_state) | |
149 | return blk; | |
150 | ||
151 | bit += GFS2_BIT_SIZE; | |
152 | if (bit >= 8) { | |
153 | bit = 0; | |
154 | byte++; | |
155 | } | |
156 | ||
157 | blk++; | |
158 | } | |
159 | ||
160 | return BFITNOENT; | |
161 | } | |
162 | ||
163 | /** | |
164 | * gfs2_bitcount - count the number of bits in a certain state | |
165 | * @buffer: the buffer that holds the bitmaps | |
166 | * @buflen: the length (in bytes) of the buffer | |
167 | * @state: the state of the block we're looking for | |
168 | * | |
169 | * Returns: The number of bits | |
170 | */ | |
171 | ||
3efd7534 SW |
172 | static uint32_t gfs2_bitcount(struct gfs2_rgrpd *rgd, unsigned char *buffer, |
173 | unsigned int buflen, unsigned char state) | |
88c8ab1f SW |
174 | { |
175 | unsigned char *byte = buffer; | |
176 | unsigned char *end = buffer + buflen; | |
177 | unsigned char state1 = state << 2; | |
178 | unsigned char state2 = state << 4; | |
179 | unsigned char state3 = state << 6; | |
180 | uint32_t count = 0; | |
181 | ||
182 | for (; byte < end; byte++) { | |
183 | if (((*byte) & 0x03) == state) | |
184 | count++; | |
185 | if (((*byte) & 0x0C) == state1) | |
186 | count++; | |
187 | if (((*byte) & 0x30) == state2) | |
188 | count++; | |
189 | if (((*byte) & 0xC0) == state3) | |
190 | count++; | |
191 | } | |
192 | ||
193 | return count; | |
194 | } | |
195 | ||
b3b94faa DT |
196 | /** |
197 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
198 | * @sdp: the filesystem | |
199 | * @rgd: the rgrp | |
200 | * | |
201 | */ | |
202 | ||
203 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
204 | { | |
205 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
206 | struct gfs2_bitmap *bi = NULL; | |
207 | uint32_t length = rgd->rd_ri.ri_length; | |
208 | uint32_t count[4], tmp; | |
209 | int buf, x; | |
210 | ||
211 | memset(count, 0, 4 * sizeof(uint32_t)); | |
212 | ||
213 | /* Count # blocks in each of 4 possible allocation states */ | |
214 | for (buf = 0; buf < length; buf++) { | |
215 | bi = rgd->rd_bits + buf; | |
216 | for (x = 0; x < 4; x++) | |
217 | count[x] += gfs2_bitcount(rgd, | |
218 | bi->bi_bh->b_data + | |
219 | bi->bi_offset, | |
220 | bi->bi_len, x); | |
221 | } | |
222 | ||
223 | if (count[0] != rgd->rd_rg.rg_free) { | |
224 | if (gfs2_consist_rgrpd(rgd)) | |
225 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
226 | count[0], rgd->rd_rg.rg_free); | |
227 | return; | |
228 | } | |
229 | ||
230 | tmp = rgd->rd_ri.ri_data - | |
231 | rgd->rd_rg.rg_free - | |
232 | rgd->rd_rg.rg_dinodes; | |
feaa7bba | 233 | if (count[1] + count[2] != tmp) { |
b3b94faa DT |
234 | if (gfs2_consist_rgrpd(rgd)) |
235 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
236 | count[1], tmp); | |
237 | return; | |
238 | } | |
239 | ||
feaa7bba | 240 | if (count[3] != rgd->rd_rg.rg_dinodes) { |
b3b94faa | 241 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba SW |
242 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
243 | count[3], rgd->rd_rg.rg_dinodes); | |
b3b94faa DT |
244 | return; |
245 | } | |
246 | ||
feaa7bba | 247 | if (count[2] > count[3]) { |
b3b94faa | 248 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba SW |
249 | fs_err(sdp, "unlinked inodes > inodes: %u\n", |
250 | count[2]); | |
b3b94faa DT |
251 | return; |
252 | } | |
feaa7bba | 253 | |
b3b94faa DT |
254 | } |
255 | ||
256 | static inline int rgrp_contains_block(struct gfs2_rindex *ri, uint64_t block) | |
257 | { | |
258 | uint64_t first = ri->ri_data0; | |
259 | uint64_t last = first + ri->ri_data; | |
260 | return !!(first <= block && block < last); | |
261 | } | |
262 | ||
263 | /** | |
264 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
265 | * @sdp: The GFS2 superblock | |
266 | * @n: The data block number | |
267 | * | |
268 | * Returns: The resource group, or NULL if not found | |
269 | */ | |
270 | ||
271 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, uint64_t blk) | |
272 | { | |
273 | struct gfs2_rgrpd *rgd; | |
274 | ||
275 | spin_lock(&sdp->sd_rindex_spin); | |
276 | ||
277 | list_for_each_entry(rgd, &sdp->sd_rindex_mru_list, rd_list_mru) { | |
278 | if (rgrp_contains_block(&rgd->rd_ri, blk)) { | |
279 | list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list); | |
280 | spin_unlock(&sdp->sd_rindex_spin); | |
281 | return rgd; | |
282 | } | |
283 | } | |
284 | ||
285 | spin_unlock(&sdp->sd_rindex_spin); | |
286 | ||
287 | return NULL; | |
288 | } | |
289 | ||
290 | /** | |
291 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
292 | * @sdp: The GFS2 superblock | |
293 | * | |
294 | * Returns: The first rgrp in the filesystem | |
295 | */ | |
296 | ||
297 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
298 | { | |
299 | gfs2_assert(sdp, !list_empty(&sdp->sd_rindex_list)); | |
300 | return list_entry(sdp->sd_rindex_list.next, struct gfs2_rgrpd, rd_list); | |
301 | } | |
302 | ||
303 | /** | |
304 | * gfs2_rgrpd_get_next - get the next RG | |
305 | * @rgd: A RG | |
306 | * | |
307 | * Returns: The next rgrp | |
308 | */ | |
309 | ||
310 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
311 | { | |
312 | if (rgd->rd_list.next == &rgd->rd_sbd->sd_rindex_list) | |
313 | return NULL; | |
314 | return list_entry(rgd->rd_list.next, struct gfs2_rgrpd, rd_list); | |
315 | } | |
316 | ||
317 | static void clear_rgrpdi(struct gfs2_sbd *sdp) | |
318 | { | |
319 | struct list_head *head; | |
320 | struct gfs2_rgrpd *rgd; | |
321 | struct gfs2_glock *gl; | |
322 | ||
323 | spin_lock(&sdp->sd_rindex_spin); | |
324 | sdp->sd_rindex_forward = NULL; | |
325 | head = &sdp->sd_rindex_recent_list; | |
326 | while (!list_empty(head)) { | |
327 | rgd = list_entry(head->next, struct gfs2_rgrpd, rd_recent); | |
328 | list_del(&rgd->rd_recent); | |
329 | } | |
330 | spin_unlock(&sdp->sd_rindex_spin); | |
331 | ||
332 | head = &sdp->sd_rindex_list; | |
333 | while (!list_empty(head)) { | |
334 | rgd = list_entry(head->next, struct gfs2_rgrpd, rd_list); | |
335 | gl = rgd->rd_gl; | |
336 | ||
337 | list_del(&rgd->rd_list); | |
338 | list_del(&rgd->rd_list_mru); | |
339 | ||
340 | if (gl) { | |
5c676f6d | 341 | gl->gl_object = NULL; |
b3b94faa DT |
342 | gfs2_glock_put(gl); |
343 | } | |
344 | ||
345 | kfree(rgd->rd_bits); | |
346 | kfree(rgd); | |
347 | } | |
348 | } | |
349 | ||
350 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) | |
351 | { | |
f55ab26a | 352 | mutex_lock(&sdp->sd_rindex_mutex); |
b3b94faa | 353 | clear_rgrpdi(sdp); |
f55ab26a | 354 | mutex_unlock(&sdp->sd_rindex_mutex); |
b3b94faa DT |
355 | } |
356 | ||
357 | /** | |
358 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
359 | * @rgd: The resource group descriptor | |
360 | * | |
361 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
362 | * | |
363 | * Returns: errno | |
364 | */ | |
365 | ||
366 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
367 | { | |
368 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
369 | struct gfs2_bitmap *bi; | |
370 | uint32_t length = rgd->rd_ri.ri_length; /* # blocks in hdr & bitmap */ | |
371 | uint32_t bytes_left, bytes; | |
372 | int x; | |
373 | ||
feaa7bba SW |
374 | if (!length) |
375 | return -EINVAL; | |
376 | ||
b3b94faa DT |
377 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_KERNEL); |
378 | if (!rgd->rd_bits) | |
379 | return -ENOMEM; | |
380 | ||
381 | bytes_left = rgd->rd_ri.ri_bitbytes; | |
382 | ||
383 | for (x = 0; x < length; x++) { | |
384 | bi = rgd->rd_bits + x; | |
385 | ||
386 | /* small rgrp; bitmap stored completely in header block */ | |
387 | if (length == 1) { | |
388 | bytes = bytes_left; | |
389 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
390 | bi->bi_start = 0; | |
391 | bi->bi_len = bytes; | |
392 | /* header block */ | |
393 | } else if (x == 0) { | |
394 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
395 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
396 | bi->bi_start = 0; | |
397 | bi->bi_len = bytes; | |
398 | /* last block */ | |
399 | } else if (x + 1 == length) { | |
400 | bytes = bytes_left; | |
401 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
402 | bi->bi_start = rgd->rd_ri.ri_bitbytes - bytes_left; | |
403 | bi->bi_len = bytes; | |
404 | /* other blocks */ | |
405 | } else { | |
568f4c96 SW |
406 | bytes = sdp->sd_sb.sb_bsize - |
407 | sizeof(struct gfs2_meta_header); | |
b3b94faa DT |
408 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
409 | bi->bi_start = rgd->rd_ri.ri_bitbytes - bytes_left; | |
410 | bi->bi_len = bytes; | |
411 | } | |
412 | ||
413 | bytes_left -= bytes; | |
414 | } | |
415 | ||
416 | if (bytes_left) { | |
417 | gfs2_consist_rgrpd(rgd); | |
418 | return -EIO; | |
419 | } | |
420 | bi = rgd->rd_bits + (length - 1); | |
421 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_ri.ri_data) { | |
422 | if (gfs2_consist_rgrpd(rgd)) { | |
423 | gfs2_rindex_print(&rgd->rd_ri); | |
424 | fs_err(sdp, "start=%u len=%u offset=%u\n", | |
425 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
426 | } | |
427 | return -EIO; | |
428 | } | |
429 | ||
430 | return 0; | |
431 | } | |
432 | ||
433 | /** | |
434 | * gfs2_ri_update - Pull in a new resource index from the disk | |
435 | * @gl: The glock covering the rindex inode | |
436 | * | |
437 | * Returns: 0 on successful update, error code otherwise | |
438 | */ | |
439 | ||
440 | static int gfs2_ri_update(struct gfs2_inode *ip) | |
441 | { | |
feaa7bba SW |
442 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
443 | struct inode *inode = &ip->i_inode; | |
b3b94faa DT |
444 | struct gfs2_rgrpd *rgd; |
445 | char buf[sizeof(struct gfs2_rindex)]; | |
f42faf4f | 446 | struct file_ra_state ra_state; |
b3b94faa DT |
447 | uint64_t junk = ip->i_di.di_size; |
448 | int error; | |
449 | ||
450 | if (do_div(junk, sizeof(struct gfs2_rindex))) { | |
451 | gfs2_consist_inode(ip); | |
452 | return -EIO; | |
453 | } | |
454 | ||
455 | clear_rgrpdi(sdp); | |
456 | ||
f42faf4f | 457 | file_ra_state_init(&ra_state, inode->i_mapping); |
b3b94faa | 458 | for (sdp->sd_rgrps = 0;; sdp->sd_rgrps++) { |
f42faf4f SW |
459 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); |
460 | error = gfs2_internal_read(ip, &ra_state, buf, &pos, | |
b3b94faa DT |
461 | sizeof(struct gfs2_rindex)); |
462 | if (!error) | |
463 | break; | |
464 | if (error != sizeof(struct gfs2_rindex)) { | |
465 | if (error > 0) | |
466 | error = -EIO; | |
467 | goto fail; | |
468 | } | |
469 | ||
470 | rgd = kzalloc(sizeof(struct gfs2_rgrpd), GFP_KERNEL); | |
471 | error = -ENOMEM; | |
472 | if (!rgd) | |
473 | goto fail; | |
474 | ||
f55ab26a | 475 | mutex_init(&rgd->rd_mutex); |
b3b94faa DT |
476 | lops_init_le(&rgd->rd_le, &gfs2_rg_lops); |
477 | rgd->rd_sbd = sdp; | |
478 | ||
479 | list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list); | |
480 | list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list); | |
481 | ||
482 | gfs2_rindex_in(&rgd->rd_ri, buf); | |
b3b94faa DT |
483 | error = compute_bitstructs(rgd); |
484 | if (error) | |
485 | goto fail; | |
486 | ||
487 | error = gfs2_glock_get(sdp, rgd->rd_ri.ri_addr, | |
488 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); | |
489 | if (error) | |
490 | goto fail; | |
491 | ||
5c676f6d | 492 | rgd->rd_gl->gl_object = rgd; |
b3b94faa DT |
493 | rgd->rd_rg_vn = rgd->rd_gl->gl_vn - 1; |
494 | } | |
495 | ||
496 | sdp->sd_rindex_vn = ip->i_gl->gl_vn; | |
b3b94faa DT |
497 | return 0; |
498 | ||
feaa7bba | 499 | fail: |
b3b94faa | 500 | clear_rgrpdi(sdp); |
b3b94faa DT |
501 | return error; |
502 | } | |
503 | ||
504 | /** | |
505 | * gfs2_rindex_hold - Grab a lock on the rindex | |
506 | * @sdp: The GFS2 superblock | |
507 | * @ri_gh: the glock holder | |
508 | * | |
509 | * We grab a lock on the rindex inode to make sure that it doesn't | |
510 | * change whilst we are performing an operation. We keep this lock | |
511 | * for quite long periods of time compared to other locks. This | |
512 | * doesn't matter, since it is shared and it is very, very rarely | |
513 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
514 | * | |
515 | * This makes sure that we're using the latest copy of the resource index | |
516 | * special file, which might have been updated if someone expanded the | |
517 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
518 | * | |
519 | * Returns: 0 on success, error code otherwise | |
520 | */ | |
521 | ||
522 | int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh) | |
523 | { | |
feaa7bba | 524 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa DT |
525 | struct gfs2_glock *gl = ip->i_gl; |
526 | int error; | |
527 | ||
528 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh); | |
529 | if (error) | |
530 | return error; | |
531 | ||
532 | /* Read new copy from disk if we don't have the latest */ | |
533 | if (sdp->sd_rindex_vn != gl->gl_vn) { | |
f55ab26a | 534 | mutex_lock(&sdp->sd_rindex_mutex); |
b3b94faa DT |
535 | if (sdp->sd_rindex_vn != gl->gl_vn) { |
536 | error = gfs2_ri_update(ip); | |
537 | if (error) | |
538 | gfs2_glock_dq_uninit(ri_gh); | |
539 | } | |
f55ab26a | 540 | mutex_unlock(&sdp->sd_rindex_mutex); |
b3b94faa DT |
541 | } |
542 | ||
543 | return error; | |
544 | } | |
545 | ||
546 | /** | |
547 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps | |
548 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
549 | * | |
550 | * Read in all of a Resource Group's header and bitmap blocks. | |
551 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
552 | * | |
553 | * Returns: errno | |
554 | */ | |
555 | ||
556 | int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) | |
557 | { | |
558 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
559 | struct gfs2_glock *gl = rgd->rd_gl; | |
560 | unsigned int length = rgd->rd_ri.ri_length; | |
561 | struct gfs2_bitmap *bi; | |
562 | unsigned int x, y; | |
563 | int error; | |
564 | ||
f55ab26a | 565 | mutex_lock(&rgd->rd_mutex); |
b3b94faa DT |
566 | |
567 | spin_lock(&sdp->sd_rindex_spin); | |
568 | if (rgd->rd_bh_count) { | |
569 | rgd->rd_bh_count++; | |
570 | spin_unlock(&sdp->sd_rindex_spin); | |
f55ab26a | 571 | mutex_unlock(&rgd->rd_mutex); |
b3b94faa DT |
572 | return 0; |
573 | } | |
574 | spin_unlock(&sdp->sd_rindex_spin); | |
575 | ||
576 | for (x = 0; x < length; x++) { | |
577 | bi = rgd->rd_bits + x; | |
578 | error = gfs2_meta_read(gl, rgd->rd_ri.ri_addr + x, DIO_START, | |
579 | &bi->bi_bh); | |
580 | if (error) | |
581 | goto fail; | |
582 | } | |
583 | ||
584 | for (y = length; y--;) { | |
585 | bi = rgd->rd_bits + y; | |
586 | error = gfs2_meta_reread(sdp, bi->bi_bh, DIO_WAIT); | |
587 | if (error) | |
588 | goto fail; | |
feaa7bba | 589 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
590 | GFS2_METATYPE_RG)) { |
591 | error = -EIO; | |
592 | goto fail; | |
593 | } | |
594 | } | |
595 | ||
596 | if (rgd->rd_rg_vn != gl->gl_vn) { | |
597 | gfs2_rgrp_in(&rgd->rd_rg, (rgd->rd_bits[0].bi_bh)->b_data); | |
598 | rgd->rd_rg_vn = gl->gl_vn; | |
599 | } | |
600 | ||
601 | spin_lock(&sdp->sd_rindex_spin); | |
602 | rgd->rd_free_clone = rgd->rd_rg.rg_free; | |
603 | rgd->rd_bh_count++; | |
604 | spin_unlock(&sdp->sd_rindex_spin); | |
605 | ||
f55ab26a | 606 | mutex_unlock(&rgd->rd_mutex); |
b3b94faa DT |
607 | |
608 | return 0; | |
609 | ||
feaa7bba | 610 | fail: |
b3b94faa DT |
611 | while (x--) { |
612 | bi = rgd->rd_bits + x; | |
613 | brelse(bi->bi_bh); | |
614 | bi->bi_bh = NULL; | |
615 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
616 | } | |
f55ab26a | 617 | mutex_unlock(&rgd->rd_mutex); |
b3b94faa DT |
618 | |
619 | return error; | |
620 | } | |
621 | ||
622 | void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd) | |
623 | { | |
624 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
625 | ||
626 | spin_lock(&sdp->sd_rindex_spin); | |
627 | gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count); | |
628 | rgd->rd_bh_count++; | |
629 | spin_unlock(&sdp->sd_rindex_spin); | |
630 | } | |
631 | ||
632 | /** | |
633 | * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get() | |
634 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
635 | * | |
636 | */ | |
637 | ||
638 | void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd) | |
639 | { | |
640 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
641 | int x, length = rgd->rd_ri.ri_length; | |
642 | ||
643 | spin_lock(&sdp->sd_rindex_spin); | |
644 | gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count); | |
645 | if (--rgd->rd_bh_count) { | |
646 | spin_unlock(&sdp->sd_rindex_spin); | |
647 | return; | |
648 | } | |
649 | ||
650 | for (x = 0; x < length; x++) { | |
651 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
652 | kfree(bi->bi_clone); | |
653 | bi->bi_clone = NULL; | |
654 | brelse(bi->bi_bh); | |
655 | bi->bi_bh = NULL; | |
656 | } | |
657 | ||
658 | spin_unlock(&sdp->sd_rindex_spin); | |
659 | } | |
660 | ||
661 | void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd) | |
662 | { | |
663 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
664 | unsigned int length = rgd->rd_ri.ri_length; | |
665 | unsigned int x; | |
666 | ||
667 | for (x = 0; x < length; x++) { | |
668 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
669 | if (!bi->bi_clone) | |
670 | continue; | |
671 | memcpy(bi->bi_clone + bi->bi_offset, | |
feaa7bba | 672 | bi->bi_bh->b_data + bi->bi_offset, bi->bi_len); |
b3b94faa DT |
673 | } |
674 | ||
675 | spin_lock(&sdp->sd_rindex_spin); | |
676 | rgd->rd_free_clone = rgd->rd_rg.rg_free; | |
677 | spin_unlock(&sdp->sd_rindex_spin); | |
678 | } | |
679 | ||
680 | /** | |
681 | * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode | |
682 | * @ip: the incore GFS2 inode structure | |
683 | * | |
684 | * Returns: the struct gfs2_alloc | |
685 | */ | |
686 | ||
687 | struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip) | |
688 | { | |
689 | struct gfs2_alloc *al = &ip->i_alloc; | |
690 | ||
691 | /* FIXME: Should assert that the correct locks are held here... */ | |
692 | memset(al, 0, sizeof(*al)); | |
693 | return al; | |
694 | } | |
695 | ||
696 | /** | |
697 | * gfs2_alloc_put - throw away the struct gfs2_alloc for an inode | |
698 | * @ip: the inode | |
699 | * | |
700 | */ | |
701 | ||
702 | void gfs2_alloc_put(struct gfs2_inode *ip) | |
703 | { | |
704 | return; | |
705 | } | |
706 | ||
707 | /** | |
708 | * try_rgrp_fit - See if a given reservation will fit in a given RG | |
709 | * @rgd: the RG data | |
710 | * @al: the struct gfs2_alloc structure describing the reservation | |
711 | * | |
712 | * If there's room for the requested blocks to be allocated from the RG: | |
713 | * Sets the $al_reserved_data field in @al. | |
714 | * Sets the $al_reserved_meta field in @al. | |
715 | * Sets the $al_rgd field in @al. | |
716 | * | |
717 | * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) | |
718 | */ | |
719 | ||
720 | static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al) | |
721 | { | |
722 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
723 | int ret = 0; | |
724 | ||
725 | spin_lock(&sdp->sd_rindex_spin); | |
726 | if (rgd->rd_free_clone >= al->al_requested) { | |
727 | al->al_rgd = rgd; | |
728 | ret = 1; | |
729 | } | |
730 | spin_unlock(&sdp->sd_rindex_spin); | |
731 | ||
732 | return ret; | |
733 | } | |
734 | ||
735 | /** | |
736 | * recent_rgrp_first - get first RG from "recent" list | |
737 | * @sdp: The GFS2 superblock | |
738 | * @rglast: address of the rgrp used last | |
739 | * | |
740 | * Returns: The first rgrp in the recent list | |
741 | */ | |
742 | ||
743 | static struct gfs2_rgrpd *recent_rgrp_first(struct gfs2_sbd *sdp, | |
744 | uint64_t rglast) | |
745 | { | |
746 | struct gfs2_rgrpd *rgd = NULL; | |
747 | ||
748 | spin_lock(&sdp->sd_rindex_spin); | |
749 | ||
750 | if (list_empty(&sdp->sd_rindex_recent_list)) | |
751 | goto out; | |
752 | ||
753 | if (!rglast) | |
754 | goto first; | |
755 | ||
756 | list_for_each_entry(rgd, &sdp->sd_rindex_recent_list, rd_recent) { | |
757 | if (rgd->rd_ri.ri_addr == rglast) | |
758 | goto out; | |
759 | } | |
760 | ||
feaa7bba | 761 | first: |
b3b94faa DT |
762 | rgd = list_entry(sdp->sd_rindex_recent_list.next, struct gfs2_rgrpd, |
763 | rd_recent); | |
feaa7bba | 764 | out: |
b3b94faa | 765 | spin_unlock(&sdp->sd_rindex_spin); |
b3b94faa DT |
766 | return rgd; |
767 | } | |
768 | ||
769 | /** | |
770 | * recent_rgrp_next - get next RG from "recent" list | |
771 | * @cur_rgd: current rgrp | |
772 | * @remove: | |
773 | * | |
774 | * Returns: The next rgrp in the recent list | |
775 | */ | |
776 | ||
777 | static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd, | |
778 | int remove) | |
779 | { | |
780 | struct gfs2_sbd *sdp = cur_rgd->rd_sbd; | |
781 | struct list_head *head; | |
782 | struct gfs2_rgrpd *rgd; | |
783 | ||
784 | spin_lock(&sdp->sd_rindex_spin); | |
785 | ||
786 | head = &sdp->sd_rindex_recent_list; | |
787 | ||
788 | list_for_each_entry(rgd, head, rd_recent) { | |
789 | if (rgd == cur_rgd) { | |
790 | if (cur_rgd->rd_recent.next != head) | |
791 | rgd = list_entry(cur_rgd->rd_recent.next, | |
792 | struct gfs2_rgrpd, rd_recent); | |
793 | else | |
794 | rgd = NULL; | |
795 | ||
796 | if (remove) | |
797 | list_del(&cur_rgd->rd_recent); | |
798 | ||
799 | goto out; | |
800 | } | |
801 | } | |
802 | ||
803 | rgd = NULL; | |
804 | if (!list_empty(head)) | |
805 | rgd = list_entry(head->next, struct gfs2_rgrpd, rd_recent); | |
806 | ||
feaa7bba | 807 | out: |
b3b94faa | 808 | spin_unlock(&sdp->sd_rindex_spin); |
b3b94faa DT |
809 | return rgd; |
810 | } | |
811 | ||
812 | /** | |
813 | * recent_rgrp_add - add an RG to tail of "recent" list | |
814 | * @new_rgd: The rgrp to add | |
815 | * | |
816 | */ | |
817 | ||
818 | static void recent_rgrp_add(struct gfs2_rgrpd *new_rgd) | |
819 | { | |
820 | struct gfs2_sbd *sdp = new_rgd->rd_sbd; | |
821 | struct gfs2_rgrpd *rgd; | |
822 | unsigned int count = 0; | |
823 | unsigned int max = sdp->sd_rgrps / gfs2_jindex_size(sdp); | |
824 | ||
825 | spin_lock(&sdp->sd_rindex_spin); | |
826 | ||
827 | list_for_each_entry(rgd, &sdp->sd_rindex_recent_list, rd_recent) { | |
828 | if (rgd == new_rgd) | |
829 | goto out; | |
830 | ||
831 | if (++count >= max) | |
832 | goto out; | |
833 | } | |
834 | list_add_tail(&new_rgd->rd_recent, &sdp->sd_rindex_recent_list); | |
835 | ||
feaa7bba | 836 | out: |
b3b94faa DT |
837 | spin_unlock(&sdp->sd_rindex_spin); |
838 | } | |
839 | ||
840 | /** | |
841 | * forward_rgrp_get - get an rgrp to try next from full list | |
842 | * @sdp: The GFS2 superblock | |
843 | * | |
844 | * Returns: The rgrp to try next | |
845 | */ | |
846 | ||
847 | static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp) | |
848 | { | |
849 | struct gfs2_rgrpd *rgd; | |
850 | unsigned int journals = gfs2_jindex_size(sdp); | |
851 | unsigned int rg = 0, x; | |
852 | ||
853 | spin_lock(&sdp->sd_rindex_spin); | |
854 | ||
855 | rgd = sdp->sd_rindex_forward; | |
856 | if (!rgd) { | |
857 | if (sdp->sd_rgrps >= journals) | |
858 | rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals; | |
859 | ||
860 | for (x = 0, rgd = gfs2_rgrpd_get_first(sdp); | |
861 | x < rg; | |
862 | x++, rgd = gfs2_rgrpd_get_next(rgd)) | |
863 | /* Do Nothing */; | |
864 | ||
865 | sdp->sd_rindex_forward = rgd; | |
866 | } | |
867 | ||
868 | spin_unlock(&sdp->sd_rindex_spin); | |
869 | ||
870 | return rgd; | |
871 | } | |
872 | ||
873 | /** | |
874 | * forward_rgrp_set - set the forward rgrp pointer | |
875 | * @sdp: the filesystem | |
876 | * @rgd: The new forward rgrp | |
877 | * | |
878 | */ | |
879 | ||
880 | static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd) | |
881 | { | |
882 | spin_lock(&sdp->sd_rindex_spin); | |
883 | sdp->sd_rindex_forward = rgd; | |
884 | spin_unlock(&sdp->sd_rindex_spin); | |
885 | } | |
886 | ||
887 | /** | |
888 | * get_local_rgrp - Choose and lock a rgrp for allocation | |
889 | * @ip: the inode to reserve space for | |
890 | * @rgp: the chosen and locked rgrp | |
891 | * | |
892 | * Try to acquire rgrp in way which avoids contending with others. | |
893 | * | |
894 | * Returns: errno | |
895 | */ | |
896 | ||
897 | static int get_local_rgrp(struct gfs2_inode *ip) | |
898 | { | |
feaa7bba | 899 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
900 | struct gfs2_rgrpd *rgd, *begin = NULL; |
901 | struct gfs2_alloc *al = &ip->i_alloc; | |
902 | int flags = LM_FLAG_TRY; | |
903 | int skipped = 0; | |
904 | int loops = 0; | |
905 | int error; | |
906 | ||
907 | /* Try recently successful rgrps */ | |
908 | ||
909 | rgd = recent_rgrp_first(sdp, ip->i_last_rg_alloc); | |
910 | ||
911 | while (rgd) { | |
912 | error = gfs2_glock_nq_init(rgd->rd_gl, | |
913 | LM_ST_EXCLUSIVE, LM_FLAG_TRY, | |
914 | &al->al_rgd_gh); | |
915 | switch (error) { | |
916 | case 0: | |
917 | if (try_rgrp_fit(rgd, al)) | |
918 | goto out; | |
919 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
920 | rgd = recent_rgrp_next(rgd, 1); | |
921 | break; | |
922 | ||
923 | case GLR_TRYFAILED: | |
924 | rgd = recent_rgrp_next(rgd, 0); | |
925 | break; | |
926 | ||
927 | default: | |
928 | return error; | |
929 | } | |
930 | } | |
931 | ||
932 | /* Go through full list of rgrps */ | |
933 | ||
934 | begin = rgd = forward_rgrp_get(sdp); | |
935 | ||
936 | for (;;) { | |
937 | error = gfs2_glock_nq_init(rgd->rd_gl, | |
938 | LM_ST_EXCLUSIVE, flags, | |
939 | &al->al_rgd_gh); | |
940 | switch (error) { | |
941 | case 0: | |
942 | if (try_rgrp_fit(rgd, al)) | |
943 | goto out; | |
944 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
945 | break; | |
946 | ||
947 | case GLR_TRYFAILED: | |
948 | skipped++; | |
949 | break; | |
950 | ||
951 | default: | |
952 | return error; | |
953 | } | |
954 | ||
955 | rgd = gfs2_rgrpd_get_next(rgd); | |
956 | if (!rgd) | |
957 | rgd = gfs2_rgrpd_get_first(sdp); | |
958 | ||
959 | if (rgd == begin) { | |
960 | if (++loops >= 2 || !skipped) | |
961 | return -ENOSPC; | |
962 | flags = 0; | |
963 | } | |
964 | } | |
965 | ||
feaa7bba | 966 | out: |
b3b94faa DT |
967 | ip->i_last_rg_alloc = rgd->rd_ri.ri_addr; |
968 | ||
969 | if (begin) { | |
970 | recent_rgrp_add(rgd); | |
971 | rgd = gfs2_rgrpd_get_next(rgd); | |
972 | if (!rgd) | |
973 | rgd = gfs2_rgrpd_get_first(sdp); | |
974 | forward_rgrp_set(sdp, rgd); | |
975 | } | |
976 | ||
977 | return 0; | |
978 | } | |
979 | ||
980 | /** | |
981 | * gfs2_inplace_reserve_i - Reserve space in the filesystem | |
982 | * @ip: the inode to reserve space for | |
983 | * | |
984 | * Returns: errno | |
985 | */ | |
986 | ||
987 | int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line) | |
988 | { | |
feaa7bba | 989 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
990 | struct gfs2_alloc *al = &ip->i_alloc; |
991 | int error; | |
992 | ||
993 | if (gfs2_assert_warn(sdp, al->al_requested)) | |
994 | return -EINVAL; | |
995 | ||
996 | error = gfs2_rindex_hold(sdp, &al->al_ri_gh); | |
997 | if (error) | |
998 | return error; | |
999 | ||
1000 | error = get_local_rgrp(ip); | |
1001 | if (error) { | |
1002 | gfs2_glock_dq_uninit(&al->al_ri_gh); | |
1003 | return error; | |
1004 | } | |
1005 | ||
1006 | al->al_file = file; | |
1007 | al->al_line = line; | |
1008 | ||
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | /** | |
1013 | * gfs2_inplace_release - release an inplace reservation | |
1014 | * @ip: the inode the reservation was taken out on | |
1015 | * | |
1016 | * Release a reservation made by gfs2_inplace_reserve(). | |
1017 | */ | |
1018 | ||
1019 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1020 | { | |
feaa7bba | 1021 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1022 | struct gfs2_alloc *al = &ip->i_alloc; |
1023 | ||
1024 | if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1) | |
1025 | fs_warn(sdp, "al_alloced = %u, al_requested = %u " | |
1026 | "al_file = %s, al_line = %u\n", | |
1027 | al->al_alloced, al->al_requested, al->al_file, | |
1028 | al->al_line); | |
1029 | ||
1030 | al->al_rgd = NULL; | |
1031 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
1032 | gfs2_glock_dq_uninit(&al->al_ri_gh); | |
1033 | } | |
1034 | ||
1035 | /** | |
1036 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1037 | * @rgd: the resource group holding the block | |
1038 | * @block: the block number | |
1039 | * | |
1040 | * Returns: The block type (GFS2_BLKST_*) | |
1041 | */ | |
1042 | ||
1043 | unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, uint64_t block) | |
1044 | { | |
1045 | struct gfs2_bitmap *bi = NULL; | |
1046 | uint32_t length, rgrp_block, buf_block; | |
1047 | unsigned int buf; | |
1048 | unsigned char type; | |
1049 | ||
1050 | length = rgd->rd_ri.ri_length; | |
1051 | rgrp_block = block - rgd->rd_ri.ri_data0; | |
1052 | ||
1053 | for (buf = 0; buf < length; buf++) { | |
1054 | bi = rgd->rd_bits + buf; | |
1055 | if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1056 | break; | |
1057 | } | |
1058 | ||
1059 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1060 | buf_block = rgrp_block - bi->bi_start * GFS2_NBBY; | |
1061 | ||
feaa7bba | 1062 | type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset, |
b3b94faa DT |
1063 | bi->bi_len, buf_block); |
1064 | ||
1065 | return type; | |
1066 | } | |
1067 | ||
1068 | /** | |
1069 | * rgblk_search - find a block in @old_state, change allocation | |
1070 | * state to @new_state | |
1071 | * @rgd: the resource group descriptor | |
1072 | * @goal: the goal block within the RG (start here to search for avail block) | |
1073 | * @old_state: GFS2_BLKST_XXX the before-allocation state to find | |
1074 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1075 | * | |
1076 | * Walk rgrp's bitmap to find bits that represent a block in @old_state. | |
1077 | * Add the found bitmap buffer to the transaction. | |
1078 | * Set the found bits to @new_state to change block's allocation state. | |
1079 | * | |
1080 | * This function never fails, because we wouldn't call it unless we | |
1081 | * know (from reservation results, etc.) that a block is available. | |
1082 | * | |
1083 | * Scope of @goal and returned block is just within rgrp, not the whole | |
1084 | * filesystem. | |
1085 | * | |
1086 | * Returns: the block number allocated | |
1087 | */ | |
1088 | ||
1089 | static uint32_t rgblk_search(struct gfs2_rgrpd *rgd, uint32_t goal, | |
1090 | unsigned char old_state, unsigned char new_state) | |
1091 | { | |
1092 | struct gfs2_bitmap *bi = NULL; | |
1093 | uint32_t length = rgd->rd_ri.ri_length; | |
1094 | uint32_t blk = 0; | |
1095 | unsigned int buf, x; | |
1096 | ||
1097 | /* Find bitmap block that contains bits for goal block */ | |
1098 | for (buf = 0; buf < length; buf++) { | |
1099 | bi = rgd->rd_bits + buf; | |
1100 | if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1101 | break; | |
1102 | } | |
1103 | ||
1104 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1105 | ||
1106 | /* Convert scope of "goal" from rgrp-wide to within found bit block */ | |
1107 | goal -= bi->bi_start * GFS2_NBBY; | |
1108 | ||
1109 | /* Search (up to entire) bitmap in this rgrp for allocatable block. | |
1110 | "x <= length", instead of "x < length", because we typically start | |
1111 | the search in the middle of a bit block, but if we can't find an | |
1112 | allocatable block anywhere else, we want to be able wrap around and | |
1113 | search in the first part of our first-searched bit block. */ | |
1114 | for (x = 0; x <= length; x++) { | |
1115 | if (bi->bi_clone) | |
fd88de56 | 1116 | blk = gfs2_bitfit(rgd, bi->bi_clone + bi->bi_offset, |
b3b94faa DT |
1117 | bi->bi_len, goal, old_state); |
1118 | else | |
1119 | blk = gfs2_bitfit(rgd, | |
1120 | bi->bi_bh->b_data + bi->bi_offset, | |
1121 | bi->bi_len, goal, old_state); | |
1122 | if (blk != BFITNOENT) | |
1123 | break; | |
1124 | ||
1125 | /* Try next bitmap block (wrap back to rgrp header if at end) */ | |
1126 | buf = (buf + 1) % length; | |
1127 | bi = rgd->rd_bits + buf; | |
1128 | goal = 0; | |
1129 | } | |
1130 | ||
1131 | if (gfs2_assert_withdraw(rgd->rd_sbd, x <= length)) | |
1132 | blk = 0; | |
1133 | ||
d4e9c4c3 | 1134 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
fd88de56 | 1135 | gfs2_setbit(rgd, bi->bi_bh->b_data + bi->bi_offset, |
b3b94faa DT |
1136 | bi->bi_len, blk, new_state); |
1137 | if (bi->bi_clone) | |
fd88de56 | 1138 | gfs2_setbit(rgd, bi->bi_clone + bi->bi_offset, |
b3b94faa DT |
1139 | bi->bi_len, blk, new_state); |
1140 | ||
1141 | return bi->bi_start * GFS2_NBBY + blk; | |
1142 | } | |
1143 | ||
1144 | /** | |
1145 | * rgblk_free - Change alloc state of given block(s) | |
1146 | * @sdp: the filesystem | |
1147 | * @bstart: the start of a run of blocks to free | |
1148 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1149 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1150 | * | |
1151 | * Returns: Resource group containing the block(s) | |
1152 | */ | |
1153 | ||
1154 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, uint64_t bstart, | |
1155 | uint32_t blen, unsigned char new_state) | |
1156 | { | |
1157 | struct gfs2_rgrpd *rgd; | |
1158 | struct gfs2_bitmap *bi = NULL; | |
1159 | uint32_t length, rgrp_blk, buf_blk; | |
1160 | unsigned int buf; | |
1161 | ||
1162 | rgd = gfs2_blk2rgrpd(sdp, bstart); | |
1163 | if (!rgd) { | |
1164 | if (gfs2_consist(sdp)) | |
382066da | 1165 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
1166 | return NULL; |
1167 | } | |
1168 | ||
1169 | length = rgd->rd_ri.ri_length; | |
1170 | ||
1171 | rgrp_blk = bstart - rgd->rd_ri.ri_data0; | |
1172 | ||
1173 | while (blen--) { | |
1174 | for (buf = 0; buf < length; buf++) { | |
1175 | bi = rgd->rd_bits + buf; | |
1176 | if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1177 | break; | |
1178 | } | |
1179 | ||
1180 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1181 | ||
1182 | buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY; | |
1183 | rgrp_blk++; | |
1184 | ||
1185 | if (!bi->bi_clone) { | |
1186 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
1187 | GFP_KERNEL | __GFP_NOFAIL); | |
1188 | memcpy(bi->bi_clone + bi->bi_offset, | |
1189 | bi->bi_bh->b_data + bi->bi_offset, | |
1190 | bi->bi_len); | |
1191 | } | |
d4e9c4c3 | 1192 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
b3b94faa DT |
1193 | gfs2_setbit(rgd, |
1194 | bi->bi_bh->b_data + bi->bi_offset, | |
1195 | bi->bi_len, buf_blk, new_state); | |
1196 | } | |
1197 | ||
1198 | return rgd; | |
1199 | } | |
1200 | ||
1201 | /** | |
1202 | * gfs2_alloc_data - Allocate a data block | |
1203 | * @ip: the inode to allocate the data block for | |
1204 | * | |
1205 | * Returns: the allocated block | |
1206 | */ | |
1207 | ||
4340fe62 | 1208 | u64 gfs2_alloc_data(struct gfs2_inode *ip) |
b3b94faa | 1209 | { |
feaa7bba | 1210 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1211 | struct gfs2_alloc *al = &ip->i_alloc; |
1212 | struct gfs2_rgrpd *rgd = al->al_rgd; | |
1213 | uint32_t goal, blk; | |
1214 | uint64_t block; | |
1215 | ||
1216 | if (rgrp_contains_block(&rgd->rd_ri, ip->i_di.di_goal_data)) | |
1217 | goal = ip->i_di.di_goal_data - rgd->rd_ri.ri_data0; | |
1218 | else | |
1219 | goal = rgd->rd_last_alloc_data; | |
1220 | ||
fd88de56 | 1221 | blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED); |
b3b94faa DT |
1222 | rgd->rd_last_alloc_data = blk; |
1223 | ||
1224 | block = rgd->rd_ri.ri_data0 + blk; | |
1225 | ip->i_di.di_goal_data = block; | |
1226 | ||
1227 | gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free); | |
1228 | rgd->rd_rg.rg_free--; | |
1229 | ||
d4e9c4c3 | 1230 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
b3b94faa DT |
1231 | gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data); |
1232 | ||
1233 | al->al_alloced++; | |
1234 | ||
1235 | gfs2_statfs_change(sdp, 0, -1, 0); | |
1236 | gfs2_quota_change(ip, +1, ip->i_di.di_uid, ip->i_di.di_gid); | |
1237 | ||
1238 | spin_lock(&sdp->sd_rindex_spin); | |
1239 | rgd->rd_free_clone--; | |
1240 | spin_unlock(&sdp->sd_rindex_spin); | |
1241 | ||
1242 | return block; | |
1243 | } | |
1244 | ||
1245 | /** | |
1246 | * gfs2_alloc_meta - Allocate a metadata block | |
1247 | * @ip: the inode to allocate the metadata block for | |
1248 | * | |
1249 | * Returns: the allocated block | |
1250 | */ | |
1251 | ||
4340fe62 | 1252 | u64 gfs2_alloc_meta(struct gfs2_inode *ip) |
b3b94faa | 1253 | { |
feaa7bba | 1254 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1255 | struct gfs2_alloc *al = &ip->i_alloc; |
1256 | struct gfs2_rgrpd *rgd = al->al_rgd; | |
1257 | uint32_t goal, blk; | |
1258 | uint64_t block; | |
1259 | ||
1260 | if (rgrp_contains_block(&rgd->rd_ri, ip->i_di.di_goal_meta)) | |
1261 | goal = ip->i_di.di_goal_meta - rgd->rd_ri.ri_data0; | |
1262 | else | |
1263 | goal = rgd->rd_last_alloc_meta; | |
1264 | ||
fd88de56 | 1265 | blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED); |
b3b94faa DT |
1266 | rgd->rd_last_alloc_meta = blk; |
1267 | ||
1268 | block = rgd->rd_ri.ri_data0 + blk; | |
1269 | ip->i_di.di_goal_meta = block; | |
1270 | ||
1271 | gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free); | |
1272 | rgd->rd_rg.rg_free--; | |
1273 | ||
d4e9c4c3 | 1274 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
b3b94faa DT |
1275 | gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data); |
1276 | ||
1277 | al->al_alloced++; | |
1278 | ||
1279 | gfs2_statfs_change(sdp, 0, -1, 0); | |
1280 | gfs2_quota_change(ip, +1, ip->i_di.di_uid, ip->i_di.di_gid); | |
1281 | gfs2_trans_add_unrevoke(sdp, block); | |
1282 | ||
1283 | spin_lock(&sdp->sd_rindex_spin); | |
1284 | rgd->rd_free_clone--; | |
1285 | spin_unlock(&sdp->sd_rindex_spin); | |
1286 | ||
1287 | return block; | |
1288 | } | |
1289 | ||
1290 | /** | |
1291 | * gfs2_alloc_di - Allocate a dinode | |
1292 | * @dip: the directory that the inode is going in | |
1293 | * | |
1294 | * Returns: the block allocated | |
1295 | */ | |
1296 | ||
4340fe62 | 1297 | u64 gfs2_alloc_di(struct gfs2_inode *dip, u64 *generation) |
b3b94faa | 1298 | { |
feaa7bba | 1299 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
b3b94faa DT |
1300 | struct gfs2_alloc *al = &dip->i_alloc; |
1301 | struct gfs2_rgrpd *rgd = al->al_rgd; | |
4340fe62 SW |
1302 | u32 blk; |
1303 | u64 block; | |
b3b94faa DT |
1304 | |
1305 | blk = rgblk_search(rgd, rgd->rd_last_alloc_meta, | |
1306 | GFS2_BLKST_FREE, GFS2_BLKST_DINODE); | |
1307 | ||
1308 | rgd->rd_last_alloc_meta = blk; | |
1309 | ||
1310 | block = rgd->rd_ri.ri_data0 + blk; | |
1311 | ||
1312 | gfs2_assert_withdraw(sdp, rgd->rd_rg.rg_free); | |
1313 | rgd->rd_rg.rg_free--; | |
1314 | rgd->rd_rg.rg_dinodes++; | |
4340fe62 | 1315 | *generation = rgd->rd_rg.rg_igeneration++; |
d4e9c4c3 | 1316 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
b3b94faa DT |
1317 | gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data); |
1318 | ||
1319 | al->al_alloced++; | |
1320 | ||
1321 | gfs2_statfs_change(sdp, 0, -1, +1); | |
1322 | gfs2_trans_add_unrevoke(sdp, block); | |
1323 | ||
1324 | spin_lock(&sdp->sd_rindex_spin); | |
1325 | rgd->rd_free_clone--; | |
1326 | spin_unlock(&sdp->sd_rindex_spin); | |
1327 | ||
1328 | return block; | |
1329 | } | |
1330 | ||
1331 | /** | |
1332 | * gfs2_free_data - free a contiguous run of data block(s) | |
1333 | * @ip: the inode these blocks are being freed from | |
1334 | * @bstart: first block of a run of contiguous blocks | |
1335 | * @blen: the length of the block run | |
1336 | * | |
1337 | */ | |
1338 | ||
1339 | void gfs2_free_data(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen) | |
1340 | { | |
feaa7bba | 1341 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1342 | struct gfs2_rgrpd *rgd; |
1343 | ||
1344 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
1345 | if (!rgd) | |
1346 | return; | |
1347 | ||
1348 | rgd->rd_rg.rg_free += blen; | |
1349 | ||
d4e9c4c3 | 1350 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
b3b94faa DT |
1351 | gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data); |
1352 | ||
1353 | gfs2_trans_add_rg(rgd); | |
1354 | ||
1355 | gfs2_statfs_change(sdp, 0, +blen, 0); | |
1356 | gfs2_quota_change(ip, -(int64_t)blen, | |
1357 | ip->i_di.di_uid, ip->i_di.di_gid); | |
1358 | } | |
1359 | ||
1360 | /** | |
1361 | * gfs2_free_meta - free a contiguous run of data block(s) | |
1362 | * @ip: the inode these blocks are being freed from | |
1363 | * @bstart: first block of a run of contiguous blocks | |
1364 | * @blen: the length of the block run | |
1365 | * | |
1366 | */ | |
1367 | ||
1368 | void gfs2_free_meta(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen) | |
1369 | { | |
feaa7bba | 1370 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1371 | struct gfs2_rgrpd *rgd; |
1372 | ||
1373 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
1374 | if (!rgd) | |
1375 | return; | |
1376 | ||
1377 | rgd->rd_rg.rg_free += blen; | |
1378 | ||
d4e9c4c3 | 1379 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
b3b94faa DT |
1380 | gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data); |
1381 | ||
1382 | gfs2_trans_add_rg(rgd); | |
1383 | ||
1384 | gfs2_statfs_change(sdp, 0, +blen, 0); | |
feaa7bba | 1385 | gfs2_quota_change(ip, -(int64_t)blen, ip->i_di.di_uid, ip->i_di.di_gid); |
b3b94faa DT |
1386 | gfs2_meta_wipe(ip, bstart, blen); |
1387 | } | |
1388 | ||
feaa7bba SW |
1389 | void gfs2_unlink_di(struct inode *inode) |
1390 | { | |
1391 | struct gfs2_inode *ip = GFS2_I(inode); | |
1392 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1393 | struct gfs2_rgrpd *rgd; | |
1394 | u64 blkno = ip->i_num.no_addr; | |
1395 | ||
1396 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
1397 | if (!rgd) | |
1398 | return; | |
1399 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); | |
1400 | gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data); | |
1401 | gfs2_trans_add_rg(rgd); | |
1402 | } | |
1403 | ||
43f5d210 | 1404 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, uint64_t blkno) |
b3b94faa DT |
1405 | { |
1406 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1407 | struct gfs2_rgrpd *tmp_rgd; | |
1408 | ||
1409 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
1410 | if (!tmp_rgd) | |
1411 | return; | |
1412 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
1413 | ||
1414 | if (!rgd->rd_rg.rg_dinodes) | |
1415 | gfs2_consist_rgrpd(rgd); | |
1416 | rgd->rd_rg.rg_dinodes--; | |
1417 | rgd->rd_rg.rg_free++; | |
1418 | ||
d4e9c4c3 | 1419 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
b3b94faa DT |
1420 | gfs2_rgrp_out(&rgd->rd_rg, rgd->rd_bits[0].bi_bh->b_data); |
1421 | ||
1422 | gfs2_statfs_change(sdp, 0, +1, -1); | |
1423 | gfs2_trans_add_rg(rgd); | |
1424 | } | |
1425 | ||
b3b94faa DT |
1426 | |
1427 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
1428 | { | |
1429 | gfs2_free_uninit_di(rgd, ip->i_num.no_addr); | |
1430 | gfs2_quota_change(ip, -1, ip->i_di.di_uid, ip->i_di.di_gid); | |
1431 | gfs2_meta_wipe(ip, ip->i_num.no_addr, 1); | |
1432 | } | |
1433 | ||
1434 | /** | |
1435 | * gfs2_rlist_add - add a RG to a list of RGs | |
1436 | * @sdp: the filesystem | |
1437 | * @rlist: the list of resource groups | |
1438 | * @block: the block | |
1439 | * | |
1440 | * Figure out what RG a block belongs to and add that RG to the list | |
1441 | * | |
1442 | * FIXME: Don't use NOFAIL | |
1443 | * | |
1444 | */ | |
1445 | ||
1446 | void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist, | |
1447 | uint64_t block) | |
1448 | { | |
1449 | struct gfs2_rgrpd *rgd; | |
1450 | struct gfs2_rgrpd **tmp; | |
1451 | unsigned int new_space; | |
1452 | unsigned int x; | |
1453 | ||
1454 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
1455 | return; | |
1456 | ||
1457 | rgd = gfs2_blk2rgrpd(sdp, block); | |
1458 | if (!rgd) { | |
1459 | if (gfs2_consist(sdp)) | |
382066da | 1460 | fs_err(sdp, "block = %llu\n", (unsigned long long)block); |
b3b94faa DT |
1461 | return; |
1462 | } | |
1463 | ||
1464 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1465 | if (rlist->rl_rgd[x] == rgd) | |
1466 | return; | |
1467 | ||
1468 | if (rlist->rl_rgrps == rlist->rl_space) { | |
1469 | new_space = rlist->rl_space + 10; | |
1470 | ||
1471 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
1472 | GFP_KERNEL | __GFP_NOFAIL); | |
1473 | ||
1474 | if (rlist->rl_rgd) { | |
1475 | memcpy(tmp, rlist->rl_rgd, | |
1476 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
1477 | kfree(rlist->rl_rgd); | |
1478 | } | |
1479 | ||
1480 | rlist->rl_space = new_space; | |
1481 | rlist->rl_rgd = tmp; | |
1482 | } | |
1483 | ||
1484 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
1485 | } | |
1486 | ||
1487 | /** | |
1488 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
1489 | * and initialize an array of glock holders for them | |
1490 | * @rlist: the list of resource groups | |
1491 | * @state: the lock state to acquire the RG lock in | |
1492 | * @flags: the modifier flags for the holder structures | |
1493 | * | |
1494 | * FIXME: Don't use NOFAIL | |
1495 | * | |
1496 | */ | |
1497 | ||
1498 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state, | |
1499 | int flags) | |
1500 | { | |
1501 | unsigned int x; | |
1502 | ||
1503 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
1504 | GFP_KERNEL | __GFP_NOFAIL); | |
1505 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1506 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
1507 | state, flags, | |
1508 | &rlist->rl_ghs[x]); | |
1509 | } | |
1510 | ||
1511 | /** | |
1512 | * gfs2_rlist_free - free a resource group list | |
1513 | * @list: the list of resource groups | |
1514 | * | |
1515 | */ | |
1516 | ||
1517 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
1518 | { | |
1519 | unsigned int x; | |
1520 | ||
1521 | kfree(rlist->rl_rgd); | |
1522 | ||
1523 | if (rlist->rl_ghs) { | |
1524 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1525 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
1526 | kfree(rlist->rl_ghs); | |
1527 | } | |
1528 | } | |
1529 |