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> |
9dbe9610 | 19 | #include <linux/random.h> |
b3b94faa DT |
20 | |
21 | #include "gfs2.h" | |
5c676f6d | 22 | #include "incore.h" |
b3b94faa DT |
23 | #include "glock.h" |
24 | #include "glops.h" | |
b3b94faa DT |
25 | #include "lops.h" |
26 | #include "meta_io.h" | |
27 | #include "quota.h" | |
28 | #include "rgrp.h" | |
29 | #include "super.h" | |
30 | #include "trans.h" | |
5c676f6d | 31 | #include "util.h" |
172e045a | 32 | #include "log.h" |
c8cdf479 | 33 | #include "inode.h" |
63997775 | 34 | #include "trace_gfs2.h" |
b3b94faa | 35 | |
2c1e52aa | 36 | #define BFITNOENT ((u32)~0) |
6760bdcd | 37 | #define NO_BLOCK ((u64)~0) |
88c8ab1f | 38 | |
1f466a47 BP |
39 | #if BITS_PER_LONG == 32 |
40 | #define LBITMASK (0x55555555UL) | |
41 | #define LBITSKIP55 (0x55555555UL) | |
42 | #define LBITSKIP00 (0x00000000UL) | |
43 | #else | |
44 | #define LBITMASK (0x5555555555555555UL) | |
45 | #define LBITSKIP55 (0x5555555555555555UL) | |
46 | #define LBITSKIP00 (0x0000000000000000UL) | |
47 | #endif | |
48 | ||
88c8ab1f SW |
49 | /* |
50 | * These routines are used by the resource group routines (rgrp.c) | |
51 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
52 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
53 | * | |
54 | * 0 = Free | |
55 | * 1 = Used (not metadata) | |
56 | * 2 = Unlinked (still in use) inode | |
57 | * 3 = Used (metadata) | |
88c8ab1f SW |
58 | */ |
59 | ||
60 | static const char valid_change[16] = { | |
61 | /* current */ | |
feaa7bba | 62 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 63 | /* e */ 1, 0, 0, 0, |
feaa7bba | 64 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
65 | 1, 0, 0, 0 |
66 | }; | |
67 | ||
ff7f4cb4 SW |
68 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 minext, |
69 | const struct gfs2_inode *ip, bool nowrap); | |
70 | ||
71 | ||
88c8ab1f SW |
72 | /** |
73 | * gfs2_setbit - Set a bit in the bitmaps | |
3e6339dd SW |
74 | * @rbm: The position of the bit to set |
75 | * @do_clone: Also set the clone bitmap, if it exists | |
88c8ab1f SW |
76 | * @new_state: the new state of the block |
77 | * | |
78 | */ | |
79 | ||
3e6339dd | 80 | static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone, |
06344b91 | 81 | unsigned char new_state) |
88c8ab1f | 82 | { |
b45e41d7 | 83 | unsigned char *byte1, *byte2, *end, cur_state; |
3e6339dd SW |
84 | unsigned int buflen = rbm->bi->bi_len; |
85 | const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 86 | |
3e6339dd SW |
87 | byte1 = rbm->bi->bi_bh->b_data + rbm->bi->bi_offset + (rbm->offset / GFS2_NBBY); |
88 | end = rbm->bi->bi_bh->b_data + rbm->bi->bi_offset + buflen; | |
88c8ab1f | 89 | |
b45e41d7 | 90 | BUG_ON(byte1 >= end); |
88c8ab1f | 91 | |
b45e41d7 | 92 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 93 | |
b45e41d7 | 94 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
3e6339dd SW |
95 | printk(KERN_WARNING "GFS2: buf_blk = 0x%x old_state=%d, " |
96 | "new_state=%d\n", rbm->offset, cur_state, new_state); | |
97 | printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%x\n", | |
98 | (unsigned long long)rbm->rgd->rd_addr, | |
99 | rbm->bi->bi_start); | |
100 | printk(KERN_WARNING "GFS2: bi_offset=0x%x bi_len=0x%x\n", | |
101 | rbm->bi->bi_offset, rbm->bi->bi_len); | |
95c8e17f | 102 | dump_stack(); |
3e6339dd | 103 | gfs2_consist_rgrpd(rbm->rgd); |
b45e41d7 SW |
104 | return; |
105 | } | |
106 | *byte1 ^= (cur_state ^ new_state) << bit; | |
107 | ||
3e6339dd SW |
108 | if (do_clone && rbm->bi->bi_clone) { |
109 | byte2 = rbm->bi->bi_clone + rbm->bi->bi_offset + (rbm->offset / GFS2_NBBY); | |
b45e41d7 SW |
110 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
111 | *byte2 ^= (cur_state ^ new_state) << bit; | |
112 | } | |
88c8ab1f SW |
113 | } |
114 | ||
115 | /** | |
116 | * gfs2_testbit - test a bit in the bitmaps | |
c04a2ef3 | 117 | * @rbm: The bit to test |
88c8ab1f | 118 | * |
c04a2ef3 | 119 | * Returns: The two bit block state of the requested bit |
88c8ab1f SW |
120 | */ |
121 | ||
c04a2ef3 | 122 | static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm) |
88c8ab1f | 123 | { |
c04a2ef3 SW |
124 | const u8 *buffer = rbm->bi->bi_bh->b_data + rbm->bi->bi_offset; |
125 | const u8 *byte; | |
88c8ab1f SW |
126 | unsigned int bit; |
127 | ||
c04a2ef3 SW |
128 | byte = buffer + (rbm->offset / GFS2_NBBY); |
129 | bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 130 | |
c04a2ef3 | 131 | return (*byte >> bit) & GFS2_BIT_MASK; |
88c8ab1f SW |
132 | } |
133 | ||
223b2b88 SW |
134 | /** |
135 | * gfs2_bit_search | |
136 | * @ptr: Pointer to bitmap data | |
137 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
138 | * @state: The state we are searching for | |
139 | * | |
140 | * We xor the bitmap data with a patter which is the bitwise opposite | |
141 | * of what we are looking for, this gives rise to a pattern of ones | |
142 | * wherever there is a match. Since we have two bits per entry, we | |
143 | * take this pattern, shift it down by one place and then and it with | |
144 | * the original. All the even bit positions (0,2,4, etc) then represent | |
145 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
146 | * odd bit positions. | |
147 | * | |
148 | * This allows searching of a whole u64 at once (32 blocks) with a | |
149 | * single test (on 64 bit arches). | |
150 | */ | |
151 | ||
152 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
153 | { | |
154 | u64 tmp; | |
155 | static const u64 search[] = { | |
075ac448 HE |
156 | [0] = 0xffffffffffffffffULL, |
157 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
158 | [2] = 0x5555555555555555ULL, | |
159 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
160 | }; |
161 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
162 | tmp &= (tmp >> 1); | |
163 | tmp &= mask; | |
164 | return tmp; | |
165 | } | |
166 | ||
8e2e0047 BP |
167 | /** |
168 | * rs_cmp - multi-block reservation range compare | |
169 | * @blk: absolute file system block number of the new reservation | |
170 | * @len: number of blocks in the new reservation | |
171 | * @rs: existing reservation to compare against | |
172 | * | |
173 | * returns: 1 if the block range is beyond the reach of the reservation | |
174 | * -1 if the block range is before the start of the reservation | |
175 | * 0 if the block range overlaps with the reservation | |
176 | */ | |
177 | static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs) | |
178 | { | |
4a993fb1 | 179 | u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 BP |
180 | |
181 | if (blk >= startblk + rs->rs_free) | |
182 | return 1; | |
183 | if (blk + len - 1 < startblk) | |
184 | return -1; | |
185 | return 0; | |
186 | } | |
187 | ||
88c8ab1f SW |
188 | /** |
189 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
190 | * a block in a given allocation state. | |
886b1416 | 191 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 192 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 193 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 194 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
195 | * |
196 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
197 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
198 | * beginning of a bitmap block buffer, skipping any header structures, but |
199 | * headers are always a multiple of 64 bits long so that the buffer is | |
200 | * always aligned to a 64 bit boundary. | |
201 | * | |
202 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 203 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 204 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
205 | * |
206 | * Return: the block number (bitmap buffer scope) that was found | |
207 | */ | |
208 | ||
02ab1721 HE |
209 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
210 | u32 goal, u8 state) | |
88c8ab1f | 211 | { |
223b2b88 SW |
212 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
213 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
214 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
215 | u64 tmp; | |
075ac448 | 216 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
217 | u32 bit; |
218 | ||
223b2b88 SW |
219 | /* Mask off bits we don't care about at the start of the search */ |
220 | mask <<= spoint; | |
221 | tmp = gfs2_bit_search(ptr, mask, state); | |
222 | ptr++; | |
223 | while(tmp == 0 && ptr < end) { | |
075ac448 | 224 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 225 | ptr++; |
1f466a47 | 226 | } |
223b2b88 SW |
227 | /* Mask off any bits which are more than len bytes from the start */ |
228 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
229 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
230 | /* Didn't find anything, so return */ | |
231 | if (tmp == 0) | |
232 | return BFITNOENT; | |
233 | ptr--; | |
d8bd504a | 234 | bit = __ffs64(tmp); |
223b2b88 SW |
235 | bit /= 2; /* two bits per entry in the bitmap */ |
236 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
237 | } |
238 | ||
ff7f4cb4 SW |
239 | /** |
240 | * gfs2_rbm_from_block - Set the rbm based upon rgd and block number | |
241 | * @rbm: The rbm with rgd already set correctly | |
242 | * @block: The block number (filesystem relative) | |
243 | * | |
244 | * This sets the bi and offset members of an rbm based on a | |
245 | * resource group and a filesystem relative block number. The | |
246 | * resource group must be set in the rbm on entry, the bi and | |
247 | * offset members will be set by this function. | |
248 | * | |
249 | * Returns: 0 on success, or an error code | |
250 | */ | |
251 | ||
252 | static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block) | |
253 | { | |
254 | u64 rblock = block - rbm->rgd->rd_data0; | |
a68a0a35 | 255 | u32 x; |
ff7f4cb4 SW |
256 | |
257 | if (WARN_ON_ONCE(rblock > UINT_MAX)) | |
258 | return -EINVAL; | |
259 | if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data) | |
260 | return -E2BIG; | |
261 | ||
a68a0a35 BP |
262 | rbm->bi = rbm->rgd->rd_bits; |
263 | rbm->offset = (u32)(rblock); | |
264 | /* Check if the block is within the first block */ | |
265 | if (rbm->offset < (rbm->bi->bi_start + rbm->bi->bi_len) * GFS2_NBBY) | |
266 | return 0; | |
267 | ||
268 | /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */ | |
269 | rbm->offset += (sizeof(struct gfs2_rgrp) - | |
270 | sizeof(struct gfs2_meta_header)) * GFS2_NBBY; | |
271 | x = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap; | |
272 | rbm->offset -= x * rbm->rgd->rd_sbd->sd_blocks_per_bitmap; | |
273 | rbm->bi += x; | |
ff7f4cb4 SW |
274 | return 0; |
275 | } | |
276 | ||
277 | /** | |
278 | * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned | |
279 | * @rbm: Position to search (value/result) | |
280 | * @n_unaligned: Number of unaligned blocks to check | |
281 | * @len: Decremented for each block found (terminate on zero) | |
282 | * | |
283 | * Returns: true if a non-free block is encountered | |
284 | */ | |
285 | ||
286 | static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len) | |
287 | { | |
288 | u64 block; | |
289 | u32 n; | |
290 | u8 res; | |
291 | ||
292 | for (n = 0; n < n_unaligned; n++) { | |
293 | res = gfs2_testbit(rbm); | |
294 | if (res != GFS2_BLKST_FREE) | |
295 | return true; | |
296 | (*len)--; | |
297 | if (*len == 0) | |
298 | return true; | |
299 | block = gfs2_rbm_to_block(rbm); | |
300 | if (gfs2_rbm_from_block(rbm, block + 1)) | |
301 | return true; | |
302 | } | |
303 | ||
304 | return false; | |
305 | } | |
306 | ||
307 | /** | |
308 | * gfs2_free_extlen - Return extent length of free blocks | |
309 | * @rbm: Starting position | |
310 | * @len: Max length to check | |
311 | * | |
312 | * Starting at the block specified by the rbm, see how many free blocks | |
313 | * there are, not reading more than len blocks ahead. This can be done | |
314 | * using memchr_inv when the blocks are byte aligned, but has to be done | |
315 | * on a block by block basis in case of unaligned blocks. Also this | |
316 | * function can cope with bitmap boundaries (although it must stop on | |
317 | * a resource group boundary) | |
318 | * | |
319 | * Returns: Number of free blocks in the extent | |
320 | */ | |
321 | ||
322 | static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len) | |
323 | { | |
324 | struct gfs2_rbm rbm = *rrbm; | |
325 | u32 n_unaligned = rbm.offset & 3; | |
326 | u32 size = len; | |
327 | u32 bytes; | |
328 | u32 chunk_size; | |
329 | u8 *ptr, *start, *end; | |
330 | u64 block; | |
331 | ||
332 | if (n_unaligned && | |
333 | gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len)) | |
334 | goto out; | |
335 | ||
3701530a | 336 | n_unaligned = len & 3; |
ff7f4cb4 SW |
337 | /* Start is now byte aligned */ |
338 | while (len > 3) { | |
339 | start = rbm.bi->bi_bh->b_data; | |
340 | if (rbm.bi->bi_clone) | |
341 | start = rbm.bi->bi_clone; | |
342 | end = start + rbm.bi->bi_bh->b_size; | |
343 | start += rbm.bi->bi_offset; | |
344 | BUG_ON(rbm.offset & 3); | |
345 | start += (rbm.offset / GFS2_NBBY); | |
346 | bytes = min_t(u32, len / GFS2_NBBY, (end - start)); | |
347 | ptr = memchr_inv(start, 0, bytes); | |
348 | chunk_size = ((ptr == NULL) ? bytes : (ptr - start)); | |
349 | chunk_size *= GFS2_NBBY; | |
350 | BUG_ON(len < chunk_size); | |
351 | len -= chunk_size; | |
352 | block = gfs2_rbm_to_block(&rbm); | |
15bd50ad BP |
353 | if (gfs2_rbm_from_block(&rbm, block + chunk_size)) { |
354 | n_unaligned = 0; | |
ff7f4cb4 | 355 | break; |
15bd50ad BP |
356 | } |
357 | if (ptr) { | |
358 | n_unaligned = 3; | |
359 | break; | |
360 | } | |
ff7f4cb4 SW |
361 | n_unaligned = len & 3; |
362 | } | |
363 | ||
364 | /* Deal with any bits left over at the end */ | |
365 | if (n_unaligned) | |
366 | gfs2_unaligned_extlen(&rbm, n_unaligned, &len); | |
367 | out: | |
368 | return size - len; | |
369 | } | |
370 | ||
88c8ab1f SW |
371 | /** |
372 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 373 | * @rgd: the resource group descriptor |
88c8ab1f SW |
374 | * @buffer: the buffer that holds the bitmaps |
375 | * @buflen: the length (in bytes) of the buffer | |
376 | * @state: the state of the block we're looking for | |
377 | * | |
378 | * Returns: The number of bits | |
379 | */ | |
380 | ||
110acf38 SW |
381 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
382 | unsigned int buflen, u8 state) | |
88c8ab1f | 383 | { |
110acf38 SW |
384 | const u8 *byte = buffer; |
385 | const u8 *end = buffer + buflen; | |
386 | const u8 state1 = state << 2; | |
387 | const u8 state2 = state << 4; | |
388 | const u8 state3 = state << 6; | |
cd915493 | 389 | u32 count = 0; |
88c8ab1f SW |
390 | |
391 | for (; byte < end; byte++) { | |
392 | if (((*byte) & 0x03) == state) | |
393 | count++; | |
394 | if (((*byte) & 0x0C) == state1) | |
395 | count++; | |
396 | if (((*byte) & 0x30) == state2) | |
397 | count++; | |
398 | if (((*byte) & 0xC0) == state3) | |
399 | count++; | |
400 | } | |
401 | ||
402 | return count; | |
403 | } | |
404 | ||
b3b94faa DT |
405 | /** |
406 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
407 | * @rgd: the rgrp |
408 | * | |
409 | */ | |
410 | ||
411 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
412 | { | |
413 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
414 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 415 | u32 length = rgd->rd_length; |
cd915493 | 416 | u32 count[4], tmp; |
b3b94faa DT |
417 | int buf, x; |
418 | ||
cd915493 | 419 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
420 | |
421 | /* Count # blocks in each of 4 possible allocation states */ | |
422 | for (buf = 0; buf < length; buf++) { | |
423 | bi = rgd->rd_bits + buf; | |
424 | for (x = 0; x < 4; x++) | |
425 | count[x] += gfs2_bitcount(rgd, | |
426 | bi->bi_bh->b_data + | |
427 | bi->bi_offset, | |
428 | bi->bi_len, x); | |
429 | } | |
430 | ||
cfc8b549 | 431 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
432 | if (gfs2_consist_rgrpd(rgd)) |
433 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 434 | count[0], rgd->rd_free); |
b3b94faa DT |
435 | return; |
436 | } | |
437 | ||
73f74948 | 438 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 439 | if (count[1] != tmp) { |
b3b94faa DT |
440 | if (gfs2_consist_rgrpd(rgd)) |
441 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
442 | count[1], tmp); | |
443 | return; | |
444 | } | |
445 | ||
6b946170 | 446 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 447 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 448 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 449 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
450 | return; |
451 | } | |
b3b94faa DT |
452 | } |
453 | ||
bb8d8a6f | 454 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 455 | { |
bb8d8a6f SW |
456 | u64 first = rgd->rd_data0; |
457 | u64 last = first + rgd->rd_data; | |
16910427 | 458 | return first <= block && block < last; |
b3b94faa DT |
459 | } |
460 | ||
461 | /** | |
462 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
463 | * @sdp: The GFS2 superblock | |
886b1416 BP |
464 | * @blk: The data block number |
465 | * @exact: True if this needs to be an exact match | |
b3b94faa DT |
466 | * |
467 | * Returns: The resource group, or NULL if not found | |
468 | */ | |
469 | ||
66fc061b | 470 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 471 | { |
66fc061b | 472 | struct rb_node *n, *next; |
f75bbfb4 | 473 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
474 | |
475 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
476 | n = sdp->sd_rindex_tree.rb_node; |
477 | while (n) { | |
478 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
479 | next = NULL; | |
7c9ca621 | 480 | if (blk < cur->rd_addr) |
66fc061b | 481 | next = n->rb_left; |
f75bbfb4 | 482 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
483 | next = n->rb_right; |
484 | if (next == NULL) { | |
b3b94faa | 485 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
486 | if (exact) { |
487 | if (blk < cur->rd_addr) | |
488 | return NULL; | |
489 | if (blk >= cur->rd_data0 + cur->rd_data) | |
490 | return NULL; | |
491 | } | |
7c9ca621 | 492 | return cur; |
b3b94faa | 493 | } |
66fc061b | 494 | n = next; |
b3b94faa | 495 | } |
b3b94faa DT |
496 | spin_unlock(&sdp->sd_rindex_spin); |
497 | ||
498 | return NULL; | |
499 | } | |
500 | ||
501 | /** | |
502 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
503 | * @sdp: The GFS2 superblock | |
504 | * | |
505 | * Returns: The first rgrp in the filesystem | |
506 | */ | |
507 | ||
508 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
509 | { | |
7c9ca621 BP |
510 | const struct rb_node *n; |
511 | struct gfs2_rgrpd *rgd; | |
512 | ||
8339ee54 | 513 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
514 | n = rb_first(&sdp->sd_rindex_tree); |
515 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 516 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
517 | |
518 | return rgd; | |
b3b94faa DT |
519 | } |
520 | ||
521 | /** | |
522 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 523 | * @rgd: the resource group descriptor |
b3b94faa DT |
524 | * |
525 | * Returns: The next rgrp | |
526 | */ | |
527 | ||
528 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
529 | { | |
7c9ca621 BP |
530 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
531 | const struct rb_node *n; | |
532 | ||
533 | spin_lock(&sdp->sd_rindex_spin); | |
534 | n = rb_next(&rgd->rd_node); | |
535 | if (n == NULL) | |
536 | n = rb_first(&sdp->sd_rindex_tree); | |
537 | ||
538 | if (unlikely(&rgd->rd_node == n)) { | |
539 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 540 | return NULL; |
7c9ca621 BP |
541 | } |
542 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
543 | spin_unlock(&sdp->sd_rindex_spin); | |
544 | return rgd; | |
b3b94faa DT |
545 | } |
546 | ||
8339ee54 SW |
547 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
548 | { | |
549 | int x; | |
550 | ||
551 | for (x = 0; x < rgd->rd_length; x++) { | |
552 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
553 | kfree(bi->bi_clone); | |
554 | bi->bi_clone = NULL; | |
555 | } | |
556 | } | |
557 | ||
0a305e49 BP |
558 | /** |
559 | * gfs2_rs_alloc - make sure we have a reservation assigned to the inode | |
560 | * @ip: the inode for this reservation | |
561 | */ | |
562 | int gfs2_rs_alloc(struct gfs2_inode *ip) | |
563 | { | |
f1213cac | 564 | int error = 0; |
8e2e0047 | 565 | |
f1213cac | 566 | down_write(&ip->i_rw_mutex); |
8e2e0047 | 567 | if (ip->i_res) |
f1213cac | 568 | goto out; |
0a305e49 | 569 | |
f1213cac AD |
570 | ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); |
571 | if (!ip->i_res) { | |
572 | error = -ENOMEM; | |
573 | goto out; | |
574 | } | |
4a993fb1 | 575 | |
f1213cac AD |
576 | RB_CLEAR_NODE(&ip->i_res->rs_node); |
577 | out: | |
0a305e49 | 578 | up_write(&ip->i_rw_mutex); |
441362d0 | 579 | return error; |
0a305e49 BP |
580 | } |
581 | ||
9e733d39 | 582 | static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs) |
8e2e0047 | 583 | { |
9e733d39 SW |
584 | gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n", |
585 | (unsigned long long)rs->rs_inum, | |
586 | (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm), | |
4a993fb1 | 587 | rs->rs_rbm.offset, rs->rs_free); |
8e2e0047 BP |
588 | } |
589 | ||
0a305e49 | 590 | /** |
8e2e0047 BP |
591 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
592 | * @rs: The reservation to remove | |
593 | * | |
594 | */ | |
20095218 | 595 | static void __rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
596 | { |
597 | struct gfs2_rgrpd *rgd; | |
598 | ||
599 | if (!gfs2_rs_active(rs)) | |
600 | return; | |
601 | ||
4a993fb1 | 602 | rgd = rs->rs_rbm.rgd; |
9e733d39 | 603 | trace_gfs2_rs(rs, TRACE_RS_TREEDEL); |
4a993fb1 | 604 | rb_erase(&rs->rs_node, &rgd->rd_rstree); |
24d634e8 | 605 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 BP |
606 | |
607 | if (rs->rs_free) { | |
20095218 | 608 | /* return reserved blocks to the rgrp */ |
4a993fb1 SW |
609 | BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free); |
610 | rs->rs_rbm.rgd->rd_reserved -= rs->rs_free; | |
8e2e0047 | 611 | rs->rs_free = 0; |
4a993fb1 | 612 | clear_bit(GBF_FULL, &rs->rs_rbm.bi->bi_flags); |
8e2e0047 BP |
613 | smp_mb__after_clear_bit(); |
614 | } | |
8e2e0047 BP |
615 | } |
616 | ||
617 | /** | |
618 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
619 | * @rs: The reservation to remove | |
620 | * | |
621 | */ | |
20095218 | 622 | void gfs2_rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
623 | { |
624 | struct gfs2_rgrpd *rgd; | |
625 | ||
4a993fb1 SW |
626 | rgd = rs->rs_rbm.rgd; |
627 | if (rgd) { | |
628 | spin_lock(&rgd->rd_rsspin); | |
20095218 | 629 | __rs_deltree(rs); |
4a993fb1 SW |
630 | spin_unlock(&rgd->rd_rsspin); |
631 | } | |
8e2e0047 BP |
632 | } |
633 | ||
634 | /** | |
635 | * gfs2_rs_delete - delete a multi-block reservation | |
0a305e49 BP |
636 | * @ip: The inode for this reservation |
637 | * | |
638 | */ | |
639 | void gfs2_rs_delete(struct gfs2_inode *ip) | |
640 | { | |
2b3dcf35 BP |
641 | struct inode *inode = &ip->i_inode; |
642 | ||
0a305e49 | 643 | down_write(&ip->i_rw_mutex); |
2b3dcf35 | 644 | if (ip->i_res && atomic_read(&inode->i_writecount) <= 1) { |
20095218 | 645 | gfs2_rs_deltree(ip->i_res); |
8e2e0047 | 646 | BUG_ON(ip->i_res->rs_free); |
0a305e49 BP |
647 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); |
648 | ip->i_res = NULL; | |
649 | } | |
650 | up_write(&ip->i_rw_mutex); | |
651 | } | |
652 | ||
8e2e0047 BP |
653 | /** |
654 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
655 | * @rgd: the rgrp that needs its space back | |
656 | * | |
657 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
658 | * give them back. This leave the reservation structures in tact, but removes | |
659 | * all of their corresponding "no-fly zones". | |
660 | */ | |
661 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
662 | { | |
663 | struct rb_node *n; | |
664 | struct gfs2_blkreserv *rs; | |
665 | ||
666 | spin_lock(&rgd->rd_rsspin); | |
667 | while ((n = rb_first(&rgd->rd_rstree))) { | |
668 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
20095218 | 669 | __rs_deltree(rs); |
8e2e0047 BP |
670 | } |
671 | spin_unlock(&rgd->rd_rsspin); | |
672 | } | |
673 | ||
8339ee54 | 674 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 675 | { |
7c9ca621 | 676 | struct rb_node *n; |
b3b94faa DT |
677 | struct gfs2_rgrpd *rgd; |
678 | struct gfs2_glock *gl; | |
679 | ||
7c9ca621 BP |
680 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
681 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
682 | gl = rgd->rd_gl; |
683 | ||
7c9ca621 | 684 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
685 | |
686 | if (gl) { | |
8339ee54 | 687 | spin_lock(&gl->gl_spin); |
5c676f6d | 688 | gl->gl_object = NULL; |
8339ee54 | 689 | spin_unlock(&gl->gl_spin); |
29687a2a | 690 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
691 | gfs2_glock_put(gl); |
692 | } | |
693 | ||
8339ee54 | 694 | gfs2_free_clones(rgd); |
b3b94faa | 695 | kfree(rgd->rd_bits); |
8e2e0047 | 696 | return_all_reservations(rgd); |
6bdd9be6 | 697 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
698 | } |
699 | } | |
700 | ||
bb8d8a6f SW |
701 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
702 | { | |
703 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
704 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
705 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
706 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
707 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
708 | } | |
709 | ||
b3b94faa DT |
710 | /** |
711 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
712 | * @rgd: The resource group descriptor | |
713 | * | |
714 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
715 | * | |
716 | * Returns: errno | |
717 | */ | |
718 | ||
719 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
720 | { | |
721 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
722 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 723 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 724 | u32 bytes_left, bytes; |
b3b94faa DT |
725 | int x; |
726 | ||
feaa7bba SW |
727 | if (!length) |
728 | return -EINVAL; | |
729 | ||
dd894be8 | 730 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
731 | if (!rgd->rd_bits) |
732 | return -ENOMEM; | |
733 | ||
bb8d8a6f | 734 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
735 | |
736 | for (x = 0; x < length; x++) { | |
737 | bi = rgd->rd_bits + x; | |
738 | ||
60a0b8f9 | 739 | bi->bi_flags = 0; |
b3b94faa DT |
740 | /* small rgrp; bitmap stored completely in header block */ |
741 | if (length == 1) { | |
742 | bytes = bytes_left; | |
743 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
744 | bi->bi_start = 0; | |
745 | bi->bi_len = bytes; | |
746 | /* header block */ | |
747 | } else if (x == 0) { | |
748 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
749 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
750 | bi->bi_start = 0; | |
751 | bi->bi_len = bytes; | |
752 | /* last block */ | |
753 | } else if (x + 1 == length) { | |
754 | bytes = bytes_left; | |
755 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 756 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
757 | bi->bi_len = bytes; |
758 | /* other blocks */ | |
759 | } else { | |
568f4c96 SW |
760 | bytes = sdp->sd_sb.sb_bsize - |
761 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 762 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 763 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa DT |
764 | bi->bi_len = bytes; |
765 | } | |
766 | ||
767 | bytes_left -= bytes; | |
768 | } | |
769 | ||
770 | if (bytes_left) { | |
771 | gfs2_consist_rgrpd(rgd); | |
772 | return -EIO; | |
773 | } | |
774 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 775 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 776 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 777 | gfs2_rindex_print(rgd); |
b3b94faa DT |
778 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
779 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
780 | } | |
781 | return -EIO; | |
782 | } | |
783 | ||
784 | return 0; | |
785 | } | |
786 | ||
7ae8fa84 RP |
787 | /** |
788 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 789 | * @sdp: the filesystem |
7ae8fa84 RP |
790 | * |
791 | */ | |
792 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
793 | { | |
794 | u64 total_data = 0; | |
795 | struct inode *inode = sdp->sd_rindex; | |
796 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 797 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
798 | int error, rgrps; |
799 | ||
7ae8fa84 RP |
800 | for (rgrps = 0;; rgrps++) { |
801 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
802 | ||
bcd7278d | 803 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 804 | break; |
4306629e | 805 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
806 | sizeof(struct gfs2_rindex)); |
807 | if (error != sizeof(struct gfs2_rindex)) | |
808 | break; | |
bb8d8a6f | 809 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 810 | } |
7ae8fa84 RP |
811 | return total_data; |
812 | } | |
813 | ||
6aad1c3d | 814 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
815 | { |
816 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
817 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
818 | ||
819 | /* Figure out where to put new node */ | |
820 | while (*newn) { | |
821 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
822 | rd_node); | |
823 | ||
824 | parent = *newn; | |
825 | if (rgd->rd_addr < cur->rd_addr) | |
826 | newn = &((*newn)->rb_left); | |
827 | else if (rgd->rd_addr > cur->rd_addr) | |
828 | newn = &((*newn)->rb_right); | |
829 | else | |
6aad1c3d | 830 | return -EEXIST; |
7c9ca621 BP |
831 | } |
832 | ||
833 | rb_link_node(&rgd->rd_node, parent, newn); | |
834 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
835 | sdp->sd_rgrps++; |
836 | return 0; | |
7c9ca621 BP |
837 | } |
838 | ||
b3b94faa | 839 | /** |
6c53267f | 840 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 841 | * @ip: Pointer to the rindex inode |
b3b94faa | 842 | * |
8339ee54 | 843 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
844 | */ |
845 | ||
4306629e | 846 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
847 | { |
848 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
849 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
8339ee54 | 850 | struct gfs2_rindex buf; |
6c53267f RP |
851 | int error; |
852 | struct gfs2_rgrpd *rgd; | |
853 | ||
8339ee54 SW |
854 | if (pos >= i_size_read(&ip->i_inode)) |
855 | return 1; | |
856 | ||
4306629e | 857 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 858 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
859 | |
860 | if (error != sizeof(struct gfs2_rindex)) | |
861 | return (error == 0) ? 1 : error; | |
6c53267f | 862 | |
6bdd9be6 | 863 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
864 | error = -ENOMEM; |
865 | if (!rgd) | |
866 | return error; | |
867 | ||
6c53267f | 868 | rgd->rd_sbd = sdp; |
8339ee54 SW |
869 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
870 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
871 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
872 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
873 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 874 | spin_lock_init(&rgd->rd_rsspin); |
7c9ca621 | 875 | |
6c53267f RP |
876 | error = compute_bitstructs(rgd); |
877 | if (error) | |
8339ee54 | 878 | goto fail; |
6c53267f | 879 | |
bb8d8a6f | 880 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
881 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
882 | if (error) | |
8339ee54 | 883 | goto fail; |
6c53267f RP |
884 | |
885 | rgd->rd_gl->gl_object = rgd; | |
4e2f8849 | 886 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr; |
cf45b752 | 887 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
7c9ca621 BP |
888 | if (rgd->rd_data > sdp->sd_max_rg_data) |
889 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 890 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 891 | error = rgd_insert(rgd); |
8339ee54 | 892 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
893 | if (!error) |
894 | return 0; | |
895 | ||
896 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 897 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
898 | |
899 | fail: | |
900 | kfree(rgd->rd_bits); | |
901 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
902 | return error; |
903 | } | |
904 | ||
905 | /** | |
906 | * gfs2_ri_update - Pull in a new resource index from the disk | |
907 | * @ip: pointer to the rindex inode | |
908 | * | |
b3b94faa DT |
909 | * Returns: 0 on successful update, error code otherwise |
910 | */ | |
911 | ||
8339ee54 | 912 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 913 | { |
feaa7bba | 914 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
915 | int error; |
916 | ||
8339ee54 | 917 | do { |
4306629e | 918 | error = read_rindex_entry(ip); |
8339ee54 SW |
919 | } while (error == 0); |
920 | ||
921 | if (error < 0) | |
922 | return error; | |
b3b94faa | 923 | |
cf45b752 | 924 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
925 | return 0; |
926 | } | |
b3b94faa | 927 | |
b3b94faa | 928 | /** |
8339ee54 | 929 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 930 | * @sdp: The GFS2 superblock |
b3b94faa DT |
931 | * |
932 | * We grab a lock on the rindex inode to make sure that it doesn't | |
933 | * change whilst we are performing an operation. We keep this lock | |
934 | * for quite long periods of time compared to other locks. This | |
935 | * doesn't matter, since it is shared and it is very, very rarely | |
936 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
937 | * | |
938 | * This makes sure that we're using the latest copy of the resource index | |
939 | * special file, which might have been updated if someone expanded the | |
940 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
941 | * | |
8339ee54 | 942 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
943 | */ |
944 | ||
8339ee54 | 945 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 946 | { |
feaa7bba | 947 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 948 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
949 | struct gfs2_holder ri_gh; |
950 | int error = 0; | |
a365fbf3 | 951 | int unlock_required = 0; |
b3b94faa DT |
952 | |
953 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 954 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
955 | if (!gfs2_glock_is_locked_by_me(gl)) { |
956 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
957 | if (error) | |
6aad1c3d | 958 | return error; |
a365fbf3 SW |
959 | unlock_required = 1; |
960 | } | |
8339ee54 | 961 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 962 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
963 | if (unlock_required) |
964 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
965 | } |
966 | ||
967 | return error; | |
968 | } | |
969 | ||
42d52e38 | 970 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
971 | { |
972 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 973 | u32 rg_flags; |
bb8d8a6f | 974 | |
42d52e38 | 975 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 976 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
977 | rgd->rd_flags &= GFS2_RDF_MASK; |
978 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 979 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 980 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 981 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
982 | } |
983 | ||
42d52e38 | 984 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
985 | { |
986 | struct gfs2_rgrp *str = buf; | |
987 | ||
09010978 | 988 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 989 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 990 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 991 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 992 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
993 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
994 | } | |
995 | ||
90306c41 BM |
996 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
997 | { | |
998 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
999 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
1000 | ||
1001 | if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free || | |
1002 | rgl->rl_dinodes != str->rg_dinodes || | |
1003 | rgl->rl_igeneration != str->rg_igeneration) | |
1004 | return 0; | |
1005 | return 1; | |
1006 | } | |
1007 | ||
1008 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) | |
1009 | { | |
1010 | const struct gfs2_rgrp *str = buf; | |
1011 | ||
1012 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
1013 | rgl->rl_flags = str->rg_flags; | |
1014 | rgl->rl_free = str->rg_free; | |
1015 | rgl->rl_dinodes = str->rg_dinodes; | |
1016 | rgl->rl_igeneration = str->rg_igeneration; | |
1017 | rgl->__pad = 0UL; | |
1018 | } | |
1019 | ||
1020 | static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change) | |
1021 | { | |
1022 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1023 | u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change; | |
1024 | rgl->rl_unlinked = cpu_to_be32(unlinked); | |
1025 | } | |
1026 | ||
1027 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) | |
1028 | { | |
1029 | struct gfs2_bitmap *bi; | |
1030 | const u32 length = rgd->rd_length; | |
1031 | const u8 *buffer = NULL; | |
1032 | u32 i, goal, count = 0; | |
1033 | ||
1034 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
1035 | goal = 0; | |
1036 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1037 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
1038 | while (goal < bi->bi_len * GFS2_NBBY) { | |
1039 | goal = gfs2_bitfit(buffer, bi->bi_len, goal, | |
1040 | GFS2_BLKST_UNLINKED); | |
1041 | if (goal == BFITNOENT) | |
1042 | break; | |
1043 | count++; | |
1044 | goal++; | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | return count; | |
1049 | } | |
1050 | ||
1051 | ||
b3b94faa | 1052 | /** |
90306c41 BM |
1053 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
1054 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
1055 | * |
1056 | * Read in all of a Resource Group's header and bitmap blocks. | |
1057 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
1058 | * | |
1059 | * Returns: errno | |
1060 | */ | |
1061 | ||
90306c41 | 1062 | int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
1063 | { |
1064 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1065 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 1066 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
1067 | struct gfs2_bitmap *bi; |
1068 | unsigned int x, y; | |
1069 | int error; | |
1070 | ||
90306c41 BM |
1071 | if (rgd->rd_bits[0].bi_bh != NULL) |
1072 | return 0; | |
1073 | ||
b3b94faa DT |
1074 | for (x = 0; x < length; x++) { |
1075 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 1076 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
1077 | if (error) |
1078 | goto fail; | |
1079 | } | |
1080 | ||
1081 | for (y = length; y--;) { | |
1082 | bi = rgd->rd_bits + y; | |
7276b3b0 | 1083 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
1084 | if (error) |
1085 | goto fail; | |
feaa7bba | 1086 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
1087 | GFS2_METATYPE_RG)) { |
1088 | error = -EIO; | |
1089 | goto fail; | |
1090 | } | |
1091 | } | |
1092 | ||
cf45b752 | 1093 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
1094 | for (x = 0; x < length; x++) |
1095 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 1096 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 1097 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 1098 | rgd->rd_free_clone = rgd->rd_free; |
b3b94faa | 1099 | } |
90306c41 BM |
1100 | if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
1101 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); | |
1102 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
1103 | rgd->rd_bits[0].bi_bh->b_data); | |
1104 | } | |
1105 | else if (sdp->sd_args.ar_rgrplvb) { | |
1106 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
1107 | gfs2_consist_rgrpd(rgd); | |
1108 | error = -EIO; | |
1109 | goto fail; | |
1110 | } | |
1111 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1112 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1113 | } | |
b3b94faa DT |
1114 | return 0; |
1115 | ||
feaa7bba | 1116 | fail: |
b3b94faa DT |
1117 | while (x--) { |
1118 | bi = rgd->rd_bits + x; | |
1119 | brelse(bi->bi_bh); | |
1120 | bi->bi_bh = NULL; | |
1121 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
1122 | } | |
b3b94faa DT |
1123 | |
1124 | return error; | |
1125 | } | |
1126 | ||
90306c41 BM |
1127 | int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
1128 | { | |
1129 | u32 rl_flags; | |
1130 | ||
1131 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1132 | return 0; | |
1133 | ||
1134 | if (be32_to_cpu(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) | |
1135 | return gfs2_rgrp_bh_get(rgd); | |
1136 | ||
1137 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1138 | rl_flags &= ~GFS2_RDF_MASK; | |
1139 | rgd->rd_flags &= GFS2_RDF_MASK; | |
1140 | rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); | |
1141 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1142 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1143 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
1144 | rgd->rd_free_clone = rgd->rd_free; | |
1145 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); | |
1146 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1147 | return 0; | |
1148 | } | |
1149 | ||
1150 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1151 | { | |
1152 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1153 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1154 | ||
1155 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1156 | return 0; | |
1157 | return gfs2_rgrp_bh_get((struct gfs2_rgrpd *)gh->gh_gl->gl_object); | |
1158 | } | |
1159 | ||
b3b94faa | 1160 | /** |
7c9ca621 | 1161 | * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
886b1416 | 1162 | * @gh: The glock holder for the resource group |
b3b94faa DT |
1163 | * |
1164 | */ | |
1165 | ||
7c9ca621 | 1166 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) |
b3b94faa | 1167 | { |
7c9ca621 | 1168 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
bb8d8a6f | 1169 | int x, length = rgd->rd_length; |
b3b94faa | 1170 | |
b3b94faa DT |
1171 | for (x = 0; x < length; x++) { |
1172 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1173 | if (bi->bi_bh) { |
1174 | brelse(bi->bi_bh); | |
1175 | bi->bi_bh = NULL; | |
1176 | } | |
b3b94faa DT |
1177 | } |
1178 | ||
b3b94faa DT |
1179 | } |
1180 | ||
66fc061b | 1181 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1182 | struct buffer_head *bh, |
66fc061b | 1183 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1184 | { |
1185 | struct super_block *sb = sdp->sd_vfs; | |
f15ab561 | 1186 | u64 blk; |
64d576ba | 1187 | sector_t start = 0; |
b2c87cae | 1188 | sector_t nr_blks = 0; |
f15ab561 SW |
1189 | int rv; |
1190 | unsigned int x; | |
66fc061b SW |
1191 | u32 trimmed = 0; |
1192 | u8 diff; | |
f15ab561 SW |
1193 | |
1194 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
1195 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1196 | clone += bi->bi_offset; | |
1197 | clone += x; | |
1198 | if (bh) { | |
1199 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1200 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1201 | } else { | |
1202 | diff = ~(*clone | (*clone >> 1)); | |
1203 | } | |
f15ab561 SW |
1204 | diff &= 0x55; |
1205 | if (diff == 0) | |
1206 | continue; | |
1207 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
f15ab561 SW |
1208 | while(diff) { |
1209 | if (diff & 1) { | |
b2c87cae | 1210 | if (nr_blks == 0) |
f15ab561 | 1211 | goto start_new_extent; |
b2c87cae BP |
1212 | if ((start + nr_blks) != blk) { |
1213 | if (nr_blks >= minlen) { | |
1214 | rv = sb_issue_discard(sb, | |
1215 | start, nr_blks, | |
66fc061b SW |
1216 | GFP_NOFS, 0); |
1217 | if (rv) | |
1218 | goto fail; | |
b2c87cae | 1219 | trimmed += nr_blks; |
66fc061b | 1220 | } |
b2c87cae | 1221 | nr_blks = 0; |
f15ab561 SW |
1222 | start_new_extent: |
1223 | start = blk; | |
1224 | } | |
b2c87cae | 1225 | nr_blks++; |
f15ab561 SW |
1226 | } |
1227 | diff >>= 2; | |
b2c87cae | 1228 | blk++; |
f15ab561 SW |
1229 | } |
1230 | } | |
b2c87cae BP |
1231 | if (nr_blks >= minlen) { |
1232 | rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0); | |
f15ab561 SW |
1233 | if (rv) |
1234 | goto fail; | |
b2c87cae | 1235 | trimmed += nr_blks; |
f15ab561 | 1236 | } |
66fc061b SW |
1237 | if (ptrimmed) |
1238 | *ptrimmed = trimmed; | |
1239 | return 0; | |
1240 | ||
f15ab561 | 1241 | fail: |
66fc061b SW |
1242 | if (sdp->sd_args.ar_discard) |
1243 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 1244 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1245 | return -EIO; |
1246 | } | |
1247 | ||
1248 | /** | |
1249 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1250 | * @filp: Any file on the filesystem | |
1251 | * @argp: Pointer to the arguments (also used to pass result) | |
1252 | * | |
1253 | * Returns: 0 on success, otherwise error code | |
1254 | */ | |
1255 | ||
1256 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1257 | { | |
496ad9aa | 1258 | struct inode *inode = file_inode(filp); |
66fc061b SW |
1259 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1260 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1261 | struct buffer_head *bh; | |
1262 | struct gfs2_rgrpd *rgd; | |
1263 | struct gfs2_rgrpd *rgd_end; | |
1264 | struct gfs2_holder gh; | |
1265 | struct fstrim_range r; | |
1266 | int ret = 0; | |
1267 | u64 amt; | |
1268 | u64 trimmed = 0; | |
076f0faa | 1269 | u64 start, end, minlen; |
66fc061b | 1270 | unsigned int x; |
076f0faa | 1271 | unsigned bs_shift = sdp->sd_sb.sb_bsize_shift; |
66fc061b SW |
1272 | |
1273 | if (!capable(CAP_SYS_ADMIN)) | |
1274 | return -EPERM; | |
1275 | ||
1276 | if (!blk_queue_discard(q)) | |
1277 | return -EOPNOTSUPP; | |
1278 | ||
3a238ade | 1279 | if (copy_from_user(&r, argp, sizeof(r))) |
66fc061b SW |
1280 | return -EFAULT; |
1281 | ||
5e2f7d61 BP |
1282 | ret = gfs2_rindex_update(sdp); |
1283 | if (ret) | |
1284 | return ret; | |
1285 | ||
076f0faa LC |
1286 | start = r.start >> bs_shift; |
1287 | end = start + (r.len >> bs_shift); | |
1288 | minlen = max_t(u64, r.minlen, | |
1289 | q->limits.discard_granularity) >> bs_shift; | |
1290 | ||
6a98c333 AD |
1291 | if (end <= start || minlen > sdp->sd_max_rg_data) |
1292 | return -EINVAL; | |
1293 | ||
076f0faa | 1294 | rgd = gfs2_blk2rgrpd(sdp, start, 0); |
6a98c333 | 1295 | rgd_end = gfs2_blk2rgrpd(sdp, end, 0); |
076f0faa | 1296 | |
6a98c333 AD |
1297 | if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end)) |
1298 | && (start > rgd_end->rd_data0 + rgd_end->rd_data)) | |
1299 | return -EINVAL; /* start is beyond the end of the fs */ | |
66fc061b SW |
1300 | |
1301 | while (1) { | |
1302 | ||
1303 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1304 | if (ret) | |
1305 | goto out; | |
1306 | ||
1307 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1308 | /* Trim each bitmap in the rgrp */ | |
1309 | for (x = 0; x < rgd->rd_length; x++) { | |
1310 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
076f0faa LC |
1311 | ret = gfs2_rgrp_send_discards(sdp, |
1312 | rgd->rd_data0, NULL, bi, minlen, | |
1313 | &amt); | |
66fc061b SW |
1314 | if (ret) { |
1315 | gfs2_glock_dq_uninit(&gh); | |
1316 | goto out; | |
1317 | } | |
1318 | trimmed += amt; | |
1319 | } | |
1320 | ||
1321 | /* Mark rgrp as having been trimmed */ | |
1322 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1323 | if (ret == 0) { | |
1324 | bh = rgd->rd_bits[0].bi_bh; | |
1325 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
350a9b0a | 1326 | gfs2_trans_add_meta(rgd->rd_gl, bh); |
66fc061b | 1327 | gfs2_rgrp_out(rgd, bh->b_data); |
90306c41 | 1328 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data); |
66fc061b SW |
1329 | gfs2_trans_end(sdp); |
1330 | } | |
1331 | } | |
1332 | gfs2_glock_dq_uninit(&gh); | |
1333 | ||
1334 | if (rgd == rgd_end) | |
1335 | break; | |
1336 | ||
1337 | rgd = gfs2_rgrpd_get_next(rgd); | |
1338 | } | |
1339 | ||
1340 | out: | |
6a98c333 | 1341 | r.len = trimmed << bs_shift; |
3a238ade | 1342 | if (copy_to_user(argp, &r, sizeof(r))) |
66fc061b SW |
1343 | return -EFAULT; |
1344 | ||
1345 | return ret; | |
f15ab561 SW |
1346 | } |
1347 | ||
8e2e0047 BP |
1348 | /** |
1349 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
8e2e0047 | 1350 | * @ip: the inode structure |
8e2e0047 | 1351 | * |
8e2e0047 | 1352 | */ |
ff7f4cb4 | 1353 | static void rs_insert(struct gfs2_inode *ip) |
8e2e0047 BP |
1354 | { |
1355 | struct rb_node **newn, *parent = NULL; | |
1356 | int rc; | |
1357 | struct gfs2_blkreserv *rs = ip->i_res; | |
4a993fb1 | 1358 | struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd; |
ff7f4cb4 | 1359 | u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 | 1360 | |
8e2e0047 | 1361 | BUG_ON(gfs2_rs_active(rs)); |
c743ffd0 | 1362 | |
ff7f4cb4 SW |
1363 | spin_lock(&rgd->rd_rsspin); |
1364 | newn = &rgd->rd_rstree.rb_node; | |
8e2e0047 BP |
1365 | while (*newn) { |
1366 | struct gfs2_blkreserv *cur = | |
1367 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1368 | ||
1369 | parent = *newn; | |
ff7f4cb4 | 1370 | rc = rs_cmp(fsblock, rs->rs_free, cur); |
8e2e0047 BP |
1371 | if (rc > 0) |
1372 | newn = &((*newn)->rb_right); | |
1373 | else if (rc < 0) | |
1374 | newn = &((*newn)->rb_left); | |
1375 | else { | |
1376 | spin_unlock(&rgd->rd_rsspin); | |
ff7f4cb4 SW |
1377 | WARN_ON(1); |
1378 | return; | |
8e2e0047 BP |
1379 | } |
1380 | } | |
1381 | ||
8e2e0047 BP |
1382 | rb_link_node(&rs->rs_node, parent, newn); |
1383 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1384 | ||
8e2e0047 | 1385 | /* Do our rgrp accounting for the reservation */ |
ff7f4cb4 | 1386 | rgd->rd_reserved += rs->rs_free; /* blocks reserved */ |
8e2e0047 | 1387 | spin_unlock(&rgd->rd_rsspin); |
9e733d39 | 1388 | trace_gfs2_rs(rs, TRACE_RS_INSERT); |
8e2e0047 BP |
1389 | } |
1390 | ||
1391 | /** | |
ff7f4cb4 | 1392 | * rg_mblk_search - find a group of multiple free blocks to form a reservation |
8e2e0047 | 1393 | * @rgd: the resource group descriptor |
8e2e0047 | 1394 | * @ip: pointer to the inode for which we're reserving blocks |
c743ffd0 | 1395 | * @requested: number of blocks required for this allocation |
8e2e0047 | 1396 | * |
8e2e0047 BP |
1397 | */ |
1398 | ||
ff7f4cb4 SW |
1399 | static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
1400 | unsigned requested) | |
8e2e0047 | 1401 | { |
ff7f4cb4 SW |
1402 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1403 | u64 goal; | |
1404 | struct gfs2_blkreserv *rs = ip->i_res; | |
1405 | u32 extlen; | |
1406 | u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved; | |
1407 | int ret; | |
af21ca8e | 1408 | struct inode *inode = &ip->i_inode; |
8e2e0047 | 1409 | |
af21ca8e BP |
1410 | if (S_ISDIR(inode->i_mode)) |
1411 | extlen = 1; | |
1412 | else { | |
1413 | extlen = max_t(u32, atomic_read(&rs->rs_sizehint), requested); | |
1414 | extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks); | |
1415 | } | |
ff7f4cb4 | 1416 | if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen)) |
c743ffd0 SW |
1417 | return; |
1418 | ||
8e2e0047 BP |
1419 | /* Find bitmap block that contains bits for goal block */ |
1420 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
ff7f4cb4 | 1421 | goal = ip->i_goal; |
8e2e0047 | 1422 | else |
ff7f4cb4 | 1423 | goal = rgd->rd_last_alloc + rgd->rd_data0; |
8e2e0047 | 1424 | |
ff7f4cb4 SW |
1425 | if (WARN_ON(gfs2_rbm_from_block(&rbm, goal))) |
1426 | return; | |
8e2e0047 | 1427 | |
ff7f4cb4 SW |
1428 | ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, extlen, ip, true); |
1429 | if (ret == 0) { | |
1430 | rs->rs_rbm = rbm; | |
1431 | rs->rs_free = extlen; | |
1432 | rs->rs_inum = ip->i_no_addr; | |
1433 | rs_insert(ip); | |
13d2eb01 BP |
1434 | } else { |
1435 | if (goal == rgd->rd_last_alloc + rgd->rd_data0) | |
1436 | rgd->rd_last_alloc = 0; | |
8e2e0047 | 1437 | } |
b3e47ca0 BP |
1438 | } |
1439 | ||
5b924ae2 SW |
1440 | /** |
1441 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1442 | * @rgd: The resource group | |
1443 | * @block: The starting block | |
ff7f4cb4 | 1444 | * @length: The required length |
5b924ae2 SW |
1445 | * @ip: Ignore any reservations for this inode |
1446 | * | |
1447 | * If the block does not appear in any reservation, then return the | |
1448 | * block number unchanged. If it does appear in the reservation, then | |
1449 | * keep looking through the tree of reservations in order to find the | |
1450 | * first block number which is not reserved. | |
1451 | */ | |
1452 | ||
1453 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
ff7f4cb4 | 1454 | u32 length, |
5b924ae2 SW |
1455 | const struct gfs2_inode *ip) |
1456 | { | |
1457 | struct gfs2_blkreserv *rs; | |
1458 | struct rb_node *n; | |
1459 | int rc; | |
1460 | ||
1461 | spin_lock(&rgd->rd_rsspin); | |
ff7f4cb4 | 1462 | n = rgd->rd_rstree.rb_node; |
5b924ae2 SW |
1463 | while (n) { |
1464 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
ff7f4cb4 | 1465 | rc = rs_cmp(block, length, rs); |
5b924ae2 SW |
1466 | if (rc < 0) |
1467 | n = n->rb_left; | |
1468 | else if (rc > 0) | |
1469 | n = n->rb_right; | |
1470 | else | |
1471 | break; | |
1472 | } | |
1473 | ||
1474 | if (n) { | |
ff7f4cb4 | 1475 | while ((rs_cmp(block, length, rs) == 0) && (ip->i_res != rs)) { |
5b924ae2 | 1476 | block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free; |
ff7f4cb4 | 1477 | n = n->rb_right; |
5b924ae2 SW |
1478 | if (n == NULL) |
1479 | break; | |
1480 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | spin_unlock(&rgd->rd_rsspin); | |
1485 | return block; | |
1486 | } | |
1487 | ||
5b924ae2 SW |
1488 | /** |
1489 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1490 | * @rbm: The current position in the resource group | |
ff7f4cb4 SW |
1491 | * @ip: The inode for which we are searching for blocks |
1492 | * @minext: The minimum extent length | |
5b924ae2 SW |
1493 | * |
1494 | * This checks the current position in the rgrp to see whether there is | |
1495 | * a reservation covering this block. If not then this function is a | |
1496 | * no-op. If there is, then the position is moved to the end of the | |
1497 | * contiguous reservation(s) so that we are pointing at the first | |
1498 | * non-reserved block. | |
1499 | * | |
1500 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1501 | */ | |
1502 | ||
1503 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
ff7f4cb4 SW |
1504 | const struct gfs2_inode *ip, |
1505 | u32 minext) | |
5b924ae2 SW |
1506 | { |
1507 | u64 block = gfs2_rbm_to_block(rbm); | |
ff7f4cb4 | 1508 | u32 extlen = 1; |
5b924ae2 SW |
1509 | u64 nblock; |
1510 | int ret; | |
1511 | ||
ff7f4cb4 SW |
1512 | /* |
1513 | * If we have a minimum extent length, then skip over any extent | |
1514 | * which is less than the min extent length in size. | |
1515 | */ | |
1516 | if (minext) { | |
1517 | extlen = gfs2_free_extlen(rbm, minext); | |
1518 | nblock = block + extlen; | |
1519 | if (extlen < minext) | |
1520 | goto fail; | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Check the extent which has been found against the reservations | |
1525 | * and skip if parts of it are already reserved | |
1526 | */ | |
1527 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip); | |
5b924ae2 SW |
1528 | if (nblock == block) |
1529 | return 0; | |
ff7f4cb4 | 1530 | fail: |
5b924ae2 SW |
1531 | ret = gfs2_rbm_from_block(rbm, nblock); |
1532 | if (ret < 0) | |
1533 | return ret; | |
1534 | return 1; | |
1535 | } | |
1536 | ||
1537 | /** | |
1538 | * gfs2_rbm_find - Look for blocks of a particular state | |
1539 | * @rbm: Value/result starting position and final position | |
1540 | * @state: The state which we want to find | |
ff7f4cb4 | 1541 | * @minext: The requested extent length (0 for a single block) |
5b924ae2 SW |
1542 | * @ip: If set, check for reservations |
1543 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping | |
1544 | * around until we've reached the starting point. | |
1545 | * | |
1546 | * Side effects: | |
1547 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1548 | * has no free blocks in it. | |
1549 | * | |
1550 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1551 | */ | |
1552 | ||
ff7f4cb4 | 1553 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 minext, |
5b924ae2 SW |
1554 | const struct gfs2_inode *ip, bool nowrap) |
1555 | { | |
1556 | struct buffer_head *bh; | |
1557 | struct gfs2_bitmap *initial_bi; | |
1558 | u32 initial_offset; | |
1559 | u32 offset; | |
1560 | u8 *buffer; | |
1561 | int index; | |
1562 | int n = 0; | |
1563 | int iters = rbm->rgd->rd_length; | |
1564 | int ret; | |
1565 | ||
1566 | /* If we are not starting at the beginning of a bitmap, then we | |
1567 | * need to add one to the bitmap count to ensure that we search | |
1568 | * the starting bitmap twice. | |
1569 | */ | |
1570 | if (rbm->offset != 0) | |
1571 | iters++; | |
1572 | ||
1573 | while(1) { | |
1574 | if (test_bit(GBF_FULL, &rbm->bi->bi_flags) && | |
1575 | (state == GFS2_BLKST_FREE)) | |
1576 | goto next_bitmap; | |
1577 | ||
1578 | bh = rbm->bi->bi_bh; | |
1579 | buffer = bh->b_data + rbm->bi->bi_offset; | |
1580 | WARN_ON(!buffer_uptodate(bh)); | |
1581 | if (state != GFS2_BLKST_UNLINKED && rbm->bi->bi_clone) | |
1582 | buffer = rbm->bi->bi_clone + rbm->bi->bi_offset; | |
5b924ae2 SW |
1583 | initial_offset = rbm->offset; |
1584 | offset = gfs2_bitfit(buffer, rbm->bi->bi_len, rbm->offset, state); | |
1585 | if (offset == BFITNOENT) | |
1586 | goto bitmap_full; | |
1587 | rbm->offset = offset; | |
1588 | if (ip == NULL) | |
1589 | return 0; | |
1590 | ||
1591 | initial_bi = rbm->bi; | |
ff7f4cb4 | 1592 | ret = gfs2_reservation_check_and_update(rbm, ip, minext); |
5b924ae2 SW |
1593 | if (ret == 0) |
1594 | return 0; | |
1595 | if (ret > 0) { | |
1596 | n += (rbm->bi - initial_bi); | |
8d8b752a | 1597 | goto next_iter; |
5b924ae2 | 1598 | } |
5d50d532 SW |
1599 | if (ret == -E2BIG) { |
1600 | index = 0; | |
1601 | rbm->offset = 0; | |
1602 | n += (rbm->bi - initial_bi); | |
1603 | goto res_covered_end_of_rgrp; | |
1604 | } | |
5b924ae2 SW |
1605 | return ret; |
1606 | ||
1607 | bitmap_full: /* Mark bitmap as full and fall through */ | |
1608 | if ((state == GFS2_BLKST_FREE) && initial_offset == 0) | |
1609 | set_bit(GBF_FULL, &rbm->bi->bi_flags); | |
1610 | ||
1611 | next_bitmap: /* Find next bitmap in the rgrp */ | |
1612 | rbm->offset = 0; | |
1613 | index = rbm->bi - rbm->rgd->rd_bits; | |
1614 | index++; | |
1615 | if (index == rbm->rgd->rd_length) | |
1616 | index = 0; | |
5d50d532 | 1617 | res_covered_end_of_rgrp: |
5b924ae2 SW |
1618 | rbm->bi = &rbm->rgd->rd_bits[index]; |
1619 | if ((index == 0) && nowrap) | |
1620 | break; | |
1621 | n++; | |
8d8b752a | 1622 | next_iter: |
5b924ae2 SW |
1623 | if (n >= iters) |
1624 | break; | |
1625 | } | |
1626 | ||
1627 | return -ENOSPC; | |
1628 | } | |
1629 | ||
c8cdf479 SW |
1630 | /** |
1631 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1632 | * @rgd: The rgrp | |
886b1416 BP |
1633 | * @last_unlinked: block address of the last dinode we unlinked |
1634 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1635 | * |
1a0eae88 BP |
1636 | * Returns: 0 if no error |
1637 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1638 | */ |
1639 | ||
044b9414 | 1640 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1641 | { |
5b924ae2 | 1642 | u64 block; |
5f3eae75 | 1643 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1644 | struct gfs2_glock *gl; |
1645 | struct gfs2_inode *ip; | |
1646 | int error; | |
1647 | int found = 0; | |
5b924ae2 | 1648 | struct gfs2_rbm rbm = { .rgd = rgd, .bi = rgd->rd_bits, .offset = 0 }; |
c8cdf479 | 1649 | |
5b924ae2 | 1650 | while (1) { |
5f3eae75 | 1651 | down_write(&sdp->sd_log_flush_lock); |
ff7f4cb4 | 1652 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, 0, NULL, true); |
5f3eae75 | 1653 | up_write(&sdp->sd_log_flush_lock); |
5b924ae2 SW |
1654 | if (error == -ENOSPC) |
1655 | break; | |
1656 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1657 | break; |
b3e47ca0 | 1658 | |
5b924ae2 SW |
1659 | block = gfs2_rbm_to_block(&rbm); |
1660 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1661 | break; | |
1662 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1663 | continue; |
5b924ae2 | 1664 | if (block == skip) |
1e19a195 | 1665 | continue; |
5b924ae2 | 1666 | *last_unlinked = block; |
044b9414 | 1667 | |
5b924ae2 | 1668 | error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl); |
044b9414 SW |
1669 | if (error) |
1670 | continue; | |
1671 | ||
1672 | /* If the inode is already in cache, we can ignore it here | |
1673 | * because the existing inode disposal code will deal with | |
1674 | * it when all refs have gone away. Accessing gl_object like | |
1675 | * this is not safe in general. Here it is ok because we do | |
1676 | * not dereference the pointer, and we only need an approx | |
1677 | * answer to whether it is NULL or not. | |
1678 | */ | |
1679 | ip = gl->gl_object; | |
1680 | ||
1681 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1682 | gfs2_glock_put(gl); | |
1683 | else | |
1684 | found++; | |
1685 | ||
1686 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1687 | if (found > NR_CPUS) |
044b9414 | 1688 | return; |
c8cdf479 SW |
1689 | } |
1690 | ||
1691 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1692 | return; |
c8cdf479 SW |
1693 | } |
1694 | ||
bcd97c06 SW |
1695 | /** |
1696 | * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested | |
1697 | * @rgd: The rgrp in question | |
1698 | * @loops: An indication of how picky we can be (0=very, 1=less so) | |
1699 | * | |
1700 | * This function uses the recently added glock statistics in order to | |
1701 | * figure out whether a parciular resource group is suffering from | |
1702 | * contention from multiple nodes. This is done purely on the basis | |
1703 | * of timings, since this is the only data we have to work with and | |
1704 | * our aim here is to reject a resource group which is highly contended | |
1705 | * but (very important) not to do this too often in order to ensure that | |
1706 | * we do not land up introducing fragmentation by changing resource | |
1707 | * groups when not actually required. | |
1708 | * | |
1709 | * The calculation is fairly simple, we want to know whether the SRTTB | |
1710 | * (i.e. smoothed round trip time for blocking operations) to acquire | |
1711 | * the lock for this rgrp's glock is significantly greater than the | |
1712 | * time taken for resource groups on average. We introduce a margin in | |
1713 | * the form of the variable @var which is computed as the sum of the two | |
1714 | * respective variences, and multiplied by a factor depending on @loops | |
1715 | * and whether we have a lot of data to base the decision on. This is | |
1716 | * then tested against the square difference of the means in order to | |
1717 | * decide whether the result is statistically significant or not. | |
1718 | * | |
1719 | * Returns: A boolean verdict on the congestion status | |
1720 | */ | |
1721 | ||
1722 | static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops) | |
1723 | { | |
1724 | const struct gfs2_glock *gl = rgd->rd_gl; | |
1725 | const struct gfs2_sbd *sdp = gl->gl_sbd; | |
1726 | struct gfs2_lkstats *st; | |
1727 | s64 r_dcount, l_dcount; | |
1728 | s64 r_srttb, l_srttb; | |
1729 | s64 srttb_diff; | |
1730 | s64 sqr_diff; | |
1731 | s64 var; | |
1732 | ||
1733 | preempt_disable(); | |
1734 | st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP]; | |
1735 | r_srttb = st->stats[GFS2_LKS_SRTTB]; | |
1736 | r_dcount = st->stats[GFS2_LKS_DCOUNT]; | |
1737 | var = st->stats[GFS2_LKS_SRTTVARB] + | |
1738 | gl->gl_stats.stats[GFS2_LKS_SRTTVARB]; | |
1739 | preempt_enable(); | |
1740 | ||
1741 | l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB]; | |
1742 | l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT]; | |
1743 | ||
1744 | if ((l_dcount < 1) || (r_dcount < 1) || (r_srttb == 0)) | |
1745 | return false; | |
1746 | ||
1747 | srttb_diff = r_srttb - l_srttb; | |
1748 | sqr_diff = srttb_diff * srttb_diff; | |
1749 | ||
1750 | var *= 2; | |
1751 | if (l_dcount < 8 || r_dcount < 8) | |
1752 | var *= 2; | |
1753 | if (loops == 1) | |
1754 | var *= 2; | |
1755 | ||
1756 | return ((srttb_diff < 0) && (sqr_diff > var)); | |
1757 | } | |
1758 | ||
1759 | /** | |
1760 | * gfs2_rgrp_used_recently | |
1761 | * @rs: The block reservation with the rgrp to test | |
1762 | * @msecs: The time limit in milliseconds | |
1763 | * | |
1764 | * Returns: True if the rgrp glock has been used within the time limit | |
1765 | */ | |
1766 | static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs, | |
1767 | u64 msecs) | |
1768 | { | |
1769 | u64 tdiff; | |
1770 | ||
1771 | tdiff = ktime_to_ns(ktime_sub(ktime_get_real(), | |
1772 | rs->rs_rbm.rgd->rd_gl->gl_dstamp)); | |
1773 | ||
1774 | return tdiff > (msecs * 1000 * 1000); | |
1775 | } | |
1776 | ||
9dbe9610 SW |
1777 | static u32 gfs2_orlov_skip(const struct gfs2_inode *ip) |
1778 | { | |
1779 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1780 | u32 skip; | |
1781 | ||
1782 | get_random_bytes(&skip, sizeof(skip)); | |
1783 | return skip % sdp->sd_rgrps; | |
1784 | } | |
1785 | ||
c743ffd0 SW |
1786 | static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin) |
1787 | { | |
1788 | struct gfs2_rgrpd *rgd = *pos; | |
aa8920c9 | 1789 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
c743ffd0 SW |
1790 | |
1791 | rgd = gfs2_rgrpd_get_next(rgd); | |
1792 | if (rgd == NULL) | |
aa8920c9 | 1793 | rgd = gfs2_rgrpd_get_first(sdp); |
c743ffd0 SW |
1794 | *pos = rgd; |
1795 | if (rgd != begin) /* If we didn't wrap */ | |
1796 | return true; | |
1797 | return false; | |
1798 | } | |
1799 | ||
b3b94faa | 1800 | /** |
666d1d8a | 1801 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1802 | * @ip: the inode to reserve space for |
666d1d8a | 1803 | * @requested: the number of blocks to be reserved |
b3b94faa DT |
1804 | * |
1805 | * Returns: errno | |
1806 | */ | |
1807 | ||
9dbe9610 | 1808 | int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested, u32 aflags) |
b3b94faa | 1809 | { |
feaa7bba | 1810 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 1811 | struct gfs2_rgrpd *begin = NULL; |
564e12b1 | 1812 | struct gfs2_blkreserv *rs = ip->i_res; |
bcd97c06 | 1813 | int error = 0, rg_locked, flags = 0; |
666d1d8a | 1814 | u64 last_unlinked = NO_BLOCK; |
7c9ca621 | 1815 | int loops = 0; |
9dbe9610 | 1816 | u32 skip = 0; |
b3b94faa | 1817 | |
90306c41 BM |
1818 | if (sdp->sd_args.ar_rgrplvb) |
1819 | flags |= GL_SKIP; | |
c743ffd0 SW |
1820 | if (gfs2_assert_warn(sdp, requested)) |
1821 | return -EINVAL; | |
8e2e0047 | 1822 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1823 | begin = rs->rs_rbm.rgd; |
8e2e0047 BP |
1824 | flags = 0; /* Yoda: Do or do not. There is no try */ |
1825 | } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) { | |
4a993fb1 | 1826 | rs->rs_rbm.rgd = begin = ip->i_rgd; |
8e2e0047 | 1827 | } else { |
4a993fb1 | 1828 | rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 1829 | } |
9dbe9610 SW |
1830 | if (S_ISDIR(ip->i_inode.i_mode) && (aflags & GFS2_AF_ORLOV)) |
1831 | skip = gfs2_orlov_skip(ip); | |
4a993fb1 | 1832 | if (rs->rs_rbm.rgd == NULL) |
7c9ca621 BP |
1833 | return -EBADSLT; |
1834 | ||
1835 | while (loops < 3) { | |
c743ffd0 SW |
1836 | rg_locked = 1; |
1837 | ||
1838 | if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) { | |
1839 | rg_locked = 0; | |
9dbe9610 SW |
1840 | if (skip && skip--) |
1841 | goto next_rgrp; | |
bcd97c06 SW |
1842 | if (!gfs2_rs_active(rs) && (loops < 2) && |
1843 | gfs2_rgrp_used_recently(rs, 1000) && | |
1844 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
1845 | goto next_rgrp; | |
4a993fb1 | 1846 | error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl, |
8e2e0047 BP |
1847 | LM_ST_EXCLUSIVE, flags, |
1848 | &rs->rs_rgd_gh); | |
c743ffd0 SW |
1849 | if (unlikely(error)) |
1850 | return error; | |
bcd97c06 SW |
1851 | if (!gfs2_rs_active(rs) && (loops < 2) && |
1852 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
1853 | goto skip_rgrp; | |
c743ffd0 | 1854 | if (sdp->sd_args.ar_rgrplvb) { |
4a993fb1 | 1855 | error = update_rgrp_lvb(rs->rs_rbm.rgd); |
c743ffd0 | 1856 | if (unlikely(error)) { |
90306c41 BM |
1857 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
1858 | return error; | |
1859 | } | |
1860 | } | |
292c8c14 | 1861 | } |
666d1d8a | 1862 | |
c743ffd0 SW |
1863 | /* Skip unuseable resource groups */ |
1864 | if (rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) | |
1865 | goto skip_rgrp; | |
1866 | ||
1867 | if (sdp->sd_args.ar_rgrplvb) | |
1868 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); | |
1869 | ||
1870 | /* Get a reservation if we don't already have one */ | |
1871 | if (!gfs2_rs_active(rs)) | |
1872 | rg_mblk_search(rs->rs_rbm.rgd, ip, requested); | |
1873 | ||
1874 | /* Skip rgrps when we can't get a reservation on first pass */ | |
1875 | if (!gfs2_rs_active(rs) && (loops < 1)) | |
1876 | goto check_rgrp; | |
1877 | ||
1878 | /* If rgrp has enough free space, use it */ | |
1879 | if (rs->rs_rbm.rgd->rd_free_clone >= requested) { | |
1880 | ip->i_rgd = rs->rs_rbm.rgd; | |
1881 | return 0; | |
b3b94faa | 1882 | } |
c743ffd0 SW |
1883 | |
1884 | /* Drop reservation, if we couldn't use reserved rgrp */ | |
1885 | if (gfs2_rs_active(rs)) | |
20095218 | 1886 | gfs2_rs_deltree(rs); |
c743ffd0 SW |
1887 | check_rgrp: |
1888 | /* Check for unlinked inodes which can be reclaimed */ | |
1889 | if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK) | |
1890 | try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked, | |
1891 | ip->i_no_addr); | |
1892 | skip_rgrp: | |
1893 | /* Unlock rgrp if required */ | |
1894 | if (!rg_locked) | |
1895 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
1896 | next_rgrp: | |
1897 | /* Find the next rgrp, and continue looking */ | |
1898 | if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin)) | |
1899 | continue; | |
9dbe9610 SW |
1900 | if (skip) |
1901 | continue; | |
c743ffd0 SW |
1902 | |
1903 | /* If we've scanned all the rgrps, but found no free blocks | |
1904 | * then this checks for some less likely conditions before | |
1905 | * trying again. | |
1906 | */ | |
c743ffd0 SW |
1907 | loops++; |
1908 | /* Check that fs hasn't grown if writing to rindex */ | |
1909 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
1910 | error = gfs2_ri_update(ip); | |
1911 | if (error) | |
1912 | return error; | |
1913 | } | |
1914 | /* Flushing the log may release space */ | |
1915 | if (loops == 2) | |
1916 | gfs2_log_flush(sdp, NULL); | |
b3b94faa | 1917 | } |
b3b94faa | 1918 | |
c743ffd0 | 1919 | return -ENOSPC; |
b3b94faa DT |
1920 | } |
1921 | ||
1922 | /** | |
1923 | * gfs2_inplace_release - release an inplace reservation | |
1924 | * @ip: the inode the reservation was taken out on | |
1925 | * | |
1926 | * Release a reservation made by gfs2_inplace_reserve(). | |
1927 | */ | |
1928 | ||
1929 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
1930 | { | |
564e12b1 | 1931 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 1932 | |
564e12b1 BP |
1933 | if (rs->rs_rgd_gh.gh_gl) |
1934 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
b3b94faa DT |
1935 | } |
1936 | ||
1937 | /** | |
1938 | * gfs2_get_block_type - Check a block in a RG is of given type | |
1939 | * @rgd: the resource group holding the block | |
1940 | * @block: the block number | |
1941 | * | |
1942 | * Returns: The block type (GFS2_BLKST_*) | |
1943 | */ | |
1944 | ||
acf7e244 | 1945 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 1946 | { |
3983903a SW |
1947 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1948 | int ret; | |
b3b94faa | 1949 | |
3983903a SW |
1950 | ret = gfs2_rbm_from_block(&rbm, block); |
1951 | WARN_ON_ONCE(ret != 0); | |
b3b94faa | 1952 | |
c04a2ef3 | 1953 | return gfs2_testbit(&rbm); |
b3b94faa DT |
1954 | } |
1955 | ||
60a0b8f9 | 1956 | |
b3e47ca0 BP |
1957 | /** |
1958 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 1959 | * @rbm: the resource group information |
b3e47ca0 | 1960 | * @dinode: TRUE if the first block we allocate is for a dinode |
c04a2ef3 | 1961 | * @n: The extent length (value/result) |
b3e47ca0 | 1962 | * |
c04a2ef3 | 1963 | * Add the bitmap buffer to the transaction. |
b3e47ca0 | 1964 | * Set the found bits to @new_state to change block's allocation state. |
b3e47ca0 | 1965 | */ |
c04a2ef3 | 1966 | static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
4a993fb1 | 1967 | unsigned int *n) |
b3e47ca0 | 1968 | { |
c04a2ef3 | 1969 | struct gfs2_rbm pos = { .rgd = rbm->rgd, }; |
b3e47ca0 | 1970 | const unsigned int elen = *n; |
c04a2ef3 SW |
1971 | u64 block; |
1972 | int ret; | |
b3e47ca0 | 1973 | |
c04a2ef3 SW |
1974 | *n = 1; |
1975 | block = gfs2_rbm_to_block(rbm); | |
350a9b0a | 1976 | gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm->bi->bi_bh); |
3e6339dd | 1977 | gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
c04a2ef3 | 1978 | block++; |
60a0b8f9 | 1979 | while (*n < elen) { |
c04a2ef3 | 1980 | ret = gfs2_rbm_from_block(&pos, block); |
0688a5ec | 1981 | if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE) |
60a0b8f9 | 1982 | break; |
350a9b0a | 1983 | gfs2_trans_add_meta(pos.rgd->rd_gl, pos.bi->bi_bh); |
3e6339dd | 1984 | gfs2_setbit(&pos, true, GFS2_BLKST_USED); |
60a0b8f9 | 1985 | (*n)++; |
c04a2ef3 | 1986 | block++; |
c8cdf479 | 1987 | } |
b3b94faa DT |
1988 | } |
1989 | ||
1990 | /** | |
1991 | * rgblk_free - Change alloc state of given block(s) | |
1992 | * @sdp: the filesystem | |
1993 | * @bstart: the start of a run of blocks to free | |
1994 | * @blen: the length of the block run (all must lie within ONE RG!) | |
1995 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
1996 | * | |
1997 | * Returns: Resource group containing the block(s) | |
1998 | */ | |
1999 | ||
cd915493 SW |
2000 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
2001 | u32 blen, unsigned char new_state) | |
b3b94faa | 2002 | { |
3b1d0b9d | 2003 | struct gfs2_rbm rbm; |
b3b94faa | 2004 | |
3b1d0b9d SW |
2005 | rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
2006 | if (!rbm.rgd) { | |
b3b94faa | 2007 | if (gfs2_consist(sdp)) |
382066da | 2008 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
2009 | return NULL; |
2010 | } | |
2011 | ||
b3b94faa | 2012 | while (blen--) { |
3b1d0b9d SW |
2013 | gfs2_rbm_from_block(&rbm, bstart); |
2014 | bstart++; | |
2015 | if (!rbm.bi->bi_clone) { | |
2016 | rbm.bi->bi_clone = kmalloc(rbm.bi->bi_bh->b_size, | |
2017 | GFP_NOFS | __GFP_NOFAIL); | |
2018 | memcpy(rbm.bi->bi_clone + rbm.bi->bi_offset, | |
2019 | rbm.bi->bi_bh->b_data + rbm.bi->bi_offset, | |
2020 | rbm.bi->bi_len); | |
b3b94faa | 2021 | } |
350a9b0a | 2022 | gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.bi->bi_bh); |
3e6339dd | 2023 | gfs2_setbit(&rbm, false, new_state); |
b3b94faa DT |
2024 | } |
2025 | ||
3b1d0b9d | 2026 | return rbm.rgd; |
b3b94faa DT |
2027 | } |
2028 | ||
2029 | /** | |
09010978 SW |
2030 | * gfs2_rgrp_dump - print out an rgrp |
2031 | * @seq: The iterator | |
2032 | * @gl: The glock in question | |
2033 | * | |
2034 | */ | |
2035 | ||
2036 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
2037 | { | |
8e2e0047 BP |
2038 | struct gfs2_rgrpd *rgd = gl->gl_object; |
2039 | struct gfs2_blkreserv *trs; | |
2040 | const struct rb_node *n; | |
2041 | ||
09010978 SW |
2042 | if (rgd == NULL) |
2043 | return 0; | |
8e2e0047 | 2044 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u\n", |
09010978 | 2045 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 BP |
2046 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
2047 | rgd->rd_reserved); | |
2048 | spin_lock(&rgd->rd_rsspin); | |
2049 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { | |
2050 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
2051 | dump_rs(seq, trs); | |
2052 | } | |
2053 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
2054 | return 0; |
2055 | } | |
2056 | ||
6050b9c7 SW |
2057 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
2058 | { | |
2059 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2060 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 2061 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
2062 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
2063 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
2064 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
2065 | } | |
2066 | ||
8e2e0047 | 2067 | /** |
5b924ae2 SW |
2068 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
2069 | * @ip: The inode we have just allocated blocks for | |
2070 | * @rbm: The start of the allocated blocks | |
2071 | * @len: The extent length | |
8e2e0047 | 2072 | * |
5b924ae2 SW |
2073 | * Adjusts a reservation after an allocation has taken place. If the |
2074 | * reservation does not match the allocation, or if it is now empty | |
2075 | * then it is removed. | |
8e2e0047 | 2076 | */ |
5b924ae2 SW |
2077 | |
2078 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
2079 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 BP |
2080 | { |
2081 | struct gfs2_blkreserv *rs = ip->i_res; | |
5b924ae2 SW |
2082 | struct gfs2_rgrpd *rgd = rbm->rgd; |
2083 | unsigned rlen; | |
2084 | u64 block; | |
2085 | int ret; | |
8e2e0047 | 2086 | |
5b924ae2 SW |
2087 | spin_lock(&rgd->rd_rsspin); |
2088 | if (gfs2_rs_active(rs)) { | |
2089 | if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) { | |
2090 | block = gfs2_rbm_to_block(rbm); | |
2091 | ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len); | |
2092 | rlen = min(rs->rs_free, len); | |
2093 | rs->rs_free -= rlen; | |
2094 | rgd->rd_reserved -= rlen; | |
9e733d39 | 2095 | trace_gfs2_rs(rs, TRACE_RS_CLAIM); |
5b924ae2 SW |
2096 | if (rs->rs_free && !ret) |
2097 | goto out; | |
2098 | } | |
20095218 | 2099 | __rs_deltree(rs); |
8e2e0047 | 2100 | } |
5b924ae2 SW |
2101 | out: |
2102 | spin_unlock(&rgd->rd_rsspin); | |
8e2e0047 BP |
2103 | } |
2104 | ||
09010978 | 2105 | /** |
6e87ed0f | 2106 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 2107 | * @ip: the inode to allocate the block for |
09010978 | 2108 | * @bn: Used to return the starting block number |
8e2e0047 | 2109 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 2110 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 2111 | * @generation: the generation number of the inode |
b3b94faa | 2112 | * |
09010978 | 2113 | * Returns: 0 or error |
b3b94faa DT |
2114 | */ |
2115 | ||
6a8099ed | 2116 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 2117 | bool dinode, u64 *generation) |
b3b94faa | 2118 | { |
feaa7bba | 2119 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 2120 | struct buffer_head *dibh; |
4a993fb1 | 2121 | struct gfs2_rbm rbm = { .rgd = ip->i_rgd, }; |
6a8099ed | 2122 | unsigned int ndata; |
5b924ae2 | 2123 | u64 goal; |
3c5d785a | 2124 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 2125 | int error; |
b3b94faa | 2126 | |
5b924ae2 SW |
2127 | if (gfs2_rs_active(ip->i_res)) |
2128 | goal = gfs2_rbm_to_block(&ip->i_res->rs_rbm); | |
2129 | else if (!dinode && rgrp_contains_block(rbm.rgd, ip->i_goal)) | |
2130 | goal = ip->i_goal; | |
62e252ee | 2131 | else |
5b924ae2 | 2132 | goal = rbm.rgd->rd_last_alloc + rbm.rgd->rd_data0; |
62e252ee | 2133 | |
5b924ae2 | 2134 | gfs2_rbm_from_block(&rbm, goal); |
ff7f4cb4 | 2135 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, 0, ip, false); |
62e252ee | 2136 | |
137834a6 SW |
2137 | if (error == -ENOSPC) { |
2138 | gfs2_rbm_from_block(&rbm, goal); | |
ff7f4cb4 | 2139 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, 0, NULL, false); |
137834a6 SW |
2140 | } |
2141 | ||
62e252ee | 2142 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
5b924ae2 | 2143 | if (error) { |
9e733d39 SW |
2144 | fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d\n", |
2145 | (unsigned long long)ip->i_no_addr, error, *nblocks, | |
5b924ae2 | 2146 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags)); |
62e252ee | 2147 | goto rgrp_error; |
8e2e0047 | 2148 | } |
62e252ee | 2149 | |
c04a2ef3 SW |
2150 | gfs2_alloc_extent(&rbm, dinode, nblocks); |
2151 | block = gfs2_rbm_to_block(&rbm); | |
c743ffd0 | 2152 | rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0; |
5b924ae2 SW |
2153 | if (gfs2_rs_active(ip->i_res)) |
2154 | gfs2_adjust_reservation(ip, &rbm, *nblocks); | |
6a8099ed SW |
2155 | ndata = *nblocks; |
2156 | if (dinode) | |
2157 | ndata--; | |
b3e47ca0 | 2158 | |
3c5d785a | 2159 | if (!dinode) { |
6a8099ed | 2160 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
2161 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2162 | if (error == 0) { | |
2163 | struct gfs2_dinode *di = | |
2164 | (struct gfs2_dinode *)dibh->b_data; | |
350a9b0a | 2165 | gfs2_trans_add_meta(ip->i_gl, dibh); |
3c5d785a BP |
2166 | di->di_goal_meta = di->di_goal_data = |
2167 | cpu_to_be64(ip->i_goal); | |
2168 | brelse(dibh); | |
2169 | } | |
d9ba7615 | 2170 | } |
4a993fb1 | 2171 | if (rbm.rgd->rd_free < *nblocks) { |
8e2e0047 | 2172 | printk(KERN_WARNING "nblocks=%u\n", *nblocks); |
09010978 | 2173 | goto rgrp_error; |
8e2e0047 | 2174 | } |
09010978 | 2175 | |
4a993fb1 | 2176 | rbm.rgd->rd_free -= *nblocks; |
3c5d785a | 2177 | if (dinode) { |
4a993fb1 SW |
2178 | rbm.rgd->rd_dinodes++; |
2179 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2180 | if (*generation == 0) |
4a993fb1 | 2181 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2182 | } |
b3b94faa | 2183 | |
350a9b0a | 2184 | gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh); |
4a993fb1 SW |
2185 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); |
2186 | gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
b3b94faa | 2187 | |
6a8099ed | 2188 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a BP |
2189 | if (dinode) |
2190 | gfs2_trans_add_unrevoke(sdp, block, 1); | |
6a8099ed | 2191 | |
fd4b4e04 | 2192 | gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa | 2193 | |
4a993fb1 SW |
2194 | rbm.rgd->rd_free_clone -= *nblocks; |
2195 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, | |
6e87ed0f | 2196 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2197 | *bn = block; |
2198 | return 0; | |
2199 | ||
2200 | rgrp_error: | |
4a993fb1 | 2201 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2202 | return -EIO; |
b3b94faa DT |
2203 | } |
2204 | ||
2205 | /** | |
46fcb2ed | 2206 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
2207 | * @ip: the inode these blocks are being freed from |
2208 | * @bstart: first block of a run of contiguous blocks | |
2209 | * @blen: the length of the block run | |
46fcb2ed | 2210 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2211 | * |
2212 | */ | |
2213 | ||
46fcb2ed | 2214 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 2215 | { |
feaa7bba | 2216 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2217 | struct gfs2_rgrpd *rgd; |
2218 | ||
2219 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
2220 | if (!rgd) | |
2221 | return; | |
41db1ab9 | 2222 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2223 | rgd->rd_free += blen; |
66fc061b | 2224 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
350a9b0a | 2225 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2226 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 | 2227 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 2228 | |
6d3117b4 | 2229 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 2230 | if (meta || ip->i_depth) |
6d3117b4 | 2231 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 2232 | } |
b3b94faa | 2233 | |
4c16c36a BP |
2234 | /** |
2235 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2236 | * @ip: the inode these blocks are being freed from | |
2237 | * @bstart: first block of a run of contiguous blocks | |
2238 | * @blen: the length of the block run | |
2239 | * | |
2240 | */ | |
2241 | ||
2242 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
2243 | { | |
2244 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2245 | ||
46fcb2ed | 2246 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 2247 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2248 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2249 | } |
2250 | ||
feaa7bba SW |
2251 | void gfs2_unlink_di(struct inode *inode) |
2252 | { | |
2253 | struct gfs2_inode *ip = GFS2_I(inode); | |
2254 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2255 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2256 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
2257 | |
2258 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
2259 | if (!rgd) | |
2260 | return; | |
41db1ab9 | 2261 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
350a9b0a | 2262 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2263 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2264 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2265 | update_rgrp_lvb_unlinked(rgd, 1); | |
feaa7bba SW |
2266 | } |
2267 | ||
cd915493 | 2268 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
2269 | { |
2270 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2271 | struct gfs2_rgrpd *tmp_rgd; | |
2272 | ||
2273 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
2274 | if (!tmp_rgd) | |
2275 | return; | |
2276 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
2277 | ||
73f74948 | 2278 | if (!rgd->rd_dinodes) |
b3b94faa | 2279 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2280 | rgd->rd_dinodes--; |
cfc8b549 | 2281 | rgd->rd_free++; |
b3b94faa | 2282 | |
350a9b0a | 2283 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2284 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2285 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2286 | update_rgrp_lvb_unlinked(rgd, -1); | |
b3b94faa DT |
2287 | |
2288 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
2289 | } |
2290 | ||
b3b94faa DT |
2291 | |
2292 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
2293 | { | |
dbb7cae2 | 2294 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
41db1ab9 | 2295 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2296 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 2297 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2298 | } |
2299 | ||
acf7e244 SW |
2300 | /** |
2301 | * gfs2_check_blk_type - Check the type of a block | |
2302 | * @sdp: The superblock | |
2303 | * @no_addr: The block number to check | |
2304 | * @type: The block type we are looking for | |
2305 | * | |
2306 | * Returns: 0 if the block type matches the expected type | |
2307 | * -ESTALE if it doesn't match | |
2308 | * or -ve errno if something went wrong while checking | |
2309 | */ | |
2310 | ||
2311 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2312 | { | |
2313 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2314 | struct gfs2_holder rgd_gh; |
58884c4d | 2315 | int error = -EINVAL; |
acf7e244 | 2316 | |
66fc061b | 2317 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2318 | if (!rgd) |
8339ee54 | 2319 | goto fail; |
acf7e244 SW |
2320 | |
2321 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2322 | if (error) | |
8339ee54 | 2323 | goto fail; |
acf7e244 SW |
2324 | |
2325 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
2326 | error = -ESTALE; | |
2327 | ||
2328 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
2329 | fail: |
2330 | return error; | |
2331 | } | |
2332 | ||
b3b94faa DT |
2333 | /** |
2334 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2335 | * @ip: the inode |
b3b94faa DT |
2336 | * @rlist: the list of resource groups |
2337 | * @block: the block | |
2338 | * | |
2339 | * Figure out what RG a block belongs to and add that RG to the list | |
2340 | * | |
2341 | * FIXME: Don't use NOFAIL | |
2342 | * | |
2343 | */ | |
2344 | ||
70b0c365 | 2345 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2346 | u64 block) |
b3b94faa | 2347 | { |
70b0c365 | 2348 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2349 | struct gfs2_rgrpd *rgd; |
2350 | struct gfs2_rgrpd **tmp; | |
2351 | unsigned int new_space; | |
2352 | unsigned int x; | |
2353 | ||
2354 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2355 | return; | |
2356 | ||
70b0c365 SW |
2357 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
2358 | rgd = ip->i_rgd; | |
2359 | else | |
66fc061b | 2360 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 2361 | if (!rgd) { |
70b0c365 | 2362 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
2363 | return; |
2364 | } | |
70b0c365 | 2365 | ip->i_rgd = rgd; |
b3b94faa DT |
2366 | |
2367 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2368 | if (rlist->rl_rgd[x] == rgd) | |
2369 | return; | |
2370 | ||
2371 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2372 | new_space = rlist->rl_space + 10; | |
2373 | ||
2374 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2375 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2376 | |
2377 | if (rlist->rl_rgd) { | |
2378 | memcpy(tmp, rlist->rl_rgd, | |
2379 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2380 | kfree(rlist->rl_rgd); | |
2381 | } | |
2382 | ||
2383 | rlist->rl_space = new_space; | |
2384 | rlist->rl_rgd = tmp; | |
2385 | } | |
2386 | ||
2387 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2388 | } | |
2389 | ||
2390 | /** | |
2391 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2392 | * and initialize an array of glock holders for them | |
2393 | * @rlist: the list of resource groups | |
2394 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
2395 | * |
2396 | * FIXME: Don't use NOFAIL | |
2397 | * | |
2398 | */ | |
2399 | ||
fe6c991c | 2400 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
2401 | { |
2402 | unsigned int x; | |
2403 | ||
2404 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 2405 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2406 | for (x = 0; x < rlist->rl_rgrps; x++) |
2407 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 2408 | state, 0, |
b3b94faa DT |
2409 | &rlist->rl_ghs[x]); |
2410 | } | |
2411 | ||
2412 | /** | |
2413 | * gfs2_rlist_free - free a resource group list | |
2414 | * @list: the list of resource groups | |
2415 | * | |
2416 | */ | |
2417 | ||
2418 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2419 | { | |
2420 | unsigned int x; | |
2421 | ||
2422 | kfree(rlist->rl_rgd); | |
2423 | ||
2424 | if (rlist->rl_ghs) { | |
2425 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2426 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2427 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2428 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2429 | } |
2430 | } | |
2431 |