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 | |
3e6339dd SW |
72 | * @rbm: The position of the bit to set |
73 | * @do_clone: Also set the clone bitmap, if it exists | |
88c8ab1f SW |
74 | * @new_state: the new state of the block |
75 | * | |
76 | */ | |
77 | ||
3e6339dd | 78 | static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone, |
06344b91 | 79 | unsigned char new_state) |
88c8ab1f | 80 | { |
b45e41d7 | 81 | unsigned char *byte1, *byte2, *end, cur_state; |
3e6339dd SW |
82 | unsigned int buflen = rbm->bi->bi_len; |
83 | const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 84 | |
3e6339dd SW |
85 | byte1 = rbm->bi->bi_bh->b_data + rbm->bi->bi_offset + (rbm->offset / GFS2_NBBY); |
86 | end = rbm->bi->bi_bh->b_data + rbm->bi->bi_offset + buflen; | |
88c8ab1f | 87 | |
b45e41d7 | 88 | BUG_ON(byte1 >= end); |
88c8ab1f | 89 | |
b45e41d7 | 90 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 91 | |
b45e41d7 | 92 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
3e6339dd SW |
93 | printk(KERN_WARNING "GFS2: buf_blk = 0x%x old_state=%d, " |
94 | "new_state=%d\n", rbm->offset, cur_state, new_state); | |
95 | printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%x\n", | |
96 | (unsigned long long)rbm->rgd->rd_addr, | |
97 | rbm->bi->bi_start); | |
98 | printk(KERN_WARNING "GFS2: bi_offset=0x%x bi_len=0x%x\n", | |
99 | rbm->bi->bi_offset, rbm->bi->bi_len); | |
95c8e17f | 100 | dump_stack(); |
3e6339dd | 101 | gfs2_consist_rgrpd(rbm->rgd); |
b45e41d7 SW |
102 | return; |
103 | } | |
104 | *byte1 ^= (cur_state ^ new_state) << bit; | |
105 | ||
3e6339dd SW |
106 | if (do_clone && rbm->bi->bi_clone) { |
107 | byte2 = rbm->bi->bi_clone + rbm->bi->bi_offset + (rbm->offset / GFS2_NBBY); | |
b45e41d7 SW |
108 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
109 | *byte2 ^= (cur_state ^ new_state) << bit; | |
110 | } | |
88c8ab1f SW |
111 | } |
112 | ||
113 | /** | |
114 | * gfs2_testbit - test a bit in the bitmaps | |
c04a2ef3 | 115 | * @rbm: The bit to test |
88c8ab1f | 116 | * |
c04a2ef3 | 117 | * Returns: The two bit block state of the requested bit |
88c8ab1f SW |
118 | */ |
119 | ||
c04a2ef3 | 120 | static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm) |
88c8ab1f | 121 | { |
c04a2ef3 SW |
122 | const u8 *buffer = rbm->bi->bi_bh->b_data + rbm->bi->bi_offset; |
123 | const u8 *byte; | |
88c8ab1f SW |
124 | unsigned int bit; |
125 | ||
c04a2ef3 SW |
126 | byte = buffer + (rbm->offset / GFS2_NBBY); |
127 | bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 128 | |
c04a2ef3 | 129 | return (*byte >> bit) & GFS2_BIT_MASK; |
88c8ab1f SW |
130 | } |
131 | ||
223b2b88 SW |
132 | /** |
133 | * gfs2_bit_search | |
134 | * @ptr: Pointer to bitmap data | |
135 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
136 | * @state: The state we are searching for | |
137 | * | |
138 | * We xor the bitmap data with a patter which is the bitwise opposite | |
139 | * of what we are looking for, this gives rise to a pattern of ones | |
140 | * wherever there is a match. Since we have two bits per entry, we | |
141 | * take this pattern, shift it down by one place and then and it with | |
142 | * the original. All the even bit positions (0,2,4, etc) then represent | |
143 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
144 | * odd bit positions. | |
145 | * | |
146 | * This allows searching of a whole u64 at once (32 blocks) with a | |
147 | * single test (on 64 bit arches). | |
148 | */ | |
149 | ||
150 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
151 | { | |
152 | u64 tmp; | |
153 | static const u64 search[] = { | |
075ac448 HE |
154 | [0] = 0xffffffffffffffffULL, |
155 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
156 | [2] = 0x5555555555555555ULL, | |
157 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
158 | }; |
159 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
160 | tmp &= (tmp >> 1); | |
161 | tmp &= mask; | |
162 | return tmp; | |
163 | } | |
164 | ||
8e2e0047 BP |
165 | /** |
166 | * rs_cmp - multi-block reservation range compare | |
167 | * @blk: absolute file system block number of the new reservation | |
168 | * @len: number of blocks in the new reservation | |
169 | * @rs: existing reservation to compare against | |
170 | * | |
171 | * returns: 1 if the block range is beyond the reach of the reservation | |
172 | * -1 if the block range is before the start of the reservation | |
173 | * 0 if the block range overlaps with the reservation | |
174 | */ | |
175 | static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs) | |
176 | { | |
4a993fb1 | 177 | u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 BP |
178 | |
179 | if (blk >= startblk + rs->rs_free) | |
180 | return 1; | |
181 | if (blk + len - 1 < startblk) | |
182 | return -1; | |
183 | return 0; | |
184 | } | |
185 | ||
88c8ab1f SW |
186 | /** |
187 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
188 | * a block in a given allocation state. | |
886b1416 | 189 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 190 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 191 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 192 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
193 | * |
194 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
195 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
196 | * beginning of a bitmap block buffer, skipping any header structures, but |
197 | * headers are always a multiple of 64 bits long so that the buffer is | |
198 | * always aligned to a 64 bit boundary. | |
199 | * | |
200 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 201 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 202 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
203 | * |
204 | * Return: the block number (bitmap buffer scope) that was found | |
205 | */ | |
206 | ||
02ab1721 HE |
207 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
208 | u32 goal, u8 state) | |
88c8ab1f | 209 | { |
223b2b88 SW |
210 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
211 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
212 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
213 | u64 tmp; | |
075ac448 | 214 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
215 | u32 bit; |
216 | ||
223b2b88 SW |
217 | /* Mask off bits we don't care about at the start of the search */ |
218 | mask <<= spoint; | |
219 | tmp = gfs2_bit_search(ptr, mask, state); | |
220 | ptr++; | |
221 | while(tmp == 0 && ptr < end) { | |
075ac448 | 222 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 223 | ptr++; |
1f466a47 | 224 | } |
223b2b88 SW |
225 | /* Mask off any bits which are more than len bytes from the start */ |
226 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
227 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
228 | /* Didn't find anything, so return */ | |
229 | if (tmp == 0) | |
230 | return BFITNOENT; | |
231 | ptr--; | |
d8bd504a | 232 | bit = __ffs64(tmp); |
223b2b88 SW |
233 | bit /= 2; /* two bits per entry in the bitmap */ |
234 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
235 | } |
236 | ||
237 | /** | |
238 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 239 | * @rgd: the resource group descriptor |
88c8ab1f SW |
240 | * @buffer: the buffer that holds the bitmaps |
241 | * @buflen: the length (in bytes) of the buffer | |
242 | * @state: the state of the block we're looking for | |
243 | * | |
244 | * Returns: The number of bits | |
245 | */ | |
246 | ||
110acf38 SW |
247 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
248 | unsigned int buflen, u8 state) | |
88c8ab1f | 249 | { |
110acf38 SW |
250 | const u8 *byte = buffer; |
251 | const u8 *end = buffer + buflen; | |
252 | const u8 state1 = state << 2; | |
253 | const u8 state2 = state << 4; | |
254 | const u8 state3 = state << 6; | |
cd915493 | 255 | u32 count = 0; |
88c8ab1f SW |
256 | |
257 | for (; byte < end; byte++) { | |
258 | if (((*byte) & 0x03) == state) | |
259 | count++; | |
260 | if (((*byte) & 0x0C) == state1) | |
261 | count++; | |
262 | if (((*byte) & 0x30) == state2) | |
263 | count++; | |
264 | if (((*byte) & 0xC0) == state3) | |
265 | count++; | |
266 | } | |
267 | ||
268 | return count; | |
269 | } | |
270 | ||
b3b94faa DT |
271 | /** |
272 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
273 | * @rgd: the rgrp |
274 | * | |
275 | */ | |
276 | ||
277 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
278 | { | |
279 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
280 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 281 | u32 length = rgd->rd_length; |
cd915493 | 282 | u32 count[4], tmp; |
b3b94faa DT |
283 | int buf, x; |
284 | ||
cd915493 | 285 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
286 | |
287 | /* Count # blocks in each of 4 possible allocation states */ | |
288 | for (buf = 0; buf < length; buf++) { | |
289 | bi = rgd->rd_bits + buf; | |
290 | for (x = 0; x < 4; x++) | |
291 | count[x] += gfs2_bitcount(rgd, | |
292 | bi->bi_bh->b_data + | |
293 | bi->bi_offset, | |
294 | bi->bi_len, x); | |
295 | } | |
296 | ||
cfc8b549 | 297 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
298 | if (gfs2_consist_rgrpd(rgd)) |
299 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 300 | count[0], rgd->rd_free); |
b3b94faa DT |
301 | return; |
302 | } | |
303 | ||
73f74948 | 304 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 305 | if (count[1] != tmp) { |
b3b94faa DT |
306 | if (gfs2_consist_rgrpd(rgd)) |
307 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
308 | count[1], tmp); | |
309 | return; | |
310 | } | |
311 | ||
6b946170 | 312 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 313 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 314 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 315 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
316 | return; |
317 | } | |
b3b94faa DT |
318 | } |
319 | ||
bb8d8a6f | 320 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 321 | { |
bb8d8a6f SW |
322 | u64 first = rgd->rd_data0; |
323 | u64 last = first + rgd->rd_data; | |
16910427 | 324 | return first <= block && block < last; |
b3b94faa DT |
325 | } |
326 | ||
327 | /** | |
328 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
329 | * @sdp: The GFS2 superblock | |
886b1416 BP |
330 | * @blk: The data block number |
331 | * @exact: True if this needs to be an exact match | |
b3b94faa DT |
332 | * |
333 | * Returns: The resource group, or NULL if not found | |
334 | */ | |
335 | ||
66fc061b | 336 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 337 | { |
66fc061b | 338 | struct rb_node *n, *next; |
f75bbfb4 | 339 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
340 | |
341 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
342 | n = sdp->sd_rindex_tree.rb_node; |
343 | while (n) { | |
344 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
345 | next = NULL; | |
7c9ca621 | 346 | if (blk < cur->rd_addr) |
66fc061b | 347 | next = n->rb_left; |
f75bbfb4 | 348 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
349 | next = n->rb_right; |
350 | if (next == NULL) { | |
b3b94faa | 351 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
352 | if (exact) { |
353 | if (blk < cur->rd_addr) | |
354 | return NULL; | |
355 | if (blk >= cur->rd_data0 + cur->rd_data) | |
356 | return NULL; | |
357 | } | |
7c9ca621 | 358 | return cur; |
b3b94faa | 359 | } |
66fc061b | 360 | n = next; |
b3b94faa | 361 | } |
b3b94faa DT |
362 | spin_unlock(&sdp->sd_rindex_spin); |
363 | ||
364 | return NULL; | |
365 | } | |
366 | ||
367 | /** | |
368 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
369 | * @sdp: The GFS2 superblock | |
370 | * | |
371 | * Returns: The first rgrp in the filesystem | |
372 | */ | |
373 | ||
374 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
375 | { | |
7c9ca621 BP |
376 | const struct rb_node *n; |
377 | struct gfs2_rgrpd *rgd; | |
378 | ||
8339ee54 | 379 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
380 | n = rb_first(&sdp->sd_rindex_tree); |
381 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 382 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
383 | |
384 | return rgd; | |
b3b94faa DT |
385 | } |
386 | ||
387 | /** | |
388 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 389 | * @rgd: the resource group descriptor |
b3b94faa DT |
390 | * |
391 | * Returns: The next rgrp | |
392 | */ | |
393 | ||
394 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
395 | { | |
7c9ca621 BP |
396 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
397 | const struct rb_node *n; | |
398 | ||
399 | spin_lock(&sdp->sd_rindex_spin); | |
400 | n = rb_next(&rgd->rd_node); | |
401 | if (n == NULL) | |
402 | n = rb_first(&sdp->sd_rindex_tree); | |
403 | ||
404 | if (unlikely(&rgd->rd_node == n)) { | |
405 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 406 | return NULL; |
7c9ca621 BP |
407 | } |
408 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
409 | spin_unlock(&sdp->sd_rindex_spin); | |
410 | return rgd; | |
b3b94faa DT |
411 | } |
412 | ||
8339ee54 SW |
413 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
414 | { | |
415 | int x; | |
416 | ||
417 | for (x = 0; x < rgd->rd_length; x++) { | |
418 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
419 | kfree(bi->bi_clone); | |
420 | bi->bi_clone = NULL; | |
421 | } | |
422 | } | |
423 | ||
0a305e49 BP |
424 | /** |
425 | * gfs2_rs_alloc - make sure we have a reservation assigned to the inode | |
426 | * @ip: the inode for this reservation | |
427 | */ | |
428 | int gfs2_rs_alloc(struct gfs2_inode *ip) | |
429 | { | |
430 | int error = 0; | |
8e2e0047 BP |
431 | struct gfs2_blkreserv *res; |
432 | ||
433 | if (ip->i_res) | |
434 | return 0; | |
435 | ||
436 | res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); | |
437 | if (!res) | |
438 | error = -ENOMEM; | |
0a305e49 | 439 | |
24d634e8 | 440 | RB_CLEAR_NODE(&res->rs_node); |
4a993fb1 | 441 | |
0a305e49 | 442 | down_write(&ip->i_rw_mutex); |
8e2e0047 BP |
443 | if (ip->i_res) |
444 | kmem_cache_free(gfs2_rsrv_cachep, res); | |
445 | else | |
446 | ip->i_res = res; | |
0a305e49 BP |
447 | up_write(&ip->i_rw_mutex); |
448 | return error; | |
449 | } | |
450 | ||
8e2e0047 BP |
451 | static void dump_rs(struct seq_file *seq, struct gfs2_blkreserv *rs) |
452 | { | |
453 | gfs2_print_dbg(seq, " r: %llu s:%llu b:%u f:%u\n", | |
8d8b752a | 454 | rs->rs_rbm.rgd->rd_addr, gfs2_rbm_to_block(&rs->rs_rbm), |
4a993fb1 | 455 | rs->rs_rbm.offset, rs->rs_free); |
8e2e0047 BP |
456 | } |
457 | ||
0a305e49 | 458 | /** |
8e2e0047 BP |
459 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
460 | * @rs: The reservation to remove | |
461 | * | |
462 | */ | |
4a993fb1 | 463 | static void __rs_deltree(struct gfs2_inode *ip, struct gfs2_blkreserv *rs) |
8e2e0047 BP |
464 | { |
465 | struct gfs2_rgrpd *rgd; | |
466 | ||
467 | if (!gfs2_rs_active(rs)) | |
468 | return; | |
469 | ||
4a993fb1 SW |
470 | rgd = rs->rs_rbm.rgd; |
471 | trace_gfs2_rs(ip, rs, TRACE_RS_TREEDEL); | |
472 | rb_erase(&rs->rs_node, &rgd->rd_rstree); | |
24d634e8 | 473 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 BP |
474 | BUG_ON(!rgd->rd_rs_cnt); |
475 | rgd->rd_rs_cnt--; | |
476 | ||
477 | if (rs->rs_free) { | |
478 | /* return reserved blocks to the rgrp and the ip */ | |
4a993fb1 SW |
479 | BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free); |
480 | rs->rs_rbm.rgd->rd_reserved -= rs->rs_free; | |
8e2e0047 | 481 | rs->rs_free = 0; |
4a993fb1 | 482 | clear_bit(GBF_FULL, &rs->rs_rbm.bi->bi_flags); |
8e2e0047 BP |
483 | smp_mb__after_clear_bit(); |
484 | } | |
8e2e0047 BP |
485 | } |
486 | ||
487 | /** | |
488 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
489 | * @rs: The reservation to remove | |
490 | * | |
491 | */ | |
4a993fb1 | 492 | void gfs2_rs_deltree(struct gfs2_inode *ip, struct gfs2_blkreserv *rs) |
8e2e0047 BP |
493 | { |
494 | struct gfs2_rgrpd *rgd; | |
495 | ||
4a993fb1 SW |
496 | rgd = rs->rs_rbm.rgd; |
497 | if (rgd) { | |
498 | spin_lock(&rgd->rd_rsspin); | |
499 | __rs_deltree(ip, rs); | |
500 | spin_unlock(&rgd->rd_rsspin); | |
501 | } | |
8e2e0047 BP |
502 | } |
503 | ||
504 | /** | |
505 | * gfs2_rs_delete - delete a multi-block reservation | |
0a305e49 BP |
506 | * @ip: The inode for this reservation |
507 | * | |
508 | */ | |
509 | void gfs2_rs_delete(struct gfs2_inode *ip) | |
510 | { | |
511 | down_write(&ip->i_rw_mutex); | |
512 | if (ip->i_res) { | |
4a993fb1 | 513 | gfs2_rs_deltree(ip, ip->i_res); |
8e2e0047 BP |
514 | trace_gfs2_rs(ip, ip->i_res, TRACE_RS_DELETE); |
515 | BUG_ON(ip->i_res->rs_free); | |
0a305e49 BP |
516 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); |
517 | ip->i_res = NULL; | |
518 | } | |
519 | up_write(&ip->i_rw_mutex); | |
520 | } | |
521 | ||
8e2e0047 BP |
522 | /** |
523 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
524 | * @rgd: the rgrp that needs its space back | |
525 | * | |
526 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
527 | * give them back. This leave the reservation structures in tact, but removes | |
528 | * all of their corresponding "no-fly zones". | |
529 | */ | |
530 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
531 | { | |
532 | struct rb_node *n; | |
533 | struct gfs2_blkreserv *rs; | |
534 | ||
535 | spin_lock(&rgd->rd_rsspin); | |
536 | while ((n = rb_first(&rgd->rd_rstree))) { | |
537 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
4a993fb1 | 538 | __rs_deltree(NULL, rs); |
8e2e0047 BP |
539 | } |
540 | spin_unlock(&rgd->rd_rsspin); | |
541 | } | |
542 | ||
8339ee54 | 543 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 544 | { |
7c9ca621 | 545 | struct rb_node *n; |
b3b94faa DT |
546 | struct gfs2_rgrpd *rgd; |
547 | struct gfs2_glock *gl; | |
548 | ||
7c9ca621 BP |
549 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
550 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
551 | gl = rgd->rd_gl; |
552 | ||
7c9ca621 | 553 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
554 | |
555 | if (gl) { | |
8339ee54 | 556 | spin_lock(&gl->gl_spin); |
5c676f6d | 557 | gl->gl_object = NULL; |
8339ee54 | 558 | spin_unlock(&gl->gl_spin); |
29687a2a | 559 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
560 | gfs2_glock_put(gl); |
561 | } | |
562 | ||
8339ee54 | 563 | gfs2_free_clones(rgd); |
b3b94faa | 564 | kfree(rgd->rd_bits); |
8e2e0047 | 565 | return_all_reservations(rgd); |
6bdd9be6 | 566 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
567 | } |
568 | } | |
569 | ||
bb8d8a6f SW |
570 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
571 | { | |
572 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
573 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
574 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
575 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
576 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
577 | } | |
578 | ||
b3b94faa DT |
579 | /** |
580 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
581 | * @rgd: The resource group descriptor | |
582 | * | |
583 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
584 | * | |
585 | * Returns: errno | |
586 | */ | |
587 | ||
588 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
589 | { | |
590 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
591 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 592 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 593 | u32 bytes_left, bytes; |
b3b94faa DT |
594 | int x; |
595 | ||
feaa7bba SW |
596 | if (!length) |
597 | return -EINVAL; | |
598 | ||
dd894be8 | 599 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
600 | if (!rgd->rd_bits) |
601 | return -ENOMEM; | |
602 | ||
bb8d8a6f | 603 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
604 | |
605 | for (x = 0; x < length; x++) { | |
606 | bi = rgd->rd_bits + x; | |
607 | ||
60a0b8f9 | 608 | bi->bi_flags = 0; |
b3b94faa DT |
609 | /* small rgrp; bitmap stored completely in header block */ |
610 | if (length == 1) { | |
611 | bytes = bytes_left; | |
612 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
613 | bi->bi_start = 0; | |
614 | bi->bi_len = bytes; | |
615 | /* header block */ | |
616 | } else if (x == 0) { | |
617 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
618 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
619 | bi->bi_start = 0; | |
620 | bi->bi_len = bytes; | |
621 | /* last block */ | |
622 | } else if (x + 1 == length) { | |
623 | bytes = bytes_left; | |
624 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 625 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
626 | bi->bi_len = bytes; |
627 | /* other blocks */ | |
628 | } else { | |
568f4c96 SW |
629 | bytes = sdp->sd_sb.sb_bsize - |
630 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 631 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 632 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
633 | bi->bi_len = bytes; |
634 | } | |
635 | ||
636 | bytes_left -= bytes; | |
637 | } | |
638 | ||
639 | if (bytes_left) { | |
640 | gfs2_consist_rgrpd(rgd); | |
641 | return -EIO; | |
642 | } | |
643 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 644 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 645 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 646 | gfs2_rindex_print(rgd); |
b3b94faa DT |
647 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
648 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
649 | } | |
650 | return -EIO; | |
651 | } | |
652 | ||
653 | return 0; | |
654 | } | |
655 | ||
7ae8fa84 RP |
656 | /** |
657 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 658 | * @sdp: the filesystem |
7ae8fa84 RP |
659 | * |
660 | */ | |
661 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
662 | { | |
663 | u64 total_data = 0; | |
664 | struct inode *inode = sdp->sd_rindex; | |
665 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 666 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
667 | int error, rgrps; |
668 | ||
7ae8fa84 RP |
669 | for (rgrps = 0;; rgrps++) { |
670 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
671 | ||
bcd7278d | 672 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 673 | break; |
4306629e | 674 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
675 | sizeof(struct gfs2_rindex)); |
676 | if (error != sizeof(struct gfs2_rindex)) | |
677 | break; | |
bb8d8a6f | 678 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 679 | } |
7ae8fa84 RP |
680 | return total_data; |
681 | } | |
682 | ||
6aad1c3d | 683 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
684 | { |
685 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
686 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
687 | ||
688 | /* Figure out where to put new node */ | |
689 | while (*newn) { | |
690 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
691 | rd_node); | |
692 | ||
693 | parent = *newn; | |
694 | if (rgd->rd_addr < cur->rd_addr) | |
695 | newn = &((*newn)->rb_left); | |
696 | else if (rgd->rd_addr > cur->rd_addr) | |
697 | newn = &((*newn)->rb_right); | |
698 | else | |
6aad1c3d | 699 | return -EEXIST; |
7c9ca621 BP |
700 | } |
701 | ||
702 | rb_link_node(&rgd->rd_node, parent, newn); | |
703 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
704 | sdp->sd_rgrps++; |
705 | return 0; | |
7c9ca621 BP |
706 | } |
707 | ||
b3b94faa | 708 | /** |
6c53267f | 709 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 710 | * @ip: Pointer to the rindex inode |
b3b94faa | 711 | * |
8339ee54 | 712 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
713 | */ |
714 | ||
4306629e | 715 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
716 | { |
717 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
718 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
8339ee54 | 719 | struct gfs2_rindex buf; |
6c53267f RP |
720 | int error; |
721 | struct gfs2_rgrpd *rgd; | |
722 | ||
8339ee54 SW |
723 | if (pos >= i_size_read(&ip->i_inode)) |
724 | return 1; | |
725 | ||
4306629e | 726 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 727 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
728 | |
729 | if (error != sizeof(struct gfs2_rindex)) | |
730 | return (error == 0) ? 1 : error; | |
6c53267f | 731 | |
6bdd9be6 | 732 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
733 | error = -ENOMEM; |
734 | if (!rgd) | |
735 | return error; | |
736 | ||
6c53267f | 737 | rgd->rd_sbd = sdp; |
8339ee54 SW |
738 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
739 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
740 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
741 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
742 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 743 | spin_lock_init(&rgd->rd_rsspin); |
7c9ca621 | 744 | |
6c53267f RP |
745 | error = compute_bitstructs(rgd); |
746 | if (error) | |
8339ee54 | 747 | goto fail; |
6c53267f | 748 | |
bb8d8a6f | 749 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
750 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
751 | if (error) | |
8339ee54 | 752 | goto fail; |
6c53267f RP |
753 | |
754 | rgd->rd_gl->gl_object = rgd; | |
90306c41 | 755 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lvb; |
cf45b752 | 756 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
7c9ca621 BP |
757 | if (rgd->rd_data > sdp->sd_max_rg_data) |
758 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 759 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 760 | error = rgd_insert(rgd); |
8339ee54 | 761 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
762 | if (!error) |
763 | return 0; | |
764 | ||
765 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 766 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
767 | |
768 | fail: | |
769 | kfree(rgd->rd_bits); | |
770 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
771 | return error; |
772 | } | |
773 | ||
774 | /** | |
775 | * gfs2_ri_update - Pull in a new resource index from the disk | |
776 | * @ip: pointer to the rindex inode | |
777 | * | |
b3b94faa DT |
778 | * Returns: 0 on successful update, error code otherwise |
779 | */ | |
780 | ||
8339ee54 | 781 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 782 | { |
feaa7bba | 783 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
784 | int error; |
785 | ||
8339ee54 | 786 | do { |
4306629e | 787 | error = read_rindex_entry(ip); |
8339ee54 SW |
788 | } while (error == 0); |
789 | ||
790 | if (error < 0) | |
791 | return error; | |
b3b94faa | 792 | |
cf45b752 | 793 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
794 | return 0; |
795 | } | |
b3b94faa | 796 | |
b3b94faa | 797 | /** |
8339ee54 | 798 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 799 | * @sdp: The GFS2 superblock |
b3b94faa DT |
800 | * |
801 | * We grab a lock on the rindex inode to make sure that it doesn't | |
802 | * change whilst we are performing an operation. We keep this lock | |
803 | * for quite long periods of time compared to other locks. This | |
804 | * doesn't matter, since it is shared and it is very, very rarely | |
805 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
806 | * | |
807 | * This makes sure that we're using the latest copy of the resource index | |
808 | * special file, which might have been updated if someone expanded the | |
809 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
810 | * | |
8339ee54 | 811 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
812 | */ |
813 | ||
8339ee54 | 814 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 815 | { |
feaa7bba | 816 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 817 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
818 | struct gfs2_holder ri_gh; |
819 | int error = 0; | |
a365fbf3 | 820 | int unlock_required = 0; |
b3b94faa DT |
821 | |
822 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 823 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
824 | if (!gfs2_glock_is_locked_by_me(gl)) { |
825 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
826 | if (error) | |
6aad1c3d | 827 | return error; |
a365fbf3 SW |
828 | unlock_required = 1; |
829 | } | |
8339ee54 | 830 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 831 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
832 | if (unlock_required) |
833 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
834 | } |
835 | ||
836 | return error; | |
837 | } | |
838 | ||
42d52e38 | 839 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
840 | { |
841 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 842 | u32 rg_flags; |
bb8d8a6f | 843 | |
42d52e38 | 844 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 845 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
846 | rgd->rd_flags &= GFS2_RDF_MASK; |
847 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 848 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 849 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 850 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
851 | } |
852 | ||
42d52e38 | 853 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
854 | { |
855 | struct gfs2_rgrp *str = buf; | |
856 | ||
09010978 | 857 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 858 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 859 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 860 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 861 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
862 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
863 | } | |
864 | ||
90306c41 BM |
865 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
866 | { | |
867 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
868 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
869 | ||
870 | if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free || | |
871 | rgl->rl_dinodes != str->rg_dinodes || | |
872 | rgl->rl_igeneration != str->rg_igeneration) | |
873 | return 0; | |
874 | return 1; | |
875 | } | |
876 | ||
877 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) | |
878 | { | |
879 | const struct gfs2_rgrp *str = buf; | |
880 | ||
881 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
882 | rgl->rl_flags = str->rg_flags; | |
883 | rgl->rl_free = str->rg_free; | |
884 | rgl->rl_dinodes = str->rg_dinodes; | |
885 | rgl->rl_igeneration = str->rg_igeneration; | |
886 | rgl->__pad = 0UL; | |
887 | } | |
888 | ||
889 | static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change) | |
890 | { | |
891 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
892 | u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change; | |
893 | rgl->rl_unlinked = cpu_to_be32(unlinked); | |
894 | } | |
895 | ||
896 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) | |
897 | { | |
898 | struct gfs2_bitmap *bi; | |
899 | const u32 length = rgd->rd_length; | |
900 | const u8 *buffer = NULL; | |
901 | u32 i, goal, count = 0; | |
902 | ||
903 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
904 | goal = 0; | |
905 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
906 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
907 | while (goal < bi->bi_len * GFS2_NBBY) { | |
908 | goal = gfs2_bitfit(buffer, bi->bi_len, goal, | |
909 | GFS2_BLKST_UNLINKED); | |
910 | if (goal == BFITNOENT) | |
911 | break; | |
912 | count++; | |
913 | goal++; | |
914 | } | |
915 | } | |
916 | ||
917 | return count; | |
918 | } | |
919 | ||
920 | ||
b3b94faa | 921 | /** |
90306c41 BM |
922 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
923 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
924 | * |
925 | * Read in all of a Resource Group's header and bitmap blocks. | |
926 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
927 | * | |
928 | * Returns: errno | |
929 | */ | |
930 | ||
90306c41 | 931 | int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
932 | { |
933 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
934 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 935 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
936 | struct gfs2_bitmap *bi; |
937 | unsigned int x, y; | |
938 | int error; | |
939 | ||
90306c41 BM |
940 | if (rgd->rd_bits[0].bi_bh != NULL) |
941 | return 0; | |
942 | ||
b3b94faa DT |
943 | for (x = 0; x < length; x++) { |
944 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 945 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
946 | if (error) |
947 | goto fail; | |
948 | } | |
949 | ||
950 | for (y = length; y--;) { | |
951 | bi = rgd->rd_bits + y; | |
7276b3b0 | 952 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
953 | if (error) |
954 | goto fail; | |
feaa7bba | 955 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
956 | GFS2_METATYPE_RG)) { |
957 | error = -EIO; | |
958 | goto fail; | |
959 | } | |
960 | } | |
961 | ||
cf45b752 | 962 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
963 | for (x = 0; x < length; x++) |
964 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 965 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 966 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 967 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa | 968 | } |
90306c41 BM |
969 | if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
970 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); | |
971 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
972 | rgd->rd_bits[0].bi_bh->b_data); | |
973 | } | |
974 | else if (sdp->sd_args.ar_rgrplvb) { | |
975 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
976 | gfs2_consist_rgrpd(rgd); | |
977 | error = -EIO; | |
978 | goto fail; | |
979 | } | |
980 | if (rgd->rd_rgl->rl_unlinked == 0) | |
981 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
982 | } | |
b3b94faa DT |
983 | return 0; |
984 | ||
feaa7bba | 985 | fail: |
b3b94faa DT |
986 | while (x--) { |
987 | bi = rgd->rd_bits + x; | |
988 | brelse(bi->bi_bh); | |
989 | bi->bi_bh = NULL; | |
990 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
991 | } | |
b3b94faa DT |
992 | |
993 | return error; | |
994 | } | |
995 | ||
90306c41 BM |
996 | int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
997 | { | |
998 | u32 rl_flags; | |
999 | ||
1000 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1001 | return 0; | |
1002 | ||
1003 | if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) | |
1004 | return gfs2_rgrp_bh_get(rgd); | |
1005 | ||
1006 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1007 | rl_flags &= ~GFS2_RDF_MASK; | |
1008 | rgd->rd_flags &= GFS2_RDF_MASK; | |
1009 | rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); | |
1010 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1011 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1012 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
1013 | rgd->rd_free_clone = rgd->rd_free; | |
1014 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); | |
1015 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1020 | { | |
1021 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1022 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1023 | ||
1024 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1025 | return 0; | |
1026 | return gfs2_rgrp_bh_get((struct gfs2_rgrpd *)gh->gh_gl->gl_object); | |
1027 | } | |
1028 | ||
b3b94faa | 1029 | /** |
7c9ca621 | 1030 | * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
886b1416 | 1031 | * @gh: The glock holder for the resource group |
b3b94faa DT |
1032 | * |
1033 | */ | |
1034 | ||
7c9ca621 | 1035 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) |
b3b94faa | 1036 | { |
7c9ca621 | 1037 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
bb8d8a6f | 1038 | int x, length = rgd->rd_length; |
b3b94faa | 1039 | |
b3b94faa DT |
1040 | for (x = 0; x < length; x++) { |
1041 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1042 | if (bi->bi_bh) { |
1043 | brelse(bi->bi_bh); | |
1044 | bi->bi_bh = NULL; | |
1045 | } | |
b3b94faa DT |
1046 | } |
1047 | ||
b3b94faa DT |
1048 | } |
1049 | ||
66fc061b | 1050 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1051 | struct buffer_head *bh, |
66fc061b | 1052 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1053 | { |
1054 | struct super_block *sb = sdp->sd_vfs; | |
1055 | struct block_device *bdev = sb->s_bdev; | |
1056 | const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / | |
e1defc4f | 1057 | bdev_logical_block_size(sb->s_bdev); |
f15ab561 | 1058 | u64 blk; |
64d576ba | 1059 | sector_t start = 0; |
f15ab561 SW |
1060 | sector_t nr_sects = 0; |
1061 | int rv; | |
1062 | unsigned int x; | |
66fc061b SW |
1063 | u32 trimmed = 0; |
1064 | u8 diff; | |
f15ab561 SW |
1065 | |
1066 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
1067 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1068 | clone += bi->bi_offset; | |
1069 | clone += x; | |
1070 | if (bh) { | |
1071 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1072 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1073 | } else { | |
1074 | diff = ~(*clone | (*clone >> 1)); | |
1075 | } | |
f15ab561 SW |
1076 | diff &= 0x55; |
1077 | if (diff == 0) | |
1078 | continue; | |
1079 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
1080 | blk *= sects_per_blk; /* convert to sectors */ | |
1081 | while(diff) { | |
1082 | if (diff & 1) { | |
1083 | if (nr_sects == 0) | |
1084 | goto start_new_extent; | |
1085 | if ((start + nr_sects) != blk) { | |
66fc061b SW |
1086 | if (nr_sects >= minlen) { |
1087 | rv = blkdev_issue_discard(bdev, | |
1088 | start, nr_sects, | |
1089 | GFP_NOFS, 0); | |
1090 | if (rv) | |
1091 | goto fail; | |
1092 | trimmed += nr_sects; | |
1093 | } | |
f15ab561 SW |
1094 | nr_sects = 0; |
1095 | start_new_extent: | |
1096 | start = blk; | |
1097 | } | |
1098 | nr_sects += sects_per_blk; | |
1099 | } | |
1100 | diff >>= 2; | |
1101 | blk += sects_per_blk; | |
1102 | } | |
1103 | } | |
66fc061b | 1104 | if (nr_sects >= minlen) { |
dd3932ed | 1105 | rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0); |
f15ab561 SW |
1106 | if (rv) |
1107 | goto fail; | |
66fc061b | 1108 | trimmed += nr_sects; |
f15ab561 | 1109 | } |
66fc061b SW |
1110 | if (ptrimmed) |
1111 | *ptrimmed = trimmed; | |
1112 | return 0; | |
1113 | ||
f15ab561 | 1114 | fail: |
66fc061b SW |
1115 | if (sdp->sd_args.ar_discard) |
1116 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 1117 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1118 | return -EIO; |
1119 | } | |
1120 | ||
1121 | /** | |
1122 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1123 | * @filp: Any file on the filesystem | |
1124 | * @argp: Pointer to the arguments (also used to pass result) | |
1125 | * | |
1126 | * Returns: 0 on success, otherwise error code | |
1127 | */ | |
1128 | ||
1129 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1130 | { | |
1131 | struct inode *inode = filp->f_dentry->d_inode; | |
1132 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1133 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1134 | struct buffer_head *bh; | |
1135 | struct gfs2_rgrpd *rgd; | |
1136 | struct gfs2_rgrpd *rgd_end; | |
1137 | struct gfs2_holder gh; | |
1138 | struct fstrim_range r; | |
1139 | int ret = 0; | |
1140 | u64 amt; | |
1141 | u64 trimmed = 0; | |
1142 | unsigned int x; | |
1143 | ||
1144 | if (!capable(CAP_SYS_ADMIN)) | |
1145 | return -EPERM; | |
1146 | ||
1147 | if (!blk_queue_discard(q)) | |
1148 | return -EOPNOTSUPP; | |
1149 | ||
66fc061b SW |
1150 | if (argp == NULL) { |
1151 | r.start = 0; | |
1152 | r.len = ULLONG_MAX; | |
1153 | r.minlen = 0; | |
1154 | } else if (copy_from_user(&r, argp, sizeof(r))) | |
1155 | return -EFAULT; | |
1156 | ||
5e2f7d61 BP |
1157 | ret = gfs2_rindex_update(sdp); |
1158 | if (ret) | |
1159 | return ret; | |
1160 | ||
66fc061b SW |
1161 | rgd = gfs2_blk2rgrpd(sdp, r.start, 0); |
1162 | rgd_end = gfs2_blk2rgrpd(sdp, r.start + r.len, 0); | |
1163 | ||
1164 | while (1) { | |
1165 | ||
1166 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1167 | if (ret) | |
1168 | goto out; | |
1169 | ||
1170 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1171 | /* Trim each bitmap in the rgrp */ | |
1172 | for (x = 0; x < rgd->rd_length; x++) { | |
1173 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
1174 | ret = gfs2_rgrp_send_discards(sdp, rgd->rd_data0, NULL, bi, r.minlen, &amt); | |
1175 | if (ret) { | |
1176 | gfs2_glock_dq_uninit(&gh); | |
1177 | goto out; | |
1178 | } | |
1179 | trimmed += amt; | |
1180 | } | |
1181 | ||
1182 | /* Mark rgrp as having been trimmed */ | |
1183 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1184 | if (ret == 0) { | |
1185 | bh = rgd->rd_bits[0].bi_bh; | |
1186 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
1187 | gfs2_trans_add_bh(rgd->rd_gl, bh, 1); | |
1188 | gfs2_rgrp_out(rgd, bh->b_data); | |
90306c41 | 1189 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data); |
66fc061b SW |
1190 | gfs2_trans_end(sdp); |
1191 | } | |
1192 | } | |
1193 | gfs2_glock_dq_uninit(&gh); | |
1194 | ||
1195 | if (rgd == rgd_end) | |
1196 | break; | |
1197 | ||
1198 | rgd = gfs2_rgrpd_get_next(rgd); | |
1199 | } | |
1200 | ||
1201 | out: | |
1202 | r.len = trimmed << 9; | |
1203 | if (argp && copy_to_user(argp, &r, sizeof(r))) | |
1204 | return -EFAULT; | |
1205 | ||
1206 | return ret; | |
f15ab561 SW |
1207 | } |
1208 | ||
8e2e0047 BP |
1209 | /** |
1210 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
1211 | * @bi: the bitmap with the blocks | |
1212 | * @ip: the inode structure | |
1213 | * @biblk: the 32-bit block number relative to the start of the bitmap | |
1214 | * @amount: the number of blocks to reserve | |
1215 | * | |
1216 | * Returns: NULL - reservation was already taken, so not inserted | |
1217 | * pointer to the inserted reservation | |
1218 | */ | |
1219 | static struct gfs2_blkreserv *rs_insert(struct gfs2_bitmap *bi, | |
1220 | struct gfs2_inode *ip, u32 biblk, | |
1221 | int amount) | |
1222 | { | |
1223 | struct rb_node **newn, *parent = NULL; | |
1224 | int rc; | |
1225 | struct gfs2_blkreserv *rs = ip->i_res; | |
4a993fb1 | 1226 | struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd; |
8e2e0047 BP |
1227 | u64 fsblock = gfs2_bi2rgd_blk(bi, biblk) + rgd->rd_data0; |
1228 | ||
1229 | spin_lock(&rgd->rd_rsspin); | |
1230 | newn = &rgd->rd_rstree.rb_node; | |
1231 | BUG_ON(!ip->i_res); | |
1232 | BUG_ON(gfs2_rs_active(rs)); | |
1233 | /* Figure out where to put new node */ | |
1234 | /*BUG_ON(!gfs2_glock_is_locked_by_me(rgd->rd_gl));*/ | |
1235 | while (*newn) { | |
1236 | struct gfs2_blkreserv *cur = | |
1237 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1238 | ||
1239 | parent = *newn; | |
1240 | rc = rs_cmp(fsblock, amount, cur); | |
1241 | if (rc > 0) | |
1242 | newn = &((*newn)->rb_right); | |
1243 | else if (rc < 0) | |
1244 | newn = &((*newn)->rb_left); | |
1245 | else { | |
1246 | spin_unlock(&rgd->rd_rsspin); | |
1247 | return NULL; /* reservation already in use */ | |
1248 | } | |
1249 | } | |
1250 | ||
1251 | /* Do our reservation work */ | |
1252 | rs = ip->i_res; | |
1253 | rs->rs_free = amount; | |
4a993fb1 SW |
1254 | rs->rs_rbm.offset = biblk; |
1255 | rs->rs_rbm.bi = bi; | |
8e2e0047 BP |
1256 | rb_link_node(&rs->rs_node, parent, newn); |
1257 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1258 | ||
8e2e0047 BP |
1259 | /* Do our rgrp accounting for the reservation */ |
1260 | rgd->rd_reserved += amount; /* blocks reserved */ | |
1261 | rgd->rd_rs_cnt++; /* number of in-tree reservations */ | |
1262 | spin_unlock(&rgd->rd_rsspin); | |
1263 | trace_gfs2_rs(ip, rs, TRACE_RS_INSERT); | |
1264 | return rs; | |
1265 | } | |
1266 | ||
1267 | /** | |
1268 | * unclaimed_blocks - return number of blocks that aren't spoken for | |
1269 | */ | |
1270 | static u32 unclaimed_blocks(struct gfs2_rgrpd *rgd) | |
1271 | { | |
1272 | return rgd->rd_free_clone - rgd->rd_reserved; | |
1273 | } | |
1274 | ||
1275 | /** | |
1276 | * rg_mblk_search - find a group of multiple free blocks | |
1277 | * @rgd: the resource group descriptor | |
1278 | * @rs: the block reservation | |
1279 | * @ip: pointer to the inode for which we're reserving blocks | |
1280 | * | |
1281 | * This is very similar to rgblk_search, except we're looking for whole | |
1282 | * 64-bit words that represent a chunk of 32 free blocks. I'm only focusing | |
1283 | * on aligned dwords for speed's sake. | |
1284 | * | |
1285 | * Returns: 0 if successful or BFITNOENT if there isn't enough free space | |
1286 | */ | |
1287 | ||
71f890f7 | 1288 | static int rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, unsigned requested) |
8e2e0047 BP |
1289 | { |
1290 | struct gfs2_bitmap *bi = rgd->rd_bits; | |
1291 | const u32 length = rgd->rd_length; | |
1292 | u32 blk; | |
1293 | unsigned int buf, x, search_bytes; | |
1294 | u8 *buffer = NULL; | |
1295 | u8 *ptr, *end, *nonzero; | |
1296 | u32 goal, rsv_bytes; | |
1297 | struct gfs2_blkreserv *rs; | |
1298 | u32 best_rs_bytes, unclaimed; | |
1299 | int best_rs_blocks; | |
1300 | ||
1301 | /* Find bitmap block that contains bits for goal block */ | |
1302 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
1303 | goal = ip->i_goal - rgd->rd_data0; | |
1304 | else | |
1305 | goal = rgd->rd_last_alloc; | |
1306 | for (buf = 0; buf < length; buf++) { | |
1307 | bi = rgd->rd_bits + buf; | |
1308 | /* Convert scope of "goal" from rgrp-wide to within | |
1309 | found bit block */ | |
1310 | if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { | |
1311 | goal -= bi->bi_start * GFS2_NBBY; | |
1312 | goto do_search; | |
1313 | } | |
1314 | } | |
1315 | buf = 0; | |
1316 | goal = 0; | |
1317 | ||
1318 | do_search: | |
1319 | best_rs_blocks = max_t(int, atomic_read(&ip->i_res->rs_sizehint), | |
1320 | (RGRP_RSRV_MINBLKS * rgd->rd_length)); | |
1321 | best_rs_bytes = (best_rs_blocks * | |
1322 | (1 + (RSRV_CONTENTION_FACTOR * rgd->rd_rs_cnt))) / | |
1323 | GFS2_NBBY; /* 1 + is for our not-yet-created reservation */ | |
1324 | best_rs_bytes = ALIGN(best_rs_bytes, sizeof(u64)); | |
1325 | unclaimed = unclaimed_blocks(rgd); | |
1326 | if (best_rs_bytes * GFS2_NBBY > unclaimed) | |
1327 | best_rs_bytes = unclaimed >> GFS2_BIT_SIZE; | |
1328 | ||
1329 | for (x = 0; x <= length; x++) { | |
1330 | bi = rgd->rd_bits + buf; | |
1331 | ||
1332 | if (test_bit(GBF_FULL, &bi->bi_flags)) | |
1333 | goto skip; | |
1334 | ||
1335 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
1336 | if (bi->bi_clone) | |
1337 | buffer = bi->bi_clone + bi->bi_offset; | |
1338 | else | |
1339 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1340 | ||
1341 | /* We have to keep the reservations aligned on u64 boundaries | |
1342 | otherwise we could get situations where a byte can't be | |
1343 | used because it's after a reservation, but a free bit still | |
1344 | is within the reservation's area. */ | |
1345 | ptr = buffer + ALIGN(goal >> GFS2_BIT_SIZE, sizeof(u64)); | |
1346 | end = (buffer + bi->bi_len); | |
1347 | while (ptr < end) { | |
1348 | rsv_bytes = 0; | |
1349 | if ((ptr + best_rs_bytes) <= end) | |
1350 | search_bytes = best_rs_bytes; | |
1351 | else | |
1352 | search_bytes = end - ptr; | |
1353 | BUG_ON(!search_bytes); | |
1354 | nonzero = memchr_inv(ptr, 0, search_bytes); | |
1355 | /* If the lot is all zeroes, reserve the whole size. If | |
1356 | there's enough zeroes to satisfy the request, use | |
1357 | what we can. If there's not enough, keep looking. */ | |
1358 | if (nonzero == NULL) | |
1359 | rsv_bytes = search_bytes; | |
71f890f7 | 1360 | else if ((nonzero - ptr) * GFS2_NBBY >= requested) |
8e2e0047 BP |
1361 | rsv_bytes = (nonzero - ptr); |
1362 | ||
1363 | if (rsv_bytes) { | |
1364 | blk = ((ptr - buffer) * GFS2_NBBY); | |
1365 | BUG_ON(blk >= bi->bi_len * GFS2_NBBY); | |
1366 | rs = rs_insert(bi, ip, blk, | |
1367 | rsv_bytes * GFS2_NBBY); | |
1368 | if (IS_ERR(rs)) | |
1369 | return PTR_ERR(rs); | |
1370 | if (rs) | |
1371 | return 0; | |
1372 | } | |
1373 | ptr += ALIGN(search_bytes, sizeof(u64)); | |
1374 | } | |
1375 | skip: | |
1376 | /* Try next bitmap block (wrap back to rgrp header | |
1377 | if at end) */ | |
1378 | buf++; | |
1379 | buf %= length; | |
1380 | goal = 0; | |
1381 | } | |
1382 | ||
1383 | return BFITNOENT; | |
1384 | } | |
1385 | ||
b3b94faa DT |
1386 | /** |
1387 | * try_rgrp_fit - See if a given reservation will fit in a given RG | |
1388 | * @rgd: the RG data | |
54335b1f | 1389 | * @ip: the inode |
b3b94faa DT |
1390 | * |
1391 | * If there's room for the requested blocks to be allocated from the RG: | |
8e2e0047 BP |
1392 | * This will try to get a multi-block reservation first, and if that doesn't |
1393 | * fit, it will take what it can. | |
b3b94faa DT |
1394 | * |
1395 | * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) | |
1396 | */ | |
1397 | ||
71f890f7 SW |
1398 | static int try_rgrp_fit(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
1399 | unsigned requested) | |
b3b94faa | 1400 | { |
09010978 | 1401 | if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) |
a43a4906 | 1402 | return 0; |
8e2e0047 BP |
1403 | /* Look for a multi-block reservation. */ |
1404 | if (unclaimed_blocks(rgd) >= RGRP_RSRV_MINBLKS && | |
71f890f7 | 1405 | rg_mblk_search(rgd, ip, requested) != BFITNOENT) |
8e2e0047 | 1406 | return 1; |
71f890f7 | 1407 | if (unclaimed_blocks(rgd) >= requested) |
7c9ca621 | 1408 | return 1; |
b3b94faa | 1409 | |
8e2e0047 | 1410 | return 0; |
b3e47ca0 BP |
1411 | } |
1412 | ||
5b924ae2 SW |
1413 | /** |
1414 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1415 | * @rgd: The resource group | |
1416 | * @block: The starting block | |
1417 | * @ip: Ignore any reservations for this inode | |
1418 | * | |
1419 | * If the block does not appear in any reservation, then return the | |
1420 | * block number unchanged. If it does appear in the reservation, then | |
1421 | * keep looking through the tree of reservations in order to find the | |
1422 | * first block number which is not reserved. | |
1423 | */ | |
1424 | ||
1425 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
1426 | const struct gfs2_inode *ip) | |
1427 | { | |
1428 | struct gfs2_blkreserv *rs; | |
1429 | struct rb_node *n; | |
1430 | int rc; | |
1431 | ||
1432 | spin_lock(&rgd->rd_rsspin); | |
1433 | n = rb_first(&rgd->rd_rstree); | |
1434 | while (n) { | |
1435 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1436 | rc = rs_cmp(block, 1, rs); | |
1437 | if (rc < 0) | |
1438 | n = n->rb_left; | |
1439 | else if (rc > 0) | |
1440 | n = n->rb_right; | |
1441 | else | |
1442 | break; | |
1443 | } | |
1444 | ||
1445 | if (n) { | |
1446 | while ((rs_cmp(block, 1, rs) == 0) && (ip->i_res != rs)) { | |
1447 | block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free; | |
1448 | n = rb_next(&rs->rs_node); | |
1449 | if (n == NULL) | |
1450 | break; | |
1451 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1452 | } | |
1453 | } | |
1454 | ||
1455 | spin_unlock(&rgd->rd_rsspin); | |
1456 | return block; | |
1457 | } | |
1458 | ||
1459 | /** | |
1460 | * gfs2_rbm_from_block - Set the rbm based upon rgd and block number | |
1461 | * @rbm: The rbm with rgd already set correctly | |
1462 | * @block: The block number (filesystem relative) | |
1463 | * | |
1464 | * This sets the bi and offset members of an rbm based on a | |
1465 | * resource group and a filesystem relative block number. The | |
1466 | * resource group must be set in the rbm on entry, the bi and | |
1467 | * offset members will be set by this function. | |
1468 | * | |
1469 | * Returns: 0 on success, or an error code | |
1470 | */ | |
1471 | ||
1472 | static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block) | |
1473 | { | |
1474 | u64 rblock = block - rbm->rgd->rd_data0; | |
1475 | u32 goal = (u32)rblock; | |
1476 | int x; | |
1477 | ||
1478 | if (WARN_ON_ONCE(rblock > UINT_MAX)) | |
1479 | return -EINVAL; | |
8d8b752a BP |
1480 | if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data) |
1481 | return -E2BIG; | |
5b924ae2 SW |
1482 | |
1483 | for (x = 0; x < rbm->rgd->rd_length; x++) { | |
1484 | rbm->bi = rbm->rgd->rd_bits + x; | |
1485 | if (goal < (rbm->bi->bi_start + rbm->bi->bi_len) * GFS2_NBBY) { | |
1486 | rbm->offset = goal - (rbm->bi->bi_start * GFS2_NBBY); | |
8d8b752a | 1487 | break; |
5b924ae2 SW |
1488 | } |
1489 | } | |
1490 | ||
8d8b752a | 1491 | return 0; |
5b924ae2 SW |
1492 | } |
1493 | ||
1494 | /** | |
1495 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1496 | * @rbm: The current position in the resource group | |
1497 | * | |
1498 | * This checks the current position in the rgrp to see whether there is | |
1499 | * a reservation covering this block. If not then this function is a | |
1500 | * no-op. If there is, then the position is moved to the end of the | |
1501 | * contiguous reservation(s) so that we are pointing at the first | |
1502 | * non-reserved block. | |
1503 | * | |
1504 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1505 | */ | |
1506 | ||
1507 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
1508 | const struct gfs2_inode *ip) | |
1509 | { | |
1510 | u64 block = gfs2_rbm_to_block(rbm); | |
1511 | u64 nblock; | |
1512 | int ret; | |
1513 | ||
1514 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, ip); | |
1515 | if (nblock == block) | |
1516 | return 0; | |
1517 | ret = gfs2_rbm_from_block(rbm, nblock); | |
1518 | if (ret < 0) | |
1519 | return ret; | |
1520 | return 1; | |
1521 | } | |
1522 | ||
1523 | /** | |
1524 | * gfs2_rbm_find - Look for blocks of a particular state | |
1525 | * @rbm: Value/result starting position and final position | |
1526 | * @state: The state which we want to find | |
1527 | * @ip: If set, check for reservations | |
1528 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping | |
1529 | * around until we've reached the starting point. | |
1530 | * | |
1531 | * Side effects: | |
1532 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1533 | * has no free blocks in it. | |
1534 | * | |
1535 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1536 | */ | |
1537 | ||
1538 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, | |
1539 | const struct gfs2_inode *ip, bool nowrap) | |
1540 | { | |
1541 | struct buffer_head *bh; | |
1542 | struct gfs2_bitmap *initial_bi; | |
1543 | u32 initial_offset; | |
1544 | u32 offset; | |
1545 | u8 *buffer; | |
1546 | int index; | |
1547 | int n = 0; | |
1548 | int iters = rbm->rgd->rd_length; | |
1549 | int ret; | |
1550 | ||
1551 | /* If we are not starting at the beginning of a bitmap, then we | |
1552 | * need to add one to the bitmap count to ensure that we search | |
1553 | * the starting bitmap twice. | |
1554 | */ | |
1555 | if (rbm->offset != 0) | |
1556 | iters++; | |
1557 | ||
1558 | while(1) { | |
1559 | if (test_bit(GBF_FULL, &rbm->bi->bi_flags) && | |
1560 | (state == GFS2_BLKST_FREE)) | |
1561 | goto next_bitmap; | |
1562 | ||
1563 | bh = rbm->bi->bi_bh; | |
1564 | buffer = bh->b_data + rbm->bi->bi_offset; | |
1565 | WARN_ON(!buffer_uptodate(bh)); | |
1566 | if (state != GFS2_BLKST_UNLINKED && rbm->bi->bi_clone) | |
1567 | buffer = rbm->bi->bi_clone + rbm->bi->bi_offset; | |
5b924ae2 SW |
1568 | initial_offset = rbm->offset; |
1569 | offset = gfs2_bitfit(buffer, rbm->bi->bi_len, rbm->offset, state); | |
1570 | if (offset == BFITNOENT) | |
1571 | goto bitmap_full; | |
1572 | rbm->offset = offset; | |
1573 | if (ip == NULL) | |
1574 | return 0; | |
1575 | ||
1576 | initial_bi = rbm->bi; | |
1577 | ret = gfs2_reservation_check_and_update(rbm, ip); | |
1578 | if (ret == 0) | |
1579 | return 0; | |
1580 | if (ret > 0) { | |
1581 | n += (rbm->bi - initial_bi); | |
8d8b752a | 1582 | goto next_iter; |
5b924ae2 | 1583 | } |
5d50d532 SW |
1584 | if (ret == -E2BIG) { |
1585 | index = 0; | |
1586 | rbm->offset = 0; | |
1587 | n += (rbm->bi - initial_bi); | |
1588 | goto res_covered_end_of_rgrp; | |
1589 | } | |
5b924ae2 SW |
1590 | return ret; |
1591 | ||
1592 | bitmap_full: /* Mark bitmap as full and fall through */ | |
1593 | if ((state == GFS2_BLKST_FREE) && initial_offset == 0) | |
1594 | set_bit(GBF_FULL, &rbm->bi->bi_flags); | |
1595 | ||
1596 | next_bitmap: /* Find next bitmap in the rgrp */ | |
1597 | rbm->offset = 0; | |
1598 | index = rbm->bi - rbm->rgd->rd_bits; | |
1599 | index++; | |
1600 | if (index == rbm->rgd->rd_length) | |
1601 | index = 0; | |
5d50d532 | 1602 | res_covered_end_of_rgrp: |
5b924ae2 SW |
1603 | rbm->bi = &rbm->rgd->rd_bits[index]; |
1604 | if ((index == 0) && nowrap) | |
1605 | break; | |
1606 | n++; | |
8d8b752a | 1607 | next_iter: |
5b924ae2 SW |
1608 | if (n >= iters) |
1609 | break; | |
1610 | } | |
1611 | ||
1612 | return -ENOSPC; | |
1613 | } | |
1614 | ||
c8cdf479 SW |
1615 | /** |
1616 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1617 | * @rgd: The rgrp | |
886b1416 BP |
1618 | * @last_unlinked: block address of the last dinode we unlinked |
1619 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1620 | * |
1a0eae88 BP |
1621 | * Returns: 0 if no error |
1622 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1623 | */ |
1624 | ||
044b9414 | 1625 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1626 | { |
5b924ae2 | 1627 | u64 block; |
5f3eae75 | 1628 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1629 | struct gfs2_glock *gl; |
1630 | struct gfs2_inode *ip; | |
1631 | int error; | |
1632 | int found = 0; | |
5b924ae2 | 1633 | struct gfs2_rbm rbm = { .rgd = rgd, .bi = rgd->rd_bits, .offset = 0 }; |
c8cdf479 | 1634 | |
5b924ae2 | 1635 | while (1) { |
5f3eae75 | 1636 | down_write(&sdp->sd_log_flush_lock); |
5b924ae2 | 1637 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, true); |
5f3eae75 | 1638 | up_write(&sdp->sd_log_flush_lock); |
5b924ae2 SW |
1639 | if (error == -ENOSPC) |
1640 | break; | |
1641 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1642 | break; |
b3e47ca0 | 1643 | |
5b924ae2 SW |
1644 | block = gfs2_rbm_to_block(&rbm); |
1645 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1646 | break; | |
1647 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1648 | continue; |
5b924ae2 | 1649 | if (block == skip) |
1e19a195 | 1650 | continue; |
5b924ae2 | 1651 | *last_unlinked = block; |
044b9414 | 1652 | |
5b924ae2 | 1653 | error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl); |
044b9414 SW |
1654 | if (error) |
1655 | continue; | |
1656 | ||
1657 | /* If the inode is already in cache, we can ignore it here | |
1658 | * because the existing inode disposal code will deal with | |
1659 | * it when all refs have gone away. Accessing gl_object like | |
1660 | * this is not safe in general. Here it is ok because we do | |
1661 | * not dereference the pointer, and we only need an approx | |
1662 | * answer to whether it is NULL or not. | |
1663 | */ | |
1664 | ip = gl->gl_object; | |
1665 | ||
1666 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1667 | gfs2_glock_put(gl); | |
1668 | else | |
1669 | found++; | |
1670 | ||
1671 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1672 | if (found > NR_CPUS) |
044b9414 | 1673 | return; |
c8cdf479 SW |
1674 | } |
1675 | ||
1676 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1677 | return; |
c8cdf479 SW |
1678 | } |
1679 | ||
b3b94faa | 1680 | /** |
666d1d8a | 1681 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1682 | * @ip: the inode to reserve space for |
666d1d8a | 1683 | * @requested: the number of blocks to be reserved |
b3b94faa DT |
1684 | * |
1685 | * Returns: errno | |
1686 | */ | |
1687 | ||
666d1d8a | 1688 | int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested) |
b3b94faa | 1689 | { |
feaa7bba | 1690 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 1691 | struct gfs2_rgrpd *begin = NULL; |
564e12b1 | 1692 | struct gfs2_blkreserv *rs = ip->i_res; |
666d1d8a BP |
1693 | int error = 0, rg_locked, flags = LM_FLAG_TRY; |
1694 | u64 last_unlinked = NO_BLOCK; | |
7c9ca621 | 1695 | int loops = 0; |
b3b94faa | 1696 | |
90306c41 BM |
1697 | if (sdp->sd_args.ar_rgrplvb) |
1698 | flags |= GL_SKIP; | |
666d1d8a BP |
1699 | if (gfs2_assert_warn(sdp, requested)) { |
1700 | error = -EINVAL; | |
1701 | goto out; | |
1702 | } | |
8e2e0047 | 1703 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1704 | begin = rs->rs_rbm.rgd; |
8e2e0047 BP |
1705 | flags = 0; /* Yoda: Do or do not. There is no try */ |
1706 | } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) { | |
4a993fb1 | 1707 | rs->rs_rbm.rgd = begin = ip->i_rgd; |
8e2e0047 | 1708 | } else { |
4a993fb1 | 1709 | rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 1710 | } |
4a993fb1 | 1711 | if (rs->rs_rbm.rgd == NULL) |
7c9ca621 BP |
1712 | return -EBADSLT; |
1713 | ||
1714 | while (loops < 3) { | |
292c8c14 AD |
1715 | rg_locked = 0; |
1716 | ||
4a993fb1 | 1717 | if (gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) { |
292c8c14 AD |
1718 | rg_locked = 1; |
1719 | error = 0; | |
8e2e0047 | 1720 | } else if (!loops && !gfs2_rs_active(rs) && |
4a993fb1 | 1721 | rs->rs_rbm.rgd->rd_rs_cnt > RGRP_RSRV_MAX_CONTENDERS) { |
8e2e0047 BP |
1722 | /* If the rgrp already is maxed out for contenders, |
1723 | we can eliminate it as a "first pass" without even | |
1724 | requesting the rgrp glock. */ | |
1725 | error = GLR_TRYFAILED; | |
292c8c14 | 1726 | } else { |
4a993fb1 | 1727 | error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl, |
8e2e0047 BP |
1728 | LM_ST_EXCLUSIVE, flags, |
1729 | &rs->rs_rgd_gh); | |
90306c41 | 1730 | if (!error && sdp->sd_args.ar_rgrplvb) { |
4a993fb1 | 1731 | error = update_rgrp_lvb(rs->rs_rbm.rgd); |
90306c41 BM |
1732 | if (error) { |
1733 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
1734 | return error; | |
1735 | } | |
1736 | } | |
292c8c14 | 1737 | } |
b3b94faa DT |
1738 | switch (error) { |
1739 | case 0: | |
8e2e0047 | 1740 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1741 | if (unclaimed_blocks(rs->rs_rbm.rgd) + |
71f890f7 | 1742 | rs->rs_free >= requested) { |
4a993fb1 | 1743 | ip->i_rgd = rs->rs_rbm.rgd; |
8e2e0047 BP |
1744 | return 0; |
1745 | } | |
1746 | /* We have a multi-block reservation, but the | |
1747 | rgrp doesn't have enough free blocks to | |
1748 | satisfy the request. Free the reservation | |
1749 | and look for a suitable rgrp. */ | |
4a993fb1 | 1750 | gfs2_rs_deltree(ip, rs); |
8e2e0047 | 1751 | } |
4a993fb1 | 1752 | if (try_rgrp_fit(rs->rs_rbm.rgd, ip, requested)) { |
90306c41 | 1753 | if (sdp->sd_args.ar_rgrplvb) |
4a993fb1 SW |
1754 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); |
1755 | ip->i_rgd = rs->rs_rbm.rgd; | |
7c9ca621 | 1756 | return 0; |
54335b1f | 1757 | } |
4a993fb1 | 1758 | if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK) { |
90306c41 | 1759 | if (sdp->sd_args.ar_rgrplvb) |
4a993fb1 SW |
1760 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); |
1761 | try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked, | |
666d1d8a | 1762 | ip->i_no_addr); |
90306c41 | 1763 | } |
292c8c14 | 1764 | if (!rg_locked) |
564e12b1 | 1765 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
9cabcdbd | 1766 | /* fall through */ |
b3b94faa | 1767 | case GLR_TRYFAILED: |
4a993fb1 SW |
1768 | rs->rs_rbm.rgd = gfs2_rgrpd_get_next(rs->rs_rbm.rgd); |
1769 | rs->rs_rbm.rgd = rs->rs_rbm.rgd ? : begin; /* if NULL, wrap */ | |
1770 | if (rs->rs_rbm.rgd != begin) /* If we didn't wrap */ | |
666d1d8a BP |
1771 | break; |
1772 | ||
1773 | flags &= ~LM_FLAG_TRY; | |
1774 | loops++; | |
1775 | /* Check that fs hasn't grown if writing to rindex */ | |
1776 | if (ip == GFS2_I(sdp->sd_rindex) && | |
1777 | !sdp->sd_rindex_uptodate) { | |
1778 | error = gfs2_ri_update(ip); | |
1779 | if (error) | |
1780 | goto out; | |
1781 | } else if (loops == 2) | |
1782 | /* Flushing the log may release space */ | |
1783 | gfs2_log_flush(sdp, NULL); | |
b3b94faa | 1784 | break; |
b3b94faa | 1785 | default: |
666d1d8a | 1786 | goto out; |
b3b94faa | 1787 | } |
b3b94faa | 1788 | } |
666d1d8a | 1789 | error = -ENOSPC; |
b3b94faa | 1790 | |
564e12b1 | 1791 | out: |
9ae32429 | 1792 | return error; |
b3b94faa DT |
1793 | } |
1794 | ||
1795 | /** | |
1796 | * gfs2_inplace_release - release an inplace reservation | |
1797 | * @ip: the inode the reservation was taken out on | |
1798 | * | |
1799 | * Release a reservation made by gfs2_inplace_reserve(). | |
1800 | */ | |
1801 | ||
1802 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1803 | { | |
564e12b1 | 1804 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 1805 | |
564e12b1 BP |
1806 | if (rs->rs_rgd_gh.gh_gl) |
1807 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
b3b94faa DT |
1808 | } |
1809 | ||
1810 | /** | |
1811 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1812 | * @rgd: the resource group holding the block | |
1813 | * @block: the block number | |
1814 | * | |
1815 | * Returns: The block type (GFS2_BLKST_*) | |
1816 | */ | |
1817 | ||
acf7e244 | 1818 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 1819 | { |
3983903a SW |
1820 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1821 | int ret; | |
b3b94faa | 1822 | |
3983903a SW |
1823 | ret = gfs2_rbm_from_block(&rbm, block); |
1824 | WARN_ON_ONCE(ret != 0); | |
b3b94faa | 1825 | |
c04a2ef3 | 1826 | return gfs2_testbit(&rbm); |
b3b94faa DT |
1827 | } |
1828 | ||
60a0b8f9 | 1829 | |
b3e47ca0 BP |
1830 | /** |
1831 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 1832 | * @rbm: the resource group information |
b3e47ca0 | 1833 | * @dinode: TRUE if the first block we allocate is for a dinode |
c04a2ef3 | 1834 | * @n: The extent length (value/result) |
b3e47ca0 | 1835 | * |
c04a2ef3 | 1836 | * Add the bitmap buffer to the transaction. |
b3e47ca0 | 1837 | * Set the found bits to @new_state to change block's allocation state. |
b3e47ca0 | 1838 | */ |
c04a2ef3 | 1839 | static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
4a993fb1 | 1840 | unsigned int *n) |
b3e47ca0 | 1841 | { |
c04a2ef3 | 1842 | struct gfs2_rbm pos = { .rgd = rbm->rgd, }; |
b3e47ca0 | 1843 | const unsigned int elen = *n; |
c04a2ef3 SW |
1844 | u64 block; |
1845 | int ret; | |
b3e47ca0 | 1846 | |
c04a2ef3 SW |
1847 | *n = 1; |
1848 | block = gfs2_rbm_to_block(rbm); | |
1849 | gfs2_trans_add_bh(rbm->rgd->rd_gl, rbm->bi->bi_bh, 1); | |
3e6339dd | 1850 | gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
c04a2ef3 | 1851 | block++; |
60a0b8f9 | 1852 | while (*n < elen) { |
c04a2ef3 SW |
1853 | ret = gfs2_rbm_from_block(&pos, block); |
1854 | WARN_ON(ret); | |
1855 | if (gfs2_testbit(&pos) != GFS2_BLKST_FREE) | |
60a0b8f9 | 1856 | break; |
c04a2ef3 | 1857 | gfs2_trans_add_bh(pos.rgd->rd_gl, pos.bi->bi_bh, 1); |
3e6339dd | 1858 | gfs2_setbit(&pos, true, GFS2_BLKST_USED); |
60a0b8f9 | 1859 | (*n)++; |
c04a2ef3 | 1860 | block++; |
c8cdf479 | 1861 | } |
b3b94faa DT |
1862 | } |
1863 | ||
1864 | /** | |
1865 | * rgblk_free - Change alloc state of given block(s) | |
1866 | * @sdp: the filesystem | |
1867 | * @bstart: the start of a run of blocks to free | |
1868 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1869 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1870 | * | |
1871 | * Returns: Resource group containing the block(s) | |
1872 | */ | |
1873 | ||
cd915493 SW |
1874 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
1875 | u32 blen, unsigned char new_state) | |
b3b94faa | 1876 | { |
3b1d0b9d | 1877 | struct gfs2_rbm rbm; |
b3b94faa | 1878 | |
3b1d0b9d SW |
1879 | rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
1880 | if (!rbm.rgd) { | |
b3b94faa | 1881 | if (gfs2_consist(sdp)) |
382066da | 1882 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
1883 | return NULL; |
1884 | } | |
1885 | ||
b3b94faa | 1886 | while (blen--) { |
3b1d0b9d SW |
1887 | gfs2_rbm_from_block(&rbm, bstart); |
1888 | bstart++; | |
1889 | if (!rbm.bi->bi_clone) { | |
1890 | rbm.bi->bi_clone = kmalloc(rbm.bi->bi_bh->b_size, | |
1891 | GFP_NOFS | __GFP_NOFAIL); | |
1892 | memcpy(rbm.bi->bi_clone + rbm.bi->bi_offset, | |
1893 | rbm.bi->bi_bh->b_data + rbm.bi->bi_offset, | |
1894 | rbm.bi->bi_len); | |
b3b94faa | 1895 | } |
3b1d0b9d | 1896 | gfs2_trans_add_bh(rbm.rgd->rd_gl, rbm.bi->bi_bh, 1); |
3e6339dd | 1897 | gfs2_setbit(&rbm, false, new_state); |
b3b94faa DT |
1898 | } |
1899 | ||
3b1d0b9d | 1900 | return rbm.rgd; |
b3b94faa DT |
1901 | } |
1902 | ||
1903 | /** | |
09010978 SW |
1904 | * gfs2_rgrp_dump - print out an rgrp |
1905 | * @seq: The iterator | |
1906 | * @gl: The glock in question | |
1907 | * | |
1908 | */ | |
1909 | ||
1910 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
1911 | { | |
8e2e0047 BP |
1912 | struct gfs2_rgrpd *rgd = gl->gl_object; |
1913 | struct gfs2_blkreserv *trs; | |
1914 | const struct rb_node *n; | |
1915 | ||
09010978 SW |
1916 | if (rgd == NULL) |
1917 | return 0; | |
8e2e0047 | 1918 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u\n", |
09010978 | 1919 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 BP |
1920 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
1921 | rgd->rd_reserved); | |
1922 | spin_lock(&rgd->rd_rsspin); | |
1923 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { | |
1924 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1925 | dump_rs(seq, trs); | |
1926 | } | |
1927 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
1928 | return 0; |
1929 | } | |
1930 | ||
6050b9c7 SW |
1931 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
1932 | { | |
1933 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1934 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 1935 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
1936 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
1937 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
1938 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
1939 | } | |
1940 | ||
8e2e0047 | 1941 | /** |
5b924ae2 SW |
1942 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
1943 | * @ip: The inode we have just allocated blocks for | |
1944 | * @rbm: The start of the allocated blocks | |
1945 | * @len: The extent length | |
8e2e0047 | 1946 | * |
5b924ae2 SW |
1947 | * Adjusts a reservation after an allocation has taken place. If the |
1948 | * reservation does not match the allocation, or if it is now empty | |
1949 | * then it is removed. | |
8e2e0047 | 1950 | */ |
5b924ae2 SW |
1951 | |
1952 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
1953 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 BP |
1954 | { |
1955 | struct gfs2_blkreserv *rs = ip->i_res; | |
5b924ae2 SW |
1956 | struct gfs2_rgrpd *rgd = rbm->rgd; |
1957 | unsigned rlen; | |
1958 | u64 block; | |
1959 | int ret; | |
8e2e0047 | 1960 | |
5b924ae2 SW |
1961 | spin_lock(&rgd->rd_rsspin); |
1962 | if (gfs2_rs_active(rs)) { | |
1963 | if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) { | |
1964 | block = gfs2_rbm_to_block(rbm); | |
1965 | ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len); | |
1966 | rlen = min(rs->rs_free, len); | |
1967 | rs->rs_free -= rlen; | |
1968 | rgd->rd_reserved -= rlen; | |
1969 | trace_gfs2_rs(ip, rs, TRACE_RS_CLAIM); | |
1970 | if (rs->rs_free && !ret) | |
1971 | goto out; | |
1972 | } | |
1973 | __rs_deltree(ip, rs); | |
8e2e0047 | 1974 | } |
5b924ae2 SW |
1975 | out: |
1976 | spin_unlock(&rgd->rd_rsspin); | |
8e2e0047 BP |
1977 | } |
1978 | ||
09010978 | 1979 | /** |
6e87ed0f | 1980 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 1981 | * @ip: the inode to allocate the block for |
09010978 | 1982 | * @bn: Used to return the starting block number |
8e2e0047 | 1983 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 1984 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 1985 | * @generation: the generation number of the inode |
b3b94faa | 1986 | * |
09010978 | 1987 | * Returns: 0 or error |
b3b94faa DT |
1988 | */ |
1989 | ||
6a8099ed | 1990 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 1991 | bool dinode, u64 *generation) |
b3b94faa | 1992 | { |
feaa7bba | 1993 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 1994 | struct buffer_head *dibh; |
4a993fb1 | 1995 | struct gfs2_rbm rbm = { .rgd = ip->i_rgd, }; |
6a8099ed | 1996 | unsigned int ndata; |
5b924ae2 | 1997 | u64 goal; |
3c5d785a | 1998 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 1999 | int error; |
b3b94faa | 2000 | |
5b924ae2 SW |
2001 | if (gfs2_rs_active(ip->i_res)) |
2002 | goal = gfs2_rbm_to_block(&ip->i_res->rs_rbm); | |
2003 | else if (!dinode && rgrp_contains_block(rbm.rgd, ip->i_goal)) | |
2004 | goal = ip->i_goal; | |
62e252ee | 2005 | else |
5b924ae2 | 2006 | goal = rbm.rgd->rd_last_alloc + rbm.rgd->rd_data0; |
62e252ee | 2007 | |
8d8b752a BP |
2008 | if ((goal < rbm.rgd->rd_data0) || |
2009 | (goal >= rbm.rgd->rd_data0 + rbm.rgd->rd_data)) | |
2010 | rbm.rgd = gfs2_blk2rgrpd(sdp, goal, 1); | |
2011 | ||
5b924ae2 SW |
2012 | gfs2_rbm_from_block(&rbm, goal); |
2013 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, ip, false); | |
62e252ee | 2014 | |
137834a6 SW |
2015 | if (error == -ENOSPC) { |
2016 | gfs2_rbm_from_block(&rbm, goal); | |
2017 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, false); | |
2018 | } | |
2019 | ||
62e252ee | 2020 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
5b924ae2 SW |
2021 | if (error) { |
2022 | fs_warn(sdp, "error=%d, nblocks=%u, full=%d\n", error, *nblocks, | |
2023 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags)); | |
62e252ee | 2024 | goto rgrp_error; |
8e2e0047 | 2025 | } |
62e252ee | 2026 | |
c04a2ef3 SW |
2027 | gfs2_alloc_extent(&rbm, dinode, nblocks); |
2028 | block = gfs2_rbm_to_block(&rbm); | |
5b924ae2 SW |
2029 | if (gfs2_rs_active(ip->i_res)) |
2030 | gfs2_adjust_reservation(ip, &rbm, *nblocks); | |
6a8099ed SW |
2031 | ndata = *nblocks; |
2032 | if (dinode) | |
2033 | ndata--; | |
b3e47ca0 | 2034 | |
3c5d785a | 2035 | if (!dinode) { |
6a8099ed | 2036 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
2037 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2038 | if (error == 0) { | |
2039 | struct gfs2_dinode *di = | |
2040 | (struct gfs2_dinode *)dibh->b_data; | |
2041 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
2042 | di->di_goal_meta = di->di_goal_data = | |
2043 | cpu_to_be64(ip->i_goal); | |
2044 | brelse(dibh); | |
2045 | } | |
d9ba7615 | 2046 | } |
4a993fb1 | 2047 | if (rbm.rgd->rd_free < *nblocks) { |
8e2e0047 | 2048 | printk(KERN_WARNING "nblocks=%u\n", *nblocks); |
09010978 | 2049 | goto rgrp_error; |
8e2e0047 | 2050 | } |
09010978 | 2051 | |
4a993fb1 | 2052 | rbm.rgd->rd_free -= *nblocks; |
3c5d785a | 2053 | if (dinode) { |
4a993fb1 SW |
2054 | rbm.rgd->rd_dinodes++; |
2055 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2056 | if (*generation == 0) |
4a993fb1 | 2057 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2058 | } |
b3b94faa | 2059 | |
4a993fb1 SW |
2060 | gfs2_trans_add_bh(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh, 1); |
2061 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
2062 | gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
b3b94faa | 2063 | |
6a8099ed | 2064 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a BP |
2065 | if (dinode) |
2066 | gfs2_trans_add_unrevoke(sdp, block, 1); | |
6a8099ed SW |
2067 | |
2068 | /* | |
2069 | * This needs reviewing to see why we cannot do the quota change | |
2070 | * at this point in the dinode case. | |
2071 | */ | |
2072 | if (ndata) | |
2073 | gfs2_quota_change(ip, ndata, ip->i_inode.i_uid, | |
3c5d785a | 2074 | ip->i_inode.i_gid); |
b3b94faa | 2075 | |
4a993fb1 SW |
2076 | rbm.rgd->rd_free_clone -= *nblocks; |
2077 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, | |
6e87ed0f | 2078 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2079 | *bn = block; |
2080 | return 0; | |
2081 | ||
2082 | rgrp_error: | |
4a993fb1 | 2083 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2084 | return -EIO; |
b3b94faa DT |
2085 | } |
2086 | ||
2087 | /** | |
46fcb2ed | 2088 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
2089 | * @ip: the inode these blocks are being freed from |
2090 | * @bstart: first block of a run of contiguous blocks | |
2091 | * @blen: the length of the block run | |
46fcb2ed | 2092 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2093 | * |
2094 | */ | |
2095 | ||
46fcb2ed | 2096 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 2097 | { |
feaa7bba | 2098 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2099 | struct gfs2_rgrpd *rgd; |
2100 | ||
2101 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
2102 | if (!rgd) | |
2103 | return; | |
41db1ab9 | 2104 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2105 | rgd->rd_free += blen; |
66fc061b | 2106 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
d4e9c4c3 | 2107 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 2108 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 | 2109 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 2110 | |
6d3117b4 | 2111 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 2112 | if (meta || ip->i_depth) |
6d3117b4 | 2113 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 2114 | } |
b3b94faa | 2115 | |
4c16c36a BP |
2116 | /** |
2117 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2118 | * @ip: the inode these blocks are being freed from | |
2119 | * @bstart: first block of a run of contiguous blocks | |
2120 | * @blen: the length of the block run | |
2121 | * | |
2122 | */ | |
2123 | ||
2124 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
2125 | { | |
2126 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2127 | ||
46fcb2ed | 2128 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 2129 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2130 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2131 | } |
2132 | ||
feaa7bba SW |
2133 | void gfs2_unlink_di(struct inode *inode) |
2134 | { | |
2135 | struct gfs2_inode *ip = GFS2_I(inode); | |
2136 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2137 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2138 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
2139 | |
2140 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
2141 | if (!rgd) | |
2142 | return; | |
41db1ab9 | 2143 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
feaa7bba | 2144 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 2145 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2146 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2147 | update_rgrp_lvb_unlinked(rgd, 1); | |
feaa7bba SW |
2148 | } |
2149 | ||
cd915493 | 2150 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
2151 | { |
2152 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2153 | struct gfs2_rgrpd *tmp_rgd; | |
2154 | ||
2155 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
2156 | if (!tmp_rgd) | |
2157 | return; | |
2158 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
2159 | ||
73f74948 | 2160 | if (!rgd->rd_dinodes) |
b3b94faa | 2161 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2162 | rgd->rd_dinodes--; |
cfc8b549 | 2163 | rgd->rd_free++; |
b3b94faa | 2164 | |
d4e9c4c3 | 2165 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 2166 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2167 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2168 | update_rgrp_lvb_unlinked(rgd, -1); | |
b3b94faa DT |
2169 | |
2170 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
2171 | } |
2172 | ||
b3b94faa DT |
2173 | |
2174 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
2175 | { | |
dbb7cae2 | 2176 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
41db1ab9 | 2177 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2178 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 2179 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2180 | } |
2181 | ||
acf7e244 SW |
2182 | /** |
2183 | * gfs2_check_blk_type - Check the type of a block | |
2184 | * @sdp: The superblock | |
2185 | * @no_addr: The block number to check | |
2186 | * @type: The block type we are looking for | |
2187 | * | |
2188 | * Returns: 0 if the block type matches the expected type | |
2189 | * -ESTALE if it doesn't match | |
2190 | * or -ve errno if something went wrong while checking | |
2191 | */ | |
2192 | ||
2193 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2194 | { | |
2195 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2196 | struct gfs2_holder rgd_gh; |
58884c4d | 2197 | int error = -EINVAL; |
acf7e244 | 2198 | |
66fc061b | 2199 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2200 | if (!rgd) |
8339ee54 | 2201 | goto fail; |
acf7e244 SW |
2202 | |
2203 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2204 | if (error) | |
8339ee54 | 2205 | goto fail; |
acf7e244 SW |
2206 | |
2207 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
2208 | error = -ESTALE; | |
2209 | ||
2210 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
2211 | fail: |
2212 | return error; | |
2213 | } | |
2214 | ||
b3b94faa DT |
2215 | /** |
2216 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2217 | * @ip: the inode |
b3b94faa DT |
2218 | * @rlist: the list of resource groups |
2219 | * @block: the block | |
2220 | * | |
2221 | * Figure out what RG a block belongs to and add that RG to the list | |
2222 | * | |
2223 | * FIXME: Don't use NOFAIL | |
2224 | * | |
2225 | */ | |
2226 | ||
70b0c365 | 2227 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2228 | u64 block) |
b3b94faa | 2229 | { |
70b0c365 | 2230 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2231 | struct gfs2_rgrpd *rgd; |
2232 | struct gfs2_rgrpd **tmp; | |
2233 | unsigned int new_space; | |
2234 | unsigned int x; | |
2235 | ||
2236 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2237 | return; | |
2238 | ||
70b0c365 SW |
2239 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
2240 | rgd = ip->i_rgd; | |
2241 | else | |
66fc061b | 2242 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 2243 | if (!rgd) { |
70b0c365 | 2244 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
2245 | return; |
2246 | } | |
70b0c365 | 2247 | ip->i_rgd = rgd; |
b3b94faa DT |
2248 | |
2249 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2250 | if (rlist->rl_rgd[x] == rgd) | |
2251 | return; | |
2252 | ||
2253 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2254 | new_space = rlist->rl_space + 10; | |
2255 | ||
2256 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2257 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2258 | |
2259 | if (rlist->rl_rgd) { | |
2260 | memcpy(tmp, rlist->rl_rgd, | |
2261 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2262 | kfree(rlist->rl_rgd); | |
2263 | } | |
2264 | ||
2265 | rlist->rl_space = new_space; | |
2266 | rlist->rl_rgd = tmp; | |
2267 | } | |
2268 | ||
2269 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2270 | } | |
2271 | ||
2272 | /** | |
2273 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2274 | * and initialize an array of glock holders for them | |
2275 | * @rlist: the list of resource groups | |
2276 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
2277 | * |
2278 | * FIXME: Don't use NOFAIL | |
2279 | * | |
2280 | */ | |
2281 | ||
fe6c991c | 2282 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
2283 | { |
2284 | unsigned int x; | |
2285 | ||
2286 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 2287 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2288 | for (x = 0; x < rlist->rl_rgrps; x++) |
2289 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 2290 | state, 0, |
b3b94faa DT |
2291 | &rlist->rl_ghs[x]); |
2292 | } | |
2293 | ||
2294 | /** | |
2295 | * gfs2_rlist_free - free a resource group list | |
2296 | * @list: the list of resource groups | |
2297 | * | |
2298 | */ | |
2299 | ||
2300 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2301 | { | |
2302 | unsigned int x; | |
2303 | ||
2304 | kfree(rlist->rl_rgd); | |
2305 | ||
2306 | if (rlist->rl_ghs) { | |
2307 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2308 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2309 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2310 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2311 | } |
2312 | } | |
2313 |