Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
fe6c991c | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
b3b94faa DT |
10 | #include <linux/slab.h> |
11 | #include <linux/spinlock.h> | |
12 | #include <linux/completion.h> | |
13 | #include <linux/buffer_head.h> | |
f42faf4f | 14 | #include <linux/fs.h> |
5c676f6d | 15 | #include <linux/gfs2_ondisk.h> |
1f466a47 | 16 | #include <linux/prefetch.h> |
f15ab561 | 17 | #include <linux/blkdev.h> |
7c9ca621 | 18 | #include <linux/rbtree.h> |
9dbe9610 | 19 | #include <linux/random.h> |
b3b94faa DT |
20 | |
21 | #include "gfs2.h" | |
5c676f6d | 22 | #include "incore.h" |
b3b94faa DT |
23 | #include "glock.h" |
24 | #include "glops.h" | |
b3b94faa DT |
25 | #include "lops.h" |
26 | #include "meta_io.h" | |
27 | #include "quota.h" | |
28 | #include "rgrp.h" | |
29 | #include "super.h" | |
30 | #include "trans.h" | |
5c676f6d | 31 | #include "util.h" |
172e045a | 32 | #include "log.h" |
c8cdf479 | 33 | #include "inode.h" |
63997775 | 34 | #include "trace_gfs2.h" |
b3b94faa | 35 | |
2c1e52aa | 36 | #define BFITNOENT ((u32)~0) |
6760bdcd | 37 | #define NO_BLOCK ((u64)~0) |
88c8ab1f | 38 | |
1f466a47 BP |
39 | #if BITS_PER_LONG == 32 |
40 | #define LBITMASK (0x55555555UL) | |
41 | #define LBITSKIP55 (0x55555555UL) | |
42 | #define LBITSKIP00 (0x00000000UL) | |
43 | #else | |
44 | #define LBITMASK (0x5555555555555555UL) | |
45 | #define LBITSKIP55 (0x5555555555555555UL) | |
46 | #define LBITSKIP00 (0x0000000000000000UL) | |
47 | #endif | |
48 | ||
88c8ab1f SW |
49 | /* |
50 | * These routines are used by the resource group routines (rgrp.c) | |
51 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
52 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
53 | * | |
54 | * 0 = Free | |
55 | * 1 = Used (not metadata) | |
56 | * 2 = Unlinked (still in use) inode | |
57 | * 3 = Used (metadata) | |
88c8ab1f SW |
58 | */ |
59 | ||
5ce13431 BP |
60 | struct gfs2_extent { |
61 | struct gfs2_rbm rbm; | |
62 | u32 len; | |
63 | }; | |
64 | ||
88c8ab1f SW |
65 | static const char valid_change[16] = { |
66 | /* current */ | |
feaa7bba | 67 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 68 | /* e */ 1, 0, 0, 0, |
feaa7bba | 69 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
70 | 1, 0, 0, 0 |
71 | }; | |
72 | ||
5ce13431 BP |
73 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
74 | const struct gfs2_inode *ip, bool nowrap, | |
75 | const struct gfs2_alloc_parms *ap); | |
ff7f4cb4 SW |
76 | |
77 | ||
88c8ab1f SW |
78 | /** |
79 | * gfs2_setbit - Set a bit in the bitmaps | |
3e6339dd SW |
80 | * @rbm: The position of the bit to set |
81 | * @do_clone: Also set the clone bitmap, if it exists | |
88c8ab1f SW |
82 | * @new_state: the new state of the block |
83 | * | |
84 | */ | |
85 | ||
3e6339dd | 86 | static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone, |
06344b91 | 87 | unsigned char new_state) |
88c8ab1f | 88 | { |
b45e41d7 | 89 | unsigned char *byte1, *byte2, *end, cur_state; |
e579ed4f BP |
90 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
91 | unsigned int buflen = bi->bi_len; | |
3e6339dd | 92 | const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 93 | |
e579ed4f BP |
94 | byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY); |
95 | end = bi->bi_bh->b_data + bi->bi_offset + buflen; | |
88c8ab1f | 96 | |
b45e41d7 | 97 | BUG_ON(byte1 >= end); |
88c8ab1f | 98 | |
b45e41d7 | 99 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 100 | |
b45e41d7 | 101 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
3e6339dd SW |
102 | printk(KERN_WARNING "GFS2: buf_blk = 0x%x old_state=%d, " |
103 | "new_state=%d\n", rbm->offset, cur_state, new_state); | |
104 | printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%x\n", | |
e579ed4f | 105 | (unsigned long long)rbm->rgd->rd_addr, bi->bi_start); |
3e6339dd | 106 | printk(KERN_WARNING "GFS2: bi_offset=0x%x bi_len=0x%x\n", |
e579ed4f | 107 | bi->bi_offset, bi->bi_len); |
95c8e17f | 108 | dump_stack(); |
3e6339dd | 109 | gfs2_consist_rgrpd(rbm->rgd); |
b45e41d7 SW |
110 | return; |
111 | } | |
112 | *byte1 ^= (cur_state ^ new_state) << bit; | |
113 | ||
e579ed4f BP |
114 | if (do_clone && bi->bi_clone) { |
115 | byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY); | |
b45e41d7 SW |
116 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
117 | *byte2 ^= (cur_state ^ new_state) << bit; | |
118 | } | |
88c8ab1f SW |
119 | } |
120 | ||
121 | /** | |
122 | * gfs2_testbit - test a bit in the bitmaps | |
c04a2ef3 | 123 | * @rbm: The bit to test |
88c8ab1f | 124 | * |
c04a2ef3 | 125 | * Returns: The two bit block state of the requested bit |
88c8ab1f SW |
126 | */ |
127 | ||
c04a2ef3 | 128 | static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm) |
88c8ab1f | 129 | { |
e579ed4f BP |
130 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
131 | const u8 *buffer = bi->bi_bh->b_data + bi->bi_offset; | |
c04a2ef3 | 132 | const u8 *byte; |
88c8ab1f SW |
133 | unsigned int bit; |
134 | ||
c04a2ef3 SW |
135 | byte = buffer + (rbm->offset / GFS2_NBBY); |
136 | bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 137 | |
c04a2ef3 | 138 | return (*byte >> bit) & GFS2_BIT_MASK; |
88c8ab1f SW |
139 | } |
140 | ||
223b2b88 SW |
141 | /** |
142 | * gfs2_bit_search | |
143 | * @ptr: Pointer to bitmap data | |
144 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
145 | * @state: The state we are searching for | |
146 | * | |
147 | * We xor the bitmap data with a patter which is the bitwise opposite | |
148 | * of what we are looking for, this gives rise to a pattern of ones | |
149 | * wherever there is a match. Since we have two bits per entry, we | |
150 | * take this pattern, shift it down by one place and then and it with | |
151 | * the original. All the even bit positions (0,2,4, etc) then represent | |
152 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
153 | * odd bit positions. | |
154 | * | |
155 | * This allows searching of a whole u64 at once (32 blocks) with a | |
156 | * single test (on 64 bit arches). | |
157 | */ | |
158 | ||
159 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
160 | { | |
161 | u64 tmp; | |
162 | static const u64 search[] = { | |
075ac448 HE |
163 | [0] = 0xffffffffffffffffULL, |
164 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
165 | [2] = 0x5555555555555555ULL, | |
166 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
167 | }; |
168 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
169 | tmp &= (tmp >> 1); | |
170 | tmp &= mask; | |
171 | return tmp; | |
172 | } | |
173 | ||
8e2e0047 BP |
174 | /** |
175 | * rs_cmp - multi-block reservation range compare | |
176 | * @blk: absolute file system block number of the new reservation | |
177 | * @len: number of blocks in the new reservation | |
178 | * @rs: existing reservation to compare against | |
179 | * | |
180 | * returns: 1 if the block range is beyond the reach of the reservation | |
181 | * -1 if the block range is before the start of the reservation | |
182 | * 0 if the block range overlaps with the reservation | |
183 | */ | |
184 | static inline int rs_cmp(u64 blk, u32 len, struct gfs2_blkreserv *rs) | |
185 | { | |
4a993fb1 | 186 | u64 startblk = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 BP |
187 | |
188 | if (blk >= startblk + rs->rs_free) | |
189 | return 1; | |
190 | if (blk + len - 1 < startblk) | |
191 | return -1; | |
192 | return 0; | |
193 | } | |
194 | ||
88c8ab1f SW |
195 | /** |
196 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
197 | * a block in a given allocation state. | |
886b1416 | 198 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 199 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 200 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 201 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
202 | * |
203 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
204 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
205 | * beginning of a bitmap block buffer, skipping any header structures, but |
206 | * headers are always a multiple of 64 bits long so that the buffer is | |
207 | * always aligned to a 64 bit boundary. | |
208 | * | |
209 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 210 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 211 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
212 | * |
213 | * Return: the block number (bitmap buffer scope) that was found | |
214 | */ | |
215 | ||
02ab1721 HE |
216 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
217 | u32 goal, u8 state) | |
88c8ab1f | 218 | { |
223b2b88 SW |
219 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
220 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
221 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
222 | u64 tmp; | |
075ac448 | 223 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
224 | u32 bit; |
225 | ||
223b2b88 SW |
226 | /* Mask off bits we don't care about at the start of the search */ |
227 | mask <<= spoint; | |
228 | tmp = gfs2_bit_search(ptr, mask, state); | |
229 | ptr++; | |
230 | while(tmp == 0 && ptr < end) { | |
075ac448 | 231 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 232 | ptr++; |
1f466a47 | 233 | } |
223b2b88 SW |
234 | /* Mask off any bits which are more than len bytes from the start */ |
235 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
236 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
237 | /* Didn't find anything, so return */ | |
238 | if (tmp == 0) | |
239 | return BFITNOENT; | |
240 | ptr--; | |
d8bd504a | 241 | bit = __ffs64(tmp); |
223b2b88 SW |
242 | bit /= 2; /* two bits per entry in the bitmap */ |
243 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
244 | } |
245 | ||
ff7f4cb4 SW |
246 | /** |
247 | * gfs2_rbm_from_block - Set the rbm based upon rgd and block number | |
248 | * @rbm: The rbm with rgd already set correctly | |
249 | * @block: The block number (filesystem relative) | |
250 | * | |
251 | * This sets the bi and offset members of an rbm based on a | |
252 | * resource group and a filesystem relative block number. The | |
253 | * resource group must be set in the rbm on entry, the bi and | |
254 | * offset members will be set by this function. | |
255 | * | |
256 | * Returns: 0 on success, or an error code | |
257 | */ | |
258 | ||
259 | static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block) | |
260 | { | |
261 | u64 rblock = block - rbm->rgd->rd_data0; | |
ff7f4cb4 SW |
262 | |
263 | if (WARN_ON_ONCE(rblock > UINT_MAX)) | |
264 | return -EINVAL; | |
265 | if (block >= rbm->rgd->rd_data0 + rbm->rgd->rd_data) | |
266 | return -E2BIG; | |
267 | ||
e579ed4f | 268 | rbm->bii = 0; |
a68a0a35 BP |
269 | rbm->offset = (u32)(rblock); |
270 | /* Check if the block is within the first block */ | |
e579ed4f | 271 | if (rbm->offset < rbm_bi(rbm)->bi_blocks) |
a68a0a35 BP |
272 | return 0; |
273 | ||
274 | /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */ | |
275 | rbm->offset += (sizeof(struct gfs2_rgrp) - | |
276 | sizeof(struct gfs2_meta_header)) * GFS2_NBBY; | |
e579ed4f BP |
277 | rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap; |
278 | rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap; | |
ff7f4cb4 SW |
279 | return 0; |
280 | } | |
281 | ||
149ed7f5 BP |
282 | /** |
283 | * gfs2_rbm_incr - increment an rbm structure | |
284 | * @rbm: The rbm with rgd already set correctly | |
285 | * | |
286 | * This function takes an existing rbm structure and increments it to the next | |
287 | * viable block offset. | |
288 | * | |
289 | * Returns: If incrementing the offset would cause the rbm to go past the | |
290 | * end of the rgrp, true is returned, otherwise false. | |
291 | * | |
292 | */ | |
293 | ||
294 | static bool gfs2_rbm_incr(struct gfs2_rbm *rbm) | |
295 | { | |
296 | if (rbm->offset + 1 < rbm_bi(rbm)->bi_blocks) { /* in the same bitmap */ | |
297 | rbm->offset++; | |
298 | return false; | |
299 | } | |
300 | if (rbm->bii == rbm->rgd->rd_length - 1) /* at the last bitmap */ | |
301 | return true; | |
302 | ||
303 | rbm->offset = 0; | |
304 | rbm->bii++; | |
305 | return false; | |
306 | } | |
307 | ||
ff7f4cb4 SW |
308 | /** |
309 | * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned | |
310 | * @rbm: Position to search (value/result) | |
311 | * @n_unaligned: Number of unaligned blocks to check | |
312 | * @len: Decremented for each block found (terminate on zero) | |
313 | * | |
314 | * Returns: true if a non-free block is encountered | |
315 | */ | |
316 | ||
317 | static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len) | |
318 | { | |
ff7f4cb4 SW |
319 | u32 n; |
320 | u8 res; | |
321 | ||
322 | for (n = 0; n < n_unaligned; n++) { | |
323 | res = gfs2_testbit(rbm); | |
324 | if (res != GFS2_BLKST_FREE) | |
325 | return true; | |
326 | (*len)--; | |
327 | if (*len == 0) | |
328 | return true; | |
149ed7f5 | 329 | if (gfs2_rbm_incr(rbm)) |
ff7f4cb4 SW |
330 | return true; |
331 | } | |
332 | ||
333 | return false; | |
334 | } | |
335 | ||
336 | /** | |
337 | * gfs2_free_extlen - Return extent length of free blocks | |
338 | * @rbm: Starting position | |
339 | * @len: Max length to check | |
340 | * | |
341 | * Starting at the block specified by the rbm, see how many free blocks | |
342 | * there are, not reading more than len blocks ahead. This can be done | |
343 | * using memchr_inv when the blocks are byte aligned, but has to be done | |
344 | * on a block by block basis in case of unaligned blocks. Also this | |
345 | * function can cope with bitmap boundaries (although it must stop on | |
346 | * a resource group boundary) | |
347 | * | |
348 | * Returns: Number of free blocks in the extent | |
349 | */ | |
350 | ||
351 | static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len) | |
352 | { | |
353 | struct gfs2_rbm rbm = *rrbm; | |
354 | u32 n_unaligned = rbm.offset & 3; | |
355 | u32 size = len; | |
356 | u32 bytes; | |
357 | u32 chunk_size; | |
358 | u8 *ptr, *start, *end; | |
359 | u64 block; | |
e579ed4f | 360 | struct gfs2_bitmap *bi; |
ff7f4cb4 SW |
361 | |
362 | if (n_unaligned && | |
363 | gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len)) | |
364 | goto out; | |
365 | ||
3701530a | 366 | n_unaligned = len & 3; |
ff7f4cb4 SW |
367 | /* Start is now byte aligned */ |
368 | while (len > 3) { | |
e579ed4f BP |
369 | bi = rbm_bi(&rbm); |
370 | start = bi->bi_bh->b_data; | |
371 | if (bi->bi_clone) | |
372 | start = bi->bi_clone; | |
373 | end = start + bi->bi_bh->b_size; | |
374 | start += bi->bi_offset; | |
ff7f4cb4 SW |
375 | BUG_ON(rbm.offset & 3); |
376 | start += (rbm.offset / GFS2_NBBY); | |
377 | bytes = min_t(u32, len / GFS2_NBBY, (end - start)); | |
378 | ptr = memchr_inv(start, 0, bytes); | |
379 | chunk_size = ((ptr == NULL) ? bytes : (ptr - start)); | |
380 | chunk_size *= GFS2_NBBY; | |
381 | BUG_ON(len < chunk_size); | |
382 | len -= chunk_size; | |
383 | block = gfs2_rbm_to_block(&rbm); | |
15bd50ad BP |
384 | if (gfs2_rbm_from_block(&rbm, block + chunk_size)) { |
385 | n_unaligned = 0; | |
ff7f4cb4 | 386 | break; |
15bd50ad BP |
387 | } |
388 | if (ptr) { | |
389 | n_unaligned = 3; | |
390 | break; | |
391 | } | |
ff7f4cb4 SW |
392 | n_unaligned = len & 3; |
393 | } | |
394 | ||
395 | /* Deal with any bits left over at the end */ | |
396 | if (n_unaligned) | |
397 | gfs2_unaligned_extlen(&rbm, n_unaligned, &len); | |
398 | out: | |
399 | return size - len; | |
400 | } | |
401 | ||
88c8ab1f SW |
402 | /** |
403 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 404 | * @rgd: the resource group descriptor |
88c8ab1f SW |
405 | * @buffer: the buffer that holds the bitmaps |
406 | * @buflen: the length (in bytes) of the buffer | |
407 | * @state: the state of the block we're looking for | |
408 | * | |
409 | * Returns: The number of bits | |
410 | */ | |
411 | ||
110acf38 SW |
412 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
413 | unsigned int buflen, u8 state) | |
88c8ab1f | 414 | { |
110acf38 SW |
415 | const u8 *byte = buffer; |
416 | const u8 *end = buffer + buflen; | |
417 | const u8 state1 = state << 2; | |
418 | const u8 state2 = state << 4; | |
419 | const u8 state3 = state << 6; | |
cd915493 | 420 | u32 count = 0; |
88c8ab1f SW |
421 | |
422 | for (; byte < end; byte++) { | |
423 | if (((*byte) & 0x03) == state) | |
424 | count++; | |
425 | if (((*byte) & 0x0C) == state1) | |
426 | count++; | |
427 | if (((*byte) & 0x30) == state2) | |
428 | count++; | |
429 | if (((*byte) & 0xC0) == state3) | |
430 | count++; | |
431 | } | |
432 | ||
433 | return count; | |
434 | } | |
435 | ||
b3b94faa DT |
436 | /** |
437 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
438 | * @rgd: the rgrp |
439 | * | |
440 | */ | |
441 | ||
442 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
443 | { | |
444 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
445 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 446 | u32 length = rgd->rd_length; |
cd915493 | 447 | u32 count[4], tmp; |
b3b94faa DT |
448 | int buf, x; |
449 | ||
cd915493 | 450 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
451 | |
452 | /* Count # blocks in each of 4 possible allocation states */ | |
453 | for (buf = 0; buf < length; buf++) { | |
454 | bi = rgd->rd_bits + buf; | |
455 | for (x = 0; x < 4; x++) | |
456 | count[x] += gfs2_bitcount(rgd, | |
457 | bi->bi_bh->b_data + | |
458 | bi->bi_offset, | |
459 | bi->bi_len, x); | |
460 | } | |
461 | ||
cfc8b549 | 462 | if (count[0] != rgd->rd_free) { |
b3b94faa DT |
463 | if (gfs2_consist_rgrpd(rgd)) |
464 | fs_err(sdp, "free data mismatch: %u != %u\n", | |
cfc8b549 | 465 | count[0], rgd->rd_free); |
b3b94faa DT |
466 | return; |
467 | } | |
468 | ||
73f74948 | 469 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 470 | if (count[1] != tmp) { |
b3b94faa DT |
471 | if (gfs2_consist_rgrpd(rgd)) |
472 | fs_err(sdp, "used data mismatch: %u != %u\n", | |
473 | count[1], tmp); | |
474 | return; | |
475 | } | |
476 | ||
6b946170 | 477 | if (count[2] + count[3] != rgd->rd_dinodes) { |
b3b94faa | 478 | if (gfs2_consist_rgrpd(rgd)) |
feaa7bba | 479 | fs_err(sdp, "used metadata mismatch: %u != %u\n", |
6b946170 | 480 | count[2] + count[3], rgd->rd_dinodes); |
b3b94faa DT |
481 | return; |
482 | } | |
b3b94faa DT |
483 | } |
484 | ||
bb8d8a6f | 485 | static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 486 | { |
bb8d8a6f SW |
487 | u64 first = rgd->rd_data0; |
488 | u64 last = first + rgd->rd_data; | |
16910427 | 489 | return first <= block && block < last; |
b3b94faa DT |
490 | } |
491 | ||
492 | /** | |
493 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
494 | * @sdp: The GFS2 superblock | |
886b1416 BP |
495 | * @blk: The data block number |
496 | * @exact: True if this needs to be an exact match | |
b3b94faa DT |
497 | * |
498 | * Returns: The resource group, or NULL if not found | |
499 | */ | |
500 | ||
66fc061b | 501 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 502 | { |
66fc061b | 503 | struct rb_node *n, *next; |
f75bbfb4 | 504 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
505 | |
506 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
507 | n = sdp->sd_rindex_tree.rb_node; |
508 | while (n) { | |
509 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
510 | next = NULL; | |
7c9ca621 | 511 | if (blk < cur->rd_addr) |
66fc061b | 512 | next = n->rb_left; |
f75bbfb4 | 513 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
514 | next = n->rb_right; |
515 | if (next == NULL) { | |
b3b94faa | 516 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
517 | if (exact) { |
518 | if (blk < cur->rd_addr) | |
519 | return NULL; | |
520 | if (blk >= cur->rd_data0 + cur->rd_data) | |
521 | return NULL; | |
522 | } | |
7c9ca621 | 523 | return cur; |
b3b94faa | 524 | } |
66fc061b | 525 | n = next; |
b3b94faa | 526 | } |
b3b94faa DT |
527 | spin_unlock(&sdp->sd_rindex_spin); |
528 | ||
529 | return NULL; | |
530 | } | |
531 | ||
532 | /** | |
533 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
534 | * @sdp: The GFS2 superblock | |
535 | * | |
536 | * Returns: The first rgrp in the filesystem | |
537 | */ | |
538 | ||
539 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
540 | { | |
7c9ca621 BP |
541 | const struct rb_node *n; |
542 | struct gfs2_rgrpd *rgd; | |
543 | ||
8339ee54 | 544 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
545 | n = rb_first(&sdp->sd_rindex_tree); |
546 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 547 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
548 | |
549 | return rgd; | |
b3b94faa DT |
550 | } |
551 | ||
552 | /** | |
553 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 554 | * @rgd: the resource group descriptor |
b3b94faa DT |
555 | * |
556 | * Returns: The next rgrp | |
557 | */ | |
558 | ||
559 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
560 | { | |
7c9ca621 BP |
561 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
562 | const struct rb_node *n; | |
563 | ||
564 | spin_lock(&sdp->sd_rindex_spin); | |
565 | n = rb_next(&rgd->rd_node); | |
566 | if (n == NULL) | |
567 | n = rb_first(&sdp->sd_rindex_tree); | |
568 | ||
569 | if (unlikely(&rgd->rd_node == n)) { | |
570 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 571 | return NULL; |
7c9ca621 BP |
572 | } |
573 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
574 | spin_unlock(&sdp->sd_rindex_spin); | |
575 | return rgd; | |
b3b94faa DT |
576 | } |
577 | ||
8339ee54 SW |
578 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
579 | { | |
580 | int x; | |
581 | ||
582 | for (x = 0; x < rgd->rd_length; x++) { | |
583 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
584 | kfree(bi->bi_clone); | |
585 | bi->bi_clone = NULL; | |
586 | } | |
587 | } | |
588 | ||
0a305e49 BP |
589 | /** |
590 | * gfs2_rs_alloc - make sure we have a reservation assigned to the inode | |
591 | * @ip: the inode for this reservation | |
592 | */ | |
593 | int gfs2_rs_alloc(struct gfs2_inode *ip) | |
594 | { | |
f1213cac | 595 | int error = 0; |
8e2e0047 | 596 | |
f1213cac | 597 | down_write(&ip->i_rw_mutex); |
8e2e0047 | 598 | if (ip->i_res) |
f1213cac | 599 | goto out; |
0a305e49 | 600 | |
f1213cac AD |
601 | ip->i_res = kmem_cache_zalloc(gfs2_rsrv_cachep, GFP_NOFS); |
602 | if (!ip->i_res) { | |
603 | error = -ENOMEM; | |
604 | goto out; | |
605 | } | |
4a993fb1 | 606 | |
f1213cac AD |
607 | RB_CLEAR_NODE(&ip->i_res->rs_node); |
608 | out: | |
0a305e49 | 609 | up_write(&ip->i_rw_mutex); |
441362d0 | 610 | return error; |
0a305e49 BP |
611 | } |
612 | ||
9e733d39 | 613 | static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs) |
8e2e0047 | 614 | { |
9e733d39 SW |
615 | gfs2_print_dbg(seq, " B: n:%llu s:%llu b:%u f:%u\n", |
616 | (unsigned long long)rs->rs_inum, | |
617 | (unsigned long long)gfs2_rbm_to_block(&rs->rs_rbm), | |
4a993fb1 | 618 | rs->rs_rbm.offset, rs->rs_free); |
8e2e0047 BP |
619 | } |
620 | ||
0a305e49 | 621 | /** |
8e2e0047 BP |
622 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
623 | * @rs: The reservation to remove | |
624 | * | |
625 | */ | |
20095218 | 626 | static void __rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
627 | { |
628 | struct gfs2_rgrpd *rgd; | |
629 | ||
630 | if (!gfs2_rs_active(rs)) | |
631 | return; | |
632 | ||
4a993fb1 | 633 | rgd = rs->rs_rbm.rgd; |
9e733d39 | 634 | trace_gfs2_rs(rs, TRACE_RS_TREEDEL); |
4a993fb1 | 635 | rb_erase(&rs->rs_node, &rgd->rd_rstree); |
24d634e8 | 636 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 BP |
637 | |
638 | if (rs->rs_free) { | |
e579ed4f BP |
639 | struct gfs2_bitmap *bi = rbm_bi(&rs->rs_rbm); |
640 | ||
20095218 | 641 | /* return reserved blocks to the rgrp */ |
4a993fb1 SW |
642 | BUG_ON(rs->rs_rbm.rgd->rd_reserved < rs->rs_free); |
643 | rs->rs_rbm.rgd->rd_reserved -= rs->rs_free; | |
5ea5050c BP |
644 | /* The rgrp extent failure point is likely not to increase; |
645 | it will only do so if the freed blocks are somehow | |
646 | contiguous with a span of free blocks that follows. Still, | |
647 | it will force the number to be recalculated later. */ | |
648 | rgd->rd_extfail_pt += rs->rs_free; | |
8e2e0047 | 649 | rs->rs_free = 0; |
e579ed4f | 650 | clear_bit(GBF_FULL, &bi->bi_flags); |
8e2e0047 | 651 | } |
8e2e0047 BP |
652 | } |
653 | ||
654 | /** | |
655 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
656 | * @rs: The reservation to remove | |
657 | * | |
658 | */ | |
20095218 | 659 | void gfs2_rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
660 | { |
661 | struct gfs2_rgrpd *rgd; | |
662 | ||
4a993fb1 SW |
663 | rgd = rs->rs_rbm.rgd; |
664 | if (rgd) { | |
665 | spin_lock(&rgd->rd_rsspin); | |
20095218 | 666 | __rs_deltree(rs); |
4a993fb1 SW |
667 | spin_unlock(&rgd->rd_rsspin); |
668 | } | |
8e2e0047 BP |
669 | } |
670 | ||
671 | /** | |
672 | * gfs2_rs_delete - delete a multi-block reservation | |
0a305e49 | 673 | * @ip: The inode for this reservation |
af5c2697 | 674 | * @wcount: The inode's write count, or NULL |
0a305e49 BP |
675 | * |
676 | */ | |
af5c2697 | 677 | void gfs2_rs_delete(struct gfs2_inode *ip, atomic_t *wcount) |
0a305e49 BP |
678 | { |
679 | down_write(&ip->i_rw_mutex); | |
af5c2697 | 680 | if (ip->i_res && ((wcount == NULL) || (atomic_read(wcount) <= 1))) { |
20095218 | 681 | gfs2_rs_deltree(ip->i_res); |
8e2e0047 | 682 | BUG_ON(ip->i_res->rs_free); |
0a305e49 BP |
683 | kmem_cache_free(gfs2_rsrv_cachep, ip->i_res); |
684 | ip->i_res = NULL; | |
685 | } | |
686 | up_write(&ip->i_rw_mutex); | |
687 | } | |
688 | ||
8e2e0047 BP |
689 | /** |
690 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
691 | * @rgd: the rgrp that needs its space back | |
692 | * | |
693 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
694 | * give them back. This leave the reservation structures in tact, but removes | |
695 | * all of their corresponding "no-fly zones". | |
696 | */ | |
697 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
698 | { | |
699 | struct rb_node *n; | |
700 | struct gfs2_blkreserv *rs; | |
701 | ||
702 | spin_lock(&rgd->rd_rsspin); | |
703 | while ((n = rb_first(&rgd->rd_rstree))) { | |
704 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
20095218 | 705 | __rs_deltree(rs); |
8e2e0047 BP |
706 | } |
707 | spin_unlock(&rgd->rd_rsspin); | |
708 | } | |
709 | ||
8339ee54 | 710 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 711 | { |
7c9ca621 | 712 | struct rb_node *n; |
b3b94faa DT |
713 | struct gfs2_rgrpd *rgd; |
714 | struct gfs2_glock *gl; | |
715 | ||
7c9ca621 BP |
716 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
717 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
718 | gl = rgd->rd_gl; |
719 | ||
7c9ca621 | 720 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
721 | |
722 | if (gl) { | |
8339ee54 | 723 | spin_lock(&gl->gl_spin); |
5c676f6d | 724 | gl->gl_object = NULL; |
8339ee54 | 725 | spin_unlock(&gl->gl_spin); |
29687a2a | 726 | gfs2_glock_add_to_lru(gl); |
b3b94faa DT |
727 | gfs2_glock_put(gl); |
728 | } | |
729 | ||
8339ee54 | 730 | gfs2_free_clones(rgd); |
b3b94faa | 731 | kfree(rgd->rd_bits); |
8e2e0047 | 732 | return_all_reservations(rgd); |
6bdd9be6 | 733 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
734 | } |
735 | } | |
736 | ||
bb8d8a6f SW |
737 | static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) |
738 | { | |
739 | printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); | |
740 | printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); | |
741 | printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); | |
742 | printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); | |
743 | printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); | |
744 | } | |
745 | ||
b3b94faa DT |
746 | /** |
747 | * gfs2_compute_bitstructs - Compute the bitmap sizes | |
748 | * @rgd: The resource group descriptor | |
749 | * | |
750 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
751 | * | |
752 | * Returns: errno | |
753 | */ | |
754 | ||
755 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
756 | { | |
757 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
758 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 759 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 760 | u32 bytes_left, bytes; |
b3b94faa DT |
761 | int x; |
762 | ||
feaa7bba SW |
763 | if (!length) |
764 | return -EINVAL; | |
765 | ||
dd894be8 | 766 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
767 | if (!rgd->rd_bits) |
768 | return -ENOMEM; | |
769 | ||
bb8d8a6f | 770 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
771 | |
772 | for (x = 0; x < length; x++) { | |
773 | bi = rgd->rd_bits + x; | |
774 | ||
60a0b8f9 | 775 | bi->bi_flags = 0; |
b3b94faa DT |
776 | /* small rgrp; bitmap stored completely in header block */ |
777 | if (length == 1) { | |
778 | bytes = bytes_left; | |
779 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
780 | bi->bi_start = 0; | |
781 | bi->bi_len = bytes; | |
7e230f57 | 782 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
783 | /* header block */ |
784 | } else if (x == 0) { | |
785 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
786 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
787 | bi->bi_start = 0; | |
788 | bi->bi_len = bytes; | |
7e230f57 | 789 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
790 | /* last block */ |
791 | } else if (x + 1 == length) { | |
792 | bytes = bytes_left; | |
793 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 794 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 795 | bi->bi_len = bytes; |
7e230f57 | 796 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
797 | /* other blocks */ |
798 | } else { | |
568f4c96 SW |
799 | bytes = sdp->sd_sb.sb_bsize - |
800 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 801 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 802 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
b3b94faa | 803 | bi->bi_len = bytes; |
7e230f57 | 804 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
805 | } |
806 | ||
807 | bytes_left -= bytes; | |
808 | } | |
809 | ||
810 | if (bytes_left) { | |
811 | gfs2_consist_rgrpd(rgd); | |
812 | return -EIO; | |
813 | } | |
814 | bi = rgd->rd_bits + (length - 1); | |
bb8d8a6f | 815 | if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { |
b3b94faa | 816 | if (gfs2_consist_rgrpd(rgd)) { |
bb8d8a6f | 817 | gfs2_rindex_print(rgd); |
b3b94faa DT |
818 | fs_err(sdp, "start=%u len=%u offset=%u\n", |
819 | bi->bi_start, bi->bi_len, bi->bi_offset); | |
820 | } | |
821 | return -EIO; | |
822 | } | |
823 | ||
824 | return 0; | |
825 | } | |
826 | ||
7ae8fa84 RP |
827 | /** |
828 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 829 | * @sdp: the filesystem |
7ae8fa84 RP |
830 | * |
831 | */ | |
832 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
833 | { | |
834 | u64 total_data = 0; | |
835 | struct inode *inode = sdp->sd_rindex; | |
836 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 837 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
838 | int error, rgrps; |
839 | ||
7ae8fa84 RP |
840 | for (rgrps = 0;; rgrps++) { |
841 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
842 | ||
bcd7278d | 843 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 844 | break; |
4306629e | 845 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
846 | sizeof(struct gfs2_rindex)); |
847 | if (error != sizeof(struct gfs2_rindex)) | |
848 | break; | |
bb8d8a6f | 849 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 850 | } |
7ae8fa84 RP |
851 | return total_data; |
852 | } | |
853 | ||
6aad1c3d | 854 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
855 | { |
856 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
857 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
858 | ||
859 | /* Figure out where to put new node */ | |
860 | while (*newn) { | |
861 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
862 | rd_node); | |
863 | ||
864 | parent = *newn; | |
865 | if (rgd->rd_addr < cur->rd_addr) | |
866 | newn = &((*newn)->rb_left); | |
867 | else if (rgd->rd_addr > cur->rd_addr) | |
868 | newn = &((*newn)->rb_right); | |
869 | else | |
6aad1c3d | 870 | return -EEXIST; |
7c9ca621 BP |
871 | } |
872 | ||
873 | rb_link_node(&rgd->rd_node, parent, newn); | |
874 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
875 | sdp->sd_rgrps++; |
876 | return 0; | |
7c9ca621 BP |
877 | } |
878 | ||
b3b94faa | 879 | /** |
6c53267f | 880 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 881 | * @ip: Pointer to the rindex inode |
b3b94faa | 882 | * |
8339ee54 | 883 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
884 | */ |
885 | ||
4306629e | 886 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
887 | { |
888 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
889 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
8339ee54 | 890 | struct gfs2_rindex buf; |
6c53267f RP |
891 | int error; |
892 | struct gfs2_rgrpd *rgd; | |
893 | ||
8339ee54 SW |
894 | if (pos >= i_size_read(&ip->i_inode)) |
895 | return 1; | |
896 | ||
4306629e | 897 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 898 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
899 | |
900 | if (error != sizeof(struct gfs2_rindex)) | |
901 | return (error == 0) ? 1 : error; | |
6c53267f | 902 | |
6bdd9be6 | 903 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
904 | error = -ENOMEM; |
905 | if (!rgd) | |
906 | return error; | |
907 | ||
6c53267f | 908 | rgd->rd_sbd = sdp; |
8339ee54 SW |
909 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
910 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
911 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
912 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
913 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 914 | spin_lock_init(&rgd->rd_rsspin); |
7c9ca621 | 915 | |
6c53267f RP |
916 | error = compute_bitstructs(rgd); |
917 | if (error) | |
8339ee54 | 918 | goto fail; |
6c53267f | 919 | |
bb8d8a6f | 920 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
921 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
922 | if (error) | |
8339ee54 | 923 | goto fail; |
6c53267f RP |
924 | |
925 | rgd->rd_gl->gl_object = rgd; | |
4e2f8849 | 926 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr; |
cf45b752 | 927 | rgd->rd_flags &= ~GFS2_RDF_UPTODATE; |
7c9ca621 BP |
928 | if (rgd->rd_data > sdp->sd_max_rg_data) |
929 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 930 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 931 | error = rgd_insert(rgd); |
8339ee54 | 932 | spin_unlock(&sdp->sd_rindex_spin); |
6aad1c3d BP |
933 | if (!error) |
934 | return 0; | |
935 | ||
936 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 937 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
938 | |
939 | fail: | |
940 | kfree(rgd->rd_bits); | |
941 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); | |
6c53267f RP |
942 | return error; |
943 | } | |
944 | ||
945 | /** | |
946 | * gfs2_ri_update - Pull in a new resource index from the disk | |
947 | * @ip: pointer to the rindex inode | |
948 | * | |
b3b94faa DT |
949 | * Returns: 0 on successful update, error code otherwise |
950 | */ | |
951 | ||
8339ee54 | 952 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 953 | { |
feaa7bba | 954 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
955 | int error; |
956 | ||
8339ee54 | 957 | do { |
4306629e | 958 | error = read_rindex_entry(ip); |
8339ee54 SW |
959 | } while (error == 0); |
960 | ||
961 | if (error < 0) | |
962 | return error; | |
b3b94faa | 963 | |
cf45b752 | 964 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
965 | return 0; |
966 | } | |
b3b94faa | 967 | |
b3b94faa | 968 | /** |
8339ee54 | 969 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 970 | * @sdp: The GFS2 superblock |
b3b94faa DT |
971 | * |
972 | * We grab a lock on the rindex inode to make sure that it doesn't | |
973 | * change whilst we are performing an operation. We keep this lock | |
974 | * for quite long periods of time compared to other locks. This | |
975 | * doesn't matter, since it is shared and it is very, very rarely | |
976 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
977 | * | |
978 | * This makes sure that we're using the latest copy of the resource index | |
979 | * special file, which might have been updated if someone expanded the | |
980 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
981 | * | |
8339ee54 | 982 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
983 | */ |
984 | ||
8339ee54 | 985 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 986 | { |
feaa7bba | 987 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 988 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
989 | struct gfs2_holder ri_gh; |
990 | int error = 0; | |
a365fbf3 | 991 | int unlock_required = 0; |
b3b94faa DT |
992 | |
993 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 994 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
995 | if (!gfs2_glock_is_locked_by_me(gl)) { |
996 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
997 | if (error) | |
6aad1c3d | 998 | return error; |
a365fbf3 SW |
999 | unlock_required = 1; |
1000 | } | |
8339ee54 | 1001 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 1002 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
1003 | if (unlock_required) |
1004 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
1005 | } |
1006 | ||
1007 | return error; | |
1008 | } | |
1009 | ||
42d52e38 | 1010 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
1011 | { |
1012 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 1013 | u32 rg_flags; |
bb8d8a6f | 1014 | |
42d52e38 | 1015 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 1016 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
1017 | rgd->rd_flags &= GFS2_RDF_MASK; |
1018 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 1019 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 1020 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 1021 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
bb8d8a6f SW |
1022 | } |
1023 | ||
42d52e38 | 1024 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f SW |
1025 | { |
1026 | struct gfs2_rgrp *str = buf; | |
1027 | ||
09010978 | 1028 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 1029 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 1030 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
bb8d8a6f | 1031 | str->__pad = cpu_to_be32(0); |
d8b71f73 | 1032 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
bb8d8a6f SW |
1033 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
1034 | } | |
1035 | ||
90306c41 BM |
1036 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
1037 | { | |
1038 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1039 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
1040 | ||
1041 | if (rgl->rl_flags != str->rg_flags || rgl->rl_free != str->rg_free || | |
1042 | rgl->rl_dinodes != str->rg_dinodes || | |
1043 | rgl->rl_igeneration != str->rg_igeneration) | |
1044 | return 0; | |
1045 | return 1; | |
1046 | } | |
1047 | ||
1048 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) | |
1049 | { | |
1050 | const struct gfs2_rgrp *str = buf; | |
1051 | ||
1052 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
1053 | rgl->rl_flags = str->rg_flags; | |
1054 | rgl->rl_free = str->rg_free; | |
1055 | rgl->rl_dinodes = str->rg_dinodes; | |
1056 | rgl->rl_igeneration = str->rg_igeneration; | |
1057 | rgl->__pad = 0UL; | |
1058 | } | |
1059 | ||
1060 | static void update_rgrp_lvb_unlinked(struct gfs2_rgrpd *rgd, u32 change) | |
1061 | { | |
1062 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1063 | u32 unlinked = be32_to_cpu(rgl->rl_unlinked) + change; | |
1064 | rgl->rl_unlinked = cpu_to_be32(unlinked); | |
1065 | } | |
1066 | ||
1067 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) | |
1068 | { | |
1069 | struct gfs2_bitmap *bi; | |
1070 | const u32 length = rgd->rd_length; | |
1071 | const u8 *buffer = NULL; | |
1072 | u32 i, goal, count = 0; | |
1073 | ||
1074 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
1075 | goal = 0; | |
1076 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1077 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
1078 | while (goal < bi->bi_len * GFS2_NBBY) { | |
1079 | goal = gfs2_bitfit(buffer, bi->bi_len, goal, | |
1080 | GFS2_BLKST_UNLINKED); | |
1081 | if (goal == BFITNOENT) | |
1082 | break; | |
1083 | count++; | |
1084 | goal++; | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | return count; | |
1089 | } | |
1090 | ||
1091 | ||
b3b94faa | 1092 | /** |
90306c41 BM |
1093 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
1094 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
1095 | * |
1096 | * Read in all of a Resource Group's header and bitmap blocks. | |
1097 | * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. | |
1098 | * | |
1099 | * Returns: errno | |
1100 | */ | |
1101 | ||
90306c41 | 1102 | int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
1103 | { |
1104 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1105 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 1106 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
1107 | struct gfs2_bitmap *bi; |
1108 | unsigned int x, y; | |
1109 | int error; | |
1110 | ||
90306c41 BM |
1111 | if (rgd->rd_bits[0].bi_bh != NULL) |
1112 | return 0; | |
1113 | ||
b3b94faa DT |
1114 | for (x = 0; x < length; x++) { |
1115 | bi = rgd->rd_bits + x; | |
bb8d8a6f | 1116 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); |
b3b94faa DT |
1117 | if (error) |
1118 | goto fail; | |
1119 | } | |
1120 | ||
1121 | for (y = length; y--;) { | |
1122 | bi = rgd->rd_bits + y; | |
7276b3b0 | 1123 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
1124 | if (error) |
1125 | goto fail; | |
feaa7bba | 1126 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
1127 | GFS2_METATYPE_RG)) { |
1128 | error = -EIO; | |
1129 | goto fail; | |
1130 | } | |
1131 | } | |
1132 | ||
cf45b752 | 1133 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
60a0b8f9 SW |
1134 | for (x = 0; x < length; x++) |
1135 | clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); | |
42d52e38 | 1136 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
1ce97e56 | 1137 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 1138 | rgd->rd_free_clone = rgd->rd_free; |
5ea5050c BP |
1139 | /* max out the rgrp allocation failure point */ |
1140 | rgd->rd_extfail_pt = rgd->rd_free; | |
b3b94faa | 1141 | } |
951b4bd5 | 1142 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
90306c41 BM |
1143 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); |
1144 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
1145 | rgd->rd_bits[0].bi_bh->b_data); | |
1146 | } | |
1147 | else if (sdp->sd_args.ar_rgrplvb) { | |
1148 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
1149 | gfs2_consist_rgrpd(rgd); | |
1150 | error = -EIO; | |
1151 | goto fail; | |
1152 | } | |
1153 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1154 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1155 | } | |
b3b94faa DT |
1156 | return 0; |
1157 | ||
feaa7bba | 1158 | fail: |
b3b94faa DT |
1159 | while (x--) { |
1160 | bi = rgd->rd_bits + x; | |
1161 | brelse(bi->bi_bh); | |
1162 | bi->bi_bh = NULL; | |
1163 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
1164 | } | |
b3b94faa DT |
1165 | |
1166 | return error; | |
1167 | } | |
1168 | ||
90306c41 BM |
1169 | int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
1170 | { | |
1171 | u32 rl_flags; | |
1172 | ||
1173 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1174 | return 0; | |
1175 | ||
951b4bd5 | 1176 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) |
90306c41 BM |
1177 | return gfs2_rgrp_bh_get(rgd); |
1178 | ||
1179 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1180 | rl_flags &= ~GFS2_RDF_MASK; | |
1181 | rgd->rd_flags &= GFS2_RDF_MASK; | |
1182 | rgd->rd_flags |= (rl_flags | GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); | |
1183 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1184 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1185 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
1186 | rgd->rd_free_clone = rgd->rd_free; | |
1187 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); | |
1188 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1189 | return 0; | |
1190 | } | |
1191 | ||
1192 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1193 | { | |
1194 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1195 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1196 | ||
1197 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1198 | return 0; | |
1199 | return gfs2_rgrp_bh_get((struct gfs2_rgrpd *)gh->gh_gl->gl_object); | |
1200 | } | |
1201 | ||
b3b94faa | 1202 | /** |
7c9ca621 | 1203 | * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
886b1416 | 1204 | * @gh: The glock holder for the resource group |
b3b94faa DT |
1205 | * |
1206 | */ | |
1207 | ||
7c9ca621 | 1208 | void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) |
b3b94faa | 1209 | { |
7c9ca621 | 1210 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; |
bb8d8a6f | 1211 | int x, length = rgd->rd_length; |
b3b94faa | 1212 | |
b3b94faa DT |
1213 | for (x = 0; x < length; x++) { |
1214 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1215 | if (bi->bi_bh) { |
1216 | brelse(bi->bi_bh); | |
1217 | bi->bi_bh = NULL; | |
1218 | } | |
b3b94faa DT |
1219 | } |
1220 | ||
b3b94faa DT |
1221 | } |
1222 | ||
66fc061b | 1223 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1224 | struct buffer_head *bh, |
66fc061b | 1225 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1226 | { |
1227 | struct super_block *sb = sdp->sd_vfs; | |
f15ab561 | 1228 | u64 blk; |
64d576ba | 1229 | sector_t start = 0; |
b2c87cae | 1230 | sector_t nr_blks = 0; |
f15ab561 SW |
1231 | int rv; |
1232 | unsigned int x; | |
66fc061b SW |
1233 | u32 trimmed = 0; |
1234 | u8 diff; | |
f15ab561 SW |
1235 | |
1236 | for (x = 0; x < bi->bi_len; x++) { | |
66fc061b SW |
1237 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1238 | clone += bi->bi_offset; | |
1239 | clone += x; | |
1240 | if (bh) { | |
1241 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1242 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1243 | } else { | |
1244 | diff = ~(*clone | (*clone >> 1)); | |
1245 | } | |
f15ab561 SW |
1246 | diff &= 0x55; |
1247 | if (diff == 0) | |
1248 | continue; | |
1249 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
f15ab561 SW |
1250 | while(diff) { |
1251 | if (diff & 1) { | |
b2c87cae | 1252 | if (nr_blks == 0) |
f15ab561 | 1253 | goto start_new_extent; |
b2c87cae BP |
1254 | if ((start + nr_blks) != blk) { |
1255 | if (nr_blks >= minlen) { | |
1256 | rv = sb_issue_discard(sb, | |
1257 | start, nr_blks, | |
66fc061b SW |
1258 | GFP_NOFS, 0); |
1259 | if (rv) | |
1260 | goto fail; | |
b2c87cae | 1261 | trimmed += nr_blks; |
66fc061b | 1262 | } |
b2c87cae | 1263 | nr_blks = 0; |
f15ab561 SW |
1264 | start_new_extent: |
1265 | start = blk; | |
1266 | } | |
b2c87cae | 1267 | nr_blks++; |
f15ab561 SW |
1268 | } |
1269 | diff >>= 2; | |
b2c87cae | 1270 | blk++; |
f15ab561 SW |
1271 | } |
1272 | } | |
b2c87cae BP |
1273 | if (nr_blks >= minlen) { |
1274 | rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0); | |
f15ab561 SW |
1275 | if (rv) |
1276 | goto fail; | |
b2c87cae | 1277 | trimmed += nr_blks; |
f15ab561 | 1278 | } |
66fc061b SW |
1279 | if (ptrimmed) |
1280 | *ptrimmed = trimmed; | |
1281 | return 0; | |
1282 | ||
f15ab561 | 1283 | fail: |
66fc061b SW |
1284 | if (sdp->sd_args.ar_discard) |
1285 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); | |
f15ab561 | 1286 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1287 | return -EIO; |
1288 | } | |
1289 | ||
1290 | /** | |
1291 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1292 | * @filp: Any file on the filesystem | |
1293 | * @argp: Pointer to the arguments (also used to pass result) | |
1294 | * | |
1295 | * Returns: 0 on success, otherwise error code | |
1296 | */ | |
1297 | ||
1298 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1299 | { | |
496ad9aa | 1300 | struct inode *inode = file_inode(filp); |
66fc061b SW |
1301 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1302 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1303 | struct buffer_head *bh; | |
1304 | struct gfs2_rgrpd *rgd; | |
1305 | struct gfs2_rgrpd *rgd_end; | |
1306 | struct gfs2_holder gh; | |
1307 | struct fstrim_range r; | |
1308 | int ret = 0; | |
1309 | u64 amt; | |
1310 | u64 trimmed = 0; | |
076f0faa | 1311 | u64 start, end, minlen; |
66fc061b | 1312 | unsigned int x; |
076f0faa | 1313 | unsigned bs_shift = sdp->sd_sb.sb_bsize_shift; |
66fc061b SW |
1314 | |
1315 | if (!capable(CAP_SYS_ADMIN)) | |
1316 | return -EPERM; | |
1317 | ||
1318 | if (!blk_queue_discard(q)) | |
1319 | return -EOPNOTSUPP; | |
1320 | ||
3a238ade | 1321 | if (copy_from_user(&r, argp, sizeof(r))) |
66fc061b SW |
1322 | return -EFAULT; |
1323 | ||
5e2f7d61 BP |
1324 | ret = gfs2_rindex_update(sdp); |
1325 | if (ret) | |
1326 | return ret; | |
1327 | ||
076f0faa LC |
1328 | start = r.start >> bs_shift; |
1329 | end = start + (r.len >> bs_shift); | |
1330 | minlen = max_t(u64, r.minlen, | |
1331 | q->limits.discard_granularity) >> bs_shift; | |
1332 | ||
6a98c333 AD |
1333 | if (end <= start || minlen > sdp->sd_max_rg_data) |
1334 | return -EINVAL; | |
1335 | ||
076f0faa | 1336 | rgd = gfs2_blk2rgrpd(sdp, start, 0); |
6a98c333 | 1337 | rgd_end = gfs2_blk2rgrpd(sdp, end, 0); |
076f0faa | 1338 | |
6a98c333 AD |
1339 | if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end)) |
1340 | && (start > rgd_end->rd_data0 + rgd_end->rd_data)) | |
1341 | return -EINVAL; /* start is beyond the end of the fs */ | |
66fc061b SW |
1342 | |
1343 | while (1) { | |
1344 | ||
1345 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1346 | if (ret) | |
1347 | goto out; | |
1348 | ||
1349 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1350 | /* Trim each bitmap in the rgrp */ | |
1351 | for (x = 0; x < rgd->rd_length; x++) { | |
1352 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
076f0faa LC |
1353 | ret = gfs2_rgrp_send_discards(sdp, |
1354 | rgd->rd_data0, NULL, bi, minlen, | |
1355 | &amt); | |
66fc061b SW |
1356 | if (ret) { |
1357 | gfs2_glock_dq_uninit(&gh); | |
1358 | goto out; | |
1359 | } | |
1360 | trimmed += amt; | |
1361 | } | |
1362 | ||
1363 | /* Mark rgrp as having been trimmed */ | |
1364 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1365 | if (ret == 0) { | |
1366 | bh = rgd->rd_bits[0].bi_bh; | |
1367 | rgd->rd_flags |= GFS2_RGF_TRIMMED; | |
350a9b0a | 1368 | gfs2_trans_add_meta(rgd->rd_gl, bh); |
66fc061b | 1369 | gfs2_rgrp_out(rgd, bh->b_data); |
90306c41 | 1370 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, bh->b_data); |
66fc061b SW |
1371 | gfs2_trans_end(sdp); |
1372 | } | |
1373 | } | |
1374 | gfs2_glock_dq_uninit(&gh); | |
1375 | ||
1376 | if (rgd == rgd_end) | |
1377 | break; | |
1378 | ||
1379 | rgd = gfs2_rgrpd_get_next(rgd); | |
1380 | } | |
1381 | ||
1382 | out: | |
6a98c333 | 1383 | r.len = trimmed << bs_shift; |
3a238ade | 1384 | if (copy_to_user(argp, &r, sizeof(r))) |
66fc061b SW |
1385 | return -EFAULT; |
1386 | ||
1387 | return ret; | |
f15ab561 SW |
1388 | } |
1389 | ||
8e2e0047 BP |
1390 | /** |
1391 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
8e2e0047 | 1392 | * @ip: the inode structure |
8e2e0047 | 1393 | * |
8e2e0047 | 1394 | */ |
ff7f4cb4 | 1395 | static void rs_insert(struct gfs2_inode *ip) |
8e2e0047 BP |
1396 | { |
1397 | struct rb_node **newn, *parent = NULL; | |
1398 | int rc; | |
1399 | struct gfs2_blkreserv *rs = ip->i_res; | |
4a993fb1 | 1400 | struct gfs2_rgrpd *rgd = rs->rs_rbm.rgd; |
ff7f4cb4 | 1401 | u64 fsblock = gfs2_rbm_to_block(&rs->rs_rbm); |
8e2e0047 | 1402 | |
8e2e0047 | 1403 | BUG_ON(gfs2_rs_active(rs)); |
c743ffd0 | 1404 | |
ff7f4cb4 SW |
1405 | spin_lock(&rgd->rd_rsspin); |
1406 | newn = &rgd->rd_rstree.rb_node; | |
8e2e0047 BP |
1407 | while (*newn) { |
1408 | struct gfs2_blkreserv *cur = | |
1409 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1410 | ||
1411 | parent = *newn; | |
ff7f4cb4 | 1412 | rc = rs_cmp(fsblock, rs->rs_free, cur); |
8e2e0047 BP |
1413 | if (rc > 0) |
1414 | newn = &((*newn)->rb_right); | |
1415 | else if (rc < 0) | |
1416 | newn = &((*newn)->rb_left); | |
1417 | else { | |
1418 | spin_unlock(&rgd->rd_rsspin); | |
ff7f4cb4 SW |
1419 | WARN_ON(1); |
1420 | return; | |
8e2e0047 BP |
1421 | } |
1422 | } | |
1423 | ||
8e2e0047 BP |
1424 | rb_link_node(&rs->rs_node, parent, newn); |
1425 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1426 | ||
8e2e0047 | 1427 | /* Do our rgrp accounting for the reservation */ |
ff7f4cb4 | 1428 | rgd->rd_reserved += rs->rs_free; /* blocks reserved */ |
8e2e0047 | 1429 | spin_unlock(&rgd->rd_rsspin); |
9e733d39 | 1430 | trace_gfs2_rs(rs, TRACE_RS_INSERT); |
8e2e0047 BP |
1431 | } |
1432 | ||
1433 | /** | |
ff7f4cb4 | 1434 | * rg_mblk_search - find a group of multiple free blocks to form a reservation |
8e2e0047 | 1435 | * @rgd: the resource group descriptor |
8e2e0047 | 1436 | * @ip: pointer to the inode for which we're reserving blocks |
7b9cff46 | 1437 | * @ap: the allocation parameters |
8e2e0047 | 1438 | * |
8e2e0047 BP |
1439 | */ |
1440 | ||
ff7f4cb4 | 1441 | static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
7b9cff46 | 1442 | const struct gfs2_alloc_parms *ap) |
8e2e0047 | 1443 | { |
ff7f4cb4 SW |
1444 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1445 | u64 goal; | |
1446 | struct gfs2_blkreserv *rs = ip->i_res; | |
1447 | u32 extlen; | |
1448 | u32 free_blocks = rgd->rd_free_clone - rgd->rd_reserved; | |
1449 | int ret; | |
af21ca8e | 1450 | struct inode *inode = &ip->i_inode; |
8e2e0047 | 1451 | |
af21ca8e BP |
1452 | if (S_ISDIR(inode->i_mode)) |
1453 | extlen = 1; | |
1454 | else { | |
7b9cff46 | 1455 | extlen = max_t(u32, atomic_read(&rs->rs_sizehint), ap->target); |
af21ca8e BP |
1456 | extlen = clamp(extlen, RGRP_RSRV_MINBLKS, free_blocks); |
1457 | } | |
ff7f4cb4 | 1458 | if ((rgd->rd_free_clone < rgd->rd_reserved) || (free_blocks < extlen)) |
c743ffd0 SW |
1459 | return; |
1460 | ||
8e2e0047 BP |
1461 | /* Find bitmap block that contains bits for goal block */ |
1462 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
ff7f4cb4 | 1463 | goal = ip->i_goal; |
8e2e0047 | 1464 | else |
ff7f4cb4 | 1465 | goal = rgd->rd_last_alloc + rgd->rd_data0; |
8e2e0047 | 1466 | |
ff7f4cb4 SW |
1467 | if (WARN_ON(gfs2_rbm_from_block(&rbm, goal))) |
1468 | return; | |
8e2e0047 | 1469 | |
5ce13431 | 1470 | ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, ip, true, ap); |
ff7f4cb4 SW |
1471 | if (ret == 0) { |
1472 | rs->rs_rbm = rbm; | |
1473 | rs->rs_free = extlen; | |
1474 | rs->rs_inum = ip->i_no_addr; | |
1475 | rs_insert(ip); | |
13d2eb01 BP |
1476 | } else { |
1477 | if (goal == rgd->rd_last_alloc + rgd->rd_data0) | |
1478 | rgd->rd_last_alloc = 0; | |
8e2e0047 | 1479 | } |
b3e47ca0 BP |
1480 | } |
1481 | ||
5b924ae2 SW |
1482 | /** |
1483 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1484 | * @rgd: The resource group | |
1485 | * @block: The starting block | |
ff7f4cb4 | 1486 | * @length: The required length |
5b924ae2 SW |
1487 | * @ip: Ignore any reservations for this inode |
1488 | * | |
1489 | * If the block does not appear in any reservation, then return the | |
1490 | * block number unchanged. If it does appear in the reservation, then | |
1491 | * keep looking through the tree of reservations in order to find the | |
1492 | * first block number which is not reserved. | |
1493 | */ | |
1494 | ||
1495 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
ff7f4cb4 | 1496 | u32 length, |
5b924ae2 SW |
1497 | const struct gfs2_inode *ip) |
1498 | { | |
1499 | struct gfs2_blkreserv *rs; | |
1500 | struct rb_node *n; | |
1501 | int rc; | |
1502 | ||
1503 | spin_lock(&rgd->rd_rsspin); | |
ff7f4cb4 | 1504 | n = rgd->rd_rstree.rb_node; |
5b924ae2 SW |
1505 | while (n) { |
1506 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
ff7f4cb4 | 1507 | rc = rs_cmp(block, length, rs); |
5b924ae2 SW |
1508 | if (rc < 0) |
1509 | n = n->rb_left; | |
1510 | else if (rc > 0) | |
1511 | n = n->rb_right; | |
1512 | else | |
1513 | break; | |
1514 | } | |
1515 | ||
1516 | if (n) { | |
ff7f4cb4 | 1517 | while ((rs_cmp(block, length, rs) == 0) && (ip->i_res != rs)) { |
5b924ae2 | 1518 | block = gfs2_rbm_to_block(&rs->rs_rbm) + rs->rs_free; |
ff7f4cb4 | 1519 | n = n->rb_right; |
5b924ae2 SW |
1520 | if (n == NULL) |
1521 | break; | |
1522 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1523 | } | |
1524 | } | |
1525 | ||
1526 | spin_unlock(&rgd->rd_rsspin); | |
1527 | return block; | |
1528 | } | |
1529 | ||
5b924ae2 SW |
1530 | /** |
1531 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1532 | * @rbm: The current position in the resource group | |
ff7f4cb4 SW |
1533 | * @ip: The inode for which we are searching for blocks |
1534 | * @minext: The minimum extent length | |
5ce13431 | 1535 | * @maxext: A pointer to the maximum extent structure |
5b924ae2 SW |
1536 | * |
1537 | * This checks the current position in the rgrp to see whether there is | |
1538 | * a reservation covering this block. If not then this function is a | |
1539 | * no-op. If there is, then the position is moved to the end of the | |
1540 | * contiguous reservation(s) so that we are pointing at the first | |
1541 | * non-reserved block. | |
1542 | * | |
1543 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1544 | */ | |
1545 | ||
1546 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
ff7f4cb4 | 1547 | const struct gfs2_inode *ip, |
5ce13431 BP |
1548 | u32 minext, |
1549 | struct gfs2_extent *maxext) | |
5b924ae2 SW |
1550 | { |
1551 | u64 block = gfs2_rbm_to_block(rbm); | |
ff7f4cb4 | 1552 | u32 extlen = 1; |
5b924ae2 SW |
1553 | u64 nblock; |
1554 | int ret; | |
1555 | ||
ff7f4cb4 SW |
1556 | /* |
1557 | * If we have a minimum extent length, then skip over any extent | |
1558 | * which is less than the min extent length in size. | |
1559 | */ | |
1560 | if (minext) { | |
1561 | extlen = gfs2_free_extlen(rbm, minext); | |
5ce13431 | 1562 | if (extlen <= maxext->len) |
ff7f4cb4 SW |
1563 | goto fail; |
1564 | } | |
1565 | ||
1566 | /* | |
1567 | * Check the extent which has been found against the reservations | |
1568 | * and skip if parts of it are already reserved | |
1569 | */ | |
1570 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, ip); | |
5ce13431 BP |
1571 | if (nblock == block) { |
1572 | if (!minext || extlen >= minext) | |
1573 | return 0; | |
1574 | ||
1575 | if (extlen > maxext->len) { | |
1576 | maxext->len = extlen; | |
1577 | maxext->rbm = *rbm; | |
1578 | } | |
ff7f4cb4 | 1579 | fail: |
5ce13431 BP |
1580 | nblock = block + extlen; |
1581 | } | |
5b924ae2 SW |
1582 | ret = gfs2_rbm_from_block(rbm, nblock); |
1583 | if (ret < 0) | |
1584 | return ret; | |
1585 | return 1; | |
1586 | } | |
1587 | ||
1588 | /** | |
1589 | * gfs2_rbm_find - Look for blocks of a particular state | |
1590 | * @rbm: Value/result starting position and final position | |
1591 | * @state: The state which we want to find | |
5ce13431 BP |
1592 | * @minext: Pointer to the requested extent length (NULL for a single block) |
1593 | * This is updated to be the actual reservation size. | |
5b924ae2 SW |
1594 | * @ip: If set, check for reservations |
1595 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping | |
1596 | * around until we've reached the starting point. | |
5ce13431 | 1597 | * @ap: the allocation parameters |
5b924ae2 SW |
1598 | * |
1599 | * Side effects: | |
1600 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1601 | * has no free blocks in it. | |
5ea5050c BP |
1602 | * - If looking for free blocks, we set rd_extfail_pt on each rgrp which |
1603 | * has come up short on a free block search. | |
5b924ae2 SW |
1604 | * |
1605 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1606 | */ | |
1607 | ||
5ce13431 BP |
1608 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
1609 | const struct gfs2_inode *ip, bool nowrap, | |
1610 | const struct gfs2_alloc_parms *ap) | |
5b924ae2 SW |
1611 | { |
1612 | struct buffer_head *bh; | |
e579ed4f | 1613 | int initial_bii; |
5b924ae2 | 1614 | u32 initial_offset; |
5ea5050c BP |
1615 | int first_bii = rbm->bii; |
1616 | u32 first_offset = rbm->offset; | |
5b924ae2 SW |
1617 | u32 offset; |
1618 | u8 *buffer; | |
5b924ae2 SW |
1619 | int n = 0; |
1620 | int iters = rbm->rgd->rd_length; | |
1621 | int ret; | |
e579ed4f | 1622 | struct gfs2_bitmap *bi; |
5ce13431 | 1623 | struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, }; |
5b924ae2 SW |
1624 | |
1625 | /* If we are not starting at the beginning of a bitmap, then we | |
1626 | * need to add one to the bitmap count to ensure that we search | |
1627 | * the starting bitmap twice. | |
1628 | */ | |
1629 | if (rbm->offset != 0) | |
1630 | iters++; | |
1631 | ||
1632 | while(1) { | |
e579ed4f BP |
1633 | bi = rbm_bi(rbm); |
1634 | if (test_bit(GBF_FULL, &bi->bi_flags) && | |
5b924ae2 SW |
1635 | (state == GFS2_BLKST_FREE)) |
1636 | goto next_bitmap; | |
1637 | ||
e579ed4f BP |
1638 | bh = bi->bi_bh; |
1639 | buffer = bh->b_data + bi->bi_offset; | |
5b924ae2 | 1640 | WARN_ON(!buffer_uptodate(bh)); |
e579ed4f BP |
1641 | if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
1642 | buffer = bi->bi_clone + bi->bi_offset; | |
5b924ae2 | 1643 | initial_offset = rbm->offset; |
e579ed4f | 1644 | offset = gfs2_bitfit(buffer, bi->bi_len, rbm->offset, state); |
5b924ae2 SW |
1645 | if (offset == BFITNOENT) |
1646 | goto bitmap_full; | |
1647 | rbm->offset = offset; | |
1648 | if (ip == NULL) | |
1649 | return 0; | |
1650 | ||
e579ed4f | 1651 | initial_bii = rbm->bii; |
5ce13431 BP |
1652 | ret = gfs2_reservation_check_and_update(rbm, ip, |
1653 | minext ? *minext : 0, | |
1654 | &maxext); | |
5b924ae2 SW |
1655 | if (ret == 0) |
1656 | return 0; | |
1657 | if (ret > 0) { | |
e579ed4f | 1658 | n += (rbm->bii - initial_bii); |
8d8b752a | 1659 | goto next_iter; |
5b924ae2 | 1660 | } |
5d50d532 | 1661 | if (ret == -E2BIG) { |
e579ed4f | 1662 | rbm->bii = 0; |
5d50d532 | 1663 | rbm->offset = 0; |
e579ed4f | 1664 | n += (rbm->bii - initial_bii); |
5d50d532 SW |
1665 | goto res_covered_end_of_rgrp; |
1666 | } | |
5b924ae2 SW |
1667 | return ret; |
1668 | ||
1669 | bitmap_full: /* Mark bitmap as full and fall through */ | |
e579ed4f BP |
1670 | if ((state == GFS2_BLKST_FREE) && initial_offset == 0) { |
1671 | struct gfs2_bitmap *bi = rbm_bi(rbm); | |
1672 | set_bit(GBF_FULL, &bi->bi_flags); | |
1673 | } | |
5b924ae2 SW |
1674 | |
1675 | next_bitmap: /* Find next bitmap in the rgrp */ | |
1676 | rbm->offset = 0; | |
e579ed4f BP |
1677 | rbm->bii++; |
1678 | if (rbm->bii == rbm->rgd->rd_length) | |
1679 | rbm->bii = 0; | |
5d50d532 | 1680 | res_covered_end_of_rgrp: |
e579ed4f | 1681 | if ((rbm->bii == 0) && nowrap) |
5b924ae2 SW |
1682 | break; |
1683 | n++; | |
8d8b752a | 1684 | next_iter: |
5b924ae2 SW |
1685 | if (n >= iters) |
1686 | break; | |
1687 | } | |
1688 | ||
5ce13431 BP |
1689 | if (minext == NULL || state != GFS2_BLKST_FREE) |
1690 | return -ENOSPC; | |
1691 | ||
5ea5050c BP |
1692 | /* If the extent was too small, and it's smaller than the smallest |
1693 | to have failed before, remember for future reference that it's | |
1694 | useless to search this rgrp again for this amount or more. */ | |
1695 | if ((first_offset == 0) && (first_bii == 0) && | |
1696 | (*minext < rbm->rgd->rd_extfail_pt)) | |
1697 | rbm->rgd->rd_extfail_pt = *minext; | |
1698 | ||
5ce13431 BP |
1699 | /* If the maximum extent we found is big enough to fulfill the |
1700 | minimum requirements, use it anyway. */ | |
1701 | if (maxext.len) { | |
1702 | *rbm = maxext.rbm; | |
1703 | *minext = maxext.len; | |
1704 | return 0; | |
1705 | } | |
1706 | ||
5b924ae2 SW |
1707 | return -ENOSPC; |
1708 | } | |
1709 | ||
c8cdf479 SW |
1710 | /** |
1711 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1712 | * @rgd: The rgrp | |
886b1416 BP |
1713 | * @last_unlinked: block address of the last dinode we unlinked |
1714 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1715 | * |
1a0eae88 BP |
1716 | * Returns: 0 if no error |
1717 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1718 | */ |
1719 | ||
044b9414 | 1720 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1721 | { |
5b924ae2 | 1722 | u64 block; |
5f3eae75 | 1723 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1724 | struct gfs2_glock *gl; |
1725 | struct gfs2_inode *ip; | |
1726 | int error; | |
1727 | int found = 0; | |
e579ed4f | 1728 | struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 }; |
c8cdf479 | 1729 | |
5b924ae2 | 1730 | while (1) { |
5f3eae75 | 1731 | down_write(&sdp->sd_log_flush_lock); |
5ce13431 BP |
1732 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL, |
1733 | true, NULL); | |
5f3eae75 | 1734 | up_write(&sdp->sd_log_flush_lock); |
5b924ae2 SW |
1735 | if (error == -ENOSPC) |
1736 | break; | |
1737 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1738 | break; |
b3e47ca0 | 1739 | |
5b924ae2 SW |
1740 | block = gfs2_rbm_to_block(&rbm); |
1741 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1742 | break; | |
1743 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1744 | continue; |
5b924ae2 | 1745 | if (block == skip) |
1e19a195 | 1746 | continue; |
5b924ae2 | 1747 | *last_unlinked = block; |
044b9414 | 1748 | |
5b924ae2 | 1749 | error = gfs2_glock_get(sdp, block, &gfs2_inode_glops, CREATE, &gl); |
044b9414 SW |
1750 | if (error) |
1751 | continue; | |
1752 | ||
1753 | /* If the inode is already in cache, we can ignore it here | |
1754 | * because the existing inode disposal code will deal with | |
1755 | * it when all refs have gone away. Accessing gl_object like | |
1756 | * this is not safe in general. Here it is ok because we do | |
1757 | * not dereference the pointer, and we only need an approx | |
1758 | * answer to whether it is NULL or not. | |
1759 | */ | |
1760 | ip = gl->gl_object; | |
1761 | ||
1762 | if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) | |
1763 | gfs2_glock_put(gl); | |
1764 | else | |
1765 | found++; | |
1766 | ||
1767 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1768 | if (found > NR_CPUS) |
044b9414 | 1769 | return; |
c8cdf479 SW |
1770 | } |
1771 | ||
1772 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1773 | return; |
c8cdf479 SW |
1774 | } |
1775 | ||
bcd97c06 SW |
1776 | /** |
1777 | * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested | |
1778 | * @rgd: The rgrp in question | |
1779 | * @loops: An indication of how picky we can be (0=very, 1=less so) | |
1780 | * | |
1781 | * This function uses the recently added glock statistics in order to | |
1782 | * figure out whether a parciular resource group is suffering from | |
1783 | * contention from multiple nodes. This is done purely on the basis | |
1784 | * of timings, since this is the only data we have to work with and | |
1785 | * our aim here is to reject a resource group which is highly contended | |
1786 | * but (very important) not to do this too often in order to ensure that | |
1787 | * we do not land up introducing fragmentation by changing resource | |
1788 | * groups when not actually required. | |
1789 | * | |
1790 | * The calculation is fairly simple, we want to know whether the SRTTB | |
1791 | * (i.e. smoothed round trip time for blocking operations) to acquire | |
1792 | * the lock for this rgrp's glock is significantly greater than the | |
1793 | * time taken for resource groups on average. We introduce a margin in | |
1794 | * the form of the variable @var which is computed as the sum of the two | |
1795 | * respective variences, and multiplied by a factor depending on @loops | |
1796 | * and whether we have a lot of data to base the decision on. This is | |
1797 | * then tested against the square difference of the means in order to | |
1798 | * decide whether the result is statistically significant or not. | |
1799 | * | |
1800 | * Returns: A boolean verdict on the congestion status | |
1801 | */ | |
1802 | ||
1803 | static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops) | |
1804 | { | |
1805 | const struct gfs2_glock *gl = rgd->rd_gl; | |
1806 | const struct gfs2_sbd *sdp = gl->gl_sbd; | |
1807 | struct gfs2_lkstats *st; | |
1808 | s64 r_dcount, l_dcount; | |
1809 | s64 r_srttb, l_srttb; | |
1810 | s64 srttb_diff; | |
1811 | s64 sqr_diff; | |
1812 | s64 var; | |
1813 | ||
1814 | preempt_disable(); | |
1815 | st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP]; | |
1816 | r_srttb = st->stats[GFS2_LKS_SRTTB]; | |
1817 | r_dcount = st->stats[GFS2_LKS_DCOUNT]; | |
1818 | var = st->stats[GFS2_LKS_SRTTVARB] + | |
1819 | gl->gl_stats.stats[GFS2_LKS_SRTTVARB]; | |
1820 | preempt_enable(); | |
1821 | ||
1822 | l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB]; | |
1823 | l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT]; | |
1824 | ||
1825 | if ((l_dcount < 1) || (r_dcount < 1) || (r_srttb == 0)) | |
1826 | return false; | |
1827 | ||
1828 | srttb_diff = r_srttb - l_srttb; | |
1829 | sqr_diff = srttb_diff * srttb_diff; | |
1830 | ||
1831 | var *= 2; | |
1832 | if (l_dcount < 8 || r_dcount < 8) | |
1833 | var *= 2; | |
1834 | if (loops == 1) | |
1835 | var *= 2; | |
1836 | ||
1837 | return ((srttb_diff < 0) && (sqr_diff > var)); | |
1838 | } | |
1839 | ||
1840 | /** | |
1841 | * gfs2_rgrp_used_recently | |
1842 | * @rs: The block reservation with the rgrp to test | |
1843 | * @msecs: The time limit in milliseconds | |
1844 | * | |
1845 | * Returns: True if the rgrp glock has been used within the time limit | |
1846 | */ | |
1847 | static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs, | |
1848 | u64 msecs) | |
1849 | { | |
1850 | u64 tdiff; | |
1851 | ||
1852 | tdiff = ktime_to_ns(ktime_sub(ktime_get_real(), | |
1853 | rs->rs_rbm.rgd->rd_gl->gl_dstamp)); | |
1854 | ||
1855 | return tdiff > (msecs * 1000 * 1000); | |
1856 | } | |
1857 | ||
9dbe9610 SW |
1858 | static u32 gfs2_orlov_skip(const struct gfs2_inode *ip) |
1859 | { | |
1860 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
1861 | u32 skip; | |
1862 | ||
1863 | get_random_bytes(&skip, sizeof(skip)); | |
1864 | return skip % sdp->sd_rgrps; | |
1865 | } | |
1866 | ||
c743ffd0 SW |
1867 | static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin) |
1868 | { | |
1869 | struct gfs2_rgrpd *rgd = *pos; | |
aa8920c9 | 1870 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
c743ffd0 SW |
1871 | |
1872 | rgd = gfs2_rgrpd_get_next(rgd); | |
1873 | if (rgd == NULL) | |
aa8920c9 | 1874 | rgd = gfs2_rgrpd_get_first(sdp); |
c743ffd0 SW |
1875 | *pos = rgd; |
1876 | if (rgd != begin) /* If we didn't wrap */ | |
1877 | return true; | |
1878 | return false; | |
1879 | } | |
1880 | ||
b3b94faa | 1881 | /** |
666d1d8a | 1882 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 1883 | * @ip: the inode to reserve space for |
7b9cff46 | 1884 | * @ap: the allocation parameters |
b3b94faa DT |
1885 | * |
1886 | * Returns: errno | |
1887 | */ | |
1888 | ||
7b9cff46 | 1889 | int gfs2_inplace_reserve(struct gfs2_inode *ip, const struct gfs2_alloc_parms *ap) |
b3b94faa | 1890 | { |
feaa7bba | 1891 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 1892 | struct gfs2_rgrpd *begin = NULL; |
564e12b1 | 1893 | struct gfs2_blkreserv *rs = ip->i_res; |
bcd97c06 | 1894 | int error = 0, rg_locked, flags = 0; |
666d1d8a | 1895 | u64 last_unlinked = NO_BLOCK; |
7c9ca621 | 1896 | int loops = 0; |
9dbe9610 | 1897 | u32 skip = 0; |
b3b94faa | 1898 | |
90306c41 BM |
1899 | if (sdp->sd_args.ar_rgrplvb) |
1900 | flags |= GL_SKIP; | |
7b9cff46 | 1901 | if (gfs2_assert_warn(sdp, ap->target)) |
c743ffd0 | 1902 | return -EINVAL; |
8e2e0047 | 1903 | if (gfs2_rs_active(rs)) { |
4a993fb1 | 1904 | begin = rs->rs_rbm.rgd; |
8e2e0047 | 1905 | } else if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) { |
4a993fb1 | 1906 | rs->rs_rbm.rgd = begin = ip->i_rgd; |
8e2e0047 | 1907 | } else { |
4a993fb1 | 1908 | rs->rs_rbm.rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 1909 | } |
7b9cff46 | 1910 | if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV)) |
9dbe9610 | 1911 | skip = gfs2_orlov_skip(ip); |
4a993fb1 | 1912 | if (rs->rs_rbm.rgd == NULL) |
7c9ca621 BP |
1913 | return -EBADSLT; |
1914 | ||
1915 | while (loops < 3) { | |
c743ffd0 SW |
1916 | rg_locked = 1; |
1917 | ||
1918 | if (!gfs2_glock_is_locked_by_me(rs->rs_rbm.rgd->rd_gl)) { | |
1919 | rg_locked = 0; | |
9dbe9610 SW |
1920 | if (skip && skip--) |
1921 | goto next_rgrp; | |
bcd97c06 SW |
1922 | if (!gfs2_rs_active(rs) && (loops < 2) && |
1923 | gfs2_rgrp_used_recently(rs, 1000) && | |
1924 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
1925 | goto next_rgrp; | |
4a993fb1 | 1926 | error = gfs2_glock_nq_init(rs->rs_rbm.rgd->rd_gl, |
8e2e0047 BP |
1927 | LM_ST_EXCLUSIVE, flags, |
1928 | &rs->rs_rgd_gh); | |
c743ffd0 SW |
1929 | if (unlikely(error)) |
1930 | return error; | |
bcd97c06 SW |
1931 | if (!gfs2_rs_active(rs) && (loops < 2) && |
1932 | gfs2_rgrp_congested(rs->rs_rbm.rgd, loops)) | |
1933 | goto skip_rgrp; | |
c743ffd0 | 1934 | if (sdp->sd_args.ar_rgrplvb) { |
4a993fb1 | 1935 | error = update_rgrp_lvb(rs->rs_rbm.rgd); |
c743ffd0 | 1936 | if (unlikely(error)) { |
90306c41 BM |
1937 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); |
1938 | return error; | |
1939 | } | |
1940 | } | |
292c8c14 | 1941 | } |
666d1d8a | 1942 | |
c743ffd0 | 1943 | /* Skip unuseable resource groups */ |
5ea5050c BP |
1944 | if ((rs->rs_rbm.rgd->rd_flags & (GFS2_RGF_NOALLOC | |
1945 | GFS2_RDF_ERROR)) || | |
1946 | (ap && (ap->target > rs->rs_rbm.rgd->rd_extfail_pt))) | |
c743ffd0 SW |
1947 | goto skip_rgrp; |
1948 | ||
1949 | if (sdp->sd_args.ar_rgrplvb) | |
1950 | gfs2_rgrp_bh_get(rs->rs_rbm.rgd); | |
1951 | ||
1952 | /* Get a reservation if we don't already have one */ | |
1953 | if (!gfs2_rs_active(rs)) | |
7b9cff46 | 1954 | rg_mblk_search(rs->rs_rbm.rgd, ip, ap); |
c743ffd0 SW |
1955 | |
1956 | /* Skip rgrps when we can't get a reservation on first pass */ | |
1957 | if (!gfs2_rs_active(rs) && (loops < 1)) | |
1958 | goto check_rgrp; | |
1959 | ||
1960 | /* If rgrp has enough free space, use it */ | |
7b9cff46 | 1961 | if (rs->rs_rbm.rgd->rd_free_clone >= ap->target) { |
c743ffd0 SW |
1962 | ip->i_rgd = rs->rs_rbm.rgd; |
1963 | return 0; | |
b3b94faa | 1964 | } |
c743ffd0 | 1965 | |
c743ffd0 SW |
1966 | check_rgrp: |
1967 | /* Check for unlinked inodes which can be reclaimed */ | |
1968 | if (rs->rs_rbm.rgd->rd_flags & GFS2_RDF_CHECK) | |
1969 | try_rgrp_unlink(rs->rs_rbm.rgd, &last_unlinked, | |
1970 | ip->i_no_addr); | |
1971 | skip_rgrp: | |
1330edbe BP |
1972 | /* Drop reservation, if we couldn't use reserved rgrp */ |
1973 | if (gfs2_rs_active(rs)) | |
1974 | gfs2_rs_deltree(rs); | |
1975 | ||
c743ffd0 SW |
1976 | /* Unlock rgrp if required */ |
1977 | if (!rg_locked) | |
1978 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
1979 | next_rgrp: | |
1980 | /* Find the next rgrp, and continue looking */ | |
1981 | if (gfs2_select_rgrp(&rs->rs_rbm.rgd, begin)) | |
1982 | continue; | |
9dbe9610 SW |
1983 | if (skip) |
1984 | continue; | |
c743ffd0 SW |
1985 | |
1986 | /* If we've scanned all the rgrps, but found no free blocks | |
1987 | * then this checks for some less likely conditions before | |
1988 | * trying again. | |
1989 | */ | |
c743ffd0 SW |
1990 | loops++; |
1991 | /* Check that fs hasn't grown if writing to rindex */ | |
1992 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
1993 | error = gfs2_ri_update(ip); | |
1994 | if (error) | |
1995 | return error; | |
1996 | } | |
1997 | /* Flushing the log may release space */ | |
1998 | if (loops == 2) | |
1999 | gfs2_log_flush(sdp, NULL); | |
b3b94faa | 2000 | } |
b3b94faa | 2001 | |
c743ffd0 | 2002 | return -ENOSPC; |
b3b94faa DT |
2003 | } |
2004 | ||
2005 | /** | |
2006 | * gfs2_inplace_release - release an inplace reservation | |
2007 | * @ip: the inode the reservation was taken out on | |
2008 | * | |
2009 | * Release a reservation made by gfs2_inplace_reserve(). | |
2010 | */ | |
2011 | ||
2012 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
2013 | { | |
564e12b1 | 2014 | struct gfs2_blkreserv *rs = ip->i_res; |
b3b94faa | 2015 | |
564e12b1 BP |
2016 | if (rs->rs_rgd_gh.gh_gl) |
2017 | gfs2_glock_dq_uninit(&rs->rs_rgd_gh); | |
b3b94faa DT |
2018 | } |
2019 | ||
2020 | /** | |
2021 | * gfs2_get_block_type - Check a block in a RG is of given type | |
2022 | * @rgd: the resource group holding the block | |
2023 | * @block: the block number | |
2024 | * | |
2025 | * Returns: The block type (GFS2_BLKST_*) | |
2026 | */ | |
2027 | ||
acf7e244 | 2028 | static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) |
b3b94faa | 2029 | { |
3983903a SW |
2030 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
2031 | int ret; | |
b3b94faa | 2032 | |
3983903a SW |
2033 | ret = gfs2_rbm_from_block(&rbm, block); |
2034 | WARN_ON_ONCE(ret != 0); | |
b3b94faa | 2035 | |
c04a2ef3 | 2036 | return gfs2_testbit(&rbm); |
b3b94faa DT |
2037 | } |
2038 | ||
60a0b8f9 | 2039 | |
b3e47ca0 BP |
2040 | /** |
2041 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 2042 | * @rbm: the resource group information |
b3e47ca0 | 2043 | * @dinode: TRUE if the first block we allocate is for a dinode |
c04a2ef3 | 2044 | * @n: The extent length (value/result) |
b3e47ca0 | 2045 | * |
c04a2ef3 | 2046 | * Add the bitmap buffer to the transaction. |
b3e47ca0 | 2047 | * Set the found bits to @new_state to change block's allocation state. |
b3e47ca0 | 2048 | */ |
c04a2ef3 | 2049 | static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
4a993fb1 | 2050 | unsigned int *n) |
b3e47ca0 | 2051 | { |
c04a2ef3 | 2052 | struct gfs2_rbm pos = { .rgd = rbm->rgd, }; |
b3e47ca0 | 2053 | const unsigned int elen = *n; |
c04a2ef3 SW |
2054 | u64 block; |
2055 | int ret; | |
b3e47ca0 | 2056 | |
c04a2ef3 SW |
2057 | *n = 1; |
2058 | block = gfs2_rbm_to_block(rbm); | |
e579ed4f | 2059 | gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh); |
3e6339dd | 2060 | gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
c04a2ef3 | 2061 | block++; |
60a0b8f9 | 2062 | while (*n < elen) { |
c04a2ef3 | 2063 | ret = gfs2_rbm_from_block(&pos, block); |
0688a5ec | 2064 | if (ret || gfs2_testbit(&pos) != GFS2_BLKST_FREE) |
60a0b8f9 | 2065 | break; |
e579ed4f | 2066 | gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh); |
3e6339dd | 2067 | gfs2_setbit(&pos, true, GFS2_BLKST_USED); |
60a0b8f9 | 2068 | (*n)++; |
c04a2ef3 | 2069 | block++; |
c8cdf479 | 2070 | } |
b3b94faa DT |
2071 | } |
2072 | ||
2073 | /** | |
2074 | * rgblk_free - Change alloc state of given block(s) | |
2075 | * @sdp: the filesystem | |
2076 | * @bstart: the start of a run of blocks to free | |
2077 | * @blen: the length of the block run (all must lie within ONE RG!) | |
2078 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
2079 | * | |
2080 | * Returns: Resource group containing the block(s) | |
2081 | */ | |
2082 | ||
cd915493 SW |
2083 | static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, |
2084 | u32 blen, unsigned char new_state) | |
b3b94faa | 2085 | { |
3b1d0b9d | 2086 | struct gfs2_rbm rbm; |
e579ed4f | 2087 | struct gfs2_bitmap *bi; |
b3b94faa | 2088 | |
3b1d0b9d SW |
2089 | rbm.rgd = gfs2_blk2rgrpd(sdp, bstart, 1); |
2090 | if (!rbm.rgd) { | |
b3b94faa | 2091 | if (gfs2_consist(sdp)) |
382066da | 2092 | fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); |
b3b94faa DT |
2093 | return NULL; |
2094 | } | |
2095 | ||
b3b94faa | 2096 | while (blen--) { |
3b1d0b9d | 2097 | gfs2_rbm_from_block(&rbm, bstart); |
e579ed4f | 2098 | bi = rbm_bi(&rbm); |
3b1d0b9d | 2099 | bstart++; |
e579ed4f BP |
2100 | if (!bi->bi_clone) { |
2101 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
2102 | GFP_NOFS | __GFP_NOFAIL); | |
2103 | memcpy(bi->bi_clone + bi->bi_offset, | |
2104 | bi->bi_bh->b_data + bi->bi_offset, bi->bi_len); | |
b3b94faa | 2105 | } |
e579ed4f | 2106 | gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh); |
3e6339dd | 2107 | gfs2_setbit(&rbm, false, new_state); |
b3b94faa DT |
2108 | } |
2109 | ||
3b1d0b9d | 2110 | return rbm.rgd; |
b3b94faa DT |
2111 | } |
2112 | ||
2113 | /** | |
09010978 SW |
2114 | * gfs2_rgrp_dump - print out an rgrp |
2115 | * @seq: The iterator | |
2116 | * @gl: The glock in question | |
2117 | * | |
2118 | */ | |
2119 | ||
2120 | int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) | |
2121 | { | |
8e2e0047 BP |
2122 | struct gfs2_rgrpd *rgd = gl->gl_object; |
2123 | struct gfs2_blkreserv *trs; | |
2124 | const struct rb_node *n; | |
2125 | ||
09010978 SW |
2126 | if (rgd == NULL) |
2127 | return 0; | |
5ea5050c | 2128 | gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u r:%u e:%u\n", |
09010978 | 2129 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 | 2130 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
5ea5050c | 2131 | rgd->rd_reserved, rgd->rd_extfail_pt); |
8e2e0047 BP |
2132 | spin_lock(&rgd->rd_rsspin); |
2133 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { | |
2134 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
2135 | dump_rs(seq, trs); | |
2136 | } | |
2137 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
2138 | return 0; |
2139 | } | |
2140 | ||
6050b9c7 SW |
2141 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
2142 | { | |
2143 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2144 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", | |
86d00636 | 2145 | (unsigned long long)rgd->rd_addr); |
6050b9c7 SW |
2146 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
2147 | gfs2_rgrp_dump(NULL, rgd->rd_gl); | |
2148 | rgd->rd_flags |= GFS2_RDF_ERROR; | |
2149 | } | |
2150 | ||
8e2e0047 | 2151 | /** |
5b924ae2 SW |
2152 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
2153 | * @ip: The inode we have just allocated blocks for | |
2154 | * @rbm: The start of the allocated blocks | |
2155 | * @len: The extent length | |
8e2e0047 | 2156 | * |
5b924ae2 SW |
2157 | * Adjusts a reservation after an allocation has taken place. If the |
2158 | * reservation does not match the allocation, or if it is now empty | |
2159 | * then it is removed. | |
8e2e0047 | 2160 | */ |
5b924ae2 SW |
2161 | |
2162 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
2163 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 BP |
2164 | { |
2165 | struct gfs2_blkreserv *rs = ip->i_res; | |
5b924ae2 SW |
2166 | struct gfs2_rgrpd *rgd = rbm->rgd; |
2167 | unsigned rlen; | |
2168 | u64 block; | |
2169 | int ret; | |
8e2e0047 | 2170 | |
5b924ae2 SW |
2171 | spin_lock(&rgd->rd_rsspin); |
2172 | if (gfs2_rs_active(rs)) { | |
2173 | if (gfs2_rbm_eq(&rs->rs_rbm, rbm)) { | |
2174 | block = gfs2_rbm_to_block(rbm); | |
2175 | ret = gfs2_rbm_from_block(&rs->rs_rbm, block + len); | |
2176 | rlen = min(rs->rs_free, len); | |
2177 | rs->rs_free -= rlen; | |
2178 | rgd->rd_reserved -= rlen; | |
9e733d39 | 2179 | trace_gfs2_rs(rs, TRACE_RS_CLAIM); |
5b924ae2 SW |
2180 | if (rs->rs_free && !ret) |
2181 | goto out; | |
2182 | } | |
20095218 | 2183 | __rs_deltree(rs); |
8e2e0047 | 2184 | } |
5b924ae2 SW |
2185 | out: |
2186 | spin_unlock(&rgd->rd_rsspin); | |
8e2e0047 BP |
2187 | } |
2188 | ||
9e07f2cb SW |
2189 | /** |
2190 | * gfs2_set_alloc_start - Set starting point for block allocation | |
2191 | * @rbm: The rbm which will be set to the required location | |
2192 | * @ip: The gfs2 inode | |
2193 | * @dinode: Flag to say if allocation includes a new inode | |
2194 | * | |
2195 | * This sets the starting point from the reservation if one is active | |
2196 | * otherwise it falls back to guessing a start point based on the | |
2197 | * inode's goal block or the last allocation point in the rgrp. | |
2198 | */ | |
2199 | ||
2200 | static void gfs2_set_alloc_start(struct gfs2_rbm *rbm, | |
2201 | const struct gfs2_inode *ip, bool dinode) | |
2202 | { | |
2203 | u64 goal; | |
2204 | ||
2205 | if (gfs2_rs_active(ip->i_res)) { | |
2206 | *rbm = ip->i_res->rs_rbm; | |
2207 | return; | |
2208 | } | |
2209 | ||
2210 | if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal)) | |
2211 | goal = ip->i_goal; | |
2212 | else | |
2213 | goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0; | |
2214 | ||
2215 | gfs2_rbm_from_block(rbm, goal); | |
2216 | } | |
2217 | ||
09010978 | 2218 | /** |
6e87ed0f | 2219 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 2220 | * @ip: the inode to allocate the block for |
09010978 | 2221 | * @bn: Used to return the starting block number |
8e2e0047 | 2222 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 2223 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 2224 | * @generation: the generation number of the inode |
b3b94faa | 2225 | * |
09010978 | 2226 | * Returns: 0 or error |
b3b94faa DT |
2227 | */ |
2228 | ||
6a8099ed | 2229 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 2230 | bool dinode, u64 *generation) |
b3b94faa | 2231 | { |
feaa7bba | 2232 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 2233 | struct buffer_head *dibh; |
4a993fb1 | 2234 | struct gfs2_rbm rbm = { .rgd = ip->i_rgd, }; |
6a8099ed | 2235 | unsigned int ndata; |
3c5d785a | 2236 | u64 block; /* block, within the file system scope */ |
d9ba7615 | 2237 | int error; |
b3b94faa | 2238 | |
9e07f2cb | 2239 | gfs2_set_alloc_start(&rbm, ip, dinode); |
5ce13431 | 2240 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, ip, false, NULL); |
62e252ee | 2241 | |
137834a6 | 2242 | if (error == -ENOSPC) { |
9e07f2cb | 2243 | gfs2_set_alloc_start(&rbm, ip, dinode); |
5ce13431 BP |
2244 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, NULL, NULL, false, |
2245 | NULL); | |
137834a6 SW |
2246 | } |
2247 | ||
62e252ee | 2248 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
5b924ae2 | 2249 | if (error) { |
5ea5050c | 2250 | fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n", |
9e733d39 | 2251 | (unsigned long long)ip->i_no_addr, error, *nblocks, |
5ea5050c BP |
2252 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags), |
2253 | rbm.rgd->rd_extfail_pt); | |
62e252ee | 2254 | goto rgrp_error; |
8e2e0047 | 2255 | } |
62e252ee | 2256 | |
c04a2ef3 SW |
2257 | gfs2_alloc_extent(&rbm, dinode, nblocks); |
2258 | block = gfs2_rbm_to_block(&rbm); | |
c743ffd0 | 2259 | rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0; |
5b924ae2 SW |
2260 | if (gfs2_rs_active(ip->i_res)) |
2261 | gfs2_adjust_reservation(ip, &rbm, *nblocks); | |
6a8099ed SW |
2262 | ndata = *nblocks; |
2263 | if (dinode) | |
2264 | ndata--; | |
b3e47ca0 | 2265 | |
3c5d785a | 2266 | if (!dinode) { |
6a8099ed | 2267 | ip->i_goal = block + ndata - 1; |
3c5d785a BP |
2268 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2269 | if (error == 0) { | |
2270 | struct gfs2_dinode *di = | |
2271 | (struct gfs2_dinode *)dibh->b_data; | |
350a9b0a | 2272 | gfs2_trans_add_meta(ip->i_gl, dibh); |
3c5d785a BP |
2273 | di->di_goal_meta = di->di_goal_data = |
2274 | cpu_to_be64(ip->i_goal); | |
2275 | brelse(dibh); | |
2276 | } | |
d9ba7615 | 2277 | } |
4a993fb1 | 2278 | if (rbm.rgd->rd_free < *nblocks) { |
8e2e0047 | 2279 | printk(KERN_WARNING "nblocks=%u\n", *nblocks); |
09010978 | 2280 | goto rgrp_error; |
8e2e0047 | 2281 | } |
09010978 | 2282 | |
4a993fb1 | 2283 | rbm.rgd->rd_free -= *nblocks; |
3c5d785a | 2284 | if (dinode) { |
4a993fb1 SW |
2285 | rbm.rgd->rd_dinodes++; |
2286 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2287 | if (*generation == 0) |
4a993fb1 | 2288 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2289 | } |
b3b94faa | 2290 | |
350a9b0a | 2291 | gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh); |
4a993fb1 SW |
2292 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); |
2293 | gfs2_rgrp_ondisk2lvb(rbm.rgd->rd_rgl, rbm.rgd->rd_bits[0].bi_bh->b_data); | |
b3b94faa | 2294 | |
6a8099ed | 2295 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a BP |
2296 | if (dinode) |
2297 | gfs2_trans_add_unrevoke(sdp, block, 1); | |
6a8099ed | 2298 | |
fd4b4e04 | 2299 | gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa | 2300 | |
4a993fb1 SW |
2301 | rbm.rgd->rd_free_clone -= *nblocks; |
2302 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, | |
6e87ed0f | 2303 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2304 | *bn = block; |
2305 | return 0; | |
2306 | ||
2307 | rgrp_error: | |
4a993fb1 | 2308 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2309 | return -EIO; |
b3b94faa DT |
2310 | } |
2311 | ||
2312 | /** | |
46fcb2ed | 2313 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa DT |
2314 | * @ip: the inode these blocks are being freed from |
2315 | * @bstart: first block of a run of contiguous blocks | |
2316 | * @blen: the length of the block run | |
46fcb2ed | 2317 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2318 | * |
2319 | */ | |
2320 | ||
46fcb2ed | 2321 | void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) |
b3b94faa | 2322 | { |
feaa7bba | 2323 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2324 | struct gfs2_rgrpd *rgd; |
2325 | ||
2326 | rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); | |
2327 | if (!rgd) | |
2328 | return; | |
41db1ab9 | 2329 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2330 | rgd->rd_free += blen; |
66fc061b | 2331 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
350a9b0a | 2332 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2333 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 | 2334 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
b3b94faa | 2335 | |
6d3117b4 | 2336 | /* Directories keep their data in the metadata address space */ |
46fcb2ed | 2337 | if (meta || ip->i_depth) |
6d3117b4 | 2338 | gfs2_meta_wipe(ip, bstart, blen); |
4c16c36a | 2339 | } |
b3b94faa | 2340 | |
4c16c36a BP |
2341 | /** |
2342 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2343 | * @ip: the inode these blocks are being freed from | |
2344 | * @bstart: first block of a run of contiguous blocks | |
2345 | * @blen: the length of the block run | |
2346 | * | |
2347 | */ | |
2348 | ||
2349 | void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) | |
2350 | { | |
2351 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2352 | ||
46fcb2ed | 2353 | __gfs2_free_blocks(ip, bstart, blen, 1); |
b3b94faa | 2354 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2355 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2356 | } |
2357 | ||
feaa7bba SW |
2358 | void gfs2_unlink_di(struct inode *inode) |
2359 | { | |
2360 | struct gfs2_inode *ip = GFS2_I(inode); | |
2361 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2362 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2363 | u64 blkno = ip->i_no_addr; |
feaa7bba SW |
2364 | |
2365 | rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); | |
2366 | if (!rgd) | |
2367 | return; | |
41db1ab9 | 2368 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
350a9b0a | 2369 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2370 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2371 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2372 | update_rgrp_lvb_unlinked(rgd, 1); | |
feaa7bba SW |
2373 | } |
2374 | ||
cd915493 | 2375 | static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) |
b3b94faa DT |
2376 | { |
2377 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
2378 | struct gfs2_rgrpd *tmp_rgd; | |
2379 | ||
2380 | tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); | |
2381 | if (!tmp_rgd) | |
2382 | return; | |
2383 | gfs2_assert_withdraw(sdp, rgd == tmp_rgd); | |
2384 | ||
73f74948 | 2385 | if (!rgd->rd_dinodes) |
b3b94faa | 2386 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2387 | rgd->rd_dinodes--; |
cfc8b549 | 2388 | rgd->rd_free++; |
b3b94faa | 2389 | |
350a9b0a | 2390 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2391 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
90306c41 BM |
2392 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, rgd->rd_bits[0].bi_bh->b_data); |
2393 | update_rgrp_lvb_unlinked(rgd, -1); | |
b3b94faa DT |
2394 | |
2395 | gfs2_statfs_change(sdp, 0, +1, -1); | |
b3b94faa DT |
2396 | } |
2397 | ||
b3b94faa DT |
2398 | |
2399 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) | |
2400 | { | |
dbb7cae2 | 2401 | gfs2_free_uninit_di(rgd, ip->i_no_addr); |
41db1ab9 | 2402 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2403 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
dbb7cae2 | 2404 | gfs2_meta_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2405 | } |
2406 | ||
acf7e244 SW |
2407 | /** |
2408 | * gfs2_check_blk_type - Check the type of a block | |
2409 | * @sdp: The superblock | |
2410 | * @no_addr: The block number to check | |
2411 | * @type: The block type we are looking for | |
2412 | * | |
2413 | * Returns: 0 if the block type matches the expected type | |
2414 | * -ESTALE if it doesn't match | |
2415 | * or -ve errno if something went wrong while checking | |
2416 | */ | |
2417 | ||
2418 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2419 | { | |
2420 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2421 | struct gfs2_holder rgd_gh; |
58884c4d | 2422 | int error = -EINVAL; |
acf7e244 | 2423 | |
66fc061b | 2424 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2425 | if (!rgd) |
8339ee54 | 2426 | goto fail; |
acf7e244 SW |
2427 | |
2428 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2429 | if (error) | |
8339ee54 | 2430 | goto fail; |
acf7e244 SW |
2431 | |
2432 | if (gfs2_get_block_type(rgd, no_addr) != type) | |
2433 | error = -ESTALE; | |
2434 | ||
2435 | gfs2_glock_dq_uninit(&rgd_gh); | |
acf7e244 SW |
2436 | fail: |
2437 | return error; | |
2438 | } | |
2439 | ||
b3b94faa DT |
2440 | /** |
2441 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2442 | * @ip: the inode |
b3b94faa DT |
2443 | * @rlist: the list of resource groups |
2444 | * @block: the block | |
2445 | * | |
2446 | * Figure out what RG a block belongs to and add that RG to the list | |
2447 | * | |
2448 | * FIXME: Don't use NOFAIL | |
2449 | * | |
2450 | */ | |
2451 | ||
70b0c365 | 2452 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2453 | u64 block) |
b3b94faa | 2454 | { |
70b0c365 | 2455 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2456 | struct gfs2_rgrpd *rgd; |
2457 | struct gfs2_rgrpd **tmp; | |
2458 | unsigned int new_space; | |
2459 | unsigned int x; | |
2460 | ||
2461 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2462 | return; | |
2463 | ||
70b0c365 SW |
2464 | if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) |
2465 | rgd = ip->i_rgd; | |
2466 | else | |
66fc061b | 2467 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
b3b94faa | 2468 | if (!rgd) { |
70b0c365 | 2469 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); |
b3b94faa DT |
2470 | return; |
2471 | } | |
70b0c365 | 2472 | ip->i_rgd = rgd; |
b3b94faa DT |
2473 | |
2474 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2475 | if (rlist->rl_rgd[x] == rgd) | |
2476 | return; | |
2477 | ||
2478 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2479 | new_space = rlist->rl_space + 10; | |
2480 | ||
2481 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2482 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2483 | |
2484 | if (rlist->rl_rgd) { | |
2485 | memcpy(tmp, rlist->rl_rgd, | |
2486 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2487 | kfree(rlist->rl_rgd); | |
2488 | } | |
2489 | ||
2490 | rlist->rl_space = new_space; | |
2491 | rlist->rl_rgd = tmp; | |
2492 | } | |
2493 | ||
2494 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2495 | } | |
2496 | ||
2497 | /** | |
2498 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2499 | * and initialize an array of glock holders for them | |
2500 | * @rlist: the list of resource groups | |
2501 | * @state: the lock state to acquire the RG lock in | |
b3b94faa DT |
2502 | * |
2503 | * FIXME: Don't use NOFAIL | |
2504 | * | |
2505 | */ | |
2506 | ||
fe6c991c | 2507 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) |
b3b94faa DT |
2508 | { |
2509 | unsigned int x; | |
2510 | ||
2511 | rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), | |
dd894be8 | 2512 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2513 | for (x = 0; x < rlist->rl_rgrps; x++) |
2514 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, | |
fe6c991c | 2515 | state, 0, |
b3b94faa DT |
2516 | &rlist->rl_ghs[x]); |
2517 | } | |
2518 | ||
2519 | /** | |
2520 | * gfs2_rlist_free - free a resource group list | |
2521 | * @list: the list of resource groups | |
2522 | * | |
2523 | */ | |
2524 | ||
2525 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2526 | { | |
2527 | unsigned int x; | |
2528 | ||
2529 | kfree(rlist->rl_rgd); | |
2530 | ||
2531 | if (rlist->rl_ghs) { | |
2532 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2533 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2534 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2535 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2536 | } |
2537 | } | |
2538 |