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 | |
73 | * @buffer: the buffer that holds the bitmaps | |
74 | * @buflen: the length (in bytes) of the buffer | |
75 | * @block: the block to set | |
76 | * @new_state: the new state of the block | |
77 | * | |
78 | */ | |
79 | ||
b45e41d7 SW |
80 | static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1, |
81 | unsigned char *buf2, unsigned int offset, | |
95c8e17f | 82 | struct gfs2_bitmap *bi, u32 block, |
b45e41d7 | 83 | unsigned char new_state) |
88c8ab1f | 84 | { |
b45e41d7 | 85 | unsigned char *byte1, *byte2, *end, cur_state; |
95c8e17f | 86 | unsigned int buflen = bi->bi_len; |
b45e41d7 | 87 | const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 88 | |
b45e41d7 SW |
89 | byte1 = buf1 + offset + (block / GFS2_NBBY); |
90 | end = buf1 + offset + buflen; | |
88c8ab1f | 91 | |
b45e41d7 | 92 | BUG_ON(byte1 >= end); |
88c8ab1f | 93 | |
b45e41d7 | 94 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 95 | |
b45e41d7 | 96 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
95c8e17f BP |
97 | printk(KERN_WARNING "GFS2: buf_blk = 0x%llx old_state=%d, " |
98 | "new_state=%d\n", | |
99 | (unsigned long long)block, cur_state, new_state); | |
100 | printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%lx\n", | |
101 | (unsigned long long)rgd->rd_addr, | |
102 | (unsigned long)bi->bi_start); | |
103 | printk(KERN_WARNING "GFS2: bi_offset=0x%lx bi_len=0x%lx\n", | |
104 | (unsigned long)bi->bi_offset, | |
105 | (unsigned long)bi->bi_len); | |
106 | dump_stack(); | |
88c8ab1f | 107 | gfs2_consist_rgrpd(rgd); |
b45e41d7 SW |
108 | return; |
109 | } | |
110 | *byte1 ^= (cur_state ^ new_state) << bit; | |
111 | ||
112 | if (buf2) { | |
113 | byte2 = buf2 + offset + (block / GFS2_NBBY); | |
114 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; | |
115 | *byte2 ^= (cur_state ^ new_state) << bit; | |
116 | } | |
88c8ab1f SW |
117 | } |
118 | ||
119 | /** | |
120 | * gfs2_testbit - test a bit in the bitmaps | |
121 | * @buffer: the buffer that holds the bitmaps | |
122 | * @buflen: the length (in bytes) of the buffer | |
123 | * @block: the block to read | |
124 | * | |
125 | */ | |
126 | ||
b45e41d7 SW |
127 | static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd, |
128 | const unsigned char *buffer, | |
129 | unsigned int buflen, u32 block) | |
88c8ab1f | 130 | { |
b45e41d7 SW |
131 | const unsigned char *byte, *end; |
132 | unsigned char cur_state; | |
88c8ab1f SW |
133 | unsigned int bit; |
134 | ||
135 | byte = buffer + (block / GFS2_NBBY); | |
136 | bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; | |
137 | end = buffer + buflen; | |
138 | ||
139 | gfs2_assert(rgd->rd_sbd, byte < end); | |
140 | ||
141 | cur_state = (*byte >> bit) & GFS2_BIT_MASK; | |
142 | ||
143 | return cur_state; | |
144 | } | |
145 | ||
223b2b88 SW |
146 | /** |
147 | * gfs2_bit_search | |
148 | * @ptr: Pointer to bitmap data | |
149 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
150 | * @state: The state we are searching for | |
151 | * | |
152 | * We xor the bitmap data with a patter which is the bitwise opposite | |
153 | * of what we are looking for, this gives rise to a pattern of ones | |
154 | * wherever there is a match. Since we have two bits per entry, we | |
155 | * take this pattern, shift it down by one place and then and it with | |
156 | * the original. All the even bit positions (0,2,4, etc) then represent | |
157 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
158 | * odd bit positions. | |
159 | * | |
160 | * This allows searching of a whole u64 at once (32 blocks) with a | |
161 | * single test (on 64 bit arches). | |
162 | */ | |
163 | ||
164 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
165 | { | |
166 | u64 tmp; | |
167 | static const u64 search[] = { | |
075ac448 HE |
168 | [0] = 0xffffffffffffffffULL, |
169 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
170 | [2] = 0x5555555555555555ULL, | |
171 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
172 | }; |
173 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
174 | tmp &= (tmp >> 1); | |
175 | tmp &= mask; | |
176 | return tmp; | |
177 | } | |
178 | ||
88c8ab1f SW |
179 | /** |
180 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
181 | * a block in a given allocation state. | |
182 | * @buffer: the buffer that holds the bitmaps | |
223b2b88 | 183 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 184 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 185 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
186 | * |
187 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
188 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
189 | * beginning of a bitmap block buffer, skipping any header structures, but |
190 | * headers are always a multiple of 64 bits long so that the buffer is | |
191 | * always aligned to a 64 bit boundary. | |
192 | * | |
193 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 194 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 195 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
196 | * |
197 | * Return: the block number (bitmap buffer scope) that was found | |
198 | */ | |
199 | ||
02ab1721 HE |
200 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
201 | u32 goal, u8 state) | |
88c8ab1f | 202 | { |
223b2b88 SW |
203 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
204 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
205 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
206 | u64 tmp; | |
075ac448 | 207 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
208 | u32 bit; |
209 | ||
210 | BUG_ON(state > 3); | |
211 | ||
212 | /* Mask off bits we don't care about at the start of the search */ | |
213 | mask <<= spoint; | |
214 | tmp = gfs2_bit_search(ptr, mask, state); | |
215 | ptr++; | |
216 | while(tmp == 0 && ptr < end) { | |
075ac448 | 217 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 218 | ptr++; |
1f466a47 | 219 | } |
223b2b88 SW |
220 | /* Mask off any bits which are more than len bytes from the start */ |
221 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
222 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
223 | /* Didn't find anything, so return */ | |
224 | if (tmp == 0) | |
225 | return BFITNOENT; | |
226 | ptr--; | |
d8bd504a | 227 | bit = __ffs64(tmp); |
223b2b88 SW |
228 | bit /= 2; /* two bits per entry in the bitmap */ |
229 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
230 | } |
231 | ||
232 | /** | |
233 | * gfs2_bitcount - count the number of bits in a certain state | |
234 | * @buffer: the buffer that holds the bitmaps | |
235 | * @buflen: the length (in bytes) of the buffer | |
236 | * @state: the state of the block we're looking for | |
237 | * | |
238 | * Returns: The number of bits | |
239 | */ | |
240 | ||
110acf38 SW |
241 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
242 | unsigned int buflen, u8 state) | |
88c8ab1f | 243 | { |
110acf38 SW |
244 | const u8 *byte = buffer; |
245 | const u8 *end = buffer + buflen; | |
246 | const u8 state1 = state << 2; | |
247 | const u8 state2 = state << 4; | |
248 | const u8 state3 = state << 6; | |
cd915493 | 249 | u32 count = 0; |
88c8ab1f SW |
250 | |
251 | for (; byte < end; byte++) { | |
252 | if (((*byte) & 0x03) == state) | |
253 | count++; | |
254 | if (((*byte) & 0x0C) == state1) | |
255 | count++; | |
256 | if (((*byte) & 0x30) == state2) | |
257 | count++; | |
258 | if (((*byte) & 0xC0) == state3) | |
259 | count++; | |
260 | } | |
261 | ||
262 | return count; | |
263 | } | |
264 | ||
b3b94faa DT |
265 | /** |
266 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
267 | * @sdp: the filesystem | |
268 | * @rgd: the rgrp | |
269 | * | |
270 | */ | |
271 | ||
272 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
273 | { | |
274 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
275 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 276 | u32 length = rgd->rd_length; |
cd915493 | 277 | u32 count[4], tmp; |
b3b94faa DT |
278 | int buf, x; |
279 | ||
cd915493 | 280 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
281 | |
282 | /* Count # blocks in each of 4 possible allocation states */ | |
283 | for (buf = 0; buf < length; buf++) { | |
284 | bi = rgd->rd_bits + buf; | |
285 | for (x = 0; x < 4; x++) | |
286 | count[x] += gfs2_bitcount(rgd, | |
287 | bi->bi_bh->b_data + | |
288 | bi->bi_offset, | |
289 | bi->bi_len, x); | |
290 | } | |
291 | ||
cfc8b549 | 292 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
293 | if (gfs2_consist_rgrpd(rgd)) |
294 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 295 | count[0], rgd->rd_free); |
b3b94faa DT |
296 | return; |
297 | } | |
298 | ||
73f74948 | 299 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 300 | if (count[1] != tmp) { |
b3b94faa DT |
301 | if (gfs2_consist_rgrpd(rgd)) |
302 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
303 | count[1], tmp); | |
304 | return; | |
305 | } | |
306 | ||
6b946170 | 307 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 308 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 309 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 310 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
311 | return; |
312 | } | |
b3b94faa DT |
313 | } |
314 | ||
bb8d8a6f | 315 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 316 | { |
bb8d8a6f SW |
317 | u64 first = rgd->rd_data0; |
318 | u64 last = first + rgd->rd_data; | |
16910427 | 319 | return first <= block && block < last; |
b3b94faa DT |
320 | } |
321 | ||
322 | /** | |
323 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
324 | * @sdp: The GFS2 superblock | |
325 | * @n: The data block number | |
326 | * | |
327 | * Returns: The resource group, or NULL if not found | |
328 | */ | |
329 | ||
66fc061b | 330 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 331 | { |
66fc061b | 332 | struct rb_node *n, *next; |
f75bbfb4 | 333 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
334 | |
335 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
336 | n = sdp->sd_rindex_tree.rb_node; |
337 | while (n) { | |
338 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
339 | next = NULL; | |
7c9ca621 | 340 | if (blk < cur->rd_addr) |
66fc061b | 341 | next = n->rb_left; |
f75bbfb4 | 342 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
343 | next = n->rb_right; |
344 | if (next == NULL) { | |
b3b94faa | 345 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
346 | if (exact) { |
347 | if (blk < cur->rd_addr) | |
348 | return NULL; | |
349 | if (blk >= cur->rd_data0 + cur->rd_data) | |
350 | return NULL; | |
351 | } | |
7c9ca621 | 352 | return cur; |
b3b94faa | 353 | } |
66fc061b | 354 | n = next; |
b3b94faa | 355 | } |
b3b94faa DT |
356 | spin_unlock(&sdp->sd_rindex_spin); |
357 | ||
358 | return NULL; | |
359 | } | |
360 | ||
361 | /** | |
362 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
363 | * @sdp: The GFS2 superblock | |
364 | * | |
365 | * Returns: The first rgrp in the filesystem | |
366 | */ | |
367 | ||
368 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
369 | { | |
7c9ca621 BP |
370 | const struct rb_node *n; |
371 | struct gfs2_rgrpd *rgd; | |
372 | ||
8339ee54 | 373 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
374 | n = rb_first(&sdp->sd_rindex_tree); |
375 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 376 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
377 | |
378 | return rgd; | |
b3b94faa DT |
379 | } |
380 | ||
381 | /** | |
382 | * gfs2_rgrpd_get_next - get the next RG | |
383 | * @rgd: A RG | |
384 | * | |
385 | * Returns: The next rgrp | |
386 | */ | |
387 | ||
388 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
389 | { | |
7c9ca621 BP |
390 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
391 | const struct rb_node *n; | |
392 | ||
393 | spin_lock(&sdp->sd_rindex_spin); | |
394 | n = rb_next(&rgd->rd_node); | |
395 | if (n == NULL) | |
396 | n = rb_first(&sdp->sd_rindex_tree); | |
397 | ||
398 | if (unlikely(&rgd->rd_node == n)) { | |
399 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 400 | return NULL; |
7c9ca621 BP |
401 | } |
402 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
403 | spin_unlock(&sdp->sd_rindex_spin); | |
404 | return rgd; | |
b3b94faa DT |
405 | } |
406 | ||
8339ee54 SW |
407 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
408 | { | |
409 | int x; | |
410 | ||
411 | for (x = 0; x < rgd->rd_length; x++) { | |
412 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
413 | kfree(bi->bi_clone); | |
414 | bi->bi_clone = NULL; | |
415 | } | |
416 | } | |
417 | ||
418 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) | |
b3b94faa | 419 | { |
7c9ca621 | 420 | struct rb_node *n; |
b3b94faa DT |
421 | struct gfs2_rgrpd *rgd; |
422 | struct gfs2_glock *gl; | |
423 | ||
7c9ca621 BP |
424 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
425 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
426 | gl = rgd->rd_gl; |
427 | ||
7c9ca621 | 428 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
429 | |
430 | if (gl) { | |
8339ee54 | 431 | spin_lock(&gl->gl_spin); |
5c676f6d | 432 | gl->gl_object = NULL; |
8339ee54 | 433 | spin_unlock(&gl->gl_spin); |
29687a2a | 434 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
435 | gfs2_glock_put(gl); |
436 | } | |
437 | ||
8339ee54 | 438 | gfs2_free_clones(rgd); |
b3b94faa | 439 | kfree(rgd->rd_bits); |
6bdd9be6 | 440 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
441 | } |
442 | } | |
443 | ||
bb8d8a6f SW |
444 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
445 | { | |
446 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
447 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
448 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
449 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
450 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
451 | } | |
452 | ||
b3b94faa DT |
453 | /** |
454 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
455 | * @rgd: The resource group descriptor | |
456 | * | |
457 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
458 | * | |
459 | * Returns: errno | |
460 | */ | |
461 | ||
462 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
463 | { | |
464 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
465 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 466 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 467 | u32 bytes_left, bytes; |
b3b94faa DT |
468 | int x; |
469 | ||
feaa7bba SW |
470 | if (!length) |
471 | return -EINVAL; | |
472 | ||
dd894be8 | 473 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
474 | if (!rgd->rd_bits) |
475 | return -ENOMEM; | |
476 | ||
bb8d8a6f | 477 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
478 | |
479 | for (x = 0; x < length; x++) { | |
480 | bi = rgd->rd_bits + x; | |
481 | ||
60a0b8f9 | 482 | bi->bi_flags = 0; |
b3b94faa DT |
483 | /* small rgrp; bitmap stored completely in header block */ |
484 | if (length == 1) { | |
485 | bytes = bytes_left; | |
486 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
487 | bi->bi_start = 0; | |
488 | bi->bi_len = bytes; | |
489 | /* header block */ | |
490 | } else if (x == 0) { | |
491 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
492 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
493 | bi->bi_start = 0; | |
494 | bi->bi_len = bytes; | |
495 | /* last block */ | |
496 | } else if (x + 1 == length) { | |
497 | bytes = bytes_left; | |
498 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 499 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
500 | bi->bi_len = bytes; |
501 | /* other blocks */ | |
502 | } else { | |
568f4c96 SW |
503 | bytes = sdp->sd_sb.sb_bsize - |
504 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 505 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 506 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
507 | bi->bi_len = bytes; |
508 | } | |
509 | ||
510 | bytes_left -= bytes; | |
511 | } | |
512 | ||
513 | if (bytes_left) { | |
514 | gfs2_consist_rgrpd(rgd); | |
515 | return -EIO; | |
516 | } | |
517 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 518 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 519 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 520 | gfs2_rindex_print(rgd); |
b3b94faa DT |
521 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
522 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
523 | } | |
524 | return -EIO; | |
525 | } | |
526 | ||
527 | return 0; | |
528 | } | |
529 | ||
7ae8fa84 RP |
530 | /** |
531 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
532 | * | |
533 | */ | |
534 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
535 | { | |
536 | u64 total_data = 0; | |
537 | struct inode *inode = sdp->sd_rindex; | |
538 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 RP |
539 | char buf[sizeof(struct gfs2_rindex)]; |
540 | struct file_ra_state ra_state; | |
541 | int error, rgrps; | |
542 | ||
7ae8fa84 RP |
543 | file_ra_state_init(&ra_state, inode->i_mapping); |
544 | for (rgrps = 0;; rgrps++) { | |
545 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
546 | ||
bcd7278d | 547 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 RP |
548 | break; |
549 | error = gfs2_internal_read(ip, &ra_state, buf, &pos, | |
550 | sizeof(struct gfs2_rindex)); | |
551 | if (error != sizeof(struct gfs2_rindex)) | |
552 | break; | |
bb8d8a6f | 553 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 554 | } |
7ae8fa84 RP |
555 | return total_data; |
556 | } | |
557 | ||
6aad1c3d | 558 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
559 | { |
560 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
561 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
562 | ||
563 | /* Figure out where to put new node */ | |
564 | while (*newn) { | |
565 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
566 | rd_node); | |
567 | ||
568 | parent = *newn; | |
569 | if (rgd->rd_addr < cur->rd_addr) | |
570 | newn = &((*newn)->rb_left); | |
571 | else if (rgd->rd_addr > cur->rd_addr) | |
572 | newn = &((*newn)->rb_right); | |
573 | else | |
6aad1c3d | 574 | return -EEXIST; |
7c9ca621 BP |
575 | } |
576 | ||
577 | rb_link_node(&rgd->rd_node, parent, newn); | |
578 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
579 | sdp->sd_rgrps++; |
580 | return 0; | |
7c9ca621 BP |
581 | } |
582 | ||
b3b94faa | 583 | /** |
6c53267f | 584 | * read_rindex_entry - Pull in a new resource index entry from the disk |
b3b94faa DT |
585 | * @gl: The glock covering the rindex inode |
586 | * | |
8339ee54 | 587 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
588 | */ |
589 | ||
590 | static int read_rindex_entry(struct gfs2_inode *ip, | |
591 | struct file_ra_state *ra_state) | |
592 | { | |
593 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
594 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
8339ee54 | 595 | struct gfs2_rindex buf; |
6c53267f RP |
596 | int error; |
597 | struct gfs2_rgrpd *rgd; | |
598 | ||
8339ee54 SW |
599 | if (pos >= i_size_read(&ip->i_inode)) |
600 | return 1; | |
601 | ||
602 | error = gfs2_internal_read(ip, ra_state, (char *)&buf, &pos, | |
6c53267f | 603 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
604 | |
605 | if (error != sizeof(struct gfs2_rindex)) | |
606 | return (error == 0) ? 1 : error; | |
6c53267f | 607 | |
6bdd9be6 | 608 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
609 | error = -ENOMEM; |
610 | if (!rgd) | |
611 | return error; | |
612 | ||
6c53267f | 613 | rgd->rd_sbd = sdp; |
8339ee54 SW |
614 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
615 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
616 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
617 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
618 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
7c9ca621 | 619 | |
6c53267f RP |
620 | error = compute_bitstructs(rgd); |
621 | if (error) | |
8339ee54 | 622 | goto fail; |
6c53267f | 623 | |
bb8d8a6f | 624 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
625 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
626 | if (error) | |
8339ee54 | 627 | goto fail; |
6c53267f RP |
628 | |
629 | rgd->rd_gl->gl_object = rgd; | |
cf45b752 | 630 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
7c9ca621 BP |
631 | if (rgd->rd_data > sdp->sd_max_rg_data) |
632 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 633 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 634 | error = rgd_insert(rgd); |
8339ee54 | 635 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
636 | if (!error) |
637 | return 0; | |
638 | ||
639 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 640 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
641 | |
642 | fail: | |
643 | kfree(rgd->rd_bits); | |
644 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
645 | return error; |
646 | } | |
647 | ||
648 | /** | |
649 | * gfs2_ri_update - Pull in a new resource index from the disk | |
650 | * @ip: pointer to the rindex inode | |
651 | * | |
b3b94faa DT |
652 | * Returns: 0 on successful update, error code otherwise |
653 | */ | |
654 | ||
8339ee54 | 655 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 656 | { |
feaa7bba SW |
657 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
658 | struct inode *inode = &ip->i_inode; | |
f42faf4f | 659 | struct file_ra_state ra_state; |
b3b94faa DT |
660 | int error; |
661 | ||
f42faf4f | 662 | file_ra_state_init(&ra_state, inode->i_mapping); |
8339ee54 | 663 | do { |
6c53267f | 664 | error = read_rindex_entry(ip, &ra_state); |
8339ee54 SW |
665 | } while (error == 0); |
666 | ||
667 | if (error < 0) | |
668 | return error; | |
b3b94faa | 669 | |
cf45b752 | 670 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
671 | return 0; |
672 | } | |
b3b94faa | 673 | |
b3b94faa | 674 | /** |
8339ee54 | 675 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 676 | * @sdp: The GFS2 superblock |
b3b94faa DT |
677 | * |
678 | * We grab a lock on the rindex inode to make sure that it doesn't | |
679 | * change whilst we are performing an operation. We keep this lock | |
680 | * for quite long periods of time compared to other locks. This | |
681 | * doesn't matter, since it is shared and it is very, very rarely | |
682 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
683 | * | |
684 | * This makes sure that we're using the latest copy of the resource index | |
685 | * special file, which might have been updated if someone expanded the | |
686 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
687 | * | |
8339ee54 | 688 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
689 | */ |
690 | ||
8339ee54 | 691 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 692 | { |
feaa7bba | 693 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 694 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
695 | struct gfs2_holder ri_gh; |
696 | int error = 0; | |
a365fbf3 | 697 | int unlock_required = 0; |
b3b94faa DT |
698 | |
699 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 700 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
701 | if (!gfs2_glock_is_locked_by_me(gl)) { |
702 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
703 | if (error) | |
6aad1c3d | 704 | return error; |
a365fbf3 SW |
705 | unlock_required = 1; |
706 | } | |
8339ee54 | 707 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 708 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
709 | if (unlock_required) |
710 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
711 | } |
712 | ||
713 | return error; | |
714 | } | |
715 | ||
42d52e38 | 716 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
717 | { |
718 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 719 | u32 rg_flags; |
bb8d8a6f | 720 | |
42d52e38 | 721 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 722 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
723 | rgd->rd_flags &= GFS2_RDF_MASK; |
724 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 725 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 726 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 727 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
728 | } |
729 | ||
42d52e38 | 730 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
731 | { |
732 | struct gfs2_rgrp *str = buf; | |
733 | ||
09010978 | 734 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 735 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 736 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 737 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 738 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
739 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
740 | } | |
741 | ||
b3b94faa | 742 | /** |
7c9ca621 | 743 | * gfs2_rgrp_go_lock - Read in a RG's header and bitmaps |
b3b94faa DT |
744 | * @rgd: the struct gfs2_rgrpd describing the RG to read in |
745 | * | |
746 | * Read in all of a Resource Group's header and bitmap blocks. | |
747 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
748 | * | |
749 | * Returns: errno | |
750 | */ | |
751 | ||
7c9ca621 | 752 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) |
b3b94faa | 753 | { |
7c9ca621 | 754 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
b3b94faa DT |
755 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
756 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 757 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
758 | struct gfs2_bitmap *bi; |
759 | unsigned int x, y; | |
760 | int error; | |
761 | ||
b3b94faa DT |
762 | for (x = 0; x < length; x++) { |
763 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 764 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
765 | if (error) |
766 | goto fail; | |
767 | } | |
768 | ||
769 | for (y = length; y--;) { | |
770 | bi = rgd->rd_bits + y; | |
7276b3b0 | 771 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
772 | if (error) |
773 | goto fail; | |
feaa7bba | 774 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
775 | GFS2_METATYPE_RG)) { |
776 | error = -EIO; | |
777 | goto fail; | |
778 | } | |
779 | } | |
780 | ||
cf45b752 | 781 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
782 | for (x = 0; x < length; x++) |
783 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 784 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 785 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 786 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa DT |
787 | } |
788 | ||
b3b94faa DT |
789 | return 0; |
790 | ||
feaa7bba | 791 | fail: |
b3b94faa DT |
792 | while (x--) { |
793 | bi = rgd->rd_bits + x; | |
794 | brelse(bi->bi_bh); | |
795 | bi->bi_bh = NULL; | |
796 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
797 | } | |
b3b94faa DT |
798 | |
799 | return error; | |
800 | } | |
801 | ||
b3b94faa | 802 | /** |
7c9ca621 | 803 | * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
b3b94faa DT |
804 | * @rgd: the struct gfs2_rgrpd describing the RG to read in |
805 | * | |
806 | */ | |
807 | ||
7c9ca621 | 808 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) |
b3b94faa | 809 | { |
7c9ca621 | 810 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
bb8d8a6f | 811 | int x, length = rgd->rd_length; |
b3b94faa | 812 | |
b3b94faa DT |
813 | for (x = 0; x < length; x++) { |
814 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
b3b94faa DT |
815 | brelse(bi->bi_bh); |
816 | bi->bi_bh = NULL; | |
817 | } | |
818 | ||
b3b94faa DT |
819 | } |
820 | ||
66fc061b | 821 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 822 | struct buffer_head *bh, |
66fc061b | 823 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
824 | { |
825 | struct super_block *sb = sdp->sd_vfs; | |
826 | struct block_device *bdev = sb->s_bdev; | |
827 | const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / | |
e1defc4f | 828 | bdev_logical_block_size(sb->s_bdev); |
f15ab561 | 829 | u64 blk; |
64d576ba | 830 | sector_t start = 0; |
f15ab561 SW |
831 | sector_t nr_sects = 0; |
832 | int rv; | |
833 | unsigned int x; | |
66fc061b SW |
834 | u32 trimmed = 0; |
835 | u8 diff; | |
f15ab561 SW |
836 | |
837 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
838 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
839 | clone += bi->bi_offset; | |
840 | clone += x; | |
841 | if (bh) { | |
842 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
843 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
844 | } else { | |
845 | diff = ~(*clone | (*clone >> 1)); | |
846 | } | |
f15ab561 SW |
847 | diff &= 0x55; |
848 | if (diff == 0) | |
849 | continue; | |
850 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
851 | blk *= sects_per_blk; /* convert to sectors */ | |
852 | while(diff) { | |
853 | if (diff & 1) { | |
854 | if (nr_sects == 0) | |
855 | goto start_new_extent; | |
856 | if ((start + nr_sects) != blk) { | |
66fc061b SW |
857 | if (nr_sects >= minlen) { |
858 | rv = blkdev_issue_discard(bdev, | |
859 | start, nr_sects, | |
860 | GFP_NOFS, 0); | |
861 | if (rv) | |
862 | goto fail; | |
863 | trimmed += nr_sects; | |
864 | } | |
f15ab561 SW |
865 | nr_sects = 0; |
866 | start_new_extent: | |
867 | start = blk; | |
868 | } | |
869 | nr_sects += sects_per_blk; | |
870 | } | |
871 | diff >>= 2; | |
872 | blk += sects_per_blk; | |
873 | } | |
874 | } | |
66fc061b | 875 | if (nr_sects >= minlen) { |
dd3932ed | 876 | rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0); |
f15ab561 SW |
877 | if (rv) |
878 | goto fail; | |
66fc061b | 879 | trimmed += nr_sects; |
f15ab561 | 880 | } |
66fc061b SW |
881 | if (ptrimmed) |
882 | *ptrimmed = trimmed; | |
883 | return 0; | |
884 | ||
f15ab561 | 885 | fail: |
66fc061b SW |
886 | if (sdp->sd_args.ar_discard) |
887 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 888 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
889 | return -EIO; |
890 | } | |
891 | ||
892 | /** | |
893 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
894 | * @filp: Any file on the filesystem | |
895 | * @argp: Pointer to the arguments (also used to pass result) | |
896 | * | |
897 | * Returns: 0 on success, otherwise error code | |
898 | */ | |
899 | ||
900 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
901 | { | |
902 | struct inode *inode = filp->f_dentry->d_inode; | |
903 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
904 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
905 | struct buffer_head *bh; | |
906 | struct gfs2_rgrpd *rgd; | |
907 | struct gfs2_rgrpd *rgd_end; | |
908 | struct gfs2_holder gh; | |
909 | struct fstrim_range r; | |
910 | int ret = 0; | |
911 | u64 amt; | |
912 | u64 trimmed = 0; | |
913 | unsigned int x; | |
914 | ||
915 | if (!capable(CAP_SYS_ADMIN)) | |
916 | return -EPERM; | |
917 | ||
918 | if (!blk_queue_discard(q)) | |
919 | return -EOPNOTSUPP; | |
920 | ||
66fc061b SW |
921 | if (argp == NULL) { |
922 | r.start = 0; | |
923 | r.len = ULLONG_MAX; | |
924 | r.minlen = 0; | |
925 | } else if (copy_from_user(&r, argp, sizeof(r))) | |
926 | return -EFAULT; | |
927 | ||
5e2f7d61 BP |
928 | ret = gfs2_rindex_update(sdp); |
929 | if (ret) | |
930 | return ret; | |
931 | ||
66fc061b SW |
932 | rgd = gfs2_blk2rgrpd(sdp, r.start, 0); |
933 | rgd_end = gfs2_blk2rgrpd(sdp, r.start + r.len, 0); | |
934 | ||
935 | while (1) { | |
936 | ||
937 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
938 | if (ret) | |
939 | goto out; | |
940 | ||
941 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
942 | /* Trim each bitmap in the rgrp */ | |
943 | for (x = 0; x < rgd->rd_length; x++) { | |
944 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
945 | ret = gfs2_rgrp_send_discards(sdp, rgd->rd_data0, NULL, bi, r.minlen, &amt); | |
946 | if (ret) { | |
947 | gfs2_glock_dq_uninit(&gh); | |
948 | goto out; | |
949 | } | |
950 | trimmed += amt; | |
951 | } | |
952 | ||
953 | /* Mark rgrp as having been trimmed */ | |
954 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
955 | if (ret == 0) { | |
956 | bh = rgd->rd_bits[0].bi_bh; | |
957 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
958 | gfs2_trans_add_bh(rgd->rd_gl, bh, 1); | |
959 | gfs2_rgrp_out(rgd, bh->b_data); | |
960 | gfs2_trans_end(sdp); | |
961 | } | |
962 | } | |
963 | gfs2_glock_dq_uninit(&gh); | |
964 | ||
965 | if (rgd == rgd_end) | |
966 | break; | |
967 | ||
968 | rgd = gfs2_rgrpd_get_next(rgd); | |
969 | } | |
970 | ||
971 | out: | |
972 | r.len = trimmed << 9; | |
973 | if (argp && copy_to_user(argp, &r, sizeof(r))) | |
974 | return -EFAULT; | |
975 | ||
976 | return ret; | |
f15ab561 SW |
977 | } |
978 | ||
b3b94faa | 979 | /** |
564e12b1 | 980 | * gfs2_qadata_get - get the struct gfs2_qadata structure for an inode |
b3b94faa DT |
981 | * @ip: the incore GFS2 inode structure |
982 | * | |
564e12b1 | 983 | * Returns: the struct gfs2_qadata |
b3b94faa DT |
984 | */ |
985 | ||
564e12b1 | 986 | struct gfs2_qadata *gfs2_qadata_get(struct gfs2_inode *ip) |
b3b94faa | 987 | { |
8339ee54 SW |
988 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
989 | int error; | |
564e12b1 BP |
990 | BUG_ON(ip->i_qadata != NULL); |
991 | ip->i_qadata = kzalloc(sizeof(struct gfs2_qadata), GFP_NOFS); | |
8339ee54 SW |
992 | error = gfs2_rindex_update(sdp); |
993 | if (error) | |
994 | fs_warn(sdp, "rindex update returns %d\n", error); | |
564e12b1 BP |
995 | return ip->i_qadata; |
996 | } | |
997 | ||
998 | /** | |
999 | * gfs2_blkrsv_get - get the struct gfs2_blkreserv structure for an inode | |
1000 | * @ip: the incore GFS2 inode structure | |
1001 | * | |
1002 | * Returns: the struct gfs2_qadata | |
1003 | */ | |
1004 | ||
36f5580b | 1005 | static int gfs2_blkrsv_get(struct gfs2_inode *ip) |
564e12b1 BP |
1006 | { |
1007 | BUG_ON(ip->i_res != NULL); | |
36f5580b BP |
1008 | ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); |
1009 | if (!ip->i_res) | |
1010 | return -ENOMEM; | |
1011 | return 0; | |
b3b94faa DT |
1012 | } |
1013 | ||
b3b94faa DT |
1014 | /** |
1015 | * try_rgrp_fit - See if a given reservation will fit in a given RG | |
1016 | * @rgd: the RG data | |
54335b1f | 1017 | * @ip: the inode |
b3b94faa DT |
1018 | * |
1019 | * If there's room for the requested blocks to be allocated from the RG: | |
b3b94faa DT |
1020 | * |
1021 | * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) | |
1022 | */ | |
1023 | ||
54335b1f | 1024 | static int try_rgrp_fit(const struct gfs2_rgrpd *rgd, const struct gfs2_inode *ip) |
b3b94faa | 1025 | { |
564e12b1 | 1026 | const struct gfs2_blkreserv *rs = ip->i_res; |
54335b1f | 1027 | |
09010978 | 1028 | if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) |
a43a4906 | 1029 | return 0; |
564e12b1 | 1030 | if (rgd->rd_free_clone >= rs->rs_requested) |
7c9ca621 | 1031 | return 1; |
7c9ca621 | 1032 | return 0; |
b3b94faa DT |
1033 | } |
1034 | ||
b3e47ca0 BP |
1035 | static inline u32 gfs2_bi2rgd_blk(struct gfs2_bitmap *bi, u32 blk) |
1036 | { | |
1037 | return (bi->bi_start * GFS2_NBBY) + blk; | |
1038 | } | |
1039 | ||
c8cdf479 SW |
1040 | /** |
1041 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1042 | * @rgd: The rgrp | |
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 | |
1107 | * @rgp: the chosen and locked rgrp | |
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 DT |
1175 | * @ip: the inode to reserve space for |
1176 | * | |
1177 | * Returns: errno | |
1178 | */ | |
1179 | ||
564e12b1 | 1180 | int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested) |
b3b94faa | 1181 | { |
feaa7bba | 1182 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
564e12b1 | 1183 | struct gfs2_blkreserv *rs; |
36f5580b | 1184 | int error; |
044b9414 SW |
1185 | u64 last_unlinked = NO_BLOCK; |
1186 | int tries = 0; | |
b3b94faa | 1187 | |
36f5580b BP |
1188 | error = gfs2_blkrsv_get(ip); |
1189 | if (error) | |
1190 | return error; | |
564e12b1 | 1191 | |
36f5580b | 1192 | rs = ip->i_res; |
564e12b1 BP |
1193 | rs->rs_requested = requested; |
1194 | if (gfs2_assert_warn(sdp, requested)) { | |
1195 | error = -EINVAL; | |
1196 | goto out; | |
1197 | } | |
b3b94faa | 1198 | |
044b9414 SW |
1199 | do { |
1200 | error = get_local_rgrp(ip, &last_unlinked); | |
54335b1f SW |
1201 | if (error != -ENOSPC) |
1202 | break; | |
1203 | /* Check that fs hasn't grown if writing to rindex */ | |
1204 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
1205 | error = gfs2_ri_update(ip); | |
1206 | if (error) | |
1207 | break; | |
1208 | continue; | |
0489b3f5 | 1209 | } |
54335b1f SW |
1210 | /* Flushing the log may release space */ |
1211 | gfs2_log_flush(sdp, NULL); | |
1212 | } while (tries++ < 3); | |
b3b94faa | 1213 | |
564e12b1 BP |
1214 | out: |
1215 | if (error) | |
1216 | gfs2_blkrsv_put(ip); | |
9ae32429 | 1217 | return error; |
b3b94faa DT |
1218 | } |
1219 | ||
1220 | /** | |
1221 | * gfs2_inplace_release - release an inplace reservation | |
1222 | * @ip: the inode the reservation was taken out on | |
1223 | * | |
1224 | * Release a reservation made by gfs2_inplace_reserve(). | |
1225 | */ | |
1226 | ||
1227 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1228 | { | |
564e12b1 | 1229 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 1230 | |
564e12b1 BP |
1231 | if (rs->rs_rgd_gh.gh_gl) |
1232 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
49528b4e | 1233 | gfs2_blkrsv_put(ip); |
b3b94faa DT |
1234 | } |
1235 | ||
1236 | /** | |
1237 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1238 | * @rgd: the resource group holding the block | |
1239 | * @block: the block number | |
1240 | * | |
1241 | * Returns: The block type (GFS2_BLKST_*) | |
1242 | */ | |
1243 | ||
acf7e244 | 1244 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa DT |
1245 | { |
1246 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1247 | u32 length, rgrp_block, buf_block; |
b3b94faa DT |
1248 | unsigned int buf; |
1249 | unsigned char type; | |
1250 | ||
bb8d8a6f SW |
1251 | length = rgd->rd_length; |
1252 | rgrp_block = block - rgd->rd_data0; | |
b3b94faa DT |
1253 | |
1254 | for (buf = 0; buf < length; buf++) { | |
1255 | bi = rgd->rd_bits + buf; | |
1256 | if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1257 | break; | |
1258 | } | |
1259 | ||
1260 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1261 | buf_block = rgrp_block - bi->bi_start * GFS2_NBBY; | |
1262 | ||
feaa7bba | 1263 | type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset, |
b3b94faa DT |
1264 | bi->bi_len, buf_block); |
1265 | ||
1266 | return type; | |
1267 | } | |
1268 | ||
1269 | /** | |
6a8099ed | 1270 | * rgblk_search - find a block in @state |
b3b94faa DT |
1271 | * @rgd: the resource group descriptor |
1272 | * @goal: the goal block within the RG (start here to search for avail block) | |
6a8099ed | 1273 | * @state: GFS2_BLKST_XXX the before-allocation state to find |
6e87ed0f | 1274 | * @dinode: TRUE if the first block we allocate is for a dinode |
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 | ||
cd915493 | 1288 | static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, |
6a8099ed | 1289 | unsigned char state, |
b3e47ca0 | 1290 | struct gfs2_bitmap **rbi) |
b3b94faa DT |
1291 | { |
1292 | struct gfs2_bitmap *bi = NULL; | |
b45e41d7 | 1293 | const u32 length = rgd->rd_length; |
60a0b8f9 | 1294 | u32 blk = BFITNOENT; |
b3b94faa | 1295 | unsigned int buf, x; |
60a0b8f9 | 1296 | const u8 *buffer = NULL; |
b3b94faa | 1297 | |
b3e47ca0 | 1298 | *rbi = NULL; |
b3b94faa DT |
1299 | /* Find bitmap block that contains bits for goal block */ |
1300 | for (buf = 0; buf < length; buf++) { | |
1301 | bi = rgd->rd_bits + buf; | |
60a0b8f9 SW |
1302 | /* Convert scope of "goal" from rgrp-wide to within found bit block */ |
1303 | if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { | |
1304 | goal -= bi->bi_start * GFS2_NBBY; | |
1305 | goto do_search; | |
1306 | } | |
b3b94faa | 1307 | } |
60a0b8f9 SW |
1308 | buf = 0; |
1309 | goal = 0; | |
b3b94faa | 1310 | |
60a0b8f9 | 1311 | do_search: |
b3b94faa DT |
1312 | /* Search (up to entire) bitmap in this rgrp for allocatable block. |
1313 | "x <= length", instead of "x < length", because we typically start | |
1314 | the search in the middle of a bit block, but if we can't find an | |
1315 | allocatable block anywhere else, we want to be able wrap around and | |
1316 | search in the first part of our first-searched bit block. */ | |
1317 | for (x = 0; x <= length; x++) { | |
60a0b8f9 SW |
1318 | bi = rgd->rd_bits + buf; |
1319 | ||
1320 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
6a8099ed | 1321 | (state == GFS2_BLKST_FREE)) |
60a0b8f9 SW |
1322 | goto skip; |
1323 | ||
5f3eae75 BP |
1324 | /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone |
1325 | bitmaps, so we must search the originals for that. */ | |
b45e41d7 | 1326 | buffer = bi->bi_bh->b_data + bi->bi_offset; |
7c9ca621 | 1327 | WARN_ON(!buffer_uptodate(bi->bi_bh)); |
6a8099ed | 1328 | if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
110acf38 SW |
1329 | buffer = bi->bi_clone + bi->bi_offset; |
1330 | ||
6a8099ed | 1331 | blk = gfs2_bitfit(buffer, bi->bi_len, goal, state); |
b3b94faa DT |
1332 | if (blk != BFITNOENT) |
1333 | break; | |
1334 | ||
6a8099ed | 1335 | if ((goal == 0) && (state == GFS2_BLKST_FREE)) |
60a0b8f9 SW |
1336 | set_bit(GBF_FULL, &bi->bi_flags); |
1337 | ||
b3b94faa | 1338 | /* Try next bitmap block (wrap back to rgrp header if at end) */ |
60a0b8f9 SW |
1339 | skip: |
1340 | buf++; | |
1341 | buf %= length; | |
b3b94faa DT |
1342 | goal = 0; |
1343 | } | |
1344 | ||
b3e47ca0 BP |
1345 | if (blk != BFITNOENT) |
1346 | *rbi = bi; | |
7c9ca621 | 1347 | |
b3e47ca0 BP |
1348 | return blk; |
1349 | } | |
60a0b8f9 | 1350 | |
b3e47ca0 BP |
1351 | /** |
1352 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
1353 | * @rgd: the resource group descriptor | |
1354 | * @bi: the bitmap within the rgrp | |
1355 | * @blk: the block within the bitmap | |
1356 | * @dinode: TRUE if the first block we allocate is for a dinode | |
1357 | * @n: The extent length | |
1358 | * | |
1359 | * Add the found bitmap buffer to the transaction. | |
1360 | * Set the found bits to @new_state to change block's allocation state. | |
1361 | * Returns: starting block number of the extent (fs scope) | |
1362 | */ | |
1363 | static u64 gfs2_alloc_extent(struct gfs2_rgrpd *rgd, struct gfs2_bitmap *bi, | |
1364 | u32 blk, bool dinode, unsigned int *n) | |
1365 | { | |
1366 | const unsigned int elen = *n; | |
1367 | u32 goal; | |
1368 | const u8 *buffer = NULL; | |
1369 | ||
6a8099ed | 1370 | *n = 0; |
b3e47ca0 | 1371 | buffer = bi->bi_bh->b_data + bi->bi_offset; |
60a0b8f9 SW |
1372 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
1373 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, | |
6e87ed0f | 1374 | bi, blk, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6a8099ed | 1375 | (*n)++; |
60a0b8f9 SW |
1376 | goal = blk; |
1377 | while (*n < elen) { | |
1378 | goal++; | |
1379 | if (goal >= (bi->bi_len * GFS2_NBBY)) | |
1380 | break; | |
1381 | if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) != | |
1382 | GFS2_BLKST_FREE) | |
1383 | break; | |
b45e41d7 | 1384 | gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, |
6e87ed0f | 1385 | bi, goal, GFS2_BLKST_USED); |
60a0b8f9 | 1386 | (*n)++; |
c8cdf479 | 1387 | } |
b3e47ca0 | 1388 | blk = gfs2_bi2rgd_blk(bi, blk); |
6a8099ed | 1389 | rgd->rd_last_alloc = blk + *n - 1; |
b3e47ca0 | 1390 | return rgd->rd_data0 + blk; |
b3b94faa DT |
1391 | } |
1392 | ||
1393 | /** | |
1394 | * rgblk_free - Change alloc state of given block(s) | |
1395 | * @sdp: the filesystem | |
1396 | * @bstart: the start of a run of blocks to free | |
1397 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1398 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1399 | * | |
1400 | * Returns: Resource group containing the block(s) | |
1401 | */ | |
1402 | ||
cd915493 SW |
1403 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
1404 | u32 blen, unsigned char new_state) | |
b3b94faa DT |
1405 | { |
1406 | struct gfs2_rgrpd *rgd; | |
1407 | struct gfs2_bitmap *bi = NULL; | |
cd915493 | 1408 | u32 length, rgrp_blk, buf_blk; |
b3b94faa DT |
1409 | unsigned int buf; |
1410 | ||
66fc061b | 1411 | rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
b3b94faa DT |
1412 | if (!rgd) { |
1413 | if (gfs2_consist(sdp)) | |
382066da | 1414 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
1415 | return NULL; |
1416 | } | |
1417 | ||
bb8d8a6f | 1418 | length = rgd->rd_length; |
b3b94faa | 1419 | |
bb8d8a6f | 1420 | rgrp_blk = bstart - rgd->rd_data0; |
b3b94faa DT |
1421 | |
1422 | while (blen--) { | |
1423 | for (buf = 0; buf < length; buf++) { | |
1424 | bi = rgd->rd_bits + buf; | |
1425 | if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY) | |
1426 | break; | |
1427 | } | |
1428 | ||
1429 | gfs2_assert(rgd->rd_sbd, buf < length); | |
1430 | ||
1431 | buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY; | |
1432 | rgrp_blk++; | |
1433 | ||
1434 | if (!bi->bi_clone) { | |
1435 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
dd894be8 | 1436 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1437 | memcpy(bi->bi_clone + bi->bi_offset, |
1438 | bi->bi_bh->b_data + bi->bi_offset, | |
1439 | bi->bi_len); | |
1440 | } | |
d4e9c4c3 | 1441 | gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); |
b45e41d7 | 1442 | gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset, |
95c8e17f | 1443 | bi, buf_blk, new_state); |
b3b94faa DT |
1444 | } |
1445 | ||
1446 | return rgd; | |
1447 | } | |
1448 | ||
1449 | /** | |
09010978 SW |
1450 | * gfs2_rgrp_dump - print out an rgrp |
1451 | * @seq: The iterator | |
1452 | * @gl: The glock in question | |
1453 | * | |
1454 | */ | |
1455 | ||
1456 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
1457 | { | |
1458 | const struct gfs2_rgrpd *rgd = gl->gl_object; | |
1459 | if (rgd == NULL) | |
1460 | return 0; | |
1461 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n", | |
1462 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, | |
1463 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes); | |
1464 | return 0; | |
1465 | } | |
1466 | ||
6050b9c7 SW |
1467 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
1468 | { | |
1469 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1470 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 1471 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
1472 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
1473 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
1474 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
1475 | } | |
1476 | ||
09010978 | 1477 | /** |
6e87ed0f | 1478 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 1479 | * @ip: the inode to allocate the block for |
09010978 | 1480 | * @bn: Used to return the starting block number |
6a8099ed | 1481 | * @ndata: requested number of blocks/extent length (value/result) |
6e87ed0f | 1482 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 1483 | * @generation: the generation number of the inode |
b3b94faa | 1484 | * |
09010978 | 1485 | * Returns: 0 or error |
b3b94faa DT |
1486 | */ |
1487 | ||
6a8099ed | 1488 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 1489 | bool dinode, u64 *generation) |
b3b94faa | 1490 | { |
feaa7bba | 1491 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 1492 | struct buffer_head *dibh; |
9a3f236d | 1493 | struct gfs2_rgrpd *rgd; |
6a8099ed SW |
1494 | unsigned int ndata; |
1495 | u32 goal, blk; /* block, within the rgrp scope */ | |
3c5d785a | 1496 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 1497 | int error; |
b3e47ca0 | 1498 | struct gfs2_bitmap *bi; |
b3b94faa | 1499 | |
9a3f236d SW |
1500 | /* Only happens if there is a bug in gfs2, return something distinctive |
1501 | * to ensure that it is noticed. | |
1502 | */ | |
564e12b1 | 1503 | if (ip->i_res == NULL) |
9a3f236d SW |
1504 | return -ECANCELED; |
1505 | ||
54335b1f | 1506 | rgd = ip->i_rgd; |
9a3f236d | 1507 | |
3c5d785a | 1508 | if (!dinode && rgrp_contains_block(rgd, ip->i_goal)) |
ce276b06 | 1509 | goal = ip->i_goal - rgd->rd_data0; |
b3b94faa | 1510 | else |
ac576cc5 | 1511 | goal = rgd->rd_last_alloc; |
b3b94faa | 1512 | |
6a8099ed | 1513 | blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, &bi); |
09010978 SW |
1514 | |
1515 | /* Since all blocks are reserved in advance, this shouldn't happen */ | |
1516 | if (blk == BFITNOENT) | |
1517 | goto rgrp_error; | |
b3b94faa | 1518 | |
6a8099ed SW |
1519 | block = gfs2_alloc_extent(rgd, bi, blk, dinode, nblocks); |
1520 | ndata = *nblocks; | |
1521 | if (dinode) | |
1522 | ndata--; | |
b3e47ca0 | 1523 | |
3c5d785a | 1524 | if (!dinode) { |
6a8099ed | 1525 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
1526 | error = gfs2_meta_inode_buffer(ip, &dibh); |
1527 | if (error == 0) { | |
1528 | struct gfs2_dinode *di = | |
1529 | (struct gfs2_dinode *)dibh->b_data; | |
1530 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
1531 | di->di_goal_meta = di->di_goal_data = | |
1532 | cpu_to_be64(ip->i_goal); | |
1533 | brelse(dibh); | |
1534 | } | |
d9ba7615 | 1535 | } |
6a8099ed | 1536 | if (rgd->rd_free < *nblocks) |
09010978 SW |
1537 | goto rgrp_error; |
1538 | ||
6a8099ed | 1539 | rgd->rd_free -= *nblocks; |
3c5d785a BP |
1540 | if (dinode) { |
1541 | rgd->rd_dinodes++; | |
1542 | *generation = rgd->rd_igeneration++; | |
1543 | if (*generation == 0) | |
1544 | *generation = rgd->rd_igeneration++; | |
1545 | } | |
b3b94faa | 1546 | |
d4e9c4c3 | 1547 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1548 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 1549 | |
6a8099ed | 1550 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a BP |
1551 | if (dinode) |
1552 | gfs2_trans_add_unrevoke(sdp, block, 1); | |
6a8099ed SW |
1553 | |
1554 | /* | |
1555 | * This needs reviewing to see why we cannot do the quota change | |
1556 | * at this point in the dinode case. | |
1557 | */ | |
1558 | if (ndata) | |
1559 | gfs2_quota_change(ip, ndata, ip->i_inode.i_uid, | |
3c5d785a | 1560 | ip->i_inode.i_gid); |
b3b94faa | 1561 | |
6a8099ed SW |
1562 | rgd->rd_free_clone -= *nblocks; |
1563 | trace_gfs2_block_alloc(ip, block, *nblocks, | |
6e87ed0f | 1564 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
1565 | *bn = block; |
1566 | return 0; | |
1567 | ||
1568 | rgrp_error: | |
1569 | gfs2_rgrp_error(rgd); | |
1570 | return -EIO; | |
b3b94faa DT |
1571 | } |
1572 | ||
1573 | /** | |
46fcb2ed | 1574 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
1575 | * @ip: the inode these blocks are being freed from |
1576 | * @bstart: first block of a run of contiguous blocks | |
1577 | * @blen: the length of the block run | |
46fcb2ed | 1578 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
1579 | * |
1580 | */ | |
1581 | ||
46fcb2ed | 1582 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 1583 | { |
feaa7bba | 1584 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1585 | struct gfs2_rgrpd *rgd; |
1586 | ||
1587 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
1588 | if (!rgd) | |
1589 | return; | |
63997775 | 1590 | trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 1591 | rgd->rd_free += blen; |
66fc061b | 1592 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
d4e9c4c3 | 1593 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1594 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 1595 | |
6d3117b4 | 1596 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 1597 | if (meta || ip->i_depth) |
6d3117b4 | 1598 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 1599 | } |
b3b94faa | 1600 | |
4c16c36a BP |
1601 | /** |
1602 | * gfs2_free_meta - free a contiguous run of data block(s) | |
1603 | * @ip: the inode these blocks are being freed from | |
1604 | * @bstart: first block of a run of contiguous blocks | |
1605 | * @blen: the length of the block run | |
1606 | * | |
1607 | */ | |
1608 | ||
1609 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
1610 | { | |
1611 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1612 | ||
46fcb2ed | 1613 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 1614 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 1615 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
1616 | } |
1617 | ||
feaa7bba SW |
1618 | void gfs2_unlink_di(struct inode *inode) |
1619 | { | |
1620 | struct gfs2_inode *ip = GFS2_I(inode); | |
1621 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1622 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 1623 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
1624 | |
1625 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
1626 | if (!rgd) | |
1627 | return; | |
63997775 | 1628 | trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED); |
feaa7bba | 1629 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1630 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
feaa7bba SW |
1631 | } |
1632 | ||
cd915493 | 1633 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
1634 | { |
1635 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1636 | struct gfs2_rgrpd *tmp_rgd; | |
1637 | ||
1638 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
1639 | if (!tmp_rgd) | |
1640 | return; | |
1641 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
1642 | ||
73f74948 | 1643 | if (!rgd->rd_dinodes) |
b3b94faa | 1644 | gfs2_consist_rgrpd(rgd); |
73f74948 | 1645 | rgd->rd_dinodes--; |
cfc8b549 | 1646 | rgd->rd_free++; |
b3b94faa | 1647 | |
d4e9c4c3 | 1648 | gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); |
42d52e38 | 1649 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa DT |
1650 | |
1651 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
1652 | } |
1653 | ||
b3b94faa DT |
1654 | |
1655 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
1656 | { | |
dbb7cae2 | 1657 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
63997775 | 1658 | trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 1659 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 1660 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
1661 | } |
1662 | ||
acf7e244 SW |
1663 | /** |
1664 | * gfs2_check_blk_type - Check the type of a block | |
1665 | * @sdp: The superblock | |
1666 | * @no_addr: The block number to check | |
1667 | * @type: The block type we are looking for | |
1668 | * | |
1669 | * Returns: 0 if the block type matches the expected type | |
1670 | * -ESTALE if it doesn't match | |
1671 | * or -ve errno if something went wrong while checking | |
1672 | */ | |
1673 | ||
1674 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
1675 | { | |
1676 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 1677 | struct gfs2_holder rgd_gh; |
58884c4d | 1678 | int error = -EINVAL; |
acf7e244 | 1679 | |
66fc061b | 1680 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 1681 | if (!rgd) |
8339ee54 | 1682 | goto fail; |
acf7e244 SW |
1683 | |
1684 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
1685 | if (error) | |
8339ee54 | 1686 | goto fail; |
acf7e244 SW |
1687 | |
1688 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
1689 | error = -ESTALE; | |
1690 | ||
1691 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
1692 | fail: |
1693 | return error; | |
1694 | } | |
1695 | ||
b3b94faa DT |
1696 | /** |
1697 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 1698 | * @ip: the inode |
b3b94faa DT |
1699 | * @rlist: the list of resource groups |
1700 | * @block: the block | |
1701 | * | |
1702 | * Figure out what RG a block belongs to and add that RG to the list | |
1703 | * | |
1704 | * FIXME: Don't use NOFAIL | |
1705 | * | |
1706 | */ | |
1707 | ||
70b0c365 | 1708 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 1709 | u64 block) |
b3b94faa | 1710 | { |
70b0c365 | 1711 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1712 | struct gfs2_rgrpd *rgd; |
1713 | struct gfs2_rgrpd **tmp; | |
1714 | unsigned int new_space; | |
1715 | unsigned int x; | |
1716 | ||
1717 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
1718 | return; | |
1719 | ||
70b0c365 SW |
1720 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
1721 | rgd = ip->i_rgd; | |
1722 | else | |
66fc061b | 1723 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 1724 | if (!rgd) { |
70b0c365 | 1725 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
1726 | return; |
1727 | } | |
70b0c365 | 1728 | ip->i_rgd = rgd; |
b3b94faa DT |
1729 | |
1730 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1731 | if (rlist->rl_rgd[x] == rgd) | |
1732 | return; | |
1733 | ||
1734 | if (rlist->rl_rgrps == rlist->rl_space) { | |
1735 | new_space = rlist->rl_space + 10; | |
1736 | ||
1737 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 1738 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1739 | |
1740 | if (rlist->rl_rgd) { | |
1741 | memcpy(tmp, rlist->rl_rgd, | |
1742 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
1743 | kfree(rlist->rl_rgd); | |
1744 | } | |
1745 | ||
1746 | rlist->rl_space = new_space; | |
1747 | rlist->rl_rgd = tmp; | |
1748 | } | |
1749 | ||
1750 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
1751 | } | |
1752 | ||
1753 | /** | |
1754 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
1755 | * and initialize an array of glock holders for them | |
1756 | * @rlist: the list of resource groups | |
1757 | * @state: the lock state to acquire the RG lock in | |
1758 | * @flags: the modifier flags for the holder structures | |
1759 | * | |
1760 | * FIXME: Don't use NOFAIL | |
1761 | * | |
1762 | */ | |
1763 | ||
fe6c991c | 1764 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
1765 | { |
1766 | unsigned int x; | |
1767 | ||
1768 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 1769 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
1770 | for (x = 0; x < rlist->rl_rgrps; x++) |
1771 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 1772 | state, 0, |
b3b94faa DT |
1773 | &rlist->rl_ghs[x]); |
1774 | } | |
1775 | ||
1776 | /** | |
1777 | * gfs2_rlist_free - free a resource group list | |
1778 | * @list: the list of resource groups | |
1779 | * | |
1780 | */ | |
1781 | ||
1782 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
1783 | { | |
1784 | unsigned int x; | |
1785 | ||
1786 | kfree(rlist->rl_rgd); | |
1787 | ||
1788 | if (rlist->rl_ghs) { | |
1789 | for (x = 0; x < rlist->rl_rgrps; x++) | |
1790 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
1791 | kfree(rlist->rl_ghs); | |
1792 | } | |
1793 | } | |
1794 |