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> |
7c9ca621 | 18 | #include <linux/rbtree.h> |
b3b94faa DT |
19 | |
20 | #include "gfs2.h" | |
5c676f6d | 21 | #include "incore.h" |
b3b94faa DT |
22 | #include "glock.h" |
23 | #include "glops.h" | |
b3b94faa DT |
24 | #include "lops.h" |
25 | #include "meta_io.h" | |
26 | #include "quota.h" | |
27 | #include "rgrp.h" | |
28 | #include "super.h" | |
29 | #include "trans.h" | |
5c676f6d | 30 | #include "util.h" |
172e045a | 31 | #include "log.h" |
c8cdf479 | 32 | #include "inode.h" |
63997775 | 33 | #include "trace_gfs2.h" |
b3b94faa | 34 | |
2c1e52aa | 35 | #define BFITNOENT ((u32)~0) |
6760bdcd | 36 | #define NO_BLOCK ((u64)~0) |
88c8ab1f | 37 | |
1f466a47 BP |
38 | #if BITS_PER_LONG == 32 |
39 | #define LBITMASK (0x55555555UL) | |
40 | #define LBITSKIP55 (0x55555555UL) | |
41 | #define LBITSKIP00 (0x00000000UL) | |
42 | #else | |
43 | #define LBITMASK (0x5555555555555555UL) | |
44 | #define LBITSKIP55 (0x5555555555555555UL) | |
45 | #define LBITSKIP00 (0x0000000000000000UL) | |
46 | #endif | |
47 | ||
88c8ab1f SW |
48 | /* |
49 | * These routines are used by the resource group routines (rgrp.c) | |
50 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
51 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
52 | * | |
53 | * 0 = Free | |
54 | * 1 = Used (not metadata) | |
55 | * 2 = Unlinked (still in use) inode | |
56 | * 3 = Used (metadata) | |
88c8ab1f SW |
57 | */ |
58 | ||
59 | static const char valid_change[16] = { | |
60 | /* current */ | |
feaa7bba | 61 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 62 | /* e */ 1, 0, 0, 0, |
feaa7bba | 63 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
64 | 1, 0, 0, 0 |
65 | }; | |
66 | ||
c8cdf479 | 67 | static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, |
b45e41d7 SW |
68 | unsigned char old_state, unsigned char new_state, |
69 | unsigned int *n); | |
c8cdf479 | 70 | |
88c8ab1f SW |
71 | /** |
72 | * gfs2_setbit - Set a bit in the bitmaps | |
73 | * @buffer: the buffer that holds the bitmaps | |
74 | * @buflen: the length (in bytes) of the buffer | |
75 | * @block: the block to set | |
76 | * @new_state: the new state of the block | |
77 | * | |
78 | */ | |
79 | ||
b45e41d7 SW |
80 | static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1, |
81 | unsigned char *buf2, unsigned int offset, | |
95c8e17f | 82 | struct gfs2_bitmap *bi, u32 block, |
b45e41d7 | 83 | unsigned char new_state) |
88c8ab1f | 84 | { |
b45e41d7 | 85 | unsigned char *byte1, *byte2, *end, cur_state; |
95c8e17f | 86 | unsigned int buflen = bi->bi_len; |
b45e41d7 | 87 | const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 88 | |
b45e41d7 SW |
89 | byte1 = buf1 + offset + (block / GFS2_NBBY); |
90 | end = buf1 + offset + buflen; | |
88c8ab1f | 91 | |
b45e41d7 | 92 | BUG_ON(byte1 >= end); |
88c8ab1f | 93 | |
b45e41d7 | 94 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 95 | |
b45e41d7 | 96 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
95c8e17f BP |
97 | printk(KERN_WARNING "GFS2: buf_blk = 0x%llx old_state=%d, " |
98 | "new_state=%d\n", | |
99 | (unsigned long long)block, cur_state, new_state); | |
100 | printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%lx\n", | |
101 | (unsigned long long)rgd->rd_addr, | |
102 | (unsigned long)bi->bi_start); | |
103 | printk(KERN_WARNING "GFS2: bi_offset=0x%lx bi_len=0x%lx\n", | |
104 | (unsigned long)bi->bi_offset, | |
105 | (unsigned long)bi->bi_len); | |
106 | dump_stack(); | |
88c8ab1f | 107 | gfs2_consist_rgrpd(rgd); |
b45e41d7 SW |
108 | return; |
109 | } | |
110 | *byte1 ^= (cur_state ^ new_state) << bit; | |
111 | ||
112 | if (buf2) { | |
113 | byte2 = buf2 + offset + (block / GFS2_NBBY); | |
114 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; | |
115 | *byte2 ^= (cur_state ^ new_state) << bit; | |
116 | } | |
88c8ab1f SW |
117 | } |
118 | ||
119 | /** | |
120 | * gfs2_testbit - test a bit in the bitmaps | |
121 | * @buffer: the buffer that holds the bitmaps | |
122 | * @buflen: the length (in bytes) of the buffer | |
123 | * @block: the block to read | |
124 | * | |
125 | */ | |
126 | ||
b45e41d7 SW |
127 | static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd, |
128 | const unsigned char *buffer, | |
129 | unsigned int buflen, u32 block) | |
88c8ab1f | 130 | { |
b45e41d7 SW |
131 | const unsigned char *byte, *end; |
132 | unsigned char cur_state; | |
88c8ab1f SW |
133 | unsigned int bit; |
134 | ||
135 | byte = buffer + (block / GFS2_NBBY); | |
136 | bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; | |
137 | end = buffer + buflen; | |
138 | ||
139 | gfs2_assert(rgd->rd_sbd, byte < end); | |
140 | ||
141 | cur_state = (*byte >> bit) & GFS2_BIT_MASK; | |
142 | ||
143 | return cur_state; | |
144 | } | |
145 | ||
223b2b88 SW |
146 | /** |
147 | * gfs2_bit_search | |
148 | * @ptr: Pointer to bitmap data | |
149 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
150 | * @state: The state we are searching for | |
151 | * | |
152 | * We xor the bitmap data with a patter which is the bitwise opposite | |
153 | * of what we are looking for, this gives rise to a pattern of ones | |
154 | * wherever there is a match. Since we have two bits per entry, we | |
155 | * take this pattern, shift it down by one place and then and it with | |
156 | * the original. All the even bit positions (0,2,4, etc) then represent | |
157 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
158 | * odd bit positions. | |
159 | * | |
160 | * This allows searching of a whole u64 at once (32 blocks) with a | |
161 | * single test (on 64 bit arches). | |
162 | */ | |
163 | ||
164 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
165 | { | |
166 | u64 tmp; | |
167 | static const u64 search[] = { | |
075ac448 HE |
168 | [0] = 0xffffffffffffffffULL, |
169 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
170 | [2] = 0x5555555555555555ULL, | |
171 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
172 | }; |
173 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
174 | tmp &= (tmp >> 1); | |
175 | tmp &= mask; | |
176 | return tmp; | |
177 | } | |
178 | ||
88c8ab1f SW |
179 | /** |
180 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
181 | * a block in a given allocation state. | |
182 | * @buffer: the buffer that holds the bitmaps | |
223b2b88 | 183 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 184 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 185 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
186 | * |
187 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
188 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
189 | * beginning of a bitmap block buffer, skipping any header structures, but |
190 | * headers are always a multiple of 64 bits long so that the buffer is | |
191 | * always aligned to a 64 bit boundary. | |
192 | * | |
193 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 194 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 195 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
196 | * |
197 | * Return: the block number (bitmap buffer scope) that was found | |
198 | */ | |
199 | ||
02ab1721 HE |
200 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
201 | u32 goal, u8 state) | |
88c8ab1f | 202 | { |
223b2b88 SW |
203 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
204 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
205 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
206 | u64 tmp; | |
075ac448 | 207 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
208 | u32 bit; |
209 | ||
210 | BUG_ON(state > 3); | |
211 | ||
212 | /* Mask off bits we don't care about at the start of the search */ | |
213 | mask <<= spoint; | |
214 | tmp = gfs2_bit_search(ptr, mask, state); | |
215 | ptr++; | |
216 | while(tmp == 0 && ptr < end) { | |
075ac448 | 217 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 218 | ptr++; |
1f466a47 | 219 | } |
223b2b88 SW |
220 | /* Mask off any bits which are more than len bytes from the start */ |
221 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
222 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
223 | /* Didn't find anything, so return */ | |
224 | if (tmp == 0) | |
225 | return BFITNOENT; | |
226 | ptr--; | |
d8bd504a | 227 | bit = __ffs64(tmp); |
223b2b88 SW |
228 | bit /= 2; /* two bits per entry in the bitmap */ |
229 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
230 | } |
231 | ||
232 | /** | |
233 | * gfs2_bitcount - count the number of bits in a certain state | |
234 | * @buffer: the buffer that holds the bitmaps | |
235 | * @buflen: the length (in bytes) of the buffer | |
236 | * @state: the state of the block we're looking for | |
237 | * | |
238 | * Returns: The number of bits | |
239 | */ | |
240 | ||
110acf38 SW |
241 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
242 | unsigned int buflen, u8 state) | |
88c8ab1f | 243 | { |
110acf38 SW |
244 | const u8 *byte = buffer; |
245 | const u8 *end = buffer + buflen; | |
246 | const u8 state1 = state << 2; | |
247 | const u8 state2 = state << 4; | |
248 | const u8 state3 = state << 6; | |
cd915493 | 249 | u32 count = 0; |
88c8ab1f SW |
250 | |
251 | for (; byte < end; byte++) { | |
252 | if (((*byte) & 0x03) == state) | |
253 | count++; | |
254 | if (((*byte) & 0x0C) == state1) | |
255 | count++; | |
256 | if (((*byte) & 0x30) == state2) | |
257 | count++; | |
258 | if (((*byte) & 0xC0) == state3) | |
259 | count++; | |
260 | } | |
261 | ||
262 | return count; | |
263 | } | |
264 | ||
b3b94faa DT |
265 | /** |
266 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
267 | * @sdp: the filesystem | |
268 | * @rgd: the rgrp | |
269 | * | |
270 | */ | |
271 | ||
272 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
273 | { | |
274 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
275 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 276 | u32 length = rgd->rd_length; |
cd915493 | 277 | u32 count[4], tmp; |
b3b94faa DT |
278 | int buf, x; |
279 | ||
cd915493 | 280 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
281 | |
282 | /* Count # blocks in each of 4 possible allocation states */ | |
283 | for (buf = 0; buf < length; buf++) { | |
284 | bi = rgd->rd_bits + buf; | |
285 | for (x = 0; x < 4; x++) | |
286 | count[x] += gfs2_bitcount(rgd, | |
287 | bi->bi_bh->b_data + | |
288 | bi->bi_offset, | |
289 | bi->bi_len, x); | |
290 | } | |
291 | ||
cfc8b549 | 292 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
293 | if (gfs2_consist_rgrpd(rgd)) |
294 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 295 | count[0], rgd->rd_free); |
b3b94faa DT |
296 | return; |
297 | } | |
298 | ||
73f74948 | 299 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 300 | if (count[1] != tmp) { |
b3b94faa DT |
301 | if (gfs2_consist_rgrpd(rgd)) |
302 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
303 | count[1], tmp); | |
304 | return; | |
305 | } | |
306 | ||
6b946170 | 307 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 308 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 309 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 310 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
311 | return; |
312 | } | |
b3b94faa DT |
313 | } |
314 | ||
bb8d8a6f | 315 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 316 | { |
bb8d8a6f SW |
317 | u64 first = rgd->rd_data0; |
318 | u64 last = first + rgd->rd_data; | |
16910427 | 319 | return first <= block && block < last; |
b3b94faa DT |
320 | } |
321 | ||
322 | /** | |
323 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
324 | * @sdp: The GFS2 superblock | |
325 | * @n: The data block number | |
326 | * | |
327 | * Returns: The resource group, or NULL if not found | |
328 | */ | |
329 | ||
cd915493 | 330 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk) |
b3b94faa | 331 | { |
7c9ca621 | 332 | struct rb_node **newn, *parent = NULL; |
b3b94faa DT |
333 | |
334 | spin_lock(&sdp->sd_rindex_spin); | |
335 | ||
7c9ca621 BP |
336 | newn = &sdp->sd_rindex_tree.rb_node; |
337 | ||
338 | /* Figure out where to put new node */ | |
339 | while (*newn) { | |
340 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
341 | rd_node); | |
342 | ||
343 | parent = *newn; | |
344 | if (blk < cur->rd_addr) | |
345 | newn = &((*newn)->rb_left); | |
346 | else if (blk > cur->rd_data0 + cur->rd_data) | |
347 | newn = &((*newn)->rb_right); | |
348 | else { | |
b3b94faa | 349 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 | 350 | return cur; |
b3b94faa DT |
351 | } |
352 | } | |
353 | ||
354 | spin_unlock(&sdp->sd_rindex_spin); | |
355 | ||
356 | return NULL; | |
357 | } | |
358 | ||
359 | /** | |
360 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
361 | * @sdp: The GFS2 superblock | |
362 | * | |
363 | * Returns: The first rgrp in the filesystem | |
364 | */ | |
365 | ||
366 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
367 | { | |
7c9ca621 BP |
368 | const struct rb_node *n; |
369 | struct gfs2_rgrpd *rgd; | |
370 | ||
371 | n = rb_first(&sdp->sd_rindex_tree); | |
372 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
373 | ||
374 | return rgd; | |
b3b94faa DT |
375 | } |
376 | ||
377 | /** | |
378 | * gfs2_rgrpd_get_next - get the next RG | |
379 | * @rgd: A RG | |
380 | * | |
381 | * Returns: The next rgrp | |
382 | */ | |
383 | ||
384 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
385 | { | |
7c9ca621 BP |
386 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
387 | const struct rb_node *n; | |
388 | ||
389 | spin_lock(&sdp->sd_rindex_spin); | |
390 | n = rb_next(&rgd->rd_node); | |
391 | if (n == NULL) | |
392 | n = rb_first(&sdp->sd_rindex_tree); | |
393 | ||
394 | if (unlikely(&rgd->rd_node == n)) { | |
395 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 396 | return NULL; |
7c9ca621 BP |
397 | } |
398 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
399 | spin_unlock(&sdp->sd_rindex_spin); | |
400 | return rgd; | |
b3b94faa DT |
401 | } |
402 | ||
403 | static void clear_rgrpdi(struct gfs2_sbd *sdp) | |
404 | { | |
7c9ca621 | 405 | struct rb_node *n; |
b3b94faa DT |
406 | struct gfs2_rgrpd *rgd; |
407 | struct gfs2_glock *gl; | |
408 | ||
7c9ca621 BP |
409 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
410 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
411 | gl = rgd->rd_gl; |
412 | ||
7c9ca621 | 413 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
414 | |
415 | if (gl) { | |
5c676f6d | 416 | gl->gl_object = NULL; |
29687a2a | 417 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
418 | gfs2_glock_put(gl); |
419 | } | |
420 | ||
421 | kfree(rgd->rd_bits); | |
6bdd9be6 | 422 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
423 | } |
424 | } | |
425 | ||
426 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) | |
427 | { | |
f55ab26a | 428 | mutex_lock(&sdp->sd_rindex_mutex); |
b3b94faa | 429 | clear_rgrpdi(sdp); |
f55ab26a | 430 | mutex_unlock(&sdp->sd_rindex_mutex); |
b3b94faa DT |
431 | } |
432 | ||
bb8d8a6f SW |
433 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
434 | { | |
435 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
436 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
437 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
438 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
439 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
440 | } | |
441 | ||
b3b94faa DT |
442 | /** |
443 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
444 | * @rgd: The resource group descriptor | |
445 | * | |
446 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
447 | * | |
448 | * Returns: errno | |
449 | */ | |
450 | ||
451 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
452 | { | |
453 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
454 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 455 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 456 | u32 bytes_left, bytes; |
b3b94faa DT |
457 | int x; |
458 | ||
feaa7bba SW |
459 | if (!length) |
460 | return -EINVAL; | |
461 | ||
dd894be8 | 462 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
463 | if (!rgd->rd_bits) |
464 | return -ENOMEM; | |
465 | ||
bb8d8a6f | 466 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
467 | |
468 | for (x = 0; x < length; x++) { | |
469 | bi = rgd->rd_bits + x; | |
470 | ||
60a0b8f9 | 471 | bi->bi_flags = 0; |
b3b94faa DT |
472 | /* small rgrp; bitmap stored completely in header block */ |
473 | if (length == 1) { | |
474 | bytes = bytes_left; | |
475 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
476 | bi->bi_start = 0; | |
477 | bi->bi_len = bytes; | |
478 | /* header block */ | |
479 | } else if (x == 0) { | |
480 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
481 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
482 | bi->bi_start = 0; | |
483 | bi->bi_len = bytes; | |
484 | /* last block */ | |
485 | } else if (x + 1 == length) { | |
486 | bytes = bytes_left; | |
487 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 488 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
489 | bi->bi_len = bytes; |
490 | /* other blocks */ | |
491 | } else { | |
568f4c96 SW |
492 | bytes = sdp->sd_sb.sb_bsize - |
493 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 494 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 495 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
496 | bi->bi_len = bytes; |
497 | } | |
498 | ||
499 | bytes_left -= bytes; | |
500 | } | |
501 | ||
502 | if (bytes_left) { | |
503 | gfs2_consist_rgrpd(rgd); | |
504 | return -EIO; | |
505 | } | |
506 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 507 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 508 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 509 | gfs2_rindex_print(rgd); |
b3b94faa DT |
510 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
511 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
512 | } | |
513 | return -EIO; | |
514 | } | |
515 | ||
516 | return 0; | |
517 | } | |
518 | ||
7ae8fa84 RP |
519 | /** |
520 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
521 | * | |
522 | */ | |
523 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
524 | { | |
525 | u64 total_data = 0; | |
526 | struct inode *inode = sdp->sd_rindex; | |
527 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 RP |
528 | char buf[sizeof(struct gfs2_rindex)]; |
529 | struct file_ra_state ra_state; | |
530 | int error, rgrps; | |
531 | ||
532 | mutex_lock(&sdp->sd_rindex_mutex); | |
533 | file_ra_state_init(&ra_state, inode->i_mapping); | |
534 | for (rgrps = 0;; rgrps++) { | |
535 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
536 | ||
bcd7278d | 537 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 RP |
538 | break; |
539 | error = gfs2_internal_read(ip, &ra_state, buf, &pos, | |
540 | sizeof(struct gfs2_rindex)); | |
541 | if (error != sizeof(struct gfs2_rindex)) | |
542 | break; | |
bb8d8a6f | 543 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 RP |
544 | } |
545 | mutex_unlock(&sdp->sd_rindex_mutex); | |
546 | return total_data; | |
547 | } | |
548 | ||
bb8d8a6f SW |
549 | static void gfs2_rindex_in(struct gfs2_rgrpd *rgd, const void *buf) |
550 | { | |
551 | const struct gfs2_rindex *str = buf; | |
552 | ||
553 | rgd->rd_addr = be64_to_cpu(str->ri_addr); | |
554 | rgd->rd_length = be32_to_cpu(str->ri_length); | |
555 | rgd->rd_data0 = be64_to_cpu(str->ri_data0); | |
556 | rgd->rd_data = be32_to_cpu(str->ri_data); | |
557 | rgd->rd_bitbytes = be32_to_cpu(str->ri_bitbytes); | |
558 | } | |
559 | ||
7c9ca621 BP |
560 | static void rgd_insert(struct gfs2_rgrpd *rgd) |
561 | { | |
562 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
563 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
564 | ||
565 | /* Figure out where to put new node */ | |
566 | while (*newn) { | |
567 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
568 | rd_node); | |
569 | ||
570 | parent = *newn; | |
571 | if (rgd->rd_addr < cur->rd_addr) | |
572 | newn = &((*newn)->rb_left); | |
573 | else if (rgd->rd_addr > cur->rd_addr) | |
574 | newn = &((*newn)->rb_right); | |
575 | else | |
576 | return; | |
577 | } | |
578 | ||
579 | rb_link_node(&rgd->rd_node, parent, newn); | |
580 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
581 | } | |
582 | ||
b3b94faa | 583 | /** |
6c53267f | 584 | * read_rindex_entry - Pull in a new resource index entry from the disk |
b3b94faa DT |
585 | * @gl: The glock covering the rindex inode |
586 | * | |
6c53267f RP |
587 | * Returns: 0 on success, error code otherwise |
588 | */ | |
589 | ||
590 | static int read_rindex_entry(struct gfs2_inode *ip, | |
591 | struct file_ra_state *ra_state) | |
592 | { | |
593 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
594 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
595 | char buf[sizeof(struct gfs2_rindex)]; | |
596 | int error; | |
597 | struct gfs2_rgrpd *rgd; | |
598 | ||
599 | error = gfs2_internal_read(ip, ra_state, buf, &pos, | |
600 | sizeof(struct gfs2_rindex)); | |
601 | if (!error) | |
602 | return 0; | |
603 | if (error != sizeof(struct gfs2_rindex)) { | |
604 | if (error > 0) | |
605 | error = -EIO; | |
606 | return error; | |
607 | } | |
608 | ||
6bdd9be6 | 609 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
610 | error = -ENOMEM; |
611 | if (!rgd) | |
612 | return error; | |
613 | ||
6c53267f RP |
614 | rgd->rd_sbd = sdp; |
615 | ||
bb8d8a6f | 616 | gfs2_rindex_in(rgd, buf); |
7c9ca621 BP |
617 | rgd_insert(rgd); |
618 | ||
6c53267f RP |
619 | error = compute_bitstructs(rgd); |
620 | if (error) | |
621 | return error; | |
622 | ||
bb8d8a6f | 623 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
624 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
625 | if (error) | |
626 | return error; | |
627 | ||
628 | rgd->rd_gl->gl_object = rgd; | |
cf45b752 | 629 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
7c9ca621 BP |
630 | if (rgd->rd_data > sdp->sd_max_rg_data) |
631 | sdp->sd_max_rg_data = rgd->rd_data; | |
6c53267f RP |
632 | return error; |
633 | } | |
634 | ||
635 | /** | |
636 | * gfs2_ri_update - Pull in a new resource index from the disk | |
637 | * @ip: pointer to the rindex inode | |
638 | * | |
b3b94faa DT |
639 | * Returns: 0 on successful update, error code otherwise |
640 | */ | |
641 | ||
086d8334 | 642 | int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 643 | { |
feaa7bba SW |
644 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
645 | struct inode *inode = &ip->i_inode; | |
f42faf4f | 646 | struct file_ra_state ra_state; |
a2e0f799 | 647 | u64 rgrp_count = i_size_read(inode); |
b3b94faa DT |
648 | int error; |
649 | ||
7fe3ec6f | 650 | do_div(rgrp_count, sizeof(struct gfs2_rindex)); |
b3b94faa DT |
651 | clear_rgrpdi(sdp); |
652 | ||
f42faf4f | 653 | file_ra_state_init(&ra_state, inode->i_mapping); |
cd81a4ba | 654 | for (sdp->sd_rgrps = 0; sdp->sd_rgrps < rgrp_count; sdp->sd_rgrps++) { |
6c53267f RP |
655 | error = read_rindex_entry(ip, &ra_state); |
656 | if (error) { | |
657 | clear_rgrpdi(sdp); | |
658 | return error; | |
b3b94faa | 659 | } |
6c53267f | 660 | } |
b3b94faa | 661 | |
cf45b752 | 662 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
663 | return 0; |
664 | } | |
b3b94faa | 665 | |
b3b94faa DT |
666 | /** |
667 | * gfs2_rindex_hold - Grab a lock on the rindex | |
668 | * @sdp: The GFS2 superblock | |
669 | * @ri_gh: the glock holder | |
670 | * | |
671 | * We grab a lock on the rindex inode to make sure that it doesn't | |
672 | * change whilst we are performing an operation. We keep this lock | |
673 | * for quite long periods of time compared to other locks. This | |
674 | * doesn't matter, since it is shared and it is very, very rarely | |
675 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
676 | * | |
677 | * This makes sure that we're using the latest copy of the resource index | |
678 | * special file, which might have been updated if someone expanded the | |
679 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
680 | * | |
681 | * Returns: 0 on success, error code otherwise | |
682 | */ | |
683 | ||
684 | int gfs2_rindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ri_gh) | |
685 | { | |
feaa7bba | 686 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa DT |
687 | struct gfs2_glock *gl = ip->i_gl; |
688 | int error; | |
689 | ||
690 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, ri_gh); | |
691 | if (error) | |
692 | return error; | |
693 | ||
694 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 695 | if (!sdp->sd_rindex_uptodate) { |
f55ab26a | 696 | mutex_lock(&sdp->sd_rindex_mutex); |
cf45b752 | 697 | if (!sdp->sd_rindex_uptodate) { |
b3b94faa DT |
698 | error = gfs2_ri_update(ip); |
699 | if (error) | |
700 | gfs2_glock_dq_uninit(ri_gh); | |
701 | } | |
f55ab26a | 702 | mutex_unlock(&sdp->sd_rindex_mutex); |
b3b94faa DT |
703 | } |
704 | ||
705 | return error; | |
706 | } | |
707 | ||
42d52e38 | 708 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
709 | { |
710 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 711 | u32 rg_flags; |
bb8d8a6f | 712 | |
42d52e38 | 713 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 714 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
715 | rgd->rd_flags &= GFS2_RDF_MASK; |
716 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 717 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 718 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 719 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
720 | } |
721 | ||
42d52e38 | 722 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
723 | { |
724 | struct gfs2_rgrp *str = buf; | |
725 | ||
09010978 | 726 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 727 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 728 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 729 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 730 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
731 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
732 | } | |
733 | ||
b3b94faa | 734 | /** |
7c9ca621 | 735 | * gfs2_rgrp_go_lock - Read in a RG's header and bitmaps |
b3b94faa DT |
736 | * @rgd: the struct gfs2_rgrpd describing the RG to read in |
737 | * | |
738 | * Read in all of a Resource Group's header and bitmap blocks. | |
739 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
740 | * | |
741 | * Returns: errno | |
742 | */ | |
743 | ||
7c9ca621 | 744 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) |
b3b94faa | 745 | { |
7c9ca621 | 746 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
b3b94faa DT |
747 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
748 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 749 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
750 | struct gfs2_bitmap *bi; |
751 | unsigned int x, y; | |
752 | int error; | |
753 | ||
b3b94faa DT |
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); |
7c9ca621 | 778 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa DT |
779 | } |
780 | ||
b3b94faa DT |
781 | return 0; |
782 | ||
feaa7bba | 783 | fail: |
b3b94faa DT |
784 | while (x--) { |
785 | bi = rgd->rd_bits + x; | |
786 | brelse(bi->bi_bh); | |
787 | bi->bi_bh = NULL; | |
788 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
789 | } | |
b3b94faa DT |
790 | |
791 | return error; | |
792 | } | |
793 | ||
b3b94faa | 794 | /** |
7c9ca621 | 795 | * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
b3b94faa DT |
796 | * @rgd: the struct gfs2_rgrpd describing the RG to read in |
797 | * | |
798 | */ | |
799 | ||
7c9ca621 | 800 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) |
b3b94faa | 801 | { |
7c9ca621 | 802 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
bb8d8a6f | 803 | int x, length = rgd->rd_length; |
b3b94faa | 804 | |
b3b94faa DT |
805 | for (x = 0; x < length; x++) { |
806 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
b3b94faa DT |
807 | brelse(bi->bi_bh); |
808 | bi->bi_bh = NULL; | |
809 | } | |
810 | ||
b3b94faa DT |
811 | } |
812 | ||
7c9ca621 BP |
813 | void gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
814 | struct buffer_head *bh, | |
815 | const struct gfs2_bitmap *bi) | |
f15ab561 SW |
816 | { |
817 | struct super_block *sb = sdp->sd_vfs; | |
818 | struct block_device *bdev = sb->s_bdev; | |
819 | const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / | |
e1defc4f | 820 | bdev_logical_block_size(sb->s_bdev); |
f15ab561 | 821 | u64 blk; |
64d576ba | 822 | sector_t start = 0; |
f15ab561 SW |
823 | sector_t nr_sects = 0; |
824 | int rv; | |
825 | unsigned int x; | |
826 | ||
827 | for (x = 0; x < bi->bi_len; x++) { | |
7c9ca621 | 828 | const u8 *orig = bh->b_data + bi->bi_offset + x; |
f15ab561 SW |
829 | const u8 *clone = bi->bi_clone + bi->bi_offset + x; |
830 | u8 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
831 | diff &= 0x55; | |
832 | if (diff == 0) | |
833 | continue; | |
834 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
835 | blk *= sects_per_blk; /* convert to sectors */ | |
836 | while(diff) { | |
837 | if (diff & 1) { | |
838 | if (nr_sects == 0) | |
839 | goto start_new_extent; | |
840 | if ((start + nr_sects) != blk) { | |
841 | rv = blkdev_issue_discard(bdev, start, | |
746cd1e7 | 842 | nr_sects, GFP_NOFS, |
dd3932ed | 843 | 0); |
f15ab561 SW |
844 | if (rv) |
845 | goto fail; | |
846 | nr_sects = 0; | |
847 | start_new_extent: | |
848 | start = blk; | |
849 | } | |
850 | nr_sects += sects_per_blk; | |
851 | } | |
852 | diff >>= 2; | |
853 | blk += sects_per_blk; | |
854 | } | |
855 | } | |
856 | if (nr_sects) { | |
dd3932ed | 857 | rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0); |
f15ab561 SW |
858 | if (rv) |
859 | goto fail; | |
860 | } | |
861 | return; | |
862 | fail: | |
863 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
864 | sdp->sd_args.ar_discard = 0; | |
865 | } | |
866 | ||
b3b94faa DT |
867 | /** |
868 | * gfs2_alloc_get - get the struct gfs2_alloc structure for an inode | |
869 | * @ip: the incore GFS2 inode structure | |
870 | * | |
871 | * Returns: the struct gfs2_alloc | |
872 | */ | |
873 | ||
874 | struct gfs2_alloc *gfs2_alloc_get(struct gfs2_inode *ip) | |
875 | { | |
6dbd8224 | 876 | BUG_ON(ip->i_alloc != NULL); |
ea8d62da | 877 | ip->i_alloc = kzalloc(sizeof(struct gfs2_alloc), GFP_NOFS); |
6dbd8224 | 878 | return ip->i_alloc; |
b3b94faa DT |
879 | } |
880 | ||
b3b94faa DT |
881 | /** |
882 | * try_rgrp_fit - See if a given reservation will fit in a given RG | |
883 | * @rgd: the RG data | |
884 | * @al: the struct gfs2_alloc structure describing the reservation | |
885 | * | |
886 | * If there's room for the requested blocks to be allocated from the RG: | |
b3b94faa DT |
887 | * Sets the $al_rgd field in @al. |
888 | * | |
889 | * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) | |
890 | */ | |
891 | ||
892 | static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_alloc *al) | |
893 | { | |
09010978 | 894 | if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) |
a43a4906 SW |
895 | return 0; |
896 | ||
b3b94faa DT |
897 | if (rgd->rd_free_clone >= al->al_requested) { |
898 | al->al_rgd = rgd; | |
7c9ca621 | 899 | return 1; |
b3b94faa | 900 | } |
b3b94faa | 901 | |
7c9ca621 | 902 | return 0; |
b3b94faa DT |
903 | } |
904 | ||
c8cdf479 SW |
905 | /** |
906 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
907 | * @rgd: The rgrp | |
908 | * | |
1a0eae88 BP |
909 | * Returns: 0 if no error |
910 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
911 | */ |
912 | ||
044b9414 | 913 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 914 | { |
6760bdcd | 915 | u32 goal = 0, block; |
bb9bcf06 | 916 | u64 no_addr; |
5f3eae75 | 917 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
b45e41d7 | 918 | unsigned int n; |
044b9414 SW |
919 | struct gfs2_glock *gl; |
920 | struct gfs2_inode *ip; | |
921 | int error; | |
922 | int found = 0; | |
c8cdf479 | 923 | |
044b9414 | 924 | while (goal < rgd->rd_data) { |
5f3eae75 | 925 | down_write(&sdp->sd_log_flush_lock); |
b45e41d7 | 926 | n = 1; |
6760bdcd | 927 | block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, |
b45e41d7 | 928 | GFS2_BLKST_UNLINKED, &n); |
5f3eae75 | 929 | up_write(&sdp->sd_log_flush_lock); |
6760bdcd | 930 | if (block == BFITNOENT) |
24c73873 | 931 | break; |
6760bdcd BP |
932 | /* rgblk_search can return a block < goal, so we need to |
933 | keep it marching forward. */ | |
934 | no_addr = block + rgd->rd_data0; | |
44ad37d6 | 935 | goal = max(block + 1, goal + 1); |
6760bdcd | 936 | if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked) |
c8cdf479 | 937 | continue; |
1e19a195 SW |
938 | if (no_addr == skip) |
939 | continue; | |
bb9bcf06 | 940 | *last_unlinked = no_addr; |
044b9414 SW |
941 | |
942 | error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &gl); | |
943 | if (error) | |
944 | continue; | |
945 | ||
946 | /* If the inode is already in cache, we can ignore it here | |
947 | * because the existing inode disposal code will deal with | |
948 | * it when all refs have gone away. Accessing gl_object like | |
949 | * this is not safe in general. Here it is ok because we do | |
950 | * not dereference the pointer, and we only need an approx | |
951 | * answer to whether it is NULL or not. | |
952 | */ | |
953 | ip = gl->gl_object; | |
954 | ||
955 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
956 | gfs2_glock_put(gl); | |
957 | else | |
958 | found++; | |
959 | ||
960 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 961 | if (found > NR_CPUS) |
044b9414 | 962 | return; |
c8cdf479 SW |
963 | } |
964 | ||
965 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 966 | return; |
c8cdf479 SW |
967 | } |
968 | ||
b3b94faa DT |
969 | /** |
970 | * get_local_rgrp - Choose and lock a rgrp for allocation | |
971 | * @ip: the inode to reserve space for | |
972 | * @rgp: the chosen and locked rgrp | |
973 | * | |
974 | * Try to acquire rgrp in way which avoids contending with others. | |
975 | * | |
976 | * Returns: errno | |
977 | */ | |
978 | ||
044b9414 | 979 | static int get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked) |
b3b94faa | 980 | { |
feaa7bba | 981 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa | 982 | struct gfs2_rgrpd *rgd, *begin = NULL; |
6dbd8224 | 983 | struct gfs2_alloc *al = ip->i_alloc; |
292c8c14 | 984 | int error, rg_locked; |
7c9ca621 | 985 | int loops = 0; |
b3b94faa | 986 | |
7c9ca621 | 987 | rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal); |
b3b94faa | 988 | |
7c9ca621 BP |
989 | if (rgd == NULL) |
990 | return -EBADSLT; | |
991 | ||
992 | while (loops < 3) { | |
292c8c14 AD |
993 | rg_locked = 0; |
994 | ||
995 | if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) { | |
996 | rg_locked = 1; | |
997 | error = 0; | |
998 | } else { | |
999 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, | |
1000 | LM_FLAG_TRY, &al->al_rgd_gh); | |
1001 | } | |
b3b94faa DT |
1002 | switch (error) { |
1003 | case 0: | |
1004 | if (try_rgrp_fit(rgd, al)) | |
7c9ca621 | 1005 | return 0; |
044b9414 SW |
1006 | if (rgd->rd_flags & GFS2_RDF_CHECK) |
1007 | try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr); | |
292c8c14 AD |
1008 | if (!rg_locked) |
1009 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
9cabcdbd | 1010 | /* fall through */ |
b3b94faa | 1011 | case GLR_TRYFAILED: |
7c9ca621 BP |
1012 | rgd = gfs2_rgrpd_get_next(rgd); |
1013 | if (rgd == begin) | |
1014 | loops++; | |
b3b94faa DT |
1015 | break; |
1016 | ||
1017 | default: | |
cc0581bd | 1018 | return error; |
b3b94faa | 1019 | } |
b3b94faa DT |
1020 | } |
1021 | ||
7c9ca621 | 1022 | return -ENOSPC; |
b3b94faa DT |
1023 | } |
1024 | ||
1025 | /** | |
1026 | * gfs2_inplace_reserve_i - Reserve space in the filesystem | |
1027 | * @ip: the inode to reserve space for | |
1028 | * | |
1029 | * Returns: errno | |
1030 | */ | |
1031 | ||
46290341 BP |
1032 | int gfs2_inplace_reserve_i(struct gfs2_inode *ip, int hold_rindex, |
1033 | char *file, unsigned int line) | |
b3b94faa | 1034 | { |
feaa7bba | 1035 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
6dbd8224 | 1036 | struct gfs2_alloc *al = ip->i_alloc; |
7ae8fa84 | 1037 | int error = 0; |
044b9414 SW |
1038 | u64 last_unlinked = NO_BLOCK; |
1039 | int tries = 0; | |
b3b94faa DT |
1040 | |
1041 | if (gfs2_assert_warn(sdp, al->al_requested)) | |
1042 | return -EINVAL; | |
1043 | ||
46290341 BP |
1044 | if (hold_rindex) { |
1045 | /* We need to hold the rindex unless the inode we're using is | |
1046 | the rindex itself, in which case it's already held. */ | |
1047 | if (ip != GFS2_I(sdp->sd_rindex)) | |
1048 | error = gfs2_rindex_hold(sdp, &al->al_ri_gh); | |
1049 | else if (!sdp->sd_rgrps) /* We may not have the rindex read | |
1050 | in, so: */ | |
0489b3f5 | 1051 | error = gfs2_ri_update(ip); |
044b9414 SW |
1052 | if (error) |
1053 | return error; | |
46290341 | 1054 | } |
7ae8fa84 | 1055 | |
0489b3f5 | 1056 | try_again: |
044b9414 SW |
1057 | do { |
1058 | error = get_local_rgrp(ip, &last_unlinked); | |
1059 | /* If there is no space, flushing the log may release some */ | |
0489b3f5 BM |
1060 | if (error) { |
1061 | if (ip == GFS2_I(sdp->sd_rindex) && | |
1062 | !sdp->sd_rindex_uptodate) { | |
1063 | error = gfs2_ri_update(ip); | |
1064 | if (error) | |
1065 | return error; | |
1066 | goto try_again; | |
1067 | } | |
044b9414 | 1068 | gfs2_log_flush(sdp, NULL); |
0489b3f5 | 1069 | } |
044b9414 | 1070 | } while (error && tries++ < 3); |
b3b94faa | 1071 | |
cc0581bd | 1072 | if (error) { |
46290341 | 1073 | if (hold_rindex && ip != GFS2_I(sdp->sd_rindex)) |
7ae8fa84 | 1074 | gfs2_glock_dq_uninit(&al->al_ri_gh); |
044b9414 | 1075 | return error; |
b3b94faa | 1076 | } |
044b9414 | 1077 | |
ed4878e8 | 1078 | /* no error, so we have the rgrp set in the inode's allocation. */ |
b3b94faa DT |
1079 | al->al_file = file; |
1080 | al->al_line = line; | |
1081 | ||
1082 | return 0; | |
1083 | } | |
1084 | ||
1085 | /** | |
1086 | * gfs2_inplace_release - release an inplace reservation | |
1087 | * @ip: the inode the reservation was taken out on | |
1088 | * | |
1089 | * Release a reservation made by gfs2_inplace_reserve(). | |
1090 | */ | |
1091 | ||
1092 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1093 | { | |
feaa7bba | 1094 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
6dbd8224 | 1095 | struct gfs2_alloc *al = ip->i_alloc; |
b3b94faa DT |
1096 | |
1097 | if (gfs2_assert_warn(sdp, al->al_alloced <= al->al_requested) == -1) | |
1098 | fs_warn(sdp, "al_alloced = %u, al_requested = %u " | |
1099 | "al_file = %s, al_line = %u\n", | |
1100 | al->al_alloced, al->al_requested, al->al_file, | |
1101 | al->al_line); | |
1102 | ||
1103 | al->al_rgd = NULL; | |
292c8c14 AD |
1104 | if (al->al_rgd_gh.gh_gl) |
1105 | gfs2_glock_dq_uninit(&al->al_rgd_gh); | |
46290341 | 1106 | if (ip != GFS2_I(sdp->sd_rindex) && al->al_ri_gh.gh_gl) |
7ae8fa84 | 1107 | gfs2_glock_dq_uninit(&al->al_ri_gh); |
b3b94faa DT |
1108 | } |
1109 | ||
1110 | /** | |
1111 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1112 | * @rgd: the resource group holding the block | |
1113 | * @block: the block number | |
1114 | * | |
1115 | * Returns: The block type (GFS2_BLKST_*) | |
1116 | */ | |
1117 | ||
acf7e244 | 1118 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa DT |
1119 | { |
1120 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1121 | u32 length, rgrp_block, buf_block; |
b3b94faa DT |
1122 | unsigned int buf; |
1123 | unsigned char type; | |
1124 | ||
bb8d8a6f SW |
1125 | length = rgd->rd_length; |
1126 | rgrp_block = block - rgd->rd_data0; | |
b3b94faa DT |
1127 | |
1128 | for (buf = 0; buf < length; buf++) { | |
1129 | bi = rgd->rd_bits + buf; | |
1130 | if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1131 | break; | |
1132 | } | |
1133 | ||
1134 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1135 | buf_block = rgrp_block - bi->bi_start * GFS2_NBBY; | |
1136 | ||
feaa7bba | 1137 | type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset, |
b3b94faa DT |
1138 | bi->bi_len, buf_block); |
1139 | ||
1140 | return type; | |
1141 | } | |
1142 | ||
1143 | /** | |
1144 | * rgblk_search - find a block in @old_state, change allocation | |
1145 | * state to @new_state | |
1146 | * @rgd: the resource group descriptor | |
1147 | * @goal: the goal block within the RG (start here to search for avail block) | |
1148 | * @old_state: GFS2_BLKST_XXX the before-allocation state to find | |
1149 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
b45e41d7 | 1150 | * @n: The extent length |
b3b94faa DT |
1151 | * |
1152 | * Walk rgrp's bitmap to find bits that represent a block in @old_state. | |
1153 | * Add the found bitmap buffer to the transaction. | |
1154 | * Set the found bits to @new_state to change block's allocation state. | |
1155 | * | |
1156 | * This function never fails, because we wouldn't call it unless we | |
1157 | * know (from reservation results, etc.) that a block is available. | |
1158 | * | |
1159 | * Scope of @goal and returned block is just within rgrp, not the whole | |
1160 | * filesystem. | |
1161 | * | |
1162 | * Returns: the block number allocated | |
1163 | */ | |
1164 | ||
cd915493 | 1165 | static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, |
b45e41d7 SW |
1166 | unsigned char old_state, unsigned char new_state, |
1167 | unsigned int *n) | |
b3b94faa DT |
1168 | { |
1169 | struct gfs2_bitmap *bi = NULL; | |
b45e41d7 | 1170 | const u32 length = rgd->rd_length; |
60a0b8f9 | 1171 | u32 blk = BFITNOENT; |
b3b94faa | 1172 | unsigned int buf, x; |
b45e41d7 | 1173 | const unsigned int elen = *n; |
60a0b8f9 | 1174 | const u8 *buffer = NULL; |
b3b94faa | 1175 | |
b45e41d7 | 1176 | *n = 0; |
b3b94faa DT |
1177 | /* Find bitmap block that contains bits for goal block */ |
1178 | for (buf = 0; buf < length; buf++) { | |
1179 | bi = rgd->rd_bits + buf; | |
60a0b8f9 SW |
1180 | /* Convert scope of "goal" from rgrp-wide to within found bit block */ |
1181 | if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { | |
1182 | goal -= bi->bi_start * GFS2_NBBY; | |
1183 | goto do_search; | |
1184 | } | |
b3b94faa | 1185 | } |
60a0b8f9 SW |
1186 | buf = 0; |
1187 | goal = 0; | |
b3b94faa | 1188 | |
60a0b8f9 | 1189 | do_search: |
b3b94faa DT |
1190 | /* Search (up to entire) bitmap in this rgrp for allocatable block. |
1191 | "x <= length", instead of "x < length", because we typically start | |
1192 | the search in the middle of a bit block, but if we can't find an | |
1193 | allocatable block anywhere else, we want to be able wrap around and | |
1194 | search in the first part of our first-searched bit block. */ | |
1195 | for (x = 0; x <= length; x++) { | |
60a0b8f9 SW |
1196 | bi = rgd->rd_bits + buf; |
1197 | ||
1198 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
1199 | (old_state == GFS2_BLKST_FREE)) | |
1200 | goto skip; | |
1201 | ||
5f3eae75 BP |
1202 | /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone |
1203 | bitmaps, so we must search the originals for that. */ | |
b45e41d7 | 1204 | buffer = bi->bi_bh->b_data + bi->bi_offset; |
7c9ca621 | 1205 | WARN_ON(!buffer_uptodate(bi->bi_bh)); |
5f3eae75 | 1206 | if (old_state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
110acf38 SW |
1207 | buffer = bi->bi_clone + bi->bi_offset; |
1208 | ||
1209 | blk = gfs2_bitfit(buffer, bi->bi_len, goal, old_state); | |
b3b94faa DT |
1210 | if (blk != BFITNOENT) |
1211 | break; | |
1212 | ||
60a0b8f9 SW |
1213 | if ((goal == 0) && (old_state == GFS2_BLKST_FREE)) |
1214 | set_bit(GBF_FULL, &bi->bi_flags); | |
1215 | ||
b3b94faa | 1216 | /* Try next bitmap block (wrap back to rgrp header if at end) */ |
60a0b8f9 SW |
1217 | skip: |
1218 | buf++; | |
1219 | buf %= length; | |
b3b94faa DT |
1220 | goal = 0; |
1221 | } | |
1222 | ||
60a0b8f9 SW |
1223 | if (blk == BFITNOENT) |
1224 | return blk; | |
7c9ca621 | 1225 | |
60a0b8f9 SW |
1226 | *n = 1; |
1227 | if (old_state == new_state) | |
1228 | goto out; | |
1229 | ||
1230 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); | |
1231 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, | |
95c8e17f | 1232 | bi, blk, new_state); |
60a0b8f9 SW |
1233 | goal = blk; |
1234 | while (*n < elen) { | |
1235 | goal++; | |
1236 | if (goal >= (bi->bi_len * GFS2_NBBY)) | |
1237 | break; | |
1238 | if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) != | |
1239 | GFS2_BLKST_FREE) | |
1240 | break; | |
b45e41d7 | 1241 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, |
95c8e17f | 1242 | bi, goal, new_state); |
60a0b8f9 | 1243 | (*n)++; |
c8cdf479 | 1244 | } |
60a0b8f9 SW |
1245 | out: |
1246 | return (bi->bi_start * GFS2_NBBY) + blk; | |
b3b94faa DT |
1247 | } |
1248 | ||
1249 | /** | |
1250 | * rgblk_free - Change alloc state of given block(s) | |
1251 | * @sdp: the filesystem | |
1252 | * @bstart: the start of a run of blocks to free | |
1253 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1254 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1255 | * | |
1256 | * Returns: Resource group containing the block(s) | |
1257 | */ | |
1258 | ||
cd915493 SW |
1259 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
1260 | u32 blen, unsigned char new_state) | |
b3b94faa DT |
1261 | { |
1262 | struct gfs2_rgrpd *rgd; | |
1263 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1264 | u32 length, rgrp_blk, buf_blk; |
b3b94faa DT |
1265 | unsigned int buf; |
1266 | ||
1267 | rgd = gfs2_blk2rgrpd(sdp, bstart); | |
1268 | if (!rgd) { | |
1269 | if (gfs2_consist(sdp)) | |
382066da | 1270 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
1271 | return NULL; |
1272 | } | |
1273 | ||
bb8d8a6f | 1274 | length = rgd->rd_length; |
b3b94faa | 1275 | |
bb8d8a6f | 1276 | rgrp_blk = bstart - rgd->rd_data0; |
b3b94faa DT |
1277 | |
1278 | while (blen--) { | |
1279 | for (buf = 0; buf < length; buf++) { | |
1280 | bi = rgd->rd_bits + buf; | |
1281 | if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1282 | break; | |
1283 | } | |
1284 | ||
1285 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1286 | ||
1287 | buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY; | |
1288 | rgrp_blk++; | |
1289 | ||
1290 | if (!bi->bi_clone) { | |
1291 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
dd894be8 | 1292 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1293 | memcpy(bi->bi_clone + bi->bi_offset, |
1294 | bi->bi_bh->b_data + bi->bi_offset, | |
1295 | bi->bi_len); | |
1296 | } | |
d4e9c4c3 | 1297 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
b45e41d7 | 1298 | gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset, |
95c8e17f | 1299 | bi, buf_blk, new_state); |
b3b94faa DT |
1300 | } |
1301 | ||
1302 | return rgd; | |
1303 | } | |
1304 | ||
1305 | /** | |
09010978 SW |
1306 | * gfs2_rgrp_dump - print out an rgrp |
1307 | * @seq: The iterator | |
1308 | * @gl: The glock in question | |
1309 | * | |
1310 | */ | |
1311 | ||
1312 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
1313 | { | |
1314 | const struct gfs2_rgrpd *rgd = gl->gl_object; | |
1315 | if (rgd == NULL) | |
1316 | return 0; | |
1317 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n", | |
1318 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, | |
1319 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes); | |
1320 | return 0; | |
1321 | } | |
1322 | ||
6050b9c7 SW |
1323 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
1324 | { | |
1325 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1326 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 1327 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
1328 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
1329 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
1330 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
1331 | } | |
1332 | ||
09010978 SW |
1333 | /** |
1334 | * gfs2_alloc_block - Allocate one or more blocks | |
1639431a | 1335 | * @ip: the inode to allocate the block for |
09010978 SW |
1336 | * @bn: Used to return the starting block number |
1337 | * @n: requested number of blocks/extent length (value/result) | |
b3b94faa | 1338 | * |
09010978 | 1339 | * Returns: 0 or error |
b3b94faa DT |
1340 | */ |
1341 | ||
09010978 | 1342 | int gfs2_alloc_block(struct gfs2_inode *ip, u64 *bn, unsigned int *n) |
b3b94faa | 1343 | { |
feaa7bba | 1344 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 1345 | struct buffer_head *dibh; |
6dbd8224 | 1346 | struct gfs2_alloc *al = ip->i_alloc; |
9a3f236d | 1347 | struct gfs2_rgrpd *rgd; |
cd915493 SW |
1348 | u32 goal, blk; |
1349 | u64 block; | |
d9ba7615 | 1350 | int error; |
b3b94faa | 1351 | |
9a3f236d SW |
1352 | /* Only happens if there is a bug in gfs2, return something distinctive |
1353 | * to ensure that it is noticed. | |
1354 | */ | |
1355 | if (al == NULL) | |
1356 | return -ECANCELED; | |
1357 | ||
1358 | rgd = al->al_rgd; | |
1359 | ||
ce276b06 SW |
1360 | if (rgrp_contains_block(rgd, ip->i_goal)) |
1361 | goal = ip->i_goal - rgd->rd_data0; | |
b3b94faa | 1362 | else |
ac576cc5 | 1363 | goal = rgd->rd_last_alloc; |
b3b94faa | 1364 | |
b45e41d7 | 1365 | blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, GFS2_BLKST_USED, n); |
09010978 SW |
1366 | |
1367 | /* Since all blocks are reserved in advance, this shouldn't happen */ | |
1368 | if (blk == BFITNOENT) | |
1369 | goto rgrp_error; | |
b3b94faa | 1370 | |
b45e41d7 | 1371 | rgd->rd_last_alloc = blk; |
bb8d8a6f | 1372 | block = rgd->rd_data0 + blk; |
ce276b06 | 1373 | ip->i_goal = block; |
d9ba7615 SW |
1374 | error = gfs2_meta_inode_buffer(ip, &dibh); |
1375 | if (error == 0) { | |
1376 | struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data; | |
1377 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
1378 | di->di_goal_meta = di->di_goal_data = cpu_to_be64(ip->i_goal); | |
1379 | brelse(dibh); | |
1380 | } | |
09010978 SW |
1381 | if (rgd->rd_free < *n) |
1382 | goto rgrp_error; | |
1383 | ||
cfc8b549 | 1384 | rgd->rd_free -= *n; |
b3b94faa | 1385 | |
d4e9c4c3 | 1386 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1387 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 1388 | |
b45e41d7 | 1389 | al->al_alloced += *n; |
b3b94faa | 1390 | |
ad99f777 | 1391 | gfs2_statfs_change(sdp, 0, -(s64)*n, 0); |
b45e41d7 | 1392 | gfs2_quota_change(ip, *n, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa | 1393 | |
b45e41d7 | 1394 | rgd->rd_free_clone -= *n; |
63997775 | 1395 | trace_gfs2_block_alloc(ip, block, *n, GFS2_BLKST_USED); |
09010978 SW |
1396 | *bn = block; |
1397 | return 0; | |
1398 | ||
1399 | rgrp_error: | |
6050b9c7 | 1400 | gfs2_rgrp_error(rgd); |
09010978 | 1401 | return -EIO; |
b3b94faa DT |
1402 | } |
1403 | ||
1404 | /** | |
1405 | * gfs2_alloc_di - Allocate a dinode | |
1406 | * @dip: the directory that the inode is going in | |
6050b9c7 SW |
1407 | * @bn: the block number which is allocated |
1408 | * @generation: the generation number of the inode | |
b3b94faa | 1409 | * |
6050b9c7 | 1410 | * Returns: 0 on success or error |
b3b94faa DT |
1411 | */ |
1412 | ||
6050b9c7 | 1413 | int gfs2_alloc_di(struct gfs2_inode *dip, u64 *bn, u64 *generation) |
b3b94faa | 1414 | { |
feaa7bba | 1415 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
6dbd8224 | 1416 | struct gfs2_alloc *al = dip->i_alloc; |
b3b94faa | 1417 | struct gfs2_rgrpd *rgd = al->al_rgd; |
4340fe62 SW |
1418 | u32 blk; |
1419 | u64 block; | |
b45e41d7 | 1420 | unsigned int n = 1; |
b3b94faa | 1421 | |
ac576cc5 | 1422 | blk = rgblk_search(rgd, rgd->rd_last_alloc, |
b45e41d7 | 1423 | GFS2_BLKST_FREE, GFS2_BLKST_DINODE, &n); |
b3b94faa | 1424 | |
6050b9c7 SW |
1425 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
1426 | if (blk == BFITNOENT) | |
1427 | goto rgrp_error; | |
b3b94faa | 1428 | |
6050b9c7 | 1429 | rgd->rd_last_alloc = blk; |
bb8d8a6f | 1430 | block = rgd->rd_data0 + blk; |
05164e5b SW |
1431 | if (rgd->rd_free == 0) |
1432 | goto rgrp_error; | |
1433 | ||
cfc8b549 | 1434 | rgd->rd_free--; |
73f74948 | 1435 | rgd->rd_dinodes++; |
d8b71f73 | 1436 | *generation = rgd->rd_igeneration++; |
8d8291ae SW |
1437 | if (*generation == 0) |
1438 | *generation = rgd->rd_igeneration++; | |
d4e9c4c3 | 1439 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1440 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1441 | |
1442 | al->al_alloced++; | |
1443 | ||
1444 | gfs2_statfs_change(sdp, 0, -1, +1); | |
5731be53 | 1445 | gfs2_trans_add_unrevoke(sdp, block, 1); |
b3b94faa | 1446 | |
b3b94faa | 1447 | rgd->rd_free_clone--; |
63997775 | 1448 | trace_gfs2_block_alloc(dip, block, 1, GFS2_BLKST_DINODE); |
6050b9c7 SW |
1449 | *bn = block; |
1450 | return 0; | |
1451 | ||
1452 | rgrp_error: | |
1453 | gfs2_rgrp_error(rgd); | |
1454 | return -EIO; | |
b3b94faa DT |
1455 | } |
1456 | ||
1457 | /** | |
46fcb2ed | 1458 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
1459 | * @ip: the inode these blocks are being freed from |
1460 | * @bstart: first block of a run of contiguous blocks | |
1461 | * @blen: the length of the block run | |
46fcb2ed | 1462 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
1463 | * |
1464 | */ | |
1465 | ||
46fcb2ed | 1466 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 1467 | { |
feaa7bba | 1468 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1469 | struct gfs2_rgrpd *rgd; |
1470 | ||
1471 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
1472 | if (!rgd) | |
1473 | return; | |
63997775 | 1474 | trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 1475 | rgd->rd_free += blen; |
b3b94faa | 1476 | |
d4e9c4c3 | 1477 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1478 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 1479 | |
6d3117b4 | 1480 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 1481 | if (meta || ip->i_depth) |
6d3117b4 | 1482 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 1483 | } |
b3b94faa | 1484 | |
4c16c36a BP |
1485 | /** |
1486 | * gfs2_free_meta - free a contiguous run of data block(s) | |
1487 | * @ip: the inode these blocks are being freed from | |
1488 | * @bstart: first block of a run of contiguous blocks | |
1489 | * @blen: the length of the block run | |
1490 | * | |
1491 | */ | |
1492 | ||
1493 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
1494 | { | |
1495 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1496 | ||
46fcb2ed | 1497 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 1498 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 1499 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
1500 | } |
1501 | ||
feaa7bba SW |
1502 | void gfs2_unlink_di(struct inode *inode) |
1503 | { | |
1504 | struct gfs2_inode *ip = GFS2_I(inode); | |
1505 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1506 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 1507 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
1508 | |
1509 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
1510 | if (!rgd) | |
1511 | return; | |
63997775 | 1512 | trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED); |
feaa7bba | 1513 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1514 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
feaa7bba SW |
1515 | } |
1516 | ||
cd915493 | 1517 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
1518 | { |
1519 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1520 | struct gfs2_rgrpd *tmp_rgd; | |
1521 | ||
1522 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
1523 | if (!tmp_rgd) | |
1524 | return; | |
1525 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
1526 | ||
73f74948 | 1527 | if (!rgd->rd_dinodes) |
b3b94faa | 1528 | gfs2_consist_rgrpd(rgd); |
73f74948 | 1529 | rgd->rd_dinodes--; |
cfc8b549 | 1530 | rgd->rd_free++; |
b3b94faa | 1531 | |
d4e9c4c3 | 1532 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1533 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1534 | |
1535 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
1536 | } |
1537 | ||
b3b94faa DT |
1538 | |
1539 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
1540 | { | |
dbb7cae2 | 1541 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
63997775 | 1542 | trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 1543 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 1544 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
1545 | } |
1546 | ||
acf7e244 SW |
1547 | /** |
1548 | * gfs2_check_blk_type - Check the type of a block | |
1549 | * @sdp: The superblock | |
1550 | * @no_addr: The block number to check | |
1551 | * @type: The block type we are looking for | |
1552 | * | |
1553 | * Returns: 0 if the block type matches the expected type | |
1554 | * -ESTALE if it doesn't match | |
1555 | * or -ve errno if something went wrong while checking | |
1556 | */ | |
1557 | ||
1558 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
1559 | { | |
1560 | struct gfs2_rgrpd *rgd; | |
1561 | struct gfs2_holder ri_gh, rgd_gh; | |
2c776349 SW |
1562 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
1563 | int ri_locked = 0; | |
acf7e244 SW |
1564 | int error; |
1565 | ||
2c776349 SW |
1566 | if (!gfs2_glock_is_locked_by_me(ip->i_gl)) { |
1567 | error = gfs2_rindex_hold(sdp, &ri_gh); | |
1568 | if (error) | |
1569 | goto fail; | |
1570 | ri_locked = 1; | |
1571 | } | |
acf7e244 SW |
1572 | |
1573 | error = -EINVAL; | |
1574 | rgd = gfs2_blk2rgrpd(sdp, no_addr); | |
1575 | if (!rgd) | |
1576 | goto fail_rindex; | |
1577 | ||
1578 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
1579 | if (error) | |
1580 | goto fail_rindex; | |
1581 | ||
1582 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
1583 | error = -ESTALE; | |
1584 | ||
1585 | gfs2_glock_dq_uninit(&rgd_gh); | |
1586 | fail_rindex: | |
2c776349 SW |
1587 | if (ri_locked) |
1588 | gfs2_glock_dq_uninit(&ri_gh); | |
acf7e244 SW |
1589 | fail: |
1590 | return error; | |
1591 | } | |
1592 | ||
b3b94faa DT |
1593 | /** |
1594 | * gfs2_rlist_add - add a RG to a list of RGs | |
1595 | * @sdp: the filesystem | |
1596 | * @rlist: the list of resource groups | |
1597 | * @block: the block | |
1598 | * | |
1599 | * Figure out what RG a block belongs to and add that RG to the list | |
1600 | * | |
1601 | * FIXME: Don't use NOFAIL | |
1602 | * | |
1603 | */ | |
1604 | ||
1605 | void gfs2_rlist_add(struct gfs2_sbd *sdp, struct gfs2_rgrp_list *rlist, | |
cd915493 | 1606 | u64 block) |
b3b94faa DT |
1607 | { |
1608 | struct gfs2_rgrpd *rgd; | |
1609 | struct gfs2_rgrpd **tmp; | |
1610 | unsigned int new_space; | |
1611 | unsigned int x; | |
1612 | ||
1613 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
1614 | return; | |
1615 | ||
1616 | rgd = gfs2_blk2rgrpd(sdp, block); | |
1617 | if (!rgd) { | |
1618 | if (gfs2_consist(sdp)) | |
382066da | 1619 | fs_err(sdp, "block = %llu\n", (unsigned long long)block); |
b3b94faa DT |
1620 | return; |
1621 | } | |
1622 | ||
1623 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1624 | if (rlist->rl_rgd[x] == rgd) | |
1625 | return; | |
1626 | ||
1627 | if (rlist->rl_rgrps == rlist->rl_space) { | |
1628 | new_space = rlist->rl_space + 10; | |
1629 | ||
1630 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 1631 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1632 | |
1633 | if (rlist->rl_rgd) { | |
1634 | memcpy(tmp, rlist->rl_rgd, | |
1635 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
1636 | kfree(rlist->rl_rgd); | |
1637 | } | |
1638 | ||
1639 | rlist->rl_space = new_space; | |
1640 | rlist->rl_rgd = tmp; | |
1641 | } | |
1642 | ||
1643 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
1644 | } | |
1645 | ||
1646 | /** | |
1647 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
1648 | * and initialize an array of glock holders for them | |
1649 | * @rlist: the list of resource groups | |
1650 | * @state: the lock state to acquire the RG lock in | |
1651 | * @flags: the modifier flags for the holder structures | |
1652 | * | |
1653 | * FIXME: Don't use NOFAIL | |
1654 | * | |
1655 | */ | |
1656 | ||
fe6c991c | 1657 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
1658 | { |
1659 | unsigned int x; | |
1660 | ||
1661 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 1662 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1663 | for (x = 0; x < rlist->rl_rgrps; x++) |
1664 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 1665 | state, 0, |
b3b94faa DT |
1666 | &rlist->rl_ghs[x]); |
1667 | } | |
1668 | ||
1669 | /** | |
1670 | * gfs2_rlist_free - free a resource group list | |
1671 | * @list: the list of resource groups | |
1672 | * | |
1673 | */ | |
1674 | ||
1675 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
1676 | { | |
1677 | unsigned int x; | |
1678 | ||
1679 | kfree(rlist->rl_rgd); | |
1680 | ||
1681 | if (rlist->rl_ghs) { | |
1682 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1683 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
1684 | kfree(rlist->rl_ghs); | |
1685 | } | |
1686 | } | |
1687 |