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 | |
8e2e0047 BP |
38 | #define RSRV_CONTENTION_FACTOR 4 |
39 | #define RGRP_RSRV_MAX_CONTENDERS 2 | |
40 | ||
1f466a47 BP |
41 | #if BITS_PER_LONG == 32 |
42 | #define LBITMASK (0x55555555UL) | |
43 | #define LBITSKIP55 (0x55555555UL) | |
44 | #define LBITSKIP00 (0x00000000UL) | |
45 | #else | |
46 | #define LBITMASK (0x5555555555555555UL) | |
47 | #define LBITSKIP55 (0x5555555555555555UL) | |
48 | #define LBITSKIP00 (0x0000000000000000UL) | |
49 | #endif | |
50 | ||
88c8ab1f SW |
51 | /* |
52 | * These routines are used by the resource group routines (rgrp.c) | |
53 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
54 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
55 | * | |
56 | * 0 = Free | |
57 | * 1 = Used (not metadata) | |
58 | * 2 = Unlinked (still in use) inode | |
59 | * 3 = Used (metadata) | |
88c8ab1f SW |
60 | */ |
61 | ||
62 | static const char valid_change[16] = { | |
63 | /* current */ | |
feaa7bba | 64 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 65 | /* e */ 1, 0, 0, 0, |
feaa7bba | 66 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
67 | 1, 0, 0, 0 |
68 | }; | |
69 | ||
70 | /** | |
71 | * gfs2_setbit - Set a bit in the bitmaps | |
29c578f5 | 72 | * @rgd: the resource group descriptor |
29c578f5 BP |
73 | * @buf2: the clone buffer that holds the bitmaps |
74 | * @bi: the bitmap structure | |
88c8ab1f SW |
75 | * @block: the block to set |
76 | * @new_state: the new state of the block | |
77 | * | |
78 | */ | |
79 | ||
06344b91 BP |
80 | static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf2, |
81 | struct gfs2_bitmap *bi, u32 block, | |
82 | unsigned char new_state) | |
88c8ab1f | 83 | { |
b45e41d7 | 84 | unsigned char *byte1, *byte2, *end, cur_state; |
95c8e17f | 85 | unsigned int buflen = bi->bi_len; |
b45e41d7 | 86 | const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 87 | |
06344b91 BP |
88 | byte1 = bi->bi_bh->b_data + bi->bi_offset + (block / GFS2_NBBY); |
89 | end = bi->bi_bh->b_data + bi->bi_offset + buflen; | |
88c8ab1f | 90 | |
b45e41d7 | 91 | BUG_ON(byte1 >= end); |
88c8ab1f | 92 | |
b45e41d7 | 93 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 94 | |
b45e41d7 | 95 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
95c8e17f BP |
96 | printk(KERN_WARNING "GFS2: buf_blk = 0x%llx old_state=%d, " |
97 | "new_state=%d\n", | |
98 | (unsigned long long)block, cur_state, new_state); | |
99 | printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%lx\n", | |
100 | (unsigned long long)rgd->rd_addr, | |
101 | (unsigned long)bi->bi_start); | |
102 | printk(KERN_WARNING "GFS2: bi_offset=0x%lx bi_len=0x%lx\n", | |
103 | (unsigned long)bi->bi_offset, | |
104 | (unsigned long)bi->bi_len); | |
105 | dump_stack(); | |
88c8ab1f | 106 | gfs2_consist_rgrpd(rgd); |
b45e41d7 SW |
107 | return; |
108 | } | |
109 | *byte1 ^= (cur_state ^ new_state) << bit; | |
110 | ||
111 | if (buf2) { | |
29c578f5 | 112 | byte2 = buf2 + bi->bi_offset + (block / GFS2_NBBY); |
b45e41d7 SW |
113 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
114 | *byte2 ^= (cur_state ^ new_state) << bit; | |
115 | } | |
88c8ab1f SW |
116 | } |
117 | ||
118 | /** | |
119 | * gfs2_testbit - test a bit in the bitmaps | |
886b1416 | 120 | * @rgd: the resource group descriptor |
88c8ab1f SW |
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 | ||
8e2e0047 BP |
179 | /** |
180 | * rs_cmp - multi-block reservation range compare | |
181 | * @blk: absolute file system block number of the new reservation | |
182 | * @len: number of blocks in the new reservation | |
183 | * @rs: existing reservation to compare against | |
184 | * | |
185 | * returns: 1 if the block range is beyond the reach of the reservation | |
186 | * -1 if the block range is before the start of the reservation | |
187 | * 0 if the block range overlaps with the reservation | |
188 | */ | |
189 | static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs) | |
190 | { | |
4a993fb1 | 191 | u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 BP |
192 | |
193 | if (blk >= startblk + rs->rs_free) | |
194 | return 1; | |
195 | if (blk + len - 1 < startblk) | |
196 | return -1; | |
197 | return 0; | |
198 | } | |
199 | ||
88c8ab1f SW |
200 | /** |
201 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
202 | * a block in a given allocation state. | |
886b1416 | 203 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 204 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 205 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 206 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
207 | * |
208 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
209 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
210 | * beginning of a bitmap block buffer, skipping any header structures, but |
211 | * headers are always a multiple of 64 bits long so that the buffer is | |
212 | * always aligned to a 64 bit boundary. | |
213 | * | |
214 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 215 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 216 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
217 | * |
218 | * Return: the block number (bitmap buffer scope) that was found | |
219 | */ | |
220 | ||
02ab1721 HE |
221 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
222 | u32 goal, u8 state) | |
88c8ab1f | 223 | { |
223b2b88 SW |
224 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
225 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
226 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
227 | u64 tmp; | |
075ac448 | 228 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
229 | u32 bit; |
230 | ||
231 | BUG_ON(state > 3); | |
232 | ||
233 | /* Mask off bits we don't care about at the start of the search */ | |
234 | mask <<= spoint; | |
235 | tmp = gfs2_bit_search(ptr, mask, state); | |
236 | ptr++; | |
237 | while(tmp == 0 && ptr < end) { | |
075ac448 | 238 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 239 | ptr++; |
1f466a47 | 240 | } |
223b2b88 SW |
241 | /* Mask off any bits which are more than len bytes from the start */ |
242 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
243 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
244 | /* Didn't find anything, so return */ | |
245 | if (tmp == 0) | |
246 | return BFITNOENT; | |
247 | ptr--; | |
d8bd504a | 248 | bit = __ffs64(tmp); |
223b2b88 SW |
249 | bit /= 2; /* two bits per entry in the bitmap */ |
250 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
251 | } |
252 | ||
253 | /** | |
254 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 255 | * @rgd: the resource group descriptor |
88c8ab1f SW |
256 | * @buffer: the buffer that holds the bitmaps |
257 | * @buflen: the length (in bytes) of the buffer | |
258 | * @state: the state of the block we're looking for | |
259 | * | |
260 | * Returns: The number of bits | |
261 | */ | |
262 | ||
110acf38 SW |
263 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
264 | unsigned int buflen, u8 state) | |
88c8ab1f | 265 | { |
110acf38 SW |
266 | const u8 *byte = buffer; |
267 | const u8 *end = buffer + buflen; | |
268 | const u8 state1 = state << 2; | |
269 | const u8 state2 = state << 4; | |
270 | const u8 state3 = state << 6; | |
cd915493 | 271 | u32 count = 0; |
88c8ab1f SW |
272 | |
273 | for (; byte < end; byte++) { | |
274 | if (((*byte) & 0x03) == state) | |
275 | count++; | |
276 | if (((*byte) & 0x0C) == state1) | |
277 | count++; | |
278 | if (((*byte) & 0x30) == state2) | |
279 | count++; | |
280 | if (((*byte) & 0xC0) == state3) | |
281 | count++; | |
282 | } | |
283 | ||
284 | return count; | |
285 | } | |
286 | ||
b3b94faa DT |
287 | /** |
288 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
289 | * @rgd: the rgrp |
290 | * | |
291 | */ | |
292 | ||
293 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
294 | { | |
295 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
296 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 297 | u32 length = rgd->rd_length; |
cd915493 | 298 | u32 count[4], tmp; |
b3b94faa DT |
299 | int buf, x; |
300 | ||
cd915493 | 301 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
302 | |
303 | /* Count # blocks in each of 4 possible allocation states */ | |
304 | for (buf = 0; buf < length; buf++) { | |
305 | bi = rgd->rd_bits + buf; | |
306 | for (x = 0; x < 4; x++) | |
307 | count[x] += gfs2_bitcount(rgd, | |
308 | bi->bi_bh->b_data + | |
309 | bi->bi_offset, | |
310 | bi->bi_len, x); | |
311 | } | |
312 | ||
cfc8b549 | 313 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
314 | if (gfs2_consist_rgrpd(rgd)) |
315 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 316 | count[0], rgd->rd_free); |
b3b94faa DT |
317 | return; |
318 | } | |
319 | ||
73f74948 | 320 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 321 | if (count[1] != tmp) { |
b3b94faa DT |
322 | if (gfs2_consist_rgrpd(rgd)) |
323 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
324 | count[1], tmp); | |
325 | return; | |
326 | } | |
327 | ||
6b946170 | 328 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 329 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 330 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 331 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
332 | return; |
333 | } | |
b3b94faa DT |
334 | } |
335 | ||
bb8d8a6f | 336 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 337 | { |
bb8d8a6f SW |
338 | u64 first = rgd->rd_data0; |
339 | u64 last = first + rgd->rd_data; | |
16910427 | 340 | return first <= block && block < last; |
b3b94faa DT |
341 | } |
342 | ||
343 | /** | |
344 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
345 | * @sdp: The GFS2 superblock | |
886b1416 BP |
346 | * @blk: The data block number |
347 | * @exact: True if this needs to be an exact match | |
b3b94faa DT |
348 | * |
349 | * Returns: The resource group, or NULL if not found | |
350 | */ | |
351 | ||
66fc061b | 352 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 353 | { |
66fc061b | 354 | struct rb_node *n, *next; |
f75bbfb4 | 355 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
356 | |
357 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
358 | n = sdp->sd_rindex_tree.rb_node; |
359 | while (n) { | |
360 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
361 | next = NULL; | |
7c9ca621 | 362 | if (blk < cur->rd_addr) |
66fc061b | 363 | next = n->rb_left; |
f75bbfb4 | 364 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
365 | next = n->rb_right; |
366 | if (next == NULL) { | |
b3b94faa | 367 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
368 | if (exact) { |
369 | if (blk < cur->rd_addr) | |
370 | return NULL; | |
371 | if (blk >= cur->rd_data0 + cur->rd_data) | |
372 | return NULL; | |
373 | } | |
7c9ca621 | 374 | return cur; |
b3b94faa | 375 | } |
66fc061b | 376 | n = next; |
b3b94faa | 377 | } |
b3b94faa DT |
378 | spin_unlock(&sdp->sd_rindex_spin); |
379 | ||
380 | return NULL; | |
381 | } | |
382 | ||
383 | /** | |
384 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
385 | * @sdp: The GFS2 superblock | |
386 | * | |
387 | * Returns: The first rgrp in the filesystem | |
388 | */ | |
389 | ||
390 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
391 | { | |
7c9ca621 BP |
392 | const struct rb_node *n; |
393 | struct gfs2_rgrpd *rgd; | |
394 | ||
8339ee54 | 395 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
396 | n = rb_first(&sdp->sd_rindex_tree); |
397 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 398 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
399 | |
400 | return rgd; | |
b3b94faa DT |
401 | } |
402 | ||
403 | /** | |
404 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 405 | * @rgd: the resource group descriptor |
b3b94faa DT |
406 | * |
407 | * Returns: The next rgrp | |
408 | */ | |
409 | ||
410 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
411 | { | |
7c9ca621 BP |
412 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
413 | const struct rb_node *n; | |
414 | ||
415 | spin_lock(&sdp->sd_rindex_spin); | |
416 | n = rb_next(&rgd->rd_node); | |
417 | if (n == NULL) | |
418 | n = rb_first(&sdp->sd_rindex_tree); | |
419 | ||
420 | if (unlikely(&rgd->rd_node == n)) { | |
421 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 422 | return NULL; |
7c9ca621 BP |
423 | } |
424 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
425 | spin_unlock(&sdp->sd_rindex_spin); | |
426 | return rgd; | |
b3b94faa DT |
427 | } |
428 | ||
8339ee54 SW |
429 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
430 | { | |
431 | int x; | |
432 | ||
433 | for (x = 0; x < rgd->rd_length; x++) { | |
434 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
435 | kfree(bi->bi_clone); | |
436 | bi->bi_clone = NULL; | |
437 | } | |
438 | } | |
439 | ||
0a305e49 BP |
440 | /** |
441 | * gfs2_rs_alloc - make sure we have a reservation assigned to the inode | |
442 | * @ip: the inode for this reservation | |
443 | */ | |
444 | int gfs2_rs_alloc(struct gfs2_inode *ip) | |
445 | { | |
446 | int error = 0; | |
8e2e0047 BP |
447 | struct gfs2_blkreserv *res; |
448 | ||
449 | if (ip->i_res) | |
450 | return 0; | |
451 | ||
452 | res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); | |
453 | if (!res) | |
454 | error = -ENOMEM; | |
0a305e49 | 455 | |
24d634e8 | 456 | RB_CLEAR_NODE(&res->rs_node); |
4a993fb1 | 457 | |
0a305e49 | 458 | down_write(&ip->i_rw_mutex); |
8e2e0047 BP |
459 | if (ip->i_res) |
460 | kmem_cache_free(gfs2_rsrv_cachep, res); | |
461 | else | |
462 | ip->i_res = res; | |
0a305e49 BP |
463 | up_write(&ip->i_rw_mutex); |
464 | return error; | |
465 | } | |
466 | ||
8e2e0047 BP |
467 | static void dump_rs(struct seq_file *seq, struct gfs2_blkreserv *rs) |
468 | { | |
469 | gfs2_print_dbg(seq, " r: %llu s:%llu b:%u f:%u\n", | |
4a993fb1 SW |
470 | rs->rs_rbm.rgd->rd_addr, gfs2_rbm_to_block(&rs->rs_rbm), |
471 | rs->rs_rbm.offset, rs->rs_free); | |
8e2e0047 BP |
472 | } |
473 | ||
0a305e49 | 474 | /** |
8e2e0047 BP |
475 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
476 | * @rs: The reservation to remove | |
477 | * | |
478 | */ | |
4a993fb1 | 479 | static void __rs_deltree(struct gfs2_inode *ip, struct gfs2_blkreserv *rs) |
8e2e0047 BP |
480 | { |
481 | struct gfs2_rgrpd *rgd; | |
482 | ||
483 | if (!gfs2_rs_active(rs)) | |
484 | return; | |
485 | ||
4a993fb1 SW |
486 | rgd = rs->rs_rbm.rgd; |
487 | trace_gfs2_rs(ip, rs, TRACE_RS_TREEDEL); | |
488 | rb_erase(&rs->rs_node, &rgd->rd_rstree); | |
24d634e8 | 489 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 BP |
490 | BUG_ON(!rgd->rd_rs_cnt); |
491 | rgd->rd_rs_cnt--; | |
492 | ||
493 | if (rs->rs_free) { | |
494 | /* return reserved blocks to the rgrp and the ip */ | |
4a993fb1 SW |
495 | BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free); |
496 | rs->rs_rbm.rgd->rd_reserved -= rs->rs_free; | |
8e2e0047 | 497 | rs->rs_free = 0; |
4a993fb1 | 498 | clear_bit(GBF_FULL, &rs->rs_rbm.bi->bi_flags); |
8e2e0047 BP |
499 | smp_mb__after_clear_bit(); |
500 | } | |
8e2e0047 BP |
501 | } |
502 | ||
503 | /** | |
504 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
505 | * @rs: The reservation to remove | |
506 | * | |
507 | */ | |
4a993fb1 | 508 | void gfs2_rs_deltree(struct gfs2_inode *ip, struct gfs2_blkreserv *rs) |
8e2e0047 BP |
509 | { |
510 | struct gfs2_rgrpd *rgd; | |
511 | ||
4a993fb1 SW |
512 | rgd = rs->rs_rbm.rgd; |
513 | if (rgd) { | |
514 | spin_lock(&rgd->rd_rsspin); | |
515 | __rs_deltree(ip, rs); | |
516 | spin_unlock(&rgd->rd_rsspin); | |
517 | } | |
8e2e0047 BP |
518 | } |
519 | ||
520 | /** | |
521 | * gfs2_rs_delete - delete a multi-block reservation | |
0a305e49 BP |
522 | * @ip: The inode for this reservation |
523 | * | |
524 | */ | |
525 | void gfs2_rs_delete(struct gfs2_inode *ip) | |
526 | { | |
527 | down_write(&ip->i_rw_mutex); | |
528 | if (ip->i_res) { | |
4a993fb1 | 529 | gfs2_rs_deltree(ip, ip->i_res); |
8e2e0047 BP |
530 | trace_gfs2_rs(ip, ip->i_res, TRACE_RS_DELETE); |
531 | BUG_ON(ip->i_res->rs_free); | |
0a305e49 BP |
532 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); |
533 | ip->i_res = NULL; | |
534 | } | |
535 | up_write(&ip->i_rw_mutex); | |
536 | } | |
537 | ||
8e2e0047 BP |
538 | /** |
539 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
540 | * @rgd: the rgrp that needs its space back | |
541 | * | |
542 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
543 | * give them back. This leave the reservation structures in tact, but removes | |
544 | * all of their corresponding "no-fly zones". | |
545 | */ | |
546 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
547 | { | |
548 | struct rb_node *n; | |
549 | struct gfs2_blkreserv *rs; | |
550 | ||
551 | spin_lock(&rgd->rd_rsspin); | |
552 | while ((n = rb_first(&rgd->rd_rstree))) { | |
553 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
4a993fb1 | 554 | __rs_deltree(NULL, rs); |
8e2e0047 BP |
555 | } |
556 | spin_unlock(&rgd->rd_rsspin); | |
557 | } | |
558 | ||
8339ee54 | 559 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 560 | { |
7c9ca621 | 561 | struct rb_node *n; |
b3b94faa DT |
562 | struct gfs2_rgrpd *rgd; |
563 | struct gfs2_glock *gl; | |
564 | ||
7c9ca621 BP |
565 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
566 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
567 | gl = rgd->rd_gl; |
568 | ||
7c9ca621 | 569 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
570 | |
571 | if (gl) { | |
8339ee54 | 572 | spin_lock(&gl->gl_spin); |
5c676f6d | 573 | gl->gl_object = NULL; |
8339ee54 | 574 | spin_unlock(&gl->gl_spin); |
29687a2a | 575 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
576 | gfs2_glock_put(gl); |
577 | } | |
578 | ||
8339ee54 | 579 | gfs2_free_clones(rgd); |
b3b94faa | 580 | kfree(rgd->rd_bits); |
8e2e0047 | 581 | return_all_reservations(rgd); |
6bdd9be6 | 582 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
583 | } |
584 | } | |
585 | ||
bb8d8a6f SW |
586 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
587 | { | |
588 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
589 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
590 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
591 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
592 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
593 | } | |
594 | ||
b3b94faa DT |
595 | /** |
596 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
597 | * @rgd: The resource group descriptor | |
598 | * | |
599 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
600 | * | |
601 | * Returns: errno | |
602 | */ | |
603 | ||
604 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
605 | { | |
606 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
607 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 608 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 609 | u32 bytes_left, bytes; |
b3b94faa DT |
610 | int x; |
611 | ||
feaa7bba SW |
612 | if (!length) |
613 | return -EINVAL; | |
614 | ||
dd894be8 | 615 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
616 | if (!rgd->rd_bits) |
617 | return -ENOMEM; | |
618 | ||
bb8d8a6f | 619 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
620 | |
621 | for (x = 0; x < length; x++) { | |
622 | bi = rgd->rd_bits + x; | |
623 | ||
60a0b8f9 | 624 | bi->bi_flags = 0; |
b3b94faa DT |
625 | /* small rgrp; bitmap stored completely in header block */ |
626 | if (length == 1) { | |
627 | bytes = bytes_left; | |
628 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
629 | bi->bi_start = 0; | |
630 | bi->bi_len = bytes; | |
631 | /* header block */ | |
632 | } else if (x == 0) { | |
633 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
634 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
635 | bi->bi_start = 0; | |
636 | bi->bi_len = bytes; | |
637 | /* last block */ | |
638 | } else if (x + 1 == length) { | |
639 | bytes = bytes_left; | |
640 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 641 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
642 | bi->bi_len = bytes; |
643 | /* other blocks */ | |
644 | } else { | |
568f4c96 SW |
645 | bytes = sdp->sd_sb.sb_bsize - |
646 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 647 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 648 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
649 | bi->bi_len = bytes; |
650 | } | |
651 | ||
652 | bytes_left -= bytes; | |
653 | } | |
654 | ||
655 | if (bytes_left) { | |
656 | gfs2_consist_rgrpd(rgd); | |
657 | return -EIO; | |
658 | } | |
659 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 660 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 661 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 662 | gfs2_rindex_print(rgd); |
b3b94faa DT |
663 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
664 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
665 | } | |
666 | return -EIO; | |
667 | } | |
668 | ||
669 | return 0; | |
670 | } | |
671 | ||
7ae8fa84 RP |
672 | /** |
673 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 674 | * @sdp: the filesystem |
7ae8fa84 RP |
675 | * |
676 | */ | |
677 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
678 | { | |
679 | u64 total_data = 0; | |
680 | struct inode *inode = sdp->sd_rindex; | |
681 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 682 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
683 | int error, rgrps; |
684 | ||
7ae8fa84 RP |
685 | for (rgrps = 0;; rgrps++) { |
686 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
687 | ||
bcd7278d | 688 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 689 | break; |
4306629e | 690 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
691 | sizeof(struct gfs2_rindex)); |
692 | if (error != sizeof(struct gfs2_rindex)) | |
693 | break; | |
bb8d8a6f | 694 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 695 | } |
7ae8fa84 RP |
696 | return total_data; |
697 | } | |
698 | ||
6aad1c3d | 699 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
700 | { |
701 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
702 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
703 | ||
704 | /* Figure out where to put new node */ | |
705 | while (*newn) { | |
706 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
707 | rd_node); | |
708 | ||
709 | parent = *newn; | |
710 | if (rgd->rd_addr < cur->rd_addr) | |
711 | newn = &((*newn)->rb_left); | |
712 | else if (rgd->rd_addr > cur->rd_addr) | |
713 | newn = &((*newn)->rb_right); | |
714 | else | |
6aad1c3d | 715 | return -EEXIST; |
7c9ca621 BP |
716 | } |
717 | ||
718 | rb_link_node(&rgd->rd_node, parent, newn); | |
719 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
720 | sdp->sd_rgrps++; |
721 | return 0; | |
7c9ca621 BP |
722 | } |
723 | ||
b3b94faa | 724 | /** |
6c53267f | 725 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 726 | * @ip: Pointer to the rindex inode |
b3b94faa | 727 | * |
8339ee54 | 728 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
729 | */ |
730 | ||
4306629e | 731 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
732 | { |
733 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
734 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
8339ee54 | 735 | struct gfs2_rindex buf; |
6c53267f RP |
736 | int error; |
737 | struct gfs2_rgrpd *rgd; | |
738 | ||
8339ee54 SW |
739 | if (pos >= i_size_read(&ip->i_inode)) |
740 | return 1; | |
741 | ||
4306629e | 742 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 743 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
744 | |
745 | if (error != sizeof(struct gfs2_rindex)) | |
746 | return (error == 0) ? 1 : error; | |
6c53267f | 747 | |
6bdd9be6 | 748 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
749 | error = -ENOMEM; |
750 | if (!rgd) | |
751 | return error; | |
752 | ||
6c53267f | 753 | rgd->rd_sbd = sdp; |
8339ee54 SW |
754 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
755 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
756 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
757 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
758 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 759 | spin_lock_init(&rgd->rd_rsspin); |
7c9ca621 | 760 | |
6c53267f RP |
761 | error = compute_bitstructs(rgd); |
762 | if (error) | |
8339ee54 | 763 | goto fail; |
6c53267f | 764 | |
bb8d8a6f | 765 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
766 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
767 | if (error) | |
8339ee54 | 768 | goto fail; |
6c53267f RP |
769 | |
770 | rgd->rd_gl->gl_object = rgd; | |
90306c41 | 771 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lvb; |
cf45b752 | 772 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
7c9ca621 BP |
773 | if (rgd->rd_data > sdp->sd_max_rg_data) |
774 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 775 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 776 | error = rgd_insert(rgd); |
8339ee54 | 777 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
778 | if (!error) |
779 | return 0; | |
780 | ||
781 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 782 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
783 | |
784 | fail: | |
785 | kfree(rgd->rd_bits); | |
786 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
787 | return error; |
788 | } | |
789 | ||
790 | /** | |
791 | * gfs2_ri_update - Pull in a new resource index from the disk | |
792 | * @ip: pointer to the rindex inode | |
793 | * | |
b3b94faa DT |
794 | * Returns: 0 on successful update, error code otherwise |
795 | */ | |
796 | ||
8339ee54 | 797 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 798 | { |
feaa7bba | 799 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
800 | int error; |
801 | ||
8339ee54 | 802 | do { |
4306629e | 803 | error = read_rindex_entry(ip); |
8339ee54 SW |
804 | } while (error == 0); |
805 | ||
806 | if (error < 0) | |
807 | return error; | |
b3b94faa | 808 | |
cf45b752 | 809 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
810 | return 0; |
811 | } | |
b3b94faa | 812 | |
b3b94faa | 813 | /** |
8339ee54 | 814 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 815 | * @sdp: The GFS2 superblock |
b3b94faa DT |
816 | * |
817 | * We grab a lock on the rindex inode to make sure that it doesn't | |
818 | * change whilst we are performing an operation. We keep this lock | |
819 | * for quite long periods of time compared to other locks. This | |
820 | * doesn't matter, since it is shared and it is very, very rarely | |
821 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
822 | * | |
823 | * This makes sure that we're using the latest copy of the resource index | |
824 | * special file, which might have been updated if someone expanded the | |
825 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
826 | * | |
8339ee54 | 827 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
828 | */ |
829 | ||
8339ee54 | 830 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 831 | { |
feaa7bba | 832 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 833 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
834 | struct gfs2_holder ri_gh; |
835 | int error = 0; | |
a365fbf3 | 836 | int unlock_required = 0; |
b3b94faa DT |
837 | |
838 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 839 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
840 | if (!gfs2_glock_is_locked_by_me(gl)) { |
841 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
842 | if (error) | |
6aad1c3d | 843 | return error; |
a365fbf3 SW |
844 | unlock_required = 1; |
845 | } | |
8339ee54 | 846 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 847 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
848 | if (unlock_required) |
849 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
850 | } |
851 | ||
852 | return error; | |
853 | } | |
854 | ||
42d52e38 | 855 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
856 | { |
857 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 858 | u32 rg_flags; |
bb8d8a6f | 859 | |
42d52e38 | 860 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 861 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
862 | rgd->rd_flags &= GFS2_RDF_MASK; |
863 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 864 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 865 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 866 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
867 | } |
868 | ||
42d52e38 | 869 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
870 | { |
871 | struct gfs2_rgrp *str = buf; | |
872 | ||
09010978 | 873 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 874 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 875 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 876 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 877 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
878 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
879 | } | |
880 | ||
90306c41 BM |
881 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
882 | { | |
883 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
884 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
885 | ||
886 | if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free || | |
887 | rgl->rl_dinodes != str->rg_dinodes || | |
888 | rgl->rl_igeneration != str->rg_igeneration) | |
889 | return 0; | |
890 | return 1; | |
891 | } | |
892 | ||
893 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) | |
894 | { | |
895 | const struct gfs2_rgrp *str = buf; | |
896 | ||
897 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
898 | rgl->rl_flags = str->rg_flags; | |
899 | rgl->rl_free = str->rg_free; | |
900 | rgl->rl_dinodes = str->rg_dinodes; | |
901 | rgl->rl_igeneration = str->rg_igeneration; | |
902 | rgl->__pad = 0UL; | |
903 | } | |
904 | ||
905 | static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change) | |
906 | { | |
907 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
908 | u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change; | |
909 | rgl->rl_unlinked = cpu_to_be32(unlinked); | |
910 | } | |
911 | ||
912 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) | |
913 | { | |
914 | struct gfs2_bitmap *bi; | |
915 | const u32 length = rgd->rd_length; | |
916 | const u8 *buffer = NULL; | |
917 | u32 i, goal, count = 0; | |
918 | ||
919 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
920 | goal = 0; | |
921 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
922 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
923 | while (goal < bi->bi_len * GFS2_NBBY) { | |
924 | goal = gfs2_bitfit(buffer, bi->bi_len, goal, | |
925 | GFS2_BLKST_UNLINKED); | |
926 | if (goal == BFITNOENT) | |
927 | break; | |
928 | count++; | |
929 | goal++; | |
930 | } | |
931 | } | |
932 | ||
933 | return count; | |
934 | } | |
935 | ||
936 | ||
b3b94faa | 937 | /** |
90306c41 BM |
938 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
939 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
940 | * |
941 | * Read in all of a Resource Group's header and bitmap blocks. | |
942 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
943 | * | |
944 | * Returns: errno | |
945 | */ | |
946 | ||
90306c41 | 947 | int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
948 | { |
949 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
950 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 951 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
952 | struct gfs2_bitmap *bi; |
953 | unsigned int x, y; | |
954 | int error; | |
955 | ||
90306c41 BM |
956 | if (rgd->rd_bits[0].bi_bh != NULL) |
957 | return 0; | |
958 | ||
b3b94faa DT |
959 | for (x = 0; x < length; x++) { |
960 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 961 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
962 | if (error) |
963 | goto fail; | |
964 | } | |
965 | ||
966 | for (y = length; y--;) { | |
967 | bi = rgd->rd_bits + y; | |
7276b3b0 | 968 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
969 | if (error) |
970 | goto fail; | |
feaa7bba | 971 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
972 | GFS2_METATYPE_RG)) { |
973 | error = -EIO; | |
974 | goto fail; | |
975 | } | |
976 | } | |
977 | ||
cf45b752 | 978 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
979 | for (x = 0; x < length; x++) |
980 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 981 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 982 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 983 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa | 984 | } |
90306c41 BM |
985 | if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
986 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); | |
987 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
988 | rgd->rd_bits[0].bi_bh->b_data); | |
989 | } | |
990 | else if (sdp->sd_args.ar_rgrplvb) { | |
991 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
992 | gfs2_consist_rgrpd(rgd); | |
993 | error = -EIO; | |
994 | goto fail; | |
995 | } | |
996 | if (rgd->rd_rgl->rl_unlinked == 0) | |
997 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
998 | } | |
b3b94faa DT |
999 | return 0; |
1000 | ||
feaa7bba | 1001 | fail: |
b3b94faa DT |
1002 | while (x--) { |
1003 | bi = rgd->rd_bits + x; | |
1004 | brelse(bi->bi_bh); | |
1005 | bi->bi_bh = NULL; | |
1006 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
1007 | } | |
b3b94faa DT |
1008 | |
1009 | return error; | |
1010 | } | |
1011 | ||
90306c41 BM |
1012 | int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
1013 | { | |
1014 | u32 rl_flags; | |
1015 | ||
1016 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1017 | return 0; | |
1018 | ||
1019 | if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) | |
1020 | return gfs2_rgrp_bh_get(rgd); | |
1021 | ||
1022 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1023 | rl_flags &= ~GFS2_RDF_MASK; | |
1024 | rgd->rd_flags &= GFS2_RDF_MASK; | |
1025 | rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); | |
1026 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1027 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1028 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
1029 | rgd->rd_free_clone = rgd->rd_free; | |
1030 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); | |
1031 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1032 | return 0; | |
1033 | } | |
1034 | ||
1035 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1036 | { | |
1037 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1038 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1039 | ||
1040 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1041 | return 0; | |
1042 | return gfs2_rgrp_bh_get((struct gfs2_rgrpd *)gh->gh_gl->gl_object); | |
1043 | } | |
1044 | ||
b3b94faa | 1045 | /** |
7c9ca621 | 1046 | * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
886b1416 | 1047 | * @gh: The glock holder for the resource group |
b3b94faa DT |
1048 | * |
1049 | */ | |
1050 | ||
7c9ca621 | 1051 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) |
b3b94faa | 1052 | { |
7c9ca621 | 1053 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
bb8d8a6f | 1054 | int x, length = rgd->rd_length; |
b3b94faa | 1055 | |
b3b94faa DT |
1056 | for (x = 0; x < length; x++) { |
1057 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1058 | if (bi->bi_bh) { |
1059 | brelse(bi->bi_bh); | |
1060 | bi->bi_bh = NULL; | |
1061 | } | |
b3b94faa DT |
1062 | } |
1063 | ||
b3b94faa DT |
1064 | } |
1065 | ||
66fc061b | 1066 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1067 | struct buffer_head *bh, |
66fc061b | 1068 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1069 | { |
1070 | struct super_block *sb = sdp->sd_vfs; | |
1071 | struct block_device *bdev = sb->s_bdev; | |
1072 | const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / | |
e1defc4f | 1073 | bdev_logical_block_size(sb->s_bdev); |
f15ab561 | 1074 | u64 blk; |
64d576ba | 1075 | sector_t start = 0; |
f15ab561 SW |
1076 | sector_t nr_sects = 0; |
1077 | int rv; | |
1078 | unsigned int x; | |
66fc061b SW |
1079 | u32 trimmed = 0; |
1080 | u8 diff; | |
f15ab561 SW |
1081 | |
1082 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
1083 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1084 | clone += bi->bi_offset; | |
1085 | clone += x; | |
1086 | if (bh) { | |
1087 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1088 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1089 | } else { | |
1090 | diff = ~(*clone | (*clone >> 1)); | |
1091 | } | |
f15ab561 SW |
1092 | diff &= 0x55; |
1093 | if (diff == 0) | |
1094 | continue; | |
1095 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
1096 | blk *= sects_per_blk; /* convert to sectors */ | |
1097 | while(diff) { | |
1098 | if (diff & 1) { | |
1099 | if (nr_sects == 0) | |
1100 | goto start_new_extent; | |
1101 | if ((start + nr_sects) != blk) { | |
66fc061b SW |
1102 | if (nr_sects >= minlen) { |
1103 | rv = blkdev_issue_discard(bdev, | |
1104 | start, nr_sects, | |
1105 | GFP_NOFS, 0); | |
1106 | if (rv) | |
1107 | goto fail; | |
1108 | trimmed += nr_sects; | |
1109 | } | |
f15ab561 SW |
1110 | nr_sects = 0; |
1111 | start_new_extent: | |
1112 | start = blk; | |
1113 | } | |
1114 | nr_sects += sects_per_blk; | |
1115 | } | |
1116 | diff >>= 2; | |
1117 | blk += sects_per_blk; | |
1118 | } | |
1119 | } | |
66fc061b | 1120 | if (nr_sects >= minlen) { |
dd3932ed | 1121 | rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0); |
f15ab561 SW |
1122 | if (rv) |
1123 | goto fail; | |
66fc061b | 1124 | trimmed += nr_sects; |
f15ab561 | 1125 | } |
66fc061b SW |
1126 | if (ptrimmed) |
1127 | *ptrimmed = trimmed; | |
1128 | return 0; | |
1129 | ||
f15ab561 | 1130 | fail: |
66fc061b SW |
1131 | if (sdp->sd_args.ar_discard) |
1132 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 1133 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1134 | return -EIO; |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1139 | * @filp: Any file on the filesystem | |
1140 | * @argp: Pointer to the arguments (also used to pass result) | |
1141 | * | |
1142 | * Returns: 0 on success, otherwise error code | |
1143 | */ | |
1144 | ||
1145 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1146 | { | |
1147 | struct inode *inode = filp->f_dentry->d_inode; | |
1148 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1149 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1150 | struct buffer_head *bh; | |
1151 | struct gfs2_rgrpd *rgd; | |
1152 | struct gfs2_rgrpd *rgd_end; | |
1153 | struct gfs2_holder gh; | |
1154 | struct fstrim_range r; | |
1155 | int ret = 0; | |
1156 | u64 amt; | |
1157 | u64 trimmed = 0; | |
1158 | unsigned int x; | |
1159 | ||
1160 | if (!capable(CAP_SYS_ADMIN)) | |
1161 | return -EPERM; | |
1162 | ||
1163 | if (!blk_queue_discard(q)) | |
1164 | return -EOPNOTSUPP; | |
1165 | ||
66fc061b SW |
1166 | if (argp == NULL) { |
1167 | r.start = 0; | |
1168 | r.len = ULLONG_MAX; | |
1169 | r.minlen = 0; | |
1170 | } else if (copy_from_user(&r, argp, sizeof(r))) | |
1171 | return -EFAULT; | |
1172 | ||
5e2f7d61 BP |
1173 | ret = gfs2_rindex_update(sdp); |
1174 | if (ret) | |
1175 | return ret; | |
1176 | ||
66fc061b SW |
1177 | rgd = gfs2_blk2rgrpd(sdp, r.start, 0); |
1178 | rgd_end = gfs2_blk2rgrpd(sdp, r.start + r.len, 0); | |
1179 | ||
1180 | while (1) { | |
1181 | ||
1182 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1183 | if (ret) | |
1184 | goto out; | |
1185 | ||
1186 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1187 | /* Trim each bitmap in the rgrp */ | |
1188 | for (x = 0; x < rgd->rd_length; x++) { | |
1189 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
1190 | ret = gfs2_rgrp_send_discards(sdp, rgd->rd_data0, NULL, bi, r.minlen, &amt); | |
1191 | if (ret) { | |
1192 | gfs2_glock_dq_uninit(&gh); | |
1193 | goto out; | |
1194 | } | |
1195 | trimmed += amt; | |
1196 | } | |
1197 | ||
1198 | /* Mark rgrp as having been trimmed */ | |
1199 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1200 | if (ret == 0) { | |
1201 | bh = rgd->rd_bits[0].bi_bh; | |
1202 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
1203 | gfs2_trans_add_bh(rgd->rd_gl, bh, 1); | |
1204 | gfs2_rgrp_out(rgd, bh->b_data); | |
90306c41 | 1205 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data); |
66fc061b SW |
1206 | gfs2_trans_end(sdp); |
1207 | } | |
1208 | } | |
1209 | gfs2_glock_dq_uninit(&gh); | |
1210 | ||
1211 | if (rgd == rgd_end) | |
1212 | break; | |
1213 | ||
1214 | rgd = gfs2_rgrpd_get_next(rgd); | |
1215 | } | |
1216 | ||
1217 | out: | |
1218 | r.len = trimmed << 9; | |
1219 | if (argp && copy_to_user(argp, &r, sizeof(r))) | |
1220 | return -EFAULT; | |
1221 | ||
1222 | return ret; | |
f15ab561 SW |
1223 | } |
1224 | ||
8e2e0047 BP |
1225 | /** |
1226 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
1227 | * @bi: the bitmap with the blocks | |
1228 | * @ip: the inode structure | |
1229 | * @biblk: the 32-bit block number relative to the start of the bitmap | |
1230 | * @amount: the number of blocks to reserve | |
1231 | * | |
1232 | * Returns: NULL - reservation was already taken, so not inserted | |
1233 | * pointer to the inserted reservation | |
1234 | */ | |
1235 | static struct gfs2_blkreserv *rs_insert(struct gfs2_bitmap *bi, | |
1236 | struct gfs2_inode *ip, u32 biblk, | |
1237 | int amount) | |
1238 | { | |
1239 | struct rb_node **newn, *parent = NULL; | |
1240 | int rc; | |
1241 | struct gfs2_blkreserv *rs = ip->i_res; | |
4a993fb1 | 1242 | struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd; |
8e2e0047 BP |
1243 | u64 fsblock = gfs2_bi2rgd_blk(bi, biblk) + rgd->rd_data0; |
1244 | ||
1245 | spin_lock(&rgd->rd_rsspin); | |
1246 | newn = &rgd->rd_rstree.rb_node; | |
1247 | BUG_ON(!ip->i_res); | |
1248 | BUG_ON(gfs2_rs_active(rs)); | |
1249 | /* Figure out where to put new node */ | |
1250 | /*BUG_ON(!gfs2_glock_is_locked_by_me(rgd->rd_gl));*/ | |
1251 | while (*newn) { | |
1252 | struct gfs2_blkreserv *cur = | |
1253 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1254 | ||
1255 | parent = *newn; | |
1256 | rc = rs_cmp(fsblock, amount, cur); | |
1257 | if (rc > 0) | |
1258 | newn = &((*newn)->rb_right); | |
1259 | else if (rc < 0) | |
1260 | newn = &((*newn)->rb_left); | |
1261 | else { | |
1262 | spin_unlock(&rgd->rd_rsspin); | |
1263 | return NULL; /* reservation already in use */ | |
1264 | } | |
1265 | } | |
1266 | ||
1267 | /* Do our reservation work */ | |
1268 | rs = ip->i_res; | |
1269 | rs->rs_free = amount; | |
4a993fb1 SW |
1270 | rs->rs_rbm.offset = biblk; |
1271 | rs->rs_rbm.bi = bi; | |
8e2e0047 BP |
1272 | rb_link_node(&rs->rs_node, parent, newn); |
1273 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1274 | ||
8e2e0047 BP |
1275 | /* Do our rgrp accounting for the reservation */ |
1276 | rgd->rd_reserved += amount; /* blocks reserved */ | |
1277 | rgd->rd_rs_cnt++; /* number of in-tree reservations */ | |
1278 | spin_unlock(&rgd->rd_rsspin); | |
1279 | trace_gfs2_rs(ip, rs, TRACE_RS_INSERT); | |
1280 | return rs; | |
1281 | } | |
1282 | ||
1283 | /** | |
1284 | * unclaimed_blocks - return number of blocks that aren't spoken for | |
1285 | */ | |
1286 | static u32 unclaimed_blocks(struct gfs2_rgrpd *rgd) | |
1287 | { | |
1288 | return rgd->rd_free_clone - rgd->rd_reserved; | |
1289 | } | |
1290 | ||
1291 | /** | |
1292 | * rg_mblk_search - find a group of multiple free blocks | |
1293 | * @rgd: the resource group descriptor | |
1294 | * @rs: the block reservation | |
1295 | * @ip: pointer to the inode for which we're reserving blocks | |
1296 | * | |
1297 | * This is very similar to rgblk_search, except we're looking for whole | |
1298 | * 64-bit words that represent a chunk of 32 free blocks. I'm only focusing | |
1299 | * on aligned dwords for speed's sake. | |
1300 | * | |
1301 | * Returns: 0 if successful or BFITNOENT if there isn't enough free space | |
1302 | */ | |
1303 | ||
71f890f7 | 1304 | static int rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, unsigned requested) |
8e2e0047 BP |
1305 | { |
1306 | struct gfs2_bitmap *bi = rgd->rd_bits; | |
1307 | const u32 length = rgd->rd_length; | |
1308 | u32 blk; | |
1309 | unsigned int buf, x, search_bytes; | |
1310 | u8 *buffer = NULL; | |
1311 | u8 *ptr, *end, *nonzero; | |
1312 | u32 goal, rsv_bytes; | |
1313 | struct gfs2_blkreserv *rs; | |
1314 | u32 best_rs_bytes, unclaimed; | |
1315 | int best_rs_blocks; | |
1316 | ||
1317 | /* Find bitmap block that contains bits for goal block */ | |
1318 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
1319 | goal = ip->i_goal - rgd->rd_data0; | |
1320 | else | |
1321 | goal = rgd->rd_last_alloc; | |
1322 | for (buf = 0; buf < length; buf++) { | |
1323 | bi = rgd->rd_bits + buf; | |
1324 | /* Convert scope of "goal" from rgrp-wide to within | |
1325 | found bit block */ | |
1326 | if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { | |
1327 | goal -= bi->bi_start * GFS2_NBBY; | |
1328 | goto do_search; | |
1329 | } | |
1330 | } | |
1331 | buf = 0; | |
1332 | goal = 0; | |
1333 | ||
1334 | do_search: | |
1335 | best_rs_blocks = max_t(int, atomic_read(&ip->i_res->rs_sizehint), | |
1336 | (RGRP_RSRV_MINBLKS * rgd->rd_length)); | |
1337 | best_rs_bytes = (best_rs_blocks * | |
1338 | (1 + (RSRV_CONTENTION_FACTOR * rgd->rd_rs_cnt))) / | |
1339 | GFS2_NBBY; /* 1 + is for our not-yet-created reservation */ | |
1340 | best_rs_bytes = ALIGN(best_rs_bytes, sizeof(u64)); | |
1341 | unclaimed = unclaimed_blocks(rgd); | |
1342 | if (best_rs_bytes * GFS2_NBBY > unclaimed) | |
1343 | best_rs_bytes = unclaimed >> GFS2_BIT_SIZE; | |
1344 | ||
1345 | for (x = 0; x <= length; x++) { | |
1346 | bi = rgd->rd_bits + buf; | |
1347 | ||
1348 | if (test_bit(GBF_FULL, &bi->bi_flags)) | |
1349 | goto skip; | |
1350 | ||
1351 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
1352 | if (bi->bi_clone) | |
1353 | buffer = bi->bi_clone + bi->bi_offset; | |
1354 | else | |
1355 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1356 | ||
1357 | /* We have to keep the reservations aligned on u64 boundaries | |
1358 | otherwise we could get situations where a byte can't be | |
1359 | used because it's after a reservation, but a free bit still | |
1360 | is within the reservation's area. */ | |
1361 | ptr = buffer + ALIGN(goal >> GFS2_BIT_SIZE, sizeof(u64)); | |
1362 | end = (buffer + bi->bi_len); | |
1363 | while (ptr < end) { | |
1364 | rsv_bytes = 0; | |
1365 | if ((ptr + best_rs_bytes) <= end) | |
1366 | search_bytes = best_rs_bytes; | |
1367 | else | |
1368 | search_bytes = end - ptr; | |
1369 | BUG_ON(!search_bytes); | |
1370 | nonzero = memchr_inv(ptr, 0, search_bytes); | |
1371 | /* If the lot is all zeroes, reserve the whole size. If | |
1372 | there's enough zeroes to satisfy the request, use | |
1373 | what we can. If there's not enough, keep looking. */ | |
1374 | if (nonzero == NULL) | |
1375 | rsv_bytes = search_bytes; | |
71f890f7 | 1376 | else if ((nonzero - ptr) * GFS2_NBBY >= requested) |
8e2e0047 BP |
1377 | rsv_bytes = (nonzero - ptr); |
1378 | ||
1379 | if (rsv_bytes) { | |
1380 | blk = ((ptr - buffer) * GFS2_NBBY); | |
1381 | BUG_ON(blk >= bi->bi_len * GFS2_NBBY); | |
1382 | rs = rs_insert(bi, ip, blk, | |
1383 | rsv_bytes * GFS2_NBBY); | |
1384 | if (IS_ERR(rs)) | |
1385 | return PTR_ERR(rs); | |
1386 | if (rs) | |
1387 | return 0; | |
1388 | } | |
1389 | ptr += ALIGN(search_bytes, sizeof(u64)); | |
1390 | } | |
1391 | skip: | |
1392 | /* Try next bitmap block (wrap back to rgrp header | |
1393 | if at end) */ | |
1394 | buf++; | |
1395 | buf %= length; | |
1396 | goal = 0; | |
1397 | } | |
1398 | ||
1399 | return BFITNOENT; | |
1400 | } | |
1401 | ||
b3b94faa DT |
1402 | /** |
1403 | * try_rgrp_fit - See if a given reservation will fit in a given RG | |
1404 | * @rgd: the RG data | |
54335b1f | 1405 | * @ip: the inode |
b3b94faa DT |
1406 | * |
1407 | * If there's room for the requested blocks to be allocated from the RG: | |
8e2e0047 BP |
1408 | * This will try to get a multi-block reservation first, and if that doesn't |
1409 | * fit, it will take what it can. | |
b3b94faa DT |
1410 | * |
1411 | * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) | |
1412 | */ | |
1413 | ||
71f890f7 SW |
1414 | static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
1415 | unsigned requested) | |
b3b94faa | 1416 | { |
09010978 | 1417 | if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) |
a43a4906 | 1418 | return 0; |
8e2e0047 BP |
1419 | /* Look for a multi-block reservation. */ |
1420 | if (unclaimed_blocks(rgd) >= RGRP_RSRV_MINBLKS && | |
71f890f7 | 1421 | rg_mblk_search(rgd, ip, requested) != BFITNOENT) |
8e2e0047 | 1422 | return 1; |
71f890f7 | 1423 | if (unclaimed_blocks(rgd) >= requested) |
7c9ca621 | 1424 | return 1; |
b3b94faa | 1425 | |
8e2e0047 | 1426 | return 0; |
b3e47ca0 BP |
1427 | } |
1428 | ||
5b924ae2 SW |
1429 | /** |
1430 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1431 | * @rgd: The resource group | |
1432 | * @block: The starting block | |
1433 | * @ip: Ignore any reservations for this inode | |
1434 | * | |
1435 | * If the block does not appear in any reservation, then return the | |
1436 | * block number unchanged. If it does appear in the reservation, then | |
1437 | * keep looking through the tree of reservations in order to find the | |
1438 | * first block number which is not reserved. | |
1439 | */ | |
1440 | ||
1441 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
1442 | const struct gfs2_inode *ip) | |
1443 | { | |
1444 | struct gfs2_blkreserv *rs; | |
1445 | struct rb_node *n; | |
1446 | int rc; | |
1447 | ||
1448 | spin_lock(&rgd->rd_rsspin); | |
1449 | n = rb_first(&rgd->rd_rstree); | |
1450 | while (n) { | |
1451 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1452 | rc = rs_cmp(block, 1, rs); | |
1453 | if (rc < 0) | |
1454 | n = n->rb_left; | |
1455 | else if (rc > 0) | |
1456 | n = n->rb_right; | |
1457 | else | |
1458 | break; | |
1459 | } | |
1460 | ||
1461 | if (n) { | |
1462 | while ((rs_cmp(block, 1, rs) == 0) && (ip->i_res != rs)) { | |
1463 | block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free; | |
1464 | n = rb_next(&rs->rs_node); | |
1465 | if (n == NULL) | |
1466 | break; | |
1467 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1468 | } | |
1469 | } | |
1470 | ||
1471 | spin_unlock(&rgd->rd_rsspin); | |
1472 | return block; | |
1473 | } | |
1474 | ||
1475 | /** | |
1476 | * gfs2_rbm_from_block - Set the rbm based upon rgd and block number | |
1477 | * @rbm: The rbm with rgd already set correctly | |
1478 | * @block: The block number (filesystem relative) | |
1479 | * | |
1480 | * This sets the bi and offset members of an rbm based on a | |
1481 | * resource group and a filesystem relative block number. The | |
1482 | * resource group must be set in the rbm on entry, the bi and | |
1483 | * offset members will be set by this function. | |
1484 | * | |
1485 | * Returns: 0 on success, or an error code | |
1486 | */ | |
1487 | ||
1488 | static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block) | |
1489 | { | |
1490 | u64 rblock = block - rbm->rgd->rd_data0; | |
1491 | u32 goal = (u32)rblock; | |
1492 | int x; | |
1493 | ||
1494 | if (WARN_ON_ONCE(rblock > UINT_MAX)) | |
1495 | return -EINVAL; | |
1496 | ||
1497 | for (x = 0; x < rbm->rgd->rd_length; x++) { | |
1498 | rbm->bi = rbm->rgd->rd_bits + x; | |
1499 | if (goal < (rbm->bi->bi_start + rbm->bi->bi_len) * GFS2_NBBY) { | |
1500 | rbm->offset = goal - (rbm->bi->bi_start * GFS2_NBBY); | |
1501 | return 0; | |
1502 | } | |
1503 | } | |
1504 | ||
1505 | return -E2BIG; | |
1506 | } | |
1507 | ||
1508 | /** | |
1509 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1510 | * @rbm: The current position in the resource group | |
1511 | * | |
1512 | * This checks the current position in the rgrp to see whether there is | |
1513 | * a reservation covering this block. If not then this function is a | |
1514 | * no-op. If there is, then the position is moved to the end of the | |
1515 | * contiguous reservation(s) so that we are pointing at the first | |
1516 | * non-reserved block. | |
1517 | * | |
1518 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1519 | */ | |
1520 | ||
1521 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
1522 | const struct gfs2_inode *ip) | |
1523 | { | |
1524 | u64 block = gfs2_rbm_to_block(rbm); | |
1525 | u64 nblock; | |
1526 | int ret; | |
1527 | ||
1528 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, ip); | |
1529 | if (nblock == block) | |
1530 | return 0; | |
1531 | ret = gfs2_rbm_from_block(rbm, nblock); | |
1532 | if (ret < 0) | |
1533 | return ret; | |
1534 | return 1; | |
1535 | } | |
1536 | ||
1537 | /** | |
1538 | * gfs2_rbm_find - Look for blocks of a particular state | |
1539 | * @rbm: Value/result starting position and final position | |
1540 | * @state: The state which we want to find | |
1541 | * @ip: If set, check for reservations | |
1542 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping | |
1543 | * around until we've reached the starting point. | |
1544 | * | |
1545 | * Side effects: | |
1546 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1547 | * has no free blocks in it. | |
1548 | * | |
1549 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1550 | */ | |
1551 | ||
1552 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, | |
1553 | const struct gfs2_inode *ip, bool nowrap) | |
1554 | { | |
1555 | struct buffer_head *bh; | |
1556 | struct gfs2_bitmap *initial_bi; | |
1557 | u32 initial_offset; | |
1558 | u32 offset; | |
1559 | u8 *buffer; | |
1560 | int index; | |
1561 | int n = 0; | |
1562 | int iters = rbm->rgd->rd_length; | |
1563 | int ret; | |
1564 | ||
1565 | /* If we are not starting at the beginning of a bitmap, then we | |
1566 | * need to add one to the bitmap count to ensure that we search | |
1567 | * the starting bitmap twice. | |
1568 | */ | |
1569 | if (rbm->offset != 0) | |
1570 | iters++; | |
1571 | ||
1572 | while(1) { | |
1573 | if (test_bit(GBF_FULL, &rbm->bi->bi_flags) && | |
1574 | (state == GFS2_BLKST_FREE)) | |
1575 | goto next_bitmap; | |
1576 | ||
1577 | bh = rbm->bi->bi_bh; | |
1578 | buffer = bh->b_data + rbm->bi->bi_offset; | |
1579 | WARN_ON(!buffer_uptodate(bh)); | |
1580 | if (state != GFS2_BLKST_UNLINKED && rbm->bi->bi_clone) | |
1581 | buffer = rbm->bi->bi_clone + rbm->bi->bi_offset; | |
1582 | find_next: | |
1583 | initial_offset = rbm->offset; | |
1584 | offset = gfs2_bitfit(buffer, rbm->bi->bi_len, rbm->offset, state); | |
1585 | if (offset == BFITNOENT) | |
1586 | goto bitmap_full; | |
1587 | rbm->offset = offset; | |
1588 | if (ip == NULL) | |
1589 | return 0; | |
1590 | ||
1591 | initial_bi = rbm->bi; | |
1592 | ret = gfs2_reservation_check_and_update(rbm, ip); | |
1593 | if (ret == 0) | |
1594 | return 0; | |
1595 | if (ret > 0) { | |
1596 | n += (rbm->bi - initial_bi); | |
1597 | goto find_next; | |
1598 | } | |
1599 | return ret; | |
1600 | ||
1601 | bitmap_full: /* Mark bitmap as full and fall through */ | |
1602 | if ((state == GFS2_BLKST_FREE) && initial_offset == 0) | |
1603 | set_bit(GBF_FULL, &rbm->bi->bi_flags); | |
1604 | ||
1605 | next_bitmap: /* Find next bitmap in the rgrp */ | |
1606 | rbm->offset = 0; | |
1607 | index = rbm->bi - rbm->rgd->rd_bits; | |
1608 | index++; | |
1609 | if (index == rbm->rgd->rd_length) | |
1610 | index = 0; | |
1611 | rbm->bi = &rbm->rgd->rd_bits[index]; | |
1612 | if ((index == 0) && nowrap) | |
1613 | break; | |
1614 | n++; | |
1615 | if (n >= iters) | |
1616 | break; | |
1617 | } | |
1618 | ||
1619 | return -ENOSPC; | |
1620 | } | |
1621 | ||
c8cdf479 SW |
1622 | /** |
1623 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1624 | * @rgd: The rgrp | |
886b1416 BP |
1625 | * @last_unlinked: block address of the last dinode we unlinked |
1626 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1627 | * |
1a0eae88 BP |
1628 | * Returns: 0 if no error |
1629 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1630 | */ |
1631 | ||
044b9414 | 1632 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1633 | { |
5b924ae2 | 1634 | u64 block; |
5f3eae75 | 1635 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1636 | struct gfs2_glock *gl; |
1637 | struct gfs2_inode *ip; | |
1638 | int error; | |
1639 | int found = 0; | |
5b924ae2 | 1640 | struct gfs2_rbm rbm = { .rgd = rgd, .bi = rgd->rd_bits, .offset = 0 }; |
c8cdf479 | 1641 | |
5b924ae2 | 1642 | while (1) { |
5f3eae75 | 1643 | down_write(&sdp->sd_log_flush_lock); |
5b924ae2 | 1644 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, true); |
5f3eae75 | 1645 | up_write(&sdp->sd_log_flush_lock); |
5b924ae2 SW |
1646 | if (error == -ENOSPC) |
1647 | break; | |
1648 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1649 | break; |
b3e47ca0 | 1650 | |
5b924ae2 SW |
1651 | block = gfs2_rbm_to_block(&rbm); |
1652 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1653 | break; | |
1654 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1655 | continue; |
5b924ae2 | 1656 | if (block == skip) |
1e19a195 | 1657 | continue; |
5b924ae2 | 1658 | *last_unlinked = block; |
044b9414 | 1659 | |
5b924ae2 | 1660 | error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl); |
044b9414 SW |
1661 | if (error) |
1662 | continue; | |
1663 | ||
1664 | /* If the inode is already in cache, we can ignore it here | |
1665 | * because the existing inode disposal code will deal with | |
1666 | * it when all refs have gone away. Accessing gl_object like | |
1667 | * this is not safe in general. Here it is ok because we do | |
1668 | * not dereference the pointer, and we only need an approx | |
1669 | * answer to whether it is NULL or not. | |
1670 | */ | |
1671 | ip = gl->gl_object; | |
1672 | ||
1673 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1674 | gfs2_glock_put(gl); | |
1675 | else | |
1676 | found++; | |
1677 | ||
1678 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1679 | if (found > NR_CPUS) |
044b9414 | 1680 | return; |
c8cdf479 SW |
1681 | } |
1682 | ||
1683 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1684 | return; |
c8cdf479 SW |
1685 | } |
1686 | ||
b3b94faa | 1687 | /** |
666d1d8a | 1688 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1689 | * @ip: the inode to reserve space for |
666d1d8a | 1690 | * @requested: the number of blocks to be reserved |
b3b94faa DT |
1691 | * |
1692 | * Returns: errno | |
1693 | */ | |
1694 | ||
666d1d8a | 1695 | int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested) |
b3b94faa | 1696 | { |
feaa7bba | 1697 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 1698 | struct gfs2_rgrpd *begin = NULL; |
564e12b1 | 1699 | struct gfs2_blkreserv *rs = ip->i_res; |
666d1d8a BP |
1700 | int error = 0, rg_locked, flags = LM_FLAG_TRY; |
1701 | u64 last_unlinked = NO_BLOCK; | |
7c9ca621 | 1702 | int loops = 0; |
b3b94faa | 1703 | |
90306c41 BM |
1704 | if (sdp->sd_args.ar_rgrplvb) |
1705 | flags |= GL_SKIP; | |
666d1d8a BP |
1706 | if (gfs2_assert_warn(sdp, requested)) { |
1707 | error = -EINVAL; | |
1708 | goto out; | |
1709 | } | |
8e2e0047 | 1710 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1711 | begin = rs->rs_rbm.rgd; |
8e2e0047 BP |
1712 | flags = 0; /* Yoda: Do or do not. There is no try */ |
1713 | } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) { | |
4a993fb1 | 1714 | rs->rs_rbm.rgd = begin = ip->i_rgd; |
8e2e0047 | 1715 | } else { |
4a993fb1 | 1716 | rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 1717 | } |
4a993fb1 | 1718 | if (rs->rs_rbm.rgd == NULL) |
7c9ca621 BP |
1719 | return -EBADSLT; |
1720 | ||
1721 | while (loops < 3) { | |
292c8c14 AD |
1722 | rg_locked = 0; |
1723 | ||
4a993fb1 | 1724 | if (gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) { |
292c8c14 AD |
1725 | rg_locked = 1; |
1726 | error = 0; | |
8e2e0047 | 1727 | } else if (!loops && !gfs2_rs_active(rs) && |
4a993fb1 | 1728 | rs->rs_rbm.rgd->rd_rs_cnt > RGRP_RSRV_MAX_CONTENDERS) { |
8e2e0047 BP |
1729 | /* If the rgrp already is maxed out for contenders, |
1730 | we can eliminate it as a "first pass" without even | |
1731 | requesting the rgrp glock. */ | |
1732 | error = GLR_TRYFAILED; | |
292c8c14 | 1733 | } else { |
4a993fb1 | 1734 | error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl, |
8e2e0047 BP |
1735 | LM_ST_EXCLUSIVE, flags, |
1736 | &rs->rs_rgd_gh); | |
90306c41 | 1737 | if (!error && sdp->sd_args.ar_rgrplvb) { |
4a993fb1 | 1738 | error = update_rgrp_lvb(rs->rs_rbm.rgd); |
90306c41 BM |
1739 | if (error) { |
1740 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
1741 | return error; | |
1742 | } | |
1743 | } | |
292c8c14 | 1744 | } |
b3b94faa DT |
1745 | switch (error) { |
1746 | case 0: | |
8e2e0047 | 1747 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1748 | if (unclaimed_blocks(rs->rs_rbm.rgd) + |
71f890f7 | 1749 | rs->rs_free >= requested) { |
4a993fb1 | 1750 | ip->i_rgd = rs->rs_rbm.rgd; |
8e2e0047 BP |
1751 | return 0; |
1752 | } | |
1753 | /* We have a multi-block reservation, but the | |
1754 | rgrp doesn't have enough free blocks to | |
1755 | satisfy the request. Free the reservation | |
1756 | and look for a suitable rgrp. */ | |
4a993fb1 | 1757 | gfs2_rs_deltree(ip, rs); |
8e2e0047 | 1758 | } |
4a993fb1 | 1759 | if (try_rgrp_fit(rs->rs_rbm.rgd, ip, requested)) { |
90306c41 | 1760 | if (sdp->sd_args.ar_rgrplvb) |
4a993fb1 SW |
1761 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); |
1762 | ip->i_rgd = rs->rs_rbm.rgd; | |
7c9ca621 | 1763 | return 0; |
54335b1f | 1764 | } |
4a993fb1 | 1765 | if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK) { |
90306c41 | 1766 | if (sdp->sd_args.ar_rgrplvb) |
4a993fb1 SW |
1767 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); |
1768 | try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked, | |
666d1d8a | 1769 | ip->i_no_addr); |
90306c41 | 1770 | } |
292c8c14 | 1771 | if (!rg_locked) |
564e12b1 | 1772 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
9cabcdbd | 1773 | /* fall through */ |
b3b94faa | 1774 | case GLR_TRYFAILED: |
4a993fb1 SW |
1775 | rs->rs_rbm.rgd = gfs2_rgrpd_get_next(rs->rs_rbm.rgd); |
1776 | rs->rs_rbm.rgd = rs->rs_rbm.rgd ? : begin; /* if NULL, wrap */ | |
1777 | if (rs->rs_rbm.rgd != begin) /* If we didn't wrap */ | |
666d1d8a BP |
1778 | break; |
1779 | ||
1780 | flags &= ~LM_FLAG_TRY; | |
1781 | loops++; | |
1782 | /* Check that fs hasn't grown if writing to rindex */ | |
1783 | if (ip == GFS2_I(sdp->sd_rindex) && | |
1784 | !sdp->sd_rindex_uptodate) { | |
1785 | error = gfs2_ri_update(ip); | |
1786 | if (error) | |
1787 | goto out; | |
1788 | } else if (loops == 2) | |
1789 | /* Flushing the log may release space */ | |
1790 | gfs2_log_flush(sdp, NULL); | |
b3b94faa | 1791 | break; |
b3b94faa | 1792 | default: |
666d1d8a | 1793 | goto out; |
b3b94faa | 1794 | } |
b3b94faa | 1795 | } |
666d1d8a | 1796 | error = -ENOSPC; |
b3b94faa | 1797 | |
564e12b1 | 1798 | out: |
9ae32429 | 1799 | return error; |
b3b94faa DT |
1800 | } |
1801 | ||
1802 | /** | |
1803 | * gfs2_inplace_release - release an inplace reservation | |
1804 | * @ip: the inode the reservation was taken out on | |
1805 | * | |
1806 | * Release a reservation made by gfs2_inplace_reserve(). | |
1807 | */ | |
1808 | ||
1809 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1810 | { | |
564e12b1 | 1811 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 1812 | |
564e12b1 BP |
1813 | if (rs->rs_rgd_gh.gh_gl) |
1814 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
b3b94faa DT |
1815 | } |
1816 | ||
1817 | /** | |
1818 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1819 | * @rgd: the resource group holding the block | |
1820 | * @block: the block number | |
1821 | * | |
1822 | * Returns: The block type (GFS2_BLKST_*) | |
1823 | */ | |
1824 | ||
acf7e244 | 1825 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 1826 | { |
3983903a SW |
1827 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1828 | int ret; | |
b3b94faa | 1829 | |
3983903a SW |
1830 | ret = gfs2_rbm_from_block(&rbm, block); |
1831 | WARN_ON_ONCE(ret != 0); | |
b3b94faa | 1832 | |
3983903a SW |
1833 | return gfs2_testbit(rgd, rbm.bi->bi_bh->b_data + rbm.bi->bi_offset, |
1834 | rbm.bi->bi_len, rbm.offset); | |
b3b94faa DT |
1835 | } |
1836 | ||
60a0b8f9 | 1837 | |
b3e47ca0 BP |
1838 | /** |
1839 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 1840 | * @rbm: the resource group information |
b3e47ca0 BP |
1841 | * @dinode: TRUE if the first block we allocate is for a dinode |
1842 | * @n: The extent length | |
1843 | * | |
1844 | * Add the found bitmap buffer to the transaction. | |
1845 | * Set the found bits to @new_state to change block's allocation state. | |
1846 | * Returns: starting block number of the extent (fs scope) | |
1847 | */ | |
4a993fb1 SW |
1848 | static u64 gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
1849 | unsigned int *n) | |
b3e47ca0 | 1850 | { |
4a993fb1 SW |
1851 | struct gfs2_rgrpd *rgd = rbm->rgd; |
1852 | struct gfs2_bitmap *bi = rbm->bi; | |
1853 | u32 blk = rbm->offset; | |
b3e47ca0 | 1854 | const unsigned int elen = *n; |
5b924ae2 | 1855 | u32 goal; |
b3e47ca0 BP |
1856 | const u8 *buffer = NULL; |
1857 | ||
6a8099ed | 1858 | *n = 0; |
b3e47ca0 | 1859 | buffer = bi->bi_bh->b_data + bi->bi_offset; |
60a0b8f9 | 1860 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
06344b91 BP |
1861 | gfs2_setbit(rgd, bi->bi_clone, bi, blk, |
1862 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); | |
6a8099ed | 1863 | (*n)++; |
60a0b8f9 SW |
1864 | goal = blk; |
1865 | while (*n < elen) { | |
1866 | goal++; | |
1867 | if (goal >= (bi->bi_len * GFS2_NBBY)) | |
1868 | break; | |
1869 | if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) != | |
1870 | GFS2_BLKST_FREE) | |
1871 | break; | |
06344b91 | 1872 | gfs2_setbit(rgd, bi->bi_clone, bi, goal, GFS2_BLKST_USED); |
60a0b8f9 | 1873 | (*n)++; |
c8cdf479 | 1874 | } |
b3e47ca0 | 1875 | blk = gfs2_bi2rgd_blk(bi, blk); |
6a8099ed | 1876 | rgd->rd_last_alloc = blk + *n - 1; |
b3e47ca0 | 1877 | return rgd->rd_data0 + blk; |
b3b94faa DT |
1878 | } |
1879 | ||
1880 | /** | |
1881 | * rgblk_free - Change alloc state of given block(s) | |
1882 | * @sdp: the filesystem | |
1883 | * @bstart: the start of a run of blocks to free | |
1884 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1885 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1886 | * | |
1887 | * Returns: Resource group containing the block(s) | |
1888 | */ | |
1889 | ||
cd915493 SW |
1890 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
1891 | u32 blen, unsigned char new_state) | |
b3b94faa | 1892 | { |
3b1d0b9d | 1893 | struct gfs2_rbm rbm; |
b3b94faa | 1894 | |
3b1d0b9d SW |
1895 | rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
1896 | if (!rbm.rgd) { | |
b3b94faa | 1897 | if (gfs2_consist(sdp)) |
382066da | 1898 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
1899 | return NULL; |
1900 | } | |
1901 | ||
b3b94faa | 1902 | while (blen--) { |
3b1d0b9d SW |
1903 | gfs2_rbm_from_block(&rbm, bstart); |
1904 | bstart++; | |
1905 | if (!rbm.bi->bi_clone) { | |
1906 | rbm.bi->bi_clone = kmalloc(rbm.bi->bi_bh->b_size, | |
1907 | GFP_NOFS | __GFP_NOFAIL); | |
1908 | memcpy(rbm.bi->bi_clone + rbm.bi->bi_offset, | |
1909 | rbm.bi->bi_bh->b_data + rbm.bi->bi_offset, | |
1910 | rbm.bi->bi_len); | |
b3b94faa | 1911 | } |
3b1d0b9d SW |
1912 | gfs2_trans_add_bh(rbm.rgd->rd_gl, rbm.bi->bi_bh, 1); |
1913 | gfs2_setbit(rbm.rgd, NULL, rbm.bi, rbm.offset, new_state); | |
b3b94faa DT |
1914 | } |
1915 | ||
3b1d0b9d | 1916 | return rbm.rgd; |
b3b94faa DT |
1917 | } |
1918 | ||
1919 | /** | |
09010978 SW |
1920 | * gfs2_rgrp_dump - print out an rgrp |
1921 | * @seq: The iterator | |
1922 | * @gl: The glock in question | |
1923 | * | |
1924 | */ | |
1925 | ||
1926 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
1927 | { | |
8e2e0047 BP |
1928 | struct gfs2_rgrpd *rgd = gl->gl_object; |
1929 | struct gfs2_blkreserv *trs; | |
1930 | const struct rb_node *n; | |
1931 | ||
09010978 SW |
1932 | if (rgd == NULL) |
1933 | return 0; | |
8e2e0047 | 1934 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u\n", |
09010978 | 1935 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 BP |
1936 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
1937 | rgd->rd_reserved); | |
1938 | spin_lock(&rgd->rd_rsspin); | |
1939 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { | |
1940 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1941 | dump_rs(seq, trs); | |
1942 | } | |
1943 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
1944 | return 0; |
1945 | } | |
1946 | ||
6050b9c7 SW |
1947 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
1948 | { | |
1949 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1950 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 1951 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
1952 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
1953 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
1954 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
1955 | } | |
1956 | ||
8e2e0047 | 1957 | /** |
5b924ae2 SW |
1958 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
1959 | * @ip: The inode we have just allocated blocks for | |
1960 | * @rbm: The start of the allocated blocks | |
1961 | * @len: The extent length | |
8e2e0047 | 1962 | * |
5b924ae2 SW |
1963 | * Adjusts a reservation after an allocation has taken place. If the |
1964 | * reservation does not match the allocation, or if it is now empty | |
1965 | * then it is removed. | |
8e2e0047 | 1966 | */ |
5b924ae2 SW |
1967 | |
1968 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
1969 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 BP |
1970 | { |
1971 | struct gfs2_blkreserv *rs = ip->i_res; | |
5b924ae2 SW |
1972 | struct gfs2_rgrpd *rgd = rbm->rgd; |
1973 | unsigned rlen; | |
1974 | u64 block; | |
1975 | int ret; | |
8e2e0047 | 1976 | |
5b924ae2 SW |
1977 | spin_lock(&rgd->rd_rsspin); |
1978 | if (gfs2_rs_active(rs)) { | |
1979 | if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) { | |
1980 | block = gfs2_rbm_to_block(rbm); | |
1981 | ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len); | |
1982 | rlen = min(rs->rs_free, len); | |
1983 | rs->rs_free -= rlen; | |
1984 | rgd->rd_reserved -= rlen; | |
1985 | trace_gfs2_rs(ip, rs, TRACE_RS_CLAIM); | |
1986 | if (rs->rs_free && !ret) | |
1987 | goto out; | |
1988 | } | |
1989 | __rs_deltree(ip, rs); | |
8e2e0047 | 1990 | } |
5b924ae2 SW |
1991 | out: |
1992 | spin_unlock(&rgd->rd_rsspin); | |
8e2e0047 BP |
1993 | } |
1994 | ||
09010978 | 1995 | /** |
6e87ed0f | 1996 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 1997 | * @ip: the inode to allocate the block for |
09010978 | 1998 | * @bn: Used to return the starting block number |
8e2e0047 | 1999 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 2000 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 2001 | * @generation: the generation number of the inode |
b3b94faa | 2002 | * |
09010978 | 2003 | * Returns: 0 or error |
b3b94faa DT |
2004 | */ |
2005 | ||
6a8099ed | 2006 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 2007 | bool dinode, u64 *generation) |
b3b94faa | 2008 | { |
feaa7bba | 2009 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 2010 | struct buffer_head *dibh; |
4a993fb1 | 2011 | struct gfs2_rbm rbm = { .rgd = ip->i_rgd, }; |
6a8099ed | 2012 | unsigned int ndata; |
5b924ae2 | 2013 | u64 goal; |
3c5d785a | 2014 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 2015 | int error; |
b3b94faa | 2016 | |
5b924ae2 SW |
2017 | if (gfs2_rs_active(ip->i_res)) |
2018 | goal = gfs2_rbm_to_block(&ip->i_res->rs_rbm); | |
2019 | else if (!dinode && rgrp_contains_block(rbm.rgd, ip->i_goal)) | |
2020 | goal = ip->i_goal; | |
62e252ee | 2021 | else |
5b924ae2 | 2022 | goal = rbm.rgd->rd_last_alloc + rbm.rgd->rd_data0; |
62e252ee | 2023 | |
5b924ae2 SW |
2024 | gfs2_rbm_from_block(&rbm, goal); |
2025 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, ip, false); | |
62e252ee SW |
2026 | |
2027 | /* Since all blocks are reserved in advance, this shouldn't happen */ | |
5b924ae2 SW |
2028 | if (error) { |
2029 | fs_warn(sdp, "error=%d, nblocks=%u, full=%d\n", error, *nblocks, | |
2030 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags)); | |
62e252ee | 2031 | goto rgrp_error; |
8e2e0047 | 2032 | } |
62e252ee | 2033 | |
4a993fb1 | 2034 | block = gfs2_alloc_extent(&rbm, dinode, nblocks); |
5b924ae2 SW |
2035 | if (gfs2_rs_active(ip->i_res)) |
2036 | gfs2_adjust_reservation(ip, &rbm, *nblocks); | |
6a8099ed SW |
2037 | ndata = *nblocks; |
2038 | if (dinode) | |
2039 | ndata--; | |
b3e47ca0 | 2040 | |
3c5d785a | 2041 | if (!dinode) { |
6a8099ed | 2042 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
2043 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2044 | if (error == 0) { | |
2045 | struct gfs2_dinode *di = | |
2046 | (struct gfs2_dinode *)dibh->b_data; | |
2047 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
2048 | di->di_goal_meta = di->di_goal_data = | |
2049 | cpu_to_be64(ip->i_goal); | |
2050 | brelse(dibh); | |
2051 | } | |
d9ba7615 | 2052 | } |
4a993fb1 | 2053 | if (rbm.rgd->rd_free < *nblocks) { |
8e2e0047 | 2054 | printk(KERN_WARNING "nblocks=%u\n", *nblocks); |
09010978 | 2055 | goto rgrp_error; |
8e2e0047 | 2056 | } |
09010978 | 2057 | |
4a993fb1 | 2058 | rbm.rgd->rd_free -= *nblocks; |
3c5d785a | 2059 | if (dinode) { |
4a993fb1 SW |
2060 | rbm.rgd->rd_dinodes++; |
2061 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2062 | if (*generation == 0) |
4a993fb1 | 2063 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2064 | } |
b3b94faa | 2065 | |
4a993fb1 SW |
2066 | gfs2_trans_add_bh(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh, 1); |
2067 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
2068 | gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
b3b94faa | 2069 | |
6a8099ed | 2070 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a BP |
2071 | if (dinode) |
2072 | gfs2_trans_add_unrevoke(sdp, block, 1); | |
6a8099ed SW |
2073 | |
2074 | /* | |
2075 | * This needs reviewing to see why we cannot do the quota change | |
2076 | * at this point in the dinode case. | |
2077 | */ | |
2078 | if (ndata) | |
2079 | gfs2_quota_change(ip, ndata, ip->i_inode.i_uid, | |
3c5d785a | 2080 | ip->i_inode.i_gid); |
b3b94faa | 2081 | |
4a993fb1 SW |
2082 | rbm.rgd->rd_free_clone -= *nblocks; |
2083 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, | |
6e87ed0f | 2084 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2085 | *bn = block; |
2086 | return 0; | |
2087 | ||
2088 | rgrp_error: | |
4a993fb1 | 2089 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2090 | return -EIO; |
b3b94faa DT |
2091 | } |
2092 | ||
2093 | /** | |
46fcb2ed | 2094 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
2095 | * @ip: the inode these blocks are being freed from |
2096 | * @bstart: first block of a run of contiguous blocks | |
2097 | * @blen: the length of the block run | |
46fcb2ed | 2098 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2099 | * |
2100 | */ | |
2101 | ||
46fcb2ed | 2102 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 2103 | { |
feaa7bba | 2104 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2105 | struct gfs2_rgrpd *rgd; |
2106 | ||
2107 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
2108 | if (!rgd) | |
2109 | return; | |
41db1ab9 | 2110 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2111 | rgd->rd_free += blen; |
66fc061b | 2112 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
d4e9c4c3 | 2113 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 2114 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 | 2115 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 2116 | |
6d3117b4 | 2117 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 2118 | if (meta || ip->i_depth) |
6d3117b4 | 2119 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 2120 | } |
b3b94faa | 2121 | |
4c16c36a BP |
2122 | /** |
2123 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2124 | * @ip: the inode these blocks are being freed from | |
2125 | * @bstart: first block of a run of contiguous blocks | |
2126 | * @blen: the length of the block run | |
2127 | * | |
2128 | */ | |
2129 | ||
2130 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
2131 | { | |
2132 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2133 | ||
46fcb2ed | 2134 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 2135 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2136 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2137 | } |
2138 | ||
feaa7bba SW |
2139 | void gfs2_unlink_di(struct inode *inode) |
2140 | { | |
2141 | struct gfs2_inode *ip = GFS2_I(inode); | |
2142 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2143 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2144 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
2145 | |
2146 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
2147 | if (!rgd) | |
2148 | return; | |
41db1ab9 | 2149 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
feaa7bba | 2150 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 2151 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2152 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2153 | update_rgrp_lvb_unlinked(rgd, 1); | |
feaa7bba SW |
2154 | } |
2155 | ||
cd915493 | 2156 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
2157 | { |
2158 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2159 | struct gfs2_rgrpd *tmp_rgd; | |
2160 | ||
2161 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
2162 | if (!tmp_rgd) | |
2163 | return; | |
2164 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
2165 | ||
73f74948 | 2166 | if (!rgd->rd_dinodes) |
b3b94faa | 2167 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2168 | rgd->rd_dinodes--; |
cfc8b549 | 2169 | rgd->rd_free++; |
b3b94faa | 2170 | |
d4e9c4c3 | 2171 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 2172 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2173 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2174 | update_rgrp_lvb_unlinked(rgd, -1); | |
b3b94faa DT |
2175 | |
2176 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
2177 | } |
2178 | ||
b3b94faa DT |
2179 | |
2180 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
2181 | { | |
dbb7cae2 | 2182 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
41db1ab9 | 2183 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2184 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 2185 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2186 | } |
2187 | ||
acf7e244 SW |
2188 | /** |
2189 | * gfs2_check_blk_type - Check the type of a block | |
2190 | * @sdp: The superblock | |
2191 | * @no_addr: The block number to check | |
2192 | * @type: The block type we are looking for | |
2193 | * | |
2194 | * Returns: 0 if the block type matches the expected type | |
2195 | * -ESTALE if it doesn't match | |
2196 | * or -ve errno if something went wrong while checking | |
2197 | */ | |
2198 | ||
2199 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2200 | { | |
2201 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2202 | struct gfs2_holder rgd_gh; |
58884c4d | 2203 | int error = -EINVAL; |
acf7e244 | 2204 | |
66fc061b | 2205 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2206 | if (!rgd) |
8339ee54 | 2207 | goto fail; |
acf7e244 SW |
2208 | |
2209 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2210 | if (error) | |
8339ee54 | 2211 | goto fail; |
acf7e244 SW |
2212 | |
2213 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
2214 | error = -ESTALE; | |
2215 | ||
2216 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
2217 | fail: |
2218 | return error; | |
2219 | } | |
2220 | ||
b3b94faa DT |
2221 | /** |
2222 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2223 | * @ip: the inode |
b3b94faa DT |
2224 | * @rlist: the list of resource groups |
2225 | * @block: the block | |
2226 | * | |
2227 | * Figure out what RG a block belongs to and add that RG to the list | |
2228 | * | |
2229 | * FIXME: Don't use NOFAIL | |
2230 | * | |
2231 | */ | |
2232 | ||
70b0c365 | 2233 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2234 | u64 block) |
b3b94faa | 2235 | { |
70b0c365 | 2236 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2237 | struct gfs2_rgrpd *rgd; |
2238 | struct gfs2_rgrpd **tmp; | |
2239 | unsigned int new_space; | |
2240 | unsigned int x; | |
2241 | ||
2242 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2243 | return; | |
2244 | ||
70b0c365 SW |
2245 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
2246 | rgd = ip->i_rgd; | |
2247 | else | |
66fc061b | 2248 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 2249 | if (!rgd) { |
70b0c365 | 2250 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
2251 | return; |
2252 | } | |
70b0c365 | 2253 | ip->i_rgd = rgd; |
b3b94faa DT |
2254 | |
2255 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2256 | if (rlist->rl_rgd[x] == rgd) | |
2257 | return; | |
2258 | ||
2259 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2260 | new_space = rlist->rl_space + 10; | |
2261 | ||
2262 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2263 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2264 | |
2265 | if (rlist->rl_rgd) { | |
2266 | memcpy(tmp, rlist->rl_rgd, | |
2267 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2268 | kfree(rlist->rl_rgd); | |
2269 | } | |
2270 | ||
2271 | rlist->rl_space = new_space; | |
2272 | rlist->rl_rgd = tmp; | |
2273 | } | |
2274 | ||
2275 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2276 | } | |
2277 | ||
2278 | /** | |
2279 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2280 | * and initialize an array of glock holders for them | |
2281 | * @rlist: the list of resource groups | |
2282 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
2283 | * |
2284 | * FIXME: Don't use NOFAIL | |
2285 | * | |
2286 | */ | |
2287 | ||
fe6c991c | 2288 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
2289 | { |
2290 | unsigned int x; | |
2291 | ||
2292 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 2293 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2294 | for (x = 0; x < rlist->rl_rgrps; x++) |
2295 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 2296 | state, 0, |
b3b94faa DT |
2297 | &rlist->rl_ghs[x]); |
2298 | } | |
2299 | ||
2300 | /** | |
2301 | * gfs2_rlist_free - free a resource group list | |
2302 | * @list: the list of resource groups | |
2303 | * | |
2304 | */ | |
2305 | ||
2306 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2307 | { | |
2308 | unsigned int x; | |
2309 | ||
2310 | kfree(rlist->rl_rgd); | |
2311 | ||
2312 | if (rlist->rl_ghs) { | |
2313 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2314 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2315 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2316 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2317 | } |
2318 | } | |
2319 |