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 BP |
75 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
76 | const struct gfs2_inode *ip, bool nowrap, | |
77 | const struct gfs2_alloc_parms *ap); | |
ff7f4cb4 SW |
78 | |
79 | ||
88c8ab1f SW |
80 | /** |
81 | * gfs2_setbit - Set a bit in the bitmaps | |
3e6339dd SW |
82 | * @rbm: The position of the bit to set |
83 | * @do_clone: Also set the clone bitmap, if it exists | |
88c8ab1f SW |
84 | * @new_state: the new state of the block |
85 | * | |
86 | */ | |
87 | ||
3e6339dd | 88 | static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone, |
06344b91 | 89 | unsigned char new_state) |
88c8ab1f | 90 | { |
b45e41d7 | 91 | unsigned char *byte1, *byte2, *end, cur_state; |
e579ed4f BP |
92 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
93 | unsigned int buflen = bi->bi_len; | |
3e6339dd | 94 | const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 95 | |
e579ed4f BP |
96 | byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY); |
97 | end = bi->bi_bh->b_data + bi->bi_offset + buflen; | |
88c8ab1f | 98 | |
b45e41d7 | 99 | BUG_ON(byte1 >= end); |
88c8ab1f | 100 | |
b45e41d7 | 101 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 102 | |
b45e41d7 | 103 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
d77d1b58 JP |
104 | pr_warn("buf_blk = 0x%x old_state=%d, new_state=%d\n", |
105 | rbm->offset, cur_state, new_state); | |
106 | pr_warn("rgrp=0x%llx bi_start=0x%x\n", | |
107 | (unsigned long long)rbm->rgd->rd_addr, bi->bi_start); | |
108 | pr_warn("bi_offset=0x%x bi_len=0x%x\n", | |
109 | bi->bi_offset, bi->bi_len); | |
95c8e17f | 110 | dump_stack(); |
3e6339dd | 111 | gfs2_consist_rgrpd(rbm->rgd); |
b45e41d7 SW |
112 | return; |
113 | } | |
114 | *byte1 ^= (cur_state ^ new_state) << bit; | |
115 | ||
e579ed4f BP |
116 | if (do_clone && bi->bi_clone) { |
117 | byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY); | |
b45e41d7 SW |
118 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
119 | *byte2 ^= (cur_state ^ new_state) << bit; | |
120 | } | |
88c8ab1f SW |
121 | } |
122 | ||
123 | /** | |
124 | * gfs2_testbit - test a bit in the bitmaps | |
c04a2ef3 | 125 | * @rbm: The bit to test |
88c8ab1f | 126 | * |
c04a2ef3 | 127 | * Returns: The two bit block state of the requested bit |
88c8ab1f SW |
128 | */ |
129 | ||
c04a2ef3 | 130 | static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm) |
88c8ab1f | 131 | { |
e579ed4f BP |
132 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
133 | const u8 *buffer = bi->bi_bh->b_data + bi->bi_offset; | |
c04a2ef3 | 134 | const u8 *byte; |
88c8ab1f SW |
135 | unsigned int bit; |
136 | ||
c04a2ef3 SW |
137 | byte = buffer + (rbm->offset / GFS2_NBBY); |
138 | bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 139 | |
c04a2ef3 | 140 | return (*byte >> bit) & GFS2_BIT_MASK; |
88c8ab1f SW |
141 | } |
142 | ||
223b2b88 SW |
143 | /** |
144 | * gfs2_bit_search | |
145 | * @ptr: Pointer to bitmap data | |
146 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
147 | * @state: The state we are searching for | |
148 | * | |
149 | * We xor the bitmap data with a patter which is the bitwise opposite | |
150 | * of what we are looking for, this gives rise to a pattern of ones | |
151 | * wherever there is a match. Since we have two bits per entry, we | |
152 | * take this pattern, shift it down by one place and then and it with | |
153 | * the original. All the even bit positions (0,2,4, etc) then represent | |
154 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
155 | * odd bit positions. | |
156 | * | |
157 | * This allows searching of a whole u64 at once (32 blocks) with a | |
158 | * single test (on 64 bit arches). | |
159 | */ | |
160 | ||
161 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
162 | { | |
163 | u64 tmp; | |
164 | static const u64 search[] = { | |
075ac448 HE |
165 | [0] = 0xffffffffffffffffULL, |
166 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
167 | [2] = 0x5555555555555555ULL, | |
168 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
169 | }; |
170 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
171 | tmp &= (tmp >> 1); | |
172 | tmp &= mask; | |
173 | return tmp; | |
174 | } | |
175 | ||
8e2e0047 BP |
176 | /** |
177 | * rs_cmp - multi-block reservation range compare | |
178 | * @blk: absolute file system block number of the new reservation | |
179 | * @len: number of blocks in the new reservation | |
180 | * @rs: existing reservation to compare against | |
181 | * | |
182 | * returns: 1 if the block range is beyond the reach of the reservation | |
183 | * -1 if the block range is before the start of the reservation | |
184 | * 0 if the block range overlaps with the reservation | |
185 | */ | |
186 | static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs) | |
187 | { | |
4a993fb1 | 188 | u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 BP |
189 | |
190 | if (blk >= startblk + rs->rs_free) | |
191 | return 1; | |
192 | if (blk + len - 1 < startblk) | |
193 | return -1; | |
194 | return 0; | |
195 | } | |
196 | ||
88c8ab1f SW |
197 | /** |
198 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
199 | * a block in a given allocation state. | |
886b1416 | 200 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 201 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 202 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 203 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
204 | * |
205 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
206 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
207 | * beginning of a bitmap block buffer, skipping any header structures, but |
208 | * headers are always a multiple of 64 bits long so that the buffer is | |
209 | * always aligned to a 64 bit boundary. | |
210 | * | |
211 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 212 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 213 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
214 | * |
215 | * Return: the block number (bitmap buffer scope) that was found | |
216 | */ | |
217 | ||
02ab1721 HE |
218 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
219 | u32 goal, u8 state) | |
88c8ab1f | 220 | { |
223b2b88 SW |
221 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
222 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
223 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
224 | u64 tmp; | |
075ac448 | 225 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
226 | u32 bit; |
227 | ||
223b2b88 SW |
228 | /* Mask off bits we don't care about at the start of the search */ |
229 | mask <<= spoint; | |
230 | tmp = gfs2_bit_search(ptr, mask, state); | |
231 | ptr++; | |
232 | while(tmp == 0 && ptr < end) { | |
075ac448 | 233 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 234 | ptr++; |
1f466a47 | 235 | } |
223b2b88 SW |
236 | /* Mask off any bits which are more than len bytes from the start */ |
237 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
238 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
239 | /* Didn't find anything, so return */ | |
240 | if (tmp == 0) | |
241 | return BFITNOENT; | |
242 | ptr--; | |
d8bd504a | 243 | bit = __ffs64(tmp); |
223b2b88 SW |
244 | bit /= 2; /* two bits per entry in the bitmap */ |
245 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
246 | } |
247 | ||
ff7f4cb4 SW |
248 | /** |
249 | * gfs2_rbm_from_block - Set the rbm based upon rgd and block number | |
250 | * @rbm: The rbm with rgd already set correctly | |
251 | * @block: The block number (filesystem relative) | |
252 | * | |
253 | * This sets the bi and offset members of an rbm based on a | |
254 | * resource group and a filesystem relative block number. The | |
255 | * resource group must be set in the rbm on entry, the bi and | |
256 | * offset members will be set by this function. | |
257 | * | |
258 | * Returns: 0 on success, or an error code | |
259 | */ | |
260 | ||
261 | static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block) | |
262 | { | |
263 | u64 rblock = block - rbm->rgd->rd_data0; | |
ff7f4cb4 SW |
264 | |
265 | if (WARN_ON_ONCE(rblock > UINT_MAX)) | |
266 | return -EINVAL; | |
267 | if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data) | |
268 | return -E2BIG; | |
269 | ||
e579ed4f | 270 | rbm->bii = 0; |
a68a0a35 BP |
271 | rbm->offset = (u32)(rblock); |
272 | /* Check if the block is within the first block */ | |
e579ed4f | 273 | if (rbm->offset < rbm_bi(rbm)->bi_blocks) |
a68a0a35 BP |
274 | return 0; |
275 | ||
276 | /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */ | |
277 | rbm->offset += (sizeof(struct gfs2_rgrp) - | |
278 | sizeof(struct gfs2_meta_header)) * GFS2_NBBY; | |
e579ed4f BP |
279 | rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap; |
280 | rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap; | |
ff7f4cb4 SW |
281 | return 0; |
282 | } | |
283 | ||
149ed7f5 BP |
284 | /** |
285 | * gfs2_rbm_incr - increment an rbm structure | |
286 | * @rbm: The rbm with rgd already set correctly | |
287 | * | |
288 | * This function takes an existing rbm structure and increments it to the next | |
289 | * viable block offset. | |
290 | * | |
291 | * Returns: If incrementing the offset would cause the rbm to go past the | |
292 | * end of the rgrp, true is returned, otherwise false. | |
293 | * | |
294 | */ | |
295 | ||
296 | static bool gfs2_rbm_incr(struct gfs2_rbm *rbm) | |
297 | { | |
298 | if (rbm->offset + 1 < rbm_bi(rbm)->bi_blocks) { /* in the same bitmap */ | |
299 | rbm->offset++; | |
300 | return false; | |
301 | } | |
302 | if (rbm->bii == rbm->rgd->rd_length - 1) /* at the last bitmap */ | |
303 | return true; | |
304 | ||
305 | rbm->offset = 0; | |
306 | rbm->bii++; | |
307 | return false; | |
308 | } | |
309 | ||
ff7f4cb4 SW |
310 | /** |
311 | * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned | |
312 | * @rbm: Position to search (value/result) | |
313 | * @n_unaligned: Number of unaligned blocks to check | |
314 | * @len: Decremented for each block found (terminate on zero) | |
315 | * | |
316 | * Returns: true if a non-free block is encountered | |
317 | */ | |
318 | ||
319 | static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len) | |
320 | { | |
ff7f4cb4 SW |
321 | u32 n; |
322 | u8 res; | |
323 | ||
324 | for (n = 0; n < n_unaligned; n++) { | |
325 | res = gfs2_testbit(rbm); | |
326 | if (res != GFS2_BLKST_FREE) | |
327 | return true; | |
328 | (*len)--; | |
329 | if (*len == 0) | |
330 | return true; | |
149ed7f5 | 331 | if (gfs2_rbm_incr(rbm)) |
ff7f4cb4 SW |
332 | return true; |
333 | } | |
334 | ||
335 | return false; | |
336 | } | |
337 | ||
338 | /** | |
339 | * gfs2_free_extlen - Return extent length of free blocks | |
27ff6a0f | 340 | * @rrbm: Starting position |
ff7f4cb4 SW |
341 | * @len: Max length to check |
342 | * | |
343 | * Starting at the block specified by the rbm, see how many free blocks | |
344 | * there are, not reading more than len blocks ahead. This can be done | |
345 | * using memchr_inv when the blocks are byte aligned, but has to be done | |
346 | * on a block by block basis in case of unaligned blocks. Also this | |
347 | * function can cope with bitmap boundaries (although it must stop on | |
348 | * a resource group boundary) | |
349 | * | |
350 | * Returns: Number of free blocks in the extent | |
351 | */ | |
352 | ||
353 | static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len) | |
354 | { | |
355 | struct gfs2_rbm rbm = *rrbm; | |
356 | u32 n_unaligned = rbm.offset & 3; | |
357 | u32 size = len; | |
358 | u32 bytes; | |
359 | u32 chunk_size; | |
360 | u8 *ptr, *start, *end; | |
361 | u64 block; | |
e579ed4f | 362 | struct gfs2_bitmap *bi; |
ff7f4cb4 SW |
363 | |
364 | if (n_unaligned && | |
365 | gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len)) | |
366 | goto out; | |
367 | ||
3701530a | 368 | n_unaligned = len & 3; |
ff7f4cb4 SW |
369 | /* Start is now byte aligned */ |
370 | while (len > 3) { | |
e579ed4f BP |
371 | bi = rbm_bi(&rbm); |
372 | start = bi->bi_bh->b_data; | |
373 | if (bi->bi_clone) | |
374 | start = bi->bi_clone; | |
375 | end = start + bi->bi_bh->b_size; | |
376 | start += bi->bi_offset; | |
ff7f4cb4 SW |
377 | BUG_ON(rbm.offset & 3); |
378 | start += (rbm.offset / GFS2_NBBY); | |
379 | bytes = min_t(u32, len / GFS2_NBBY, (end - start)); | |
380 | ptr = memchr_inv(start, 0, bytes); | |
381 | chunk_size = ((ptr == NULL) ? bytes : (ptr - start)); | |
382 | chunk_size *= GFS2_NBBY; | |
383 | BUG_ON(len < chunk_size); | |
384 | len -= chunk_size; | |
385 | block = gfs2_rbm_to_block(&rbm); | |
15bd50ad BP |
386 | if (gfs2_rbm_from_block(&rbm, block + chunk_size)) { |
387 | n_unaligned = 0; | |
ff7f4cb4 | 388 | break; |
15bd50ad BP |
389 | } |
390 | if (ptr) { | |
391 | n_unaligned = 3; | |
392 | break; | |
393 | } | |
ff7f4cb4 SW |
394 | n_unaligned = len & 3; |
395 | } | |
396 | ||
397 | /* Deal with any bits left over at the end */ | |
398 | if (n_unaligned) | |
399 | gfs2_unaligned_extlen(&rbm, n_unaligned, &len); | |
400 | out: | |
401 | return size - len; | |
402 | } | |
403 | ||
88c8ab1f SW |
404 | /** |
405 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 406 | * @rgd: the resource group descriptor |
88c8ab1f SW |
407 | * @buffer: the buffer that holds the bitmaps |
408 | * @buflen: the length (in bytes) of the buffer | |
409 | * @state: the state of the block we're looking for | |
410 | * | |
411 | * Returns: The number of bits | |
412 | */ | |
413 | ||
110acf38 SW |
414 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
415 | unsigned int buflen, u8 state) | |
88c8ab1f | 416 | { |
110acf38 SW |
417 | const u8 *byte = buffer; |
418 | const u8 *end = buffer + buflen; | |
419 | const u8 state1 = state << 2; | |
420 | const u8 state2 = state << 4; | |
421 | const u8 state3 = state << 6; | |
cd915493 | 422 | u32 count = 0; |
88c8ab1f SW |
423 | |
424 | for (; byte < end; byte++) { | |
425 | if (((*byte) & 0x03) == state) | |
426 | count++; | |
427 | if (((*byte) & 0x0C) == state1) | |
428 | count++; | |
429 | if (((*byte) & 0x30) == state2) | |
430 | count++; | |
431 | if (((*byte) & 0xC0) == state3) | |
432 | count++; | |
433 | } | |
434 | ||
435 | return count; | |
436 | } | |
437 | ||
b3b94faa DT |
438 | /** |
439 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
440 | * @rgd: the rgrp |
441 | * | |
442 | */ | |
443 | ||
444 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
445 | { | |
446 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
447 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 448 | u32 length = rgd->rd_length; |
cd915493 | 449 | u32 count[4], tmp; |
b3b94faa DT |
450 | int buf, x; |
451 | ||
cd915493 | 452 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
453 | |
454 | /* Count # blocks in each of 4 possible allocation states */ | |
455 | for (buf = 0; buf < length; buf++) { | |
456 | bi = rgd->rd_bits + buf; | |
457 | for (x = 0; x < 4; x++) | |
458 | count[x] += gfs2_bitcount(rgd, | |
459 | bi->bi_bh->b_data + | |
460 | bi->bi_offset, | |
461 | bi->bi_len, x); | |
462 | } | |
463 | ||
cfc8b549 | 464 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
465 | if (gfs2_consist_rgrpd(rgd)) |
466 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 467 | count[0], rgd->rd_free); |
b3b94faa DT |
468 | return; |
469 | } | |
470 | ||
73f74948 | 471 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 472 | if (count[1] != tmp) { |
b3b94faa DT |
473 | if (gfs2_consist_rgrpd(rgd)) |
474 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
475 | count[1], tmp); | |
476 | return; | |
477 | } | |
478 | ||
6b946170 | 479 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 480 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 481 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 482 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
483 | return; |
484 | } | |
b3b94faa DT |
485 | } |
486 | ||
bb8d8a6f | 487 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 488 | { |
bb8d8a6f SW |
489 | u64 first = rgd->rd_data0; |
490 | u64 last = first + rgd->rd_data; | |
16910427 | 491 | return first <= block && block < last; |
b3b94faa DT |
492 | } |
493 | ||
494 | /** | |
495 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
496 | * @sdp: The GFS2 superblock | |
886b1416 BP |
497 | * @blk: The data block number |
498 | * @exact: True if this needs to be an exact match | |
b3b94faa DT |
499 | * |
500 | * Returns: The resource group, or NULL if not found | |
501 | */ | |
502 | ||
66fc061b | 503 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 504 | { |
66fc061b | 505 | struct rb_node *n, *next; |
f75bbfb4 | 506 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
507 | |
508 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
509 | n = sdp->sd_rindex_tree.rb_node; |
510 | while (n) { | |
511 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
512 | next = NULL; | |
7c9ca621 | 513 | if (blk < cur->rd_addr) |
66fc061b | 514 | next = n->rb_left; |
f75bbfb4 | 515 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
516 | next = n->rb_right; |
517 | if (next == NULL) { | |
b3b94faa | 518 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
519 | if (exact) { |
520 | if (blk < cur->rd_addr) | |
521 | return NULL; | |
522 | if (blk >= cur->rd_data0 + cur->rd_data) | |
523 | return NULL; | |
524 | } | |
7c9ca621 | 525 | return cur; |
b3b94faa | 526 | } |
66fc061b | 527 | n = next; |
b3b94faa | 528 | } |
b3b94faa DT |
529 | spin_unlock(&sdp->sd_rindex_spin); |
530 | ||
531 | return NULL; | |
532 | } | |
533 | ||
534 | /** | |
535 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
536 | * @sdp: The GFS2 superblock | |
537 | * | |
538 | * Returns: The first rgrp in the filesystem | |
539 | */ | |
540 | ||
541 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
542 | { | |
7c9ca621 BP |
543 | const struct rb_node *n; |
544 | struct gfs2_rgrpd *rgd; | |
545 | ||
8339ee54 | 546 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
547 | n = rb_first(&sdp->sd_rindex_tree); |
548 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 549 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
550 | |
551 | return rgd; | |
b3b94faa DT |
552 | } |
553 | ||
554 | /** | |
555 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 556 | * @rgd: the resource group descriptor |
b3b94faa DT |
557 | * |
558 | * Returns: The next rgrp | |
559 | */ | |
560 | ||
561 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
562 | { | |
7c9ca621 BP |
563 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
564 | const struct rb_node *n; | |
565 | ||
566 | spin_lock(&sdp->sd_rindex_spin); | |
567 | n = rb_next(&rgd->rd_node); | |
568 | if (n == NULL) | |
569 | n = rb_first(&sdp->sd_rindex_tree); | |
570 | ||
571 | if (unlikely(&rgd->rd_node == n)) { | |
572 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 573 | return NULL; |
7c9ca621 BP |
574 | } |
575 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
576 | spin_unlock(&sdp->sd_rindex_spin); | |
577 | return rgd; | |
b3b94faa DT |
578 | } |
579 | ||
00a158be AD |
580 | void check_and_update_goal(struct gfs2_inode *ip) |
581 | { | |
582 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
583 | if (!ip->i_goal || gfs2_blk2rgrpd(sdp, ip->i_goal, 1) == NULL) | |
584 | ip->i_goal = ip->i_no_addr; | |
585 | } | |
586 | ||
8339ee54 SW |
587 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
588 | { | |
589 | int x; | |
590 | ||
591 | for (x = 0; x < rgd->rd_length; x++) { | |
592 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
593 | kfree(bi->bi_clone); | |
594 | bi->bi_clone = NULL; | |
595 | } | |
596 | } | |
597 | ||
0a305e49 BP |
598 | /** |
599 | * gfs2_rs_alloc - make sure we have a reservation assigned to the inode | |
600 | * @ip: the inode for this reservation | |
601 | */ | |
602 | int gfs2_rs_alloc(struct gfs2_inode *ip) | |
603 | { | |
f1213cac | 604 | int error = 0; |
8e2e0047 | 605 | |
f1213cac | 606 | down_write(&ip->i_rw_mutex); |
8e2e0047 | 607 | if (ip->i_res) |
f1213cac | 608 | goto out; |
0a305e49 | 609 | |
f1213cac AD |
610 | ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); |
611 | if (!ip->i_res) { | |
612 | error = -ENOMEM; | |
613 | goto out; | |
614 | } | |
4a993fb1 | 615 | |
f1213cac AD |
616 | RB_CLEAR_NODE(&ip->i_res->rs_node); |
617 | out: | |
0a305e49 | 618 | up_write(&ip->i_rw_mutex); |
441362d0 | 619 | return error; |
0a305e49 BP |
620 | } |
621 | ||
9e733d39 | 622 | static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs) |
8e2e0047 | 623 | { |
9e733d39 SW |
624 | gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n", |
625 | (unsigned long long)rs->rs_inum, | |
626 | (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm), | |
4a993fb1 | 627 | rs->rs_rbm.offset, rs->rs_free); |
8e2e0047 BP |
628 | } |
629 | ||
0a305e49 | 630 | /** |
8e2e0047 BP |
631 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
632 | * @rs: The reservation to remove | |
633 | * | |
634 | */ | |
20095218 | 635 | static void __rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
636 | { |
637 | struct gfs2_rgrpd *rgd; | |
638 | ||
639 | if (!gfs2_rs_active(rs)) | |
640 | return; | |
641 | ||
4a993fb1 | 642 | rgd = rs->rs_rbm.rgd; |
9e733d39 | 643 | trace_gfs2_rs(rs, TRACE_RS_TREEDEL); |
4a993fb1 | 644 | rb_erase(&rs->rs_node, &rgd->rd_rstree); |
24d634e8 | 645 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 BP |
646 | |
647 | if (rs->rs_free) { | |
e579ed4f BP |
648 | struct gfs2_bitmap *bi = rbm_bi(&rs->rs_rbm); |
649 | ||
20095218 | 650 | /* return reserved blocks to the rgrp */ |
4a993fb1 SW |
651 | BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free); |
652 | rs->rs_rbm.rgd->rd_reserved -= rs->rs_free; | |
5ea5050c BP |
653 | /* The rgrp extent failure point is likely not to increase; |
654 | it will only do so if the freed blocks are somehow | |
655 | contiguous with a span of free blocks that follows. Still, | |
656 | it will force the number to be recalculated later. */ | |
657 | rgd->rd_extfail_pt += rs->rs_free; | |
8e2e0047 | 658 | rs->rs_free = 0; |
e579ed4f | 659 | clear_bit(GBF_FULL, &bi->bi_flags); |
8e2e0047 | 660 | } |
8e2e0047 BP |
661 | } |
662 | ||
663 | /** | |
664 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
665 | * @rs: The reservation to remove | |
666 | * | |
667 | */ | |
20095218 | 668 | void gfs2_rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
669 | { |
670 | struct gfs2_rgrpd *rgd; | |
671 | ||
4a993fb1 SW |
672 | rgd = rs->rs_rbm.rgd; |
673 | if (rgd) { | |
674 | spin_lock(&rgd->rd_rsspin); | |
20095218 | 675 | __rs_deltree(rs); |
4a993fb1 SW |
676 | spin_unlock(&rgd->rd_rsspin); |
677 | } | |
8e2e0047 BP |
678 | } |
679 | ||
680 | /** | |
681 | * gfs2_rs_delete - delete a multi-block reservation | |
0a305e49 | 682 | * @ip: The inode for this reservation |
af5c2697 | 683 | * @wcount: The inode's write count, or NULL |
0a305e49 BP |
684 | * |
685 | */ | |
af5c2697 | 686 | void gfs2_rs_delete(struct gfs2_inode *ip, atomic_t *wcount) |
0a305e49 BP |
687 | { |
688 | down_write(&ip->i_rw_mutex); | |
af5c2697 | 689 | if (ip->i_res && ((wcount == NULL) || (atomic_read(wcount) <= 1))) { |
20095218 | 690 | gfs2_rs_deltree(ip->i_res); |
8e2e0047 | 691 | BUG_ON(ip->i_res->rs_free); |
0a305e49 BP |
692 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); |
693 | ip->i_res = NULL; | |
694 | } | |
695 | up_write(&ip->i_rw_mutex); | |
696 | } | |
697 | ||
8e2e0047 BP |
698 | /** |
699 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
700 | * @rgd: the rgrp that needs its space back | |
701 | * | |
702 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
703 | * give them back. This leave the reservation structures in tact, but removes | |
704 | * all of their corresponding "no-fly zones". | |
705 | */ | |
706 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
707 | { | |
708 | struct rb_node *n; | |
709 | struct gfs2_blkreserv *rs; | |
710 | ||
711 | spin_lock(&rgd->rd_rsspin); | |
712 | while ((n = rb_first(&rgd->rd_rstree))) { | |
713 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
20095218 | 714 | __rs_deltree(rs); |
8e2e0047 BP |
715 | } |
716 | spin_unlock(&rgd->rd_rsspin); | |
717 | } | |
718 | ||
8339ee54 | 719 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 720 | { |
7c9ca621 | 721 | struct rb_node *n; |
b3b94faa DT |
722 | struct gfs2_rgrpd *rgd; |
723 | struct gfs2_glock *gl; | |
724 | ||
7c9ca621 BP |
725 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
726 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
727 | gl = rgd->rd_gl; |
728 | ||
7c9ca621 | 729 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
730 | |
731 | if (gl) { | |
8339ee54 | 732 | spin_lock(&gl->gl_spin); |
5c676f6d | 733 | gl->gl_object = NULL; |
8339ee54 | 734 | spin_unlock(&gl->gl_spin); |
29687a2a | 735 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
736 | gfs2_glock_put(gl); |
737 | } | |
738 | ||
8339ee54 | 739 | gfs2_free_clones(rgd); |
b3b94faa | 740 | kfree(rgd->rd_bits); |
8e2e0047 | 741 | return_all_reservations(rgd); |
6bdd9be6 | 742 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
743 | } |
744 | } | |
745 | ||
bb8d8a6f SW |
746 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
747 | { | |
d77d1b58 JP |
748 | pr_info("ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); |
749 | pr_info("ri_length = %u\n", rgd->rd_length); | |
750 | pr_info("ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
751 | pr_info("ri_data = %u\n", rgd->rd_data); | |
752 | pr_info("ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
bb8d8a6f SW |
753 | } |
754 | ||
b3b94faa DT |
755 | /** |
756 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
757 | * @rgd: The resource group descriptor | |
758 | * | |
759 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
760 | * | |
761 | * Returns: errno | |
762 | */ | |
763 | ||
764 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
765 | { | |
766 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
767 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 768 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 769 | u32 bytes_left, bytes; |
b3b94faa DT |
770 | int x; |
771 | ||
feaa7bba SW |
772 | if (!length) |
773 | return -EINVAL; | |
774 | ||
dd894be8 | 775 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
776 | if (!rgd->rd_bits) |
777 | return -ENOMEM; | |
778 | ||
bb8d8a6f | 779 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
780 | |
781 | for (x = 0; x < length; x++) { | |
782 | bi = rgd->rd_bits + x; | |
783 | ||
60a0b8f9 | 784 | bi->bi_flags = 0; |
b3b94faa DT |
785 | /* small rgrp; bitmap stored completely in header block */ |
786 | if (length == 1) { | |
787 | bytes = bytes_left; | |
788 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
789 | bi->bi_start = 0; | |
790 | bi->bi_len = bytes; | |
7e230f57 | 791 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
792 | /* header block */ |
793 | } else if (x == 0) { | |
794 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
795 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
796 | bi->bi_start = 0; | |
797 | bi->bi_len = bytes; | |
7e230f57 | 798 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
799 | /* last block */ |
800 | } else if (x + 1 == length) { | |
801 | bytes = bytes_left; | |
802 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 803 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 804 | bi->bi_len = bytes; |
7e230f57 | 805 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
806 | /* other blocks */ |
807 | } else { | |
568f4c96 SW |
808 | bytes = sdp->sd_sb.sb_bsize - |
809 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 810 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 811 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 812 | bi->bi_len = bytes; |
7e230f57 | 813 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
814 | } |
815 | ||
816 | bytes_left -= bytes; | |
817 | } | |
818 | ||
819 | if (bytes_left) { | |
820 | gfs2_consist_rgrpd(rgd); | |
821 | return -EIO; | |
822 | } | |
823 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 824 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 825 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 826 | gfs2_rindex_print(rgd); |
b3b94faa DT |
827 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
828 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
829 | } | |
830 | return -EIO; | |
831 | } | |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
7ae8fa84 RP |
836 | /** |
837 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 838 | * @sdp: the filesystem |
7ae8fa84 RP |
839 | * |
840 | */ | |
841 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
842 | { | |
843 | u64 total_data = 0; | |
844 | struct inode *inode = sdp->sd_rindex; | |
845 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 846 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
847 | int error, rgrps; |
848 | ||
7ae8fa84 RP |
849 | for (rgrps = 0;; rgrps++) { |
850 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
851 | ||
bcd7278d | 852 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 853 | break; |
4306629e | 854 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
855 | sizeof(struct gfs2_rindex)); |
856 | if (error != sizeof(struct gfs2_rindex)) | |
857 | break; | |
bb8d8a6f | 858 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 859 | } |
7ae8fa84 RP |
860 | return total_data; |
861 | } | |
862 | ||
6aad1c3d | 863 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
864 | { |
865 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
866 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
867 | ||
868 | /* Figure out where to put new node */ | |
869 | while (*newn) { | |
870 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
871 | rd_node); | |
872 | ||
873 | parent = *newn; | |
874 | if (rgd->rd_addr < cur->rd_addr) | |
875 | newn = &((*newn)->rb_left); | |
876 | else if (rgd->rd_addr > cur->rd_addr) | |
877 | newn = &((*newn)->rb_right); | |
878 | else | |
6aad1c3d | 879 | return -EEXIST; |
7c9ca621 BP |
880 | } |
881 | ||
882 | rb_link_node(&rgd->rd_node, parent, newn); | |
883 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
884 | sdp->sd_rgrps++; |
885 | return 0; | |
7c9ca621 BP |
886 | } |
887 | ||
b3b94faa | 888 | /** |
6c53267f | 889 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 890 | * @ip: Pointer to the rindex inode |
b3b94faa | 891 | * |
8339ee54 | 892 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
893 | */ |
894 | ||
4306629e | 895 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
896 | { |
897 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
7005c3e4 | 898 | const unsigned bsize = sdp->sd_sb.sb_bsize; |
6c53267f | 899 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); |
8339ee54 | 900 | struct gfs2_rindex buf; |
6c53267f RP |
901 | int error; |
902 | struct gfs2_rgrpd *rgd; | |
903 | ||
8339ee54 SW |
904 | if (pos >= i_size_read(&ip->i_inode)) |
905 | return 1; | |
906 | ||
4306629e | 907 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 908 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
909 | |
910 | if (error != sizeof(struct gfs2_rindex)) | |
911 | return (error == 0) ? 1 : error; | |
6c53267f | 912 | |
6bdd9be6 | 913 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
914 | error = -ENOMEM; |
915 | if (!rgd) | |
916 | return error; | |
917 | ||
6c53267f | 918 | rgd->rd_sbd = sdp; |
8339ee54 SW |
919 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
920 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
921 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
922 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
923 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 924 | spin_lock_init(&rgd->rd_rsspin); |
7c9ca621 | 925 | |
6c53267f RP |
926 | error = compute_bitstructs(rgd); |
927 | if (error) | |
8339ee54 | 928 | goto fail; |
6c53267f | 929 | |
bb8d8a6f | 930 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
931 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
932 | if (error) | |
8339ee54 | 933 | goto fail; |
6c53267f RP |
934 | |
935 | rgd->rd_gl->gl_object = rgd; | |
7005c3e4 SW |
936 | rgd->rd_gl->gl_vm.start = rgd->rd_addr * bsize; |
937 | rgd->rd_gl->gl_vm.end = rgd->rd_gl->gl_vm.start + (rgd->rd_length * bsize) - 1; | |
4e2f8849 | 938 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr; |
0e27c18c | 939 | rgd->rd_flags &= ~(GFS2_RDF_UPTODATE | GFS2_RDF_PREFERRED); |
7c9ca621 BP |
940 | if (rgd->rd_data > sdp->sd_max_rg_data) |
941 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 942 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 943 | error = rgd_insert(rgd); |
8339ee54 | 944 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
945 | if (!error) |
946 | return 0; | |
947 | ||
948 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 949 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
950 | |
951 | fail: | |
952 | kfree(rgd->rd_bits); | |
953 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
954 | return error; |
955 | } | |
956 | ||
0e27c18c BP |
957 | /** |
958 | * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use | |
959 | * @sdp: the GFS2 superblock | |
960 | * | |
961 | * The purpose of this function is to select a subset of the resource groups | |
962 | * and mark them as PREFERRED. We do it in such a way that each node prefers | |
963 | * to use a unique set of rgrps to minimize glock contention. | |
964 | */ | |
965 | static void set_rgrp_preferences(struct gfs2_sbd *sdp) | |
966 | { | |
967 | struct gfs2_rgrpd *rgd, *first; | |
968 | int i; | |
969 | ||
970 | /* Skip an initial number of rgrps, based on this node's journal ID. | |
971 | That should start each node out on its own set. */ | |
972 | rgd = gfs2_rgrpd_get_first(sdp); | |
973 | for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++) | |
974 | rgd = gfs2_rgrpd_get_next(rgd); | |
975 | first = rgd; | |
976 | ||
977 | do { | |
978 | rgd->rd_flags |= GFS2_RDF_PREFERRED; | |
979 | for (i = 0; i < sdp->sd_journals; i++) { | |
980 | rgd = gfs2_rgrpd_get_next(rgd); | |
959b6717 | 981 | if (!rgd || rgd == first) |
0e27c18c BP |
982 | break; |
983 | } | |
959b6717 | 984 | } while (rgd && rgd != first); |
0e27c18c BP |
985 | } |
986 | ||
6c53267f RP |
987 | /** |
988 | * gfs2_ri_update - Pull in a new resource index from the disk | |
989 | * @ip: pointer to the rindex inode | |
990 | * | |
b3b94faa DT |
991 | * Returns: 0 on successful update, error code otherwise |
992 | */ | |
993 | ||
8339ee54 | 994 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 995 | { |
feaa7bba | 996 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
997 | int error; |
998 | ||
8339ee54 | 999 | do { |
4306629e | 1000 | error = read_rindex_entry(ip); |
8339ee54 SW |
1001 | } while (error == 0); |
1002 | ||
1003 | if (error < 0) | |
1004 | return error; | |
b3b94faa | 1005 | |
0e27c18c BP |
1006 | set_rgrp_preferences(sdp); |
1007 | ||
cf45b752 | 1008 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
1009 | return 0; |
1010 | } | |
b3b94faa | 1011 | |
b3b94faa | 1012 | /** |
8339ee54 | 1013 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 1014 | * @sdp: The GFS2 superblock |
b3b94faa DT |
1015 | * |
1016 | * We grab a lock on the rindex inode to make sure that it doesn't | |
1017 | * change whilst we are performing an operation. We keep this lock | |
1018 | * for quite long periods of time compared to other locks. This | |
1019 | * doesn't matter, since it is shared and it is very, very rarely | |
1020 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
1021 | * | |
1022 | * This makes sure that we're using the latest copy of the resource index | |
1023 | * special file, which might have been updated if someone expanded the | |
1024 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
1025 | * | |
8339ee54 | 1026 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
1027 | */ |
1028 | ||
8339ee54 | 1029 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 1030 | { |
feaa7bba | 1031 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 1032 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
1033 | struct gfs2_holder ri_gh; |
1034 | int error = 0; | |
a365fbf3 | 1035 | int unlock_required = 0; |
b3b94faa DT |
1036 | |
1037 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 1038 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
1039 | if (!gfs2_glock_is_locked_by_me(gl)) { |
1040 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
1041 | if (error) | |
6aad1c3d | 1042 | return error; |
a365fbf3 SW |
1043 | unlock_required = 1; |
1044 | } | |
8339ee54 | 1045 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 1046 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
1047 | if (unlock_required) |
1048 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
1049 | } |
1050 | ||
1051 | return error; | |
1052 | } | |
1053 | ||
42d52e38 | 1054 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
1055 | { |
1056 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 1057 | u32 rg_flags; |
bb8d8a6f | 1058 | |
42d52e38 | 1059 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 1060 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
1061 | rgd->rd_flags &= GFS2_RDF_MASK; |
1062 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 1063 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 1064 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 1065 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
1066 | } |
1067 | ||
42d52e38 | 1068 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
1069 | { |
1070 | struct gfs2_rgrp *str = buf; | |
1071 | ||
09010978 | 1072 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 1073 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 1074 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 1075 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 1076 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
1077 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
1078 | } | |
1079 | ||
90306c41 BM |
1080 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
1081 | { | |
1082 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1083 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
1084 | ||
1085 | if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free || | |
1086 | rgl->rl_dinodes != str->rg_dinodes || | |
1087 | rgl->rl_igeneration != str->rg_igeneration) | |
1088 | return 0; | |
1089 | return 1; | |
1090 | } | |
1091 | ||
1092 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) | |
1093 | { | |
1094 | const struct gfs2_rgrp *str = buf; | |
1095 | ||
1096 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
1097 | rgl->rl_flags = str->rg_flags; | |
1098 | rgl->rl_free = str->rg_free; | |
1099 | rgl->rl_dinodes = str->rg_dinodes; | |
1100 | rgl->rl_igeneration = str->rg_igeneration; | |
1101 | rgl->__pad = 0UL; | |
1102 | } | |
1103 | ||
1104 | static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change) | |
1105 | { | |
1106 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1107 | u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change; | |
1108 | rgl->rl_unlinked = cpu_to_be32(unlinked); | |
1109 | } | |
1110 | ||
1111 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) | |
1112 | { | |
1113 | struct gfs2_bitmap *bi; | |
1114 | const u32 length = rgd->rd_length; | |
1115 | const u8 *buffer = NULL; | |
1116 | u32 i, goal, count = 0; | |
1117 | ||
1118 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
1119 | goal = 0; | |
1120 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1121 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
1122 | while (goal < bi->bi_len * GFS2_NBBY) { | |
1123 | goal = gfs2_bitfit(buffer, bi->bi_len, goal, | |
1124 | GFS2_BLKST_UNLINKED); | |
1125 | if (goal == BFITNOENT) | |
1126 | break; | |
1127 | count++; | |
1128 | goal++; | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | return count; | |
1133 | } | |
1134 | ||
1135 | ||
b3b94faa | 1136 | /** |
90306c41 BM |
1137 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
1138 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
1139 | * |
1140 | * Read in all of a Resource Group's header and bitmap blocks. | |
1141 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
1142 | * | |
1143 | * Returns: errno | |
1144 | */ | |
1145 | ||
c2b0b30e | 1146 | static int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
1147 | { |
1148 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1149 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 1150 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
1151 | struct gfs2_bitmap *bi; |
1152 | unsigned int x, y; | |
1153 | int error; | |
1154 | ||
90306c41 BM |
1155 | if (rgd->rd_bits[0].bi_bh != NULL) |
1156 | return 0; | |
1157 | ||
b3b94faa DT |
1158 | for (x = 0; x < length; x++) { |
1159 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 1160 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
1161 | if (error) |
1162 | goto fail; | |
1163 | } | |
1164 | ||
1165 | for (y = length; y--;) { | |
1166 | bi = rgd->rd_bits + y; | |
7276b3b0 | 1167 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
1168 | if (error) |
1169 | goto fail; | |
feaa7bba | 1170 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
1171 | GFS2_METATYPE_RG)) { |
1172 | error = -EIO; | |
1173 | goto fail; | |
1174 | } | |
1175 | } | |
1176 | ||
cf45b752 | 1177 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
1178 | for (x = 0; x < length; x++) |
1179 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 1180 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 1181 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 1182 | rgd->rd_free_clone = rgd->rd_free; |
5ea5050c BP |
1183 | /* max out the rgrp allocation failure point */ |
1184 | rgd->rd_extfail_pt = rgd->rd_free; | |
b3b94faa | 1185 | } |
951b4bd5 | 1186 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
90306c41 BM |
1187 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); |
1188 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
1189 | rgd->rd_bits[0].bi_bh->b_data); | |
1190 | } | |
1191 | else if (sdp->sd_args.ar_rgrplvb) { | |
1192 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
1193 | gfs2_consist_rgrpd(rgd); | |
1194 | error = -EIO; | |
1195 | goto fail; | |
1196 | } | |
1197 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1198 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1199 | } | |
b3b94faa DT |
1200 | return 0; |
1201 | ||
feaa7bba | 1202 | fail: |
b3b94faa DT |
1203 | while (x--) { |
1204 | bi = rgd->rd_bits + x; | |
1205 | brelse(bi->bi_bh); | |
1206 | bi->bi_bh = NULL; | |
1207 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
1208 | } | |
b3b94faa DT |
1209 | |
1210 | return error; | |
1211 | } | |
1212 | ||
c2b0b30e | 1213 | static int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
90306c41 BM |
1214 | { |
1215 | u32 rl_flags; | |
1216 | ||
1217 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1218 | return 0; | |
1219 | ||
951b4bd5 | 1220 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) |
90306c41 BM |
1221 | return gfs2_rgrp_bh_get(rgd); |
1222 | ||
1223 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1224 | rl_flags &= ~GFS2_RDF_MASK; | |
1225 | rgd->rd_flags &= GFS2_RDF_MASK; | |
1226 | rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); | |
1227 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1228 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1229 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
1230 | rgd->rd_free_clone = rgd->rd_free; | |
1231 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); | |
1232 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1237 | { | |
1238 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1239 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1240 | ||
1241 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1242 | return 0; | |
8b127d04 | 1243 | return gfs2_rgrp_bh_get(rgd); |
90306c41 BM |
1244 | } |
1245 | ||
b3b94faa | 1246 | /** |
39b0f1e9 BP |
1247 | * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
1248 | * @rgd: The resource group | |
b3b94faa DT |
1249 | * |
1250 | */ | |
1251 | ||
39b0f1e9 | 1252 | void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd) |
b3b94faa | 1253 | { |
bb8d8a6f | 1254 | int x, length = rgd->rd_length; |
b3b94faa | 1255 | |
b3b94faa DT |
1256 | for (x = 0; x < length; x++) { |
1257 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1258 | if (bi->bi_bh) { |
1259 | brelse(bi->bi_bh); | |
1260 | bi->bi_bh = NULL; | |
1261 | } | |
b3b94faa DT |
1262 | } |
1263 | ||
b3b94faa DT |
1264 | } |
1265 | ||
39b0f1e9 BP |
1266 | /** |
1267 | * gfs2_rgrp_go_unlock - Unlock a rgrp glock | |
1268 | * @gh: The glock holder for the resource group | |
1269 | * | |
1270 | */ | |
1271 | ||
1272 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) | |
1273 | { | |
1274 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1275 | int demote_requested = test_bit(GLF_DEMOTE, &gh->gh_gl->gl_flags) | | |
1276 | test_bit(GLF_PENDING_DEMOTE, &gh->gh_gl->gl_flags); | |
1277 | ||
1278 | if (rgd && demote_requested) | |
1279 | gfs2_rgrp_brelse(rgd); | |
1280 | } | |
1281 | ||
66fc061b | 1282 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1283 | struct buffer_head *bh, |
66fc061b | 1284 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1285 | { |
1286 | struct super_block *sb = sdp->sd_vfs; | |
f15ab561 | 1287 | u64 blk; |
64d576ba | 1288 | sector_t start = 0; |
b2c87cae | 1289 | sector_t nr_blks = 0; |
f15ab561 SW |
1290 | int rv; |
1291 | unsigned int x; | |
66fc061b SW |
1292 | u32 trimmed = 0; |
1293 | u8 diff; | |
f15ab561 SW |
1294 | |
1295 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
1296 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1297 | clone += bi->bi_offset; | |
1298 | clone += x; | |
1299 | if (bh) { | |
1300 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1301 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1302 | } else { | |
1303 | diff = ~(*clone | (*clone >> 1)); | |
1304 | } | |
f15ab561 SW |
1305 | diff &= 0x55; |
1306 | if (diff == 0) | |
1307 | continue; | |
1308 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
f15ab561 SW |
1309 | while(diff) { |
1310 | if (diff & 1) { | |
b2c87cae | 1311 | if (nr_blks == 0) |
f15ab561 | 1312 | goto start_new_extent; |
b2c87cae BP |
1313 | if ((start + nr_blks) != blk) { |
1314 | if (nr_blks >= minlen) { | |
1315 | rv = sb_issue_discard(sb, | |
1316 | start, nr_blks, | |
66fc061b SW |
1317 | GFP_NOFS, 0); |
1318 | if (rv) | |
1319 | goto fail; | |
b2c87cae | 1320 | trimmed += nr_blks; |
66fc061b | 1321 | } |
b2c87cae | 1322 | nr_blks = 0; |
f15ab561 SW |
1323 | start_new_extent: |
1324 | start = blk; | |
1325 | } | |
b2c87cae | 1326 | nr_blks++; |
f15ab561 SW |
1327 | } |
1328 | diff >>= 2; | |
b2c87cae | 1329 | blk++; |
f15ab561 SW |
1330 | } |
1331 | } | |
b2c87cae BP |
1332 | if (nr_blks >= minlen) { |
1333 | rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0); | |
f15ab561 SW |
1334 | if (rv) |
1335 | goto fail; | |
b2c87cae | 1336 | trimmed += nr_blks; |
f15ab561 | 1337 | } |
66fc061b SW |
1338 | if (ptrimmed) |
1339 | *ptrimmed = trimmed; | |
1340 | return 0; | |
1341 | ||
f15ab561 | 1342 | fail: |
66fc061b SW |
1343 | if (sdp->sd_args.ar_discard) |
1344 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 1345 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1346 | return -EIO; |
1347 | } | |
1348 | ||
1349 | /** | |
1350 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1351 | * @filp: Any file on the filesystem | |
1352 | * @argp: Pointer to the arguments (also used to pass result) | |
1353 | * | |
1354 | * Returns: 0 on success, otherwise error code | |
1355 | */ | |
1356 | ||
1357 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1358 | { | |
496ad9aa | 1359 | struct inode *inode = file_inode(filp); |
66fc061b SW |
1360 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1361 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1362 | struct buffer_head *bh; | |
1363 | struct gfs2_rgrpd *rgd; | |
1364 | struct gfs2_rgrpd *rgd_end; | |
1365 | struct gfs2_holder gh; | |
1366 | struct fstrim_range r; | |
1367 | int ret = 0; | |
1368 | u64 amt; | |
1369 | u64 trimmed = 0; | |
076f0faa | 1370 | u64 start, end, minlen; |
66fc061b | 1371 | unsigned int x; |
076f0faa | 1372 | unsigned bs_shift = sdp->sd_sb.sb_bsize_shift; |
66fc061b SW |
1373 | |
1374 | if (!capable(CAP_SYS_ADMIN)) | |
1375 | return -EPERM; | |
1376 | ||
1377 | if (!blk_queue_discard(q)) | |
1378 | return -EOPNOTSUPP; | |
1379 | ||
3a238ade | 1380 | if (copy_from_user(&r, argp, sizeof(r))) |
66fc061b SW |
1381 | return -EFAULT; |
1382 | ||
5e2f7d61 BP |
1383 | ret = gfs2_rindex_update(sdp); |
1384 | if (ret) | |
1385 | return ret; | |
1386 | ||
076f0faa LC |
1387 | start = r.start >> bs_shift; |
1388 | end = start + (r.len >> bs_shift); | |
1389 | minlen = max_t(u64, r.minlen, | |
1390 | q->limits.discard_granularity) >> bs_shift; | |
1391 | ||
6a98c333 AD |
1392 | if (end <= start || minlen > sdp->sd_max_rg_data) |
1393 | return -EINVAL; | |
1394 | ||
076f0faa | 1395 | rgd = gfs2_blk2rgrpd(sdp, start, 0); |
6a98c333 | 1396 | rgd_end = gfs2_blk2rgrpd(sdp, end, 0); |
076f0faa | 1397 | |
6a98c333 AD |
1398 | if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end)) |
1399 | && (start > rgd_end->rd_data0 + rgd_end->rd_data)) | |
1400 | return -EINVAL; /* start is beyond the end of the fs */ | |
66fc061b SW |
1401 | |
1402 | while (1) { | |
1403 | ||
1404 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1405 | if (ret) | |
1406 | goto out; | |
1407 | ||
1408 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1409 | /* Trim each bitmap in the rgrp */ | |
1410 | for (x = 0; x < rgd->rd_length; x++) { | |
1411 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
076f0faa LC |
1412 | ret = gfs2_rgrp_send_discards(sdp, |
1413 | rgd->rd_data0, NULL, bi, minlen, | |
1414 | &amt); | |
66fc061b SW |
1415 | if (ret) { |
1416 | gfs2_glock_dq_uninit(&gh); | |
1417 | goto out; | |
1418 | } | |
1419 | trimmed += amt; | |
1420 | } | |
1421 | ||
1422 | /* Mark rgrp as having been trimmed */ | |
1423 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1424 | if (ret == 0) { | |
1425 | bh = rgd->rd_bits[0].bi_bh; | |
1426 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
350a9b0a | 1427 | gfs2_trans_add_meta(rgd->rd_gl, bh); |
66fc061b | 1428 | gfs2_rgrp_out(rgd, bh->b_data); |
90306c41 | 1429 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data); |
66fc061b SW |
1430 | gfs2_trans_end(sdp); |
1431 | } | |
1432 | } | |
1433 | gfs2_glock_dq_uninit(&gh); | |
1434 | ||
1435 | if (rgd == rgd_end) | |
1436 | break; | |
1437 | ||
1438 | rgd = gfs2_rgrpd_get_next(rgd); | |
1439 | } | |
1440 | ||
1441 | out: | |
6a98c333 | 1442 | r.len = trimmed << bs_shift; |
3a238ade | 1443 | if (copy_to_user(argp, &r, sizeof(r))) |
66fc061b SW |
1444 | return -EFAULT; |
1445 | ||
1446 | return ret; | |
f15ab561 SW |
1447 | } |
1448 | ||
8e2e0047 BP |
1449 | /** |
1450 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
8e2e0047 | 1451 | * @ip: the inode structure |
8e2e0047 | 1452 | * |
8e2e0047 | 1453 | */ |
ff7f4cb4 | 1454 | static void rs_insert(struct gfs2_inode *ip) |
8e2e0047 BP |
1455 | { |
1456 | struct rb_node **newn, *parent = NULL; | |
1457 | int rc; | |
1458 | struct gfs2_blkreserv *rs = ip->i_res; | |
4a993fb1 | 1459 | struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd; |
ff7f4cb4 | 1460 | u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 | 1461 | |
8e2e0047 | 1462 | BUG_ON(gfs2_rs_active(rs)); |
c743ffd0 | 1463 | |
ff7f4cb4 SW |
1464 | spin_lock(&rgd->rd_rsspin); |
1465 | newn = &rgd->rd_rstree.rb_node; | |
8e2e0047 BP |
1466 | while (*newn) { |
1467 | struct gfs2_blkreserv *cur = | |
1468 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1469 | ||
1470 | parent = *newn; | |
ff7f4cb4 | 1471 | rc = rs_cmp(fsblock, rs->rs_free, cur); |
8e2e0047 BP |
1472 | if (rc > 0) |
1473 | newn = &((*newn)->rb_right); | |
1474 | else if (rc < 0) | |
1475 | newn = &((*newn)->rb_left); | |
1476 | else { | |
1477 | spin_unlock(&rgd->rd_rsspin); | |
ff7f4cb4 SW |
1478 | WARN_ON(1); |
1479 | return; | |
8e2e0047 BP |
1480 | } |
1481 | } | |
1482 | ||
8e2e0047 BP |
1483 | rb_link_node(&rs->rs_node, parent, newn); |
1484 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1485 | ||
8e2e0047 | 1486 | /* Do our rgrp accounting for the reservation */ |
ff7f4cb4 | 1487 | rgd->rd_reserved += rs->rs_free; /* blocks reserved */ |
8e2e0047 | 1488 | spin_unlock(&rgd->rd_rsspin); |
9e733d39 | 1489 | trace_gfs2_rs(rs, TRACE_RS_INSERT); |
8e2e0047 BP |
1490 | } |
1491 | ||
1492 | /** | |
ff7f4cb4 | 1493 | * rg_mblk_search - find a group of multiple free blocks to form a reservation |
8e2e0047 | 1494 | * @rgd: the resource group descriptor |
8e2e0047 | 1495 | * @ip: pointer to the inode for which we're reserving blocks |
7b9cff46 | 1496 | * @ap: the allocation parameters |
8e2e0047 | 1497 | * |
8e2e0047 BP |
1498 | */ |
1499 | ||
ff7f4cb4 | 1500 | static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
7b9cff46 | 1501 | const struct gfs2_alloc_parms *ap) |
8e2e0047 | 1502 | { |
ff7f4cb4 SW |
1503 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1504 | u64 goal; | |
1505 | struct gfs2_blkreserv *rs = ip->i_res; | |
1506 | u32 extlen; | |
1507 | u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved; | |
1508 | int ret; | |
af21ca8e | 1509 | struct inode *inode = &ip->i_inode; |
8e2e0047 | 1510 | |
af21ca8e BP |
1511 | if (S_ISDIR(inode->i_mode)) |
1512 | extlen = 1; | |
1513 | else { | |
7b9cff46 | 1514 | extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target); |
af21ca8e BP |
1515 | extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks); |
1516 | } | |
ff7f4cb4 | 1517 | if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen)) |
c743ffd0 SW |
1518 | return; |
1519 | ||
8e2e0047 BP |
1520 | /* Find bitmap block that contains bits for goal block */ |
1521 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
ff7f4cb4 | 1522 | goal = ip->i_goal; |
8e2e0047 | 1523 | else |
ff7f4cb4 | 1524 | goal = rgd->rd_last_alloc + rgd->rd_data0; |
8e2e0047 | 1525 | |
ff7f4cb4 SW |
1526 | if (WARN_ON(gfs2_rbm_from_block(&rbm, goal))) |
1527 | return; | |
8e2e0047 | 1528 | |
5ce13431 | 1529 | ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true, ap); |
ff7f4cb4 SW |
1530 | if (ret == 0) { |
1531 | rs->rs_rbm = rbm; | |
1532 | rs->rs_free = extlen; | |
1533 | rs->rs_inum = ip->i_no_addr; | |
1534 | rs_insert(ip); | |
13d2eb01 BP |
1535 | } else { |
1536 | if (goal == rgd->rd_last_alloc + rgd->rd_data0) | |
1537 | rgd->rd_last_alloc = 0; | |
8e2e0047 | 1538 | } |
b3e47ca0 BP |
1539 | } |
1540 | ||
5b924ae2 SW |
1541 | /** |
1542 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1543 | * @rgd: The resource group | |
1544 | * @block: The starting block | |
ff7f4cb4 | 1545 | * @length: The required length |
5b924ae2 SW |
1546 | * @ip: Ignore any reservations for this inode |
1547 | * | |
1548 | * If the block does not appear in any reservation, then return the | |
1549 | * block number unchanged. If it does appear in the reservation, then | |
1550 | * keep looking through the tree of reservations in order to find the | |
1551 | * first block number which is not reserved. | |
1552 | */ | |
1553 | ||
1554 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
ff7f4cb4 | 1555 | u32 length, |
5b924ae2 SW |
1556 | const struct gfs2_inode *ip) |
1557 | { | |
1558 | struct gfs2_blkreserv *rs; | |
1559 | struct rb_node *n; | |
1560 | int rc; | |
1561 | ||
1562 | spin_lock(&rgd->rd_rsspin); | |
ff7f4cb4 | 1563 | n = rgd->rd_rstree.rb_node; |
5b924ae2 SW |
1564 | while (n) { |
1565 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
ff7f4cb4 | 1566 | rc = rs_cmp(block, length, rs); |
5b924ae2 SW |
1567 | if (rc < 0) |
1568 | n = n->rb_left; | |
1569 | else if (rc > 0) | |
1570 | n = n->rb_right; | |
1571 | else | |
1572 | break; | |
1573 | } | |
1574 | ||
1575 | if (n) { | |
ff7f4cb4 | 1576 | while ((rs_cmp(block, length, rs) == 0) && (ip->i_res != rs)) { |
5b924ae2 | 1577 | block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free; |
ff7f4cb4 | 1578 | n = n->rb_right; |
5b924ae2 SW |
1579 | if (n == NULL) |
1580 | break; | |
1581 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | spin_unlock(&rgd->rd_rsspin); | |
1586 | return block; | |
1587 | } | |
1588 | ||
5b924ae2 SW |
1589 | /** |
1590 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1591 | * @rbm: The current position in the resource group | |
ff7f4cb4 SW |
1592 | * @ip: The inode for which we are searching for blocks |
1593 | * @minext: The minimum extent length | |
5ce13431 | 1594 | * @maxext: A pointer to the maximum extent structure |
5b924ae2 SW |
1595 | * |
1596 | * This checks the current position in the rgrp to see whether there is | |
1597 | * a reservation covering this block. If not then this function is a | |
1598 | * no-op. If there is, then the position is moved to the end of the | |
1599 | * contiguous reservation(s) so that we are pointing at the first | |
1600 | * non-reserved block. | |
1601 | * | |
1602 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1603 | */ | |
1604 | ||
1605 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
ff7f4cb4 | 1606 | const struct gfs2_inode *ip, |
5ce13431 BP |
1607 | u32 minext, |
1608 | struct gfs2_extent *maxext) | |
5b924ae2 SW |
1609 | { |
1610 | u64 block = gfs2_rbm_to_block(rbm); | |
ff7f4cb4 | 1611 | u32 extlen = 1; |
5b924ae2 SW |
1612 | u64 nblock; |
1613 | int ret; | |
1614 | ||
ff7f4cb4 SW |
1615 | /* |
1616 | * If we have a minimum extent length, then skip over any extent | |
1617 | * which is less than the min extent length in size. | |
1618 | */ | |
1619 | if (minext) { | |
1620 | extlen = gfs2_free_extlen(rbm, minext); | |
5ce13431 | 1621 | if (extlen <= maxext->len) |
ff7f4cb4 SW |
1622 | goto fail; |
1623 | } | |
1624 | ||
1625 | /* | |
1626 | * Check the extent which has been found against the reservations | |
1627 | * and skip if parts of it are already reserved | |
1628 | */ | |
1629 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip); | |
5ce13431 BP |
1630 | if (nblock == block) { |
1631 | if (!minext || extlen >= minext) | |
1632 | return 0; | |
1633 | ||
1634 | if (extlen > maxext->len) { | |
1635 | maxext->len = extlen; | |
1636 | maxext->rbm = *rbm; | |
1637 | } | |
ff7f4cb4 | 1638 | fail: |
5ce13431 BP |
1639 | nblock = block + extlen; |
1640 | } | |
5b924ae2 SW |
1641 | ret = gfs2_rbm_from_block(rbm, nblock); |
1642 | if (ret < 0) | |
1643 | return ret; | |
1644 | return 1; | |
1645 | } | |
1646 | ||
1647 | /** | |
1648 | * gfs2_rbm_find - Look for blocks of a particular state | |
1649 | * @rbm: Value/result starting position and final position | |
1650 | * @state: The state which we want to find | |
5ce13431 BP |
1651 | * @minext: Pointer to the requested extent length (NULL for a single block) |
1652 | * This is updated to be the actual reservation size. | |
5b924ae2 SW |
1653 | * @ip: If set, check for reservations |
1654 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping | |
1655 | * around until we've reached the starting point. | |
5ce13431 | 1656 | * @ap: the allocation parameters |
5b924ae2 SW |
1657 | * |
1658 | * Side effects: | |
1659 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1660 | * has no free blocks in it. | |
5ea5050c BP |
1661 | * - If looking for free blocks, we set rd_extfail_pt on each rgrp which |
1662 | * has come up short on a free block search. | |
5b924ae2 SW |
1663 | * |
1664 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1665 | */ | |
1666 | ||
5ce13431 BP |
1667 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
1668 | const struct gfs2_inode *ip, bool nowrap, | |
1669 | const struct gfs2_alloc_parms *ap) | |
5b924ae2 SW |
1670 | { |
1671 | struct buffer_head *bh; | |
e579ed4f | 1672 | int initial_bii; |
5b924ae2 | 1673 | u32 initial_offset; |
5ea5050c BP |
1674 | int first_bii = rbm->bii; |
1675 | u32 first_offset = rbm->offset; | |
5b924ae2 SW |
1676 | u32 offset; |
1677 | u8 *buffer; | |
5b924ae2 SW |
1678 | int n = 0; |
1679 | int iters = rbm->rgd->rd_length; | |
1680 | int ret; | |
e579ed4f | 1681 | struct gfs2_bitmap *bi; |
5ce13431 | 1682 | struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, }; |
5b924ae2 SW |
1683 | |
1684 | /* If we are not starting at the beginning of a bitmap, then we | |
1685 | * need to add one to the bitmap count to ensure that we search | |
1686 | * the starting bitmap twice. | |
1687 | */ | |
1688 | if (rbm->offset != 0) | |
1689 | iters++; | |
1690 | ||
1691 | while(1) { | |
e579ed4f BP |
1692 | bi = rbm_bi(rbm); |
1693 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
5b924ae2 SW |
1694 | (state == GFS2_BLKST_FREE)) |
1695 | goto next_bitmap; | |
1696 | ||
e579ed4f BP |
1697 | bh = bi->bi_bh; |
1698 | buffer = bh->b_data + bi->bi_offset; | |
5b924ae2 | 1699 | WARN_ON(!buffer_uptodate(bh)); |
e579ed4f BP |
1700 | if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
1701 | buffer = bi->bi_clone + bi->bi_offset; | |
5b924ae2 | 1702 | initial_offset = rbm->offset; |
e579ed4f | 1703 | offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state); |
5b924ae2 SW |
1704 | if (offset == BFITNOENT) |
1705 | goto bitmap_full; | |
1706 | rbm->offset = offset; | |
1707 | if (ip == NULL) | |
1708 | return 0; | |
1709 | ||
e579ed4f | 1710 | initial_bii = rbm->bii; |
5ce13431 BP |
1711 | ret = gfs2_reservation_check_and_update(rbm, ip, |
1712 | minext ? *minext : 0, | |
1713 | &maxext); | |
5b924ae2 SW |
1714 | if (ret == 0) |
1715 | return 0; | |
1716 | if (ret > 0) { | |
e579ed4f | 1717 | n += (rbm->bii - initial_bii); |
8d8b752a | 1718 | goto next_iter; |
5b924ae2 | 1719 | } |
5d50d532 | 1720 | if (ret == -E2BIG) { |
e579ed4f | 1721 | rbm->bii = 0; |
5d50d532 | 1722 | rbm->offset = 0; |
e579ed4f | 1723 | n += (rbm->bii - initial_bii); |
5d50d532 SW |
1724 | goto res_covered_end_of_rgrp; |
1725 | } | |
5b924ae2 SW |
1726 | return ret; |
1727 | ||
1728 | bitmap_full: /* Mark bitmap as full and fall through */ | |
a3e32136 | 1729 | if ((state == GFS2_BLKST_FREE) && initial_offset == 0) |
e579ed4f | 1730 | set_bit(GBF_FULL, &bi->bi_flags); |
5b924ae2 SW |
1731 | |
1732 | next_bitmap: /* Find next bitmap in the rgrp */ | |
1733 | rbm->offset = 0; | |
e579ed4f BP |
1734 | rbm->bii++; |
1735 | if (rbm->bii == rbm->rgd->rd_length) | |
1736 | rbm->bii = 0; | |
5d50d532 | 1737 | res_covered_end_of_rgrp: |
e579ed4f | 1738 | if ((rbm->bii == 0) && nowrap) |
5b924ae2 SW |
1739 | break; |
1740 | n++; | |
8d8b752a | 1741 | next_iter: |
5b924ae2 SW |
1742 | if (n >= iters) |
1743 | break; | |
1744 | } | |
1745 | ||
5ce13431 BP |
1746 | if (minext == NULL || state != GFS2_BLKST_FREE) |
1747 | return -ENOSPC; | |
1748 | ||
5ea5050c BP |
1749 | /* If the extent was too small, and it's smaller than the smallest |
1750 | to have failed before, remember for future reference that it's | |
1751 | useless to search this rgrp again for this amount or more. */ | |
1752 | if ((first_offset == 0) && (first_bii == 0) && | |
1753 | (*minext < rbm->rgd->rd_extfail_pt)) | |
1754 | rbm->rgd->rd_extfail_pt = *minext; | |
1755 | ||
5ce13431 BP |
1756 | /* If the maximum extent we found is big enough to fulfill the |
1757 | minimum requirements, use it anyway. */ | |
1758 | if (maxext.len) { | |
1759 | *rbm = maxext.rbm; | |
1760 | *minext = maxext.len; | |
1761 | return 0; | |
1762 | } | |
1763 | ||
5b924ae2 SW |
1764 | return -ENOSPC; |
1765 | } | |
1766 | ||
c8cdf479 SW |
1767 | /** |
1768 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1769 | * @rgd: The rgrp | |
886b1416 BP |
1770 | * @last_unlinked: block address of the last dinode we unlinked |
1771 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1772 | * |
1a0eae88 BP |
1773 | * Returns: 0 if no error |
1774 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1775 | */ |
1776 | ||
044b9414 | 1777 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1778 | { |
5b924ae2 | 1779 | u64 block; |
5f3eae75 | 1780 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1781 | struct gfs2_glock *gl; |
1782 | struct gfs2_inode *ip; | |
1783 | int error; | |
1784 | int found = 0; | |
e579ed4f | 1785 | struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 }; |
c8cdf479 | 1786 | |
5b924ae2 | 1787 | while (1) { |
5f3eae75 | 1788 | down_write(&sdp->sd_log_flush_lock); |
5ce13431 BP |
1789 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL, |
1790 | true, NULL); | |
5f3eae75 | 1791 | up_write(&sdp->sd_log_flush_lock); |
5b924ae2 SW |
1792 | if (error == -ENOSPC) |
1793 | break; | |
1794 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1795 | break; |
b3e47ca0 | 1796 | |
5b924ae2 SW |
1797 | block = gfs2_rbm_to_block(&rbm); |
1798 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1799 | break; | |
1800 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1801 | continue; |
5b924ae2 | 1802 | if (block == skip) |
1e19a195 | 1803 | continue; |
5b924ae2 | 1804 | *last_unlinked = block; |
044b9414 | 1805 | |
5b924ae2 | 1806 | error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl); |
044b9414 SW |
1807 | if (error) |
1808 | continue; | |
1809 | ||
1810 | /* If the inode is already in cache, we can ignore it here | |
1811 | * because the existing inode disposal code will deal with | |
1812 | * it when all refs have gone away. Accessing gl_object like | |
1813 | * this is not safe in general. Here it is ok because we do | |
1814 | * not dereference the pointer, and we only need an approx | |
1815 | * answer to whether it is NULL or not. | |
1816 | */ | |
1817 | ip = gl->gl_object; | |
1818 | ||
1819 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1820 | gfs2_glock_put(gl); | |
1821 | else | |
1822 | found++; | |
1823 | ||
1824 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1825 | if (found > NR_CPUS) |
044b9414 | 1826 | return; |
c8cdf479 SW |
1827 | } |
1828 | ||
1829 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1830 | return; |
c8cdf479 SW |
1831 | } |
1832 | ||
bcd97c06 SW |
1833 | /** |
1834 | * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested | |
1835 | * @rgd: The rgrp in question | |
1836 | * @loops: An indication of how picky we can be (0=very, 1=less so) | |
1837 | * | |
1838 | * This function uses the recently added glock statistics in order to | |
1839 | * figure out whether a parciular resource group is suffering from | |
1840 | * contention from multiple nodes. This is done purely on the basis | |
1841 | * of timings, since this is the only data we have to work with and | |
1842 | * our aim here is to reject a resource group which is highly contended | |
1843 | * but (very important) not to do this too often in order to ensure that | |
1844 | * we do not land up introducing fragmentation by changing resource | |
1845 | * groups when not actually required. | |
1846 | * | |
1847 | * The calculation is fairly simple, we want to know whether the SRTTB | |
1848 | * (i.e. smoothed round trip time for blocking operations) to acquire | |
1849 | * the lock for this rgrp's glock is significantly greater than the | |
1850 | * time taken for resource groups on average. We introduce a margin in | |
1851 | * the form of the variable @var which is computed as the sum of the two | |
1852 | * respective variences, and multiplied by a factor depending on @loops | |
1853 | * and whether we have a lot of data to base the decision on. This is | |
1854 | * then tested against the square difference of the means in order to | |
1855 | * decide whether the result is statistically significant or not. | |
1856 | * | |
1857 | * Returns: A boolean verdict on the congestion status | |
1858 | */ | |
1859 | ||
1860 | static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops) | |
1861 | { | |
1862 | const struct gfs2_glock *gl = rgd->rd_gl; | |
1863 | const struct gfs2_sbd *sdp = gl->gl_sbd; | |
1864 | struct gfs2_lkstats *st; | |
1865 | s64 r_dcount, l_dcount; | |
f4a3ae93 | 1866 | s64 l_srttb, a_srttb = 0; |
bcd97c06 SW |
1867 | s64 srttb_diff; |
1868 | s64 sqr_diff; | |
1869 | s64 var; | |
0166b197 | 1870 | int cpu, nonzero = 0; |
bcd97c06 SW |
1871 | |
1872 | preempt_disable(); | |
f4a3ae93 BP |
1873 | for_each_present_cpu(cpu) { |
1874 | st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP]; | |
0166b197 BP |
1875 | if (st->stats[GFS2_LKS_SRTTB]) { |
1876 | a_srttb += st->stats[GFS2_LKS_SRTTB]; | |
1877 | nonzero++; | |
1878 | } | |
f4a3ae93 | 1879 | } |
bcd97c06 | 1880 | st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP]; |
0166b197 BP |
1881 | if (nonzero) |
1882 | do_div(a_srttb, nonzero); | |
bcd97c06 SW |
1883 | r_dcount = st->stats[GFS2_LKS_DCOUNT]; |
1884 | var = st->stats[GFS2_LKS_SRTTVARB] + | |
1885 | gl->gl_stats.stats[GFS2_LKS_SRTTVARB]; | |
1886 | preempt_enable(); | |
1887 | ||
1888 | l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB]; | |
1889 | l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT]; | |
1890 | ||
f4a3ae93 | 1891 | if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0)) |
bcd97c06 SW |
1892 | return false; |
1893 | ||
f4a3ae93 | 1894 | srttb_diff = a_srttb - l_srttb; |
bcd97c06 SW |
1895 | sqr_diff = srttb_diff * srttb_diff; |
1896 | ||
1897 | var *= 2; | |
1898 | if (l_dcount < 8 || r_dcount < 8) | |
1899 | var *= 2; | |
1900 | if (loops == 1) | |
1901 | var *= 2; | |
1902 | ||
1903 | return ((srttb_diff < 0) && (sqr_diff > var)); | |
1904 | } | |
1905 | ||
1906 | /** | |
1907 | * gfs2_rgrp_used_recently | |
1908 | * @rs: The block reservation with the rgrp to test | |
1909 | * @msecs: The time limit in milliseconds | |
1910 | * | |
1911 | * Returns: True if the rgrp glock has been used within the time limit | |
1912 | */ | |
1913 | static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs, | |
1914 | u64 msecs) | |
1915 | { | |
1916 | u64 tdiff; | |
1917 | ||
1918 | tdiff = ktime_to_ns(ktime_sub(ktime_get_real(), | |
1919 | rs->rs_rbm.rgd->rd_gl->gl_dstamp)); | |
1920 | ||
1921 | return tdiff > (msecs * 1000 * 1000); | |
1922 | } | |
1923 | ||
9dbe9610 SW |
1924 | static u32 gfs2_orlov_skip(const struct gfs2_inode *ip) |
1925 | { | |
1926 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1927 | u32 skip; | |
1928 | ||
1929 | get_random_bytes(&skip, sizeof(skip)); | |
1930 | return skip % sdp->sd_rgrps; | |
1931 | } | |
1932 | ||
c743ffd0 SW |
1933 | static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin) |
1934 | { | |
1935 | struct gfs2_rgrpd *rgd = *pos; | |
aa8920c9 | 1936 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
c743ffd0 SW |
1937 | |
1938 | rgd = gfs2_rgrpd_get_next(rgd); | |
1939 | if (rgd == NULL) | |
aa8920c9 | 1940 | rgd = gfs2_rgrpd_get_first(sdp); |
c743ffd0 SW |
1941 | *pos = rgd; |
1942 | if (rgd != begin) /* If we didn't wrap */ | |
1943 | return true; | |
1944 | return false; | |
1945 | } | |
1946 | ||
0e27c18c BP |
1947 | /** |
1948 | * fast_to_acquire - determine if a resource group will be fast to acquire | |
1949 | * | |
1950 | * If this is one of our preferred rgrps, it should be quicker to acquire, | |
1951 | * because we tried to set ourselves up as dlm lock master. | |
1952 | */ | |
1953 | static inline int fast_to_acquire(struct gfs2_rgrpd *rgd) | |
1954 | { | |
1955 | struct gfs2_glock *gl = rgd->rd_gl; | |
1956 | ||
1957 | if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) && | |
1958 | !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
1959 | !test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
1960 | return 1; | |
1961 | if (rgd->rd_flags & GFS2_RDF_PREFERRED) | |
1962 | return 1; | |
1963 | return 0; | |
1964 | } | |
1965 | ||
b3b94faa | 1966 | /** |
666d1d8a | 1967 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1968 | * @ip: the inode to reserve space for |
7b9cff46 | 1969 | * @ap: the allocation parameters |
b3b94faa | 1970 | * |
25435e5e AD |
1971 | * We try our best to find an rgrp that has at least ap->target blocks |
1972 | * available. After a couple of passes (loops == 2), the prospects of finding | |
1973 | * such an rgrp diminish. At this stage, we return the first rgrp that has | |
1974 | * atleast ap->min_target blocks available. Either way, we set ap->allowed to | |
1975 | * the number of blocks available in the chosen rgrp. | |
1976 | * | |
1977 | * Returns: 0 on success, | |
1978 | * -ENOMEM if a suitable rgrp can't be found | |
1979 | * errno otherwise | |
b3b94faa DT |
1980 | */ |
1981 | ||
25435e5e | 1982 | int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap) |
b3b94faa | 1983 | { |
feaa7bba | 1984 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 1985 | struct gfs2_rgrpd *begin = NULL; |
564e12b1 | 1986 | struct gfs2_blkreserv *rs = ip->i_res; |
bcd97c06 | 1987 | int error = 0, rg_locked, flags = 0; |
666d1d8a | 1988 | u64 last_unlinked = NO_BLOCK; |
7c9ca621 | 1989 | int loops = 0; |
9dbe9610 | 1990 | u32 skip = 0; |
b3b94faa | 1991 | |
90306c41 BM |
1992 | if (sdp->sd_args.ar_rgrplvb) |
1993 | flags |= GL_SKIP; | |
7b9cff46 | 1994 | if (gfs2_assert_warn(sdp, ap->target)) |
c743ffd0 | 1995 | return -EINVAL; |
8e2e0047 | 1996 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1997 | begin = rs->rs_rbm.rgd; |
8e2e0047 | 1998 | } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) { |
4a993fb1 | 1999 | rs->rs_rbm.rgd = begin = ip->i_rgd; |
8e2e0047 | 2000 | } else { |
00a158be | 2001 | check_and_update_goal(ip); |
4a993fb1 | 2002 | rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 2003 | } |
7b9cff46 | 2004 | if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV)) |
9dbe9610 | 2005 | skip = gfs2_orlov_skip(ip); |
4a993fb1 | 2006 | if (rs->rs_rbm.rgd == NULL) |
7c9ca621 BP |
2007 | return -EBADSLT; |
2008 | ||
2009 | while (loops < 3) { | |
c743ffd0 SW |
2010 | rg_locked = 1; |
2011 | ||
2012 | if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) { | |
2013 | rg_locked = 0; | |
9dbe9610 SW |
2014 | if (skip && skip--) |
2015 | goto next_rgrp; | |
0e27c18c BP |
2016 | if (!gfs2_rs_active(rs)) { |
2017 | if (loops == 0 && | |
2018 | !fast_to_acquire(rs->rs_rbm.rgd)) | |
2019 | goto next_rgrp; | |
2020 | if ((loops < 2) && | |
2021 | gfs2_rgrp_used_recently(rs, 1000) && | |
2022 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
2023 | goto next_rgrp; | |
2024 | } | |
4a993fb1 | 2025 | error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl, |
8e2e0047 BP |
2026 | LM_ST_EXCLUSIVE, flags, |
2027 | &rs->rs_rgd_gh); | |
c743ffd0 SW |
2028 | if (unlikely(error)) |
2029 | return error; | |
bcd97c06 SW |
2030 | if (!gfs2_rs_active(rs) && (loops < 2) && |
2031 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
2032 | goto skip_rgrp; | |
c743ffd0 | 2033 | if (sdp->sd_args.ar_rgrplvb) { |
4a993fb1 | 2034 | error = update_rgrp_lvb(rs->rs_rbm.rgd); |
c743ffd0 | 2035 | if (unlikely(error)) { |
90306c41 BM |
2036 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
2037 | return error; | |
2038 | } | |
2039 | } | |
292c8c14 | 2040 | } |
666d1d8a | 2041 | |
c743ffd0 | 2042 | /* Skip unuseable resource groups */ |
5ea5050c BP |
2043 | if ((rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC | |
2044 | GFS2_RDF_ERROR)) || | |
25435e5e | 2045 | (loops == 0 && ap->target > rs->rs_rbm.rgd->rd_extfail_pt)) |
c743ffd0 SW |
2046 | goto skip_rgrp; |
2047 | ||
2048 | if (sdp->sd_args.ar_rgrplvb) | |
2049 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); | |
2050 | ||
2051 | /* Get a reservation if we don't already have one */ | |
2052 | if (!gfs2_rs_active(rs)) | |
7b9cff46 | 2053 | rg_mblk_search(rs->rs_rbm.rgd, ip, ap); |
c743ffd0 SW |
2054 | |
2055 | /* Skip rgrps when we can't get a reservation on first pass */ | |
2056 | if (!gfs2_rs_active(rs) && (loops < 1)) | |
2057 | goto check_rgrp; | |
2058 | ||
2059 | /* If rgrp has enough free space, use it */ | |
25435e5e AD |
2060 | if (rs->rs_rbm.rgd->rd_free_clone >= ap->target || |
2061 | (loops == 2 && ap->min_target && | |
2062 | rs->rs_rbm.rgd->rd_free_clone >= ap->min_target)) { | |
c743ffd0 | 2063 | ip->i_rgd = rs->rs_rbm.rgd; |
25435e5e | 2064 | ap->allowed = ip->i_rgd->rd_free_clone; |
c743ffd0 | 2065 | return 0; |
b3b94faa | 2066 | } |
c743ffd0 SW |
2067 | check_rgrp: |
2068 | /* Check for unlinked inodes which can be reclaimed */ | |
2069 | if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK) | |
2070 | try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked, | |
2071 | ip->i_no_addr); | |
2072 | skip_rgrp: | |
1330edbe BP |
2073 | /* Drop reservation, if we couldn't use reserved rgrp */ |
2074 | if (gfs2_rs_active(rs)) | |
2075 | gfs2_rs_deltree(rs); | |
2076 | ||
c743ffd0 SW |
2077 | /* Unlock rgrp if required */ |
2078 | if (!rg_locked) | |
2079 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
2080 | next_rgrp: | |
2081 | /* Find the next rgrp, and continue looking */ | |
2082 | if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin)) | |
2083 | continue; | |
9dbe9610 SW |
2084 | if (skip) |
2085 | continue; | |
c743ffd0 SW |
2086 | |
2087 | /* If we've scanned all the rgrps, but found no free blocks | |
2088 | * then this checks for some less likely conditions before | |
2089 | * trying again. | |
2090 | */ | |
c743ffd0 SW |
2091 | loops++; |
2092 | /* Check that fs hasn't grown if writing to rindex */ | |
2093 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
2094 | error = gfs2_ri_update(ip); | |
2095 | if (error) | |
2096 | return error; | |
2097 | } | |
2098 | /* Flushing the log may release space */ | |
2099 | if (loops == 2) | |
24972557 | 2100 | gfs2_log_flush(sdp, NULL, NORMAL_FLUSH); |
b3b94faa | 2101 | } |
b3b94faa | 2102 | |
c743ffd0 | 2103 | return -ENOSPC; |
b3b94faa DT |
2104 | } |
2105 | ||
2106 | /** | |
2107 | * gfs2_inplace_release - release an inplace reservation | |
2108 | * @ip: the inode the reservation was taken out on | |
2109 | * | |
2110 | * Release a reservation made by gfs2_inplace_reserve(). | |
2111 | */ | |
2112 | ||
2113 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
2114 | { | |
564e12b1 | 2115 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 2116 | |
564e12b1 BP |
2117 | if (rs->rs_rgd_gh.gh_gl) |
2118 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
b3b94faa DT |
2119 | } |
2120 | ||
2121 | /** | |
2122 | * gfs2_get_block_type - Check a block in a RG is of given type | |
2123 | * @rgd: the resource group holding the block | |
2124 | * @block: the block number | |
2125 | * | |
2126 | * Returns: The block type (GFS2_BLKST_*) | |
2127 | */ | |
2128 | ||
acf7e244 | 2129 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 2130 | { |
3983903a SW |
2131 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
2132 | int ret; | |
b3b94faa | 2133 | |
3983903a SW |
2134 | ret = gfs2_rbm_from_block(&rbm, block); |
2135 | WARN_ON_ONCE(ret != 0); | |
b3b94faa | 2136 | |
c04a2ef3 | 2137 | return gfs2_testbit(&rbm); |
b3b94faa DT |
2138 | } |
2139 | ||
60a0b8f9 | 2140 | |
b3e47ca0 BP |
2141 | /** |
2142 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 2143 | * @rbm: the resource group information |
b3e47ca0 | 2144 | * @dinode: TRUE if the first block we allocate is for a dinode |
c04a2ef3 | 2145 | * @n: The extent length (value/result) |
b3e47ca0 | 2146 | * |
c04a2ef3 | 2147 | * Add the bitmap buffer to the transaction. |
b3e47ca0 | 2148 | * Set the found bits to @new_state to change block's allocation state. |
b3e47ca0 | 2149 | */ |
c04a2ef3 | 2150 | static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
4a993fb1 | 2151 | unsigned int *n) |
b3e47ca0 | 2152 | { |
c04a2ef3 | 2153 | struct gfs2_rbm pos = { .rgd = rbm->rgd, }; |
b3e47ca0 | 2154 | const unsigned int elen = *n; |
c04a2ef3 SW |
2155 | u64 block; |
2156 | int ret; | |
b3e47ca0 | 2157 | |
c04a2ef3 SW |
2158 | *n = 1; |
2159 | block = gfs2_rbm_to_block(rbm); | |
e579ed4f | 2160 | gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh); |
3e6339dd | 2161 | gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
c04a2ef3 | 2162 | block++; |
60a0b8f9 | 2163 | while (*n < elen) { |
c04a2ef3 | 2164 | ret = gfs2_rbm_from_block(&pos, block); |
0688a5ec | 2165 | if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE) |
60a0b8f9 | 2166 | break; |
e579ed4f | 2167 | gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh); |
3e6339dd | 2168 | gfs2_setbit(&pos, true, GFS2_BLKST_USED); |
60a0b8f9 | 2169 | (*n)++; |
c04a2ef3 | 2170 | block++; |
c8cdf479 | 2171 | } |
b3b94faa DT |
2172 | } |
2173 | ||
2174 | /** | |
2175 | * rgblk_free - Change alloc state of given block(s) | |
2176 | * @sdp: the filesystem | |
2177 | * @bstart: the start of a run of blocks to free | |
2178 | * @blen: the length of the block run (all must lie within ONE RG!) | |
2179 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
2180 | * | |
2181 | * Returns: Resource group containing the block(s) | |
2182 | */ | |
2183 | ||
cd915493 SW |
2184 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
2185 | u32 blen, unsigned char new_state) | |
b3b94faa | 2186 | { |
3b1d0b9d | 2187 | struct gfs2_rbm rbm; |
d24e0569 | 2188 | struct gfs2_bitmap *bi, *bi_prev = NULL; |
b3b94faa | 2189 | |
3b1d0b9d SW |
2190 | rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
2191 | if (!rbm.rgd) { | |
b3b94faa | 2192 | if (gfs2_consist(sdp)) |
382066da | 2193 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
2194 | return NULL; |
2195 | } | |
2196 | ||
d24e0569 | 2197 | gfs2_rbm_from_block(&rbm, bstart); |
b3b94faa | 2198 | while (blen--) { |
e579ed4f | 2199 | bi = rbm_bi(&rbm); |
d24e0569 BP |
2200 | if (bi != bi_prev) { |
2201 | if (!bi->bi_clone) { | |
2202 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
2203 | GFP_NOFS | __GFP_NOFAIL); | |
2204 | memcpy(bi->bi_clone + bi->bi_offset, | |
2205 | bi->bi_bh->b_data + bi->bi_offset, | |
2206 | bi->bi_len); | |
2207 | } | |
2208 | gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh); | |
2209 | bi_prev = bi; | |
b3b94faa | 2210 | } |
3e6339dd | 2211 | gfs2_setbit(&rbm, false, new_state); |
d24e0569 | 2212 | gfs2_rbm_incr(&rbm); |
b3b94faa DT |
2213 | } |
2214 | ||
3b1d0b9d | 2215 | return rbm.rgd; |
b3b94faa DT |
2216 | } |
2217 | ||
2218 | /** | |
09010978 SW |
2219 | * gfs2_rgrp_dump - print out an rgrp |
2220 | * @seq: The iterator | |
2221 | * @gl: The glock in question | |
2222 | * | |
2223 | */ | |
2224 | ||
ac3beb6a | 2225 | void gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) |
09010978 | 2226 | { |
8e2e0047 BP |
2227 | struct gfs2_rgrpd *rgd = gl->gl_object; |
2228 | struct gfs2_blkreserv *trs; | |
2229 | const struct rb_node *n; | |
2230 | ||
09010978 | 2231 | if (rgd == NULL) |
ac3beb6a | 2232 | return; |
5ea5050c | 2233 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u e:%u\n", |
09010978 | 2234 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 | 2235 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
5ea5050c | 2236 | rgd->rd_reserved, rgd->rd_extfail_pt); |
8e2e0047 BP |
2237 | spin_lock(&rgd->rd_rsspin); |
2238 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { | |
2239 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
2240 | dump_rs(seq, trs); | |
2241 | } | |
2242 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
2243 | } |
2244 | ||
6050b9c7 SW |
2245 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
2246 | { | |
2247 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2248 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 2249 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
2250 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
2251 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
2252 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
2253 | } | |
2254 | ||
8e2e0047 | 2255 | /** |
5b924ae2 SW |
2256 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
2257 | * @ip: The inode we have just allocated blocks for | |
2258 | * @rbm: The start of the allocated blocks | |
2259 | * @len: The extent length | |
8e2e0047 | 2260 | * |
5b924ae2 SW |
2261 | * Adjusts a reservation after an allocation has taken place. If the |
2262 | * reservation does not match the allocation, or if it is now empty | |
2263 | * then it is removed. | |
8e2e0047 | 2264 | */ |
5b924ae2 SW |
2265 | |
2266 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
2267 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 BP |
2268 | { |
2269 | struct gfs2_blkreserv *rs = ip->i_res; | |
5b924ae2 SW |
2270 | struct gfs2_rgrpd *rgd = rbm->rgd; |
2271 | unsigned rlen; | |
2272 | u64 block; | |
2273 | int ret; | |
8e2e0047 | 2274 | |
5b924ae2 SW |
2275 | spin_lock(&rgd->rd_rsspin); |
2276 | if (gfs2_rs_active(rs)) { | |
2277 | if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) { | |
2278 | block = gfs2_rbm_to_block(rbm); | |
2279 | ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len); | |
2280 | rlen = min(rs->rs_free, len); | |
2281 | rs->rs_free -= rlen; | |
2282 | rgd->rd_reserved -= rlen; | |
9e733d39 | 2283 | trace_gfs2_rs(rs, TRACE_RS_CLAIM); |
5b924ae2 SW |
2284 | if (rs->rs_free && !ret) |
2285 | goto out; | |
1a855033 BP |
2286 | /* We used up our block reservation, so we should |
2287 | reserve more blocks next time. */ | |
2288 | atomic_add(RGRP_RSRV_ADDBLKS, &rs->rs_sizehint); | |
5b924ae2 | 2289 | } |
20095218 | 2290 | __rs_deltree(rs); |
8e2e0047 | 2291 | } |
5b924ae2 SW |
2292 | out: |
2293 | spin_unlock(&rgd->rd_rsspin); | |
8e2e0047 BP |
2294 | } |
2295 | ||
9e07f2cb SW |
2296 | /** |
2297 | * gfs2_set_alloc_start - Set starting point for block allocation | |
2298 | * @rbm: The rbm which will be set to the required location | |
2299 | * @ip: The gfs2 inode | |
2300 | * @dinode: Flag to say if allocation includes a new inode | |
2301 | * | |
2302 | * This sets the starting point from the reservation if one is active | |
2303 | * otherwise it falls back to guessing a start point based on the | |
2304 | * inode's goal block or the last allocation point in the rgrp. | |
2305 | */ | |
2306 | ||
2307 | static void gfs2_set_alloc_start(struct gfs2_rbm *rbm, | |
2308 | const struct gfs2_inode *ip, bool dinode) | |
2309 | { | |
2310 | u64 goal; | |
2311 | ||
2312 | if (gfs2_rs_active(ip->i_res)) { | |
2313 | *rbm = ip->i_res->rs_rbm; | |
2314 | return; | |
2315 | } | |
2316 | ||
2317 | if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal)) | |
2318 | goal = ip->i_goal; | |
2319 | else | |
2320 | goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0; | |
2321 | ||
2322 | gfs2_rbm_from_block(rbm, goal); | |
2323 | } | |
2324 | ||
09010978 | 2325 | /** |
6e87ed0f | 2326 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 2327 | * @ip: the inode to allocate the block for |
09010978 | 2328 | * @bn: Used to return the starting block number |
8e2e0047 | 2329 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 2330 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 2331 | * @generation: the generation number of the inode |
b3b94faa | 2332 | * |
09010978 | 2333 | * Returns: 0 or error |
b3b94faa DT |
2334 | */ |
2335 | ||
6a8099ed | 2336 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 2337 | bool dinode, u64 *generation) |
b3b94faa | 2338 | { |
feaa7bba | 2339 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 2340 | struct buffer_head *dibh; |
4a993fb1 | 2341 | struct gfs2_rbm rbm = { .rgd = ip->i_rgd, }; |
6a8099ed | 2342 | unsigned int ndata; |
3c5d785a | 2343 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 2344 | int error; |
b3b94faa | 2345 | |
9e07f2cb | 2346 | gfs2_set_alloc_start(&rbm, ip, dinode); |
5ce13431 | 2347 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false, NULL); |
62e252ee | 2348 | |
137834a6 | 2349 | if (error == -ENOSPC) { |
9e07f2cb | 2350 | gfs2_set_alloc_start(&rbm, ip, dinode); |
5ce13431 BP |
2351 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false, |
2352 | NULL); | |
137834a6 SW |
2353 | } |
2354 | ||
62e252ee | 2355 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
5b924ae2 | 2356 | if (error) { |
5ea5050c | 2357 | fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n", |
9e733d39 | 2358 | (unsigned long long)ip->i_no_addr, error, *nblocks, |
5ea5050c BP |
2359 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags), |
2360 | rbm.rgd->rd_extfail_pt); | |
62e252ee | 2361 | goto rgrp_error; |
8e2e0047 | 2362 | } |
62e252ee | 2363 | |
c04a2ef3 SW |
2364 | gfs2_alloc_extent(&rbm, dinode, nblocks); |
2365 | block = gfs2_rbm_to_block(&rbm); | |
c743ffd0 | 2366 | rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0; |
5b924ae2 SW |
2367 | if (gfs2_rs_active(ip->i_res)) |
2368 | gfs2_adjust_reservation(ip, &rbm, *nblocks); | |
6a8099ed SW |
2369 | ndata = *nblocks; |
2370 | if (dinode) | |
2371 | ndata--; | |
b3e47ca0 | 2372 | |
3c5d785a | 2373 | if (!dinode) { |
6a8099ed | 2374 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
2375 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2376 | if (error == 0) { | |
2377 | struct gfs2_dinode *di = | |
2378 | (struct gfs2_dinode *)dibh->b_data; | |
350a9b0a | 2379 | gfs2_trans_add_meta(ip->i_gl, dibh); |
3c5d785a BP |
2380 | di->di_goal_meta = di->di_goal_data = |
2381 | cpu_to_be64(ip->i_goal); | |
2382 | brelse(dibh); | |
2383 | } | |
d9ba7615 | 2384 | } |
4a993fb1 | 2385 | if (rbm.rgd->rd_free < *nblocks) { |
fc554ed3 | 2386 | pr_warn("nblocks=%u\n", *nblocks); |
09010978 | 2387 | goto rgrp_error; |
8e2e0047 | 2388 | } |
09010978 | 2389 | |
4a993fb1 | 2390 | rbm.rgd->rd_free -= *nblocks; |
3c5d785a | 2391 | if (dinode) { |
4a993fb1 SW |
2392 | rbm.rgd->rd_dinodes++; |
2393 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2394 | if (*generation == 0) |
4a993fb1 | 2395 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2396 | } |
b3b94faa | 2397 | |
350a9b0a | 2398 | gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh); |
4a993fb1 SW |
2399 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); |
2400 | gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
b3b94faa | 2401 | |
6a8099ed | 2402 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a | 2403 | if (dinode) |
b2c8b3ea | 2404 | gfs2_trans_add_unrevoke(sdp, block, *nblocks); |
6a8099ed | 2405 | |
fd4b4e04 | 2406 | gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa | 2407 | |
4a993fb1 SW |
2408 | rbm.rgd->rd_free_clone -= *nblocks; |
2409 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, | |
6e87ed0f | 2410 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2411 | *bn = block; |
2412 | return 0; | |
2413 | ||
2414 | rgrp_error: | |
4a993fb1 | 2415 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2416 | return -EIO; |
b3b94faa DT |
2417 | } |
2418 | ||
2419 | /** | |
46fcb2ed | 2420 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
2421 | * @ip: the inode these blocks are being freed from |
2422 | * @bstart: first block of a run of contiguous blocks | |
2423 | * @blen: the length of the block run | |
46fcb2ed | 2424 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2425 | * |
2426 | */ | |
2427 | ||
46fcb2ed | 2428 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 2429 | { |
feaa7bba | 2430 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2431 | struct gfs2_rgrpd *rgd; |
2432 | ||
2433 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
2434 | if (!rgd) | |
2435 | return; | |
41db1ab9 | 2436 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2437 | rgd->rd_free += blen; |
66fc061b | 2438 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
350a9b0a | 2439 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2440 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 | 2441 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 2442 | |
6d3117b4 | 2443 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 2444 | if (meta || ip->i_depth) |
6d3117b4 | 2445 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 2446 | } |
b3b94faa | 2447 | |
4c16c36a BP |
2448 | /** |
2449 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2450 | * @ip: the inode these blocks are being freed from | |
2451 | * @bstart: first block of a run of contiguous blocks | |
2452 | * @blen: the length of the block run | |
2453 | * | |
2454 | */ | |
2455 | ||
2456 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
2457 | { | |
2458 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2459 | ||
46fcb2ed | 2460 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 2461 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2462 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2463 | } |
2464 | ||
feaa7bba SW |
2465 | void gfs2_unlink_di(struct inode *inode) |
2466 | { | |
2467 | struct gfs2_inode *ip = GFS2_I(inode); | |
2468 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2469 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2470 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
2471 | |
2472 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
2473 | if (!rgd) | |
2474 | return; | |
41db1ab9 | 2475 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
350a9b0a | 2476 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2477 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2478 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2479 | update_rgrp_lvb_unlinked(rgd, 1); | |
feaa7bba SW |
2480 | } |
2481 | ||
cd915493 | 2482 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
2483 | { |
2484 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2485 | struct gfs2_rgrpd *tmp_rgd; | |
2486 | ||
2487 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
2488 | if (!tmp_rgd) | |
2489 | return; | |
2490 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
2491 | ||
73f74948 | 2492 | if (!rgd->rd_dinodes) |
b3b94faa | 2493 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2494 | rgd->rd_dinodes--; |
cfc8b549 | 2495 | rgd->rd_free++; |
b3b94faa | 2496 | |
350a9b0a | 2497 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2498 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2499 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2500 | update_rgrp_lvb_unlinked(rgd, -1); | |
b3b94faa DT |
2501 | |
2502 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
2503 | } |
2504 | ||
b3b94faa DT |
2505 | |
2506 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
2507 | { | |
dbb7cae2 | 2508 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
41db1ab9 | 2509 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2510 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 2511 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2512 | } |
2513 | ||
acf7e244 SW |
2514 | /** |
2515 | * gfs2_check_blk_type - Check the type of a block | |
2516 | * @sdp: The superblock | |
2517 | * @no_addr: The block number to check | |
2518 | * @type: The block type we are looking for | |
2519 | * | |
2520 | * Returns: 0 if the block type matches the expected type | |
2521 | * -ESTALE if it doesn't match | |
2522 | * or -ve errno if something went wrong while checking | |
2523 | */ | |
2524 | ||
2525 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2526 | { | |
2527 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2528 | struct gfs2_holder rgd_gh; |
58884c4d | 2529 | int error = -EINVAL; |
acf7e244 | 2530 | |
66fc061b | 2531 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2532 | if (!rgd) |
8339ee54 | 2533 | goto fail; |
acf7e244 SW |
2534 | |
2535 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2536 | if (error) | |
8339ee54 | 2537 | goto fail; |
acf7e244 SW |
2538 | |
2539 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
2540 | error = -ESTALE; | |
2541 | ||
2542 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
2543 | fail: |
2544 | return error; | |
2545 | } | |
2546 | ||
b3b94faa DT |
2547 | /** |
2548 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2549 | * @ip: the inode |
b3b94faa DT |
2550 | * @rlist: the list of resource groups |
2551 | * @block: the block | |
2552 | * | |
2553 | * Figure out what RG a block belongs to and add that RG to the list | |
2554 | * | |
2555 | * FIXME: Don't use NOFAIL | |
2556 | * | |
2557 | */ | |
2558 | ||
70b0c365 | 2559 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2560 | u64 block) |
b3b94faa | 2561 | { |
70b0c365 | 2562 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2563 | struct gfs2_rgrpd *rgd; |
2564 | struct gfs2_rgrpd **tmp; | |
2565 | unsigned int new_space; | |
2566 | unsigned int x; | |
2567 | ||
2568 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2569 | return; | |
2570 | ||
70b0c365 SW |
2571 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
2572 | rgd = ip->i_rgd; | |
2573 | else | |
66fc061b | 2574 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 2575 | if (!rgd) { |
70b0c365 | 2576 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
2577 | return; |
2578 | } | |
70b0c365 | 2579 | ip->i_rgd = rgd; |
b3b94faa DT |
2580 | |
2581 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2582 | if (rlist->rl_rgd[x] == rgd) | |
2583 | return; | |
2584 | ||
2585 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2586 | new_space = rlist->rl_space + 10; | |
2587 | ||
2588 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2589 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2590 | |
2591 | if (rlist->rl_rgd) { | |
2592 | memcpy(tmp, rlist->rl_rgd, | |
2593 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2594 | kfree(rlist->rl_rgd); | |
2595 | } | |
2596 | ||
2597 | rlist->rl_space = new_space; | |
2598 | rlist->rl_rgd = tmp; | |
2599 | } | |
2600 | ||
2601 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2602 | } | |
2603 | ||
2604 | /** | |
2605 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2606 | * and initialize an array of glock holders for them | |
2607 | * @rlist: the list of resource groups | |
2608 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
2609 | * |
2610 | * FIXME: Don't use NOFAIL | |
2611 | * | |
2612 | */ | |
2613 | ||
fe6c991c | 2614 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
2615 | { |
2616 | unsigned int x; | |
2617 | ||
2618 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 2619 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2620 | for (x = 0; x < rlist->rl_rgrps; x++) |
2621 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 2622 | state, 0, |
b3b94faa DT |
2623 | &rlist->rl_ghs[x]); |
2624 | } | |
2625 | ||
2626 | /** | |
2627 | * gfs2_rlist_free - free a resource group list | |
27ff6a0f | 2628 | * @rlist: the list of resource groups |
b3b94faa DT |
2629 | * |
2630 | */ | |
2631 | ||
2632 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2633 | { | |
2634 | unsigned int x; | |
2635 | ||
2636 | kfree(rlist->rl_rgd); | |
2637 | ||
2638 | if (rlist->rl_ghs) { | |
2639 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2640 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2641 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2642 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2643 | } |
2644 | } | |
2645 |