Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
fe6c991c | 3 | * Copyright (C) 2004-2008 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 | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
b3b94faa DT |
10 | #include <linux/slab.h> |
11 | #include <linux/spinlock.h> | |
12 | #include <linux/completion.h> | |
13 | #include <linux/buffer_head.h> | |
f42faf4f | 14 | #include <linux/fs.h> |
5c676f6d | 15 | #include <linux/gfs2_ondisk.h> |
1f466a47 | 16 | #include <linux/prefetch.h> |
f15ab561 | 17 | #include <linux/blkdev.h> |
b3b94faa DT |
18 | |
19 | #include "gfs2.h" | |
5c676f6d | 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" | |
5c676f6d | 29 | #include "util.h" |
172e045a | 30 | #include "log.h" |
c8cdf479 | 31 | #include "inode.h" |
63997775 | 32 | #include "trace_gfs2.h" |
b3b94faa | 33 | |
2c1e52aa | 34 | #define BFITNOENT ((u32)~0) |
6760bdcd | 35 | #define NO_BLOCK ((u64)~0) |
88c8ab1f | 36 | |
1f466a47 BP |
37 | #if BITS_PER_LONG == 32 |
38 | #define LBITMASK (0x55555555UL) | |
39 | #define LBITSKIP55 (0x55555555UL) | |
40 | #define LBITSKIP00 (0x00000000UL) | |
41 | #else | |
42 | #define LBITMASK (0x5555555555555555UL) | |
43 | #define LBITSKIP55 (0x5555555555555555UL) | |
44 | #define LBITSKIP00 (0x0000000000000000UL) | |
45 | #endif | |
46 | ||
88c8ab1f SW |
47 | /* |
48 | * These routines are used by the resource group routines (rgrp.c) | |
49 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
50 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
51 | * | |
52 | * 0 = Free | |
53 | * 1 = Used (not metadata) | |
54 | * 2 = Unlinked (still in use) inode | |
55 | * 3 = Used (metadata) | |
88c8ab1f SW |
56 | */ |
57 | ||
58 | static const char valid_change[16] = { | |
59 | /* current */ | |
feaa7bba | 60 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 61 | /* e */ 1, 0, 0, 0, |
feaa7bba | 62 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
63 | 1, 0, 0, 0 |
64 | }; | |
65 | ||
c8cdf479 | 66 | static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, |
b45e41d7 SW |
67 | unsigned char old_state, unsigned char new_state, |
68 | unsigned int *n); | |
c8cdf479 | 69 | |
88c8ab1f SW |
70 | /** |
71 | * gfs2_setbit - Set a bit in the bitmaps | |
72 | * @buffer: the buffer that holds the bitmaps | |
73 | * @buflen: the length (in bytes) of the buffer | |
74 | * @block: the block to set | |
75 | * @new_state: the new state of the block | |
76 | * | |
77 | */ | |
78 | ||
b45e41d7 SW |
79 | static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1, |
80 | unsigned char *buf2, unsigned int offset, | |
81 | unsigned int buflen, u32 block, | |
82 | unsigned char new_state) | |
88c8ab1f | 83 | { |
b45e41d7 SW |
84 | unsigned char *byte1, *byte2, *end, cur_state; |
85 | const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 86 | |
b45e41d7 SW |
87 | byte1 = buf1 + offset + (block / GFS2_NBBY); |
88 | end = buf1 + offset + buflen; | |
88c8ab1f | 89 | |
b45e41d7 | 90 | BUG_ON(byte1 >= end); |
88c8ab1f | 91 | |
b45e41d7 | 92 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 93 | |
b45e41d7 | 94 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
88c8ab1f | 95 | gfs2_consist_rgrpd(rgd); |
b45e41d7 SW |
96 | return; |
97 | } | |
98 | *byte1 ^= (cur_state ^ new_state) << bit; | |
99 | ||
100 | if (buf2) { | |
101 | byte2 = buf2 + offset + (block / GFS2_NBBY); | |
102 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; | |
103 | *byte2 ^= (cur_state ^ new_state) << bit; | |
104 | } | |
88c8ab1f SW |
105 | } |
106 | ||
107 | /** | |
108 | * gfs2_testbit - test a bit in the bitmaps | |
109 | * @buffer: the buffer that holds the bitmaps | |
110 | * @buflen: the length (in bytes) of the buffer | |
111 | * @block: the block to read | |
112 | * | |
113 | */ | |
114 | ||
b45e41d7 SW |
115 | static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd, |
116 | const unsigned char *buffer, | |
117 | unsigned int buflen, u32 block) | |
88c8ab1f | 118 | { |
b45e41d7 SW |
119 | const unsigned char *byte, *end; |
120 | unsigned char cur_state; | |
88c8ab1f SW |
121 | unsigned int bit; |
122 | ||
123 | byte = buffer + (block / GFS2_NBBY); | |
124 | bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; | |
125 | end = buffer + buflen; | |
126 | ||
127 | gfs2_assert(rgd->rd_sbd, byte < end); | |
128 | ||
129 | cur_state = (*byte >> bit) & GFS2_BIT_MASK; | |
130 | ||
131 | return cur_state; | |
132 | } | |
133 | ||
223b2b88 SW |
134 | /** |
135 | * gfs2_bit_search | |
136 | * @ptr: Pointer to bitmap data | |
137 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
138 | * @state: The state we are searching for | |
139 | * | |
140 | * We xor the bitmap data with a patter which is the bitwise opposite | |
141 | * of what we are looking for, this gives rise to a pattern of ones | |
142 | * wherever there is a match. Since we have two bits per entry, we | |
143 | * take this pattern, shift it down by one place and then and it with | |
144 | * the original. All the even bit positions (0,2,4, etc) then represent | |
145 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
146 | * odd bit positions. | |
147 | * | |
148 | * This allows searching of a whole u64 at once (32 blocks) with a | |
149 | * single test (on 64 bit arches). | |
150 | */ | |
151 | ||
152 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
153 | { | |
154 | u64 tmp; | |
155 | static const u64 search[] = { | |
075ac448 HE |
156 | [0] = 0xffffffffffffffffULL, |
157 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
158 | [2] = 0x5555555555555555ULL, | |
159 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
160 | }; |
161 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
162 | tmp &= (tmp >> 1); | |
163 | tmp &= mask; | |
164 | return tmp; | |
165 | } | |
166 | ||
88c8ab1f SW |
167 | /** |
168 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
169 | * a block in a given allocation state. | |
170 | * @buffer: the buffer that holds the bitmaps | |
223b2b88 | 171 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 172 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 173 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
174 | * |
175 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
176 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
177 | * beginning of a bitmap block buffer, skipping any header structures, but |
178 | * headers are always a multiple of 64 bits long so that the buffer is | |
179 | * always aligned to a 64 bit boundary. | |
180 | * | |
181 | * The size of the buffer is in bytes, but is it assumed that it is | |
182 | * always ok to to read a complete multiple of 64 bits at the end | |
183 | * of the block in case the end is no aligned to a natural boundary. | |
88c8ab1f SW |
184 | * |
185 | * Return: the block number (bitmap buffer scope) that was found | |
186 | */ | |
187 | ||
02ab1721 HE |
188 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
189 | u32 goal, u8 state) | |
88c8ab1f | 190 | { |
223b2b88 SW |
191 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
192 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
193 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
194 | u64 tmp; | |
075ac448 | 195 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
196 | u32 bit; |
197 | ||
198 | BUG_ON(state > 3); | |
199 | ||
200 | /* Mask off bits we don't care about at the start of the search */ | |
201 | mask <<= spoint; | |
202 | tmp = gfs2_bit_search(ptr, mask, state); | |
203 | ptr++; | |
204 | while(tmp == 0 && ptr < end) { | |
075ac448 | 205 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 206 | ptr++; |
1f466a47 | 207 | } |
223b2b88 SW |
208 | /* Mask off any bits which are more than len bytes from the start */ |
209 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
210 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
211 | /* Didn't find anything, so return */ | |
212 | if (tmp == 0) | |
213 | return BFITNOENT; | |
214 | ptr--; | |
d8bd504a | 215 | bit = __ffs64(tmp); |
223b2b88 SW |
216 | bit /= 2; /* two bits per entry in the bitmap */ |
217 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
218 | } |
219 | ||
220 | /** | |
221 | * gfs2_bitcount - count the number of bits in a certain state | |
222 | * @buffer: the buffer that holds the bitmaps | |
223 | * @buflen: the length (in bytes) of the buffer | |
224 | * @state: the state of the block we're looking for | |
225 | * | |
226 | * Returns: The number of bits | |
227 | */ | |
228 | ||
110acf38 SW |
229 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
230 | unsigned int buflen, u8 state) | |
88c8ab1f | 231 | { |
110acf38 SW |
232 | const u8 *byte = buffer; |
233 | const u8 *end = buffer + buflen; | |
234 | const u8 state1 = state << 2; | |
235 | const u8 state2 = state << 4; | |
236 | const u8 state3 = state << 6; | |
cd915493 | 237 | u32 count = 0; |
88c8ab1f SW |
238 | |
239 | for (; byte < end; byte++) { | |
240 | if (((*byte) & 0x03) == state) | |
241 | count++; | |
242 | if (((*byte) & 0x0C) == state1) | |
243 | count++; | |
244 | if (((*byte) & 0x30) == state2) | |
245 | count++; | |
246 | if (((*byte) & 0xC0) == state3) | |
247 | count++; | |
248 | } | |
249 | ||
250 | return count; | |
251 | } | |
252 | ||
b3b94faa DT |
253 | /** |
254 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
255 | * @sdp: the filesystem | |
256 | * @rgd: the rgrp | |
257 | * | |
258 | */ | |
259 | ||
260 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
261 | { | |
262 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
263 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 264 | u32 length = rgd->rd_length; |
cd915493 | 265 | u32 count[4], tmp; |
b3b94faa DT |
266 | int buf, x; |
267 | ||
cd915493 | 268 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
269 | |
270 | /* Count # blocks in each of 4 possible allocation states */ | |
271 | for (buf = 0; buf < length; buf++) { | |
272 | bi = rgd->rd_bits + buf; | |
273 | for (x = 0; x < 4; x++) | |
274 | count[x] += gfs2_bitcount(rgd, | |
275 | bi->bi_bh->b_data + | |
276 | bi->bi_offset, | |
277 | bi->bi_len, x); | |
278 | } | |
279 | ||
cfc8b549 | 280 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
281 | if (gfs2_consist_rgrpd(rgd)) |
282 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 283 | count[0], rgd->rd_free); |
b3b94faa DT |
284 | return; |
285 | } | |
286 | ||
73f74948 | 287 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
feaa7bba | 288 | if (count[1] + count[2] != tmp) { |
b3b94faa DT |
289 | if (gfs2_consist_rgrpd(rgd)) |
290 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
291 | count[1], tmp); | |
292 | return; | |
293 | } | |
294 | ||
73f74948 | 295 | if (count[3] != rgd->rd_dinodes) { |
b3b94faa | 296 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 297 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
73f74948 | 298 | count[3], rgd->rd_dinodes); |
b3b94faa DT |
299 | return; |
300 | } | |
301 | ||
feaa7bba | 302 | if (count[2] > count[3]) { |
b3b94faa | 303 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba SW |
304 | fs_err(sdp, "unlinked inodes > inodes: %u\n", |
305 | count[2]); | |
b3b94faa DT |
306 | return; |
307 | } | |
feaa7bba | 308 | |
b3b94faa DT |
309 | } |
310 | ||
bb8d8a6f | 311 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 312 | { |
bb8d8a6f SW |
313 | u64 first = rgd->rd_data0; |
314 | u64 last = first + rgd->rd_data; | |
16910427 | 315 | return first <= block && block < last; |
b3b94faa DT |
316 | } |
317 | ||
318 | /** | |
319 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
320 | * @sdp: The GFS2 superblock | |
321 | * @n: The data block number | |
322 | * | |
323 | * Returns: The resource group, or NULL if not found | |
324 | */ | |
325 | ||
cd915493 | 326 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk) |
b3b94faa DT |
327 | { |
328 | struct gfs2_rgrpd *rgd; | |
329 | ||
330 | spin_lock(&sdp->sd_rindex_spin); | |
331 | ||
332 | list_for_each_entry(rgd, &sdp->sd_rindex_mru_list, rd_list_mru) { | |
bb8d8a6f | 333 | if (rgrp_contains_block(rgd, blk)) { |
b3b94faa DT |
334 | list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list); |
335 | spin_unlock(&sdp->sd_rindex_spin); | |
336 | return rgd; | |
337 | } | |
338 | } | |
339 | ||
340 | spin_unlock(&sdp->sd_rindex_spin); | |
341 | ||
342 | return NULL; | |
343 | } | |
344 | ||
345 | /** | |
346 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
347 | * @sdp: The GFS2 superblock | |
348 | * | |
349 | * Returns: The first rgrp in the filesystem | |
350 | */ | |
351 | ||
352 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
353 | { | |
354 | gfs2_assert(sdp, !list_empty(&sdp->sd_rindex_list)); | |
355 | return list_entry(sdp->sd_rindex_list.next, struct gfs2_rgrpd, rd_list); | |
356 | } | |
357 | ||
358 | /** | |
359 | * gfs2_rgrpd_get_next - get the next RG | |
360 | * @rgd: A RG | |
361 | * | |
362 | * Returns: The next rgrp | |
363 | */ | |
364 | ||
365 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
366 | { | |
367 | if (rgd->rd_list.next == &rgd->rd_sbd->sd_rindex_list) | |
368 | return NULL; | |
369 | return list_entry(rgd->rd_list.next, struct gfs2_rgrpd, rd_list); | |
370 | } | |
371 | ||
372 | static void clear_rgrpdi(struct gfs2_sbd *sdp) | |
373 | { | |
374 | struct list_head *head; | |
375 | struct gfs2_rgrpd *rgd; | |
376 | struct gfs2_glock *gl; | |
377 | ||
378 | spin_lock(&sdp->sd_rindex_spin); | |
379 | sdp->sd_rindex_forward = NULL; | |
b3b94faa DT |
380 | spin_unlock(&sdp->sd_rindex_spin); |
381 | ||
382 | head = &sdp->sd_rindex_list; | |
383 | while (!list_empty(head)) { | |
384 | rgd = list_entry(head->next, struct gfs2_rgrpd, rd_list); | |
385 | gl = rgd->rd_gl; | |
386 | ||
387 | list_del(&rgd->rd_list); | |
388 | list_del(&rgd->rd_list_mru); | |
389 | ||
390 | if (gl) { | |
5c676f6d | 391 | gl->gl_object = NULL; |
b3b94faa DT |
392 | gfs2_glock_put(gl); |
393 | } | |
394 | ||
395 | kfree(rgd->rd_bits); | |
6bdd9be6 | 396 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
397 | } |
398 | } | |
399 | ||
400 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) | |
401 | { | |
f55ab26a | 402 | mutex_lock(&sdp->sd_rindex_mutex); |
b3b94faa | 403 | clear_rgrpdi(sdp); |
f55ab26a | 404 | mutex_unlock(&sdp->sd_rindex_mutex); |
b3b94faa DT |
405 | } |
406 | ||
bb8d8a6f SW |
407 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
408 | { | |
409 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
410 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
411 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
412 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
413 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
414 | } | |
415 | ||
b3b94faa DT |
416 | /** |
417 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
418 | * @rgd: The resource group descriptor | |
419 | * | |
420 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
421 | * | |
422 | * Returns: errno | |
423 | */ | |
424 | ||
425 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
426 | { | |
427 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
428 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 429 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 430 | u32 bytes_left, bytes; |
b3b94faa DT |
431 | int x; |
432 | ||
feaa7bba SW |
433 | if (!length) |
434 | return -EINVAL; | |
435 | ||
dd894be8 | 436 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
437 | if (!rgd->rd_bits) |
438 | return -ENOMEM; | |
439 | ||
bb8d8a6f | 440 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
441 | |
442 | for (x = 0; x < length; x++) { | |
443 | bi = rgd->rd_bits + x; | |
444 | ||
60a0b8f9 | 445 | bi->bi_flags = 0; |
b3b94faa DT |
446 | /* small rgrp; bitmap stored completely in header block */ |
447 | if (length == 1) { | |
448 | bytes = bytes_left; | |
449 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
450 | bi->bi_start = 0; | |
451 | bi->bi_len = bytes; | |
452 | /* header block */ | |
453 | } else if (x == 0) { | |
454 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
455 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
456 | bi->bi_start = 0; | |
457 | bi->bi_len = bytes; | |
458 | /* last block */ | |
459 | } else if (x + 1 == length) { | |
460 | bytes = bytes_left; | |
461 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 462 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
463 | bi->bi_len = bytes; |
464 | /* other blocks */ | |
465 | } else { | |
568f4c96 SW |
466 | bytes = sdp->sd_sb.sb_bsize - |
467 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 468 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 469 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
470 | bi->bi_len = bytes; |
471 | } | |
472 | ||
473 | bytes_left -= bytes; | |
474 | } | |
475 | ||
476 | if (bytes_left) { | |
477 | gfs2_consist_rgrpd(rgd); | |
478 | return -EIO; | |
479 | } | |
480 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 481 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 482 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 483 | gfs2_rindex_print(rgd); |
b3b94faa DT |
484 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
485 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
486 | } | |
487 | return -EIO; | |
488 | } | |
489 | ||
490 | return 0; | |
491 | } | |
492 | ||
7ae8fa84 RP |
493 | /** |
494 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
495 | * | |
496 | */ | |
497 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
498 | { | |
499 | u64 total_data = 0; | |
500 | struct inode *inode = sdp->sd_rindex; | |
501 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 RP |
502 | char buf[sizeof(struct gfs2_rindex)]; |
503 | struct file_ra_state ra_state; | |
504 | int error, rgrps; | |
505 | ||
506 | mutex_lock(&sdp->sd_rindex_mutex); | |
507 | file_ra_state_init(&ra_state, inode->i_mapping); | |
508 | for (rgrps = 0;; rgrps++) { | |
509 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
510 | ||
c9e98886 | 511 | if (pos + sizeof(struct gfs2_rindex) >= ip->i_disksize) |
7ae8fa84 RP |
512 | break; |
513 | error = gfs2_internal_read(ip, &ra_state, buf, &pos, | |
514 | sizeof(struct gfs2_rindex)); | |
515 | if (error != sizeof(struct gfs2_rindex)) | |
516 | break; | |
bb8d8a6f | 517 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 RP |
518 | } |
519 | mutex_unlock(&sdp->sd_rindex_mutex); | |
520 | return total_data; | |
521 | } | |
522 | ||
bb8d8a6f SW |
523 | static void gfs2_rindex_in(struct gfs2_rgrpd *rgd, const void *buf) |
524 | { | |
525 | const struct gfs2_rindex *str = buf; | |
526 | ||
527 | rgd->rd_addr = be64_to_cpu(str->ri_addr); | |
528 | rgd->rd_length = be32_to_cpu(str->ri_length); | |
529 | rgd->rd_data0 = be64_to_cpu(str->ri_data0); | |
530 | rgd->rd_data = be32_to_cpu(str->ri_data); | |
531 | rgd->rd_bitbytes = be32_to_cpu(str->ri_bitbytes); | |
532 | } | |
533 | ||
b3b94faa | 534 | /** |
6c53267f | 535 | * read_rindex_entry - Pull in a new resource index entry from the disk |
b3b94faa DT |
536 | * @gl: The glock covering the rindex inode |
537 | * | |
6c53267f RP |
538 | * Returns: 0 on success, error code otherwise |
539 | */ | |
540 | ||
541 | static int read_rindex_entry(struct gfs2_inode *ip, | |
542 | struct file_ra_state *ra_state) | |
543 | { | |
544 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
545 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
546 | char buf[sizeof(struct gfs2_rindex)]; | |
547 | int error; | |
548 | struct gfs2_rgrpd *rgd; | |
549 | ||
550 | error = gfs2_internal_read(ip, ra_state, buf, &pos, | |
551 | sizeof(struct gfs2_rindex)); | |
552 | if (!error) | |
553 | return 0; | |
554 | if (error != sizeof(struct gfs2_rindex)) { | |
555 | if (error > 0) | |
556 | error = -EIO; | |
557 | return error; | |
558 | } | |
559 | ||
6bdd9be6 | 560 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
561 | error = -ENOMEM; |
562 | if (!rgd) | |
563 | return error; | |
564 | ||
565 | mutex_init(&rgd->rd_mutex); | |
566 | lops_init_le(&rgd->rd_le, &gfs2_rg_lops); | |
567 | rgd->rd_sbd = sdp; | |
568 | ||
569 | list_add_tail(&rgd->rd_list, &sdp->sd_rindex_list); | |
570 | list_add_tail(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list); | |
571 | ||
bb8d8a6f | 572 | gfs2_rindex_in(rgd, buf); |
6c53267f RP |
573 | error = compute_bitstructs(rgd); |
574 | if (error) | |
575 | return error; | |
576 | ||
bb8d8a6f | 577 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
578 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
579 | if (error) | |
580 | return error; | |
581 | ||
582 | rgd->rd_gl->gl_object = rgd; | |
cf45b752 | 583 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
6c53267f RP |
584 | return error; |
585 | } | |
586 | ||
587 | /** | |
588 | * gfs2_ri_update - Pull in a new resource index from the disk | |
589 | * @ip: pointer to the rindex inode | |
590 | * | |
b3b94faa DT |
591 | * Returns: 0 on successful update, error code otherwise |
592 | */ | |
593 | ||
594 | static int gfs2_ri_update(struct gfs2_inode *ip) | |
595 | { | |
feaa7bba SW |
596 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
597 | struct inode *inode = &ip->i_inode; | |
f42faf4f | 598 | struct file_ra_state ra_state; |
c9e98886 | 599 | u64 rgrp_count = ip->i_disksize; |
b3b94faa DT |
600 | int error; |
601 | ||
cd81a4ba | 602 | if (do_div(rgrp_count, sizeof(struct gfs2_rindex))) { |
b3b94faa DT |
603 | gfs2_consist_inode(ip); |
604 | return -EIO; | |
605 | } | |
606 | ||
607 | clear_rgrpdi(sdp); | |
608 | ||
f42faf4f | 609 | file_ra_state_init(&ra_state, inode->i_mapping); |
cd81a4ba | 610 | for (sdp->sd_rgrps = 0; sdp->sd_rgrps < rgrp_count; sdp->sd_rgrps++) { |
6c53267f RP |
611 | error = read_rindex_entry(ip, &ra_state); |
612 | if (error) { | |
613 | clear_rgrpdi(sdp); | |
614 | return error; | |
b3b94faa | 615 | } |
6c53267f | 616 | } |
b3b94faa | 617 | |
cf45b752 | 618 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
619 | return 0; |
620 | } | |
b3b94faa | 621 | |
6c53267f RP |
622 | /** |
623 | * gfs2_ri_update_special - Pull in a new resource index from the disk | |
624 | * | |
625 | * This is a special version that's safe to call from gfs2_inplace_reserve_i. | |
626 | * In this case we know that we don't have any resource groups in memory yet. | |
627 | * | |
628 | * @ip: pointer to the rindex inode | |
629 | * | |
630 | * Returns: 0 on successful update, error code otherwise | |
631 | */ | |
632 | static int gfs2_ri_update_special(struct gfs2_inode *ip) | |
633 | { | |
634 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
635 | struct inode *inode = &ip->i_inode; | |
636 | struct file_ra_state ra_state; | |
637 | int error; | |
b3b94faa | 638 | |
6c53267f RP |
639 | file_ra_state_init(&ra_state, inode->i_mapping); |
640 | for (sdp->sd_rgrps = 0;; sdp->sd_rgrps++) { | |
641 | /* Ignore partials */ | |
642 | if ((sdp->sd_rgrps + 1) * sizeof(struct gfs2_rindex) > | |
c9e98886 | 643 | ip->i_disksize) |
6c53267f RP |
644 | break; |
645 | error = read_rindex_entry(ip, &ra_state); | |
646 | if (error) { | |
647 | clear_rgrpdi(sdp); | |
648 | return error; | |
649 | } | |
b3b94faa DT |
650 | } |
651 | ||
cf45b752 | 652 | sdp->sd_rindex_uptodate = 1; |
b3b94faa | 653 | return 0; |
b3b94faa DT |
654 | } |
655 | ||
656 | /** | |
657 | * gfs2_rindex_hold - Grab a lock on the rindex | |
658 | * @sdp: The GFS2 superblock | |
659 | * @ri_gh: the glock holder | |
660 | * | |
661 | * We grab a lock on the rindex inode to make sure that it doesn't | |
662 | * change whilst we are performing an operation. We keep this lock | |
663 | * for quite long periods of time compared to other locks. This | |
664 | * doesn't matter, since it is shared and it is very, very rarely | |
665 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
666 | * | |
667 | * This makes sure that we're using the latest copy of the resource index | |
668 | * special file, which might have been updated if someone expanded the | |
669 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
670 | * | |
671 | * Returns: 0 on success, error code otherwise | |
672 | */ | |
673 | ||
674 | int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh) | |
675 | { | |
feaa7bba | 676 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa DT |
677 | struct gfs2_glock *gl = ip->i_gl; |
678 | int error; | |
679 | ||
680 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh); | |
681 | if (error) | |
682 | return error; | |
683 | ||
684 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 685 | if (!sdp->sd_rindex_uptodate) { |
f55ab26a | 686 | mutex_lock(&sdp->sd_rindex_mutex); |
cf45b752 | 687 | if (!sdp->sd_rindex_uptodate) { |
b3b94faa DT |
688 | error = gfs2_ri_update(ip); |
689 | if (error) | |
690 | gfs2_glock_dq_uninit(ri_gh); | |
691 | } | |
f55ab26a | 692 | mutex_unlock(&sdp->sd_rindex_mutex); |
b3b94faa DT |
693 | } |
694 | ||
695 | return error; | |
696 | } | |
697 | ||
42d52e38 | 698 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
699 | { |
700 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 701 | u32 rg_flags; |
bb8d8a6f | 702 | |
42d52e38 | 703 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 704 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
705 | rgd->rd_flags &= GFS2_RDF_MASK; |
706 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 707 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 708 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 709 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
710 | } |
711 | ||
42d52e38 | 712 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
713 | { |
714 | struct gfs2_rgrp *str = buf; | |
715 | ||
09010978 | 716 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 717 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 718 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 719 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 720 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
721 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
722 | } | |
723 | ||
b3b94faa DT |
724 | /** |
725 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps | |
726 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
727 | * | |
728 | * Read in all of a Resource Group's header and bitmap blocks. | |
729 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
730 | * | |
731 | * Returns: errno | |
732 | */ | |
733 | ||
734 | int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) | |
735 | { | |
736 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
737 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 738 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
739 | struct gfs2_bitmap *bi; |
740 | unsigned int x, y; | |
741 | int error; | |
742 | ||
f55ab26a | 743 | mutex_lock(&rgd->rd_mutex); |
b3b94faa DT |
744 | |
745 | spin_lock(&sdp->sd_rindex_spin); | |
746 | if (rgd->rd_bh_count) { | |
747 | rgd->rd_bh_count++; | |
748 | spin_unlock(&sdp->sd_rindex_spin); | |
f55ab26a | 749 | mutex_unlock(&rgd->rd_mutex); |
b3b94faa DT |
750 | return 0; |
751 | } | |
752 | spin_unlock(&sdp->sd_rindex_spin); | |
753 | ||
754 | for (x = 0; x < length; x++) { | |
755 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 756 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
757 | if (error) |
758 | goto fail; | |
759 | } | |
760 | ||
761 | for (y = length; y--;) { | |
762 | bi = rgd->rd_bits + y; | |
7276b3b0 | 763 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
764 | if (error) |
765 | goto fail; | |
feaa7bba | 766 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
767 | GFS2_METATYPE_RG)) { |
768 | error = -EIO; | |
769 | goto fail; | |
770 | } | |
771 | } | |
772 | ||
cf45b752 | 773 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
774 | for (x = 0; x < length; x++) |
775 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 776 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 777 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
b3b94faa DT |
778 | } |
779 | ||
780 | spin_lock(&sdp->sd_rindex_spin); | |
cfc8b549 | 781 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa DT |
782 | rgd->rd_bh_count++; |
783 | spin_unlock(&sdp->sd_rindex_spin); | |
784 | ||
f55ab26a | 785 | mutex_unlock(&rgd->rd_mutex); |
b3b94faa DT |
786 | |
787 | return 0; | |
788 | ||
feaa7bba | 789 | fail: |
b3b94faa DT |
790 | while (x--) { |
791 | bi = rgd->rd_bits + x; | |
792 | brelse(bi->bi_bh); | |
793 | bi->bi_bh = NULL; | |
794 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
795 | } | |
f55ab26a | 796 | mutex_unlock(&rgd->rd_mutex); |
b3b94faa DT |
797 | |
798 | return error; | |
799 | } | |
800 | ||
801 | void gfs2_rgrp_bh_hold(struct gfs2_rgrpd *rgd) | |
802 | { | |
803 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
804 | ||
805 | spin_lock(&sdp->sd_rindex_spin); | |
806 | gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count); | |
807 | rgd->rd_bh_count++; | |
808 | spin_unlock(&sdp->sd_rindex_spin); | |
809 | } | |
810 | ||
811 | /** | |
812 | * gfs2_rgrp_bh_put - Release RG bitmaps read in with gfs2_rgrp_bh_get() | |
813 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
814 | * | |
815 | */ | |
816 | ||
817 | void gfs2_rgrp_bh_put(struct gfs2_rgrpd *rgd) | |
818 | { | |
819 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
bb8d8a6f | 820 | int x, length = rgd->rd_length; |
b3b94faa DT |
821 | |
822 | spin_lock(&sdp->sd_rindex_spin); | |
823 | gfs2_assert_warn(rgd->rd_sbd, rgd->rd_bh_count); | |
824 | if (--rgd->rd_bh_count) { | |
825 | spin_unlock(&sdp->sd_rindex_spin); | |
826 | return; | |
827 | } | |
828 | ||
829 | for (x = 0; x < length; x++) { | |
830 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
831 | kfree(bi->bi_clone); | |
832 | bi->bi_clone = NULL; | |
833 | brelse(bi->bi_bh); | |
834 | bi->bi_bh = NULL; | |
835 | } | |
836 | ||
837 | spin_unlock(&sdp->sd_rindex_spin); | |
838 | } | |
839 | ||
f15ab561 SW |
840 | static void gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
841 | const struct gfs2_bitmap *bi) | |
842 | { | |
843 | struct super_block *sb = sdp->sd_vfs; | |
844 | struct block_device *bdev = sb->s_bdev; | |
845 | const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / | |
e1defc4f | 846 | bdev_logical_block_size(sb->s_bdev); |
f15ab561 | 847 | u64 blk; |
64d576ba | 848 | sector_t start = 0; |
f15ab561 SW |
849 | sector_t nr_sects = 0; |
850 | int rv; | |
851 | unsigned int x; | |
852 | ||
853 | for (x = 0; x < bi->bi_len; x++) { | |
854 | const u8 *orig = bi->bi_bh->b_data + bi->bi_offset + x; | |
855 | const u8 *clone = bi->bi_clone + bi->bi_offset + x; | |
856 | u8 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
857 | diff &= 0x55; | |
858 | if (diff == 0) | |
859 | continue; | |
860 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
861 | blk *= sects_per_blk; /* convert to sectors */ | |
862 | while(diff) { | |
863 | if (diff & 1) { | |
864 | if (nr_sects == 0) | |
865 | goto start_new_extent; | |
866 | if ((start + nr_sects) != blk) { | |
867 | rv = blkdev_issue_discard(bdev, start, | |
868 | nr_sects, GFP_NOFS); | |
869 | if (rv) | |
870 | goto fail; | |
871 | nr_sects = 0; | |
872 | start_new_extent: | |
873 | start = blk; | |
874 | } | |
875 | nr_sects += sects_per_blk; | |
876 | } | |
877 | diff >>= 2; | |
878 | blk += sects_per_blk; | |
879 | } | |
880 | } | |
881 | if (nr_sects) { | |
882 | rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS); | |
883 | if (rv) | |
884 | goto fail; | |
885 | } | |
886 | return; | |
887 | fail: | |
888 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
889 | sdp->sd_args.ar_discard = 0; | |
890 | } | |
891 | ||
b3b94faa DT |
892 | void gfs2_rgrp_repolish_clones(struct gfs2_rgrpd *rgd) |
893 | { | |
894 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
bb8d8a6f | 895 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
896 | unsigned int x; |
897 | ||
898 | for (x = 0; x < length; x++) { | |
899 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
900 | if (!bi->bi_clone) | |
901 | continue; | |
f15ab561 SW |
902 | if (sdp->sd_args.ar_discard) |
903 | gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bi); | |
60a0b8f9 | 904 | clear_bit(GBF_FULL, &bi->bi_flags); |
b3b94faa | 905 | memcpy(bi->bi_clone + bi->bi_offset, |
feaa7bba | 906 | bi->bi_bh->b_data + bi->bi_offset, bi->bi_len); |
b3b94faa DT |
907 | } |
908 | ||
909 | spin_lock(&sdp->sd_rindex_spin); | |
cfc8b549 | 910 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa DT |
911 | spin_unlock(&sdp->sd_rindex_spin); |
912 | } | |
913 | ||
914 | /** | |
915 | * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode | |
916 | * @ip: the incore GFS2 inode structure | |
917 | * | |
918 | * Returns: the struct gfs2_alloc | |
919 | */ | |
920 | ||
921 | struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip) | |
922 | { | |
6dbd8224 SW |
923 | BUG_ON(ip->i_alloc != NULL); |
924 | ip->i_alloc = kzalloc(sizeof(struct gfs2_alloc), GFP_KERNEL); | |
925 | return ip->i_alloc; | |
b3b94faa DT |
926 | } |
927 | ||
b3b94faa DT |
928 | /** |
929 | * try_rgrp_fit - See if a given reservation will fit in a given RG | |
930 | * @rgd: the RG data | |
931 | * @al: the struct gfs2_alloc structure describing the reservation | |
932 | * | |
933 | * If there's room for the requested blocks to be allocated from the RG: | |
b3b94faa DT |
934 | * Sets the $al_rgd field in @al. |
935 | * | |
936 | * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) | |
937 | */ | |
938 | ||
939 | static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al) | |
940 | { | |
941 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
942 | int ret = 0; | |
943 | ||
09010978 | 944 | if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) |
a43a4906 SW |
945 | return 0; |
946 | ||
b3b94faa DT |
947 | spin_lock(&sdp->sd_rindex_spin); |
948 | if (rgd->rd_free_clone >= al->al_requested) { | |
949 | al->al_rgd = rgd; | |
950 | ret = 1; | |
951 | } | |
952 | spin_unlock(&sdp->sd_rindex_spin); | |
953 | ||
954 | return ret; | |
955 | } | |
956 | ||
c8cdf479 SW |
957 | /** |
958 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
959 | * @rgd: The rgrp | |
960 | * | |
961 | * Returns: The inode, if one has been found | |
962 | */ | |
963 | ||
1e19a195 SW |
964 | static struct inode *try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, |
965 | u64 skip) | |
c8cdf479 SW |
966 | { |
967 | struct inode *inode; | |
6760bdcd | 968 | u32 goal = 0, block; |
bb9bcf06 | 969 | u64 no_addr; |
5f3eae75 | 970 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
b45e41d7 | 971 | unsigned int n; |
c8cdf479 SW |
972 | |
973 | for(;;) { | |
24c73873 BP |
974 | if (goal >= rgd->rd_data) |
975 | break; | |
5f3eae75 | 976 | down_write(&sdp->sd_log_flush_lock); |
b45e41d7 | 977 | n = 1; |
6760bdcd | 978 | block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, |
b45e41d7 | 979 | GFS2_BLKST_UNLINKED, &n); |
5f3eae75 | 980 | up_write(&sdp->sd_log_flush_lock); |
6760bdcd | 981 | if (block == BFITNOENT) |
24c73873 | 982 | break; |
6760bdcd BP |
983 | /* rgblk_search can return a block < goal, so we need to |
984 | keep it marching forward. */ | |
985 | no_addr = block + rgd->rd_data0; | |
24c73873 | 986 | goal++; |
6760bdcd | 987 | if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked) |
c8cdf479 | 988 | continue; |
1e19a195 SW |
989 | if (no_addr == skip) |
990 | continue; | |
bb9bcf06 WC |
991 | *last_unlinked = no_addr; |
992 | inode = gfs2_inode_lookup(rgd->rd_sbd->sd_vfs, DT_UNKNOWN, | |
7a9f53b3 | 993 | no_addr, -1, 1); |
c8cdf479 SW |
994 | if (!IS_ERR(inode)) |
995 | return inode; | |
996 | } | |
997 | ||
998 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
999 | return NULL; | |
1000 | } | |
1001 | ||
b3b94faa DT |
1002 | /** |
1003 | * recent_rgrp_next - get next RG from "recent" list | |
1004 | * @cur_rgd: current rgrp | |
b3b94faa DT |
1005 | * |
1006 | * Returns: The next rgrp in the recent list | |
1007 | */ | |
1008 | ||
9cabcdbd | 1009 | static struct gfs2_rgrpd *recent_rgrp_next(struct gfs2_rgrpd *cur_rgd) |
b3b94faa DT |
1010 | { |
1011 | struct gfs2_sbd *sdp = cur_rgd->rd_sbd; | |
1012 | struct list_head *head; | |
1013 | struct gfs2_rgrpd *rgd; | |
1014 | ||
1015 | spin_lock(&sdp->sd_rindex_spin); | |
9cabcdbd SW |
1016 | head = &sdp->sd_rindex_mru_list; |
1017 | if (unlikely(cur_rgd->rd_list_mru.next == head)) { | |
1018 | spin_unlock(&sdp->sd_rindex_spin); | |
1019 | return NULL; | |
b3b94faa | 1020 | } |
9cabcdbd | 1021 | rgd = list_entry(cur_rgd->rd_list_mru.next, struct gfs2_rgrpd, rd_list_mru); |
b3b94faa | 1022 | spin_unlock(&sdp->sd_rindex_spin); |
b3b94faa DT |
1023 | return rgd; |
1024 | } | |
1025 | ||
b3b94faa DT |
1026 | /** |
1027 | * forward_rgrp_get - get an rgrp to try next from full list | |
1028 | * @sdp: The GFS2 superblock | |
1029 | * | |
1030 | * Returns: The rgrp to try next | |
1031 | */ | |
1032 | ||
1033 | static struct gfs2_rgrpd *forward_rgrp_get(struct gfs2_sbd *sdp) | |
1034 | { | |
1035 | struct gfs2_rgrpd *rgd; | |
1036 | unsigned int journals = gfs2_jindex_size(sdp); | |
1037 | unsigned int rg = 0, x; | |
1038 | ||
1039 | spin_lock(&sdp->sd_rindex_spin); | |
1040 | ||
1041 | rgd = sdp->sd_rindex_forward; | |
1042 | if (!rgd) { | |
1043 | if (sdp->sd_rgrps >= journals) | |
1044 | rg = sdp->sd_rgrps * sdp->sd_jdesc->jd_jid / journals; | |
1045 | ||
b8e1aabf | 1046 | for (x = 0, rgd = gfs2_rgrpd_get_first(sdp); x < rg; |
b3b94faa DT |
1047 | x++, rgd = gfs2_rgrpd_get_next(rgd)) |
1048 | /* Do Nothing */; | |
1049 | ||
1050 | sdp->sd_rindex_forward = rgd; | |
1051 | } | |
1052 | ||
1053 | spin_unlock(&sdp->sd_rindex_spin); | |
1054 | ||
1055 | return rgd; | |
1056 | } | |
1057 | ||
1058 | /** | |
1059 | * forward_rgrp_set - set the forward rgrp pointer | |
1060 | * @sdp: the filesystem | |
1061 | * @rgd: The new forward rgrp | |
1062 | * | |
1063 | */ | |
1064 | ||
1065 | static void forward_rgrp_set(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd) | |
1066 | { | |
1067 | spin_lock(&sdp->sd_rindex_spin); | |
1068 | sdp->sd_rindex_forward = rgd; | |
1069 | spin_unlock(&sdp->sd_rindex_spin); | |
1070 | } | |
1071 | ||
1072 | /** | |
1073 | * get_local_rgrp - Choose and lock a rgrp for allocation | |
1074 | * @ip: the inode to reserve space for | |
1075 | * @rgp: the chosen and locked rgrp | |
1076 | * | |
1077 | * Try to acquire rgrp in way which avoids contending with others. | |
1078 | * | |
1079 | * Returns: errno | |
1080 | */ | |
1081 | ||
c8cdf479 | 1082 | static struct inode *get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked) |
b3b94faa | 1083 | { |
c8cdf479 | 1084 | struct inode *inode = NULL; |
feaa7bba | 1085 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa | 1086 | struct gfs2_rgrpd *rgd, *begin = NULL; |
6dbd8224 | 1087 | struct gfs2_alloc *al = ip->i_alloc; |
b3b94faa DT |
1088 | int flags = LM_FLAG_TRY; |
1089 | int skipped = 0; | |
1090 | int loops = 0; | |
292c8c14 | 1091 | int error, rg_locked; |
b3b94faa | 1092 | |
9cabcdbd | 1093 | rgd = gfs2_blk2rgrpd(sdp, ip->i_goal); |
b3b94faa DT |
1094 | |
1095 | while (rgd) { | |
292c8c14 AD |
1096 | rg_locked = 0; |
1097 | ||
1098 | if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) { | |
1099 | rg_locked = 1; | |
1100 | error = 0; | |
1101 | } else { | |
1102 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, | |
1103 | LM_FLAG_TRY, &al->al_rgd_gh); | |
1104 | } | |
b3b94faa DT |
1105 | switch (error) { |
1106 | case 0: | |
1107 | if (try_rgrp_fit(rgd, al)) | |
1108 | goto out; | |
c8cdf479 | 1109 | if (rgd->rd_flags & GFS2_RDF_CHECK) |
1e19a195 | 1110 | inode = try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr); |
292c8c14 AD |
1111 | if (!rg_locked) |
1112 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
c8cdf479 SW |
1113 | if (inode) |
1114 | return inode; | |
9cabcdbd | 1115 | /* fall through */ |
b3b94faa | 1116 | case GLR_TRYFAILED: |
9cabcdbd | 1117 | rgd = recent_rgrp_next(rgd); |
b3b94faa DT |
1118 | break; |
1119 | ||
1120 | default: | |
c8cdf479 | 1121 | return ERR_PTR(error); |
b3b94faa DT |
1122 | } |
1123 | } | |
1124 | ||
1125 | /* Go through full list of rgrps */ | |
1126 | ||
1127 | begin = rgd = forward_rgrp_get(sdp); | |
1128 | ||
1129 | for (;;) { | |
292c8c14 AD |
1130 | rg_locked = 0; |
1131 | ||
1132 | if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) { | |
1133 | rg_locked = 1; | |
1134 | error = 0; | |
1135 | } else { | |
1136 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, flags, | |
1137 | &al->al_rgd_gh); | |
1138 | } | |
b3b94faa DT |
1139 | switch (error) { |
1140 | case 0: | |
1141 | if (try_rgrp_fit(rgd, al)) | |
1142 | goto out; | |
c8cdf479 | 1143 | if (rgd->rd_flags & GFS2_RDF_CHECK) |
1e19a195 | 1144 | inode = try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr); |
292c8c14 AD |
1145 | if (!rg_locked) |
1146 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
c8cdf479 SW |
1147 | if (inode) |
1148 | return inode; | |
b3b94faa DT |
1149 | break; |
1150 | ||
1151 | case GLR_TRYFAILED: | |
1152 | skipped++; | |
1153 | break; | |
1154 | ||
1155 | default: | |
c8cdf479 | 1156 | return ERR_PTR(error); |
b3b94faa DT |
1157 | } |
1158 | ||
1159 | rgd = gfs2_rgrpd_get_next(rgd); | |
1160 | if (!rgd) | |
1161 | rgd = gfs2_rgrpd_get_first(sdp); | |
1162 | ||
1163 | if (rgd == begin) { | |
172e045a | 1164 | if (++loops >= 3) |
c8cdf479 | 1165 | return ERR_PTR(-ENOSPC); |
172e045a BM |
1166 | if (!skipped) |
1167 | loops++; | |
b3b94faa | 1168 | flags = 0; |
172e045a BM |
1169 | if (loops == 2) |
1170 | gfs2_log_flush(sdp, NULL); | |
b3b94faa DT |
1171 | } |
1172 | } | |
1173 | ||
feaa7bba | 1174 | out: |
b3b94faa | 1175 | if (begin) { |
9cabcdbd SW |
1176 | spin_lock(&sdp->sd_rindex_spin); |
1177 | list_move(&rgd->rd_list_mru, &sdp->sd_rindex_mru_list); | |
1178 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa DT |
1179 | rgd = gfs2_rgrpd_get_next(rgd); |
1180 | if (!rgd) | |
1181 | rgd = gfs2_rgrpd_get_first(sdp); | |
1182 | forward_rgrp_set(sdp, rgd); | |
1183 | } | |
1184 | ||
c8cdf479 | 1185 | return NULL; |
b3b94faa DT |
1186 | } |
1187 | ||
1188 | /** | |
1189 | * gfs2_inplace_reserve_i - Reserve space in the filesystem | |
1190 | * @ip: the inode to reserve space for | |
1191 | * | |
1192 | * Returns: errno | |
1193 | */ | |
1194 | ||
1195 | int gfs2_inplace_reserve_i(struct gfs2_inode *ip, char *file, unsigned int line) | |
1196 | { | |
feaa7bba | 1197 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
6dbd8224 | 1198 | struct gfs2_alloc *al = ip->i_alloc; |
c8cdf479 | 1199 | struct inode *inode; |
7ae8fa84 | 1200 | int error = 0; |
6760bdcd | 1201 | u64 last_unlinked = NO_BLOCK; |
b3b94faa DT |
1202 | |
1203 | if (gfs2_assert_warn(sdp, al->al_requested)) | |
1204 | return -EINVAL; | |
1205 | ||
c8cdf479 | 1206 | try_again: |
7ae8fa84 RP |
1207 | /* We need to hold the rindex unless the inode we're using is |
1208 | the rindex itself, in which case it's already held. */ | |
1209 | if (ip != GFS2_I(sdp->sd_rindex)) | |
1210 | error = gfs2_rindex_hold(sdp, &al->al_ri_gh); | |
1211 | else if (!sdp->sd_rgrps) /* We may not have the rindex read in, so: */ | |
6c53267f | 1212 | error = gfs2_ri_update_special(ip); |
7ae8fa84 | 1213 | |
b3b94faa DT |
1214 | if (error) |
1215 | return error; | |
1216 | ||
c8cdf479 SW |
1217 | inode = get_local_rgrp(ip, &last_unlinked); |
1218 | if (inode) { | |
7ae8fa84 RP |
1219 | if (ip != GFS2_I(sdp->sd_rindex)) |
1220 | gfs2_glock_dq_uninit(&al->al_ri_gh); | |
c8cdf479 SW |
1221 | if (IS_ERR(inode)) |
1222 | return PTR_ERR(inode); | |
1223 | iput(inode); | |
1224 | gfs2_log_flush(sdp, NULL); | |
1225 | goto try_again; | |
b3b94faa DT |
1226 | } |
1227 | ||
1228 | al->al_file = file; | |
1229 | al->al_line = line; | |
1230 | ||
1231 | return 0; | |
1232 | } | |
1233 | ||
1234 | /** | |
1235 | * gfs2_inplace_release - release an inplace reservation | |
1236 | * @ip: the inode the reservation was taken out on | |
1237 | * | |
1238 | * Release a reservation made by gfs2_inplace_reserve(). | |
1239 | */ | |
1240 | ||
1241 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1242 | { | |
feaa7bba | 1243 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
6dbd8224 | 1244 | struct gfs2_alloc *al = ip->i_alloc; |
b3b94faa DT |
1245 | |
1246 | if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1) | |
1247 | fs_warn(sdp, "al_alloced = %u, al_requested = %u " | |
1248 | "al_file = %s, al_line = %u\n", | |
1249 | al->al_alloced, al->al_requested, al->al_file, | |
1250 | al->al_line); | |
1251 | ||
1252 | al->al_rgd = NULL; | |
292c8c14 AD |
1253 | if (al->al_rgd_gh.gh_gl) |
1254 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
7ae8fa84 RP |
1255 | if (ip != GFS2_I(sdp->sd_rindex)) |
1256 | gfs2_glock_dq_uninit(&al->al_ri_gh); | |
b3b94faa DT |
1257 | } |
1258 | ||
1259 | /** | |
1260 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1261 | * @rgd: the resource group holding the block | |
1262 | * @block: the block number | |
1263 | * | |
1264 | * Returns: The block type (GFS2_BLKST_*) | |
1265 | */ | |
1266 | ||
cd915493 | 1267 | unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa DT |
1268 | { |
1269 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1270 | u32 length, rgrp_block, buf_block; |
b3b94faa DT |
1271 | unsigned int buf; |
1272 | unsigned char type; | |
1273 | ||
bb8d8a6f SW |
1274 | length = rgd->rd_length; |
1275 | rgrp_block = block - rgd->rd_data0; | |
b3b94faa DT |
1276 | |
1277 | for (buf = 0; buf < length; buf++) { | |
1278 | bi = rgd->rd_bits + buf; | |
1279 | if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1280 | break; | |
1281 | } | |
1282 | ||
1283 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1284 | buf_block = rgrp_block - bi->bi_start * GFS2_NBBY; | |
1285 | ||
feaa7bba | 1286 | type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset, |
b3b94faa DT |
1287 | bi->bi_len, buf_block); |
1288 | ||
1289 | return type; | |
1290 | } | |
1291 | ||
1292 | /** | |
1293 | * rgblk_search - find a block in @old_state, change allocation | |
1294 | * state to @new_state | |
1295 | * @rgd: the resource group descriptor | |
1296 | * @goal: the goal block within the RG (start here to search for avail block) | |
1297 | * @old_state: GFS2_BLKST_XXX the before-allocation state to find | |
1298 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
b45e41d7 | 1299 | * @n: The extent length |
b3b94faa DT |
1300 | * |
1301 | * Walk rgrp's bitmap to find bits that represent a block in @old_state. | |
1302 | * Add the found bitmap buffer to the transaction. | |
1303 | * Set the found bits to @new_state to change block's allocation state. | |
1304 | * | |
1305 | * This function never fails, because we wouldn't call it unless we | |
1306 | * know (from reservation results, etc.) that a block is available. | |
1307 | * | |
1308 | * Scope of @goal and returned block is just within rgrp, not the whole | |
1309 | * filesystem. | |
1310 | * | |
1311 | * Returns: the block number allocated | |
1312 | */ | |
1313 | ||
cd915493 | 1314 | static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, |
b45e41d7 SW |
1315 | unsigned char old_state, unsigned char new_state, |
1316 | unsigned int *n) | |
b3b94faa DT |
1317 | { |
1318 | struct gfs2_bitmap *bi = NULL; | |
b45e41d7 | 1319 | const u32 length = rgd->rd_length; |
60a0b8f9 | 1320 | u32 blk = BFITNOENT; |
b3b94faa | 1321 | unsigned int buf, x; |
b45e41d7 | 1322 | const unsigned int elen = *n; |
60a0b8f9 | 1323 | const u8 *buffer = NULL; |
b3b94faa | 1324 | |
b45e41d7 | 1325 | *n = 0; |
b3b94faa DT |
1326 | /* Find bitmap block that contains bits for goal block */ |
1327 | for (buf = 0; buf < length; buf++) { | |
1328 | bi = rgd->rd_bits + buf; | |
60a0b8f9 SW |
1329 | /* Convert scope of "goal" from rgrp-wide to within found bit block */ |
1330 | if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { | |
1331 | goal -= bi->bi_start * GFS2_NBBY; | |
1332 | goto do_search; | |
1333 | } | |
b3b94faa | 1334 | } |
60a0b8f9 SW |
1335 | buf = 0; |
1336 | goal = 0; | |
b3b94faa | 1337 | |
60a0b8f9 | 1338 | do_search: |
b3b94faa DT |
1339 | /* Search (up to entire) bitmap in this rgrp for allocatable block. |
1340 | "x <= length", instead of "x < length", because we typically start | |
1341 | the search in the middle of a bit block, but if we can't find an | |
1342 | allocatable block anywhere else, we want to be able wrap around and | |
1343 | search in the first part of our first-searched bit block. */ | |
1344 | for (x = 0; x <= length; x++) { | |
60a0b8f9 SW |
1345 | bi = rgd->rd_bits + buf; |
1346 | ||
1347 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
1348 | (old_state == GFS2_BLKST_FREE)) | |
1349 | goto skip; | |
1350 | ||
5f3eae75 BP |
1351 | /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone |
1352 | bitmaps, so we must search the originals for that. */ | |
b45e41d7 | 1353 | buffer = bi->bi_bh->b_data + bi->bi_offset; |
5f3eae75 | 1354 | if (old_state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
110acf38 SW |
1355 | buffer = bi->bi_clone + bi->bi_offset; |
1356 | ||
1357 | blk = gfs2_bitfit(buffer, bi->bi_len, goal, old_state); | |
b3b94faa DT |
1358 | if (blk != BFITNOENT) |
1359 | break; | |
1360 | ||
60a0b8f9 SW |
1361 | if ((goal == 0) && (old_state == GFS2_BLKST_FREE)) |
1362 | set_bit(GBF_FULL, &bi->bi_flags); | |
1363 | ||
b3b94faa | 1364 | /* Try next bitmap block (wrap back to rgrp header if at end) */ |
60a0b8f9 SW |
1365 | skip: |
1366 | buf++; | |
1367 | buf %= length; | |
b3b94faa DT |
1368 | goal = 0; |
1369 | } | |
1370 | ||
60a0b8f9 SW |
1371 | if (blk == BFITNOENT) |
1372 | return blk; | |
1373 | *n = 1; | |
1374 | if (old_state == new_state) | |
1375 | goto out; | |
1376 | ||
1377 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); | |
1378 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, | |
1379 | bi->bi_len, blk, new_state); | |
1380 | goal = blk; | |
1381 | while (*n < elen) { | |
1382 | goal++; | |
1383 | if (goal >= (bi->bi_len * GFS2_NBBY)) | |
1384 | break; | |
1385 | if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) != | |
1386 | GFS2_BLKST_FREE) | |
1387 | break; | |
b45e41d7 | 1388 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, |
60a0b8f9 SW |
1389 | bi->bi_len, goal, new_state); |
1390 | (*n)++; | |
c8cdf479 | 1391 | } |
60a0b8f9 SW |
1392 | out: |
1393 | return (bi->bi_start * GFS2_NBBY) + blk; | |
b3b94faa DT |
1394 | } |
1395 | ||
1396 | /** | |
1397 | * rgblk_free - Change alloc state of given block(s) | |
1398 | * @sdp: the filesystem | |
1399 | * @bstart: the start of a run of blocks to free | |
1400 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1401 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1402 | * | |
1403 | * Returns: Resource group containing the block(s) | |
1404 | */ | |
1405 | ||
cd915493 SW |
1406 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
1407 | u32 blen, unsigned char new_state) | |
b3b94faa DT |
1408 | { |
1409 | struct gfs2_rgrpd *rgd; | |
1410 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1411 | u32 length, rgrp_blk, buf_blk; |
b3b94faa DT |
1412 | unsigned int buf; |
1413 | ||
1414 | rgd = gfs2_blk2rgrpd(sdp, bstart); | |
1415 | if (!rgd) { | |
1416 | if (gfs2_consist(sdp)) | |
382066da | 1417 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
1418 | return NULL; |
1419 | } | |
1420 | ||
bb8d8a6f | 1421 | length = rgd->rd_length; |
b3b94faa | 1422 | |
bb8d8a6f | 1423 | rgrp_blk = bstart - rgd->rd_data0; |
b3b94faa DT |
1424 | |
1425 | while (blen--) { | |
1426 | for (buf = 0; buf < length; buf++) { | |
1427 | bi = rgd->rd_bits + buf; | |
1428 | if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1429 | break; | |
1430 | } | |
1431 | ||
1432 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1433 | ||
1434 | buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY; | |
1435 | rgrp_blk++; | |
1436 | ||
1437 | if (!bi->bi_clone) { | |
1438 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
dd894be8 | 1439 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1440 | memcpy(bi->bi_clone + bi->bi_offset, |
1441 | bi->bi_bh->b_data + bi->bi_offset, | |
1442 | bi->bi_len); | |
1443 | } | |
d4e9c4c3 | 1444 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
b45e41d7 | 1445 | gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset, |
b3b94faa DT |
1446 | bi->bi_len, buf_blk, new_state); |
1447 | } | |
1448 | ||
1449 | return rgd; | |
1450 | } | |
1451 | ||
1452 | /** | |
09010978 SW |
1453 | * gfs2_rgrp_dump - print out an rgrp |
1454 | * @seq: The iterator | |
1455 | * @gl: The glock in question | |
1456 | * | |
1457 | */ | |
1458 | ||
1459 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
1460 | { | |
1461 | const struct gfs2_rgrpd *rgd = gl->gl_object; | |
1462 | if (rgd == NULL) | |
1463 | return 0; | |
1464 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n", | |
1465 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, | |
1466 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes); | |
1467 | return 0; | |
1468 | } | |
1469 | ||
1470 | /** | |
1471 | * gfs2_alloc_block - Allocate one or more blocks | |
1639431a | 1472 | * @ip: the inode to allocate the block for |
09010978 SW |
1473 | * @bn: Used to return the starting block number |
1474 | * @n: requested number of blocks/extent length (value/result) | |
b3b94faa | 1475 | * |
09010978 | 1476 | * Returns: 0 or error |
b3b94faa DT |
1477 | */ |
1478 | ||
09010978 | 1479 | int gfs2_alloc_block(struct gfs2_inode *ip, u64 *bn, unsigned int *n) |
b3b94faa | 1480 | { |
feaa7bba | 1481 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 1482 | struct buffer_head *dibh; |
6dbd8224 | 1483 | struct gfs2_alloc *al = ip->i_alloc; |
b3b94faa | 1484 | struct gfs2_rgrpd *rgd = al->al_rgd; |
cd915493 SW |
1485 | u32 goal, blk; |
1486 | u64 block; | |
d9ba7615 | 1487 | int error; |
b3b94faa | 1488 | |
ce276b06 SW |
1489 | if (rgrp_contains_block(rgd, ip->i_goal)) |
1490 | goal = ip->i_goal - rgd->rd_data0; | |
b3b94faa | 1491 | else |
ac576cc5 | 1492 | goal = rgd->rd_last_alloc; |
b3b94faa | 1493 | |
b45e41d7 | 1494 | blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED, n); |
09010978 SW |
1495 | |
1496 | /* Since all blocks are reserved in advance, this shouldn't happen */ | |
1497 | if (blk == BFITNOENT) | |
1498 | goto rgrp_error; | |
b3b94faa | 1499 | |
b45e41d7 | 1500 | rgd->rd_last_alloc = blk; |
bb8d8a6f | 1501 | block = rgd->rd_data0 + blk; |
ce276b06 | 1502 | ip->i_goal = block; |
d9ba7615 SW |
1503 | error = gfs2_meta_inode_buffer(ip, &dibh); |
1504 | if (error == 0) { | |
1505 | struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data; | |
1506 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
1507 | di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_goal); | |
1508 | brelse(dibh); | |
1509 | } | |
09010978 SW |
1510 | if (rgd->rd_free < *n) |
1511 | goto rgrp_error; | |
1512 | ||
cfc8b549 | 1513 | rgd->rd_free -= *n; |
b3b94faa | 1514 | |
d4e9c4c3 | 1515 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1516 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 1517 | |
b45e41d7 | 1518 | al->al_alloced += *n; |
b3b94faa | 1519 | |
ad99f777 | 1520 | gfs2_statfs_change(sdp, 0, -(s64)*n, 0); |
b45e41d7 | 1521 | gfs2_quota_change(ip, *n, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
1522 | |
1523 | spin_lock(&sdp->sd_rindex_spin); | |
b45e41d7 | 1524 | rgd->rd_free_clone -= *n; |
b3b94faa | 1525 | spin_unlock(&sdp->sd_rindex_spin); |
63997775 | 1526 | trace_gfs2_block_alloc(ip, block, *n, GFS2_BLKST_USED); |
09010978 SW |
1527 | *bn = block; |
1528 | return 0; | |
1529 | ||
1530 | rgrp_error: | |
1531 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
1532 | (unsigned long long)rgd->rd_addr); | |
1533 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); | |
1534 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
1535 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
1536 | return -EIO; | |
b3b94faa DT |
1537 | } |
1538 | ||
1539 | /** | |
1540 | * gfs2_alloc_di - Allocate a dinode | |
1541 | * @dip: the directory that the inode is going in | |
1542 | * | |
1543 | * Returns: the block allocated | |
1544 | */ | |
1545 | ||
4340fe62 | 1546 | u64 gfs2_alloc_di(struct gfs2_inode *dip, u64 *generation) |
b3b94faa | 1547 | { |
feaa7bba | 1548 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
6dbd8224 | 1549 | struct gfs2_alloc *al = dip->i_alloc; |
b3b94faa | 1550 | struct gfs2_rgrpd *rgd = al->al_rgd; |
4340fe62 SW |
1551 | u32 blk; |
1552 | u64 block; | |
b45e41d7 | 1553 | unsigned int n = 1; |
b3b94faa | 1554 | |
ac576cc5 | 1555 | blk = rgblk_search(rgd, rgd->rd_last_alloc, |
b45e41d7 | 1556 | GFS2_BLKST_FREE, GFS2_BLKST_DINODE, &n); |
6eefaf61 | 1557 | BUG_ON(blk == BFITNOENT); |
b3b94faa | 1558 | |
ac576cc5 | 1559 | rgd->rd_last_alloc = blk; |
b3b94faa | 1560 | |
bb8d8a6f | 1561 | block = rgd->rd_data0 + blk; |
b3b94faa | 1562 | |
cfc8b549 SW |
1563 | gfs2_assert_withdraw(sdp, rgd->rd_free); |
1564 | rgd->rd_free--; | |
73f74948 | 1565 | rgd->rd_dinodes++; |
d8b71f73 | 1566 | *generation = rgd->rd_igeneration++; |
d4e9c4c3 | 1567 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1568 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1569 | |
1570 | al->al_alloced++; | |
1571 | ||
1572 | gfs2_statfs_change(sdp, 0, -1, +1); | |
5731be53 | 1573 | gfs2_trans_add_unrevoke(sdp, block, 1); |
b3b94faa DT |
1574 | |
1575 | spin_lock(&sdp->sd_rindex_spin); | |
1576 | rgd->rd_free_clone--; | |
1577 | spin_unlock(&sdp->sd_rindex_spin); | |
63997775 | 1578 | trace_gfs2_block_alloc(dip, block, 1, GFS2_BLKST_DINODE); |
b3b94faa DT |
1579 | return block; |
1580 | } | |
1581 | ||
1582 | /** | |
1583 | * gfs2_free_data - free a contiguous run of data block(s) | |
1584 | * @ip: the inode these blocks are being freed from | |
1585 | * @bstart: first block of a run of contiguous blocks | |
1586 | * @blen: the length of the block run | |
1587 | * | |
1588 | */ | |
1589 | ||
cd915493 | 1590 | void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen) |
b3b94faa | 1591 | { |
feaa7bba | 1592 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1593 | struct gfs2_rgrpd *rgd; |
1594 | ||
1595 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
1596 | if (!rgd) | |
1597 | return; | |
63997775 | 1598 | trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 1599 | rgd->rd_free += blen; |
b3b94faa | 1600 | |
d4e9c4c3 | 1601 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1602 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1603 | |
1604 | gfs2_trans_add_rg(rgd); | |
1605 | ||
1606 | gfs2_statfs_change(sdp, 0, +blen, 0); | |
2933f925 | 1607 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
1608 | } |
1609 | ||
1610 | /** | |
1611 | * gfs2_free_meta - free a contiguous run of data block(s) | |
1612 | * @ip: the inode these blocks are being freed from | |
1613 | * @bstart: first block of a run of contiguous blocks | |
1614 | * @blen: the length of the block run | |
1615 | * | |
1616 | */ | |
1617 | ||
cd915493 | 1618 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) |
b3b94faa | 1619 | { |
feaa7bba | 1620 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1621 | struct gfs2_rgrpd *rgd; |
1622 | ||
1623 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
1624 | if (!rgd) | |
1625 | return; | |
63997775 | 1626 | trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 1627 | rgd->rd_free += blen; |
b3b94faa | 1628 | |
d4e9c4c3 | 1629 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1630 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1631 | |
1632 | gfs2_trans_add_rg(rgd); | |
1633 | ||
1634 | gfs2_statfs_change(sdp, 0, +blen, 0); | |
2933f925 | 1635 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
1636 | gfs2_meta_wipe(ip, bstart, blen); |
1637 | } | |
1638 | ||
feaa7bba SW |
1639 | void gfs2_unlink_di(struct inode *inode) |
1640 | { | |
1641 | struct gfs2_inode *ip = GFS2_I(inode); | |
1642 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1643 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 1644 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
1645 | |
1646 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
1647 | if (!rgd) | |
1648 | return; | |
63997775 | 1649 | trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED); |
feaa7bba | 1650 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1651 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
feaa7bba SW |
1652 | gfs2_trans_add_rg(rgd); |
1653 | } | |
1654 | ||
cd915493 | 1655 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
1656 | { |
1657 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1658 | struct gfs2_rgrpd *tmp_rgd; | |
1659 | ||
1660 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
1661 | if (!tmp_rgd) | |
1662 | return; | |
1663 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
1664 | ||
73f74948 | 1665 | if (!rgd->rd_dinodes) |
b3b94faa | 1666 | gfs2_consist_rgrpd(rgd); |
73f74948 | 1667 | rgd->rd_dinodes--; |
cfc8b549 | 1668 | rgd->rd_free++; |
b3b94faa | 1669 | |
d4e9c4c3 | 1670 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1671 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1672 | |
1673 | gfs2_statfs_change(sdp, 0, +1, -1); | |
1674 | gfs2_trans_add_rg(rgd); | |
1675 | } | |
1676 | ||
b3b94faa DT |
1677 | |
1678 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
1679 | { | |
dbb7cae2 | 1680 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
63997775 | 1681 | trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 1682 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 1683 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
1684 | } |
1685 | ||
1686 | /** | |
1687 | * gfs2_rlist_add - add a RG to a list of RGs | |
1688 | * @sdp: the filesystem | |
1689 | * @rlist: the list of resource groups | |
1690 | * @block: the block | |
1691 | * | |
1692 | * Figure out what RG a block belongs to and add that RG to the list | |
1693 | * | |
1694 | * FIXME: Don't use NOFAIL | |
1695 | * | |
1696 | */ | |
1697 | ||
1698 | void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist, | |
cd915493 | 1699 | u64 block) |
b3b94faa DT |
1700 | { |
1701 | struct gfs2_rgrpd *rgd; | |
1702 | struct gfs2_rgrpd **tmp; | |
1703 | unsigned int new_space; | |
1704 | unsigned int x; | |
1705 | ||
1706 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
1707 | return; | |
1708 | ||
1709 | rgd = gfs2_blk2rgrpd(sdp, block); | |
1710 | if (!rgd) { | |
1711 | if (gfs2_consist(sdp)) | |
382066da | 1712 | fs_err(sdp, "block = %llu\n", (unsigned long long)block); |
b3b94faa DT |
1713 | return; |
1714 | } | |
1715 | ||
1716 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1717 | if (rlist->rl_rgd[x] == rgd) | |
1718 | return; | |
1719 | ||
1720 | if (rlist->rl_rgrps == rlist->rl_space) { | |
1721 | new_space = rlist->rl_space + 10; | |
1722 | ||
1723 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 1724 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1725 | |
1726 | if (rlist->rl_rgd) { | |
1727 | memcpy(tmp, rlist->rl_rgd, | |
1728 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
1729 | kfree(rlist->rl_rgd); | |
1730 | } | |
1731 | ||
1732 | rlist->rl_space = new_space; | |
1733 | rlist->rl_rgd = tmp; | |
1734 | } | |
1735 | ||
1736 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
1737 | } | |
1738 | ||
1739 | /** | |
1740 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
1741 | * and initialize an array of glock holders for them | |
1742 | * @rlist: the list of resource groups | |
1743 | * @state: the lock state to acquire the RG lock in | |
1744 | * @flags: the modifier flags for the holder structures | |
1745 | * | |
1746 | * FIXME: Don't use NOFAIL | |
1747 | * | |
1748 | */ | |
1749 | ||
fe6c991c | 1750 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
1751 | { |
1752 | unsigned int x; | |
1753 | ||
1754 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 1755 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1756 | for (x = 0; x < rlist->rl_rgrps; x++) |
1757 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 1758 | state, 0, |
b3b94faa DT |
1759 | &rlist->rl_ghs[x]); |
1760 | } | |
1761 | ||
1762 | /** | |
1763 | * gfs2_rlist_free - free a resource group list | |
1764 | * @list: the list of resource groups | |
1765 | * | |
1766 | */ | |
1767 | ||
1768 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
1769 | { | |
1770 | unsigned int x; | |
1771 | ||
1772 | kfree(rlist->rl_rgd); | |
1773 | ||
1774 | if (rlist->rl_ghs) { | |
1775 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1776 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
1777 | kfree(rlist->rl_ghs); | |
1778 | } | |
1779 | } | |
1780 |