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 | 598 | /** |
b54e9a0b BP |
599 | * gfs2_rsqa_alloc - make sure we have a reservation assigned to the inode |
600 | * plus a quota allocations data structure, if necessary | |
0a305e49 BP |
601 | * @ip: the inode for this reservation |
602 | */ | |
b54e9a0b | 603 | int gfs2_rsqa_alloc(struct gfs2_inode *ip) |
0a305e49 | 604 | { |
f1213cac | 605 | int error = 0; |
8e2e0047 | 606 | |
f1213cac | 607 | down_write(&ip->i_rw_mutex); |
8e2e0047 | 608 | if (ip->i_res) |
f1213cac | 609 | goto out; |
0a305e49 | 610 | |
f1213cac AD |
611 | ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); |
612 | if (!ip->i_res) { | |
613 | error = -ENOMEM; | |
614 | goto out; | |
615 | } | |
4a993fb1 | 616 | |
f1213cac | 617 | RB_CLEAR_NODE(&ip->i_res->rs_node); |
b54e9a0b BP |
618 | error = gfs2_qa_alloc(ip); |
619 | if (error) { | |
620 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); | |
621 | ip->i_res = NULL; | |
622 | } | |
623 | ||
f1213cac | 624 | out: |
0a305e49 | 625 | up_write(&ip->i_rw_mutex); |
441362d0 | 626 | return error; |
0a305e49 BP |
627 | } |
628 | ||
9e733d39 | 629 | static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs) |
8e2e0047 | 630 | { |
9e733d39 SW |
631 | gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n", |
632 | (unsigned long long)rs->rs_inum, | |
633 | (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm), | |
4a993fb1 | 634 | rs->rs_rbm.offset, rs->rs_free); |
8e2e0047 BP |
635 | } |
636 | ||
0a305e49 | 637 | /** |
8e2e0047 BP |
638 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
639 | * @rs: The reservation to remove | |
640 | * | |
641 | */ | |
20095218 | 642 | static void __rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
643 | { |
644 | struct gfs2_rgrpd *rgd; | |
645 | ||
646 | if (!gfs2_rs_active(rs)) | |
647 | return; | |
648 | ||
4a993fb1 | 649 | rgd = rs->rs_rbm.rgd; |
9e733d39 | 650 | trace_gfs2_rs(rs, TRACE_RS_TREEDEL); |
4a993fb1 | 651 | rb_erase(&rs->rs_node, &rgd->rd_rstree); |
24d634e8 | 652 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 BP |
653 | |
654 | if (rs->rs_free) { | |
e579ed4f BP |
655 | struct gfs2_bitmap *bi = rbm_bi(&rs->rs_rbm); |
656 | ||
20095218 | 657 | /* return reserved blocks to the rgrp */ |
4a993fb1 SW |
658 | BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free); |
659 | rs->rs_rbm.rgd->rd_reserved -= rs->rs_free; | |
5ea5050c BP |
660 | /* The rgrp extent failure point is likely not to increase; |
661 | it will only do so if the freed blocks are somehow | |
662 | contiguous with a span of free blocks that follows. Still, | |
663 | it will force the number to be recalculated later. */ | |
664 | rgd->rd_extfail_pt += rs->rs_free; | |
8e2e0047 | 665 | rs->rs_free = 0; |
e579ed4f | 666 | clear_bit(GBF_FULL, &bi->bi_flags); |
8e2e0047 | 667 | } |
8e2e0047 BP |
668 | } |
669 | ||
670 | /** | |
671 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
672 | * @rs: The reservation to remove | |
673 | * | |
674 | */ | |
20095218 | 675 | void gfs2_rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
676 | { |
677 | struct gfs2_rgrpd *rgd; | |
678 | ||
4a993fb1 SW |
679 | rgd = rs->rs_rbm.rgd; |
680 | if (rgd) { | |
681 | spin_lock(&rgd->rd_rsspin); | |
20095218 | 682 | __rs_deltree(rs); |
4a993fb1 SW |
683 | spin_unlock(&rgd->rd_rsspin); |
684 | } | |
8e2e0047 BP |
685 | } |
686 | ||
687 | /** | |
b54e9a0b | 688 | * gfs2_rsqa_delete - delete a multi-block reservation and quota allocation |
0a305e49 | 689 | * @ip: The inode for this reservation |
af5c2697 | 690 | * @wcount: The inode's write count, or NULL |
0a305e49 BP |
691 | * |
692 | */ | |
b54e9a0b | 693 | void gfs2_rsqa_delete(struct gfs2_inode *ip, atomic_t *wcount) |
0a305e49 BP |
694 | { |
695 | down_write(&ip->i_rw_mutex); | |
af5c2697 | 696 | if (ip->i_res && ((wcount == NULL) || (atomic_read(wcount) <= 1))) { |
20095218 | 697 | gfs2_rs_deltree(ip->i_res); |
8e2e0047 | 698 | BUG_ON(ip->i_res->rs_free); |
0a305e49 BP |
699 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); |
700 | ip->i_res = NULL; | |
b54e9a0b BP |
701 | |
702 | gfs2_qa_delete(ip); | |
0a305e49 BP |
703 | } |
704 | up_write(&ip->i_rw_mutex); | |
705 | } | |
706 | ||
8e2e0047 BP |
707 | /** |
708 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
709 | * @rgd: the rgrp that needs its space back | |
710 | * | |
711 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
712 | * give them back. This leave the reservation structures in tact, but removes | |
713 | * all of their corresponding "no-fly zones". | |
714 | */ | |
715 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
716 | { | |
717 | struct rb_node *n; | |
718 | struct gfs2_blkreserv *rs; | |
719 | ||
720 | spin_lock(&rgd->rd_rsspin); | |
721 | while ((n = rb_first(&rgd->rd_rstree))) { | |
722 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
20095218 | 723 | __rs_deltree(rs); |
8e2e0047 BP |
724 | } |
725 | spin_unlock(&rgd->rd_rsspin); | |
726 | } | |
727 | ||
8339ee54 | 728 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 729 | { |
7c9ca621 | 730 | struct rb_node *n; |
b3b94faa DT |
731 | struct gfs2_rgrpd *rgd; |
732 | struct gfs2_glock *gl; | |
733 | ||
7c9ca621 BP |
734 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
735 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
736 | gl = rgd->rd_gl; |
737 | ||
7c9ca621 | 738 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
739 | |
740 | if (gl) { | |
f3dd1649 | 741 | spin_lock(&gl->gl_lockref.lock); |
5c676f6d | 742 | gl->gl_object = NULL; |
f3dd1649 | 743 | spin_unlock(&gl->gl_lockref.lock); |
29687a2a | 744 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
745 | gfs2_glock_put(gl); |
746 | } | |
747 | ||
8339ee54 | 748 | gfs2_free_clones(rgd); |
b3b94faa | 749 | kfree(rgd->rd_bits); |
8e2e0047 | 750 | return_all_reservations(rgd); |
6bdd9be6 | 751 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
752 | } |
753 | } | |
754 | ||
bb8d8a6f SW |
755 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
756 | { | |
d77d1b58 JP |
757 | pr_info("ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); |
758 | pr_info("ri_length = %u\n", rgd->rd_length); | |
759 | pr_info("ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
760 | pr_info("ri_data = %u\n", rgd->rd_data); | |
761 | pr_info("ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
bb8d8a6f SW |
762 | } |
763 | ||
b3b94faa DT |
764 | /** |
765 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
766 | * @rgd: The resource group descriptor | |
767 | * | |
768 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
769 | * | |
770 | * Returns: errno | |
771 | */ | |
772 | ||
773 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
774 | { | |
775 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
776 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 777 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 778 | u32 bytes_left, bytes; |
b3b94faa DT |
779 | int x; |
780 | ||
feaa7bba SW |
781 | if (!length) |
782 | return -EINVAL; | |
783 | ||
dd894be8 | 784 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
785 | if (!rgd->rd_bits) |
786 | return -ENOMEM; | |
787 | ||
bb8d8a6f | 788 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
789 | |
790 | for (x = 0; x < length; x++) { | |
791 | bi = rgd->rd_bits + x; | |
792 | ||
60a0b8f9 | 793 | bi->bi_flags = 0; |
b3b94faa DT |
794 | /* small rgrp; bitmap stored completely in header block */ |
795 | if (length == 1) { | |
796 | bytes = bytes_left; | |
797 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
798 | bi->bi_start = 0; | |
799 | bi->bi_len = bytes; | |
7e230f57 | 800 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
801 | /* header block */ |
802 | } else if (x == 0) { | |
803 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
804 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
805 | bi->bi_start = 0; | |
806 | bi->bi_len = bytes; | |
7e230f57 | 807 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
808 | /* last block */ |
809 | } else if (x + 1 == length) { | |
810 | bytes = bytes_left; | |
811 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 812 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 813 | bi->bi_len = bytes; |
7e230f57 | 814 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
815 | /* other blocks */ |
816 | } else { | |
568f4c96 SW |
817 | bytes = sdp->sd_sb.sb_bsize - |
818 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 819 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 820 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 821 | bi->bi_len = bytes; |
7e230f57 | 822 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
823 | } |
824 | ||
825 | bytes_left -= bytes; | |
826 | } | |
827 | ||
828 | if (bytes_left) { | |
829 | gfs2_consist_rgrpd(rgd); | |
830 | return -EIO; | |
831 | } | |
832 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 833 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 834 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 835 | gfs2_rindex_print(rgd); |
b3b94faa DT |
836 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
837 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
838 | } | |
839 | return -EIO; | |
840 | } | |
841 | ||
842 | return 0; | |
843 | } | |
844 | ||
7ae8fa84 RP |
845 | /** |
846 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 847 | * @sdp: the filesystem |
7ae8fa84 RP |
848 | * |
849 | */ | |
850 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
851 | { | |
852 | u64 total_data = 0; | |
853 | struct inode *inode = sdp->sd_rindex; | |
854 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 855 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
856 | int error, rgrps; |
857 | ||
7ae8fa84 RP |
858 | for (rgrps = 0;; rgrps++) { |
859 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
860 | ||
bcd7278d | 861 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 862 | break; |
4306629e | 863 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
864 | sizeof(struct gfs2_rindex)); |
865 | if (error != sizeof(struct gfs2_rindex)) | |
866 | break; | |
bb8d8a6f | 867 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 868 | } |
7ae8fa84 RP |
869 | return total_data; |
870 | } | |
871 | ||
6aad1c3d | 872 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
873 | { |
874 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
875 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
876 | ||
877 | /* Figure out where to put new node */ | |
878 | while (*newn) { | |
879 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
880 | rd_node); | |
881 | ||
882 | parent = *newn; | |
883 | if (rgd->rd_addr < cur->rd_addr) | |
884 | newn = &((*newn)->rb_left); | |
885 | else if (rgd->rd_addr > cur->rd_addr) | |
886 | newn = &((*newn)->rb_right); | |
887 | else | |
6aad1c3d | 888 | return -EEXIST; |
7c9ca621 BP |
889 | } |
890 | ||
891 | rb_link_node(&rgd->rd_node, parent, newn); | |
892 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
893 | sdp->sd_rgrps++; |
894 | return 0; | |
7c9ca621 BP |
895 | } |
896 | ||
b3b94faa | 897 | /** |
6c53267f | 898 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 899 | * @ip: Pointer to the rindex inode |
b3b94faa | 900 | * |
8339ee54 | 901 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
902 | */ |
903 | ||
4306629e | 904 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
905 | { |
906 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
7005c3e4 | 907 | const unsigned bsize = sdp->sd_sb.sb_bsize; |
6c53267f | 908 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); |
8339ee54 | 909 | struct gfs2_rindex buf; |
6c53267f RP |
910 | int error; |
911 | struct gfs2_rgrpd *rgd; | |
912 | ||
8339ee54 SW |
913 | if (pos >= i_size_read(&ip->i_inode)) |
914 | return 1; | |
915 | ||
4306629e | 916 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 917 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
918 | |
919 | if (error != sizeof(struct gfs2_rindex)) | |
920 | return (error == 0) ? 1 : error; | |
6c53267f | 921 | |
6bdd9be6 | 922 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
923 | error = -ENOMEM; |
924 | if (!rgd) | |
925 | return error; | |
926 | ||
6c53267f | 927 | rgd->rd_sbd = sdp; |
8339ee54 SW |
928 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
929 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
930 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
931 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
932 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 933 | spin_lock_init(&rgd->rd_rsspin); |
7c9ca621 | 934 | |
6c53267f RP |
935 | error = compute_bitstructs(rgd); |
936 | if (error) | |
8339ee54 | 937 | goto fail; |
6c53267f | 938 | |
bb8d8a6f | 939 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
940 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
941 | if (error) | |
8339ee54 | 942 | goto fail; |
6c53267f RP |
943 | |
944 | rgd->rd_gl->gl_object = rgd; | |
31dddd9e BP |
945 | rgd->rd_gl->gl_vm.start = (rgd->rd_addr * bsize) & PAGE_CACHE_MASK; |
946 | rgd->rd_gl->gl_vm.end = PAGE_CACHE_ALIGN((rgd->rd_addr + | |
947 | rgd->rd_length) * bsize) - 1; | |
4e2f8849 | 948 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr; |
0e27c18c | 949 | rgd->rd_flags &= ~(GFS2_RDF_UPTODATE | GFS2_RDF_PREFERRED); |
7c9ca621 BP |
950 | if (rgd->rd_data > sdp->sd_max_rg_data) |
951 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 952 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 953 | error = rgd_insert(rgd); |
8339ee54 | 954 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
955 | if (!error) |
956 | return 0; | |
957 | ||
958 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 959 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
960 | |
961 | fail: | |
962 | kfree(rgd->rd_bits); | |
963 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
964 | return error; |
965 | } | |
966 | ||
0e27c18c BP |
967 | /** |
968 | * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use | |
969 | * @sdp: the GFS2 superblock | |
970 | * | |
971 | * The purpose of this function is to select a subset of the resource groups | |
972 | * and mark them as PREFERRED. We do it in such a way that each node prefers | |
973 | * to use a unique set of rgrps to minimize glock contention. | |
974 | */ | |
975 | static void set_rgrp_preferences(struct gfs2_sbd *sdp) | |
976 | { | |
977 | struct gfs2_rgrpd *rgd, *first; | |
978 | int i; | |
979 | ||
980 | /* Skip an initial number of rgrps, based on this node's journal ID. | |
981 | That should start each node out on its own set. */ | |
982 | rgd = gfs2_rgrpd_get_first(sdp); | |
983 | for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++) | |
984 | rgd = gfs2_rgrpd_get_next(rgd); | |
985 | first = rgd; | |
986 | ||
987 | do { | |
988 | rgd->rd_flags |= GFS2_RDF_PREFERRED; | |
989 | for (i = 0; i < sdp->sd_journals; i++) { | |
990 | rgd = gfs2_rgrpd_get_next(rgd); | |
959b6717 | 991 | if (!rgd || rgd == first) |
0e27c18c BP |
992 | break; |
993 | } | |
959b6717 | 994 | } while (rgd && rgd != first); |
0e27c18c BP |
995 | } |
996 | ||
6c53267f RP |
997 | /** |
998 | * gfs2_ri_update - Pull in a new resource index from the disk | |
999 | * @ip: pointer to the rindex inode | |
1000 | * | |
b3b94faa DT |
1001 | * Returns: 0 on successful update, error code otherwise |
1002 | */ | |
1003 | ||
8339ee54 | 1004 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 1005 | { |
feaa7bba | 1006 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
1007 | int error; |
1008 | ||
8339ee54 | 1009 | do { |
4306629e | 1010 | error = read_rindex_entry(ip); |
8339ee54 SW |
1011 | } while (error == 0); |
1012 | ||
1013 | if (error < 0) | |
1014 | return error; | |
b3b94faa | 1015 | |
0e27c18c BP |
1016 | set_rgrp_preferences(sdp); |
1017 | ||
cf45b752 | 1018 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
1019 | return 0; |
1020 | } | |
b3b94faa | 1021 | |
b3b94faa | 1022 | /** |
8339ee54 | 1023 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 1024 | * @sdp: The GFS2 superblock |
b3b94faa DT |
1025 | * |
1026 | * We grab a lock on the rindex inode to make sure that it doesn't | |
1027 | * change whilst we are performing an operation. We keep this lock | |
1028 | * for quite long periods of time compared to other locks. This | |
1029 | * doesn't matter, since it is shared and it is very, very rarely | |
1030 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
1031 | * | |
1032 | * This makes sure that we're using the latest copy of the resource index | |
1033 | * special file, which might have been updated if someone expanded the | |
1034 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
1035 | * | |
8339ee54 | 1036 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
1037 | */ |
1038 | ||
8339ee54 | 1039 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 1040 | { |
feaa7bba | 1041 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 1042 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
1043 | struct gfs2_holder ri_gh; |
1044 | int error = 0; | |
a365fbf3 | 1045 | int unlock_required = 0; |
b3b94faa DT |
1046 | |
1047 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 1048 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
1049 | if (!gfs2_glock_is_locked_by_me(gl)) { |
1050 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
1051 | if (error) | |
6aad1c3d | 1052 | return error; |
a365fbf3 SW |
1053 | unlock_required = 1; |
1054 | } | |
8339ee54 | 1055 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 1056 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
1057 | if (unlock_required) |
1058 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
1059 | } |
1060 | ||
1061 | return error; | |
1062 | } | |
1063 | ||
42d52e38 | 1064 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
1065 | { |
1066 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 1067 | u32 rg_flags; |
bb8d8a6f | 1068 | |
42d52e38 | 1069 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 1070 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
1071 | rgd->rd_flags &= GFS2_RDF_MASK; |
1072 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 1073 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 1074 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 1075 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
1076 | } |
1077 | ||
42d52e38 | 1078 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
1079 | { |
1080 | struct gfs2_rgrp *str = buf; | |
1081 | ||
09010978 | 1082 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 1083 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 1084 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 1085 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 1086 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
1087 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
1088 | } | |
1089 | ||
90306c41 BM |
1090 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
1091 | { | |
1092 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1093 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
1094 | ||
1095 | if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free || | |
1096 | rgl->rl_dinodes != str->rg_dinodes || | |
1097 | rgl->rl_igeneration != str->rg_igeneration) | |
1098 | return 0; | |
1099 | return 1; | |
1100 | } | |
1101 | ||
1102 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) | |
1103 | { | |
1104 | const struct gfs2_rgrp *str = buf; | |
1105 | ||
1106 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
1107 | rgl->rl_flags = str->rg_flags; | |
1108 | rgl->rl_free = str->rg_free; | |
1109 | rgl->rl_dinodes = str->rg_dinodes; | |
1110 | rgl->rl_igeneration = str->rg_igeneration; | |
1111 | rgl->__pad = 0UL; | |
1112 | } | |
1113 | ||
1114 | static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change) | |
1115 | { | |
1116 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1117 | u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change; | |
1118 | rgl->rl_unlinked = cpu_to_be32(unlinked); | |
1119 | } | |
1120 | ||
1121 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) | |
1122 | { | |
1123 | struct gfs2_bitmap *bi; | |
1124 | const u32 length = rgd->rd_length; | |
1125 | const u8 *buffer = NULL; | |
1126 | u32 i, goal, count = 0; | |
1127 | ||
1128 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
1129 | goal = 0; | |
1130 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1131 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
1132 | while (goal < bi->bi_len * GFS2_NBBY) { | |
1133 | goal = gfs2_bitfit(buffer, bi->bi_len, goal, | |
1134 | GFS2_BLKST_UNLINKED); | |
1135 | if (goal == BFITNOENT) | |
1136 | break; | |
1137 | count++; | |
1138 | goal++; | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | return count; | |
1143 | } | |
1144 | ||
1145 | ||
b3b94faa | 1146 | /** |
90306c41 BM |
1147 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
1148 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
1149 | * |
1150 | * Read in all of a Resource Group's header and bitmap blocks. | |
1151 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
1152 | * | |
1153 | * Returns: errno | |
1154 | */ | |
1155 | ||
c2b0b30e | 1156 | static int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
1157 | { |
1158 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1159 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 1160 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
1161 | struct gfs2_bitmap *bi; |
1162 | unsigned int x, y; | |
1163 | int error; | |
1164 | ||
90306c41 BM |
1165 | if (rgd->rd_bits[0].bi_bh != NULL) |
1166 | return 0; | |
1167 | ||
b3b94faa DT |
1168 | for (x = 0; x < length; x++) { |
1169 | bi = rgd->rd_bits + x; | |
c8d57703 | 1170 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, 0, &bi->bi_bh); |
b3b94faa DT |
1171 | if (error) |
1172 | goto fail; | |
1173 | } | |
1174 | ||
1175 | for (y = length; y--;) { | |
1176 | bi = rgd->rd_bits + y; | |
7276b3b0 | 1177 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
1178 | if (error) |
1179 | goto fail; | |
feaa7bba | 1180 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
1181 | GFS2_METATYPE_RG)) { |
1182 | error = -EIO; | |
1183 | goto fail; | |
1184 | } | |
1185 | } | |
1186 | ||
cf45b752 | 1187 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
1188 | for (x = 0; x < length; x++) |
1189 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 1190 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 1191 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 1192 | rgd->rd_free_clone = rgd->rd_free; |
5ea5050c BP |
1193 | /* max out the rgrp allocation failure point */ |
1194 | rgd->rd_extfail_pt = rgd->rd_free; | |
b3b94faa | 1195 | } |
951b4bd5 | 1196 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
90306c41 BM |
1197 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); |
1198 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
1199 | rgd->rd_bits[0].bi_bh->b_data); | |
1200 | } | |
1201 | else if (sdp->sd_args.ar_rgrplvb) { | |
1202 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
1203 | gfs2_consist_rgrpd(rgd); | |
1204 | error = -EIO; | |
1205 | goto fail; | |
1206 | } | |
1207 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1208 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1209 | } | |
b3b94faa DT |
1210 | return 0; |
1211 | ||
feaa7bba | 1212 | fail: |
b3b94faa DT |
1213 | while (x--) { |
1214 | bi = rgd->rd_bits + x; | |
1215 | brelse(bi->bi_bh); | |
1216 | bi->bi_bh = NULL; | |
1217 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
1218 | } | |
b3b94faa DT |
1219 | |
1220 | return error; | |
1221 | } | |
1222 | ||
c2b0b30e | 1223 | static int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
90306c41 BM |
1224 | { |
1225 | u32 rl_flags; | |
1226 | ||
1227 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1228 | return 0; | |
1229 | ||
951b4bd5 | 1230 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) |
90306c41 BM |
1231 | return gfs2_rgrp_bh_get(rgd); |
1232 | ||
1233 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1234 | rl_flags &= ~GFS2_RDF_MASK; | |
1235 | rgd->rd_flags &= GFS2_RDF_MASK; | |
1236 | rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); | |
1237 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1238 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1239 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
1240 | rgd->rd_free_clone = rgd->rd_free; | |
1241 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); | |
1242 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1243 | return 0; | |
1244 | } | |
1245 | ||
1246 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1247 | { | |
1248 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1249 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1250 | ||
1251 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1252 | return 0; | |
8b127d04 | 1253 | return gfs2_rgrp_bh_get(rgd); |
90306c41 BM |
1254 | } |
1255 | ||
b3b94faa | 1256 | /** |
39b0f1e9 BP |
1257 | * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
1258 | * @rgd: The resource group | |
b3b94faa DT |
1259 | * |
1260 | */ | |
1261 | ||
39b0f1e9 | 1262 | void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd) |
b3b94faa | 1263 | { |
bb8d8a6f | 1264 | int x, length = rgd->rd_length; |
b3b94faa | 1265 | |
b3b94faa DT |
1266 | for (x = 0; x < length; x++) { |
1267 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1268 | if (bi->bi_bh) { |
1269 | brelse(bi->bi_bh); | |
1270 | bi->bi_bh = NULL; | |
1271 | } | |
b3b94faa DT |
1272 | } |
1273 | ||
b3b94faa DT |
1274 | } |
1275 | ||
39b0f1e9 BP |
1276 | /** |
1277 | * gfs2_rgrp_go_unlock - Unlock a rgrp glock | |
1278 | * @gh: The glock holder for the resource group | |
1279 | * | |
1280 | */ | |
1281 | ||
1282 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) | |
1283 | { | |
1284 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1285 | int demote_requested = test_bit(GLF_DEMOTE, &gh->gh_gl->gl_flags) | | |
1286 | test_bit(GLF_PENDING_DEMOTE, &gh->gh_gl->gl_flags); | |
1287 | ||
1288 | if (rgd && demote_requested) | |
1289 | gfs2_rgrp_brelse(rgd); | |
1290 | } | |
1291 | ||
66fc061b | 1292 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1293 | struct buffer_head *bh, |
66fc061b | 1294 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1295 | { |
1296 | struct super_block *sb = sdp->sd_vfs; | |
f15ab561 | 1297 | u64 blk; |
64d576ba | 1298 | sector_t start = 0; |
b2c87cae | 1299 | sector_t nr_blks = 0; |
f15ab561 SW |
1300 | int rv; |
1301 | unsigned int x; | |
66fc061b SW |
1302 | u32 trimmed = 0; |
1303 | u8 diff; | |
f15ab561 SW |
1304 | |
1305 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
1306 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1307 | clone += bi->bi_offset; | |
1308 | clone += x; | |
1309 | if (bh) { | |
1310 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1311 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1312 | } else { | |
1313 | diff = ~(*clone | (*clone >> 1)); | |
1314 | } | |
f15ab561 SW |
1315 | diff &= 0x55; |
1316 | if (diff == 0) | |
1317 | continue; | |
1318 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
f15ab561 SW |
1319 | while(diff) { |
1320 | if (diff & 1) { | |
b2c87cae | 1321 | if (nr_blks == 0) |
f15ab561 | 1322 | goto start_new_extent; |
b2c87cae BP |
1323 | if ((start + nr_blks) != blk) { |
1324 | if (nr_blks >= minlen) { | |
1325 | rv = sb_issue_discard(sb, | |
1326 | start, nr_blks, | |
66fc061b SW |
1327 | GFP_NOFS, 0); |
1328 | if (rv) | |
1329 | goto fail; | |
b2c87cae | 1330 | trimmed += nr_blks; |
66fc061b | 1331 | } |
b2c87cae | 1332 | nr_blks = 0; |
f15ab561 SW |
1333 | start_new_extent: |
1334 | start = blk; | |
1335 | } | |
b2c87cae | 1336 | nr_blks++; |
f15ab561 SW |
1337 | } |
1338 | diff >>= 2; | |
b2c87cae | 1339 | blk++; |
f15ab561 SW |
1340 | } |
1341 | } | |
b2c87cae BP |
1342 | if (nr_blks >= minlen) { |
1343 | rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0); | |
f15ab561 SW |
1344 | if (rv) |
1345 | goto fail; | |
b2c87cae | 1346 | trimmed += nr_blks; |
f15ab561 | 1347 | } |
66fc061b SW |
1348 | if (ptrimmed) |
1349 | *ptrimmed = trimmed; | |
1350 | return 0; | |
1351 | ||
f15ab561 | 1352 | fail: |
66fc061b SW |
1353 | if (sdp->sd_args.ar_discard) |
1354 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 1355 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1356 | return -EIO; |
1357 | } | |
1358 | ||
1359 | /** | |
1360 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1361 | * @filp: Any file on the filesystem | |
1362 | * @argp: Pointer to the arguments (also used to pass result) | |
1363 | * | |
1364 | * Returns: 0 on success, otherwise error code | |
1365 | */ | |
1366 | ||
1367 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1368 | { | |
496ad9aa | 1369 | struct inode *inode = file_inode(filp); |
66fc061b SW |
1370 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1371 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1372 | struct buffer_head *bh; | |
1373 | struct gfs2_rgrpd *rgd; | |
1374 | struct gfs2_rgrpd *rgd_end; | |
1375 | struct gfs2_holder gh; | |
1376 | struct fstrim_range r; | |
1377 | int ret = 0; | |
1378 | u64 amt; | |
1379 | u64 trimmed = 0; | |
076f0faa | 1380 | u64 start, end, minlen; |
66fc061b | 1381 | unsigned int x; |
076f0faa | 1382 | unsigned bs_shift = sdp->sd_sb.sb_bsize_shift; |
66fc061b SW |
1383 | |
1384 | if (!capable(CAP_SYS_ADMIN)) | |
1385 | return -EPERM; | |
1386 | ||
1387 | if (!blk_queue_discard(q)) | |
1388 | return -EOPNOTSUPP; | |
1389 | ||
3a238ade | 1390 | if (copy_from_user(&r, argp, sizeof(r))) |
66fc061b SW |
1391 | return -EFAULT; |
1392 | ||
5e2f7d61 BP |
1393 | ret = gfs2_rindex_update(sdp); |
1394 | if (ret) | |
1395 | return ret; | |
1396 | ||
076f0faa LC |
1397 | start = r.start >> bs_shift; |
1398 | end = start + (r.len >> bs_shift); | |
1399 | minlen = max_t(u64, r.minlen, | |
1400 | q->limits.discard_granularity) >> bs_shift; | |
1401 | ||
6a98c333 AD |
1402 | if (end <= start || minlen > sdp->sd_max_rg_data) |
1403 | return -EINVAL; | |
1404 | ||
076f0faa | 1405 | rgd = gfs2_blk2rgrpd(sdp, start, 0); |
6a98c333 | 1406 | rgd_end = gfs2_blk2rgrpd(sdp, end, 0); |
076f0faa | 1407 | |
6a98c333 AD |
1408 | if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end)) |
1409 | && (start > rgd_end->rd_data0 + rgd_end->rd_data)) | |
1410 | return -EINVAL; /* start is beyond the end of the fs */ | |
66fc061b SW |
1411 | |
1412 | while (1) { | |
1413 | ||
1414 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1415 | if (ret) | |
1416 | goto out; | |
1417 | ||
1418 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1419 | /* Trim each bitmap in the rgrp */ | |
1420 | for (x = 0; x < rgd->rd_length; x++) { | |
1421 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
076f0faa LC |
1422 | ret = gfs2_rgrp_send_discards(sdp, |
1423 | rgd->rd_data0, NULL, bi, minlen, | |
1424 | &amt); | |
66fc061b SW |
1425 | if (ret) { |
1426 | gfs2_glock_dq_uninit(&gh); | |
1427 | goto out; | |
1428 | } | |
1429 | trimmed += amt; | |
1430 | } | |
1431 | ||
1432 | /* Mark rgrp as having been trimmed */ | |
1433 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1434 | if (ret == 0) { | |
1435 | bh = rgd->rd_bits[0].bi_bh; | |
1436 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
350a9b0a | 1437 | gfs2_trans_add_meta(rgd->rd_gl, bh); |
66fc061b | 1438 | gfs2_rgrp_out(rgd, bh->b_data); |
90306c41 | 1439 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data); |
66fc061b SW |
1440 | gfs2_trans_end(sdp); |
1441 | } | |
1442 | } | |
1443 | gfs2_glock_dq_uninit(&gh); | |
1444 | ||
1445 | if (rgd == rgd_end) | |
1446 | break; | |
1447 | ||
1448 | rgd = gfs2_rgrpd_get_next(rgd); | |
1449 | } | |
1450 | ||
1451 | out: | |
6a98c333 | 1452 | r.len = trimmed << bs_shift; |
3a238ade | 1453 | if (copy_to_user(argp, &r, sizeof(r))) |
66fc061b SW |
1454 | return -EFAULT; |
1455 | ||
1456 | return ret; | |
f15ab561 SW |
1457 | } |
1458 | ||
8e2e0047 BP |
1459 | /** |
1460 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
8e2e0047 | 1461 | * @ip: the inode structure |
8e2e0047 | 1462 | * |
8e2e0047 | 1463 | */ |
ff7f4cb4 | 1464 | static void rs_insert(struct gfs2_inode *ip) |
8e2e0047 BP |
1465 | { |
1466 | struct rb_node **newn, *parent = NULL; | |
1467 | int rc; | |
1468 | struct gfs2_blkreserv *rs = ip->i_res; | |
4a993fb1 | 1469 | struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd; |
ff7f4cb4 | 1470 | u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 | 1471 | |
8e2e0047 | 1472 | BUG_ON(gfs2_rs_active(rs)); |
c743ffd0 | 1473 | |
ff7f4cb4 SW |
1474 | spin_lock(&rgd->rd_rsspin); |
1475 | newn = &rgd->rd_rstree.rb_node; | |
8e2e0047 BP |
1476 | while (*newn) { |
1477 | struct gfs2_blkreserv *cur = | |
1478 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1479 | ||
1480 | parent = *newn; | |
ff7f4cb4 | 1481 | rc = rs_cmp(fsblock, rs->rs_free, cur); |
8e2e0047 BP |
1482 | if (rc > 0) |
1483 | newn = &((*newn)->rb_right); | |
1484 | else if (rc < 0) | |
1485 | newn = &((*newn)->rb_left); | |
1486 | else { | |
1487 | spin_unlock(&rgd->rd_rsspin); | |
ff7f4cb4 SW |
1488 | WARN_ON(1); |
1489 | return; | |
8e2e0047 BP |
1490 | } |
1491 | } | |
1492 | ||
8e2e0047 BP |
1493 | rb_link_node(&rs->rs_node, parent, newn); |
1494 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1495 | ||
8e2e0047 | 1496 | /* Do our rgrp accounting for the reservation */ |
ff7f4cb4 | 1497 | rgd->rd_reserved += rs->rs_free; /* blocks reserved */ |
8e2e0047 | 1498 | spin_unlock(&rgd->rd_rsspin); |
9e733d39 | 1499 | trace_gfs2_rs(rs, TRACE_RS_INSERT); |
8e2e0047 BP |
1500 | } |
1501 | ||
1502 | /** | |
ff7f4cb4 | 1503 | * rg_mblk_search - find a group of multiple free blocks to form a reservation |
8e2e0047 | 1504 | * @rgd: the resource group descriptor |
8e2e0047 | 1505 | * @ip: pointer to the inode for which we're reserving blocks |
7b9cff46 | 1506 | * @ap: the allocation parameters |
8e2e0047 | 1507 | * |
8e2e0047 BP |
1508 | */ |
1509 | ||
ff7f4cb4 | 1510 | static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
7b9cff46 | 1511 | const struct gfs2_alloc_parms *ap) |
8e2e0047 | 1512 | { |
ff7f4cb4 SW |
1513 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1514 | u64 goal; | |
1515 | struct gfs2_blkreserv *rs = ip->i_res; | |
1516 | u32 extlen; | |
1517 | u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved; | |
1518 | int ret; | |
af21ca8e | 1519 | struct inode *inode = &ip->i_inode; |
8e2e0047 | 1520 | |
af21ca8e BP |
1521 | if (S_ISDIR(inode->i_mode)) |
1522 | extlen = 1; | |
1523 | else { | |
7b9cff46 | 1524 | extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target); |
af21ca8e BP |
1525 | extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks); |
1526 | } | |
ff7f4cb4 | 1527 | if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen)) |
c743ffd0 SW |
1528 | return; |
1529 | ||
8e2e0047 BP |
1530 | /* Find bitmap block that contains bits for goal block */ |
1531 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
ff7f4cb4 | 1532 | goal = ip->i_goal; |
8e2e0047 | 1533 | else |
ff7f4cb4 | 1534 | goal = rgd->rd_last_alloc + rgd->rd_data0; |
8e2e0047 | 1535 | |
ff7f4cb4 SW |
1536 | if (WARN_ON(gfs2_rbm_from_block(&rbm, goal))) |
1537 | return; | |
8e2e0047 | 1538 | |
5ce13431 | 1539 | ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true, ap); |
ff7f4cb4 SW |
1540 | if (ret == 0) { |
1541 | rs->rs_rbm = rbm; | |
1542 | rs->rs_free = extlen; | |
1543 | rs->rs_inum = ip->i_no_addr; | |
1544 | rs_insert(ip); | |
13d2eb01 BP |
1545 | } else { |
1546 | if (goal == rgd->rd_last_alloc + rgd->rd_data0) | |
1547 | rgd->rd_last_alloc = 0; | |
8e2e0047 | 1548 | } |
b3e47ca0 BP |
1549 | } |
1550 | ||
5b924ae2 SW |
1551 | /** |
1552 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1553 | * @rgd: The resource group | |
1554 | * @block: The starting block | |
ff7f4cb4 | 1555 | * @length: The required length |
5b924ae2 SW |
1556 | * @ip: Ignore any reservations for this inode |
1557 | * | |
1558 | * If the block does not appear in any reservation, then return the | |
1559 | * block number unchanged. If it does appear in the reservation, then | |
1560 | * keep looking through the tree of reservations in order to find the | |
1561 | * first block number which is not reserved. | |
1562 | */ | |
1563 | ||
1564 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
ff7f4cb4 | 1565 | u32 length, |
5b924ae2 SW |
1566 | const struct gfs2_inode *ip) |
1567 | { | |
1568 | struct gfs2_blkreserv *rs; | |
1569 | struct rb_node *n; | |
1570 | int rc; | |
1571 | ||
1572 | spin_lock(&rgd->rd_rsspin); | |
ff7f4cb4 | 1573 | n = rgd->rd_rstree.rb_node; |
5b924ae2 SW |
1574 | while (n) { |
1575 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
ff7f4cb4 | 1576 | rc = rs_cmp(block, length, rs); |
5b924ae2 SW |
1577 | if (rc < 0) |
1578 | n = n->rb_left; | |
1579 | else if (rc > 0) | |
1580 | n = n->rb_right; | |
1581 | else | |
1582 | break; | |
1583 | } | |
1584 | ||
1585 | if (n) { | |
ff7f4cb4 | 1586 | while ((rs_cmp(block, length, rs) == 0) && (ip->i_res != rs)) { |
5b924ae2 | 1587 | block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free; |
ff7f4cb4 | 1588 | n = n->rb_right; |
5b924ae2 SW |
1589 | if (n == NULL) |
1590 | break; | |
1591 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1592 | } | |
1593 | } | |
1594 | ||
1595 | spin_unlock(&rgd->rd_rsspin); | |
1596 | return block; | |
1597 | } | |
1598 | ||
5b924ae2 SW |
1599 | /** |
1600 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1601 | * @rbm: The current position in the resource group | |
ff7f4cb4 SW |
1602 | * @ip: The inode for which we are searching for blocks |
1603 | * @minext: The minimum extent length | |
5ce13431 | 1604 | * @maxext: A pointer to the maximum extent structure |
5b924ae2 SW |
1605 | * |
1606 | * This checks the current position in the rgrp to see whether there is | |
1607 | * a reservation covering this block. If not then this function is a | |
1608 | * no-op. If there is, then the position is moved to the end of the | |
1609 | * contiguous reservation(s) so that we are pointing at the first | |
1610 | * non-reserved block. | |
1611 | * | |
1612 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1613 | */ | |
1614 | ||
1615 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
ff7f4cb4 | 1616 | const struct gfs2_inode *ip, |
5ce13431 BP |
1617 | u32 minext, |
1618 | struct gfs2_extent *maxext) | |
5b924ae2 SW |
1619 | { |
1620 | u64 block = gfs2_rbm_to_block(rbm); | |
ff7f4cb4 | 1621 | u32 extlen = 1; |
5b924ae2 SW |
1622 | u64 nblock; |
1623 | int ret; | |
1624 | ||
ff7f4cb4 SW |
1625 | /* |
1626 | * If we have a minimum extent length, then skip over any extent | |
1627 | * which is less than the min extent length in size. | |
1628 | */ | |
1629 | if (minext) { | |
1630 | extlen = gfs2_free_extlen(rbm, minext); | |
5ce13431 | 1631 | if (extlen <= maxext->len) |
ff7f4cb4 SW |
1632 | goto fail; |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * Check the extent which has been found against the reservations | |
1637 | * and skip if parts of it are already reserved | |
1638 | */ | |
1639 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip); | |
5ce13431 BP |
1640 | if (nblock == block) { |
1641 | if (!minext || extlen >= minext) | |
1642 | return 0; | |
1643 | ||
1644 | if (extlen > maxext->len) { | |
1645 | maxext->len = extlen; | |
1646 | maxext->rbm = *rbm; | |
1647 | } | |
ff7f4cb4 | 1648 | fail: |
5ce13431 BP |
1649 | nblock = block + extlen; |
1650 | } | |
5b924ae2 SW |
1651 | ret = gfs2_rbm_from_block(rbm, nblock); |
1652 | if (ret < 0) | |
1653 | return ret; | |
1654 | return 1; | |
1655 | } | |
1656 | ||
1657 | /** | |
1658 | * gfs2_rbm_find - Look for blocks of a particular state | |
1659 | * @rbm: Value/result starting position and final position | |
1660 | * @state: The state which we want to find | |
5ce13431 BP |
1661 | * @minext: Pointer to the requested extent length (NULL for a single block) |
1662 | * This is updated to be the actual reservation size. | |
5b924ae2 SW |
1663 | * @ip: If set, check for reservations |
1664 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping | |
1665 | * around until we've reached the starting point. | |
5ce13431 | 1666 | * @ap: the allocation parameters |
5b924ae2 SW |
1667 | * |
1668 | * Side effects: | |
1669 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1670 | * has no free blocks in it. | |
5ea5050c BP |
1671 | * - If looking for free blocks, we set rd_extfail_pt on each rgrp which |
1672 | * has come up short on a free block search. | |
5b924ae2 SW |
1673 | * |
1674 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1675 | */ | |
1676 | ||
5ce13431 BP |
1677 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
1678 | const struct gfs2_inode *ip, bool nowrap, | |
1679 | const struct gfs2_alloc_parms *ap) | |
5b924ae2 SW |
1680 | { |
1681 | struct buffer_head *bh; | |
e579ed4f | 1682 | int initial_bii; |
5b924ae2 | 1683 | u32 initial_offset; |
5ea5050c BP |
1684 | int first_bii = rbm->bii; |
1685 | u32 first_offset = rbm->offset; | |
5b924ae2 SW |
1686 | u32 offset; |
1687 | u8 *buffer; | |
5b924ae2 SW |
1688 | int n = 0; |
1689 | int iters = rbm->rgd->rd_length; | |
1690 | int ret; | |
e579ed4f | 1691 | struct gfs2_bitmap *bi; |
5ce13431 | 1692 | struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, }; |
5b924ae2 SW |
1693 | |
1694 | /* If we are not starting at the beginning of a bitmap, then we | |
1695 | * need to add one to the bitmap count to ensure that we search | |
1696 | * the starting bitmap twice. | |
1697 | */ | |
1698 | if (rbm->offset != 0) | |
1699 | iters++; | |
1700 | ||
1701 | while(1) { | |
e579ed4f BP |
1702 | bi = rbm_bi(rbm); |
1703 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
5b924ae2 SW |
1704 | (state == GFS2_BLKST_FREE)) |
1705 | goto next_bitmap; | |
1706 | ||
e579ed4f BP |
1707 | bh = bi->bi_bh; |
1708 | buffer = bh->b_data + bi->bi_offset; | |
5b924ae2 | 1709 | WARN_ON(!buffer_uptodate(bh)); |
e579ed4f BP |
1710 | if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
1711 | buffer = bi->bi_clone + bi->bi_offset; | |
5b924ae2 | 1712 | initial_offset = rbm->offset; |
e579ed4f | 1713 | offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state); |
5b924ae2 SW |
1714 | if (offset == BFITNOENT) |
1715 | goto bitmap_full; | |
1716 | rbm->offset = offset; | |
1717 | if (ip == NULL) | |
1718 | return 0; | |
1719 | ||
e579ed4f | 1720 | initial_bii = rbm->bii; |
5ce13431 BP |
1721 | ret = gfs2_reservation_check_and_update(rbm, ip, |
1722 | minext ? *minext : 0, | |
1723 | &maxext); | |
5b924ae2 SW |
1724 | if (ret == 0) |
1725 | return 0; | |
1726 | if (ret > 0) { | |
e579ed4f | 1727 | n += (rbm->bii - initial_bii); |
8d8b752a | 1728 | goto next_iter; |
5b924ae2 | 1729 | } |
5d50d532 | 1730 | if (ret == -E2BIG) { |
e579ed4f | 1731 | rbm->bii = 0; |
5d50d532 | 1732 | rbm->offset = 0; |
e579ed4f | 1733 | n += (rbm->bii - initial_bii); |
5d50d532 SW |
1734 | goto res_covered_end_of_rgrp; |
1735 | } | |
5b924ae2 SW |
1736 | return ret; |
1737 | ||
1738 | bitmap_full: /* Mark bitmap as full and fall through */ | |
a3e32136 | 1739 | if ((state == GFS2_BLKST_FREE) && initial_offset == 0) |
e579ed4f | 1740 | set_bit(GBF_FULL, &bi->bi_flags); |
5b924ae2 SW |
1741 | |
1742 | next_bitmap: /* Find next bitmap in the rgrp */ | |
1743 | rbm->offset = 0; | |
e579ed4f BP |
1744 | rbm->bii++; |
1745 | if (rbm->bii == rbm->rgd->rd_length) | |
1746 | rbm->bii = 0; | |
5d50d532 | 1747 | res_covered_end_of_rgrp: |
e579ed4f | 1748 | if ((rbm->bii == 0) && nowrap) |
5b924ae2 SW |
1749 | break; |
1750 | n++; | |
8d8b752a | 1751 | next_iter: |
5b924ae2 SW |
1752 | if (n >= iters) |
1753 | break; | |
1754 | } | |
1755 | ||
5ce13431 BP |
1756 | if (minext == NULL || state != GFS2_BLKST_FREE) |
1757 | return -ENOSPC; | |
1758 | ||
5ea5050c BP |
1759 | /* If the extent was too small, and it's smaller than the smallest |
1760 | to have failed before, remember for future reference that it's | |
1761 | useless to search this rgrp again for this amount or more. */ | |
1762 | if ((first_offset == 0) && (first_bii == 0) && | |
1763 | (*minext < rbm->rgd->rd_extfail_pt)) | |
1764 | rbm->rgd->rd_extfail_pt = *minext; | |
1765 | ||
5ce13431 BP |
1766 | /* If the maximum extent we found is big enough to fulfill the |
1767 | minimum requirements, use it anyway. */ | |
1768 | if (maxext.len) { | |
1769 | *rbm = maxext.rbm; | |
1770 | *minext = maxext.len; | |
1771 | return 0; | |
1772 | } | |
1773 | ||
5b924ae2 SW |
1774 | return -ENOSPC; |
1775 | } | |
1776 | ||
c8cdf479 SW |
1777 | /** |
1778 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1779 | * @rgd: The rgrp | |
886b1416 BP |
1780 | * @last_unlinked: block address of the last dinode we unlinked |
1781 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1782 | * |
1a0eae88 BP |
1783 | * Returns: 0 if no error |
1784 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1785 | */ |
1786 | ||
044b9414 | 1787 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1788 | { |
5b924ae2 | 1789 | u64 block; |
5f3eae75 | 1790 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1791 | struct gfs2_glock *gl; |
1792 | struct gfs2_inode *ip; | |
1793 | int error; | |
1794 | int found = 0; | |
e579ed4f | 1795 | struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 }; |
c8cdf479 | 1796 | |
5b924ae2 | 1797 | while (1) { |
5f3eae75 | 1798 | down_write(&sdp->sd_log_flush_lock); |
5ce13431 BP |
1799 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL, |
1800 | true, NULL); | |
5f3eae75 | 1801 | up_write(&sdp->sd_log_flush_lock); |
5b924ae2 SW |
1802 | if (error == -ENOSPC) |
1803 | break; | |
1804 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1805 | break; |
b3e47ca0 | 1806 | |
5b924ae2 SW |
1807 | block = gfs2_rbm_to_block(&rbm); |
1808 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1809 | break; | |
1810 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1811 | continue; |
5b924ae2 | 1812 | if (block == skip) |
1e19a195 | 1813 | continue; |
5b924ae2 | 1814 | *last_unlinked = block; |
044b9414 | 1815 | |
5b924ae2 | 1816 | error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl); |
044b9414 SW |
1817 | if (error) |
1818 | continue; | |
1819 | ||
1820 | /* If the inode is already in cache, we can ignore it here | |
1821 | * because the existing inode disposal code will deal with | |
1822 | * it when all refs have gone away. Accessing gl_object like | |
1823 | * this is not safe in general. Here it is ok because we do | |
1824 | * not dereference the pointer, and we only need an approx | |
1825 | * answer to whether it is NULL or not. | |
1826 | */ | |
1827 | ip = gl->gl_object; | |
1828 | ||
1829 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1830 | gfs2_glock_put(gl); | |
1831 | else | |
1832 | found++; | |
1833 | ||
1834 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1835 | if (found > NR_CPUS) |
044b9414 | 1836 | return; |
c8cdf479 SW |
1837 | } |
1838 | ||
1839 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1840 | return; |
c8cdf479 SW |
1841 | } |
1842 | ||
bcd97c06 SW |
1843 | /** |
1844 | * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested | |
1845 | * @rgd: The rgrp in question | |
1846 | * @loops: An indication of how picky we can be (0=very, 1=less so) | |
1847 | * | |
1848 | * This function uses the recently added glock statistics in order to | |
1849 | * figure out whether a parciular resource group is suffering from | |
1850 | * contention from multiple nodes. This is done purely on the basis | |
1851 | * of timings, since this is the only data we have to work with and | |
1852 | * our aim here is to reject a resource group which is highly contended | |
1853 | * but (very important) not to do this too often in order to ensure that | |
1854 | * we do not land up introducing fragmentation by changing resource | |
1855 | * groups when not actually required. | |
1856 | * | |
1857 | * The calculation is fairly simple, we want to know whether the SRTTB | |
1858 | * (i.e. smoothed round trip time for blocking operations) to acquire | |
1859 | * the lock for this rgrp's glock is significantly greater than the | |
1860 | * time taken for resource groups on average. We introduce a margin in | |
1861 | * the form of the variable @var which is computed as the sum of the two | |
1862 | * respective variences, and multiplied by a factor depending on @loops | |
1863 | * and whether we have a lot of data to base the decision on. This is | |
1864 | * then tested against the square difference of the means in order to | |
1865 | * decide whether the result is statistically significant or not. | |
1866 | * | |
1867 | * Returns: A boolean verdict on the congestion status | |
1868 | */ | |
1869 | ||
1870 | static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops) | |
1871 | { | |
1872 | const struct gfs2_glock *gl = rgd->rd_gl; | |
15562c43 | 1873 | const struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
bcd97c06 | 1874 | struct gfs2_lkstats *st; |
4d207133 BH |
1875 | u64 r_dcount, l_dcount; |
1876 | u64 l_srttb, a_srttb = 0; | |
bcd97c06 | 1877 | s64 srttb_diff; |
4d207133 BH |
1878 | u64 sqr_diff; |
1879 | u64 var; | |
0166b197 | 1880 | int cpu, nonzero = 0; |
bcd97c06 SW |
1881 | |
1882 | preempt_disable(); | |
f4a3ae93 BP |
1883 | for_each_present_cpu(cpu) { |
1884 | st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP]; | |
0166b197 BP |
1885 | if (st->stats[GFS2_LKS_SRTTB]) { |
1886 | a_srttb += st->stats[GFS2_LKS_SRTTB]; | |
1887 | nonzero++; | |
1888 | } | |
f4a3ae93 | 1889 | } |
bcd97c06 | 1890 | st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP]; |
0166b197 BP |
1891 | if (nonzero) |
1892 | do_div(a_srttb, nonzero); | |
bcd97c06 SW |
1893 | r_dcount = st->stats[GFS2_LKS_DCOUNT]; |
1894 | var = st->stats[GFS2_LKS_SRTTVARB] + | |
1895 | gl->gl_stats.stats[GFS2_LKS_SRTTVARB]; | |
1896 | preempt_enable(); | |
1897 | ||
1898 | l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB]; | |
1899 | l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT]; | |
1900 | ||
f4a3ae93 | 1901 | if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0)) |
bcd97c06 SW |
1902 | return false; |
1903 | ||
f4a3ae93 | 1904 | srttb_diff = a_srttb - l_srttb; |
bcd97c06 SW |
1905 | sqr_diff = srttb_diff * srttb_diff; |
1906 | ||
1907 | var *= 2; | |
1908 | if (l_dcount < 8 || r_dcount < 8) | |
1909 | var *= 2; | |
1910 | if (loops == 1) | |
1911 | var *= 2; | |
1912 | ||
1913 | return ((srttb_diff < 0) && (sqr_diff > var)); | |
1914 | } | |
1915 | ||
1916 | /** | |
1917 | * gfs2_rgrp_used_recently | |
1918 | * @rs: The block reservation with the rgrp to test | |
1919 | * @msecs: The time limit in milliseconds | |
1920 | * | |
1921 | * Returns: True if the rgrp glock has been used within the time limit | |
1922 | */ | |
1923 | static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs, | |
1924 | u64 msecs) | |
1925 | { | |
1926 | u64 tdiff; | |
1927 | ||
1928 | tdiff = ktime_to_ns(ktime_sub(ktime_get_real(), | |
1929 | rs->rs_rbm.rgd->rd_gl->gl_dstamp)); | |
1930 | ||
1931 | return tdiff > (msecs * 1000 * 1000); | |
1932 | } | |
1933 | ||
9dbe9610 SW |
1934 | static u32 gfs2_orlov_skip(const struct gfs2_inode *ip) |
1935 | { | |
1936 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1937 | u32 skip; | |
1938 | ||
1939 | get_random_bytes(&skip, sizeof(skip)); | |
1940 | return skip % sdp->sd_rgrps; | |
1941 | } | |
1942 | ||
c743ffd0 SW |
1943 | static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin) |
1944 | { | |
1945 | struct gfs2_rgrpd *rgd = *pos; | |
aa8920c9 | 1946 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
c743ffd0 SW |
1947 | |
1948 | rgd = gfs2_rgrpd_get_next(rgd); | |
1949 | if (rgd == NULL) | |
aa8920c9 | 1950 | rgd = gfs2_rgrpd_get_first(sdp); |
c743ffd0 SW |
1951 | *pos = rgd; |
1952 | if (rgd != begin) /* If we didn't wrap */ | |
1953 | return true; | |
1954 | return false; | |
1955 | } | |
1956 | ||
0e27c18c BP |
1957 | /** |
1958 | * fast_to_acquire - determine if a resource group will be fast to acquire | |
1959 | * | |
1960 | * If this is one of our preferred rgrps, it should be quicker to acquire, | |
1961 | * because we tried to set ourselves up as dlm lock master. | |
1962 | */ | |
1963 | static inline int fast_to_acquire(struct gfs2_rgrpd *rgd) | |
1964 | { | |
1965 | struct gfs2_glock *gl = rgd->rd_gl; | |
1966 | ||
1967 | if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) && | |
1968 | !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
1969 | !test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
1970 | return 1; | |
1971 | if (rgd->rd_flags & GFS2_RDF_PREFERRED) | |
1972 | return 1; | |
1973 | return 0; | |
1974 | } | |
1975 | ||
b3b94faa | 1976 | /** |
666d1d8a | 1977 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1978 | * @ip: the inode to reserve space for |
7b9cff46 | 1979 | * @ap: the allocation parameters |
b3b94faa | 1980 | * |
25435e5e AD |
1981 | * We try our best to find an rgrp that has at least ap->target blocks |
1982 | * available. After a couple of passes (loops == 2), the prospects of finding | |
1983 | * such an rgrp diminish. At this stage, we return the first rgrp that has | |
1984 | * atleast ap->min_target blocks available. Either way, we set ap->allowed to | |
1985 | * the number of blocks available in the chosen rgrp. | |
1986 | * | |
1987 | * Returns: 0 on success, | |
1988 | * -ENOMEM if a suitable rgrp can't be found | |
1989 | * errno otherwise | |
b3b94faa DT |
1990 | */ |
1991 | ||
25435e5e | 1992 | int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap) |
b3b94faa | 1993 | { |
feaa7bba | 1994 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 1995 | struct gfs2_rgrpd *begin = NULL; |
564e12b1 | 1996 | struct gfs2_blkreserv *rs = ip->i_res; |
bcd97c06 | 1997 | int error = 0, rg_locked, flags = 0; |
666d1d8a | 1998 | u64 last_unlinked = NO_BLOCK; |
7c9ca621 | 1999 | int loops = 0; |
9dbe9610 | 2000 | u32 skip = 0; |
b3b94faa | 2001 | |
90306c41 BM |
2002 | if (sdp->sd_args.ar_rgrplvb) |
2003 | flags |= GL_SKIP; | |
7b9cff46 | 2004 | if (gfs2_assert_warn(sdp, ap->target)) |
c743ffd0 | 2005 | return -EINVAL; |
8e2e0047 | 2006 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 2007 | begin = rs->rs_rbm.rgd; |
8e2e0047 | 2008 | } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) { |
4a993fb1 | 2009 | rs->rs_rbm.rgd = begin = ip->i_rgd; |
8e2e0047 | 2010 | } else { |
00a158be | 2011 | check_and_update_goal(ip); |
4a993fb1 | 2012 | rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 2013 | } |
7b9cff46 | 2014 | if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV)) |
9dbe9610 | 2015 | skip = gfs2_orlov_skip(ip); |
4a993fb1 | 2016 | if (rs->rs_rbm.rgd == NULL) |
7c9ca621 BP |
2017 | return -EBADSLT; |
2018 | ||
2019 | while (loops < 3) { | |
c743ffd0 SW |
2020 | rg_locked = 1; |
2021 | ||
2022 | if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) { | |
2023 | rg_locked = 0; | |
9dbe9610 SW |
2024 | if (skip && skip--) |
2025 | goto next_rgrp; | |
0e27c18c BP |
2026 | if (!gfs2_rs_active(rs)) { |
2027 | if (loops == 0 && | |
2028 | !fast_to_acquire(rs->rs_rbm.rgd)) | |
2029 | goto next_rgrp; | |
2030 | if ((loops < 2) && | |
2031 | gfs2_rgrp_used_recently(rs, 1000) && | |
2032 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
2033 | goto next_rgrp; | |
2034 | } | |
4a993fb1 | 2035 | error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl, |
8e2e0047 BP |
2036 | LM_ST_EXCLUSIVE, flags, |
2037 | &rs->rs_rgd_gh); | |
c743ffd0 SW |
2038 | if (unlikely(error)) |
2039 | return error; | |
bcd97c06 SW |
2040 | if (!gfs2_rs_active(rs) && (loops < 2) && |
2041 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
2042 | goto skip_rgrp; | |
c743ffd0 | 2043 | if (sdp->sd_args.ar_rgrplvb) { |
4a993fb1 | 2044 | error = update_rgrp_lvb(rs->rs_rbm.rgd); |
c743ffd0 | 2045 | if (unlikely(error)) { |
90306c41 BM |
2046 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
2047 | return error; | |
2048 | } | |
2049 | } | |
292c8c14 | 2050 | } |
666d1d8a | 2051 | |
c743ffd0 | 2052 | /* Skip unuseable resource groups */ |
5ea5050c BP |
2053 | if ((rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC | |
2054 | GFS2_RDF_ERROR)) || | |
25435e5e | 2055 | (loops == 0 && ap->target > rs->rs_rbm.rgd->rd_extfail_pt)) |
c743ffd0 SW |
2056 | goto skip_rgrp; |
2057 | ||
2058 | if (sdp->sd_args.ar_rgrplvb) | |
2059 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); | |
2060 | ||
2061 | /* Get a reservation if we don't already have one */ | |
2062 | if (!gfs2_rs_active(rs)) | |
7b9cff46 | 2063 | rg_mblk_search(rs->rs_rbm.rgd, ip, ap); |
c743ffd0 SW |
2064 | |
2065 | /* Skip rgrps when we can't get a reservation on first pass */ | |
2066 | if (!gfs2_rs_active(rs) && (loops < 1)) | |
2067 | goto check_rgrp; | |
2068 | ||
2069 | /* If rgrp has enough free space, use it */ | |
25435e5e AD |
2070 | if (rs->rs_rbm.rgd->rd_free_clone >= ap->target || |
2071 | (loops == 2 && ap->min_target && | |
2072 | rs->rs_rbm.rgd->rd_free_clone >= ap->min_target)) { | |
c743ffd0 | 2073 | ip->i_rgd = rs->rs_rbm.rgd; |
25435e5e | 2074 | ap->allowed = ip->i_rgd->rd_free_clone; |
c743ffd0 | 2075 | return 0; |
b3b94faa | 2076 | } |
c743ffd0 SW |
2077 | check_rgrp: |
2078 | /* Check for unlinked inodes which can be reclaimed */ | |
2079 | if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK) | |
2080 | try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked, | |
2081 | ip->i_no_addr); | |
2082 | skip_rgrp: | |
1330edbe BP |
2083 | /* Drop reservation, if we couldn't use reserved rgrp */ |
2084 | if (gfs2_rs_active(rs)) | |
2085 | gfs2_rs_deltree(rs); | |
2086 | ||
c743ffd0 SW |
2087 | /* Unlock rgrp if required */ |
2088 | if (!rg_locked) | |
2089 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
2090 | next_rgrp: | |
2091 | /* Find the next rgrp, and continue looking */ | |
2092 | if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin)) | |
2093 | continue; | |
9dbe9610 SW |
2094 | if (skip) |
2095 | continue; | |
c743ffd0 SW |
2096 | |
2097 | /* If we've scanned all the rgrps, but found no free blocks | |
2098 | * then this checks for some less likely conditions before | |
2099 | * trying again. | |
2100 | */ | |
c743ffd0 SW |
2101 | loops++; |
2102 | /* Check that fs hasn't grown if writing to rindex */ | |
2103 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
2104 | error = gfs2_ri_update(ip); | |
2105 | if (error) | |
2106 | return error; | |
2107 | } | |
2108 | /* Flushing the log may release space */ | |
2109 | if (loops == 2) | |
24972557 | 2110 | gfs2_log_flush(sdp, NULL, NORMAL_FLUSH); |
b3b94faa | 2111 | } |
b3b94faa | 2112 | |
c743ffd0 | 2113 | return -ENOSPC; |
b3b94faa DT |
2114 | } |
2115 | ||
2116 | /** | |
2117 | * gfs2_inplace_release - release an inplace reservation | |
2118 | * @ip: the inode the reservation was taken out on | |
2119 | * | |
2120 | * Release a reservation made by gfs2_inplace_reserve(). | |
2121 | */ | |
2122 | ||
2123 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
2124 | { | |
564e12b1 | 2125 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 2126 | |
564e12b1 BP |
2127 | if (rs->rs_rgd_gh.gh_gl) |
2128 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
b3b94faa DT |
2129 | } |
2130 | ||
2131 | /** | |
2132 | * gfs2_get_block_type - Check a block in a RG is of given type | |
2133 | * @rgd: the resource group holding the block | |
2134 | * @block: the block number | |
2135 | * | |
2136 | * Returns: The block type (GFS2_BLKST_*) | |
2137 | */ | |
2138 | ||
acf7e244 | 2139 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 2140 | { |
3983903a SW |
2141 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
2142 | int ret; | |
b3b94faa | 2143 | |
3983903a SW |
2144 | ret = gfs2_rbm_from_block(&rbm, block); |
2145 | WARN_ON_ONCE(ret != 0); | |
b3b94faa | 2146 | |
c04a2ef3 | 2147 | return gfs2_testbit(&rbm); |
b3b94faa DT |
2148 | } |
2149 | ||
60a0b8f9 | 2150 | |
b3e47ca0 BP |
2151 | /** |
2152 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 2153 | * @rbm: the resource group information |
b3e47ca0 | 2154 | * @dinode: TRUE if the first block we allocate is for a dinode |
c04a2ef3 | 2155 | * @n: The extent length (value/result) |
b3e47ca0 | 2156 | * |
c04a2ef3 | 2157 | * Add the bitmap buffer to the transaction. |
b3e47ca0 | 2158 | * Set the found bits to @new_state to change block's allocation state. |
b3e47ca0 | 2159 | */ |
c04a2ef3 | 2160 | static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
4a993fb1 | 2161 | unsigned int *n) |
b3e47ca0 | 2162 | { |
c04a2ef3 | 2163 | struct gfs2_rbm pos = { .rgd = rbm->rgd, }; |
b3e47ca0 | 2164 | const unsigned int elen = *n; |
c04a2ef3 SW |
2165 | u64 block; |
2166 | int ret; | |
b3e47ca0 | 2167 | |
c04a2ef3 SW |
2168 | *n = 1; |
2169 | block = gfs2_rbm_to_block(rbm); | |
e579ed4f | 2170 | gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh); |
3e6339dd | 2171 | gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
c04a2ef3 | 2172 | block++; |
60a0b8f9 | 2173 | while (*n < elen) { |
c04a2ef3 | 2174 | ret = gfs2_rbm_from_block(&pos, block); |
0688a5ec | 2175 | if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE) |
60a0b8f9 | 2176 | break; |
e579ed4f | 2177 | gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh); |
3e6339dd | 2178 | gfs2_setbit(&pos, true, GFS2_BLKST_USED); |
60a0b8f9 | 2179 | (*n)++; |
c04a2ef3 | 2180 | block++; |
c8cdf479 | 2181 | } |
b3b94faa DT |
2182 | } |
2183 | ||
2184 | /** | |
2185 | * rgblk_free - Change alloc state of given block(s) | |
2186 | * @sdp: the filesystem | |
2187 | * @bstart: the start of a run of blocks to free | |
2188 | * @blen: the length of the block run (all must lie within ONE RG!) | |
2189 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
2190 | * | |
2191 | * Returns: Resource group containing the block(s) | |
2192 | */ | |
2193 | ||
cd915493 SW |
2194 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
2195 | u32 blen, unsigned char new_state) | |
b3b94faa | 2196 | { |
3b1d0b9d | 2197 | struct gfs2_rbm rbm; |
d24e0569 | 2198 | struct gfs2_bitmap *bi, *bi_prev = NULL; |
b3b94faa | 2199 | |
3b1d0b9d SW |
2200 | rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
2201 | if (!rbm.rgd) { | |
b3b94faa | 2202 | if (gfs2_consist(sdp)) |
382066da | 2203 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
2204 | return NULL; |
2205 | } | |
2206 | ||
d24e0569 | 2207 | gfs2_rbm_from_block(&rbm, bstart); |
b3b94faa | 2208 | while (blen--) { |
e579ed4f | 2209 | bi = rbm_bi(&rbm); |
d24e0569 BP |
2210 | if (bi != bi_prev) { |
2211 | if (!bi->bi_clone) { | |
2212 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
2213 | GFP_NOFS | __GFP_NOFAIL); | |
2214 | memcpy(bi->bi_clone + bi->bi_offset, | |
2215 | bi->bi_bh->b_data + bi->bi_offset, | |
2216 | bi->bi_len); | |
2217 | } | |
2218 | gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh); | |
2219 | bi_prev = bi; | |
b3b94faa | 2220 | } |
3e6339dd | 2221 | gfs2_setbit(&rbm, false, new_state); |
d24e0569 | 2222 | gfs2_rbm_incr(&rbm); |
b3b94faa DT |
2223 | } |
2224 | ||
3b1d0b9d | 2225 | return rbm.rgd; |
b3b94faa DT |
2226 | } |
2227 | ||
2228 | /** | |
09010978 SW |
2229 | * gfs2_rgrp_dump - print out an rgrp |
2230 | * @seq: The iterator | |
2231 | * @gl: The glock in question | |
2232 | * | |
2233 | */ | |
2234 | ||
ac3beb6a | 2235 | void gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) |
09010978 | 2236 | { |
8e2e0047 BP |
2237 | struct gfs2_rgrpd *rgd = gl->gl_object; |
2238 | struct gfs2_blkreserv *trs; | |
2239 | const struct rb_node *n; | |
2240 | ||
09010978 | 2241 | if (rgd == NULL) |
ac3beb6a | 2242 | return; |
5ea5050c | 2243 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u e:%u\n", |
09010978 | 2244 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 | 2245 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
5ea5050c | 2246 | rgd->rd_reserved, rgd->rd_extfail_pt); |
8e2e0047 BP |
2247 | spin_lock(&rgd->rd_rsspin); |
2248 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { | |
2249 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
2250 | dump_rs(seq, trs); | |
2251 | } | |
2252 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
2253 | } |
2254 | ||
6050b9c7 SW |
2255 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
2256 | { | |
2257 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2258 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 2259 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
2260 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
2261 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
2262 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
2263 | } | |
2264 | ||
8e2e0047 | 2265 | /** |
5b924ae2 SW |
2266 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
2267 | * @ip: The inode we have just allocated blocks for | |
2268 | * @rbm: The start of the allocated blocks | |
2269 | * @len: The extent length | |
8e2e0047 | 2270 | * |
5b924ae2 SW |
2271 | * Adjusts a reservation after an allocation has taken place. If the |
2272 | * reservation does not match the allocation, or if it is now empty | |
2273 | * then it is removed. | |
8e2e0047 | 2274 | */ |
5b924ae2 SW |
2275 | |
2276 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
2277 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 BP |
2278 | { |
2279 | struct gfs2_blkreserv *rs = ip->i_res; | |
5b924ae2 SW |
2280 | struct gfs2_rgrpd *rgd = rbm->rgd; |
2281 | unsigned rlen; | |
2282 | u64 block; | |
2283 | int ret; | |
8e2e0047 | 2284 | |
5b924ae2 SW |
2285 | spin_lock(&rgd->rd_rsspin); |
2286 | if (gfs2_rs_active(rs)) { | |
2287 | if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) { | |
2288 | block = gfs2_rbm_to_block(rbm); | |
2289 | ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len); | |
2290 | rlen = min(rs->rs_free, len); | |
2291 | rs->rs_free -= rlen; | |
2292 | rgd->rd_reserved -= rlen; | |
9e733d39 | 2293 | trace_gfs2_rs(rs, TRACE_RS_CLAIM); |
5b924ae2 SW |
2294 | if (rs->rs_free && !ret) |
2295 | goto out; | |
1a855033 BP |
2296 | /* We used up our block reservation, so we should |
2297 | reserve more blocks next time. */ | |
2298 | atomic_add(RGRP_RSRV_ADDBLKS, &rs->rs_sizehint); | |
5b924ae2 | 2299 | } |
20095218 | 2300 | __rs_deltree(rs); |
8e2e0047 | 2301 | } |
5b924ae2 SW |
2302 | out: |
2303 | spin_unlock(&rgd->rd_rsspin); | |
8e2e0047 BP |
2304 | } |
2305 | ||
9e07f2cb SW |
2306 | /** |
2307 | * gfs2_set_alloc_start - Set starting point for block allocation | |
2308 | * @rbm: The rbm which will be set to the required location | |
2309 | * @ip: The gfs2 inode | |
2310 | * @dinode: Flag to say if allocation includes a new inode | |
2311 | * | |
2312 | * This sets the starting point from the reservation if one is active | |
2313 | * otherwise it falls back to guessing a start point based on the | |
2314 | * inode's goal block or the last allocation point in the rgrp. | |
2315 | */ | |
2316 | ||
2317 | static void gfs2_set_alloc_start(struct gfs2_rbm *rbm, | |
2318 | const struct gfs2_inode *ip, bool dinode) | |
2319 | { | |
2320 | u64 goal; | |
2321 | ||
2322 | if (gfs2_rs_active(ip->i_res)) { | |
2323 | *rbm = ip->i_res->rs_rbm; | |
2324 | return; | |
2325 | } | |
2326 | ||
2327 | if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal)) | |
2328 | goal = ip->i_goal; | |
2329 | else | |
2330 | goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0; | |
2331 | ||
2332 | gfs2_rbm_from_block(rbm, goal); | |
2333 | } | |
2334 | ||
09010978 | 2335 | /** |
6e87ed0f | 2336 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 2337 | * @ip: the inode to allocate the block for |
09010978 | 2338 | * @bn: Used to return the starting block number |
8e2e0047 | 2339 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 2340 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 2341 | * @generation: the generation number of the inode |
b3b94faa | 2342 | * |
09010978 | 2343 | * Returns: 0 or error |
b3b94faa DT |
2344 | */ |
2345 | ||
6a8099ed | 2346 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 2347 | bool dinode, u64 *generation) |
b3b94faa | 2348 | { |
feaa7bba | 2349 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 2350 | struct buffer_head *dibh; |
4a993fb1 | 2351 | struct gfs2_rbm rbm = { .rgd = ip->i_rgd, }; |
6a8099ed | 2352 | unsigned int ndata; |
3c5d785a | 2353 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 2354 | int error; |
b3b94faa | 2355 | |
9e07f2cb | 2356 | gfs2_set_alloc_start(&rbm, ip, dinode); |
5ce13431 | 2357 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false, NULL); |
62e252ee | 2358 | |
137834a6 | 2359 | if (error == -ENOSPC) { |
9e07f2cb | 2360 | gfs2_set_alloc_start(&rbm, ip, dinode); |
5ce13431 BP |
2361 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false, |
2362 | NULL); | |
137834a6 SW |
2363 | } |
2364 | ||
62e252ee | 2365 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
5b924ae2 | 2366 | if (error) { |
5ea5050c | 2367 | fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n", |
9e733d39 | 2368 | (unsigned long long)ip->i_no_addr, error, *nblocks, |
5ea5050c BP |
2369 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags), |
2370 | rbm.rgd->rd_extfail_pt); | |
62e252ee | 2371 | goto rgrp_error; |
8e2e0047 | 2372 | } |
62e252ee | 2373 | |
c04a2ef3 SW |
2374 | gfs2_alloc_extent(&rbm, dinode, nblocks); |
2375 | block = gfs2_rbm_to_block(&rbm); | |
c743ffd0 | 2376 | rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0; |
5b924ae2 SW |
2377 | if (gfs2_rs_active(ip->i_res)) |
2378 | gfs2_adjust_reservation(ip, &rbm, *nblocks); | |
6a8099ed SW |
2379 | ndata = *nblocks; |
2380 | if (dinode) | |
2381 | ndata--; | |
b3e47ca0 | 2382 | |
3c5d785a | 2383 | if (!dinode) { |
6a8099ed | 2384 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
2385 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2386 | if (error == 0) { | |
2387 | struct gfs2_dinode *di = | |
2388 | (struct gfs2_dinode *)dibh->b_data; | |
350a9b0a | 2389 | gfs2_trans_add_meta(ip->i_gl, dibh); |
3c5d785a BP |
2390 | di->di_goal_meta = di->di_goal_data = |
2391 | cpu_to_be64(ip->i_goal); | |
2392 | brelse(dibh); | |
2393 | } | |
d9ba7615 | 2394 | } |
4a993fb1 | 2395 | if (rbm.rgd->rd_free < *nblocks) { |
fc554ed3 | 2396 | pr_warn("nblocks=%u\n", *nblocks); |
09010978 | 2397 | goto rgrp_error; |
8e2e0047 | 2398 | } |
09010978 | 2399 | |
4a993fb1 | 2400 | rbm.rgd->rd_free -= *nblocks; |
3c5d785a | 2401 | if (dinode) { |
4a993fb1 SW |
2402 | rbm.rgd->rd_dinodes++; |
2403 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2404 | if (*generation == 0) |
4a993fb1 | 2405 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2406 | } |
b3b94faa | 2407 | |
350a9b0a | 2408 | gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh); |
4a993fb1 SW |
2409 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); |
2410 | gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
b3b94faa | 2411 | |
6a8099ed | 2412 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a | 2413 | if (dinode) |
b2c8b3ea | 2414 | gfs2_trans_add_unrevoke(sdp, block, *nblocks); |
6a8099ed | 2415 | |
fd4b4e04 | 2416 | gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa | 2417 | |
4a993fb1 SW |
2418 | rbm.rgd->rd_free_clone -= *nblocks; |
2419 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, | |
6e87ed0f | 2420 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2421 | *bn = block; |
2422 | return 0; | |
2423 | ||
2424 | rgrp_error: | |
4a993fb1 | 2425 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2426 | return -EIO; |
b3b94faa DT |
2427 | } |
2428 | ||
2429 | /** | |
46fcb2ed | 2430 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
2431 | * @ip: the inode these blocks are being freed from |
2432 | * @bstart: first block of a run of contiguous blocks | |
2433 | * @blen: the length of the block run | |
46fcb2ed | 2434 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2435 | * |
2436 | */ | |
2437 | ||
46fcb2ed | 2438 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 2439 | { |
feaa7bba | 2440 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2441 | struct gfs2_rgrpd *rgd; |
2442 | ||
2443 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
2444 | if (!rgd) | |
2445 | return; | |
41db1ab9 | 2446 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2447 | rgd->rd_free += blen; |
66fc061b | 2448 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
350a9b0a | 2449 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2450 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 | 2451 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 2452 | |
6d3117b4 | 2453 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 2454 | if (meta || ip->i_depth) |
6d3117b4 | 2455 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 2456 | } |
b3b94faa | 2457 | |
4c16c36a BP |
2458 | /** |
2459 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2460 | * @ip: the inode these blocks are being freed from | |
2461 | * @bstart: first block of a run of contiguous blocks | |
2462 | * @blen: the length of the block run | |
2463 | * | |
2464 | */ | |
2465 | ||
2466 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
2467 | { | |
2468 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2469 | ||
46fcb2ed | 2470 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 2471 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2472 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2473 | } |
2474 | ||
feaa7bba SW |
2475 | void gfs2_unlink_di(struct inode *inode) |
2476 | { | |
2477 | struct gfs2_inode *ip = GFS2_I(inode); | |
2478 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2479 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2480 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
2481 | |
2482 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
2483 | if (!rgd) | |
2484 | return; | |
41db1ab9 | 2485 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
350a9b0a | 2486 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2487 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2488 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2489 | update_rgrp_lvb_unlinked(rgd, 1); | |
feaa7bba SW |
2490 | } |
2491 | ||
cd915493 | 2492 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
2493 | { |
2494 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2495 | struct gfs2_rgrpd *tmp_rgd; | |
2496 | ||
2497 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
2498 | if (!tmp_rgd) | |
2499 | return; | |
2500 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
2501 | ||
73f74948 | 2502 | if (!rgd->rd_dinodes) |
b3b94faa | 2503 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2504 | rgd->rd_dinodes--; |
cfc8b549 | 2505 | rgd->rd_free++; |
b3b94faa | 2506 | |
350a9b0a | 2507 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2508 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2509 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2510 | update_rgrp_lvb_unlinked(rgd, -1); | |
b3b94faa DT |
2511 | |
2512 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
2513 | } |
2514 | ||
b3b94faa DT |
2515 | |
2516 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
2517 | { | |
dbb7cae2 | 2518 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
41db1ab9 | 2519 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2520 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 2521 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2522 | } |
2523 | ||
acf7e244 SW |
2524 | /** |
2525 | * gfs2_check_blk_type - Check the type of a block | |
2526 | * @sdp: The superblock | |
2527 | * @no_addr: The block number to check | |
2528 | * @type: The block type we are looking for | |
2529 | * | |
2530 | * Returns: 0 if the block type matches the expected type | |
2531 | * -ESTALE if it doesn't match | |
2532 | * or -ve errno if something went wrong while checking | |
2533 | */ | |
2534 | ||
2535 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2536 | { | |
2537 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2538 | struct gfs2_holder rgd_gh; |
58884c4d | 2539 | int error = -EINVAL; |
acf7e244 | 2540 | |
66fc061b | 2541 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2542 | if (!rgd) |
8339ee54 | 2543 | goto fail; |
acf7e244 SW |
2544 | |
2545 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2546 | if (error) | |
8339ee54 | 2547 | goto fail; |
acf7e244 SW |
2548 | |
2549 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
2550 | error = -ESTALE; | |
2551 | ||
2552 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
2553 | fail: |
2554 | return error; | |
2555 | } | |
2556 | ||
b3b94faa DT |
2557 | /** |
2558 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2559 | * @ip: the inode |
b3b94faa DT |
2560 | * @rlist: the list of resource groups |
2561 | * @block: the block | |
2562 | * | |
2563 | * Figure out what RG a block belongs to and add that RG to the list | |
2564 | * | |
2565 | * FIXME: Don't use NOFAIL | |
2566 | * | |
2567 | */ | |
2568 | ||
70b0c365 | 2569 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2570 | u64 block) |
b3b94faa | 2571 | { |
70b0c365 | 2572 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2573 | struct gfs2_rgrpd *rgd; |
2574 | struct gfs2_rgrpd **tmp; | |
2575 | unsigned int new_space; | |
2576 | unsigned int x; | |
2577 | ||
2578 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2579 | return; | |
2580 | ||
70b0c365 SW |
2581 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
2582 | rgd = ip->i_rgd; | |
2583 | else | |
66fc061b | 2584 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 2585 | if (!rgd) { |
70b0c365 | 2586 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
2587 | return; |
2588 | } | |
70b0c365 | 2589 | ip->i_rgd = rgd; |
b3b94faa DT |
2590 | |
2591 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2592 | if (rlist->rl_rgd[x] == rgd) | |
2593 | return; | |
2594 | ||
2595 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2596 | new_space = rlist->rl_space + 10; | |
2597 | ||
2598 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2599 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2600 | |
2601 | if (rlist->rl_rgd) { | |
2602 | memcpy(tmp, rlist->rl_rgd, | |
2603 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2604 | kfree(rlist->rl_rgd); | |
2605 | } | |
2606 | ||
2607 | rlist->rl_space = new_space; | |
2608 | rlist->rl_rgd = tmp; | |
2609 | } | |
2610 | ||
2611 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2612 | } | |
2613 | ||
2614 | /** | |
2615 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2616 | * and initialize an array of glock holders for them | |
2617 | * @rlist: the list of resource groups | |
2618 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
2619 | * |
2620 | * FIXME: Don't use NOFAIL | |
2621 | * | |
2622 | */ | |
2623 | ||
fe6c991c | 2624 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
2625 | { |
2626 | unsigned int x; | |
2627 | ||
2628 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 2629 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2630 | for (x = 0; x < rlist->rl_rgrps; x++) |
2631 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 2632 | state, 0, |
b3b94faa DT |
2633 | &rlist->rl_ghs[x]); |
2634 | } | |
2635 | ||
2636 | /** | |
2637 | * gfs2_rlist_free - free a resource group list | |
27ff6a0f | 2638 | * @rlist: the list of resource groups |
b3b94faa DT |
2639 | * |
2640 | */ | |
2641 | ||
2642 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2643 | { | |
2644 | unsigned int x; | |
2645 | ||
2646 | kfree(rlist->rl_rgd); | |
2647 | ||
2648 | if (rlist->rl_ghs) { | |
2649 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2650 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2651 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2652 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2653 | } |
2654 | } | |
2655 |