Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
fe6c991c | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
b3b94faa DT |
10 | #include <linux/slab.h> |
11 | #include <linux/spinlock.h> | |
12 | #include <linux/completion.h> | |
13 | #include <linux/buffer_head.h> | |
f42faf4f | 14 | #include <linux/fs.h> |
5c676f6d | 15 | #include <linux/gfs2_ondisk.h> |
1f466a47 | 16 | #include <linux/prefetch.h> |
f15ab561 | 17 | #include <linux/blkdev.h> |
7c9ca621 | 18 | #include <linux/rbtree.h> |
b3b94faa DT |
19 | |
20 | #include "gfs2.h" | |
5c676f6d | 21 | #include "incore.h" |
b3b94faa DT |
22 | #include "glock.h" |
23 | #include "glops.h" | |
b3b94faa DT |
24 | #include "lops.h" |
25 | #include "meta_io.h" | |
26 | #include "quota.h" | |
27 | #include "rgrp.h" | |
28 | #include "super.h" | |
29 | #include "trans.h" | |
5c676f6d | 30 | #include "util.h" |
172e045a | 31 | #include "log.h" |
c8cdf479 | 32 | #include "inode.h" |
63997775 | 33 | #include "trace_gfs2.h" |
b3b94faa | 34 | |
2c1e52aa | 35 | #define BFITNOENT ((u32)~0) |
6760bdcd | 36 | #define NO_BLOCK ((u64)~0) |
88c8ab1f | 37 | |
1f466a47 BP |
38 | #if BITS_PER_LONG == 32 |
39 | #define LBITMASK (0x55555555UL) | |
40 | #define LBITSKIP55 (0x55555555UL) | |
41 | #define LBITSKIP00 (0x00000000UL) | |
42 | #else | |
43 | #define LBITMASK (0x5555555555555555UL) | |
44 | #define LBITSKIP55 (0x5555555555555555UL) | |
45 | #define LBITSKIP00 (0x0000000000000000UL) | |
46 | #endif | |
47 | ||
88c8ab1f SW |
48 | /* |
49 | * These routines are used by the resource group routines (rgrp.c) | |
50 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
51 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
52 | * | |
53 | * 0 = Free | |
54 | * 1 = Used (not metadata) | |
55 | * 2 = Unlinked (still in use) inode | |
56 | * 3 = Used (metadata) | |
88c8ab1f SW |
57 | */ |
58 | ||
59 | static const char valid_change[16] = { | |
60 | /* current */ | |
feaa7bba | 61 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 62 | /* e */ 1, 0, 0, 0, |
feaa7bba | 63 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
64 | 1, 0, 0, 0 |
65 | }; | |
66 | ||
c8cdf479 | 67 | static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, |
6a8099ed | 68 | unsigned char old_state, |
b3e47ca0 | 69 | struct gfs2_bitmap **rbi); |
c8cdf479 | 70 | |
88c8ab1f SW |
71 | /** |
72 | * gfs2_setbit - Set a bit in the bitmaps | |
29c578f5 BP |
73 | * @rgd: the resource group descriptor |
74 | * @buf1: the primary buffer that holds the bitmaps | |
75 | * @buf2: the clone buffer that holds the bitmaps | |
76 | * @bi: the bitmap structure | |
88c8ab1f SW |
77 | * @block: the block to set |
78 | * @new_state: the new state of the block | |
79 | * | |
80 | */ | |
81 | ||
b45e41d7 | 82 | static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1, |
29c578f5 BP |
83 | unsigned char *buf2, struct gfs2_bitmap *bi, |
84 | u32 block, unsigned char new_state) | |
88c8ab1f | 85 | { |
b45e41d7 | 86 | unsigned char *byte1, *byte2, *end, cur_state; |
95c8e17f | 87 | unsigned int buflen = bi->bi_len; |
b45e41d7 | 88 | const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 89 | |
29c578f5 BP |
90 | byte1 = buf1 + bi->bi_offset + (block / GFS2_NBBY); |
91 | end = buf1 + bi->bi_offset + buflen; | |
88c8ab1f | 92 | |
b45e41d7 | 93 | BUG_ON(byte1 >= end); |
88c8ab1f | 94 | |
b45e41d7 | 95 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 96 | |
b45e41d7 | 97 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
95c8e17f BP |
98 | printk(KERN_WARNING "GFS2: buf_blk = 0x%llx old_state=%d, " |
99 | "new_state=%d\n", | |
100 | (unsigned long long)block, cur_state, new_state); | |
101 | printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%lx\n", | |
102 | (unsigned long long)rgd->rd_addr, | |
103 | (unsigned long)bi->bi_start); | |
104 | printk(KERN_WARNING "GFS2: bi_offset=0x%lx bi_len=0x%lx\n", | |
105 | (unsigned long)bi->bi_offset, | |
106 | (unsigned long)bi->bi_len); | |
107 | dump_stack(); | |
88c8ab1f | 108 | gfs2_consist_rgrpd(rgd); |
b45e41d7 SW |
109 | return; |
110 | } | |
111 | *byte1 ^= (cur_state ^ new_state) << bit; | |
112 | ||
113 | if (buf2) { | |
29c578f5 | 114 | byte2 = buf2 + bi->bi_offset + (block / GFS2_NBBY); |
b45e41d7 SW |
115 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
116 | *byte2 ^= (cur_state ^ new_state) << bit; | |
117 | } | |
88c8ab1f SW |
118 | } |
119 | ||
120 | /** | |
121 | * gfs2_testbit - test a bit in the bitmaps | |
886b1416 | 122 | * @rgd: the resource group descriptor |
88c8ab1f SW |
123 | * @buffer: the buffer that holds the bitmaps |
124 | * @buflen: the length (in bytes) of the buffer | |
125 | * @block: the block to read | |
126 | * | |
127 | */ | |
128 | ||
b45e41d7 SW |
129 | static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd, |
130 | const unsigned char *buffer, | |
131 | unsigned int buflen, u32 block) | |
88c8ab1f | 132 | { |
b45e41d7 SW |
133 | const unsigned char *byte, *end; |
134 | unsigned char cur_state; | |
88c8ab1f SW |
135 | unsigned int bit; |
136 | ||
137 | byte = buffer + (block / GFS2_NBBY); | |
138 | bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; | |
139 | end = buffer + buflen; | |
140 | ||
141 | gfs2_assert(rgd->rd_sbd, byte < end); | |
142 | ||
143 | cur_state = (*byte >> bit) & GFS2_BIT_MASK; | |
144 | ||
145 | return cur_state; | |
146 | } | |
147 | ||
223b2b88 SW |
148 | /** |
149 | * gfs2_bit_search | |
150 | * @ptr: Pointer to bitmap data | |
151 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
152 | * @state: The state we are searching for | |
153 | * | |
154 | * We xor the bitmap data with a patter which is the bitwise opposite | |
155 | * of what we are looking for, this gives rise to a pattern of ones | |
156 | * wherever there is a match. Since we have two bits per entry, we | |
157 | * take this pattern, shift it down by one place and then and it with | |
158 | * the original. All the even bit positions (0,2,4, etc) then represent | |
159 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
160 | * odd bit positions. | |
161 | * | |
162 | * This allows searching of a whole u64 at once (32 blocks) with a | |
163 | * single test (on 64 bit arches). | |
164 | */ | |
165 | ||
166 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
167 | { | |
168 | u64 tmp; | |
169 | static const u64 search[] = { | |
075ac448 HE |
170 | [0] = 0xffffffffffffffffULL, |
171 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
172 | [2] = 0x5555555555555555ULL, | |
173 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
174 | }; |
175 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
176 | tmp &= (tmp >> 1); | |
177 | tmp &= mask; | |
178 | return tmp; | |
179 | } | |
180 | ||
88c8ab1f SW |
181 | /** |
182 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
183 | * a block in a given allocation state. | |
886b1416 | 184 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 185 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 186 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 187 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
188 | * |
189 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
190 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
191 | * beginning of a bitmap block buffer, skipping any header structures, but |
192 | * headers are always a multiple of 64 bits long so that the buffer is | |
193 | * always aligned to a 64 bit boundary. | |
194 | * | |
195 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 196 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 197 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
198 | * |
199 | * Return: the block number (bitmap buffer scope) that was found | |
200 | */ | |
201 | ||
02ab1721 HE |
202 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
203 | u32 goal, u8 state) | |
88c8ab1f | 204 | { |
223b2b88 SW |
205 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
206 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
207 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
208 | u64 tmp; | |
075ac448 | 209 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
210 | u32 bit; |
211 | ||
212 | BUG_ON(state > 3); | |
213 | ||
214 | /* Mask off bits we don't care about at the start of the search */ | |
215 | mask <<= spoint; | |
216 | tmp = gfs2_bit_search(ptr, mask, state); | |
217 | ptr++; | |
218 | while(tmp == 0 && ptr < end) { | |
075ac448 | 219 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 220 | ptr++; |
1f466a47 | 221 | } |
223b2b88 SW |
222 | /* Mask off any bits which are more than len bytes from the start */ |
223 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
224 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
225 | /* Didn't find anything, so return */ | |
226 | if (tmp == 0) | |
227 | return BFITNOENT; | |
228 | ptr--; | |
d8bd504a | 229 | bit = __ffs64(tmp); |
223b2b88 SW |
230 | bit /= 2; /* two bits per entry in the bitmap */ |
231 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
232 | } |
233 | ||
234 | /** | |
235 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 236 | * @rgd: the resource group descriptor |
88c8ab1f SW |
237 | * @buffer: the buffer that holds the bitmaps |
238 | * @buflen: the length (in bytes) of the buffer | |
239 | * @state: the state of the block we're looking for | |
240 | * | |
241 | * Returns: The number of bits | |
242 | */ | |
243 | ||
110acf38 SW |
244 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
245 | unsigned int buflen, u8 state) | |
88c8ab1f | 246 | { |
110acf38 SW |
247 | const u8 *byte = buffer; |
248 | const u8 *end = buffer + buflen; | |
249 | const u8 state1 = state << 2; | |
250 | const u8 state2 = state << 4; | |
251 | const u8 state3 = state << 6; | |
cd915493 | 252 | u32 count = 0; |
88c8ab1f SW |
253 | |
254 | for (; byte < end; byte++) { | |
255 | if (((*byte) & 0x03) == state) | |
256 | count++; | |
257 | if (((*byte) & 0x0C) == state1) | |
258 | count++; | |
259 | if (((*byte) & 0x30) == state2) | |
260 | count++; | |
261 | if (((*byte) & 0xC0) == state3) | |
262 | count++; | |
263 | } | |
264 | ||
265 | return count; | |
266 | } | |
267 | ||
b3b94faa DT |
268 | /** |
269 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
270 | * @rgd: the rgrp |
271 | * | |
272 | */ | |
273 | ||
274 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
275 | { | |
276 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
277 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 278 | u32 length = rgd->rd_length; |
cd915493 | 279 | u32 count[4], tmp; |
b3b94faa DT |
280 | int buf, x; |
281 | ||
cd915493 | 282 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
283 | |
284 | /* Count # blocks in each of 4 possible allocation states */ | |
285 | for (buf = 0; buf < length; buf++) { | |
286 | bi = rgd->rd_bits + buf; | |
287 | for (x = 0; x < 4; x++) | |
288 | count[x] += gfs2_bitcount(rgd, | |
289 | bi->bi_bh->b_data + | |
290 | bi->bi_offset, | |
291 | bi->bi_len, x); | |
292 | } | |
293 | ||
cfc8b549 | 294 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
295 | if (gfs2_consist_rgrpd(rgd)) |
296 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 297 | count[0], rgd->rd_free); |
b3b94faa DT |
298 | return; |
299 | } | |
300 | ||
73f74948 | 301 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 302 | if (count[1] != tmp) { |
b3b94faa DT |
303 | if (gfs2_consist_rgrpd(rgd)) |
304 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
305 | count[1], tmp); | |
306 | return; | |
307 | } | |
308 | ||
6b946170 | 309 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 310 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 311 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 312 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
313 | return; |
314 | } | |
b3b94faa DT |
315 | } |
316 | ||
bb8d8a6f | 317 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 318 | { |
bb8d8a6f SW |
319 | u64 first = rgd->rd_data0; |
320 | u64 last = first + rgd->rd_data; | |
16910427 | 321 | return first <= block && block < last; |
b3b94faa DT |
322 | } |
323 | ||
324 | /** | |
325 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
326 | * @sdp: The GFS2 superblock | |
886b1416 BP |
327 | * @blk: The data block number |
328 | * @exact: True if this needs to be an exact match | |
b3b94faa DT |
329 | * |
330 | * Returns: The resource group, or NULL if not found | |
331 | */ | |
332 | ||
66fc061b | 333 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 334 | { |
66fc061b | 335 | struct rb_node *n, *next; |
f75bbfb4 | 336 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
337 | |
338 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
339 | n = sdp->sd_rindex_tree.rb_node; |
340 | while (n) { | |
341 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
342 | next = NULL; | |
7c9ca621 | 343 | if (blk < cur->rd_addr) |
66fc061b | 344 | next = n->rb_left; |
f75bbfb4 | 345 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
346 | next = n->rb_right; |
347 | if (next == NULL) { | |
b3b94faa | 348 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
349 | if (exact) { |
350 | if (blk < cur->rd_addr) | |
351 | return NULL; | |
352 | if (blk >= cur->rd_data0 + cur->rd_data) | |
353 | return NULL; | |
354 | } | |
7c9ca621 | 355 | return cur; |
b3b94faa | 356 | } |
66fc061b | 357 | n = next; |
b3b94faa | 358 | } |
b3b94faa DT |
359 | spin_unlock(&sdp->sd_rindex_spin); |
360 | ||
361 | return NULL; | |
362 | } | |
363 | ||
364 | /** | |
365 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
366 | * @sdp: The GFS2 superblock | |
367 | * | |
368 | * Returns: The first rgrp in the filesystem | |
369 | */ | |
370 | ||
371 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
372 | { | |
7c9ca621 BP |
373 | const struct rb_node *n; |
374 | struct gfs2_rgrpd *rgd; | |
375 | ||
8339ee54 | 376 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
377 | n = rb_first(&sdp->sd_rindex_tree); |
378 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 379 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
380 | |
381 | return rgd; | |
b3b94faa DT |
382 | } |
383 | ||
384 | /** | |
385 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 386 | * @rgd: the resource group descriptor |
b3b94faa DT |
387 | * |
388 | * Returns: The next rgrp | |
389 | */ | |
390 | ||
391 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
392 | { | |
7c9ca621 BP |
393 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
394 | const struct rb_node *n; | |
395 | ||
396 | spin_lock(&sdp->sd_rindex_spin); | |
397 | n = rb_next(&rgd->rd_node); | |
398 | if (n == NULL) | |
399 | n = rb_first(&sdp->sd_rindex_tree); | |
400 | ||
401 | if (unlikely(&rgd->rd_node == n)) { | |
402 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 403 | return NULL; |
7c9ca621 BP |
404 | } |
405 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
406 | spin_unlock(&sdp->sd_rindex_spin); | |
407 | return rgd; | |
b3b94faa DT |
408 | } |
409 | ||
8339ee54 SW |
410 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
411 | { | |
412 | int x; | |
413 | ||
414 | for (x = 0; x < rgd->rd_length; x++) { | |
415 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
416 | kfree(bi->bi_clone); | |
417 | bi->bi_clone = NULL; | |
418 | } | |
419 | } | |
420 | ||
421 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) | |
b3b94faa | 422 | { |
7c9ca621 | 423 | struct rb_node *n; |
b3b94faa DT |
424 | struct gfs2_rgrpd *rgd; |
425 | struct gfs2_glock *gl; | |
426 | ||
7c9ca621 BP |
427 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
428 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
429 | gl = rgd->rd_gl; |
430 | ||
7c9ca621 | 431 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
432 | |
433 | if (gl) { | |
8339ee54 | 434 | spin_lock(&gl->gl_spin); |
5c676f6d | 435 | gl->gl_object = NULL; |
8339ee54 | 436 | spin_unlock(&gl->gl_spin); |
29687a2a | 437 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
438 | gfs2_glock_put(gl); |
439 | } | |
440 | ||
8339ee54 | 441 | gfs2_free_clones(rgd); |
b3b94faa | 442 | kfree(rgd->rd_bits); |
6bdd9be6 | 443 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
444 | } |
445 | } | |
446 | ||
bb8d8a6f SW |
447 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
448 | { | |
449 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
450 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
451 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
452 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
453 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
454 | } | |
455 | ||
b3b94faa DT |
456 | /** |
457 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
458 | * @rgd: The resource group descriptor | |
459 | * | |
460 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
461 | * | |
462 | * Returns: errno | |
463 | */ | |
464 | ||
465 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
466 | { | |
467 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
468 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 469 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 470 | u32 bytes_left, bytes; |
b3b94faa DT |
471 | int x; |
472 | ||
feaa7bba SW |
473 | if (!length) |
474 | return -EINVAL; | |
475 | ||
dd894be8 | 476 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
477 | if (!rgd->rd_bits) |
478 | return -ENOMEM; | |
479 | ||
bb8d8a6f | 480 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
481 | |
482 | for (x = 0; x < length; x++) { | |
483 | bi = rgd->rd_bits + x; | |
484 | ||
60a0b8f9 | 485 | bi->bi_flags = 0; |
b3b94faa DT |
486 | /* small rgrp; bitmap stored completely in header block */ |
487 | if (length == 1) { | |
488 | bytes = bytes_left; | |
489 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
490 | bi->bi_start = 0; | |
491 | bi->bi_len = bytes; | |
492 | /* header block */ | |
493 | } else if (x == 0) { | |
494 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
495 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
496 | bi->bi_start = 0; | |
497 | bi->bi_len = bytes; | |
498 | /* last block */ | |
499 | } else if (x + 1 == length) { | |
500 | bytes = bytes_left; | |
501 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 502 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
503 | bi->bi_len = bytes; |
504 | /* other blocks */ | |
505 | } else { | |
568f4c96 SW |
506 | bytes = sdp->sd_sb.sb_bsize - |
507 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 508 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 509 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
510 | bi->bi_len = bytes; |
511 | } | |
512 | ||
513 | bytes_left -= bytes; | |
514 | } | |
515 | ||
516 | if (bytes_left) { | |
517 | gfs2_consist_rgrpd(rgd); | |
518 | return -EIO; | |
519 | } | |
520 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 521 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 522 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 523 | gfs2_rindex_print(rgd); |
b3b94faa DT |
524 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
525 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
526 | } | |
527 | return -EIO; | |
528 | } | |
529 | ||
530 | return 0; | |
531 | } | |
532 | ||
7ae8fa84 RP |
533 | /** |
534 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 535 | * @sdp: the filesystem |
7ae8fa84 RP |
536 | * |
537 | */ | |
538 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
539 | { | |
540 | u64 total_data = 0; | |
541 | struct inode *inode = sdp->sd_rindex; | |
542 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 543 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
544 | int error, rgrps; |
545 | ||
7ae8fa84 RP |
546 | for (rgrps = 0;; rgrps++) { |
547 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
548 | ||
bcd7278d | 549 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 550 | break; |
4306629e | 551 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
552 | sizeof(struct gfs2_rindex)); |
553 | if (error != sizeof(struct gfs2_rindex)) | |
554 | break; | |
bb8d8a6f | 555 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 556 | } |
7ae8fa84 RP |
557 | return total_data; |
558 | } | |
559 | ||
6aad1c3d | 560 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
561 | { |
562 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
563 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
564 | ||
565 | /* Figure out where to put new node */ | |
566 | while (*newn) { | |
567 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
568 | rd_node); | |
569 | ||
570 | parent = *newn; | |
571 | if (rgd->rd_addr < cur->rd_addr) | |
572 | newn = &((*newn)->rb_left); | |
573 | else if (rgd->rd_addr > cur->rd_addr) | |
574 | newn = &((*newn)->rb_right); | |
575 | else | |
6aad1c3d | 576 | return -EEXIST; |
7c9ca621 BP |
577 | } |
578 | ||
579 | rb_link_node(&rgd->rd_node, parent, newn); | |
580 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
581 | sdp->sd_rgrps++; |
582 | return 0; | |
7c9ca621 BP |
583 | } |
584 | ||
b3b94faa | 585 | /** |
6c53267f | 586 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 587 | * @ip: Pointer to the rindex inode |
b3b94faa | 588 | * |
8339ee54 | 589 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
590 | */ |
591 | ||
4306629e | 592 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
593 | { |
594 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
595 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
8339ee54 | 596 | struct gfs2_rindex buf; |
6c53267f RP |
597 | int error; |
598 | struct gfs2_rgrpd *rgd; | |
599 | ||
8339ee54 SW |
600 | if (pos >= i_size_read(&ip->i_inode)) |
601 | return 1; | |
602 | ||
4306629e | 603 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 604 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
605 | |
606 | if (error != sizeof(struct gfs2_rindex)) | |
607 | return (error == 0) ? 1 : error; | |
6c53267f | 608 | |
6bdd9be6 | 609 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
610 | error = -ENOMEM; |
611 | if (!rgd) | |
612 | return error; | |
613 | ||
6c53267f | 614 | rgd->rd_sbd = sdp; |
8339ee54 SW |
615 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
616 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
617 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
618 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
619 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
7c9ca621 | 620 | |
6c53267f RP |
621 | error = compute_bitstructs(rgd); |
622 | if (error) | |
8339ee54 | 623 | goto fail; |
6c53267f | 624 | |
bb8d8a6f | 625 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
626 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
627 | if (error) | |
8339ee54 | 628 | goto fail; |
6c53267f RP |
629 | |
630 | rgd->rd_gl->gl_object = rgd; | |
cf45b752 | 631 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
7c9ca621 BP |
632 | if (rgd->rd_data > sdp->sd_max_rg_data) |
633 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 634 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 635 | error = rgd_insert(rgd); |
8339ee54 | 636 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
637 | if (!error) |
638 | return 0; | |
639 | ||
640 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 641 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
642 | |
643 | fail: | |
644 | kfree(rgd->rd_bits); | |
645 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
646 | return error; |
647 | } | |
648 | ||
649 | /** | |
650 | * gfs2_ri_update - Pull in a new resource index from the disk | |
651 | * @ip: pointer to the rindex inode | |
652 | * | |
b3b94faa DT |
653 | * Returns: 0 on successful update, error code otherwise |
654 | */ | |
655 | ||
8339ee54 | 656 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 657 | { |
feaa7bba | 658 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
659 | int error; |
660 | ||
8339ee54 | 661 | do { |
4306629e | 662 | error = read_rindex_entry(ip); |
8339ee54 SW |
663 | } while (error == 0); |
664 | ||
665 | if (error < 0) | |
666 | return error; | |
b3b94faa | 667 | |
cf45b752 | 668 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
669 | return 0; |
670 | } | |
b3b94faa | 671 | |
b3b94faa | 672 | /** |
8339ee54 | 673 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 674 | * @sdp: The GFS2 superblock |
b3b94faa DT |
675 | * |
676 | * We grab a lock on the rindex inode to make sure that it doesn't | |
677 | * change whilst we are performing an operation. We keep this lock | |
678 | * for quite long periods of time compared to other locks. This | |
679 | * doesn't matter, since it is shared and it is very, very rarely | |
680 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
681 | * | |
682 | * This makes sure that we're using the latest copy of the resource index | |
683 | * special file, which might have been updated if someone expanded the | |
684 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
685 | * | |
8339ee54 | 686 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
687 | */ |
688 | ||
8339ee54 | 689 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 690 | { |
feaa7bba | 691 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 692 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
693 | struct gfs2_holder ri_gh; |
694 | int error = 0; | |
a365fbf3 | 695 | int unlock_required = 0; |
b3b94faa DT |
696 | |
697 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 698 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
699 | if (!gfs2_glock_is_locked_by_me(gl)) { |
700 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
701 | if (error) | |
6aad1c3d | 702 | return error; |
a365fbf3 SW |
703 | unlock_required = 1; |
704 | } | |
8339ee54 | 705 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 706 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
707 | if (unlock_required) |
708 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
709 | } |
710 | ||
711 | return error; | |
712 | } | |
713 | ||
42d52e38 | 714 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
715 | { |
716 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 717 | u32 rg_flags; |
bb8d8a6f | 718 | |
42d52e38 | 719 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 720 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
721 | rgd->rd_flags &= GFS2_RDF_MASK; |
722 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 723 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 724 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 725 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
726 | } |
727 | ||
42d52e38 | 728 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
729 | { |
730 | struct gfs2_rgrp *str = buf; | |
731 | ||
09010978 | 732 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 733 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 734 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 735 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 736 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
737 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
738 | } | |
739 | ||
b3b94faa | 740 | /** |
7c9ca621 | 741 | * gfs2_rgrp_go_lock - Read in a RG's header and bitmaps |
886b1416 | 742 | * @gh: The glock holder for the resource group |
b3b94faa DT |
743 | * |
744 | * Read in all of a Resource Group's header and bitmap blocks. | |
745 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
746 | * | |
747 | * Returns: errno | |
748 | */ | |
749 | ||
7c9ca621 | 750 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) |
b3b94faa | 751 | { |
7c9ca621 | 752 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
b3b94faa DT |
753 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
754 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 755 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
756 | struct gfs2_bitmap *bi; |
757 | unsigned int x, y; | |
758 | int error; | |
759 | ||
b3b94faa DT |
760 | for (x = 0; x < length; x++) { |
761 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 762 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
763 | if (error) |
764 | goto fail; | |
765 | } | |
766 | ||
767 | for (y = length; y--;) { | |
768 | bi = rgd->rd_bits + y; | |
7276b3b0 | 769 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
770 | if (error) |
771 | goto fail; | |
feaa7bba | 772 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
773 | GFS2_METATYPE_RG)) { |
774 | error = -EIO; | |
775 | goto fail; | |
776 | } | |
777 | } | |
778 | ||
cf45b752 | 779 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
780 | for (x = 0; x < length; x++) |
781 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 782 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 783 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 784 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa DT |
785 | } |
786 | ||
b3b94faa DT |
787 | return 0; |
788 | ||
feaa7bba | 789 | fail: |
b3b94faa DT |
790 | while (x--) { |
791 | bi = rgd->rd_bits + x; | |
792 | brelse(bi->bi_bh); | |
793 | bi->bi_bh = NULL; | |
794 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
795 | } | |
b3b94faa DT |
796 | |
797 | return error; | |
798 | } | |
799 | ||
b3b94faa | 800 | /** |
7c9ca621 | 801 | * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
886b1416 | 802 | * @gh: The glock holder for the resource group |
b3b94faa DT |
803 | * |
804 | */ | |
805 | ||
7c9ca621 | 806 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) |
b3b94faa | 807 | { |
7c9ca621 | 808 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
bb8d8a6f | 809 | int x, length = rgd->rd_length; |
b3b94faa | 810 | |
b3b94faa DT |
811 | for (x = 0; x < length; x++) { |
812 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
b3b94faa DT |
813 | brelse(bi->bi_bh); |
814 | bi->bi_bh = NULL; | |
815 | } | |
816 | ||
b3b94faa DT |
817 | } |
818 | ||
66fc061b | 819 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 820 | struct buffer_head *bh, |
66fc061b | 821 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
822 | { |
823 | struct super_block *sb = sdp->sd_vfs; | |
824 | struct block_device *bdev = sb->s_bdev; | |
825 | const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / | |
e1defc4f | 826 | bdev_logical_block_size(sb->s_bdev); |
f15ab561 | 827 | u64 blk; |
64d576ba | 828 | sector_t start = 0; |
f15ab561 SW |
829 | sector_t nr_sects = 0; |
830 | int rv; | |
831 | unsigned int x; | |
66fc061b SW |
832 | u32 trimmed = 0; |
833 | u8 diff; | |
f15ab561 SW |
834 | |
835 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
836 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
837 | clone += bi->bi_offset; | |
838 | clone += x; | |
839 | if (bh) { | |
840 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
841 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
842 | } else { | |
843 | diff = ~(*clone | (*clone >> 1)); | |
844 | } | |
f15ab561 SW |
845 | diff &= 0x55; |
846 | if (diff == 0) | |
847 | continue; | |
848 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
849 | blk *= sects_per_blk; /* convert to sectors */ | |
850 | while(diff) { | |
851 | if (diff & 1) { | |
852 | if (nr_sects == 0) | |
853 | goto start_new_extent; | |
854 | if ((start + nr_sects) != blk) { | |
66fc061b SW |
855 | if (nr_sects >= minlen) { |
856 | rv = blkdev_issue_discard(bdev, | |
857 | start, nr_sects, | |
858 | GFP_NOFS, 0); | |
859 | if (rv) | |
860 | goto fail; | |
861 | trimmed += nr_sects; | |
862 | } | |
f15ab561 SW |
863 | nr_sects = 0; |
864 | start_new_extent: | |
865 | start = blk; | |
866 | } | |
867 | nr_sects += sects_per_blk; | |
868 | } | |
869 | diff >>= 2; | |
870 | blk += sects_per_blk; | |
871 | } | |
872 | } | |
66fc061b | 873 | if (nr_sects >= minlen) { |
dd3932ed | 874 | rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0); |
f15ab561 SW |
875 | if (rv) |
876 | goto fail; | |
66fc061b | 877 | trimmed += nr_sects; |
f15ab561 | 878 | } |
66fc061b SW |
879 | if (ptrimmed) |
880 | *ptrimmed = trimmed; | |
881 | return 0; | |
882 | ||
f15ab561 | 883 | fail: |
66fc061b SW |
884 | if (sdp->sd_args.ar_discard) |
885 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 886 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
887 | return -EIO; |
888 | } | |
889 | ||
890 | /** | |
891 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
892 | * @filp: Any file on the filesystem | |
893 | * @argp: Pointer to the arguments (also used to pass result) | |
894 | * | |
895 | * Returns: 0 on success, otherwise error code | |
896 | */ | |
897 | ||
898 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
899 | { | |
900 | struct inode *inode = filp->f_dentry->d_inode; | |
901 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
902 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
903 | struct buffer_head *bh; | |
904 | struct gfs2_rgrpd *rgd; | |
905 | struct gfs2_rgrpd *rgd_end; | |
906 | struct gfs2_holder gh; | |
907 | struct fstrim_range r; | |
908 | int ret = 0; | |
909 | u64 amt; | |
910 | u64 trimmed = 0; | |
911 | unsigned int x; | |
912 | ||
913 | if (!capable(CAP_SYS_ADMIN)) | |
914 | return -EPERM; | |
915 | ||
916 | if (!blk_queue_discard(q)) | |
917 | return -EOPNOTSUPP; | |
918 | ||
66fc061b SW |
919 | if (argp == NULL) { |
920 | r.start = 0; | |
921 | r.len = ULLONG_MAX; | |
922 | r.minlen = 0; | |
923 | } else if (copy_from_user(&r, argp, sizeof(r))) | |
924 | return -EFAULT; | |
925 | ||
5e2f7d61 BP |
926 | ret = gfs2_rindex_update(sdp); |
927 | if (ret) | |
928 | return ret; | |
929 | ||
66fc061b SW |
930 | rgd = gfs2_blk2rgrpd(sdp, r.start, 0); |
931 | rgd_end = gfs2_blk2rgrpd(sdp, r.start + r.len, 0); | |
932 | ||
933 | while (1) { | |
934 | ||
935 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
936 | if (ret) | |
937 | goto out; | |
938 | ||
939 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
940 | /* Trim each bitmap in the rgrp */ | |
941 | for (x = 0; x < rgd->rd_length; x++) { | |
942 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
943 | ret = gfs2_rgrp_send_discards(sdp, rgd->rd_data0, NULL, bi, r.minlen, &amt); | |
944 | if (ret) { | |
945 | gfs2_glock_dq_uninit(&gh); | |
946 | goto out; | |
947 | } | |
948 | trimmed += amt; | |
949 | } | |
950 | ||
951 | /* Mark rgrp as having been trimmed */ | |
952 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
953 | if (ret == 0) { | |
954 | bh = rgd->rd_bits[0].bi_bh; | |
955 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
956 | gfs2_trans_add_bh(rgd->rd_gl, bh, 1); | |
957 | gfs2_rgrp_out(rgd, bh->b_data); | |
958 | gfs2_trans_end(sdp); | |
959 | } | |
960 | } | |
961 | gfs2_glock_dq_uninit(&gh); | |
962 | ||
963 | if (rgd == rgd_end) | |
964 | break; | |
965 | ||
966 | rgd = gfs2_rgrpd_get_next(rgd); | |
967 | } | |
968 | ||
969 | out: | |
970 | r.len = trimmed << 9; | |
971 | if (argp && copy_to_user(argp, &r, sizeof(r))) | |
972 | return -EFAULT; | |
973 | ||
974 | return ret; | |
f15ab561 SW |
975 | } |
976 | ||
b3b94faa | 977 | /** |
564e12b1 | 978 | * gfs2_qadata_get - get the struct gfs2_qadata structure for an inode |
b3b94faa DT |
979 | * @ip: the incore GFS2 inode structure |
980 | * | |
564e12b1 | 981 | * Returns: the struct gfs2_qadata |
b3b94faa DT |
982 | */ |
983 | ||
564e12b1 | 984 | struct gfs2_qadata *gfs2_qadata_get(struct gfs2_inode *ip) |
b3b94faa | 985 | { |
8339ee54 SW |
986 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
987 | int error; | |
564e12b1 BP |
988 | BUG_ON(ip->i_qadata != NULL); |
989 | ip->i_qadata = kzalloc(sizeof(struct gfs2_qadata), GFP_NOFS); | |
8339ee54 SW |
990 | error = gfs2_rindex_update(sdp); |
991 | if (error) | |
992 | fs_warn(sdp, "rindex update returns %d\n", error); | |
564e12b1 BP |
993 | return ip->i_qadata; |
994 | } | |
995 | ||
996 | /** | |
997 | * gfs2_blkrsv_get - get the struct gfs2_blkreserv structure for an inode | |
998 | * @ip: the incore GFS2 inode structure | |
999 | * | |
1000 | * Returns: the struct gfs2_qadata | |
1001 | */ | |
1002 | ||
36f5580b | 1003 | static int gfs2_blkrsv_get(struct gfs2_inode *ip) |
564e12b1 BP |
1004 | { |
1005 | BUG_ON(ip->i_res != NULL); | |
36f5580b BP |
1006 | ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); |
1007 | if (!ip->i_res) | |
1008 | return -ENOMEM; | |
1009 | return 0; | |
b3b94faa DT |
1010 | } |
1011 | ||
b3b94faa DT |
1012 | /** |
1013 | * try_rgrp_fit - See if a given reservation will fit in a given RG | |
1014 | * @rgd: the RG data | |
54335b1f | 1015 | * @ip: the inode |
b3b94faa DT |
1016 | * |
1017 | * If there's room for the requested blocks to be allocated from the RG: | |
b3b94faa DT |
1018 | * |
1019 | * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) | |
1020 | */ | |
1021 | ||
54335b1f | 1022 | static int try_rgrp_fit(const struct gfs2_rgrpd *rgd, const struct gfs2_inode *ip) |
b3b94faa | 1023 | { |
564e12b1 | 1024 | const struct gfs2_blkreserv *rs = ip->i_res; |
54335b1f | 1025 | |
09010978 | 1026 | if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) |
a43a4906 | 1027 | return 0; |
564e12b1 | 1028 | if (rgd->rd_free_clone >= rs->rs_requested) |
7c9ca621 | 1029 | return 1; |
7c9ca621 | 1030 | return 0; |
b3b94faa DT |
1031 | } |
1032 | ||
b3e47ca0 BP |
1033 | static inline u32 gfs2_bi2rgd_blk(struct gfs2_bitmap *bi, u32 blk) |
1034 | { | |
1035 | return (bi->bi_start * GFS2_NBBY) + blk; | |
1036 | } | |
1037 | ||
c8cdf479 SW |
1038 | /** |
1039 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1040 | * @rgd: The rgrp | |
886b1416 BP |
1041 | * @last_unlinked: block address of the last dinode we unlinked |
1042 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1043 | * |
1a0eae88 BP |
1044 | * Returns: 0 if no error |
1045 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1046 | */ |
1047 | ||
044b9414 | 1048 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1049 | { |
6760bdcd | 1050 | u32 goal = 0, block; |
bb9bcf06 | 1051 | u64 no_addr; |
5f3eae75 | 1052 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1053 | struct gfs2_glock *gl; |
1054 | struct gfs2_inode *ip; | |
1055 | int error; | |
1056 | int found = 0; | |
b3e47ca0 | 1057 | struct gfs2_bitmap *bi; |
c8cdf479 | 1058 | |
044b9414 | 1059 | while (goal < rgd->rd_data) { |
5f3eae75 | 1060 | down_write(&sdp->sd_log_flush_lock); |
6a8099ed | 1061 | block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, &bi); |
5f3eae75 | 1062 | up_write(&sdp->sd_log_flush_lock); |
6760bdcd | 1063 | if (block == BFITNOENT) |
24c73873 | 1064 | break; |
b3e47ca0 BP |
1065 | |
1066 | block = gfs2_bi2rgd_blk(bi, block); | |
6760bdcd BP |
1067 | /* rgblk_search can return a block < goal, so we need to |
1068 | keep it marching forward. */ | |
1069 | no_addr = block + rgd->rd_data0; | |
44ad37d6 | 1070 | goal = max(block + 1, goal + 1); |
6760bdcd | 1071 | if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked) |
c8cdf479 | 1072 | continue; |
1e19a195 SW |
1073 | if (no_addr == skip) |
1074 | continue; | |
bb9bcf06 | 1075 | *last_unlinked = no_addr; |
044b9414 SW |
1076 | |
1077 | error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &gl); | |
1078 | if (error) | |
1079 | continue; | |
1080 | ||
1081 | /* If the inode is already in cache, we can ignore it here | |
1082 | * because the existing inode disposal code will deal with | |
1083 | * it when all refs have gone away. Accessing gl_object like | |
1084 | * this is not safe in general. Here it is ok because we do | |
1085 | * not dereference the pointer, and we only need an approx | |
1086 | * answer to whether it is NULL or not. | |
1087 | */ | |
1088 | ip = gl->gl_object; | |
1089 | ||
1090 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1091 | gfs2_glock_put(gl); | |
1092 | else | |
1093 | found++; | |
1094 | ||
1095 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1096 | if (found > NR_CPUS) |
044b9414 | 1097 | return; |
c8cdf479 SW |
1098 | } |
1099 | ||
1100 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1101 | return; |
c8cdf479 SW |
1102 | } |
1103 | ||
b3b94faa DT |
1104 | /** |
1105 | * get_local_rgrp - Choose and lock a rgrp for allocation | |
1106 | * @ip: the inode to reserve space for | |
886b1416 | 1107 | * @last_unlinked: the last unlinked block |
b3b94faa DT |
1108 | * |
1109 | * Try to acquire rgrp in way which avoids contending with others. | |
1110 | * | |
1111 | * Returns: errno | |
1112 | */ | |
1113 | ||
044b9414 | 1114 | static int get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked) |
b3b94faa | 1115 | { |
feaa7bba | 1116 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa | 1117 | struct gfs2_rgrpd *rgd, *begin = NULL; |
564e12b1 | 1118 | struct gfs2_blkreserv *rs = ip->i_res; |
c688b8b3 | 1119 | int error, rg_locked, flags = LM_FLAG_TRY; |
7c9ca621 | 1120 | int loops = 0; |
b3b94faa | 1121 | |
54335b1f SW |
1122 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) |
1123 | rgd = begin = ip->i_rgd; | |
1124 | else | |
66fc061b | 1125 | rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
b3b94faa | 1126 | |
7c9ca621 BP |
1127 | if (rgd == NULL) |
1128 | return -EBADSLT; | |
1129 | ||
1130 | while (loops < 3) { | |
292c8c14 AD |
1131 | rg_locked = 0; |
1132 | ||
1133 | if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) { | |
1134 | rg_locked = 1; | |
1135 | error = 0; | |
1136 | } else { | |
1137 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, | |
564e12b1 | 1138 | flags, &rs->rs_rgd_gh); |
292c8c14 | 1139 | } |
b3b94faa DT |
1140 | switch (error) { |
1141 | case 0: | |
54335b1f SW |
1142 | if (try_rgrp_fit(rgd, ip)) { |
1143 | ip->i_rgd = rgd; | |
7c9ca621 | 1144 | return 0; |
54335b1f | 1145 | } |
044b9414 SW |
1146 | if (rgd->rd_flags & GFS2_RDF_CHECK) |
1147 | try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr); | |
292c8c14 | 1148 | if (!rg_locked) |
564e12b1 | 1149 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
9cabcdbd | 1150 | /* fall through */ |
b3b94faa | 1151 | case GLR_TRYFAILED: |
7c9ca621 | 1152 | rgd = gfs2_rgrpd_get_next(rgd); |
c688b8b3 BP |
1153 | if (rgd == begin) { |
1154 | flags = 0; | |
7c9ca621 | 1155 | loops++; |
c688b8b3 | 1156 | } |
b3b94faa | 1157 | break; |
b3b94faa | 1158 | default: |
cc0581bd | 1159 | return error; |
b3b94faa | 1160 | } |
b3b94faa DT |
1161 | } |
1162 | ||
7c9ca621 | 1163 | return -ENOSPC; |
b3b94faa DT |
1164 | } |
1165 | ||
564e12b1 BP |
1166 | static void gfs2_blkrsv_put(struct gfs2_inode *ip) |
1167 | { | |
1168 | BUG_ON(ip->i_res == NULL); | |
36f5580b | 1169 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); |
564e12b1 BP |
1170 | ip->i_res = NULL; |
1171 | } | |
1172 | ||
b3b94faa | 1173 | /** |
9ae32429 | 1174 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1175 | * @ip: the inode to reserve space for |
886b1416 | 1176 | * @requested: the number of blocks to be reserved |
b3b94faa DT |
1177 | * |
1178 | * Returns: errno | |
1179 | */ | |
1180 | ||
564e12b1 | 1181 | int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested) |
b3b94faa | 1182 | { |
feaa7bba | 1183 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
564e12b1 | 1184 | struct gfs2_blkreserv *rs; |
36f5580b | 1185 | int error; |
044b9414 SW |
1186 | u64 last_unlinked = NO_BLOCK; |
1187 | int tries = 0; | |
b3b94faa | 1188 | |
36f5580b BP |
1189 | error = gfs2_blkrsv_get(ip); |
1190 | if (error) | |
1191 | return error; | |
564e12b1 | 1192 | |
36f5580b | 1193 | rs = ip->i_res; |
564e12b1 BP |
1194 | rs->rs_requested = requested; |
1195 | if (gfs2_assert_warn(sdp, requested)) { | |
1196 | error = -EINVAL; | |
1197 | goto out; | |
1198 | } | |
b3b94faa | 1199 | |
044b9414 SW |
1200 | do { |
1201 | error = get_local_rgrp(ip, &last_unlinked); | |
54335b1f SW |
1202 | if (error != -ENOSPC) |
1203 | break; | |
1204 | /* Check that fs hasn't grown if writing to rindex */ | |
1205 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
1206 | error = gfs2_ri_update(ip); | |
1207 | if (error) | |
1208 | break; | |
1209 | continue; | |
0489b3f5 | 1210 | } |
54335b1f SW |
1211 | /* Flushing the log may release space */ |
1212 | gfs2_log_flush(sdp, NULL); | |
1213 | } while (tries++ < 3); | |
b3b94faa | 1214 | |
564e12b1 BP |
1215 | out: |
1216 | if (error) | |
1217 | gfs2_blkrsv_put(ip); | |
9ae32429 | 1218 | return error; |
b3b94faa DT |
1219 | } |
1220 | ||
1221 | /** | |
1222 | * gfs2_inplace_release - release an inplace reservation | |
1223 | * @ip: the inode the reservation was taken out on | |
1224 | * | |
1225 | * Release a reservation made by gfs2_inplace_reserve(). | |
1226 | */ | |
1227 | ||
1228 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1229 | { | |
564e12b1 | 1230 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 1231 | |
564e12b1 BP |
1232 | if (rs->rs_rgd_gh.gh_gl) |
1233 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
49528b4e | 1234 | gfs2_blkrsv_put(ip); |
b3b94faa DT |
1235 | } |
1236 | ||
1237 | /** | |
1238 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1239 | * @rgd: the resource group holding the block | |
1240 | * @block: the block number | |
1241 | * | |
1242 | * Returns: The block type (GFS2_BLKST_*) | |
1243 | */ | |
1244 | ||
acf7e244 | 1245 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa DT |
1246 | { |
1247 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1248 | u32 length, rgrp_block, buf_block; |
b3b94faa DT |
1249 | unsigned int buf; |
1250 | unsigned char type; | |
1251 | ||
bb8d8a6f SW |
1252 | length = rgd->rd_length; |
1253 | rgrp_block = block - rgd->rd_data0; | |
b3b94faa DT |
1254 | |
1255 | for (buf = 0; buf < length; buf++) { | |
1256 | bi = rgd->rd_bits + buf; | |
1257 | if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1258 | break; | |
1259 | } | |
1260 | ||
1261 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1262 | buf_block = rgrp_block - bi->bi_start * GFS2_NBBY; | |
1263 | ||
feaa7bba | 1264 | type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset, |
b3b94faa DT |
1265 | bi->bi_len, buf_block); |
1266 | ||
1267 | return type; | |
1268 | } | |
1269 | ||
1270 | /** | |
6a8099ed | 1271 | * rgblk_search - find a block in @state |
b3b94faa DT |
1272 | * @rgd: the resource group descriptor |
1273 | * @goal: the goal block within the RG (start here to search for avail block) | |
6a8099ed | 1274 | * @state: GFS2_BLKST_XXX the before-allocation state to find |
b3e47ca0 | 1275 | * @rbi: address of the pointer to the bitmap containing the block found |
b3b94faa | 1276 | * |
6a8099ed | 1277 | * Walk rgrp's bitmap to find bits that represent a block in @state. |
b3b94faa DT |
1278 | * |
1279 | * This function never fails, because we wouldn't call it unless we | |
1280 | * know (from reservation results, etc.) that a block is available. | |
1281 | * | |
b3e47ca0 BP |
1282 | * Scope of @goal is just within rgrp, not the whole filesystem. |
1283 | * Scope of @returned block is just within bitmap, not the whole filesystem. | |
b3b94faa | 1284 | * |
b3e47ca0 | 1285 | * Returns: the block number found relative to the bitmap rbi |
b3b94faa DT |
1286 | */ |
1287 | ||
886b1416 | 1288 | static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, unsigned char state, |
b3e47ca0 | 1289 | struct gfs2_bitmap **rbi) |
b3b94faa DT |
1290 | { |
1291 | struct gfs2_bitmap *bi = NULL; | |
b45e41d7 | 1292 | const u32 length = rgd->rd_length; |
9598d25e | 1293 | u32 biblk = BFITNOENT; |
b3b94faa | 1294 | unsigned int buf, x; |
60a0b8f9 | 1295 | const u8 *buffer = NULL; |
b3b94faa | 1296 | |
b3e47ca0 | 1297 | *rbi = NULL; |
b3b94faa DT |
1298 | /* Find bitmap block that contains bits for goal block */ |
1299 | for (buf = 0; buf < length; buf++) { | |
1300 | bi = rgd->rd_bits + buf; | |
60a0b8f9 SW |
1301 | /* Convert scope of "goal" from rgrp-wide to within found bit block */ |
1302 | if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { | |
1303 | goal -= bi->bi_start * GFS2_NBBY; | |
1304 | goto do_search; | |
1305 | } | |
b3b94faa | 1306 | } |
60a0b8f9 SW |
1307 | buf = 0; |
1308 | goal = 0; | |
b3b94faa | 1309 | |
60a0b8f9 | 1310 | do_search: |
b3b94faa DT |
1311 | /* Search (up to entire) bitmap in this rgrp for allocatable block. |
1312 | "x <= length", instead of "x < length", because we typically start | |
1313 | the search in the middle of a bit block, but if we can't find an | |
1314 | allocatable block anywhere else, we want to be able wrap around and | |
1315 | search in the first part of our first-searched bit block. */ | |
1316 | for (x = 0; x <= length; x++) { | |
60a0b8f9 SW |
1317 | bi = rgd->rd_bits + buf; |
1318 | ||
1319 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
6a8099ed | 1320 | (state == GFS2_BLKST_FREE)) |
60a0b8f9 SW |
1321 | goto skip; |
1322 | ||
5f3eae75 BP |
1323 | /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone |
1324 | bitmaps, so we must search the originals for that. */ | |
b45e41d7 | 1325 | buffer = bi->bi_bh->b_data + bi->bi_offset; |
7c9ca621 | 1326 | WARN_ON(!buffer_uptodate(bi->bi_bh)); |
6a8099ed | 1327 | if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
110acf38 SW |
1328 | buffer = bi->bi_clone + bi->bi_offset; |
1329 | ||
9598d25e BP |
1330 | biblk = gfs2_bitfit(buffer, bi->bi_len, goal, state); |
1331 | if (biblk != BFITNOENT) | |
b3b94faa DT |
1332 | break; |
1333 | ||
6a8099ed | 1334 | if ((goal == 0) && (state == GFS2_BLKST_FREE)) |
60a0b8f9 SW |
1335 | set_bit(GBF_FULL, &bi->bi_flags); |
1336 | ||
b3b94faa | 1337 | /* Try next bitmap block (wrap back to rgrp header if at end) */ |
60a0b8f9 SW |
1338 | skip: |
1339 | buf++; | |
1340 | buf %= length; | |
b3b94faa DT |
1341 | goal = 0; |
1342 | } | |
1343 | ||
9598d25e | 1344 | if (biblk != BFITNOENT) |
b3e47ca0 | 1345 | *rbi = bi; |
7c9ca621 | 1346 | |
9598d25e | 1347 | return biblk; |
b3e47ca0 | 1348 | } |
60a0b8f9 | 1349 | |
b3e47ca0 BP |
1350 | /** |
1351 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
1352 | * @rgd: the resource group descriptor | |
1353 | * @bi: the bitmap within the rgrp | |
1354 | * @blk: the block within the bitmap | |
1355 | * @dinode: TRUE if the first block we allocate is for a dinode | |
1356 | * @n: The extent length | |
1357 | * | |
1358 | * Add the found bitmap buffer to the transaction. | |
1359 | * Set the found bits to @new_state to change block's allocation state. | |
1360 | * Returns: starting block number of the extent (fs scope) | |
1361 | */ | |
1362 | static u64 gfs2_alloc_extent(struct gfs2_rgrpd *rgd, struct gfs2_bitmap *bi, | |
1363 | u32 blk, bool dinode, unsigned int *n) | |
1364 | { | |
1365 | const unsigned int elen = *n; | |
1366 | u32 goal; | |
1367 | const u8 *buffer = NULL; | |
1368 | ||
6a8099ed | 1369 | *n = 0; |
b3e47ca0 | 1370 | buffer = bi->bi_bh->b_data + bi->bi_offset; |
60a0b8f9 | 1371 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
29c578f5 | 1372 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, |
6e87ed0f | 1373 | bi, blk, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6a8099ed | 1374 | (*n)++; |
60a0b8f9 SW |
1375 | goal = blk; |
1376 | while (*n < elen) { | |
1377 | goal++; | |
1378 | if (goal >= (bi->bi_len * GFS2_NBBY)) | |
1379 | break; | |
1380 | if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) != | |
1381 | GFS2_BLKST_FREE) | |
1382 | break; | |
29c578f5 | 1383 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, |
6e87ed0f | 1384 | bi, goal, GFS2_BLKST_USED); |
60a0b8f9 | 1385 | (*n)++; |
c8cdf479 | 1386 | } |
b3e47ca0 | 1387 | blk = gfs2_bi2rgd_blk(bi, blk); |
6a8099ed | 1388 | rgd->rd_last_alloc = blk + *n - 1; |
b3e47ca0 | 1389 | return rgd->rd_data0 + blk; |
b3b94faa DT |
1390 | } |
1391 | ||
1392 | /** | |
1393 | * rgblk_free - Change alloc state of given block(s) | |
1394 | * @sdp: the filesystem | |
1395 | * @bstart: the start of a run of blocks to free | |
1396 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1397 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1398 | * | |
1399 | * Returns: Resource group containing the block(s) | |
1400 | */ | |
1401 | ||
cd915493 SW |
1402 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
1403 | u32 blen, unsigned char new_state) | |
b3b94faa DT |
1404 | { |
1405 | struct gfs2_rgrpd *rgd; | |
1406 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1407 | u32 length, rgrp_blk, buf_blk; |
b3b94faa DT |
1408 | unsigned int buf; |
1409 | ||
66fc061b | 1410 | rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
b3b94faa DT |
1411 | if (!rgd) { |
1412 | if (gfs2_consist(sdp)) | |
382066da | 1413 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
1414 | return NULL; |
1415 | } | |
1416 | ||
bb8d8a6f | 1417 | length = rgd->rd_length; |
b3b94faa | 1418 | |
bb8d8a6f | 1419 | rgrp_blk = bstart - rgd->rd_data0; |
b3b94faa DT |
1420 | |
1421 | while (blen--) { | |
1422 | for (buf = 0; buf < length; buf++) { | |
1423 | bi = rgd->rd_bits + buf; | |
1424 | if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1425 | break; | |
1426 | } | |
1427 | ||
1428 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1429 | ||
1430 | buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY; | |
1431 | rgrp_blk++; | |
1432 | ||
1433 | if (!bi->bi_clone) { | |
1434 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
dd894be8 | 1435 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1436 | memcpy(bi->bi_clone + bi->bi_offset, |
1437 | bi->bi_bh->b_data + bi->bi_offset, | |
1438 | bi->bi_len); | |
1439 | } | |
d4e9c4c3 | 1440 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
29c578f5 | 1441 | gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, |
95c8e17f | 1442 | bi, buf_blk, new_state); |
b3b94faa DT |
1443 | } |
1444 | ||
1445 | return rgd; | |
1446 | } | |
1447 | ||
1448 | /** | |
09010978 SW |
1449 | * gfs2_rgrp_dump - print out an rgrp |
1450 | * @seq: The iterator | |
1451 | * @gl: The glock in question | |
1452 | * | |
1453 | */ | |
1454 | ||
1455 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
1456 | { | |
1457 | const struct gfs2_rgrpd *rgd = gl->gl_object; | |
1458 | if (rgd == NULL) | |
1459 | return 0; | |
1460 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n", | |
1461 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, | |
1462 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes); | |
1463 | return 0; | |
1464 | } | |
1465 | ||
6050b9c7 SW |
1466 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
1467 | { | |
1468 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1469 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 1470 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
1471 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
1472 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
1473 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
1474 | } | |
1475 | ||
09010978 | 1476 | /** |
6e87ed0f | 1477 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 1478 | * @ip: the inode to allocate the block for |
09010978 | 1479 | * @bn: Used to return the starting block number |
6a8099ed | 1480 | * @ndata: requested number of blocks/extent length (value/result) |
6e87ed0f | 1481 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 1482 | * @generation: the generation number of the inode |
b3b94faa | 1483 | * |
09010978 | 1484 | * Returns: 0 or error |
b3b94faa DT |
1485 | */ |
1486 | ||
6a8099ed | 1487 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 1488 | bool dinode, u64 *generation) |
b3b94faa | 1489 | { |
feaa7bba | 1490 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 1491 | struct buffer_head *dibh; |
9a3f236d | 1492 | struct gfs2_rgrpd *rgd; |
6a8099ed SW |
1493 | unsigned int ndata; |
1494 | u32 goal, blk; /* block, within the rgrp scope */ | |
3c5d785a | 1495 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 1496 | int error; |
b3e47ca0 | 1497 | struct gfs2_bitmap *bi; |
b3b94faa | 1498 | |
9a3f236d SW |
1499 | /* Only happens if there is a bug in gfs2, return something distinctive |
1500 | * to ensure that it is noticed. | |
1501 | */ | |
564e12b1 | 1502 | if (ip->i_res == NULL) |
9a3f236d SW |
1503 | return -ECANCELED; |
1504 | ||
54335b1f | 1505 | rgd = ip->i_rgd; |
9a3f236d | 1506 | |
3c5d785a | 1507 | if (!dinode && rgrp_contains_block(rgd, ip->i_goal)) |
ce276b06 | 1508 | goal = ip->i_goal - rgd->rd_data0; |
b3b94faa | 1509 | else |
ac576cc5 | 1510 | goal = rgd->rd_last_alloc; |
b3b94faa | 1511 | |
6a8099ed | 1512 | blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, &bi); |
09010978 SW |
1513 | |
1514 | /* Since all blocks are reserved in advance, this shouldn't happen */ | |
1515 | if (blk == BFITNOENT) | |
1516 | goto rgrp_error; | |
b3b94faa | 1517 | |
6a8099ed SW |
1518 | block = gfs2_alloc_extent(rgd, bi, blk, dinode, nblocks); |
1519 | ndata = *nblocks; | |
1520 | if (dinode) | |
1521 | ndata--; | |
b3e47ca0 | 1522 | |
3c5d785a | 1523 | if (!dinode) { |
6a8099ed | 1524 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
1525 | error = gfs2_meta_inode_buffer(ip, &dibh); |
1526 | if (error == 0) { | |
1527 | struct gfs2_dinode *di = | |
1528 | (struct gfs2_dinode *)dibh->b_data; | |
1529 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
1530 | di->di_goal_meta = di->di_goal_data = | |
1531 | cpu_to_be64(ip->i_goal); | |
1532 | brelse(dibh); | |
1533 | } | |
d9ba7615 | 1534 | } |
6a8099ed | 1535 | if (rgd->rd_free < *nblocks) |
09010978 SW |
1536 | goto rgrp_error; |
1537 | ||
6a8099ed | 1538 | rgd->rd_free -= *nblocks; |
3c5d785a BP |
1539 | if (dinode) { |
1540 | rgd->rd_dinodes++; | |
1541 | *generation = rgd->rd_igeneration++; | |
1542 | if (*generation == 0) | |
1543 | *generation = rgd->rd_igeneration++; | |
1544 | } | |
b3b94faa | 1545 | |
d4e9c4c3 | 1546 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1547 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 1548 | |
6a8099ed | 1549 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a BP |
1550 | if (dinode) |
1551 | gfs2_trans_add_unrevoke(sdp, block, 1); | |
6a8099ed SW |
1552 | |
1553 | /* | |
1554 | * This needs reviewing to see why we cannot do the quota change | |
1555 | * at this point in the dinode case. | |
1556 | */ | |
1557 | if (ndata) | |
1558 | gfs2_quota_change(ip, ndata, ip->i_inode.i_uid, | |
3c5d785a | 1559 | ip->i_inode.i_gid); |
b3b94faa | 1560 | |
6a8099ed SW |
1561 | rgd->rd_free_clone -= *nblocks; |
1562 | trace_gfs2_block_alloc(ip, block, *nblocks, | |
6e87ed0f | 1563 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
1564 | *bn = block; |
1565 | return 0; | |
1566 | ||
1567 | rgrp_error: | |
1568 | gfs2_rgrp_error(rgd); | |
1569 | return -EIO; | |
b3b94faa DT |
1570 | } |
1571 | ||
1572 | /** | |
46fcb2ed | 1573 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
1574 | * @ip: the inode these blocks are being freed from |
1575 | * @bstart: first block of a run of contiguous blocks | |
1576 | * @blen: the length of the block run | |
46fcb2ed | 1577 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
1578 | * |
1579 | */ | |
1580 | ||
46fcb2ed | 1581 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 1582 | { |
feaa7bba | 1583 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1584 | struct gfs2_rgrpd *rgd; |
1585 | ||
1586 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
1587 | if (!rgd) | |
1588 | return; | |
63997775 | 1589 | trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 1590 | rgd->rd_free += blen; |
66fc061b | 1591 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
d4e9c4c3 | 1592 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1593 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 1594 | |
6d3117b4 | 1595 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 1596 | if (meta || ip->i_depth) |
6d3117b4 | 1597 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 1598 | } |
b3b94faa | 1599 | |
4c16c36a BP |
1600 | /** |
1601 | * gfs2_free_meta - free a contiguous run of data block(s) | |
1602 | * @ip: the inode these blocks are being freed from | |
1603 | * @bstart: first block of a run of contiguous blocks | |
1604 | * @blen: the length of the block run | |
1605 | * | |
1606 | */ | |
1607 | ||
1608 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
1609 | { | |
1610 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1611 | ||
46fcb2ed | 1612 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 1613 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 1614 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
1615 | } |
1616 | ||
feaa7bba SW |
1617 | void gfs2_unlink_di(struct inode *inode) |
1618 | { | |
1619 | struct gfs2_inode *ip = GFS2_I(inode); | |
1620 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1621 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 1622 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
1623 | |
1624 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
1625 | if (!rgd) | |
1626 | return; | |
63997775 | 1627 | trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED); |
feaa7bba | 1628 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1629 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
feaa7bba SW |
1630 | } |
1631 | ||
cd915493 | 1632 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
1633 | { |
1634 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1635 | struct gfs2_rgrpd *tmp_rgd; | |
1636 | ||
1637 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
1638 | if (!tmp_rgd) | |
1639 | return; | |
1640 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
1641 | ||
73f74948 | 1642 | if (!rgd->rd_dinodes) |
b3b94faa | 1643 | gfs2_consist_rgrpd(rgd); |
73f74948 | 1644 | rgd->rd_dinodes--; |
cfc8b549 | 1645 | rgd->rd_free++; |
b3b94faa | 1646 | |
d4e9c4c3 | 1647 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1648 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1649 | |
1650 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
1651 | } |
1652 | ||
b3b94faa DT |
1653 | |
1654 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
1655 | { | |
dbb7cae2 | 1656 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
63997775 | 1657 | trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 1658 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 1659 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
1660 | } |
1661 | ||
acf7e244 SW |
1662 | /** |
1663 | * gfs2_check_blk_type - Check the type of a block | |
1664 | * @sdp: The superblock | |
1665 | * @no_addr: The block number to check | |
1666 | * @type: The block type we are looking for | |
1667 | * | |
1668 | * Returns: 0 if the block type matches the expected type | |
1669 | * -ESTALE if it doesn't match | |
1670 | * or -ve errno if something went wrong while checking | |
1671 | */ | |
1672 | ||
1673 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
1674 | { | |
1675 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 1676 | struct gfs2_holder rgd_gh; |
58884c4d | 1677 | int error = -EINVAL; |
acf7e244 | 1678 | |
66fc061b | 1679 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 1680 | if (!rgd) |
8339ee54 | 1681 | goto fail; |
acf7e244 SW |
1682 | |
1683 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
1684 | if (error) | |
8339ee54 | 1685 | goto fail; |
acf7e244 SW |
1686 | |
1687 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
1688 | error = -ESTALE; | |
1689 | ||
1690 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
1691 | fail: |
1692 | return error; | |
1693 | } | |
1694 | ||
b3b94faa DT |
1695 | /** |
1696 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 1697 | * @ip: the inode |
b3b94faa DT |
1698 | * @rlist: the list of resource groups |
1699 | * @block: the block | |
1700 | * | |
1701 | * Figure out what RG a block belongs to and add that RG to the list | |
1702 | * | |
1703 | * FIXME: Don't use NOFAIL | |
1704 | * | |
1705 | */ | |
1706 | ||
70b0c365 | 1707 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 1708 | u64 block) |
b3b94faa | 1709 | { |
70b0c365 | 1710 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1711 | struct gfs2_rgrpd *rgd; |
1712 | struct gfs2_rgrpd **tmp; | |
1713 | unsigned int new_space; | |
1714 | unsigned int x; | |
1715 | ||
1716 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
1717 | return; | |
1718 | ||
70b0c365 SW |
1719 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
1720 | rgd = ip->i_rgd; | |
1721 | else | |
66fc061b | 1722 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 1723 | if (!rgd) { |
70b0c365 | 1724 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
1725 | return; |
1726 | } | |
70b0c365 | 1727 | ip->i_rgd = rgd; |
b3b94faa DT |
1728 | |
1729 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1730 | if (rlist->rl_rgd[x] == rgd) | |
1731 | return; | |
1732 | ||
1733 | if (rlist->rl_rgrps == rlist->rl_space) { | |
1734 | new_space = rlist->rl_space + 10; | |
1735 | ||
1736 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 1737 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1738 | |
1739 | if (rlist->rl_rgd) { | |
1740 | memcpy(tmp, rlist->rl_rgd, | |
1741 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
1742 | kfree(rlist->rl_rgd); | |
1743 | } | |
1744 | ||
1745 | rlist->rl_space = new_space; | |
1746 | rlist->rl_rgd = tmp; | |
1747 | } | |
1748 | ||
1749 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
1750 | } | |
1751 | ||
1752 | /** | |
1753 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
1754 | * and initialize an array of glock holders for them | |
1755 | * @rlist: the list of resource groups | |
1756 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
1757 | * |
1758 | * FIXME: Don't use NOFAIL | |
1759 | * | |
1760 | */ | |
1761 | ||
fe6c991c | 1762 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
1763 | { |
1764 | unsigned int x; | |
1765 | ||
1766 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 1767 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1768 | for (x = 0; x < rlist->rl_rgrps; x++) |
1769 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 1770 | state, 0, |
b3b94faa DT |
1771 | &rlist->rl_ghs[x]); |
1772 | } | |
1773 | ||
1774 | /** | |
1775 | * gfs2_rlist_free - free a resource group list | |
1776 | * @list: the list of resource groups | |
1777 | * | |
1778 | */ | |
1779 | ||
1780 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
1781 | { | |
1782 | unsigned int x; | |
1783 | ||
1784 | kfree(rlist->rl_rgd); | |
1785 | ||
1786 | if (rlist->rl_ghs) { | |
1787 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1788 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
1789 | kfree(rlist->rl_ghs); | |
1790 | } | |
1791 | } | |
1792 |