]>
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 | ||
d77d1b58 JP |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
b3b94faa DT |
12 | #include <linux/slab.h> |
13 | #include <linux/spinlock.h> | |
14 | #include <linux/completion.h> | |
15 | #include <linux/buffer_head.h> | |
f42faf4f | 16 | #include <linux/fs.h> |
5c676f6d | 17 | #include <linux/gfs2_ondisk.h> |
1f466a47 | 18 | #include <linux/prefetch.h> |
f15ab561 | 19 | #include <linux/blkdev.h> |
7c9ca621 | 20 | #include <linux/rbtree.h> |
9dbe9610 | 21 | #include <linux/random.h> |
b3b94faa DT |
22 | |
23 | #include "gfs2.h" | |
5c676f6d | 24 | #include "incore.h" |
b3b94faa DT |
25 | #include "glock.h" |
26 | #include "glops.h" | |
b3b94faa DT |
27 | #include "lops.h" |
28 | #include "meta_io.h" | |
29 | #include "quota.h" | |
30 | #include "rgrp.h" | |
31 | #include "super.h" | |
32 | #include "trans.h" | |
5c676f6d | 33 | #include "util.h" |
172e045a | 34 | #include "log.h" |
c8cdf479 | 35 | #include "inode.h" |
63997775 | 36 | #include "trace_gfs2.h" |
b3b94faa | 37 | |
2c1e52aa | 38 | #define BFITNOENT ((u32)~0) |
6760bdcd | 39 | #define NO_BLOCK ((u64)~0) |
88c8ab1f | 40 | |
1f466a47 BP |
41 | #if BITS_PER_LONG == 32 |
42 | #define LBITMASK (0x55555555UL) | |
43 | #define LBITSKIP55 (0x55555555UL) | |
44 | #define LBITSKIP00 (0x00000000UL) | |
45 | #else | |
46 | #define LBITMASK (0x5555555555555555UL) | |
47 | #define LBITSKIP55 (0x5555555555555555UL) | |
48 | #define LBITSKIP00 (0x0000000000000000UL) | |
49 | #endif | |
50 | ||
88c8ab1f SW |
51 | /* |
52 | * These routines are used by the resource group routines (rgrp.c) | |
53 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
54 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
55 | * | |
56 | * 0 = Free | |
57 | * 1 = Used (not metadata) | |
58 | * 2 = Unlinked (still in use) inode | |
59 | * 3 = Used (metadata) | |
88c8ab1f SW |
60 | */ |
61 | ||
5ce13431 BP |
62 | struct gfs2_extent { |
63 | struct gfs2_rbm rbm; | |
64 | u32 len; | |
65 | }; | |
66 | ||
88c8ab1f SW |
67 | static const char valid_change[16] = { |
68 | /* current */ | |
feaa7bba | 69 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 70 | /* e */ 1, 0, 0, 0, |
feaa7bba | 71 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
72 | 1, 0, 0, 0 |
73 | }; | |
74 | ||
5ce13431 | 75 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
8381e602 | 76 | const struct gfs2_inode *ip, bool nowrap); |
ff7f4cb4 SW |
77 | |
78 | ||
88c8ab1f SW |
79 | /** |
80 | * gfs2_setbit - Set a bit in the bitmaps | |
3e6339dd SW |
81 | * @rbm: The position of the bit to set |
82 | * @do_clone: Also set the clone bitmap, if it exists | |
88c8ab1f SW |
83 | * @new_state: the new state of the block |
84 | * | |
85 | */ | |
86 | ||
3e6339dd | 87 | static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone, |
06344b91 | 88 | unsigned char new_state) |
88c8ab1f | 89 | { |
b45e41d7 | 90 | unsigned char *byte1, *byte2, *end, cur_state; |
e579ed4f BP |
91 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
92 | unsigned int buflen = bi->bi_len; | |
3e6339dd | 93 | const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 94 | |
e579ed4f BP |
95 | byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY); |
96 | end = bi->bi_bh->b_data + bi->bi_offset + buflen; | |
88c8ab1f | 97 | |
b45e41d7 | 98 | BUG_ON(byte1 >= end); |
88c8ab1f | 99 | |
b45e41d7 | 100 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 101 | |
b45e41d7 | 102 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
d77d1b58 JP |
103 | pr_warn("buf_blk = 0x%x old_state=%d, new_state=%d\n", |
104 | rbm->offset, cur_state, new_state); | |
105 | pr_warn("rgrp=0x%llx bi_start=0x%x\n", | |
106 | (unsigned long long)rbm->rgd->rd_addr, bi->bi_start); | |
107 | pr_warn("bi_offset=0x%x bi_len=0x%x\n", | |
108 | bi->bi_offset, bi->bi_len); | |
95c8e17f | 109 | dump_stack(); |
3e6339dd | 110 | gfs2_consist_rgrpd(rbm->rgd); |
b45e41d7 SW |
111 | return; |
112 | } | |
113 | *byte1 ^= (cur_state ^ new_state) << bit; | |
114 | ||
e579ed4f BP |
115 | if (do_clone && bi->bi_clone) { |
116 | byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY); | |
b45e41d7 SW |
117 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
118 | *byte2 ^= (cur_state ^ new_state) << bit; | |
119 | } | |
88c8ab1f SW |
120 | } |
121 | ||
122 | /** | |
123 | * gfs2_testbit - test a bit in the bitmaps | |
c04a2ef3 | 124 | * @rbm: The bit to test |
88c8ab1f | 125 | * |
c04a2ef3 | 126 | * Returns: The two bit block state of the requested bit |
88c8ab1f SW |
127 | */ |
128 | ||
c04a2ef3 | 129 | static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm) |
88c8ab1f | 130 | { |
e579ed4f BP |
131 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
132 | const u8 *buffer = bi->bi_bh->b_data + bi->bi_offset; | |
c04a2ef3 | 133 | const u8 *byte; |
88c8ab1f SW |
134 | unsigned int bit; |
135 | ||
c04a2ef3 SW |
136 | byte = buffer + (rbm->offset / GFS2_NBBY); |
137 | bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 138 | |
c04a2ef3 | 139 | return (*byte >> bit) & GFS2_BIT_MASK; |
88c8ab1f SW |
140 | } |
141 | ||
223b2b88 SW |
142 | /** |
143 | * gfs2_bit_search | |
144 | * @ptr: Pointer to bitmap data | |
145 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
146 | * @state: The state we are searching for | |
147 | * | |
148 | * We xor the bitmap data with a patter which is the bitwise opposite | |
149 | * of what we are looking for, this gives rise to a pattern of ones | |
150 | * wherever there is a match. Since we have two bits per entry, we | |
151 | * take this pattern, shift it down by one place and then and it with | |
152 | * the original. All the even bit positions (0,2,4, etc) then represent | |
153 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
154 | * odd bit positions. | |
155 | * | |
156 | * This allows searching of a whole u64 at once (32 blocks) with a | |
157 | * single test (on 64 bit arches). | |
158 | */ | |
159 | ||
160 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
161 | { | |
162 | u64 tmp; | |
163 | static const u64 search[] = { | |
075ac448 HE |
164 | [0] = 0xffffffffffffffffULL, |
165 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
166 | [2] = 0x5555555555555555ULL, | |
167 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
168 | }; |
169 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
170 | tmp &= (tmp >> 1); | |
171 | tmp &= mask; | |
172 | return tmp; | |
173 | } | |
174 | ||
8e2e0047 BP |
175 | /** |
176 | * rs_cmp - multi-block reservation range compare | |
177 | * @blk: absolute file system block number of the new reservation | |
178 | * @len: number of blocks in the new reservation | |
179 | * @rs: existing reservation to compare against | |
180 | * | |
181 | * returns: 1 if the block range is beyond the reach of the reservation | |
182 | * -1 if the block range is before the start of the reservation | |
183 | * 0 if the block range overlaps with the reservation | |
184 | */ | |
185 | static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs) | |
186 | { | |
4a993fb1 | 187 | u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 BP |
188 | |
189 | if (blk >= startblk + rs->rs_free) | |
190 | return 1; | |
191 | if (blk + len - 1 < startblk) | |
192 | return -1; | |
193 | return 0; | |
194 | } | |
195 | ||
88c8ab1f SW |
196 | /** |
197 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
198 | * a block in a given allocation state. | |
886b1416 | 199 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 200 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 201 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 202 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
203 | * |
204 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
205 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
206 | * beginning of a bitmap block buffer, skipping any header structures, but |
207 | * headers are always a multiple of 64 bits long so that the buffer is | |
208 | * always aligned to a 64 bit boundary. | |
209 | * | |
210 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 211 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 212 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
213 | * |
214 | * Return: the block number (bitmap buffer scope) that was found | |
215 | */ | |
216 | ||
02ab1721 HE |
217 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
218 | u32 goal, u8 state) | |
88c8ab1f | 219 | { |
223b2b88 SW |
220 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
221 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
222 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
223 | u64 tmp; | |
075ac448 | 224 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
225 | u32 bit; |
226 | ||
223b2b88 SW |
227 | /* Mask off bits we don't care about at the start of the search */ |
228 | mask <<= spoint; | |
229 | tmp = gfs2_bit_search(ptr, mask, state); | |
230 | ptr++; | |
231 | while(tmp == 0 && ptr < end) { | |
075ac448 | 232 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 233 | ptr++; |
1f466a47 | 234 | } |
223b2b88 SW |
235 | /* Mask off any bits which are more than len bytes from the start */ |
236 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
237 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
238 | /* Didn't find anything, so return */ | |
239 | if (tmp == 0) | |
240 | return BFITNOENT; | |
241 | ptr--; | |
d8bd504a | 242 | bit = __ffs64(tmp); |
223b2b88 SW |
243 | bit /= 2; /* two bits per entry in the bitmap */ |
244 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
245 | } |
246 | ||
ff7f4cb4 SW |
247 | /** |
248 | * gfs2_rbm_from_block - Set the rbm based upon rgd and block number | |
249 | * @rbm: The rbm with rgd already set correctly | |
250 | * @block: The block number (filesystem relative) | |
251 | * | |
252 | * This sets the bi and offset members of an rbm based on a | |
253 | * resource group and a filesystem relative block number. The | |
254 | * resource group must be set in the rbm on entry, the bi and | |
255 | * offset members will be set by this function. | |
256 | * | |
257 | * Returns: 0 on success, or an error code | |
258 | */ | |
259 | ||
260 | static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block) | |
261 | { | |
262 | u64 rblock = block - rbm->rgd->rd_data0; | |
ff7f4cb4 SW |
263 | |
264 | if (WARN_ON_ONCE(rblock > UINT_MAX)) | |
265 | return -EINVAL; | |
266 | if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data) | |
267 | return -E2BIG; | |
268 | ||
e579ed4f | 269 | rbm->bii = 0; |
a68a0a35 BP |
270 | rbm->offset = (u32)(rblock); |
271 | /* Check if the block is within the first block */ | |
e579ed4f | 272 | if (rbm->offset < rbm_bi(rbm)->bi_blocks) |
a68a0a35 BP |
273 | return 0; |
274 | ||
275 | /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */ | |
276 | rbm->offset += (sizeof(struct gfs2_rgrp) - | |
277 | sizeof(struct gfs2_meta_header)) * GFS2_NBBY; | |
e579ed4f BP |
278 | rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap; |
279 | rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap; | |
ff7f4cb4 SW |
280 | return 0; |
281 | } | |
282 | ||
149ed7f5 BP |
283 | /** |
284 | * gfs2_rbm_incr - increment an rbm structure | |
285 | * @rbm: The rbm with rgd already set correctly | |
286 | * | |
287 | * This function takes an existing rbm structure and increments it to the next | |
288 | * viable block offset. | |
289 | * | |
290 | * Returns: If incrementing the offset would cause the rbm to go past the | |
291 | * end of the rgrp, true is returned, otherwise false. | |
292 | * | |
293 | */ | |
294 | ||
295 | static bool gfs2_rbm_incr(struct gfs2_rbm *rbm) | |
296 | { | |
297 | if (rbm->offset + 1 < rbm_bi(rbm)->bi_blocks) { /* in the same bitmap */ | |
298 | rbm->offset++; | |
299 | return false; | |
300 | } | |
301 | if (rbm->bii == rbm->rgd->rd_length - 1) /* at the last bitmap */ | |
302 | return true; | |
303 | ||
304 | rbm->offset = 0; | |
305 | rbm->bii++; | |
306 | return false; | |
307 | } | |
308 | ||
ff7f4cb4 SW |
309 | /** |
310 | * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned | |
311 | * @rbm: Position to search (value/result) | |
312 | * @n_unaligned: Number of unaligned blocks to check | |
313 | * @len: Decremented for each block found (terminate on zero) | |
314 | * | |
315 | * Returns: true if a non-free block is encountered | |
316 | */ | |
317 | ||
318 | static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len) | |
319 | { | |
ff7f4cb4 SW |
320 | u32 n; |
321 | u8 res; | |
322 | ||
323 | for (n = 0; n < n_unaligned; n++) { | |
324 | res = gfs2_testbit(rbm); | |
325 | if (res != GFS2_BLKST_FREE) | |
326 | return true; | |
327 | (*len)--; | |
328 | if (*len == 0) | |
329 | return true; | |
149ed7f5 | 330 | if (gfs2_rbm_incr(rbm)) |
ff7f4cb4 SW |
331 | return true; |
332 | } | |
333 | ||
334 | return false; | |
335 | } | |
336 | ||
337 | /** | |
338 | * gfs2_free_extlen - Return extent length of free blocks | |
27ff6a0f | 339 | * @rrbm: Starting position |
ff7f4cb4 SW |
340 | * @len: Max length to check |
341 | * | |
342 | * Starting at the block specified by the rbm, see how many free blocks | |
343 | * there are, not reading more than len blocks ahead. This can be done | |
344 | * using memchr_inv when the blocks are byte aligned, but has to be done | |
345 | * on a block by block basis in case of unaligned blocks. Also this | |
346 | * function can cope with bitmap boundaries (although it must stop on | |
347 | * a resource group boundary) | |
348 | * | |
349 | * Returns: Number of free blocks in the extent | |
350 | */ | |
351 | ||
352 | static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len) | |
353 | { | |
354 | struct gfs2_rbm rbm = *rrbm; | |
355 | u32 n_unaligned = rbm.offset & 3; | |
356 | u32 size = len; | |
357 | u32 bytes; | |
358 | u32 chunk_size; | |
359 | u8 *ptr, *start, *end; | |
360 | u64 block; | |
e579ed4f | 361 | struct gfs2_bitmap *bi; |
ff7f4cb4 SW |
362 | |
363 | if (n_unaligned && | |
364 | gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len)) | |
365 | goto out; | |
366 | ||
3701530a | 367 | n_unaligned = len & 3; |
ff7f4cb4 SW |
368 | /* Start is now byte aligned */ |
369 | while (len > 3) { | |
e579ed4f BP |
370 | bi = rbm_bi(&rbm); |
371 | start = bi->bi_bh->b_data; | |
372 | if (bi->bi_clone) | |
373 | start = bi->bi_clone; | |
374 | end = start + bi->bi_bh->b_size; | |
375 | start += bi->bi_offset; | |
ff7f4cb4 SW |
376 | BUG_ON(rbm.offset & 3); |
377 | start += (rbm.offset / GFS2_NBBY); | |
378 | bytes = min_t(u32, len / GFS2_NBBY, (end - start)); | |
379 | ptr = memchr_inv(start, 0, bytes); | |
380 | chunk_size = ((ptr == NULL) ? bytes : (ptr - start)); | |
381 | chunk_size *= GFS2_NBBY; | |
382 | BUG_ON(len < chunk_size); | |
383 | len -= chunk_size; | |
384 | block = gfs2_rbm_to_block(&rbm); | |
15bd50ad BP |
385 | if (gfs2_rbm_from_block(&rbm, block + chunk_size)) { |
386 | n_unaligned = 0; | |
ff7f4cb4 | 387 | break; |
15bd50ad BP |
388 | } |
389 | if (ptr) { | |
390 | n_unaligned = 3; | |
391 | break; | |
392 | } | |
ff7f4cb4 SW |
393 | n_unaligned = len & 3; |
394 | } | |
395 | ||
396 | /* Deal with any bits left over at the end */ | |
397 | if (n_unaligned) | |
398 | gfs2_unaligned_extlen(&rbm, n_unaligned, &len); | |
399 | out: | |
400 | return size - len; | |
401 | } | |
402 | ||
88c8ab1f SW |
403 | /** |
404 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 405 | * @rgd: the resource group descriptor |
88c8ab1f SW |
406 | * @buffer: the buffer that holds the bitmaps |
407 | * @buflen: the length (in bytes) of the buffer | |
408 | * @state: the state of the block we're looking for | |
409 | * | |
410 | * Returns: The number of bits | |
411 | */ | |
412 | ||
110acf38 SW |
413 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
414 | unsigned int buflen, u8 state) | |
88c8ab1f | 415 | { |
110acf38 SW |
416 | const u8 *byte = buffer; |
417 | const u8 *end = buffer + buflen; | |
418 | const u8 state1 = state << 2; | |
419 | const u8 state2 = state << 4; | |
420 | const u8 state3 = state << 6; | |
cd915493 | 421 | u32 count = 0; |
88c8ab1f SW |
422 | |
423 | for (; byte < end; byte++) { | |
424 | if (((*byte) & 0x03) == state) | |
425 | count++; | |
426 | if (((*byte) & 0x0C) == state1) | |
427 | count++; | |
428 | if (((*byte) & 0x30) == state2) | |
429 | count++; | |
430 | if (((*byte) & 0xC0) == state3) | |
431 | count++; | |
432 | } | |
433 | ||
434 | return count; | |
435 | } | |
436 | ||
b3b94faa DT |
437 | /** |
438 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
439 | * @rgd: the rgrp |
440 | * | |
441 | */ | |
442 | ||
443 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
444 | { | |
445 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
446 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 447 | u32 length = rgd->rd_length; |
cd915493 | 448 | u32 count[4], tmp; |
b3b94faa DT |
449 | int buf, x; |
450 | ||
cd915493 | 451 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
452 | |
453 | /* Count # blocks in each of 4 possible allocation states */ | |
454 | for (buf = 0; buf < length; buf++) { | |
455 | bi = rgd->rd_bits + buf; | |
456 | for (x = 0; x < 4; x++) | |
457 | count[x] += gfs2_bitcount(rgd, | |
458 | bi->bi_bh->b_data + | |
459 | bi->bi_offset, | |
460 | bi->bi_len, x); | |
461 | } | |
462 | ||
cfc8b549 | 463 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
464 | if (gfs2_consist_rgrpd(rgd)) |
465 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 466 | count[0], rgd->rd_free); |
b3b94faa DT |
467 | return; |
468 | } | |
469 | ||
73f74948 | 470 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 471 | if (count[1] != tmp) { |
b3b94faa DT |
472 | if (gfs2_consist_rgrpd(rgd)) |
473 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
474 | count[1], tmp); | |
475 | return; | |
476 | } | |
477 | ||
6b946170 | 478 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 479 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 480 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 481 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
482 | return; |
483 | } | |
b3b94faa DT |
484 | } |
485 | ||
bb8d8a6f | 486 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 487 | { |
bb8d8a6f SW |
488 | u64 first = rgd->rd_data0; |
489 | u64 last = first + rgd->rd_data; | |
16910427 | 490 | return first <= block && block < last; |
b3b94faa DT |
491 | } |
492 | ||
493 | /** | |
494 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
495 | * @sdp: The GFS2 superblock | |
886b1416 BP |
496 | * @blk: The data block number |
497 | * @exact: True if this needs to be an exact match | |
b3b94faa DT |
498 | * |
499 | * Returns: The resource group, or NULL if not found | |
500 | */ | |
501 | ||
66fc061b | 502 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 503 | { |
66fc061b | 504 | struct rb_node *n, *next; |
f75bbfb4 | 505 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
506 | |
507 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
508 | n = sdp->sd_rindex_tree.rb_node; |
509 | while (n) { | |
510 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
511 | next = NULL; | |
7c9ca621 | 512 | if (blk < cur->rd_addr) |
66fc061b | 513 | next = n->rb_left; |
f75bbfb4 | 514 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
515 | next = n->rb_right; |
516 | if (next == NULL) { | |
b3b94faa | 517 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
518 | if (exact) { |
519 | if (blk < cur->rd_addr) | |
520 | return NULL; | |
521 | if (blk >= cur->rd_data0 + cur->rd_data) | |
522 | return NULL; | |
523 | } | |
7c9ca621 | 524 | return cur; |
b3b94faa | 525 | } |
66fc061b | 526 | n = next; |
b3b94faa | 527 | } |
b3b94faa DT |
528 | spin_unlock(&sdp->sd_rindex_spin); |
529 | ||
530 | return NULL; | |
531 | } | |
532 | ||
533 | /** | |
534 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
535 | * @sdp: The GFS2 superblock | |
536 | * | |
537 | * Returns: The first rgrp in the filesystem | |
538 | */ | |
539 | ||
540 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
541 | { | |
7c9ca621 BP |
542 | const struct rb_node *n; |
543 | struct gfs2_rgrpd *rgd; | |
544 | ||
8339ee54 | 545 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
546 | n = rb_first(&sdp->sd_rindex_tree); |
547 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 548 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
549 | |
550 | return rgd; | |
b3b94faa DT |
551 | } |
552 | ||
553 | /** | |
554 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 555 | * @rgd: the resource group descriptor |
b3b94faa DT |
556 | * |
557 | * Returns: The next rgrp | |
558 | */ | |
559 | ||
560 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
561 | { | |
7c9ca621 BP |
562 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
563 | const struct rb_node *n; | |
564 | ||
565 | spin_lock(&sdp->sd_rindex_spin); | |
566 | n = rb_next(&rgd->rd_node); | |
567 | if (n == NULL) | |
568 | n = rb_first(&sdp->sd_rindex_tree); | |
569 | ||
570 | if (unlikely(&rgd->rd_node == n)) { | |
571 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 572 | return NULL; |
7c9ca621 BP |
573 | } |
574 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
575 | spin_unlock(&sdp->sd_rindex_spin); | |
576 | return rgd; | |
b3b94faa DT |
577 | } |
578 | ||
00a158be AD |
579 | void check_and_update_goal(struct gfs2_inode *ip) |
580 | { | |
581 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
582 | if (!ip->i_goal || gfs2_blk2rgrpd(sdp, ip->i_goal, 1) == NULL) | |
583 | ip->i_goal = ip->i_no_addr; | |
584 | } | |
585 | ||
8339ee54 SW |
586 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
587 | { | |
588 | int x; | |
589 | ||
590 | for (x = 0; x < rgd->rd_length; x++) { | |
591 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
592 | kfree(bi->bi_clone); | |
593 | bi->bi_clone = NULL; | |
594 | } | |
595 | } | |
596 | ||
0a305e49 | 597 | /** |
b54e9a0b BP |
598 | * gfs2_rsqa_alloc - make sure we have a reservation assigned to the inode |
599 | * plus a quota allocations data structure, if necessary | |
0a305e49 BP |
600 | * @ip: the inode for this reservation |
601 | */ | |
b54e9a0b | 602 | int gfs2_rsqa_alloc(struct gfs2_inode *ip) |
0a305e49 | 603 | { |
a097dc7e | 604 | return gfs2_qa_alloc(ip); |
0a305e49 BP |
605 | } |
606 | ||
9e733d39 | 607 | static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs) |
8e2e0047 | 608 | { |
9e733d39 SW |
609 | gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n", |
610 | (unsigned long long)rs->rs_inum, | |
611 | (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm), | |
4a993fb1 | 612 | rs->rs_rbm.offset, rs->rs_free); |
8e2e0047 BP |
613 | } |
614 | ||
0a305e49 | 615 | /** |
8e2e0047 BP |
616 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
617 | * @rs: The reservation to remove | |
618 | * | |
619 | */ | |
20095218 | 620 | static void __rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
621 | { |
622 | struct gfs2_rgrpd *rgd; | |
623 | ||
624 | if (!gfs2_rs_active(rs)) | |
625 | return; | |
626 | ||
4a993fb1 | 627 | rgd = rs->rs_rbm.rgd; |
9e733d39 | 628 | trace_gfs2_rs(rs, TRACE_RS_TREEDEL); |
4a993fb1 | 629 | rb_erase(&rs->rs_node, &rgd->rd_rstree); |
24d634e8 | 630 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 BP |
631 | |
632 | if (rs->rs_free) { | |
e579ed4f BP |
633 | struct gfs2_bitmap *bi = rbm_bi(&rs->rs_rbm); |
634 | ||
20095218 | 635 | /* return reserved blocks to the rgrp */ |
4a993fb1 SW |
636 | BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free); |
637 | rs->rs_rbm.rgd->rd_reserved -= rs->rs_free; | |
5ea5050c BP |
638 | /* The rgrp extent failure point is likely not to increase; |
639 | it will only do so if the freed blocks are somehow | |
640 | contiguous with a span of free blocks that follows. Still, | |
641 | it will force the number to be recalculated later. */ | |
642 | rgd->rd_extfail_pt += rs->rs_free; | |
8e2e0047 | 643 | rs->rs_free = 0; |
e579ed4f | 644 | clear_bit(GBF_FULL, &bi->bi_flags); |
8e2e0047 | 645 | } |
8e2e0047 BP |
646 | } |
647 | ||
648 | /** | |
649 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
650 | * @rs: The reservation to remove | |
651 | * | |
652 | */ | |
20095218 | 653 | void gfs2_rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
654 | { |
655 | struct gfs2_rgrpd *rgd; | |
656 | ||
4a993fb1 SW |
657 | rgd = rs->rs_rbm.rgd; |
658 | if (rgd) { | |
659 | spin_lock(&rgd->rd_rsspin); | |
20095218 | 660 | __rs_deltree(rs); |
44f52122 | 661 | BUG_ON(rs->rs_free); |
4a993fb1 SW |
662 | spin_unlock(&rgd->rd_rsspin); |
663 | } | |
8e2e0047 BP |
664 | } |
665 | ||
666 | /** | |
b54e9a0b | 667 | * gfs2_rsqa_delete - delete a multi-block reservation and quota allocation |
0a305e49 | 668 | * @ip: The inode for this reservation |
af5c2697 | 669 | * @wcount: The inode's write count, or NULL |
0a305e49 BP |
670 | * |
671 | */ | |
b54e9a0b | 672 | void gfs2_rsqa_delete(struct gfs2_inode *ip, atomic_t *wcount) |
0a305e49 BP |
673 | { |
674 | down_write(&ip->i_rw_mutex); | |
44f52122 | 675 | if ((wcount == NULL) || (atomic_read(wcount) <= 1)) |
a097dc7e | 676 | gfs2_rs_deltree(&ip->i_res); |
0a305e49 | 677 | up_write(&ip->i_rw_mutex); |
a097dc7e | 678 | gfs2_qa_delete(ip, wcount); |
0a305e49 BP |
679 | } |
680 | ||
8e2e0047 BP |
681 | /** |
682 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
683 | * @rgd: the rgrp that needs its space back | |
684 | * | |
685 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
686 | * give them back. This leave the reservation structures in tact, but removes | |
687 | * all of their corresponding "no-fly zones". | |
688 | */ | |
689 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
690 | { | |
691 | struct rb_node *n; | |
692 | struct gfs2_blkreserv *rs; | |
693 | ||
694 | spin_lock(&rgd->rd_rsspin); | |
695 | while ((n = rb_first(&rgd->rd_rstree))) { | |
696 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
20095218 | 697 | __rs_deltree(rs); |
8e2e0047 BP |
698 | } |
699 | spin_unlock(&rgd->rd_rsspin); | |
700 | } | |
701 | ||
8339ee54 | 702 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 703 | { |
7c9ca621 | 704 | struct rb_node *n; |
b3b94faa DT |
705 | struct gfs2_rgrpd *rgd; |
706 | struct gfs2_glock *gl; | |
707 | ||
7c9ca621 BP |
708 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
709 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
710 | gl = rgd->rd_gl; |
711 | ||
7c9ca621 | 712 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
713 | |
714 | if (gl) { | |
f3dd1649 | 715 | spin_lock(&gl->gl_lockref.lock); |
5c676f6d | 716 | gl->gl_object = NULL; |
f3dd1649 | 717 | spin_unlock(&gl->gl_lockref.lock); |
29687a2a | 718 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
719 | gfs2_glock_put(gl); |
720 | } | |
721 | ||
8339ee54 | 722 | gfs2_free_clones(rgd); |
b3b94faa | 723 | kfree(rgd->rd_bits); |
36e4ad03 | 724 | rgd->rd_bits = NULL; |
8e2e0047 | 725 | return_all_reservations(rgd); |
6bdd9be6 | 726 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
727 | } |
728 | } | |
729 | ||
bb8d8a6f SW |
730 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
731 | { | |
d77d1b58 JP |
732 | pr_info("ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); |
733 | pr_info("ri_length = %u\n", rgd->rd_length); | |
734 | pr_info("ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
735 | pr_info("ri_data = %u\n", rgd->rd_data); | |
736 | pr_info("ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
bb8d8a6f SW |
737 | } |
738 | ||
b3b94faa DT |
739 | /** |
740 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
741 | * @rgd: The resource group descriptor | |
742 | * | |
743 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
744 | * | |
745 | * Returns: errno | |
746 | */ | |
747 | ||
748 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
749 | { | |
750 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
751 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 752 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 753 | u32 bytes_left, bytes; |
b3b94faa DT |
754 | int x; |
755 | ||
feaa7bba SW |
756 | if (!length) |
757 | return -EINVAL; | |
758 | ||
dd894be8 | 759 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
760 | if (!rgd->rd_bits) |
761 | return -ENOMEM; | |
762 | ||
bb8d8a6f | 763 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
764 | |
765 | for (x = 0; x < length; x++) { | |
766 | bi = rgd->rd_bits + x; | |
767 | ||
60a0b8f9 | 768 | bi->bi_flags = 0; |
b3b94faa DT |
769 | /* small rgrp; bitmap stored completely in header block */ |
770 | if (length == 1) { | |
771 | bytes = bytes_left; | |
772 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
773 | bi->bi_start = 0; | |
774 | bi->bi_len = bytes; | |
7e230f57 | 775 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
776 | /* header block */ |
777 | } else if (x == 0) { | |
778 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
779 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
780 | bi->bi_start = 0; | |
781 | bi->bi_len = bytes; | |
7e230f57 | 782 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
783 | /* last block */ |
784 | } else if (x + 1 == length) { | |
785 | bytes = bytes_left; | |
786 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 787 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 788 | bi->bi_len = bytes; |
7e230f57 | 789 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
790 | /* other blocks */ |
791 | } else { | |
568f4c96 SW |
792 | bytes = sdp->sd_sb.sb_bsize - |
793 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 794 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 795 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 796 | bi->bi_len = bytes; |
7e230f57 | 797 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
798 | } |
799 | ||
800 | bytes_left -= bytes; | |
801 | } | |
802 | ||
803 | if (bytes_left) { | |
804 | gfs2_consist_rgrpd(rgd); | |
805 | return -EIO; | |
806 | } | |
807 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 808 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 809 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 810 | gfs2_rindex_print(rgd); |
b3b94faa DT |
811 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
812 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
813 | } | |
814 | return -EIO; | |
815 | } | |
816 | ||
817 | return 0; | |
818 | } | |
819 | ||
7ae8fa84 RP |
820 | /** |
821 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 822 | * @sdp: the filesystem |
7ae8fa84 RP |
823 | * |
824 | */ | |
825 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
826 | { | |
827 | u64 total_data = 0; | |
828 | struct inode *inode = sdp->sd_rindex; | |
829 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 830 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
831 | int error, rgrps; |
832 | ||
7ae8fa84 RP |
833 | for (rgrps = 0;; rgrps++) { |
834 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
835 | ||
bcd7278d | 836 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 837 | break; |
4306629e | 838 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
839 | sizeof(struct gfs2_rindex)); |
840 | if (error != sizeof(struct gfs2_rindex)) | |
841 | break; | |
bb8d8a6f | 842 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 843 | } |
7ae8fa84 RP |
844 | return total_data; |
845 | } | |
846 | ||
6aad1c3d | 847 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
848 | { |
849 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
850 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
851 | ||
852 | /* Figure out where to put new node */ | |
853 | while (*newn) { | |
854 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
855 | rd_node); | |
856 | ||
857 | parent = *newn; | |
858 | if (rgd->rd_addr < cur->rd_addr) | |
859 | newn = &((*newn)->rb_left); | |
860 | else if (rgd->rd_addr > cur->rd_addr) | |
861 | newn = &((*newn)->rb_right); | |
862 | else | |
6aad1c3d | 863 | return -EEXIST; |
7c9ca621 BP |
864 | } |
865 | ||
866 | rb_link_node(&rgd->rd_node, parent, newn); | |
867 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
868 | sdp->sd_rgrps++; |
869 | return 0; | |
7c9ca621 BP |
870 | } |
871 | ||
b3b94faa | 872 | /** |
6c53267f | 873 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 874 | * @ip: Pointer to the rindex inode |
b3b94faa | 875 | * |
8339ee54 | 876 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
877 | */ |
878 | ||
4306629e | 879 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
880 | { |
881 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
7005c3e4 | 882 | const unsigned bsize = sdp->sd_sb.sb_bsize; |
6c53267f | 883 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); |
8339ee54 | 884 | struct gfs2_rindex buf; |
6c53267f RP |
885 | int error; |
886 | struct gfs2_rgrpd *rgd; | |
887 | ||
8339ee54 SW |
888 | if (pos >= i_size_read(&ip->i_inode)) |
889 | return 1; | |
890 | ||
4306629e | 891 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 892 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
893 | |
894 | if (error != sizeof(struct gfs2_rindex)) | |
895 | return (error == 0) ? 1 : error; | |
6c53267f | 896 | |
6bdd9be6 | 897 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
898 | error = -ENOMEM; |
899 | if (!rgd) | |
900 | return error; | |
901 | ||
6c53267f | 902 | rgd->rd_sbd = sdp; |
8339ee54 SW |
903 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
904 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
905 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
906 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
907 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 908 | spin_lock_init(&rgd->rd_rsspin); |
7c9ca621 | 909 | |
6c53267f RP |
910 | error = compute_bitstructs(rgd); |
911 | if (error) | |
8339ee54 | 912 | goto fail; |
6c53267f | 913 | |
bb8d8a6f | 914 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
915 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
916 | if (error) | |
8339ee54 | 917 | goto fail; |
6c53267f | 918 | |
4e2f8849 | 919 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr; |
0e27c18c | 920 | rgd->rd_flags &= ~(GFS2_RDF_UPTODATE | GFS2_RDF_PREFERRED); |
7c9ca621 BP |
921 | if (rgd->rd_data > sdp->sd_max_rg_data) |
922 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 923 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 924 | error = rgd_insert(rgd); |
8339ee54 | 925 | spin_unlock(&sdp->sd_rindex_spin); |
36e4ad03 BP |
926 | if (!error) { |
927 | rgd->rd_gl->gl_object = rgd; | |
928 | rgd->rd_gl->gl_vm.start = (rgd->rd_addr * bsize) & PAGE_MASK; | |
929 | rgd->rd_gl->gl_vm.end = PAGE_ALIGN((rgd->rd_addr + | |
930 | rgd->rd_length) * bsize) - 1; | |
6aad1c3d | 931 | return 0; |
36e4ad03 | 932 | } |
6aad1c3d BP |
933 | |
934 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 935 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
936 | |
937 | fail: | |
938 | kfree(rgd->rd_bits); | |
36e4ad03 | 939 | rgd->rd_bits = NULL; |
8339ee54 | 940 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
6c53267f RP |
941 | return error; |
942 | } | |
943 | ||
0e27c18c BP |
944 | /** |
945 | * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use | |
946 | * @sdp: the GFS2 superblock | |
947 | * | |
948 | * The purpose of this function is to select a subset of the resource groups | |
949 | * and mark them as PREFERRED. We do it in such a way that each node prefers | |
950 | * to use a unique set of rgrps to minimize glock contention. | |
951 | */ | |
952 | static void set_rgrp_preferences(struct gfs2_sbd *sdp) | |
953 | { | |
954 | struct gfs2_rgrpd *rgd, *first; | |
955 | int i; | |
956 | ||
957 | /* Skip an initial number of rgrps, based on this node's journal ID. | |
958 | That should start each node out on its own set. */ | |
959 | rgd = gfs2_rgrpd_get_first(sdp); | |
960 | for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++) | |
961 | rgd = gfs2_rgrpd_get_next(rgd); | |
962 | first = rgd; | |
963 | ||
964 | do { | |
965 | rgd->rd_flags |= GFS2_RDF_PREFERRED; | |
966 | for (i = 0; i < sdp->sd_journals; i++) { | |
967 | rgd = gfs2_rgrpd_get_next(rgd); | |
959b6717 | 968 | if (!rgd || rgd == first) |
0e27c18c BP |
969 | break; |
970 | } | |
959b6717 | 971 | } while (rgd && rgd != first); |
0e27c18c BP |
972 | } |
973 | ||
6c53267f RP |
974 | /** |
975 | * gfs2_ri_update - Pull in a new resource index from the disk | |
976 | * @ip: pointer to the rindex inode | |
977 | * | |
b3b94faa DT |
978 | * Returns: 0 on successful update, error code otherwise |
979 | */ | |
980 | ||
8339ee54 | 981 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 982 | { |
feaa7bba | 983 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
984 | int error; |
985 | ||
8339ee54 | 986 | do { |
4306629e | 987 | error = read_rindex_entry(ip); |
8339ee54 SW |
988 | } while (error == 0); |
989 | ||
990 | if (error < 0) | |
991 | return error; | |
b3b94faa | 992 | |
0e27c18c BP |
993 | set_rgrp_preferences(sdp); |
994 | ||
cf45b752 | 995 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
996 | return 0; |
997 | } | |
b3b94faa | 998 | |
b3b94faa | 999 | /** |
8339ee54 | 1000 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 1001 | * @sdp: The GFS2 superblock |
b3b94faa DT |
1002 | * |
1003 | * We grab a lock on the rindex inode to make sure that it doesn't | |
1004 | * change whilst we are performing an operation. We keep this lock | |
1005 | * for quite long periods of time compared to other locks. This | |
1006 | * doesn't matter, since it is shared and it is very, very rarely | |
1007 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
1008 | * | |
1009 | * This makes sure that we're using the latest copy of the resource index | |
1010 | * special file, which might have been updated if someone expanded the | |
1011 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
1012 | * | |
8339ee54 | 1013 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
1014 | */ |
1015 | ||
8339ee54 | 1016 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 1017 | { |
feaa7bba | 1018 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 1019 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
1020 | struct gfs2_holder ri_gh; |
1021 | int error = 0; | |
a365fbf3 | 1022 | int unlock_required = 0; |
b3b94faa DT |
1023 | |
1024 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 1025 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
1026 | if (!gfs2_glock_is_locked_by_me(gl)) { |
1027 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
1028 | if (error) | |
6aad1c3d | 1029 | return error; |
a365fbf3 SW |
1030 | unlock_required = 1; |
1031 | } | |
8339ee54 | 1032 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 1033 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
1034 | if (unlock_required) |
1035 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
1036 | } |
1037 | ||
1038 | return error; | |
1039 | } | |
1040 | ||
42d52e38 | 1041 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
1042 | { |
1043 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 1044 | u32 rg_flags; |
bb8d8a6f | 1045 | |
42d52e38 | 1046 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 1047 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
1048 | rgd->rd_flags &= GFS2_RDF_MASK; |
1049 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 1050 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 1051 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 1052 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
1053 | } |
1054 | ||
42d52e38 | 1055 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
1056 | { |
1057 | struct gfs2_rgrp *str = buf; | |
1058 | ||
09010978 | 1059 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 1060 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 1061 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 1062 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 1063 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
1064 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
1065 | } | |
1066 | ||
90306c41 BM |
1067 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
1068 | { | |
1069 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1070 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
1071 | ||
1072 | if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free || | |
1073 | rgl->rl_dinodes != str->rg_dinodes || | |
1074 | rgl->rl_igeneration != str->rg_igeneration) | |
1075 | return 0; | |
1076 | return 1; | |
1077 | } | |
1078 | ||
1079 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) | |
1080 | { | |
1081 | const struct gfs2_rgrp *str = buf; | |
1082 | ||
1083 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
1084 | rgl->rl_flags = str->rg_flags; | |
1085 | rgl->rl_free = str->rg_free; | |
1086 | rgl->rl_dinodes = str->rg_dinodes; | |
1087 | rgl->rl_igeneration = str->rg_igeneration; | |
1088 | rgl->__pad = 0UL; | |
1089 | } | |
1090 | ||
1091 | static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change) | |
1092 | { | |
1093 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1094 | u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change; | |
1095 | rgl->rl_unlinked = cpu_to_be32(unlinked); | |
1096 | } | |
1097 | ||
1098 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) | |
1099 | { | |
1100 | struct gfs2_bitmap *bi; | |
1101 | const u32 length = rgd->rd_length; | |
1102 | const u8 *buffer = NULL; | |
1103 | u32 i, goal, count = 0; | |
1104 | ||
1105 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
1106 | goal = 0; | |
1107 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1108 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
1109 | while (goal < bi->bi_len * GFS2_NBBY) { | |
1110 | goal = gfs2_bitfit(buffer, bi->bi_len, goal, | |
1111 | GFS2_BLKST_UNLINKED); | |
1112 | if (goal == BFITNOENT) | |
1113 | break; | |
1114 | count++; | |
1115 | goal++; | |
1116 | } | |
1117 | } | |
1118 | ||
1119 | return count; | |
1120 | } | |
1121 | ||
1122 | ||
b3b94faa | 1123 | /** |
90306c41 BM |
1124 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
1125 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
1126 | * |
1127 | * Read in all of a Resource Group's header and bitmap blocks. | |
1128 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
1129 | * | |
1130 | * Returns: errno | |
1131 | */ | |
1132 | ||
c2b0b30e | 1133 | static int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
1134 | { |
1135 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1136 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 1137 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
1138 | struct gfs2_bitmap *bi; |
1139 | unsigned int x, y; | |
1140 | int error; | |
1141 | ||
90306c41 BM |
1142 | if (rgd->rd_bits[0].bi_bh != NULL) |
1143 | return 0; | |
1144 | ||
b3b94faa DT |
1145 | for (x = 0; x < length; x++) { |
1146 | bi = rgd->rd_bits + x; | |
c8d57703 | 1147 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, 0, &bi->bi_bh); |
b3b94faa DT |
1148 | if (error) |
1149 | goto fail; | |
1150 | } | |
1151 | ||
1152 | for (y = length; y--;) { | |
1153 | bi = rgd->rd_bits + y; | |
7276b3b0 | 1154 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
1155 | if (error) |
1156 | goto fail; | |
feaa7bba | 1157 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
1158 | GFS2_METATYPE_RG)) { |
1159 | error = -EIO; | |
1160 | goto fail; | |
1161 | } | |
1162 | } | |
1163 | ||
cf45b752 | 1164 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
1165 | for (x = 0; x < length; x++) |
1166 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 1167 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 1168 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 1169 | rgd->rd_free_clone = rgd->rd_free; |
5ea5050c BP |
1170 | /* max out the rgrp allocation failure point */ |
1171 | rgd->rd_extfail_pt = rgd->rd_free; | |
b3b94faa | 1172 | } |
951b4bd5 | 1173 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
90306c41 BM |
1174 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); |
1175 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
1176 | rgd->rd_bits[0].bi_bh->b_data); | |
1177 | } | |
1178 | else if (sdp->sd_args.ar_rgrplvb) { | |
1179 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
1180 | gfs2_consist_rgrpd(rgd); | |
1181 | error = -EIO; | |
1182 | goto fail; | |
1183 | } | |
1184 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1185 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1186 | } | |
b3b94faa DT |
1187 | return 0; |
1188 | ||
feaa7bba | 1189 | fail: |
b3b94faa DT |
1190 | while (x--) { |
1191 | bi = rgd->rd_bits + x; | |
1192 | brelse(bi->bi_bh); | |
1193 | bi->bi_bh = NULL; | |
1194 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
1195 | } | |
b3b94faa DT |
1196 | |
1197 | return error; | |
1198 | } | |
1199 | ||
c2b0b30e | 1200 | static int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
90306c41 BM |
1201 | { |
1202 | u32 rl_flags; | |
1203 | ||
1204 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1205 | return 0; | |
1206 | ||
951b4bd5 | 1207 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) |
90306c41 BM |
1208 | return gfs2_rgrp_bh_get(rgd); |
1209 | ||
1210 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1211 | rl_flags &= ~GFS2_RDF_MASK; | |
1212 | rgd->rd_flags &= GFS2_RDF_MASK; | |
1213 | rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); | |
1214 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1215 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1216 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
1217 | rgd->rd_free_clone = rgd->rd_free; | |
1218 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); | |
1219 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1220 | return 0; | |
1221 | } | |
1222 | ||
1223 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1224 | { | |
1225 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1226 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1227 | ||
1228 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1229 | return 0; | |
8b127d04 | 1230 | return gfs2_rgrp_bh_get(rgd); |
90306c41 BM |
1231 | } |
1232 | ||
b3b94faa | 1233 | /** |
39b0f1e9 BP |
1234 | * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
1235 | * @rgd: The resource group | |
b3b94faa DT |
1236 | * |
1237 | */ | |
1238 | ||
39b0f1e9 | 1239 | void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd) |
b3b94faa | 1240 | { |
bb8d8a6f | 1241 | int x, length = rgd->rd_length; |
b3b94faa | 1242 | |
b3b94faa DT |
1243 | for (x = 0; x < length; x++) { |
1244 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1245 | if (bi->bi_bh) { |
1246 | brelse(bi->bi_bh); | |
1247 | bi->bi_bh = NULL; | |
1248 | } | |
b3b94faa DT |
1249 | } |
1250 | ||
b3b94faa DT |
1251 | } |
1252 | ||
39b0f1e9 BP |
1253 | /** |
1254 | * gfs2_rgrp_go_unlock - Unlock a rgrp glock | |
1255 | * @gh: The glock holder for the resource group | |
1256 | * | |
1257 | */ | |
1258 | ||
1259 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) | |
1260 | { | |
1261 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1262 | int demote_requested = test_bit(GLF_DEMOTE, &gh->gh_gl->gl_flags) | | |
1263 | test_bit(GLF_PENDING_DEMOTE, &gh->gh_gl->gl_flags); | |
1264 | ||
1265 | if (rgd && demote_requested) | |
1266 | gfs2_rgrp_brelse(rgd); | |
1267 | } | |
1268 | ||
66fc061b | 1269 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1270 | struct buffer_head *bh, |
66fc061b | 1271 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1272 | { |
1273 | struct super_block *sb = sdp->sd_vfs; | |
f15ab561 | 1274 | u64 blk; |
64d576ba | 1275 | sector_t start = 0; |
b2c87cae | 1276 | sector_t nr_blks = 0; |
f15ab561 SW |
1277 | int rv; |
1278 | unsigned int x; | |
66fc061b SW |
1279 | u32 trimmed = 0; |
1280 | u8 diff; | |
f15ab561 SW |
1281 | |
1282 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
1283 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1284 | clone += bi->bi_offset; | |
1285 | clone += x; | |
1286 | if (bh) { | |
1287 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1288 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1289 | } else { | |
1290 | diff = ~(*clone | (*clone >> 1)); | |
1291 | } | |
f15ab561 SW |
1292 | diff &= 0x55; |
1293 | if (diff == 0) | |
1294 | continue; | |
1295 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
f15ab561 SW |
1296 | while(diff) { |
1297 | if (diff & 1) { | |
b2c87cae | 1298 | if (nr_blks == 0) |
f15ab561 | 1299 | goto start_new_extent; |
b2c87cae BP |
1300 | if ((start + nr_blks) != blk) { |
1301 | if (nr_blks >= minlen) { | |
1302 | rv = sb_issue_discard(sb, | |
1303 | start, nr_blks, | |
66fc061b SW |
1304 | GFP_NOFS, 0); |
1305 | if (rv) | |
1306 | goto fail; | |
b2c87cae | 1307 | trimmed += nr_blks; |
66fc061b | 1308 | } |
b2c87cae | 1309 | nr_blks = 0; |
f15ab561 SW |
1310 | start_new_extent: |
1311 | start = blk; | |
1312 | } | |
b2c87cae | 1313 | nr_blks++; |
f15ab561 SW |
1314 | } |
1315 | diff >>= 2; | |
b2c87cae | 1316 | blk++; |
f15ab561 SW |
1317 | } |
1318 | } | |
b2c87cae BP |
1319 | if (nr_blks >= minlen) { |
1320 | rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0); | |
f15ab561 SW |
1321 | if (rv) |
1322 | goto fail; | |
b2c87cae | 1323 | trimmed += nr_blks; |
f15ab561 | 1324 | } |
66fc061b SW |
1325 | if (ptrimmed) |
1326 | *ptrimmed = trimmed; | |
1327 | return 0; | |
1328 | ||
f15ab561 | 1329 | fail: |
66fc061b SW |
1330 | if (sdp->sd_args.ar_discard) |
1331 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 1332 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1333 | return -EIO; |
1334 | } | |
1335 | ||
1336 | /** | |
1337 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1338 | * @filp: Any file on the filesystem | |
1339 | * @argp: Pointer to the arguments (also used to pass result) | |
1340 | * | |
1341 | * Returns: 0 on success, otherwise error code | |
1342 | */ | |
1343 | ||
1344 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1345 | { | |
496ad9aa | 1346 | struct inode *inode = file_inode(filp); |
66fc061b SW |
1347 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1348 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1349 | struct buffer_head *bh; | |
1350 | struct gfs2_rgrpd *rgd; | |
1351 | struct gfs2_rgrpd *rgd_end; | |
1352 | struct gfs2_holder gh; | |
1353 | struct fstrim_range r; | |
1354 | int ret = 0; | |
1355 | u64 amt; | |
1356 | u64 trimmed = 0; | |
076f0faa | 1357 | u64 start, end, minlen; |
66fc061b | 1358 | unsigned int x; |
076f0faa | 1359 | unsigned bs_shift = sdp->sd_sb.sb_bsize_shift; |
66fc061b SW |
1360 | |
1361 | if (!capable(CAP_SYS_ADMIN)) | |
1362 | return -EPERM; | |
1363 | ||
1364 | if (!blk_queue_discard(q)) | |
1365 | return -EOPNOTSUPP; | |
1366 | ||
3a238ade | 1367 | if (copy_from_user(&r, argp, sizeof(r))) |
66fc061b SW |
1368 | return -EFAULT; |
1369 | ||
5e2f7d61 BP |
1370 | ret = gfs2_rindex_update(sdp); |
1371 | if (ret) | |
1372 | return ret; | |
1373 | ||
076f0faa LC |
1374 | start = r.start >> bs_shift; |
1375 | end = start + (r.len >> bs_shift); | |
1376 | minlen = max_t(u64, r.minlen, | |
1377 | q->limits.discard_granularity) >> bs_shift; | |
1378 | ||
6a98c333 AD |
1379 | if (end <= start || minlen > sdp->sd_max_rg_data) |
1380 | return -EINVAL; | |
1381 | ||
076f0faa | 1382 | rgd = gfs2_blk2rgrpd(sdp, start, 0); |
6a98c333 | 1383 | rgd_end = gfs2_blk2rgrpd(sdp, end, 0); |
076f0faa | 1384 | |
6a98c333 AD |
1385 | if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end)) |
1386 | && (start > rgd_end->rd_data0 + rgd_end->rd_data)) | |
1387 | return -EINVAL; /* start is beyond the end of the fs */ | |
66fc061b SW |
1388 | |
1389 | while (1) { | |
1390 | ||
1391 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1392 | if (ret) | |
1393 | goto out; | |
1394 | ||
1395 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1396 | /* Trim each bitmap in the rgrp */ | |
1397 | for (x = 0; x < rgd->rd_length; x++) { | |
1398 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
076f0faa LC |
1399 | ret = gfs2_rgrp_send_discards(sdp, |
1400 | rgd->rd_data0, NULL, bi, minlen, | |
1401 | &amt); | |
66fc061b SW |
1402 | if (ret) { |
1403 | gfs2_glock_dq_uninit(&gh); | |
1404 | goto out; | |
1405 | } | |
1406 | trimmed += amt; | |
1407 | } | |
1408 | ||
1409 | /* Mark rgrp as having been trimmed */ | |
1410 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1411 | if (ret == 0) { | |
1412 | bh = rgd->rd_bits[0].bi_bh; | |
1413 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
350a9b0a | 1414 | gfs2_trans_add_meta(rgd->rd_gl, bh); |
66fc061b | 1415 | gfs2_rgrp_out(rgd, bh->b_data); |
90306c41 | 1416 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data); |
66fc061b SW |
1417 | gfs2_trans_end(sdp); |
1418 | } | |
1419 | } | |
1420 | gfs2_glock_dq_uninit(&gh); | |
1421 | ||
1422 | if (rgd == rgd_end) | |
1423 | break; | |
1424 | ||
1425 | rgd = gfs2_rgrpd_get_next(rgd); | |
1426 | } | |
1427 | ||
1428 | out: | |
6a98c333 | 1429 | r.len = trimmed << bs_shift; |
3a238ade | 1430 | if (copy_to_user(argp, &r, sizeof(r))) |
66fc061b SW |
1431 | return -EFAULT; |
1432 | ||
1433 | return ret; | |
f15ab561 SW |
1434 | } |
1435 | ||
8e2e0047 BP |
1436 | /** |
1437 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
8e2e0047 | 1438 | * @ip: the inode structure |
8e2e0047 | 1439 | * |
8e2e0047 | 1440 | */ |
ff7f4cb4 | 1441 | static void rs_insert(struct gfs2_inode *ip) |
8e2e0047 BP |
1442 | { |
1443 | struct rb_node **newn, *parent = NULL; | |
1444 | int rc; | |
a097dc7e | 1445 | struct gfs2_blkreserv *rs = &ip->i_res; |
4a993fb1 | 1446 | struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd; |
ff7f4cb4 | 1447 | u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 | 1448 | |
8e2e0047 | 1449 | BUG_ON(gfs2_rs_active(rs)); |
c743ffd0 | 1450 | |
ff7f4cb4 SW |
1451 | spin_lock(&rgd->rd_rsspin); |
1452 | newn = &rgd->rd_rstree.rb_node; | |
8e2e0047 BP |
1453 | while (*newn) { |
1454 | struct gfs2_blkreserv *cur = | |
1455 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1456 | ||
1457 | parent = *newn; | |
ff7f4cb4 | 1458 | rc = rs_cmp(fsblock, rs->rs_free, cur); |
8e2e0047 BP |
1459 | if (rc > 0) |
1460 | newn = &((*newn)->rb_right); | |
1461 | else if (rc < 0) | |
1462 | newn = &((*newn)->rb_left); | |
1463 | else { | |
1464 | spin_unlock(&rgd->rd_rsspin); | |
ff7f4cb4 SW |
1465 | WARN_ON(1); |
1466 | return; | |
8e2e0047 BP |
1467 | } |
1468 | } | |
1469 | ||
8e2e0047 BP |
1470 | rb_link_node(&rs->rs_node, parent, newn); |
1471 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1472 | ||
8e2e0047 | 1473 | /* Do our rgrp accounting for the reservation */ |
ff7f4cb4 | 1474 | rgd->rd_reserved += rs->rs_free; /* blocks reserved */ |
8e2e0047 | 1475 | spin_unlock(&rgd->rd_rsspin); |
9e733d39 | 1476 | trace_gfs2_rs(rs, TRACE_RS_INSERT); |
8e2e0047 BP |
1477 | } |
1478 | ||
1479 | /** | |
ff7f4cb4 | 1480 | * rg_mblk_search - find a group of multiple free blocks to form a reservation |
8e2e0047 | 1481 | * @rgd: the resource group descriptor |
8e2e0047 | 1482 | * @ip: pointer to the inode for which we're reserving blocks |
7b9cff46 | 1483 | * @ap: the allocation parameters |
8e2e0047 | 1484 | * |
8e2e0047 BP |
1485 | */ |
1486 | ||
ff7f4cb4 | 1487 | static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
7b9cff46 | 1488 | const struct gfs2_alloc_parms *ap) |
8e2e0047 | 1489 | { |
ff7f4cb4 SW |
1490 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1491 | u64 goal; | |
a097dc7e | 1492 | struct gfs2_blkreserv *rs = &ip->i_res; |
ff7f4cb4 SW |
1493 | u32 extlen; |
1494 | u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved; | |
1495 | int ret; | |
af21ca8e | 1496 | struct inode *inode = &ip->i_inode; |
8e2e0047 | 1497 | |
af21ca8e BP |
1498 | if (S_ISDIR(inode->i_mode)) |
1499 | extlen = 1; | |
1500 | else { | |
7b9cff46 | 1501 | extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target); |
af21ca8e BP |
1502 | extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks); |
1503 | } | |
ff7f4cb4 | 1504 | if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen)) |
c743ffd0 SW |
1505 | return; |
1506 | ||
8e2e0047 BP |
1507 | /* Find bitmap block that contains bits for goal block */ |
1508 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
ff7f4cb4 | 1509 | goal = ip->i_goal; |
8e2e0047 | 1510 | else |
ff7f4cb4 | 1511 | goal = rgd->rd_last_alloc + rgd->rd_data0; |
8e2e0047 | 1512 | |
ff7f4cb4 SW |
1513 | if (WARN_ON(gfs2_rbm_from_block(&rbm, goal))) |
1514 | return; | |
8e2e0047 | 1515 | |
8381e602 | 1516 | ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true); |
ff7f4cb4 SW |
1517 | if (ret == 0) { |
1518 | rs->rs_rbm = rbm; | |
1519 | rs->rs_free = extlen; | |
1520 | rs->rs_inum = ip->i_no_addr; | |
1521 | rs_insert(ip); | |
13d2eb01 BP |
1522 | } else { |
1523 | if (goal == rgd->rd_last_alloc + rgd->rd_data0) | |
1524 | rgd->rd_last_alloc = 0; | |
8e2e0047 | 1525 | } |
b3e47ca0 BP |
1526 | } |
1527 | ||
5b924ae2 SW |
1528 | /** |
1529 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1530 | * @rgd: The resource group | |
1531 | * @block: The starting block | |
ff7f4cb4 | 1532 | * @length: The required length |
5b924ae2 SW |
1533 | * @ip: Ignore any reservations for this inode |
1534 | * | |
1535 | * If the block does not appear in any reservation, then return the | |
1536 | * block number unchanged. If it does appear in the reservation, then | |
1537 | * keep looking through the tree of reservations in order to find the | |
1538 | * first block number which is not reserved. | |
1539 | */ | |
1540 | ||
1541 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
ff7f4cb4 | 1542 | u32 length, |
5b924ae2 SW |
1543 | const struct gfs2_inode *ip) |
1544 | { | |
1545 | struct gfs2_blkreserv *rs; | |
1546 | struct rb_node *n; | |
1547 | int rc; | |
1548 | ||
1549 | spin_lock(&rgd->rd_rsspin); | |
ff7f4cb4 | 1550 | n = rgd->rd_rstree.rb_node; |
5b924ae2 SW |
1551 | while (n) { |
1552 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
ff7f4cb4 | 1553 | rc = rs_cmp(block, length, rs); |
5b924ae2 SW |
1554 | if (rc < 0) |
1555 | n = n->rb_left; | |
1556 | else if (rc > 0) | |
1557 | n = n->rb_right; | |
1558 | else | |
1559 | break; | |
1560 | } | |
1561 | ||
1562 | if (n) { | |
a097dc7e | 1563 | while ((rs_cmp(block, length, rs) == 0) && (&ip->i_res != rs)) { |
5b924ae2 | 1564 | block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free; |
ff7f4cb4 | 1565 | n = n->rb_right; |
5b924ae2 SW |
1566 | if (n == NULL) |
1567 | break; | |
1568 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1569 | } | |
1570 | } | |
1571 | ||
1572 | spin_unlock(&rgd->rd_rsspin); | |
1573 | return block; | |
1574 | } | |
1575 | ||
5b924ae2 SW |
1576 | /** |
1577 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1578 | * @rbm: The current position in the resource group | |
ff7f4cb4 SW |
1579 | * @ip: The inode for which we are searching for blocks |
1580 | * @minext: The minimum extent length | |
5ce13431 | 1581 | * @maxext: A pointer to the maximum extent structure |
5b924ae2 SW |
1582 | * |
1583 | * This checks the current position in the rgrp to see whether there is | |
1584 | * a reservation covering this block. If not then this function is a | |
1585 | * no-op. If there is, then the position is moved to the end of the | |
1586 | * contiguous reservation(s) so that we are pointing at the first | |
1587 | * non-reserved block. | |
1588 | * | |
1589 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1590 | */ | |
1591 | ||
1592 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
ff7f4cb4 | 1593 | const struct gfs2_inode *ip, |
5ce13431 BP |
1594 | u32 minext, |
1595 | struct gfs2_extent *maxext) | |
5b924ae2 SW |
1596 | { |
1597 | u64 block = gfs2_rbm_to_block(rbm); | |
ff7f4cb4 | 1598 | u32 extlen = 1; |
5b924ae2 SW |
1599 | u64 nblock; |
1600 | int ret; | |
1601 | ||
ff7f4cb4 SW |
1602 | /* |
1603 | * If we have a minimum extent length, then skip over any extent | |
1604 | * which is less than the min extent length in size. | |
1605 | */ | |
1606 | if (minext) { | |
1607 | extlen = gfs2_free_extlen(rbm, minext); | |
5ce13431 | 1608 | if (extlen <= maxext->len) |
ff7f4cb4 SW |
1609 | goto fail; |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * Check the extent which has been found against the reservations | |
1614 | * and skip if parts of it are already reserved | |
1615 | */ | |
1616 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip); | |
5ce13431 BP |
1617 | if (nblock == block) { |
1618 | if (!minext || extlen >= minext) | |
1619 | return 0; | |
1620 | ||
1621 | if (extlen > maxext->len) { | |
1622 | maxext->len = extlen; | |
1623 | maxext->rbm = *rbm; | |
1624 | } | |
ff7f4cb4 | 1625 | fail: |
5ce13431 BP |
1626 | nblock = block + extlen; |
1627 | } | |
5b924ae2 SW |
1628 | ret = gfs2_rbm_from_block(rbm, nblock); |
1629 | if (ret < 0) | |
1630 | return ret; | |
1631 | return 1; | |
1632 | } | |
1633 | ||
1634 | /** | |
1635 | * gfs2_rbm_find - Look for blocks of a particular state | |
1636 | * @rbm: Value/result starting position and final position | |
1637 | * @state: The state which we want to find | |
5ce13431 BP |
1638 | * @minext: Pointer to the requested extent length (NULL for a single block) |
1639 | * This is updated to be the actual reservation size. | |
5b924ae2 SW |
1640 | * @ip: If set, check for reservations |
1641 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping | |
1642 | * around until we've reached the starting point. | |
1643 | * | |
1644 | * Side effects: | |
1645 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1646 | * has no free blocks in it. | |
5ea5050c BP |
1647 | * - If looking for free blocks, we set rd_extfail_pt on each rgrp which |
1648 | * has come up short on a free block search. | |
5b924ae2 SW |
1649 | * |
1650 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1651 | */ | |
1652 | ||
5ce13431 | 1653 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
8381e602 | 1654 | const struct gfs2_inode *ip, bool nowrap) |
5b924ae2 SW |
1655 | { |
1656 | struct buffer_head *bh; | |
e579ed4f | 1657 | int initial_bii; |
5b924ae2 | 1658 | u32 initial_offset; |
5ea5050c BP |
1659 | int first_bii = rbm->bii; |
1660 | u32 first_offset = rbm->offset; | |
5b924ae2 SW |
1661 | u32 offset; |
1662 | u8 *buffer; | |
5b924ae2 SW |
1663 | int n = 0; |
1664 | int iters = rbm->rgd->rd_length; | |
1665 | int ret; | |
e579ed4f | 1666 | struct gfs2_bitmap *bi; |
5ce13431 | 1667 | struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, }; |
5b924ae2 SW |
1668 | |
1669 | /* If we are not starting at the beginning of a bitmap, then we | |
1670 | * need to add one to the bitmap count to ensure that we search | |
1671 | * the starting bitmap twice. | |
1672 | */ | |
1673 | if (rbm->offset != 0) | |
1674 | iters++; | |
1675 | ||
1676 | while(1) { | |
e579ed4f BP |
1677 | bi = rbm_bi(rbm); |
1678 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
5b924ae2 SW |
1679 | (state == GFS2_BLKST_FREE)) |
1680 | goto next_bitmap; | |
1681 | ||
e579ed4f BP |
1682 | bh = bi->bi_bh; |
1683 | buffer = bh->b_data + bi->bi_offset; | |
5b924ae2 | 1684 | WARN_ON(!buffer_uptodate(bh)); |
e579ed4f BP |
1685 | if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
1686 | buffer = bi->bi_clone + bi->bi_offset; | |
5b924ae2 | 1687 | initial_offset = rbm->offset; |
e579ed4f | 1688 | offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state); |
5b924ae2 SW |
1689 | if (offset == BFITNOENT) |
1690 | goto bitmap_full; | |
1691 | rbm->offset = offset; | |
1692 | if (ip == NULL) | |
1693 | return 0; | |
1694 | ||
e579ed4f | 1695 | initial_bii = rbm->bii; |
5ce13431 BP |
1696 | ret = gfs2_reservation_check_and_update(rbm, ip, |
1697 | minext ? *minext : 0, | |
1698 | &maxext); | |
5b924ae2 SW |
1699 | if (ret == 0) |
1700 | return 0; | |
1701 | if (ret > 0) { | |
e579ed4f | 1702 | n += (rbm->bii - initial_bii); |
8d8b752a | 1703 | goto next_iter; |
5b924ae2 | 1704 | } |
5d50d532 | 1705 | if (ret == -E2BIG) { |
e579ed4f | 1706 | rbm->bii = 0; |
5d50d532 | 1707 | rbm->offset = 0; |
e579ed4f | 1708 | n += (rbm->bii - initial_bii); |
5d50d532 SW |
1709 | goto res_covered_end_of_rgrp; |
1710 | } | |
5b924ae2 SW |
1711 | return ret; |
1712 | ||
1713 | bitmap_full: /* Mark bitmap as full and fall through */ | |
a3e32136 | 1714 | if ((state == GFS2_BLKST_FREE) && initial_offset == 0) |
e579ed4f | 1715 | set_bit(GBF_FULL, &bi->bi_flags); |
5b924ae2 SW |
1716 | |
1717 | next_bitmap: /* Find next bitmap in the rgrp */ | |
1718 | rbm->offset = 0; | |
e579ed4f BP |
1719 | rbm->bii++; |
1720 | if (rbm->bii == rbm->rgd->rd_length) | |
1721 | rbm->bii = 0; | |
5d50d532 | 1722 | res_covered_end_of_rgrp: |
e579ed4f | 1723 | if ((rbm->bii == 0) && nowrap) |
5b924ae2 SW |
1724 | break; |
1725 | n++; | |
8d8b752a | 1726 | next_iter: |
5b924ae2 SW |
1727 | if (n >= iters) |
1728 | break; | |
1729 | } | |
1730 | ||
5ce13431 BP |
1731 | if (minext == NULL || state != GFS2_BLKST_FREE) |
1732 | return -ENOSPC; | |
1733 | ||
5ea5050c BP |
1734 | /* If the extent was too small, and it's smaller than the smallest |
1735 | to have failed before, remember for future reference that it's | |
1736 | useless to search this rgrp again for this amount or more. */ | |
1737 | if ((first_offset == 0) && (first_bii == 0) && | |
1738 | (*minext < rbm->rgd->rd_extfail_pt)) | |
1739 | rbm->rgd->rd_extfail_pt = *minext; | |
1740 | ||
5ce13431 BP |
1741 | /* If the maximum extent we found is big enough to fulfill the |
1742 | minimum requirements, use it anyway. */ | |
1743 | if (maxext.len) { | |
1744 | *rbm = maxext.rbm; | |
1745 | *minext = maxext.len; | |
1746 | return 0; | |
1747 | } | |
1748 | ||
5b924ae2 SW |
1749 | return -ENOSPC; |
1750 | } | |
1751 | ||
c8cdf479 SW |
1752 | /** |
1753 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1754 | * @rgd: The rgrp | |
886b1416 BP |
1755 | * @last_unlinked: block address of the last dinode we unlinked |
1756 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1757 | * |
1a0eae88 BP |
1758 | * Returns: 0 if no error |
1759 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1760 | */ |
1761 | ||
044b9414 | 1762 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1763 | { |
5b924ae2 | 1764 | u64 block; |
5f3eae75 | 1765 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1766 | struct gfs2_glock *gl; |
1767 | struct gfs2_inode *ip; | |
1768 | int error; | |
1769 | int found = 0; | |
e579ed4f | 1770 | struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 }; |
c8cdf479 | 1771 | |
5b924ae2 | 1772 | while (1) { |
5f3eae75 | 1773 | down_write(&sdp->sd_log_flush_lock); |
5ce13431 | 1774 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL, |
8381e602 | 1775 | true); |
5f3eae75 | 1776 | up_write(&sdp->sd_log_flush_lock); |
5b924ae2 SW |
1777 | if (error == -ENOSPC) |
1778 | break; | |
1779 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1780 | break; |
b3e47ca0 | 1781 | |
5b924ae2 SW |
1782 | block = gfs2_rbm_to_block(&rbm); |
1783 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1784 | break; | |
1785 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1786 | continue; |
5b924ae2 | 1787 | if (block == skip) |
1e19a195 | 1788 | continue; |
5b924ae2 | 1789 | *last_unlinked = block; |
044b9414 | 1790 | |
5ea31bc0 | 1791 | error = gfs2_glock_get(sdp, block, &gfs2_iopen_glops, CREATE, &gl); |
044b9414 SW |
1792 | if (error) |
1793 | continue; | |
1794 | ||
1795 | /* If the inode is already in cache, we can ignore it here | |
1796 | * because the existing inode disposal code will deal with | |
1797 | * it when all refs have gone away. Accessing gl_object like | |
1798 | * this is not safe in general. Here it is ok because we do | |
1799 | * not dereference the pointer, and we only need an approx | |
1800 | * answer to whether it is NULL or not. | |
1801 | */ | |
1802 | ip = gl->gl_object; | |
1803 | ||
1804 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1805 | gfs2_glock_put(gl); | |
1806 | else | |
1807 | found++; | |
1808 | ||
1809 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1810 | if (found > NR_CPUS) |
044b9414 | 1811 | return; |
c8cdf479 SW |
1812 | } |
1813 | ||
1814 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1815 | return; |
c8cdf479 SW |
1816 | } |
1817 | ||
bcd97c06 SW |
1818 | /** |
1819 | * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested | |
1820 | * @rgd: The rgrp in question | |
1821 | * @loops: An indication of how picky we can be (0=very, 1=less so) | |
1822 | * | |
1823 | * This function uses the recently added glock statistics in order to | |
1824 | * figure out whether a parciular resource group is suffering from | |
1825 | * contention from multiple nodes. This is done purely on the basis | |
1826 | * of timings, since this is the only data we have to work with and | |
1827 | * our aim here is to reject a resource group which is highly contended | |
1828 | * but (very important) not to do this too often in order to ensure that | |
1829 | * we do not land up introducing fragmentation by changing resource | |
1830 | * groups when not actually required. | |
1831 | * | |
1832 | * The calculation is fairly simple, we want to know whether the SRTTB | |
1833 | * (i.e. smoothed round trip time for blocking operations) to acquire | |
1834 | * the lock for this rgrp's glock is significantly greater than the | |
1835 | * time taken for resource groups on average. We introduce a margin in | |
1836 | * the form of the variable @var which is computed as the sum of the two | |
1837 | * respective variences, and multiplied by a factor depending on @loops | |
1838 | * and whether we have a lot of data to base the decision on. This is | |
1839 | * then tested against the square difference of the means in order to | |
1840 | * decide whether the result is statistically significant or not. | |
1841 | * | |
1842 | * Returns: A boolean verdict on the congestion status | |
1843 | */ | |
1844 | ||
1845 | static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops) | |
1846 | { | |
1847 | const struct gfs2_glock *gl = rgd->rd_gl; | |
15562c43 | 1848 | const struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
bcd97c06 | 1849 | struct gfs2_lkstats *st; |
4d207133 BH |
1850 | u64 r_dcount, l_dcount; |
1851 | u64 l_srttb, a_srttb = 0; | |
bcd97c06 | 1852 | s64 srttb_diff; |
4d207133 BH |
1853 | u64 sqr_diff; |
1854 | u64 var; | |
0166b197 | 1855 | int cpu, nonzero = 0; |
bcd97c06 SW |
1856 | |
1857 | preempt_disable(); | |
f4a3ae93 BP |
1858 | for_each_present_cpu(cpu) { |
1859 | st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP]; | |
0166b197 BP |
1860 | if (st->stats[GFS2_LKS_SRTTB]) { |
1861 | a_srttb += st->stats[GFS2_LKS_SRTTB]; | |
1862 | nonzero++; | |
1863 | } | |
f4a3ae93 | 1864 | } |
bcd97c06 | 1865 | st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP]; |
0166b197 BP |
1866 | if (nonzero) |
1867 | do_div(a_srttb, nonzero); | |
bcd97c06 SW |
1868 | r_dcount = st->stats[GFS2_LKS_DCOUNT]; |
1869 | var = st->stats[GFS2_LKS_SRTTVARB] + | |
1870 | gl->gl_stats.stats[GFS2_LKS_SRTTVARB]; | |
1871 | preempt_enable(); | |
1872 | ||
1873 | l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB]; | |
1874 | l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT]; | |
1875 | ||
f4a3ae93 | 1876 | if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0)) |
bcd97c06 SW |
1877 | return false; |
1878 | ||
f4a3ae93 | 1879 | srttb_diff = a_srttb - l_srttb; |
bcd97c06 SW |
1880 | sqr_diff = srttb_diff * srttb_diff; |
1881 | ||
1882 | var *= 2; | |
1883 | if (l_dcount < 8 || r_dcount < 8) | |
1884 | var *= 2; | |
1885 | if (loops == 1) | |
1886 | var *= 2; | |
1887 | ||
1888 | return ((srttb_diff < 0) && (sqr_diff > var)); | |
1889 | } | |
1890 | ||
1891 | /** | |
1892 | * gfs2_rgrp_used_recently | |
1893 | * @rs: The block reservation with the rgrp to test | |
1894 | * @msecs: The time limit in milliseconds | |
1895 | * | |
1896 | * Returns: True if the rgrp glock has been used within the time limit | |
1897 | */ | |
1898 | static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs, | |
1899 | u64 msecs) | |
1900 | { | |
1901 | u64 tdiff; | |
1902 | ||
1903 | tdiff = ktime_to_ns(ktime_sub(ktime_get_real(), | |
1904 | rs->rs_rbm.rgd->rd_gl->gl_dstamp)); | |
1905 | ||
1906 | return tdiff > (msecs * 1000 * 1000); | |
1907 | } | |
1908 | ||
9dbe9610 SW |
1909 | static u32 gfs2_orlov_skip(const struct gfs2_inode *ip) |
1910 | { | |
1911 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1912 | u32 skip; | |
1913 | ||
1914 | get_random_bytes(&skip, sizeof(skip)); | |
1915 | return skip % sdp->sd_rgrps; | |
1916 | } | |
1917 | ||
c743ffd0 SW |
1918 | static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin) |
1919 | { | |
1920 | struct gfs2_rgrpd *rgd = *pos; | |
aa8920c9 | 1921 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
c743ffd0 SW |
1922 | |
1923 | rgd = gfs2_rgrpd_get_next(rgd); | |
1924 | if (rgd == NULL) | |
aa8920c9 | 1925 | rgd = gfs2_rgrpd_get_first(sdp); |
c743ffd0 SW |
1926 | *pos = rgd; |
1927 | if (rgd != begin) /* If we didn't wrap */ | |
1928 | return true; | |
1929 | return false; | |
1930 | } | |
1931 | ||
0e27c18c BP |
1932 | /** |
1933 | * fast_to_acquire - determine if a resource group will be fast to acquire | |
1934 | * | |
1935 | * If this is one of our preferred rgrps, it should be quicker to acquire, | |
1936 | * because we tried to set ourselves up as dlm lock master. | |
1937 | */ | |
1938 | static inline int fast_to_acquire(struct gfs2_rgrpd *rgd) | |
1939 | { | |
1940 | struct gfs2_glock *gl = rgd->rd_gl; | |
1941 | ||
1942 | if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) && | |
1943 | !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
1944 | !test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
1945 | return 1; | |
1946 | if (rgd->rd_flags & GFS2_RDF_PREFERRED) | |
1947 | return 1; | |
1948 | return 0; | |
1949 | } | |
1950 | ||
b3b94faa | 1951 | /** |
666d1d8a | 1952 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1953 | * @ip: the inode to reserve space for |
7b9cff46 | 1954 | * @ap: the allocation parameters |
b3b94faa | 1955 | * |
25435e5e AD |
1956 | * We try our best to find an rgrp that has at least ap->target blocks |
1957 | * available. After a couple of passes (loops == 2), the prospects of finding | |
1958 | * such an rgrp diminish. At this stage, we return the first rgrp that has | |
1959 | * atleast ap->min_target blocks available. Either way, we set ap->allowed to | |
1960 | * the number of blocks available in the chosen rgrp. | |
1961 | * | |
1962 | * Returns: 0 on success, | |
1963 | * -ENOMEM if a suitable rgrp can't be found | |
1964 | * errno otherwise | |
b3b94faa DT |
1965 | */ |
1966 | ||
25435e5e | 1967 | int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap) |
b3b94faa | 1968 | { |
feaa7bba | 1969 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 1970 | struct gfs2_rgrpd *begin = NULL; |
a097dc7e | 1971 | struct gfs2_blkreserv *rs = &ip->i_res; |
bcd97c06 | 1972 | int error = 0, rg_locked, flags = 0; |
666d1d8a | 1973 | u64 last_unlinked = NO_BLOCK; |
7c9ca621 | 1974 | int loops = 0; |
9dbe9610 | 1975 | u32 skip = 0; |
b3b94faa | 1976 | |
90306c41 BM |
1977 | if (sdp->sd_args.ar_rgrplvb) |
1978 | flags |= GL_SKIP; | |
7b9cff46 | 1979 | if (gfs2_assert_warn(sdp, ap->target)) |
c743ffd0 | 1980 | return -EINVAL; |
8e2e0047 | 1981 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1982 | begin = rs->rs_rbm.rgd; |
8e2e0047 | 1983 | } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) { |
4a993fb1 | 1984 | rs->rs_rbm.rgd = begin = ip->i_rgd; |
8e2e0047 | 1985 | } else { |
00a158be | 1986 | check_and_update_goal(ip); |
4a993fb1 | 1987 | rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 1988 | } |
7b9cff46 | 1989 | if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV)) |
9dbe9610 | 1990 | skip = gfs2_orlov_skip(ip); |
4a993fb1 | 1991 | if (rs->rs_rbm.rgd == NULL) |
7c9ca621 BP |
1992 | return -EBADSLT; |
1993 | ||
1994 | while (loops < 3) { | |
c743ffd0 SW |
1995 | rg_locked = 1; |
1996 | ||
1997 | if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) { | |
1998 | rg_locked = 0; | |
9dbe9610 SW |
1999 | if (skip && skip--) |
2000 | goto next_rgrp; | |
0e27c18c BP |
2001 | if (!gfs2_rs_active(rs)) { |
2002 | if (loops == 0 && | |
2003 | !fast_to_acquire(rs->rs_rbm.rgd)) | |
2004 | goto next_rgrp; | |
2005 | if ((loops < 2) && | |
2006 | gfs2_rgrp_used_recently(rs, 1000) && | |
2007 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
2008 | goto next_rgrp; | |
2009 | } | |
4a993fb1 | 2010 | error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl, |
8e2e0047 BP |
2011 | LM_ST_EXCLUSIVE, flags, |
2012 | &rs->rs_rgd_gh); | |
c743ffd0 SW |
2013 | if (unlikely(error)) |
2014 | return error; | |
bcd97c06 SW |
2015 | if (!gfs2_rs_active(rs) && (loops < 2) && |
2016 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
2017 | goto skip_rgrp; | |
c743ffd0 | 2018 | if (sdp->sd_args.ar_rgrplvb) { |
4a993fb1 | 2019 | error = update_rgrp_lvb(rs->rs_rbm.rgd); |
c743ffd0 | 2020 | if (unlikely(error)) { |
90306c41 BM |
2021 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
2022 | return error; | |
2023 | } | |
2024 | } | |
292c8c14 | 2025 | } |
666d1d8a | 2026 | |
c743ffd0 | 2027 | /* Skip unuseable resource groups */ |
5ea5050c BP |
2028 | if ((rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC | |
2029 | GFS2_RDF_ERROR)) || | |
25435e5e | 2030 | (loops == 0 && ap->target > rs->rs_rbm.rgd->rd_extfail_pt)) |
c743ffd0 SW |
2031 | goto skip_rgrp; |
2032 | ||
2033 | if (sdp->sd_args.ar_rgrplvb) | |
2034 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); | |
2035 | ||
2036 | /* Get a reservation if we don't already have one */ | |
2037 | if (!gfs2_rs_active(rs)) | |
7b9cff46 | 2038 | rg_mblk_search(rs->rs_rbm.rgd, ip, ap); |
c743ffd0 SW |
2039 | |
2040 | /* Skip rgrps when we can't get a reservation on first pass */ | |
2041 | if (!gfs2_rs_active(rs) && (loops < 1)) | |
2042 | goto check_rgrp; | |
2043 | ||
2044 | /* If rgrp has enough free space, use it */ | |
25435e5e AD |
2045 | if (rs->rs_rbm.rgd->rd_free_clone >= ap->target || |
2046 | (loops == 2 && ap->min_target && | |
2047 | rs->rs_rbm.rgd->rd_free_clone >= ap->min_target)) { | |
c743ffd0 | 2048 | ip->i_rgd = rs->rs_rbm.rgd; |
25435e5e | 2049 | ap->allowed = ip->i_rgd->rd_free_clone; |
c743ffd0 | 2050 | return 0; |
b3b94faa | 2051 | } |
c743ffd0 SW |
2052 | check_rgrp: |
2053 | /* Check for unlinked inodes which can be reclaimed */ | |
2054 | if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK) | |
2055 | try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked, | |
2056 | ip->i_no_addr); | |
2057 | skip_rgrp: | |
1330edbe BP |
2058 | /* Drop reservation, if we couldn't use reserved rgrp */ |
2059 | if (gfs2_rs_active(rs)) | |
2060 | gfs2_rs_deltree(rs); | |
2061 | ||
c743ffd0 SW |
2062 | /* Unlock rgrp if required */ |
2063 | if (!rg_locked) | |
2064 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
2065 | next_rgrp: | |
2066 | /* Find the next rgrp, and continue looking */ | |
2067 | if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin)) | |
2068 | continue; | |
9dbe9610 SW |
2069 | if (skip) |
2070 | continue; | |
c743ffd0 SW |
2071 | |
2072 | /* If we've scanned all the rgrps, but found no free blocks | |
2073 | * then this checks for some less likely conditions before | |
2074 | * trying again. | |
2075 | */ | |
c743ffd0 SW |
2076 | loops++; |
2077 | /* Check that fs hasn't grown if writing to rindex */ | |
2078 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
2079 | error = gfs2_ri_update(ip); | |
2080 | if (error) | |
2081 | return error; | |
2082 | } | |
2083 | /* Flushing the log may release space */ | |
2084 | if (loops == 2) | |
24972557 | 2085 | gfs2_log_flush(sdp, NULL, NORMAL_FLUSH); |
b3b94faa | 2086 | } |
b3b94faa | 2087 | |
c743ffd0 | 2088 | return -ENOSPC; |
b3b94faa DT |
2089 | } |
2090 | ||
2091 | /** | |
2092 | * gfs2_inplace_release - release an inplace reservation | |
2093 | * @ip: the inode the reservation was taken out on | |
2094 | * | |
2095 | * Release a reservation made by gfs2_inplace_reserve(). | |
2096 | */ | |
2097 | ||
2098 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
2099 | { | |
a097dc7e | 2100 | struct gfs2_blkreserv *rs = &ip->i_res; |
b3b94faa | 2101 | |
6df9f9a2 | 2102 | if (gfs2_holder_initialized(&rs->rs_rgd_gh)) |
564e12b1 | 2103 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
b3b94faa DT |
2104 | } |
2105 | ||
2106 | /** | |
2107 | * gfs2_get_block_type - Check a block in a RG is of given type | |
2108 | * @rgd: the resource group holding the block | |
2109 | * @block: the block number | |
2110 | * | |
2111 | * Returns: The block type (GFS2_BLKST_*) | |
2112 | */ | |
2113 | ||
acf7e244 | 2114 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 2115 | { |
3983903a SW |
2116 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
2117 | int ret; | |
b3b94faa | 2118 | |
3983903a SW |
2119 | ret = gfs2_rbm_from_block(&rbm, block); |
2120 | WARN_ON_ONCE(ret != 0); | |
b3b94faa | 2121 | |
c04a2ef3 | 2122 | return gfs2_testbit(&rbm); |
b3b94faa DT |
2123 | } |
2124 | ||
60a0b8f9 | 2125 | |
b3e47ca0 BP |
2126 | /** |
2127 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 2128 | * @rbm: the resource group information |
b3e47ca0 | 2129 | * @dinode: TRUE if the first block we allocate is for a dinode |
c04a2ef3 | 2130 | * @n: The extent length (value/result) |
b3e47ca0 | 2131 | * |
c04a2ef3 | 2132 | * Add the bitmap buffer to the transaction. |
b3e47ca0 | 2133 | * Set the found bits to @new_state to change block's allocation state. |
b3e47ca0 | 2134 | */ |
c04a2ef3 | 2135 | static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
4a993fb1 | 2136 | unsigned int *n) |
b3e47ca0 | 2137 | { |
c04a2ef3 | 2138 | struct gfs2_rbm pos = { .rgd = rbm->rgd, }; |
b3e47ca0 | 2139 | const unsigned int elen = *n; |
c04a2ef3 SW |
2140 | u64 block; |
2141 | int ret; | |
b3e47ca0 | 2142 | |
c04a2ef3 SW |
2143 | *n = 1; |
2144 | block = gfs2_rbm_to_block(rbm); | |
e579ed4f | 2145 | gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh); |
3e6339dd | 2146 | gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
c04a2ef3 | 2147 | block++; |
60a0b8f9 | 2148 | while (*n < elen) { |
c04a2ef3 | 2149 | ret = gfs2_rbm_from_block(&pos, block); |
0688a5ec | 2150 | if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE) |
60a0b8f9 | 2151 | break; |
e579ed4f | 2152 | gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh); |
3e6339dd | 2153 | gfs2_setbit(&pos, true, GFS2_BLKST_USED); |
60a0b8f9 | 2154 | (*n)++; |
c04a2ef3 | 2155 | block++; |
c8cdf479 | 2156 | } |
b3b94faa DT |
2157 | } |
2158 | ||
2159 | /** | |
2160 | * rgblk_free - Change alloc state of given block(s) | |
2161 | * @sdp: the filesystem | |
2162 | * @bstart: the start of a run of blocks to free | |
2163 | * @blen: the length of the block run (all must lie within ONE RG!) | |
2164 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
2165 | * | |
2166 | * Returns: Resource group containing the block(s) | |
2167 | */ | |
2168 | ||
cd915493 SW |
2169 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
2170 | u32 blen, unsigned char new_state) | |
b3b94faa | 2171 | { |
3b1d0b9d | 2172 | struct gfs2_rbm rbm; |
d24e0569 | 2173 | struct gfs2_bitmap *bi, *bi_prev = NULL; |
b3b94faa | 2174 | |
3b1d0b9d SW |
2175 | rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
2176 | if (!rbm.rgd) { | |
b3b94faa | 2177 | if (gfs2_consist(sdp)) |
382066da | 2178 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
2179 | return NULL; |
2180 | } | |
2181 | ||
d24e0569 | 2182 | gfs2_rbm_from_block(&rbm, bstart); |
b3b94faa | 2183 | while (blen--) { |
e579ed4f | 2184 | bi = rbm_bi(&rbm); |
d24e0569 BP |
2185 | if (bi != bi_prev) { |
2186 | if (!bi->bi_clone) { | |
2187 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
2188 | GFP_NOFS | __GFP_NOFAIL); | |
2189 | memcpy(bi->bi_clone + bi->bi_offset, | |
2190 | bi->bi_bh->b_data + bi->bi_offset, | |
2191 | bi->bi_len); | |
2192 | } | |
2193 | gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh); | |
2194 | bi_prev = bi; | |
b3b94faa | 2195 | } |
3e6339dd | 2196 | gfs2_setbit(&rbm, false, new_state); |
d24e0569 | 2197 | gfs2_rbm_incr(&rbm); |
b3b94faa DT |
2198 | } |
2199 | ||
3b1d0b9d | 2200 | return rbm.rgd; |
b3b94faa DT |
2201 | } |
2202 | ||
2203 | /** | |
09010978 SW |
2204 | * gfs2_rgrp_dump - print out an rgrp |
2205 | * @seq: The iterator | |
2206 | * @gl: The glock in question | |
2207 | * | |
2208 | */ | |
2209 | ||
ac3beb6a | 2210 | void gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) |
09010978 | 2211 | { |
8e2e0047 BP |
2212 | struct gfs2_rgrpd *rgd = gl->gl_object; |
2213 | struct gfs2_blkreserv *trs; | |
2214 | const struct rb_node *n; | |
2215 | ||
09010978 | 2216 | if (rgd == NULL) |
ac3beb6a | 2217 | return; |
5ea5050c | 2218 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u e:%u\n", |
09010978 | 2219 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 | 2220 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
5ea5050c | 2221 | rgd->rd_reserved, rgd->rd_extfail_pt); |
8e2e0047 BP |
2222 | spin_lock(&rgd->rd_rsspin); |
2223 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { | |
2224 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
2225 | dump_rs(seq, trs); | |
2226 | } | |
2227 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
2228 | } |
2229 | ||
6050b9c7 SW |
2230 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
2231 | { | |
2232 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2233 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 2234 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
2235 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
2236 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
2237 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
2238 | } | |
2239 | ||
8e2e0047 | 2240 | /** |
5b924ae2 SW |
2241 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
2242 | * @ip: The inode we have just allocated blocks for | |
2243 | * @rbm: The start of the allocated blocks | |
2244 | * @len: The extent length | |
8e2e0047 | 2245 | * |
5b924ae2 SW |
2246 | * Adjusts a reservation after an allocation has taken place. If the |
2247 | * reservation does not match the allocation, or if it is now empty | |
2248 | * then it is removed. | |
8e2e0047 | 2249 | */ |
5b924ae2 SW |
2250 | |
2251 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
2252 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 | 2253 | { |
a097dc7e | 2254 | struct gfs2_blkreserv *rs = &ip->i_res; |
5b924ae2 SW |
2255 | struct gfs2_rgrpd *rgd = rbm->rgd; |
2256 | unsigned rlen; | |
2257 | u64 block; | |
2258 | int ret; | |
8e2e0047 | 2259 | |
5b924ae2 SW |
2260 | spin_lock(&rgd->rd_rsspin); |
2261 | if (gfs2_rs_active(rs)) { | |
2262 | if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) { | |
2263 | block = gfs2_rbm_to_block(rbm); | |
2264 | ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len); | |
2265 | rlen = min(rs->rs_free, len); | |
2266 | rs->rs_free -= rlen; | |
2267 | rgd->rd_reserved -= rlen; | |
9e733d39 | 2268 | trace_gfs2_rs(rs, TRACE_RS_CLAIM); |
5b924ae2 SW |
2269 | if (rs->rs_free && !ret) |
2270 | goto out; | |
1a855033 BP |
2271 | /* We used up our block reservation, so we should |
2272 | reserve more blocks next time. */ | |
2273 | atomic_add(RGRP_RSRV_ADDBLKS, &rs->rs_sizehint); | |
5b924ae2 | 2274 | } |
20095218 | 2275 | __rs_deltree(rs); |
8e2e0047 | 2276 | } |
5b924ae2 SW |
2277 | out: |
2278 | spin_unlock(&rgd->rd_rsspin); | |
8e2e0047 BP |
2279 | } |
2280 | ||
9e07f2cb SW |
2281 | /** |
2282 | * gfs2_set_alloc_start - Set starting point for block allocation | |
2283 | * @rbm: The rbm which will be set to the required location | |
2284 | * @ip: The gfs2 inode | |
2285 | * @dinode: Flag to say if allocation includes a new inode | |
2286 | * | |
2287 | * This sets the starting point from the reservation if one is active | |
2288 | * otherwise it falls back to guessing a start point based on the | |
2289 | * inode's goal block or the last allocation point in the rgrp. | |
2290 | */ | |
2291 | ||
2292 | static void gfs2_set_alloc_start(struct gfs2_rbm *rbm, | |
2293 | const struct gfs2_inode *ip, bool dinode) | |
2294 | { | |
2295 | u64 goal; | |
2296 | ||
a097dc7e BP |
2297 | if (gfs2_rs_active(&ip->i_res)) { |
2298 | *rbm = ip->i_res.rs_rbm; | |
9e07f2cb SW |
2299 | return; |
2300 | } | |
2301 | ||
2302 | if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal)) | |
2303 | goal = ip->i_goal; | |
2304 | else | |
2305 | goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0; | |
2306 | ||
2307 | gfs2_rbm_from_block(rbm, goal); | |
2308 | } | |
2309 | ||
09010978 | 2310 | /** |
6e87ed0f | 2311 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 2312 | * @ip: the inode to allocate the block for |
09010978 | 2313 | * @bn: Used to return the starting block number |
8e2e0047 | 2314 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 2315 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 2316 | * @generation: the generation number of the inode |
b3b94faa | 2317 | * |
09010978 | 2318 | * Returns: 0 or error |
b3b94faa DT |
2319 | */ |
2320 | ||
6a8099ed | 2321 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 2322 | bool dinode, u64 *generation) |
b3b94faa | 2323 | { |
feaa7bba | 2324 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 2325 | struct buffer_head *dibh; |
4a993fb1 | 2326 | struct gfs2_rbm rbm = { .rgd = ip->i_rgd, }; |
6a8099ed | 2327 | unsigned int ndata; |
3c5d785a | 2328 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 2329 | int error; |
b3b94faa | 2330 | |
9e07f2cb | 2331 | gfs2_set_alloc_start(&rbm, ip, dinode); |
8381e602 | 2332 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false); |
62e252ee | 2333 | |
137834a6 | 2334 | if (error == -ENOSPC) { |
9e07f2cb | 2335 | gfs2_set_alloc_start(&rbm, ip, dinode); |
8381e602 | 2336 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false); |
137834a6 SW |
2337 | } |
2338 | ||
62e252ee | 2339 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
5b924ae2 | 2340 | if (error) { |
5ea5050c | 2341 | fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n", |
9e733d39 | 2342 | (unsigned long long)ip->i_no_addr, error, *nblocks, |
5ea5050c BP |
2343 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags), |
2344 | rbm.rgd->rd_extfail_pt); | |
62e252ee | 2345 | goto rgrp_error; |
8e2e0047 | 2346 | } |
62e252ee | 2347 | |
c04a2ef3 SW |
2348 | gfs2_alloc_extent(&rbm, dinode, nblocks); |
2349 | block = gfs2_rbm_to_block(&rbm); | |
c743ffd0 | 2350 | rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0; |
a097dc7e | 2351 | if (gfs2_rs_active(&ip->i_res)) |
5b924ae2 | 2352 | gfs2_adjust_reservation(ip, &rbm, *nblocks); |
6a8099ed SW |
2353 | ndata = *nblocks; |
2354 | if (dinode) | |
2355 | ndata--; | |
b3e47ca0 | 2356 | |
3c5d785a | 2357 | if (!dinode) { |
6a8099ed | 2358 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
2359 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2360 | if (error == 0) { | |
2361 | struct gfs2_dinode *di = | |
2362 | (struct gfs2_dinode *)dibh->b_data; | |
350a9b0a | 2363 | gfs2_trans_add_meta(ip->i_gl, dibh); |
3c5d785a BP |
2364 | di->di_goal_meta = di->di_goal_data = |
2365 | cpu_to_be64(ip->i_goal); | |
2366 | brelse(dibh); | |
2367 | } | |
d9ba7615 | 2368 | } |
4a993fb1 | 2369 | if (rbm.rgd->rd_free < *nblocks) { |
fc554ed3 | 2370 | pr_warn("nblocks=%u\n", *nblocks); |
09010978 | 2371 | goto rgrp_error; |
8e2e0047 | 2372 | } |
09010978 | 2373 | |
4a993fb1 | 2374 | rbm.rgd->rd_free -= *nblocks; |
3c5d785a | 2375 | if (dinode) { |
4a993fb1 SW |
2376 | rbm.rgd->rd_dinodes++; |
2377 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2378 | if (*generation == 0) |
4a993fb1 | 2379 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2380 | } |
b3b94faa | 2381 | |
350a9b0a | 2382 | gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh); |
4a993fb1 SW |
2383 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); |
2384 | gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
b3b94faa | 2385 | |
6a8099ed | 2386 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a | 2387 | if (dinode) |
b2c8b3ea | 2388 | gfs2_trans_add_unrevoke(sdp, block, *nblocks); |
6a8099ed | 2389 | |
fd4b4e04 | 2390 | gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa | 2391 | |
4a993fb1 SW |
2392 | rbm.rgd->rd_free_clone -= *nblocks; |
2393 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, | |
6e87ed0f | 2394 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2395 | *bn = block; |
2396 | return 0; | |
2397 | ||
2398 | rgrp_error: | |
4a993fb1 | 2399 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2400 | return -EIO; |
b3b94faa DT |
2401 | } |
2402 | ||
2403 | /** | |
46fcb2ed | 2404 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
2405 | * @ip: the inode these blocks are being freed from |
2406 | * @bstart: first block of a run of contiguous blocks | |
2407 | * @blen: the length of the block run | |
46fcb2ed | 2408 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2409 | * |
2410 | */ | |
2411 | ||
46fcb2ed | 2412 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 2413 | { |
feaa7bba | 2414 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2415 | struct gfs2_rgrpd *rgd; |
2416 | ||
2417 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
2418 | if (!rgd) | |
2419 | return; | |
41db1ab9 | 2420 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2421 | rgd->rd_free += blen; |
66fc061b | 2422 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
350a9b0a | 2423 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2424 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 | 2425 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 2426 | |
6d3117b4 | 2427 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 2428 | if (meta || ip->i_depth) |
6d3117b4 | 2429 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 2430 | } |
b3b94faa | 2431 | |
4c16c36a BP |
2432 | /** |
2433 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2434 | * @ip: the inode these blocks are being freed from | |
2435 | * @bstart: first block of a run of contiguous blocks | |
2436 | * @blen: the length of the block run | |
2437 | * | |
2438 | */ | |
2439 | ||
2440 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
2441 | { | |
2442 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2443 | ||
46fcb2ed | 2444 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 2445 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2446 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2447 | } |
2448 | ||
feaa7bba SW |
2449 | void gfs2_unlink_di(struct inode *inode) |
2450 | { | |
2451 | struct gfs2_inode *ip = GFS2_I(inode); | |
2452 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2453 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2454 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
2455 | |
2456 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
2457 | if (!rgd) | |
2458 | return; | |
41db1ab9 | 2459 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
350a9b0a | 2460 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2461 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2462 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2463 | update_rgrp_lvb_unlinked(rgd, 1); | |
feaa7bba SW |
2464 | } |
2465 | ||
cd915493 | 2466 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
2467 | { |
2468 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2469 | struct gfs2_rgrpd *tmp_rgd; | |
2470 | ||
2471 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
2472 | if (!tmp_rgd) | |
2473 | return; | |
2474 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
2475 | ||
73f74948 | 2476 | if (!rgd->rd_dinodes) |
b3b94faa | 2477 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2478 | rgd->rd_dinodes--; |
cfc8b549 | 2479 | rgd->rd_free++; |
b3b94faa | 2480 | |
350a9b0a | 2481 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2482 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2483 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2484 | update_rgrp_lvb_unlinked(rgd, -1); | |
b3b94faa DT |
2485 | |
2486 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
2487 | } |
2488 | ||
b3b94faa DT |
2489 | |
2490 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
2491 | { | |
dbb7cae2 | 2492 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
41db1ab9 | 2493 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2494 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 2495 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2496 | } |
2497 | ||
acf7e244 SW |
2498 | /** |
2499 | * gfs2_check_blk_type - Check the type of a block | |
2500 | * @sdp: The superblock | |
2501 | * @no_addr: The block number to check | |
2502 | * @type: The block type we are looking for | |
2503 | * | |
2504 | * Returns: 0 if the block type matches the expected type | |
2505 | * -ESTALE if it doesn't match | |
2506 | * or -ve errno if something went wrong while checking | |
2507 | */ | |
2508 | ||
2509 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2510 | { | |
2511 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2512 | struct gfs2_holder rgd_gh; |
58884c4d | 2513 | int error = -EINVAL; |
acf7e244 | 2514 | |
66fc061b | 2515 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2516 | if (!rgd) |
8339ee54 | 2517 | goto fail; |
acf7e244 SW |
2518 | |
2519 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2520 | if (error) | |
8339ee54 | 2521 | goto fail; |
acf7e244 SW |
2522 | |
2523 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
2524 | error = -ESTALE; | |
2525 | ||
2526 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
2527 | fail: |
2528 | return error; | |
2529 | } | |
2530 | ||
b3b94faa DT |
2531 | /** |
2532 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2533 | * @ip: the inode |
b3b94faa DT |
2534 | * @rlist: the list of resource groups |
2535 | * @block: the block | |
2536 | * | |
2537 | * Figure out what RG a block belongs to and add that RG to the list | |
2538 | * | |
2539 | * FIXME: Don't use NOFAIL | |
2540 | * | |
2541 | */ | |
2542 | ||
70b0c365 | 2543 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2544 | u64 block) |
b3b94faa | 2545 | { |
70b0c365 | 2546 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2547 | struct gfs2_rgrpd *rgd; |
2548 | struct gfs2_rgrpd **tmp; | |
2549 | unsigned int new_space; | |
2550 | unsigned int x; | |
2551 | ||
2552 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2553 | return; | |
2554 | ||
70b0c365 SW |
2555 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
2556 | rgd = ip->i_rgd; | |
2557 | else | |
66fc061b | 2558 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 2559 | if (!rgd) { |
70b0c365 | 2560 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
2561 | return; |
2562 | } | |
70b0c365 | 2563 | ip->i_rgd = rgd; |
b3b94faa DT |
2564 | |
2565 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2566 | if (rlist->rl_rgd[x] == rgd) | |
2567 | return; | |
2568 | ||
2569 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2570 | new_space = rlist->rl_space + 10; | |
2571 | ||
2572 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2573 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2574 | |
2575 | if (rlist->rl_rgd) { | |
2576 | memcpy(tmp, rlist->rl_rgd, | |
2577 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2578 | kfree(rlist->rl_rgd); | |
2579 | } | |
2580 | ||
2581 | rlist->rl_space = new_space; | |
2582 | rlist->rl_rgd = tmp; | |
2583 | } | |
2584 | ||
2585 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2586 | } | |
2587 | ||
2588 | /** | |
2589 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2590 | * and initialize an array of glock holders for them | |
2591 | * @rlist: the list of resource groups | |
2592 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
2593 | * |
2594 | * FIXME: Don't use NOFAIL | |
2595 | * | |
2596 | */ | |
2597 | ||
fe6c991c | 2598 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
2599 | { |
2600 | unsigned int x; | |
2601 | ||
6df9f9a2 | 2602 | rlist->rl_ghs = kmalloc(rlist->rl_rgrps * sizeof(struct gfs2_holder), |
dd894be8 | 2603 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2604 | for (x = 0; x < rlist->rl_rgrps; x++) |
2605 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 2606 | state, 0, |
b3b94faa DT |
2607 | &rlist->rl_ghs[x]); |
2608 | } | |
2609 | ||
2610 | /** | |
2611 | * gfs2_rlist_free - free a resource group list | |
27ff6a0f | 2612 | * @rlist: the list of resource groups |
b3b94faa DT |
2613 | * |
2614 | */ | |
2615 | ||
2616 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2617 | { | |
2618 | unsigned int x; | |
2619 | ||
2620 | kfree(rlist->rl_rgd); | |
2621 | ||
2622 | if (rlist->rl_ghs) { | |
2623 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2624 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2625 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2626 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2627 | } |
2628 | } | |
2629 |