]>
Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
fe6c991c | 4 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
d77d1b58 JP |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
8 | ||
b3b94faa DT |
9 | #include <linux/slab.h> |
10 | #include <linux/spinlock.h> | |
11 | #include <linux/completion.h> | |
12 | #include <linux/buffer_head.h> | |
f42faf4f | 13 | #include <linux/fs.h> |
5c676f6d | 14 | #include <linux/gfs2_ondisk.h> |
1f466a47 | 15 | #include <linux/prefetch.h> |
f15ab561 | 16 | #include <linux/blkdev.h> |
7c9ca621 | 17 | #include <linux/rbtree.h> |
9dbe9610 | 18 | #include <linux/random.h> |
b3b94faa DT |
19 | |
20 | #include "gfs2.h" | |
5c676f6d | 21 | #include "incore.h" |
b3b94faa DT |
22 | #include "glock.h" |
23 | #include "glops.h" | |
b3b94faa DT |
24 | #include "lops.h" |
25 | #include "meta_io.h" | |
26 | #include "quota.h" | |
27 | #include "rgrp.h" | |
28 | #include "super.h" | |
29 | #include "trans.h" | |
5c676f6d | 30 | #include "util.h" |
172e045a | 31 | #include "log.h" |
c8cdf479 | 32 | #include "inode.h" |
63997775 | 33 | #include "trace_gfs2.h" |
850d2d91 | 34 | #include "dir.h" |
b3b94faa | 35 | |
2c1e52aa | 36 | #define BFITNOENT ((u32)~0) |
6760bdcd | 37 | #define NO_BLOCK ((u64)~0) |
88c8ab1f | 38 | |
c65b76b8 AG |
39 | struct gfs2_rbm { |
40 | struct gfs2_rgrpd *rgd; | |
41 | u32 offset; /* The offset is bitmap relative */ | |
42 | int bii; /* Bitmap index */ | |
43 | }; | |
44 | ||
45 | static inline struct gfs2_bitmap *rbm_bi(const struct gfs2_rbm *rbm) | |
46 | { | |
47 | return rbm->rgd->rd_bits + rbm->bii; | |
48 | } | |
49 | ||
50 | static inline u64 gfs2_rbm_to_block(const struct gfs2_rbm *rbm) | |
51 | { | |
52 | BUG_ON(rbm->offset >= rbm->rgd->rd_data); | |
53 | return rbm->rgd->rd_data0 + (rbm_bi(rbm)->bi_start * GFS2_NBBY) + | |
54 | rbm->offset; | |
55 | } | |
56 | ||
88c8ab1f SW |
57 | /* |
58 | * These routines are used by the resource group routines (rgrp.c) | |
59 | * to keep track of block allocation. Each block is represented by two | |
feaa7bba SW |
60 | * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. |
61 | * | |
62 | * 0 = Free | |
63 | * 1 = Used (not metadata) | |
64 | * 2 = Unlinked (still in use) inode | |
65 | * 3 = Used (metadata) | |
88c8ab1f SW |
66 | */ |
67 | ||
5ce13431 BP |
68 | struct gfs2_extent { |
69 | struct gfs2_rbm rbm; | |
70 | u32 len; | |
71 | }; | |
72 | ||
88c8ab1f SW |
73 | static const char valid_change[16] = { |
74 | /* current */ | |
feaa7bba | 75 | /* n */ 0, 1, 1, 1, |
88c8ab1f | 76 | /* e */ 1, 0, 0, 0, |
feaa7bba | 77 | /* w */ 0, 0, 0, 1, |
88c8ab1f SW |
78 | 1, 0, 0, 0 |
79 | }; | |
80 | ||
5ce13431 | 81 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
3d39fcd1 | 82 | struct gfs2_blkreserv *rs, bool nowrap); |
ff7f4cb4 SW |
83 | |
84 | ||
88c8ab1f SW |
85 | /** |
86 | * gfs2_setbit - Set a bit in the bitmaps | |
3e6339dd SW |
87 | * @rbm: The position of the bit to set |
88 | * @do_clone: Also set the clone bitmap, if it exists | |
88c8ab1f SW |
89 | * @new_state: the new state of the block |
90 | * | |
91 | */ | |
92 | ||
3e6339dd | 93 | static inline void gfs2_setbit(const struct gfs2_rbm *rbm, bool do_clone, |
06344b91 | 94 | unsigned char new_state) |
88c8ab1f | 95 | { |
b45e41d7 | 96 | unsigned char *byte1, *byte2, *end, cur_state; |
e579ed4f | 97 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
281b4952 | 98 | unsigned int buflen = bi->bi_bytes; |
3e6339dd | 99 | const unsigned int bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; |
88c8ab1f | 100 | |
e579ed4f BP |
101 | byte1 = bi->bi_bh->b_data + bi->bi_offset + (rbm->offset / GFS2_NBBY); |
102 | end = bi->bi_bh->b_data + bi->bi_offset + buflen; | |
88c8ab1f | 103 | |
b45e41d7 | 104 | BUG_ON(byte1 >= end); |
88c8ab1f | 105 | |
b45e41d7 | 106 | cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; |
88c8ab1f | 107 | |
b45e41d7 | 108 | if (unlikely(!valid_change[new_state * 4 + cur_state])) { |
e54c78a2 BP |
109 | struct gfs2_sbd *sdp = rbm->rgd->rd_sbd; |
110 | ||
111 | fs_warn(sdp, "buf_blk = 0x%x old_state=%d, new_state=%d\n", | |
d77d1b58 | 112 | rbm->offset, cur_state, new_state); |
e54c78a2 BP |
113 | fs_warn(sdp, "rgrp=0x%llx bi_start=0x%x biblk: 0x%llx\n", |
114 | (unsigned long long)rbm->rgd->rd_addr, bi->bi_start, | |
115 | (unsigned long long)bi->bi_bh->b_blocknr); | |
281b4952 AG |
116 | fs_warn(sdp, "bi_offset=0x%x bi_bytes=0x%x block=0x%llx\n", |
117 | bi->bi_offset, bi->bi_bytes, | |
e54c78a2 | 118 | (unsigned long long)gfs2_rbm_to_block(rbm)); |
95c8e17f | 119 | dump_stack(); |
3e6339dd | 120 | gfs2_consist_rgrpd(rbm->rgd); |
b45e41d7 SW |
121 | return; |
122 | } | |
123 | *byte1 ^= (cur_state ^ new_state) << bit; | |
124 | ||
e579ed4f BP |
125 | if (do_clone && bi->bi_clone) { |
126 | byte2 = bi->bi_clone + bi->bi_offset + (rbm->offset / GFS2_NBBY); | |
b45e41d7 SW |
127 | cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; |
128 | *byte2 ^= (cur_state ^ new_state) << bit; | |
129 | } | |
88c8ab1f SW |
130 | } |
131 | ||
132 | /** | |
133 | * gfs2_testbit - test a bit in the bitmaps | |
c04a2ef3 | 134 | * @rbm: The bit to test |
dffe12a8 BP |
135 | * @use_clone: If true, test the clone bitmap, not the official bitmap. |
136 | * | |
137 | * Some callers like gfs2_unaligned_extlen need to test the clone bitmaps, | |
138 | * not the "real" bitmaps, to avoid allocating recently freed blocks. | |
88c8ab1f | 139 | * |
c04a2ef3 | 140 | * Returns: The two bit block state of the requested bit |
88c8ab1f SW |
141 | */ |
142 | ||
dffe12a8 | 143 | static inline u8 gfs2_testbit(const struct gfs2_rbm *rbm, bool use_clone) |
88c8ab1f | 144 | { |
e579ed4f | 145 | struct gfs2_bitmap *bi = rbm_bi(rbm); |
dffe12a8 | 146 | const u8 *buffer; |
c04a2ef3 | 147 | const u8 *byte; |
88c8ab1f SW |
148 | unsigned int bit; |
149 | ||
dffe12a8 BP |
150 | if (use_clone && bi->bi_clone) |
151 | buffer = bi->bi_clone; | |
152 | else | |
153 | buffer = bi->bi_bh->b_data; | |
154 | buffer += bi->bi_offset; | |
c04a2ef3 SW |
155 | byte = buffer + (rbm->offset / GFS2_NBBY); |
156 | bit = (rbm->offset % GFS2_NBBY) * GFS2_BIT_SIZE; | |
88c8ab1f | 157 | |
c04a2ef3 | 158 | return (*byte >> bit) & GFS2_BIT_MASK; |
88c8ab1f SW |
159 | } |
160 | ||
223b2b88 SW |
161 | /** |
162 | * gfs2_bit_search | |
163 | * @ptr: Pointer to bitmap data | |
164 | * @mask: Mask to use (normally 0x55555.... but adjusted for search start) | |
165 | * @state: The state we are searching for | |
166 | * | |
167 | * We xor the bitmap data with a patter which is the bitwise opposite | |
168 | * of what we are looking for, this gives rise to a pattern of ones | |
169 | * wherever there is a match. Since we have two bits per entry, we | |
170 | * take this pattern, shift it down by one place and then and it with | |
171 | * the original. All the even bit positions (0,2,4, etc) then represent | |
172 | * successful matches, so we mask with 0x55555..... to remove the unwanted | |
173 | * odd bit positions. | |
174 | * | |
175 | * This allows searching of a whole u64 at once (32 blocks) with a | |
176 | * single test (on 64 bit arches). | |
177 | */ | |
178 | ||
179 | static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) | |
180 | { | |
181 | u64 tmp; | |
182 | static const u64 search[] = { | |
075ac448 HE |
183 | [0] = 0xffffffffffffffffULL, |
184 | [1] = 0xaaaaaaaaaaaaaaaaULL, | |
185 | [2] = 0x5555555555555555ULL, | |
186 | [3] = 0x0000000000000000ULL, | |
223b2b88 SW |
187 | }; |
188 | tmp = le64_to_cpu(*ptr) ^ search[state]; | |
189 | tmp &= (tmp >> 1); | |
190 | tmp &= mask; | |
191 | return tmp; | |
192 | } | |
193 | ||
8e2e0047 BP |
194 | /** |
195 | * rs_cmp - multi-block reservation range compare | |
c65b76b8 | 196 | * @start: start of the new reservation |
8e2e0047 BP |
197 | * @len: number of blocks in the new reservation |
198 | * @rs: existing reservation to compare against | |
199 | * | |
200 | * returns: 1 if the block range is beyond the reach of the reservation | |
201 | * -1 if the block range is before the start of the reservation | |
202 | * 0 if the block range overlaps with the reservation | |
203 | */ | |
c65b76b8 | 204 | static inline int rs_cmp(u64 start, u32 len, struct gfs2_blkreserv *rs) |
8e2e0047 | 205 | { |
07974d2a | 206 | if (start >= rs->rs_start + rs->rs_requested) |
8e2e0047 | 207 | return 1; |
c65b76b8 | 208 | if (rs->rs_start >= start + len) |
8e2e0047 BP |
209 | return -1; |
210 | return 0; | |
211 | } | |
212 | ||
88c8ab1f SW |
213 | /** |
214 | * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing | |
215 | * a block in a given allocation state. | |
886b1416 | 216 | * @buf: the buffer that holds the bitmaps |
223b2b88 | 217 | * @len: the length (in bytes) of the buffer |
88c8ab1f | 218 | * @goal: start search at this block's bit-pair (within @buffer) |
223b2b88 | 219 | * @state: GFS2_BLKST_XXX the state of the block we're looking for. |
88c8ab1f SW |
220 | * |
221 | * Scope of @goal and returned block number is only within this bitmap buffer, | |
222 | * not entire rgrp or filesystem. @buffer will be offset from the actual | |
223b2b88 SW |
223 | * beginning of a bitmap block buffer, skipping any header structures, but |
224 | * headers are always a multiple of 64 bits long so that the buffer is | |
225 | * always aligned to a 64 bit boundary. | |
226 | * | |
227 | * The size of the buffer is in bytes, but is it assumed that it is | |
fd589a8f | 228 | * always ok to read a complete multiple of 64 bits at the end |
223b2b88 | 229 | * of the block in case the end is no aligned to a natural boundary. |
88c8ab1f SW |
230 | * |
231 | * Return: the block number (bitmap buffer scope) that was found | |
232 | */ | |
233 | ||
02ab1721 HE |
234 | static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, |
235 | u32 goal, u8 state) | |
88c8ab1f | 236 | { |
223b2b88 SW |
237 | u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); |
238 | const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); | |
239 | const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); | |
240 | u64 tmp; | |
075ac448 | 241 | u64 mask = 0x5555555555555555ULL; |
223b2b88 SW |
242 | u32 bit; |
243 | ||
223b2b88 SW |
244 | /* Mask off bits we don't care about at the start of the search */ |
245 | mask <<= spoint; | |
246 | tmp = gfs2_bit_search(ptr, mask, state); | |
247 | ptr++; | |
248 | while(tmp == 0 && ptr < end) { | |
075ac448 | 249 | tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); |
223b2b88 | 250 | ptr++; |
1f466a47 | 251 | } |
223b2b88 SW |
252 | /* Mask off any bits which are more than len bytes from the start */ |
253 | if (ptr == end && (len & (sizeof(u64) - 1))) | |
254 | tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); | |
255 | /* Didn't find anything, so return */ | |
256 | if (tmp == 0) | |
257 | return BFITNOENT; | |
258 | ptr--; | |
d8bd504a | 259 | bit = __ffs64(tmp); |
223b2b88 SW |
260 | bit /= 2; /* two bits per entry in the bitmap */ |
261 | return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; | |
88c8ab1f SW |
262 | } |
263 | ||
ff7f4cb4 SW |
264 | /** |
265 | * gfs2_rbm_from_block - Set the rbm based upon rgd and block number | |
266 | * @rbm: The rbm with rgd already set correctly | |
267 | * @block: The block number (filesystem relative) | |
268 | * | |
269 | * This sets the bi and offset members of an rbm based on a | |
270 | * resource group and a filesystem relative block number. The | |
271 | * resource group must be set in the rbm on entry, the bi and | |
272 | * offset members will be set by this function. | |
273 | * | |
274 | * Returns: 0 on success, or an error code | |
275 | */ | |
276 | ||
277 | static int gfs2_rbm_from_block(struct gfs2_rbm *rbm, u64 block) | |
278 | { | |
3548fce1 | 279 | if (!rgrp_contains_block(rbm->rgd, block)) |
ff7f4cb4 | 280 | return -E2BIG; |
e579ed4f | 281 | rbm->bii = 0; |
3548fce1 | 282 | rbm->offset = block - rbm->rgd->rd_data0; |
a68a0a35 | 283 | /* Check if the block is within the first block */ |
e579ed4f | 284 | if (rbm->offset < rbm_bi(rbm)->bi_blocks) |
a68a0a35 BP |
285 | return 0; |
286 | ||
287 | /* Adjust for the size diff between gfs2_meta_header and gfs2_rgrp */ | |
288 | rbm->offset += (sizeof(struct gfs2_rgrp) - | |
289 | sizeof(struct gfs2_meta_header)) * GFS2_NBBY; | |
e579ed4f BP |
290 | rbm->bii = rbm->offset / rbm->rgd->rd_sbd->sd_blocks_per_bitmap; |
291 | rbm->offset -= rbm->bii * rbm->rgd->rd_sbd->sd_blocks_per_bitmap; | |
ff7f4cb4 SW |
292 | return 0; |
293 | } | |
294 | ||
149ed7f5 | 295 | /** |
0eacdd16 | 296 | * gfs2_rbm_add - add a number of blocks to an rbm |
149ed7f5 | 297 | * @rbm: The rbm with rgd already set correctly |
0eacdd16 | 298 | * @blocks: The number of blocks to add to rpm |
149ed7f5 | 299 | * |
0eacdd16 AG |
300 | * This function takes an existing rbm structure and adds a number of blocks to |
301 | * it. | |
149ed7f5 | 302 | * |
0eacdd16 | 303 | * Returns: True if the new rbm would point past the end of the rgrp. |
149ed7f5 BP |
304 | */ |
305 | ||
0eacdd16 | 306 | static bool gfs2_rbm_add(struct gfs2_rbm *rbm, u32 blocks) |
149ed7f5 | 307 | { |
0eacdd16 AG |
308 | struct gfs2_rgrpd *rgd = rbm->rgd; |
309 | struct gfs2_bitmap *bi = rgd->rd_bits + rbm->bii; | |
310 | ||
311 | if (rbm->offset + blocks < bi->bi_blocks) { | |
312 | rbm->offset += blocks; | |
149ed7f5 BP |
313 | return false; |
314 | } | |
0eacdd16 | 315 | blocks -= bi->bi_blocks - rbm->offset; |
149ed7f5 | 316 | |
0eacdd16 AG |
317 | for(;;) { |
318 | bi++; | |
319 | if (bi == rgd->rd_bits + rgd->rd_length) | |
320 | return true; | |
321 | if (blocks < bi->bi_blocks) { | |
322 | rbm->offset = blocks; | |
323 | rbm->bii = bi - rgd->rd_bits; | |
324 | return false; | |
325 | } | |
326 | blocks -= bi->bi_blocks; | |
327 | } | |
149ed7f5 BP |
328 | } |
329 | ||
ff7f4cb4 SW |
330 | /** |
331 | * gfs2_unaligned_extlen - Look for free blocks which are not byte aligned | |
332 | * @rbm: Position to search (value/result) | |
333 | * @n_unaligned: Number of unaligned blocks to check | |
334 | * @len: Decremented for each block found (terminate on zero) | |
335 | * | |
c65b76b8 AG |
336 | * Returns: true if a non-free block is encountered or the end of the resource |
337 | * group is reached. | |
ff7f4cb4 SW |
338 | */ |
339 | ||
340 | static bool gfs2_unaligned_extlen(struct gfs2_rbm *rbm, u32 n_unaligned, u32 *len) | |
341 | { | |
ff7f4cb4 SW |
342 | u32 n; |
343 | u8 res; | |
344 | ||
345 | for (n = 0; n < n_unaligned; n++) { | |
dffe12a8 | 346 | res = gfs2_testbit(rbm, true); |
ff7f4cb4 SW |
347 | if (res != GFS2_BLKST_FREE) |
348 | return true; | |
349 | (*len)--; | |
350 | if (*len == 0) | |
351 | return true; | |
0eacdd16 | 352 | if (gfs2_rbm_add(rbm, 1)) |
ff7f4cb4 SW |
353 | return true; |
354 | } | |
355 | ||
356 | return false; | |
357 | } | |
358 | ||
359 | /** | |
360 | * gfs2_free_extlen - Return extent length of free blocks | |
27ff6a0f | 361 | * @rrbm: Starting position |
ff7f4cb4 SW |
362 | * @len: Max length to check |
363 | * | |
364 | * Starting at the block specified by the rbm, see how many free blocks | |
365 | * there are, not reading more than len blocks ahead. This can be done | |
366 | * using memchr_inv when the blocks are byte aligned, but has to be done | |
367 | * on a block by block basis in case of unaligned blocks. Also this | |
368 | * function can cope with bitmap boundaries (although it must stop on | |
369 | * a resource group boundary) | |
370 | * | |
371 | * Returns: Number of free blocks in the extent | |
372 | */ | |
373 | ||
374 | static u32 gfs2_free_extlen(const struct gfs2_rbm *rrbm, u32 len) | |
375 | { | |
376 | struct gfs2_rbm rbm = *rrbm; | |
377 | u32 n_unaligned = rbm.offset & 3; | |
378 | u32 size = len; | |
379 | u32 bytes; | |
380 | u32 chunk_size; | |
381 | u8 *ptr, *start, *end; | |
382 | u64 block; | |
e579ed4f | 383 | struct gfs2_bitmap *bi; |
ff7f4cb4 SW |
384 | |
385 | if (n_unaligned && | |
386 | gfs2_unaligned_extlen(&rbm, 4 - n_unaligned, &len)) | |
387 | goto out; | |
388 | ||
3701530a | 389 | n_unaligned = len & 3; |
ff7f4cb4 SW |
390 | /* Start is now byte aligned */ |
391 | while (len > 3) { | |
e579ed4f BP |
392 | bi = rbm_bi(&rbm); |
393 | start = bi->bi_bh->b_data; | |
394 | if (bi->bi_clone) | |
395 | start = bi->bi_clone; | |
e579ed4f | 396 | start += bi->bi_offset; |
281b4952 | 397 | end = start + bi->bi_bytes; |
ff7f4cb4 SW |
398 | BUG_ON(rbm.offset & 3); |
399 | start += (rbm.offset / GFS2_NBBY); | |
400 | bytes = min_t(u32, len / GFS2_NBBY, (end - start)); | |
401 | ptr = memchr_inv(start, 0, bytes); | |
402 | chunk_size = ((ptr == NULL) ? bytes : (ptr - start)); | |
403 | chunk_size *= GFS2_NBBY; | |
404 | BUG_ON(len < chunk_size); | |
405 | len -= chunk_size; | |
406 | block = gfs2_rbm_to_block(&rbm); | |
15bd50ad BP |
407 | if (gfs2_rbm_from_block(&rbm, block + chunk_size)) { |
408 | n_unaligned = 0; | |
ff7f4cb4 | 409 | break; |
15bd50ad BP |
410 | } |
411 | if (ptr) { | |
412 | n_unaligned = 3; | |
413 | break; | |
414 | } | |
ff7f4cb4 SW |
415 | n_unaligned = len & 3; |
416 | } | |
417 | ||
418 | /* Deal with any bits left over at the end */ | |
419 | if (n_unaligned) | |
420 | gfs2_unaligned_extlen(&rbm, n_unaligned, &len); | |
421 | out: | |
422 | return size - len; | |
423 | } | |
424 | ||
88c8ab1f SW |
425 | /** |
426 | * gfs2_bitcount - count the number of bits in a certain state | |
886b1416 | 427 | * @rgd: the resource group descriptor |
88c8ab1f SW |
428 | * @buffer: the buffer that holds the bitmaps |
429 | * @buflen: the length (in bytes) of the buffer | |
430 | * @state: the state of the block we're looking for | |
431 | * | |
432 | * Returns: The number of bits | |
433 | */ | |
434 | ||
110acf38 SW |
435 | static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, |
436 | unsigned int buflen, u8 state) | |
88c8ab1f | 437 | { |
110acf38 SW |
438 | const u8 *byte = buffer; |
439 | const u8 *end = buffer + buflen; | |
440 | const u8 state1 = state << 2; | |
441 | const u8 state2 = state << 4; | |
442 | const u8 state3 = state << 6; | |
cd915493 | 443 | u32 count = 0; |
88c8ab1f SW |
444 | |
445 | for (; byte < end; byte++) { | |
446 | if (((*byte) & 0x03) == state) | |
447 | count++; | |
448 | if (((*byte) & 0x0C) == state1) | |
449 | count++; | |
450 | if (((*byte) & 0x30) == state2) | |
451 | count++; | |
452 | if (((*byte) & 0xC0) == state3) | |
453 | count++; | |
454 | } | |
455 | ||
456 | return count; | |
457 | } | |
458 | ||
b3b94faa DT |
459 | /** |
460 | * gfs2_rgrp_verify - Verify that a resource group is consistent | |
b3b94faa DT |
461 | * @rgd: the rgrp |
462 | * | |
463 | */ | |
464 | ||
465 | void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) | |
466 | { | |
467 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
468 | struct gfs2_bitmap *bi = NULL; | |
bb8d8a6f | 469 | u32 length = rgd->rd_length; |
cd915493 | 470 | u32 count[4], tmp; |
b3b94faa DT |
471 | int buf, x; |
472 | ||
cd915493 | 473 | memset(count, 0, 4 * sizeof(u32)); |
b3b94faa DT |
474 | |
475 | /* Count # blocks in each of 4 possible allocation states */ | |
476 | for (buf = 0; buf < length; buf++) { | |
477 | bi = rgd->rd_bits + buf; | |
478 | for (x = 0; x < 4; x++) | |
479 | count[x] += gfs2_bitcount(rgd, | |
480 | bi->bi_bh->b_data + | |
481 | bi->bi_offset, | |
281b4952 | 482 | bi->bi_bytes, x); |
b3b94faa DT |
483 | } |
484 | ||
cfc8b549 | 485 | if (count[0] != rgd->rd_free) { |
8dc88ac6 AG |
486 | gfs2_lm(sdp, "free data mismatch: %u != %u\n", |
487 | count[0], rgd->rd_free); | |
488 | gfs2_consist_rgrpd(rgd); | |
b3b94faa DT |
489 | return; |
490 | } | |
491 | ||
73f74948 | 492 | tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; |
6b946170 | 493 | if (count[1] != tmp) { |
8dc88ac6 AG |
494 | gfs2_lm(sdp, "used data mismatch: %u != %u\n", |
495 | count[1], tmp); | |
496 | gfs2_consist_rgrpd(rgd); | |
b3b94faa DT |
497 | return; |
498 | } | |
499 | ||
6b946170 | 500 | if (count[2] + count[3] != rgd->rd_dinodes) { |
8dc88ac6 AG |
501 | gfs2_lm(sdp, "used metadata mismatch: %u != %u\n", |
502 | count[2] + count[3], rgd->rd_dinodes); | |
503 | gfs2_consist_rgrpd(rgd); | |
b3b94faa DT |
504 | return; |
505 | } | |
b3b94faa DT |
506 | } |
507 | ||
b3b94faa DT |
508 | /** |
509 | * gfs2_blk2rgrpd - Find resource group for a given data/meta block number | |
510 | * @sdp: The GFS2 superblock | |
886b1416 BP |
511 | * @blk: The data block number |
512 | * @exact: True if this needs to be an exact match | |
b3b94faa | 513 | * |
90bcab99 SW |
514 | * The @exact argument should be set to true by most callers. The exception |
515 | * is when we need to match blocks which are not represented by the rgrp | |
516 | * bitmap, but which are part of the rgrp (i.e. padding blocks) which are | |
517 | * there for alignment purposes. Another way of looking at it is that @exact | |
518 | * matches only valid data/metadata blocks, but with @exact false, it will | |
519 | * match any block within the extent of the rgrp. | |
520 | * | |
b3b94faa DT |
521 | * Returns: The resource group, or NULL if not found |
522 | */ | |
523 | ||
66fc061b | 524 | struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk, bool exact) |
b3b94faa | 525 | { |
66fc061b | 526 | struct rb_node *n, *next; |
f75bbfb4 | 527 | struct gfs2_rgrpd *cur; |
b3b94faa DT |
528 | |
529 | spin_lock(&sdp->sd_rindex_spin); | |
66fc061b SW |
530 | n = sdp->sd_rindex_tree.rb_node; |
531 | while (n) { | |
532 | cur = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
533 | next = NULL; | |
7c9ca621 | 534 | if (blk < cur->rd_addr) |
66fc061b | 535 | next = n->rb_left; |
f75bbfb4 | 536 | else if (blk >= cur->rd_data0 + cur->rd_data) |
66fc061b SW |
537 | next = n->rb_right; |
538 | if (next == NULL) { | |
b3b94faa | 539 | spin_unlock(&sdp->sd_rindex_spin); |
66fc061b SW |
540 | if (exact) { |
541 | if (blk < cur->rd_addr) | |
542 | return NULL; | |
543 | if (blk >= cur->rd_data0 + cur->rd_data) | |
544 | return NULL; | |
545 | } | |
7c9ca621 | 546 | return cur; |
b3b94faa | 547 | } |
66fc061b | 548 | n = next; |
b3b94faa | 549 | } |
b3b94faa DT |
550 | spin_unlock(&sdp->sd_rindex_spin); |
551 | ||
552 | return NULL; | |
553 | } | |
554 | ||
555 | /** | |
556 | * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem | |
557 | * @sdp: The GFS2 superblock | |
558 | * | |
559 | * Returns: The first rgrp in the filesystem | |
560 | */ | |
561 | ||
562 | struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) | |
563 | { | |
7c9ca621 BP |
564 | const struct rb_node *n; |
565 | struct gfs2_rgrpd *rgd; | |
566 | ||
8339ee54 | 567 | spin_lock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
568 | n = rb_first(&sdp->sd_rindex_tree); |
569 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
8339ee54 | 570 | spin_unlock(&sdp->sd_rindex_spin); |
7c9ca621 BP |
571 | |
572 | return rgd; | |
b3b94faa DT |
573 | } |
574 | ||
575 | /** | |
576 | * gfs2_rgrpd_get_next - get the next RG | |
886b1416 | 577 | * @rgd: the resource group descriptor |
b3b94faa DT |
578 | * |
579 | * Returns: The next rgrp | |
580 | */ | |
581 | ||
582 | struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) | |
583 | { | |
7c9ca621 BP |
584 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
585 | const struct rb_node *n; | |
586 | ||
587 | spin_lock(&sdp->sd_rindex_spin); | |
588 | n = rb_next(&rgd->rd_node); | |
589 | if (n == NULL) | |
590 | n = rb_first(&sdp->sd_rindex_tree); | |
591 | ||
592 | if (unlikely(&rgd->rd_node == n)) { | |
593 | spin_unlock(&sdp->sd_rindex_spin); | |
b3b94faa | 594 | return NULL; |
7c9ca621 BP |
595 | } |
596 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
597 | spin_unlock(&sdp->sd_rindex_spin); | |
598 | return rgd; | |
b3b94faa DT |
599 | } |
600 | ||
00a158be AD |
601 | void check_and_update_goal(struct gfs2_inode *ip) |
602 | { | |
603 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
604 | if (!ip->i_goal || gfs2_blk2rgrpd(sdp, ip->i_goal, 1) == NULL) | |
605 | ip->i_goal = ip->i_no_addr; | |
606 | } | |
607 | ||
8339ee54 SW |
608 | void gfs2_free_clones(struct gfs2_rgrpd *rgd) |
609 | { | |
610 | int x; | |
611 | ||
612 | for (x = 0; x < rgd->rd_length; x++) { | |
613 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
614 | kfree(bi->bi_clone); | |
615 | bi->bi_clone = NULL; | |
616 | } | |
617 | } | |
618 | ||
3792ce97 BP |
619 | static void dump_rs(struct seq_file *seq, const struct gfs2_blkreserv *rs, |
620 | const char *fs_id_buf) | |
8e2e0047 | 621 | { |
f85c10e2 BP |
622 | struct gfs2_inode *ip = container_of(rs, struct gfs2_inode, i_res); |
623 | ||
c65b76b8 AG |
624 | gfs2_print_dbg(seq, "%s B: n:%llu s:%llu f:%u\n", |
625 | fs_id_buf, | |
f85c10e2 | 626 | (unsigned long long)ip->i_no_addr, |
c65b76b8 | 627 | (unsigned long long)rs->rs_start, |
07974d2a | 628 | rs->rs_requested); |
8e2e0047 BP |
629 | } |
630 | ||
0a305e49 | 631 | /** |
8e2e0047 BP |
632 | * __rs_deltree - remove a multi-block reservation from the rgd tree |
633 | * @rs: The reservation to remove | |
634 | * | |
635 | */ | |
20095218 | 636 | static void __rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
637 | { |
638 | struct gfs2_rgrpd *rgd; | |
639 | ||
640 | if (!gfs2_rs_active(rs)) | |
641 | return; | |
642 | ||
c65b76b8 | 643 | rgd = rs->rs_rgd; |
9e733d39 | 644 | trace_gfs2_rs(rs, TRACE_RS_TREEDEL); |
4a993fb1 | 645 | rb_erase(&rs->rs_node, &rgd->rd_rstree); |
24d634e8 | 646 | RB_CLEAR_NODE(&rs->rs_node); |
8e2e0047 | 647 | |
07974d2a AG |
648 | if (rs->rs_requested) { |
649 | /* return requested blocks to the rgrp */ | |
650 | BUG_ON(rs->rs_rgd->rd_requested < rs->rs_requested); | |
651 | rs->rs_rgd->rd_requested -= rs->rs_requested; | |
a12c6fa1 | 652 | |
5ea5050c BP |
653 | /* The rgrp extent failure point is likely not to increase; |
654 | it will only do so if the freed blocks are somehow | |
655 | contiguous with a span of free blocks that follows. Still, | |
656 | it will force the number to be recalculated later. */ | |
07974d2a AG |
657 | rgd->rd_extfail_pt += rs->rs_requested; |
658 | rs->rs_requested = 0; | |
8e2e0047 | 659 | } |
8e2e0047 BP |
660 | } |
661 | ||
662 | /** | |
663 | * gfs2_rs_deltree - remove a multi-block reservation from the rgd tree | |
664 | * @rs: The reservation to remove | |
665 | * | |
666 | */ | |
20095218 | 667 | void gfs2_rs_deltree(struct gfs2_blkreserv *rs) |
8e2e0047 BP |
668 | { |
669 | struct gfs2_rgrpd *rgd; | |
670 | ||
c65b76b8 | 671 | rgd = rs->rs_rgd; |
4a993fb1 SW |
672 | if (rgd) { |
673 | spin_lock(&rgd->rd_rsspin); | |
20095218 | 674 | __rs_deltree(rs); |
07974d2a | 675 | BUG_ON(rs->rs_requested); |
4a993fb1 SW |
676 | spin_unlock(&rgd->rd_rsspin); |
677 | } | |
8e2e0047 BP |
678 | } |
679 | ||
680 | /** | |
1595548f | 681 | * gfs2_rs_delete - delete a multi-block reservation |
0a305e49 | 682 | * @ip: The inode for this reservation |
af5c2697 | 683 | * @wcount: The inode's write count, or NULL |
0a305e49 BP |
684 | * |
685 | */ | |
1595548f | 686 | void gfs2_rs_delete(struct gfs2_inode *ip, atomic_t *wcount) |
0a305e49 BP |
687 | { |
688 | down_write(&ip->i_rw_mutex); | |
44f52122 | 689 | if ((wcount == NULL) || (atomic_read(wcount) <= 1)) |
a097dc7e | 690 | gfs2_rs_deltree(&ip->i_res); |
0a305e49 BP |
691 | up_write(&ip->i_rw_mutex); |
692 | } | |
693 | ||
8e2e0047 BP |
694 | /** |
695 | * return_all_reservations - return all reserved blocks back to the rgrp. | |
696 | * @rgd: the rgrp that needs its space back | |
697 | * | |
698 | * We previously reserved a bunch of blocks for allocation. Now we need to | |
699 | * give them back. This leave the reservation structures in tact, but removes | |
700 | * all of their corresponding "no-fly zones". | |
701 | */ | |
702 | static void return_all_reservations(struct gfs2_rgrpd *rgd) | |
703 | { | |
704 | struct rb_node *n; | |
705 | struct gfs2_blkreserv *rs; | |
706 | ||
707 | spin_lock(&rgd->rd_rsspin); | |
708 | while ((n = rb_first(&rgd->rd_rstree))) { | |
709 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
20095218 | 710 | __rs_deltree(rs); |
8e2e0047 BP |
711 | } |
712 | spin_unlock(&rgd->rd_rsspin); | |
713 | } | |
714 | ||
8339ee54 | 715 | void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) |
b3b94faa | 716 | { |
7c9ca621 | 717 | struct rb_node *n; |
b3b94faa DT |
718 | struct gfs2_rgrpd *rgd; |
719 | struct gfs2_glock *gl; | |
720 | ||
7c9ca621 BP |
721 | while ((n = rb_first(&sdp->sd_rindex_tree))) { |
722 | rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); | |
b3b94faa DT |
723 | gl = rgd->rd_gl; |
724 | ||
7c9ca621 | 725 | rb_erase(n, &sdp->sd_rindex_tree); |
b3b94faa DT |
726 | |
727 | if (gl) { | |
b3422cac BP |
728 | if (gl->gl_state != LM_ST_UNLOCKED) { |
729 | gfs2_glock_cb(gl, LM_ST_UNLOCKED); | |
730 | flush_delayed_work(&gl->gl_work); | |
731 | } | |
10283ea5 | 732 | gfs2_rgrp_brelse(rgd); |
b3422cac | 733 | glock_clear_object(gl, rgd); |
b3b94faa DT |
734 | gfs2_glock_put(gl); |
735 | } | |
736 | ||
8339ee54 | 737 | gfs2_free_clones(rgd); |
d0f17d38 | 738 | return_all_reservations(rgd); |
b3b94faa | 739 | kfree(rgd->rd_bits); |
36e4ad03 | 740 | rgd->rd_bits = NULL; |
6bdd9be6 | 741 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
b3b94faa DT |
742 | } |
743 | } | |
744 | ||
b3b94faa | 745 | /** |
c551f66c | 746 | * compute_bitstructs - Compute the bitmap sizes |
b3b94faa DT |
747 | * @rgd: The resource group descriptor |
748 | * | |
749 | * Calculates bitmap descriptors, one for each block that contains bitmap data | |
750 | * | |
751 | * Returns: errno | |
752 | */ | |
753 | ||
754 | static int compute_bitstructs(struct gfs2_rgrpd *rgd) | |
755 | { | |
756 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
757 | struct gfs2_bitmap *bi; | |
bb8d8a6f | 758 | u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ |
cd915493 | 759 | u32 bytes_left, bytes; |
b3b94faa DT |
760 | int x; |
761 | ||
feaa7bba SW |
762 | if (!length) |
763 | return -EINVAL; | |
764 | ||
dd894be8 | 765 | rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); |
b3b94faa DT |
766 | if (!rgd->rd_bits) |
767 | return -ENOMEM; | |
768 | ||
bb8d8a6f | 769 | bytes_left = rgd->rd_bitbytes; |
b3b94faa DT |
770 | |
771 | for (x = 0; x < length; x++) { | |
772 | bi = rgd->rd_bits + x; | |
773 | ||
60a0b8f9 | 774 | bi->bi_flags = 0; |
b3b94faa DT |
775 | /* small rgrp; bitmap stored completely in header block */ |
776 | if (length == 1) { | |
777 | bytes = bytes_left; | |
778 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
779 | bi->bi_start = 0; | |
281b4952 | 780 | bi->bi_bytes = bytes; |
7e230f57 | 781 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
782 | /* header block */ |
783 | } else if (x == 0) { | |
784 | bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); | |
785 | bi->bi_offset = sizeof(struct gfs2_rgrp); | |
786 | bi->bi_start = 0; | |
281b4952 | 787 | bi->bi_bytes = bytes; |
7e230f57 | 788 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
789 | /* last block */ |
790 | } else if (x + 1 == length) { | |
791 | bytes = bytes_left; | |
792 | bi->bi_offset = sizeof(struct gfs2_meta_header); | |
bb8d8a6f | 793 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
281b4952 | 794 | bi->bi_bytes = bytes; |
7e230f57 | 795 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
796 | /* other blocks */ |
797 | } else { | |
568f4c96 SW |
798 | bytes = sdp->sd_sb.sb_bsize - |
799 | sizeof(struct gfs2_meta_header); | |
b3b94faa | 800 | bi->bi_offset = sizeof(struct gfs2_meta_header); |
bb8d8a6f | 801 | bi->bi_start = rgd->rd_bitbytes - bytes_left; |
281b4952 | 802 | bi->bi_bytes = bytes; |
7e230f57 | 803 | bi->bi_blocks = bytes * GFS2_NBBY; |
b3b94faa DT |
804 | } |
805 | ||
806 | bytes_left -= bytes; | |
807 | } | |
808 | ||
809 | if (bytes_left) { | |
810 | gfs2_consist_rgrpd(rgd); | |
811 | return -EIO; | |
812 | } | |
813 | bi = rgd->rd_bits + (length - 1); | |
281b4952 | 814 | if ((bi->bi_start + bi->bi_bytes) * GFS2_NBBY != rgd->rd_data) { |
8dc88ac6 AG |
815 | gfs2_lm(sdp, |
816 | "ri_addr = %llu\n" | |
817 | "ri_length = %u\n" | |
818 | "ri_data0 = %llu\n" | |
819 | "ri_data = %u\n" | |
820 | "ri_bitbytes = %u\n" | |
821 | "start=%u len=%u offset=%u\n", | |
822 | (unsigned long long)rgd->rd_addr, | |
823 | rgd->rd_length, | |
824 | (unsigned long long)rgd->rd_data0, | |
825 | rgd->rd_data, | |
826 | rgd->rd_bitbytes, | |
827 | bi->bi_start, bi->bi_bytes, bi->bi_offset); | |
828 | gfs2_consist_rgrpd(rgd); | |
b3b94faa DT |
829 | return -EIO; |
830 | } | |
831 | ||
832 | return 0; | |
833 | } | |
834 | ||
7ae8fa84 RP |
835 | /** |
836 | * gfs2_ri_total - Total up the file system space, according to the rindex. | |
886b1416 | 837 | * @sdp: the filesystem |
7ae8fa84 RP |
838 | * |
839 | */ | |
840 | u64 gfs2_ri_total(struct gfs2_sbd *sdp) | |
841 | { | |
842 | u64 total_data = 0; | |
843 | struct inode *inode = sdp->sd_rindex; | |
844 | struct gfs2_inode *ip = GFS2_I(inode); | |
7ae8fa84 | 845 | char buf[sizeof(struct gfs2_rindex)]; |
7ae8fa84 RP |
846 | int error, rgrps; |
847 | ||
7ae8fa84 RP |
848 | for (rgrps = 0;; rgrps++) { |
849 | loff_t pos = rgrps * sizeof(struct gfs2_rindex); | |
850 | ||
bcd7278d | 851 | if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) |
7ae8fa84 | 852 | break; |
4306629e | 853 | error = gfs2_internal_read(ip, buf, &pos, |
7ae8fa84 RP |
854 | sizeof(struct gfs2_rindex)); |
855 | if (error != sizeof(struct gfs2_rindex)) | |
856 | break; | |
bb8d8a6f | 857 | total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); |
7ae8fa84 | 858 | } |
7ae8fa84 RP |
859 | return total_data; |
860 | } | |
861 | ||
6aad1c3d | 862 | static int rgd_insert(struct gfs2_rgrpd *rgd) |
7c9ca621 BP |
863 | { |
864 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
865 | struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; | |
866 | ||
867 | /* Figure out where to put new node */ | |
868 | while (*newn) { | |
869 | struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, | |
870 | rd_node); | |
871 | ||
872 | parent = *newn; | |
873 | if (rgd->rd_addr < cur->rd_addr) | |
874 | newn = &((*newn)->rb_left); | |
875 | else if (rgd->rd_addr > cur->rd_addr) | |
876 | newn = &((*newn)->rb_right); | |
877 | else | |
6aad1c3d | 878 | return -EEXIST; |
7c9ca621 BP |
879 | } |
880 | ||
881 | rb_link_node(&rgd->rd_node, parent, newn); | |
882 | rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); | |
6aad1c3d BP |
883 | sdp->sd_rgrps++; |
884 | return 0; | |
7c9ca621 BP |
885 | } |
886 | ||
b3b94faa | 887 | /** |
6c53267f | 888 | * read_rindex_entry - Pull in a new resource index entry from the disk |
4306629e | 889 | * @ip: Pointer to the rindex inode |
b3b94faa | 890 | * |
8339ee54 | 891 | * Returns: 0 on success, > 0 on EOF, error code otherwise |
6c53267f RP |
892 | */ |
893 | ||
4306629e | 894 | static int read_rindex_entry(struct gfs2_inode *ip) |
6c53267f RP |
895 | { |
896 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
897 | loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); | |
8339ee54 | 898 | struct gfs2_rindex buf; |
6c53267f RP |
899 | int error; |
900 | struct gfs2_rgrpd *rgd; | |
901 | ||
8339ee54 SW |
902 | if (pos >= i_size_read(&ip->i_inode)) |
903 | return 1; | |
904 | ||
4306629e | 905 | error = gfs2_internal_read(ip, (char *)&buf, &pos, |
6c53267f | 906 | sizeof(struct gfs2_rindex)); |
8339ee54 SW |
907 | |
908 | if (error != sizeof(struct gfs2_rindex)) | |
909 | return (error == 0) ? 1 : error; | |
6c53267f | 910 | |
6bdd9be6 | 911 | rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); |
6c53267f RP |
912 | error = -ENOMEM; |
913 | if (!rgd) | |
914 | return error; | |
915 | ||
6c53267f | 916 | rgd->rd_sbd = sdp; |
8339ee54 SW |
917 | rgd->rd_addr = be64_to_cpu(buf.ri_addr); |
918 | rgd->rd_length = be32_to_cpu(buf.ri_length); | |
919 | rgd->rd_data0 = be64_to_cpu(buf.ri_data0); | |
920 | rgd->rd_data = be32_to_cpu(buf.ri_data); | |
921 | rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); | |
8e2e0047 | 922 | spin_lock_init(&rgd->rd_rsspin); |
9e514605 | 923 | mutex_init(&rgd->rd_mutex); |
7c9ca621 | 924 | |
6c53267f RP |
925 | error = compute_bitstructs(rgd); |
926 | if (error) | |
8339ee54 | 927 | goto fail; |
6c53267f | 928 | |
bb8d8a6f | 929 | error = gfs2_glock_get(sdp, rgd->rd_addr, |
6c53267f RP |
930 | &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); |
931 | if (error) | |
8339ee54 | 932 | goto fail; |
6c53267f | 933 | |
4e2f8849 | 934 | rgd->rd_rgl = (struct gfs2_rgrp_lvb *)rgd->rd_gl->gl_lksb.sb_lvbptr; |
0e27c18c | 935 | rgd->rd_flags &= ~(GFS2_RDF_UPTODATE | GFS2_RDF_PREFERRED); |
7c9ca621 BP |
936 | if (rgd->rd_data > sdp->sd_max_rg_data) |
937 | sdp->sd_max_rg_data = rgd->rd_data; | |
8339ee54 | 938 | spin_lock(&sdp->sd_rindex_spin); |
6aad1c3d | 939 | error = rgd_insert(rgd); |
8339ee54 | 940 | spin_unlock(&sdp->sd_rindex_spin); |
36e4ad03 | 941 | if (!error) { |
6f6597ba | 942 | glock_set_object(rgd->rd_gl, rgd); |
6aad1c3d | 943 | return 0; |
36e4ad03 | 944 | } |
6aad1c3d BP |
945 | |
946 | error = 0; /* someone else read in the rgrp; free it and ignore it */ | |
c1ac539e | 947 | gfs2_glock_put(rgd->rd_gl); |
8339ee54 SW |
948 | |
949 | fail: | |
950 | kfree(rgd->rd_bits); | |
36e4ad03 | 951 | rgd->rd_bits = NULL; |
8339ee54 | 952 | kmem_cache_free(gfs2_rgrpd_cachep, rgd); |
6c53267f RP |
953 | return error; |
954 | } | |
955 | ||
0e27c18c BP |
956 | /** |
957 | * set_rgrp_preferences - Run all the rgrps, selecting some we prefer to use | |
958 | * @sdp: the GFS2 superblock | |
959 | * | |
960 | * The purpose of this function is to select a subset of the resource groups | |
961 | * and mark them as PREFERRED. We do it in such a way that each node prefers | |
962 | * to use a unique set of rgrps to minimize glock contention. | |
963 | */ | |
964 | static void set_rgrp_preferences(struct gfs2_sbd *sdp) | |
965 | { | |
966 | struct gfs2_rgrpd *rgd, *first; | |
967 | int i; | |
968 | ||
969 | /* Skip an initial number of rgrps, based on this node's journal ID. | |
970 | That should start each node out on its own set. */ | |
971 | rgd = gfs2_rgrpd_get_first(sdp); | |
972 | for (i = 0; i < sdp->sd_lockstruct.ls_jid; i++) | |
973 | rgd = gfs2_rgrpd_get_next(rgd); | |
974 | first = rgd; | |
975 | ||
976 | do { | |
977 | rgd->rd_flags |= GFS2_RDF_PREFERRED; | |
978 | for (i = 0; i < sdp->sd_journals; i++) { | |
979 | rgd = gfs2_rgrpd_get_next(rgd); | |
959b6717 | 980 | if (!rgd || rgd == first) |
0e27c18c BP |
981 | break; |
982 | } | |
959b6717 | 983 | } while (rgd && rgd != first); |
0e27c18c BP |
984 | } |
985 | ||
6c53267f RP |
986 | /** |
987 | * gfs2_ri_update - Pull in a new resource index from the disk | |
988 | * @ip: pointer to the rindex inode | |
989 | * | |
b3b94faa DT |
990 | * Returns: 0 on successful update, error code otherwise |
991 | */ | |
992 | ||
8339ee54 | 993 | static int gfs2_ri_update(struct gfs2_inode *ip) |
b3b94faa | 994 | { |
feaa7bba | 995 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
996 | int error; |
997 | ||
8339ee54 | 998 | do { |
4306629e | 999 | error = read_rindex_entry(ip); |
8339ee54 SW |
1000 | } while (error == 0); |
1001 | ||
1002 | if (error < 0) | |
1003 | return error; | |
b3b94faa | 1004 | |
77872151 BP |
1005 | if (RB_EMPTY_ROOT(&sdp->sd_rindex_tree)) { |
1006 | fs_err(sdp, "no resource groups found in the file system.\n"); | |
1007 | return -ENOENT; | |
1008 | } | |
0e27c18c BP |
1009 | set_rgrp_preferences(sdp); |
1010 | ||
cf45b752 | 1011 | sdp->sd_rindex_uptodate = 1; |
6c53267f RP |
1012 | return 0; |
1013 | } | |
b3b94faa | 1014 | |
b3b94faa | 1015 | /** |
8339ee54 | 1016 | * gfs2_rindex_update - Update the rindex if required |
b3b94faa | 1017 | * @sdp: The GFS2 superblock |
b3b94faa DT |
1018 | * |
1019 | * We grab a lock on the rindex inode to make sure that it doesn't | |
1020 | * change whilst we are performing an operation. We keep this lock | |
1021 | * for quite long periods of time compared to other locks. This | |
1022 | * doesn't matter, since it is shared and it is very, very rarely | |
1023 | * accessed in the exclusive mode (i.e. only when expanding the filesystem). | |
1024 | * | |
1025 | * This makes sure that we're using the latest copy of the resource index | |
1026 | * special file, which might have been updated if someone expanded the | |
1027 | * filesystem (via gfs2_grow utility), which adds new resource groups. | |
1028 | * | |
8339ee54 | 1029 | * Returns: 0 on succeess, error code otherwise |
b3b94faa DT |
1030 | */ |
1031 | ||
8339ee54 | 1032 | int gfs2_rindex_update(struct gfs2_sbd *sdp) |
b3b94faa | 1033 | { |
feaa7bba | 1034 | struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); |
b3b94faa | 1035 | struct gfs2_glock *gl = ip->i_gl; |
8339ee54 SW |
1036 | struct gfs2_holder ri_gh; |
1037 | int error = 0; | |
a365fbf3 | 1038 | int unlock_required = 0; |
b3b94faa DT |
1039 | |
1040 | /* Read new copy from disk if we don't have the latest */ | |
cf45b752 | 1041 | if (!sdp->sd_rindex_uptodate) { |
a365fbf3 SW |
1042 | if (!gfs2_glock_is_locked_by_me(gl)) { |
1043 | error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); | |
1044 | if (error) | |
6aad1c3d | 1045 | return error; |
a365fbf3 SW |
1046 | unlock_required = 1; |
1047 | } | |
8339ee54 | 1048 | if (!sdp->sd_rindex_uptodate) |
b3b94faa | 1049 | error = gfs2_ri_update(ip); |
a365fbf3 SW |
1050 | if (unlock_required) |
1051 | gfs2_glock_dq_uninit(&ri_gh); | |
b3b94faa DT |
1052 | } |
1053 | ||
1054 | return error; | |
1055 | } | |
1056 | ||
42d52e38 | 1057 | static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) |
bb8d8a6f SW |
1058 | { |
1059 | const struct gfs2_rgrp *str = buf; | |
42d52e38 | 1060 | u32 rg_flags; |
bb8d8a6f | 1061 | |
42d52e38 | 1062 | rg_flags = be32_to_cpu(str->rg_flags); |
09010978 | 1063 | rg_flags &= ~GFS2_RDF_MASK; |
1ce97e56 SW |
1064 | rgd->rd_flags &= GFS2_RDF_MASK; |
1065 | rgd->rd_flags |= rg_flags; | |
cfc8b549 | 1066 | rgd->rd_free = be32_to_cpu(str->rg_free); |
73f74948 | 1067 | rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); |
d8b71f73 | 1068 | rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); |
166725d9 | 1069 | /* rd_data0, rd_data and rd_bitbytes already set from rindex */ |
bb8d8a6f SW |
1070 | } |
1071 | ||
3f30f929 BP |
1072 | static void gfs2_rgrp_ondisk2lvb(struct gfs2_rgrp_lvb *rgl, const void *buf) |
1073 | { | |
1074 | const struct gfs2_rgrp *str = buf; | |
1075 | ||
1076 | rgl->rl_magic = cpu_to_be32(GFS2_MAGIC); | |
1077 | rgl->rl_flags = str->rg_flags; | |
1078 | rgl->rl_free = str->rg_free; | |
1079 | rgl->rl_dinodes = str->rg_dinodes; | |
1080 | rgl->rl_igeneration = str->rg_igeneration; | |
1081 | rgl->__pad = 0UL; | |
1082 | } | |
1083 | ||
42d52e38 | 1084 | static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) |
bb8d8a6f | 1085 | { |
65adc273 | 1086 | struct gfs2_rgrpd *next = gfs2_rgrpd_get_next(rgd); |
bb8d8a6f | 1087 | struct gfs2_rgrp *str = buf; |
850d2d91 | 1088 | u32 crc; |
bb8d8a6f | 1089 | |
09010978 | 1090 | str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); |
cfc8b549 | 1091 | str->rg_free = cpu_to_be32(rgd->rd_free); |
73f74948 | 1092 | str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); |
65adc273 AP |
1093 | if (next == NULL) |
1094 | str->rg_skip = 0; | |
1095 | else if (next->rd_addr > rgd->rd_addr) | |
1096 | str->rg_skip = cpu_to_be32(next->rd_addr - rgd->rd_addr); | |
d8b71f73 | 1097 | str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); |
166725d9 AP |
1098 | str->rg_data0 = cpu_to_be64(rgd->rd_data0); |
1099 | str->rg_data = cpu_to_be32(rgd->rd_data); | |
1100 | str->rg_bitbytes = cpu_to_be32(rgd->rd_bitbytes); | |
850d2d91 AP |
1101 | str->rg_crc = 0; |
1102 | crc = gfs2_disk_hash(buf, sizeof(struct gfs2_rgrp)); | |
1103 | str->rg_crc = cpu_to_be32(crc); | |
166725d9 | 1104 | |
bb8d8a6f | 1105 | memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); |
3f30f929 | 1106 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, buf); |
bb8d8a6f SW |
1107 | } |
1108 | ||
90306c41 BM |
1109 | static int gfs2_rgrp_lvb_valid(struct gfs2_rgrpd *rgd) |
1110 | { | |
1111 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
1112 | struct gfs2_rgrp *str = (struct gfs2_rgrp *)rgd->rd_bits[0].bi_bh->b_data; | |
f29e62ee | 1113 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
72244b6b | 1114 | int valid = 1; |
90306c41 | 1115 | |
72244b6b | 1116 | if (rgl->rl_flags != str->rg_flags) { |
f29e62ee BP |
1117 | fs_warn(sdp, "GFS2: rgd: %llu lvb flag mismatch %u/%u", |
1118 | (unsigned long long)rgd->rd_addr, | |
72244b6b BP |
1119 | be32_to_cpu(rgl->rl_flags), be32_to_cpu(str->rg_flags)); |
1120 | valid = 0; | |
1121 | } | |
1122 | if (rgl->rl_free != str->rg_free) { | |
f29e62ee BP |
1123 | fs_warn(sdp, "GFS2: rgd: %llu lvb free mismatch %u/%u", |
1124 | (unsigned long long)rgd->rd_addr, | |
1125 | be32_to_cpu(rgl->rl_free), be32_to_cpu(str->rg_free)); | |
72244b6b BP |
1126 | valid = 0; |
1127 | } | |
1128 | if (rgl->rl_dinodes != str->rg_dinodes) { | |
f29e62ee BP |
1129 | fs_warn(sdp, "GFS2: rgd: %llu lvb dinode mismatch %u/%u", |
1130 | (unsigned long long)rgd->rd_addr, | |
1131 | be32_to_cpu(rgl->rl_dinodes), | |
1132 | be32_to_cpu(str->rg_dinodes)); | |
72244b6b BP |
1133 | valid = 0; |
1134 | } | |
1135 | if (rgl->rl_igeneration != str->rg_igeneration) { | |
f29e62ee BP |
1136 | fs_warn(sdp, "GFS2: rgd: %llu lvb igen mismatch %llu/%llu", |
1137 | (unsigned long long)rgd->rd_addr, | |
1138 | (unsigned long long)be64_to_cpu(rgl->rl_igeneration), | |
1139 | (unsigned long long)be64_to_cpu(str->rg_igeneration)); | |
72244b6b BP |
1140 | valid = 0; |
1141 | } | |
1142 | return valid; | |
90306c41 BM |
1143 | } |
1144 | ||
90306c41 BM |
1145 | static u32 count_unlinked(struct gfs2_rgrpd *rgd) |
1146 | { | |
1147 | struct gfs2_bitmap *bi; | |
1148 | const u32 length = rgd->rd_length; | |
1149 | const u8 *buffer = NULL; | |
1150 | u32 i, goal, count = 0; | |
1151 | ||
1152 | for (i = 0, bi = rgd->rd_bits; i < length; i++, bi++) { | |
1153 | goal = 0; | |
1154 | buffer = bi->bi_bh->b_data + bi->bi_offset; | |
1155 | WARN_ON(!buffer_uptodate(bi->bi_bh)); | |
281b4952 AG |
1156 | while (goal < bi->bi_blocks) { |
1157 | goal = gfs2_bitfit(buffer, bi->bi_bytes, goal, | |
90306c41 BM |
1158 | GFS2_BLKST_UNLINKED); |
1159 | if (goal == BFITNOENT) | |
1160 | break; | |
1161 | count++; | |
1162 | goal++; | |
1163 | } | |
1164 | } | |
1165 | ||
1166 | return count; | |
1167 | } | |
1168 | ||
560b8eba AG |
1169 | static void rgrp_set_bitmap_flags(struct gfs2_rgrpd *rgd) |
1170 | { | |
1171 | struct gfs2_bitmap *bi; | |
1172 | int x; | |
1173 | ||
1174 | if (rgd->rd_free) { | |
1175 | for (x = 0; x < rgd->rd_length; x++) { | |
1176 | bi = rgd->rd_bits + x; | |
1177 | clear_bit(GBF_FULL, &bi->bi_flags); | |
1178 | } | |
1179 | } else { | |
1180 | for (x = 0; x < rgd->rd_length; x++) { | |
1181 | bi = rgd->rd_bits + x; | |
1182 | set_bit(GBF_FULL, &bi->bi_flags); | |
1183 | } | |
1184 | } | |
1185 | } | |
90306c41 | 1186 | |
b3b94faa | 1187 | /** |
90306c41 BM |
1188 | * gfs2_rgrp_bh_get - Read in a RG's header and bitmaps |
1189 | * @rgd: the struct gfs2_rgrpd describing the RG to read in | |
b3b94faa DT |
1190 | * |
1191 | * Read in all of a Resource Group's header and bitmap blocks. | |
10283ea5 | 1192 | * Caller must eventually call gfs2_rgrp_brelse() to free the bitmaps. |
b3b94faa DT |
1193 | * |
1194 | * Returns: errno | |
1195 | */ | |
1196 | ||
c2b0b30e | 1197 | static int gfs2_rgrp_bh_get(struct gfs2_rgrpd *rgd) |
b3b94faa DT |
1198 | { |
1199 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1200 | struct gfs2_glock *gl = rgd->rd_gl; | |
bb8d8a6f | 1201 | unsigned int length = rgd->rd_length; |
b3b94faa DT |
1202 | struct gfs2_bitmap *bi; |
1203 | unsigned int x, y; | |
1204 | int error; | |
1205 | ||
90306c41 BM |
1206 | if (rgd->rd_bits[0].bi_bh != NULL) |
1207 | return 0; | |
1208 | ||
b3b94faa DT |
1209 | for (x = 0; x < length; x++) { |
1210 | bi = rgd->rd_bits + x; | |
c8d57703 | 1211 | error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, 0, &bi->bi_bh); |
b3b94faa DT |
1212 | if (error) |
1213 | goto fail; | |
1214 | } | |
1215 | ||
1216 | for (y = length; y--;) { | |
1217 | bi = rgd->rd_bits + y; | |
7276b3b0 | 1218 | error = gfs2_meta_wait(sdp, bi->bi_bh); |
b3b94faa DT |
1219 | if (error) |
1220 | goto fail; | |
feaa7bba | 1221 | if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : |
b3b94faa DT |
1222 | GFS2_METATYPE_RG)) { |
1223 | error = -EIO; | |
1224 | goto fail; | |
1225 | } | |
1226 | } | |
1227 | ||
cf45b752 | 1228 | if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { |
42d52e38 | 1229 | gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); |
560b8eba | 1230 | rgrp_set_bitmap_flags(rgd); |
1ce97e56 | 1231 | rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); |
7c9ca621 | 1232 | rgd->rd_free_clone = rgd->rd_free; |
725d0e9d | 1233 | BUG_ON(rgd->rd_reserved); |
5ea5050c BP |
1234 | /* max out the rgrp allocation failure point */ |
1235 | rgd->rd_extfail_pt = rgd->rd_free; | |
b3b94faa | 1236 | } |
951b4bd5 | 1237 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) { |
90306c41 BM |
1238 | rgd->rd_rgl->rl_unlinked = cpu_to_be32(count_unlinked(rgd)); |
1239 | gfs2_rgrp_ondisk2lvb(rgd->rd_rgl, | |
1240 | rgd->rd_bits[0].bi_bh->b_data); | |
1241 | } | |
1242 | else if (sdp->sd_args.ar_rgrplvb) { | |
1243 | if (!gfs2_rgrp_lvb_valid(rgd)){ | |
1244 | gfs2_consist_rgrpd(rgd); | |
1245 | error = -EIO; | |
1246 | goto fail; | |
1247 | } | |
1248 | if (rgd->rd_rgl->rl_unlinked == 0) | |
1249 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1250 | } | |
b3b94faa DT |
1251 | return 0; |
1252 | ||
feaa7bba | 1253 | fail: |
b3b94faa DT |
1254 | while (x--) { |
1255 | bi = rgd->rd_bits + x; | |
1256 | brelse(bi->bi_bh); | |
1257 | bi->bi_bh = NULL; | |
1258 | gfs2_assert_warn(sdp, !bi->bi_clone); | |
1259 | } | |
b3b94faa DT |
1260 | |
1261 | return error; | |
1262 | } | |
1263 | ||
c2b0b30e | 1264 | static int update_rgrp_lvb(struct gfs2_rgrpd *rgd) |
90306c41 BM |
1265 | { |
1266 | u32 rl_flags; | |
1267 | ||
1268 | if (rgd->rd_flags & GFS2_RDF_UPTODATE) | |
1269 | return 0; | |
1270 | ||
951b4bd5 | 1271 | if (cpu_to_be32(GFS2_MAGIC) != rgd->rd_rgl->rl_magic) |
90306c41 BM |
1272 | return gfs2_rgrp_bh_get(rgd); |
1273 | ||
1274 | rl_flags = be32_to_cpu(rgd->rd_rgl->rl_flags); | |
1275 | rl_flags &= ~GFS2_RDF_MASK; | |
1276 | rgd->rd_flags &= GFS2_RDF_MASK; | |
4f36cb36 | 1277 | rgd->rd_flags |= (rl_flags | GFS2_RDF_CHECK); |
90306c41 BM |
1278 | if (rgd->rd_rgl->rl_unlinked == 0) |
1279 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
1280 | rgd->rd_free = be32_to_cpu(rgd->rd_rgl->rl_free); | |
560b8eba | 1281 | rgrp_set_bitmap_flags(rgd); |
90306c41 | 1282 | rgd->rd_free_clone = rgd->rd_free; |
725d0e9d | 1283 | BUG_ON(rgd->rd_reserved); |
f38e998f AG |
1284 | /* max out the rgrp allocation failure point */ |
1285 | rgd->rd_extfail_pt = rgd->rd_free; | |
90306c41 BM |
1286 | rgd->rd_dinodes = be32_to_cpu(rgd->rd_rgl->rl_dinodes); |
1287 | rgd->rd_igeneration = be64_to_cpu(rgd->rd_rgl->rl_igeneration); | |
1288 | return 0; | |
1289 | } | |
1290 | ||
1291 | int gfs2_rgrp_go_lock(struct gfs2_holder *gh) | |
1292 | { | |
1293 | struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; | |
1294 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
1295 | ||
1296 | if (gh->gh_flags & GL_SKIP && sdp->sd_args.ar_rgrplvb) | |
1297 | return 0; | |
8b127d04 | 1298 | return gfs2_rgrp_bh_get(rgd); |
90306c41 BM |
1299 | } |
1300 | ||
b3b94faa | 1301 | /** |
39b0f1e9 BP |
1302 | * gfs2_rgrp_brelse - Release RG bitmaps read in with gfs2_rgrp_bh_get() |
1303 | * @rgd: The resource group | |
b3b94faa DT |
1304 | * |
1305 | */ | |
1306 | ||
39b0f1e9 | 1307 | void gfs2_rgrp_brelse(struct gfs2_rgrpd *rgd) |
b3b94faa | 1308 | { |
bb8d8a6f | 1309 | int x, length = rgd->rd_length; |
b3b94faa | 1310 | |
b3b94faa DT |
1311 | for (x = 0; x < length; x++) { |
1312 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
90306c41 BM |
1313 | if (bi->bi_bh) { |
1314 | brelse(bi->bi_bh); | |
1315 | bi->bi_bh = NULL; | |
1316 | } | |
b3b94faa | 1317 | } |
39b0f1e9 BP |
1318 | } |
1319 | ||
66fc061b | 1320 | int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, |
7c9ca621 | 1321 | struct buffer_head *bh, |
66fc061b | 1322 | const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) |
f15ab561 SW |
1323 | { |
1324 | struct super_block *sb = sdp->sd_vfs; | |
f15ab561 | 1325 | u64 blk; |
64d576ba | 1326 | sector_t start = 0; |
b2c87cae | 1327 | sector_t nr_blks = 0; |
f15ab561 SW |
1328 | int rv; |
1329 | unsigned int x; | |
66fc061b SW |
1330 | u32 trimmed = 0; |
1331 | u8 diff; | |
f15ab561 | 1332 | |
281b4952 | 1333 | for (x = 0; x < bi->bi_bytes; x++) { |
66fc061b SW |
1334 | const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; |
1335 | clone += bi->bi_offset; | |
1336 | clone += x; | |
1337 | if (bh) { | |
1338 | const u8 *orig = bh->b_data + bi->bi_offset + x; | |
1339 | diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); | |
1340 | } else { | |
1341 | diff = ~(*clone | (*clone >> 1)); | |
1342 | } | |
f15ab561 SW |
1343 | diff &= 0x55; |
1344 | if (diff == 0) | |
1345 | continue; | |
1346 | blk = offset + ((bi->bi_start + x) * GFS2_NBBY); | |
f15ab561 SW |
1347 | while(diff) { |
1348 | if (diff & 1) { | |
b2c87cae | 1349 | if (nr_blks == 0) |
f15ab561 | 1350 | goto start_new_extent; |
b2c87cae BP |
1351 | if ((start + nr_blks) != blk) { |
1352 | if (nr_blks >= minlen) { | |
1353 | rv = sb_issue_discard(sb, | |
1354 | start, nr_blks, | |
66fc061b SW |
1355 | GFP_NOFS, 0); |
1356 | if (rv) | |
1357 | goto fail; | |
b2c87cae | 1358 | trimmed += nr_blks; |
66fc061b | 1359 | } |
b2c87cae | 1360 | nr_blks = 0; |
f15ab561 SW |
1361 | start_new_extent: |
1362 | start = blk; | |
1363 | } | |
b2c87cae | 1364 | nr_blks++; |
f15ab561 SW |
1365 | } |
1366 | diff >>= 2; | |
b2c87cae | 1367 | blk++; |
f15ab561 SW |
1368 | } |
1369 | } | |
b2c87cae BP |
1370 | if (nr_blks >= minlen) { |
1371 | rv = sb_issue_discard(sb, start, nr_blks, GFP_NOFS, 0); | |
f15ab561 SW |
1372 | if (rv) |
1373 | goto fail; | |
b2c87cae | 1374 | trimmed += nr_blks; |
f15ab561 | 1375 | } |
66fc061b SW |
1376 | if (ptrimmed) |
1377 | *ptrimmed = trimmed; | |
1378 | return 0; | |
1379 | ||
f15ab561 | 1380 | fail: |
66fc061b | 1381 | if (sdp->sd_args.ar_discard) |
af38816e | 1382 | fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem\n", rv); |
f15ab561 | 1383 | sdp->sd_args.ar_discard = 0; |
66fc061b SW |
1384 | return -EIO; |
1385 | } | |
1386 | ||
1387 | /** | |
1388 | * gfs2_fitrim - Generate discard requests for unused bits of the filesystem | |
1389 | * @filp: Any file on the filesystem | |
1390 | * @argp: Pointer to the arguments (also used to pass result) | |
1391 | * | |
1392 | * Returns: 0 on success, otherwise error code | |
1393 | */ | |
1394 | ||
1395 | int gfs2_fitrim(struct file *filp, void __user *argp) | |
1396 | { | |
496ad9aa | 1397 | struct inode *inode = file_inode(filp); |
66fc061b SW |
1398 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1399 | struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); | |
1400 | struct buffer_head *bh; | |
1401 | struct gfs2_rgrpd *rgd; | |
1402 | struct gfs2_rgrpd *rgd_end; | |
1403 | struct gfs2_holder gh; | |
1404 | struct fstrim_range r; | |
1405 | int ret = 0; | |
1406 | u64 amt; | |
1407 | u64 trimmed = 0; | |
076f0faa | 1408 | u64 start, end, minlen; |
66fc061b | 1409 | unsigned int x; |
076f0faa | 1410 | unsigned bs_shift = sdp->sd_sb.sb_bsize_shift; |
66fc061b SW |
1411 | |
1412 | if (!capable(CAP_SYS_ADMIN)) | |
1413 | return -EPERM; | |
1414 | ||
c5c68724 BP |
1415 | if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) |
1416 | return -EROFS; | |
1417 | ||
66fc061b SW |
1418 | if (!blk_queue_discard(q)) |
1419 | return -EOPNOTSUPP; | |
1420 | ||
3a238ade | 1421 | if (copy_from_user(&r, argp, sizeof(r))) |
66fc061b SW |
1422 | return -EFAULT; |
1423 | ||
5e2f7d61 BP |
1424 | ret = gfs2_rindex_update(sdp); |
1425 | if (ret) | |
1426 | return ret; | |
1427 | ||
076f0faa LC |
1428 | start = r.start >> bs_shift; |
1429 | end = start + (r.len >> bs_shift); | |
1430 | minlen = max_t(u64, r.minlen, | |
1431 | q->limits.discard_granularity) >> bs_shift; | |
1432 | ||
6a98c333 AD |
1433 | if (end <= start || minlen > sdp->sd_max_rg_data) |
1434 | return -EINVAL; | |
1435 | ||
076f0faa | 1436 | rgd = gfs2_blk2rgrpd(sdp, start, 0); |
6a98c333 | 1437 | rgd_end = gfs2_blk2rgrpd(sdp, end, 0); |
076f0faa | 1438 | |
6a98c333 AD |
1439 | if ((gfs2_rgrpd_get_first(sdp) == gfs2_rgrpd_get_next(rgd_end)) |
1440 | && (start > rgd_end->rd_data0 + rgd_end->rd_data)) | |
1441 | return -EINVAL; /* start is beyond the end of the fs */ | |
66fc061b SW |
1442 | |
1443 | while (1) { | |
1444 | ||
4fc7ec31 BP |
1445 | ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1446 | LM_FLAG_NODE_SCOPE, &gh); | |
66fc061b SW |
1447 | if (ret) |
1448 | goto out; | |
1449 | ||
1450 | if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { | |
1451 | /* Trim each bitmap in the rgrp */ | |
1452 | for (x = 0; x < rgd->rd_length; x++) { | |
1453 | struct gfs2_bitmap *bi = rgd->rd_bits + x; | |
9e514605 | 1454 | rgrp_lock_local(rgd); |
076f0faa LC |
1455 | ret = gfs2_rgrp_send_discards(sdp, |
1456 | rgd->rd_data0, NULL, bi, minlen, | |
1457 | &amt); | |
9e514605 | 1458 | rgrp_unlock_local(rgd); |
66fc061b SW |
1459 | if (ret) { |
1460 | gfs2_glock_dq_uninit(&gh); | |
1461 | goto out; | |
1462 | } | |
1463 | trimmed += amt; | |
1464 | } | |
1465 | ||
1466 | /* Mark rgrp as having been trimmed */ | |
1467 | ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); | |
1468 | if (ret == 0) { | |
1469 | bh = rgd->rd_bits[0].bi_bh; | |
9e514605 | 1470 | rgrp_lock_local(rgd); |
66fc061b | 1471 | rgd->rd_flags |= GFS2_RGF_TRIMMED; |
350a9b0a | 1472 | gfs2_trans_add_meta(rgd->rd_gl, bh); |
66fc061b | 1473 | gfs2_rgrp_out(rgd, bh->b_data); |
9e514605 | 1474 | rgrp_unlock_local(rgd); |
66fc061b SW |
1475 | gfs2_trans_end(sdp); |
1476 | } | |
1477 | } | |
1478 | gfs2_glock_dq_uninit(&gh); | |
1479 | ||
1480 | if (rgd == rgd_end) | |
1481 | break; | |
1482 | ||
1483 | rgd = gfs2_rgrpd_get_next(rgd); | |
1484 | } | |
1485 | ||
1486 | out: | |
6a98c333 | 1487 | r.len = trimmed << bs_shift; |
3a238ade | 1488 | if (copy_to_user(argp, &r, sizeof(r))) |
66fc061b SW |
1489 | return -EFAULT; |
1490 | ||
1491 | return ret; | |
f15ab561 SW |
1492 | } |
1493 | ||
8e2e0047 BP |
1494 | /** |
1495 | * rs_insert - insert a new multi-block reservation into the rgrp's rb_tree | |
8e2e0047 | 1496 | * @ip: the inode structure |
8e2e0047 | 1497 | * |
8e2e0047 | 1498 | */ |
ff7f4cb4 | 1499 | static void rs_insert(struct gfs2_inode *ip) |
8e2e0047 BP |
1500 | { |
1501 | struct rb_node **newn, *parent = NULL; | |
1502 | int rc; | |
a097dc7e | 1503 | struct gfs2_blkreserv *rs = &ip->i_res; |
c65b76b8 | 1504 | struct gfs2_rgrpd *rgd = rs->rs_rgd; |
8e2e0047 | 1505 | |
8e2e0047 | 1506 | BUG_ON(gfs2_rs_active(rs)); |
c743ffd0 | 1507 | |
ff7f4cb4 SW |
1508 | spin_lock(&rgd->rd_rsspin); |
1509 | newn = &rgd->rd_rstree.rb_node; | |
8e2e0047 BP |
1510 | while (*newn) { |
1511 | struct gfs2_blkreserv *cur = | |
1512 | rb_entry(*newn, struct gfs2_blkreserv, rs_node); | |
1513 | ||
1514 | parent = *newn; | |
07974d2a | 1515 | rc = rs_cmp(rs->rs_start, rs->rs_requested, cur); |
8e2e0047 BP |
1516 | if (rc > 0) |
1517 | newn = &((*newn)->rb_right); | |
1518 | else if (rc < 0) | |
1519 | newn = &((*newn)->rb_left); | |
1520 | else { | |
1521 | spin_unlock(&rgd->rd_rsspin); | |
ff7f4cb4 SW |
1522 | WARN_ON(1); |
1523 | return; | |
8e2e0047 BP |
1524 | } |
1525 | } | |
1526 | ||
8e2e0047 BP |
1527 | rb_link_node(&rs->rs_node, parent, newn); |
1528 | rb_insert_color(&rs->rs_node, &rgd->rd_rstree); | |
1529 | ||
8e2e0047 | 1530 | /* Do our rgrp accounting for the reservation */ |
07974d2a | 1531 | rgd->rd_requested += rs->rs_requested; /* blocks requested */ |
8e2e0047 | 1532 | spin_unlock(&rgd->rd_rsspin); |
9e733d39 | 1533 | trace_gfs2_rs(rs, TRACE_RS_INSERT); |
8e2e0047 BP |
1534 | } |
1535 | ||
f6753df3 | 1536 | /** |
c551f66c | 1537 | * rgd_free - return the number of free blocks we can allocate |
f6753df3 | 1538 | * @rgd: the resource group |
c551f66c | 1539 | * @rs: The reservation to free |
f6753df3 BP |
1540 | * |
1541 | * This function returns the number of free blocks for an rgrp. | |
1542 | * That's the clone-free blocks (blocks that are free, not including those | |
1543 | * still being used for unlinked files that haven't been deleted.) | |
1544 | * | |
1545 | * It also subtracts any blocks reserved by someone else, but does not | |
1546 | * include free blocks that are still part of our current reservation, | |
1547 | * because obviously we can (and will) allocate them. | |
1548 | */ | |
1549 | static inline u32 rgd_free(struct gfs2_rgrpd *rgd, struct gfs2_blkreserv *rs) | |
1550 | { | |
1551 | u32 tot_reserved, tot_free; | |
1552 | ||
07974d2a | 1553 | if (WARN_ON_ONCE(rgd->rd_requested < rs->rs_requested)) |
f6753df3 | 1554 | return 0; |
07974d2a | 1555 | tot_reserved = rgd->rd_requested - rs->rs_requested; |
f6753df3 BP |
1556 | |
1557 | if (rgd->rd_free_clone < tot_reserved) | |
1558 | tot_reserved = 0; | |
1559 | ||
1560 | tot_free = rgd->rd_free_clone - tot_reserved; | |
1561 | ||
1562 | return tot_free; | |
1563 | } | |
1564 | ||
8e2e0047 | 1565 | /** |
ff7f4cb4 | 1566 | * rg_mblk_search - find a group of multiple free blocks to form a reservation |
8e2e0047 | 1567 | * @rgd: the resource group descriptor |
8e2e0047 | 1568 | * @ip: pointer to the inode for which we're reserving blocks |
7b9cff46 | 1569 | * @ap: the allocation parameters |
8e2e0047 | 1570 | * |
8e2e0047 BP |
1571 | */ |
1572 | ||
ff7f4cb4 | 1573 | static void rg_mblk_search(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip, |
7b9cff46 | 1574 | const struct gfs2_alloc_parms *ap) |
8e2e0047 | 1575 | { |
ff7f4cb4 SW |
1576 | struct gfs2_rbm rbm = { .rgd = rgd, }; |
1577 | u64 goal; | |
a097dc7e | 1578 | struct gfs2_blkreserv *rs = &ip->i_res; |
ff7f4cb4 | 1579 | u32 extlen; |
725d0e9d | 1580 | u32 free_blocks, blocks_available; |
ff7f4cb4 | 1581 | int ret; |
af21ca8e | 1582 | struct inode *inode = &ip->i_inode; |
8e2e0047 | 1583 | |
725d0e9d AG |
1584 | spin_lock(&rgd->rd_rsspin); |
1585 | free_blocks = rgd_free(rgd, rs); | |
1586 | if (rgd->rd_free_clone < rgd->rd_requested) | |
1587 | free_blocks = 0; | |
1588 | blocks_available = rgd->rd_free_clone - rgd->rd_reserved; | |
1589 | if (rgd == rs->rs_rgd) | |
1590 | blocks_available += rs->rs_reserved; | |
1591 | spin_unlock(&rgd->rd_rsspin); | |
1592 | ||
af21ca8e BP |
1593 | if (S_ISDIR(inode->i_mode)) |
1594 | extlen = 1; | |
1595 | else { | |
21f09c43 | 1596 | extlen = max_t(u32, atomic_read(&ip->i_sizehint), ap->target); |
ad899458 | 1597 | extlen = clamp(extlen, (u32)RGRP_RSRV_MINBLKS, free_blocks); |
af21ca8e | 1598 | } |
725d0e9d | 1599 | if (free_blocks < extlen || blocks_available < extlen) |
c743ffd0 SW |
1600 | return; |
1601 | ||
8e2e0047 BP |
1602 | /* Find bitmap block that contains bits for goal block */ |
1603 | if (rgrp_contains_block(rgd, ip->i_goal)) | |
ff7f4cb4 | 1604 | goal = ip->i_goal; |
8e2e0047 | 1605 | else |
ff7f4cb4 | 1606 | goal = rgd->rd_last_alloc + rgd->rd_data0; |
8e2e0047 | 1607 | |
ff7f4cb4 SW |
1608 | if (WARN_ON(gfs2_rbm_from_block(&rbm, goal))) |
1609 | return; | |
8e2e0047 | 1610 | |
3d39fcd1 | 1611 | ret = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &extlen, &ip->i_res, true); |
ff7f4cb4 | 1612 | if (ret == 0) { |
c65b76b8 | 1613 | rs->rs_start = gfs2_rbm_to_block(&rbm); |
07974d2a | 1614 | rs->rs_requested = extlen; |
ff7f4cb4 | 1615 | rs_insert(ip); |
13d2eb01 BP |
1616 | } else { |
1617 | if (goal == rgd->rd_last_alloc + rgd->rd_data0) | |
1618 | rgd->rd_last_alloc = 0; | |
8e2e0047 | 1619 | } |
b3e47ca0 BP |
1620 | } |
1621 | ||
5b924ae2 SW |
1622 | /** |
1623 | * gfs2_next_unreserved_block - Return next block that is not reserved | |
1624 | * @rgd: The resource group | |
1625 | * @block: The starting block | |
ff7f4cb4 | 1626 | * @length: The required length |
3d39fcd1 | 1627 | * @ignore_rs: Reservation to ignore |
5b924ae2 SW |
1628 | * |
1629 | * If the block does not appear in any reservation, then return the | |
1630 | * block number unchanged. If it does appear in the reservation, then | |
1631 | * keep looking through the tree of reservations in order to find the | |
1632 | * first block number which is not reserved. | |
1633 | */ | |
1634 | ||
1635 | static u64 gfs2_next_unreserved_block(struct gfs2_rgrpd *rgd, u64 block, | |
ff7f4cb4 | 1636 | u32 length, |
3d39fcd1 | 1637 | struct gfs2_blkreserv *ignore_rs) |
5b924ae2 SW |
1638 | { |
1639 | struct gfs2_blkreserv *rs; | |
1640 | struct rb_node *n; | |
1641 | int rc; | |
1642 | ||
1643 | spin_lock(&rgd->rd_rsspin); | |
ff7f4cb4 | 1644 | n = rgd->rd_rstree.rb_node; |
5b924ae2 SW |
1645 | while (n) { |
1646 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
ff7f4cb4 | 1647 | rc = rs_cmp(block, length, rs); |
5b924ae2 SW |
1648 | if (rc < 0) |
1649 | n = n->rb_left; | |
1650 | else if (rc > 0) | |
1651 | n = n->rb_right; | |
1652 | else | |
1653 | break; | |
1654 | } | |
1655 | ||
1656 | if (n) { | |
3d39fcd1 | 1657 | while (rs_cmp(block, length, rs) == 0 && rs != ignore_rs) { |
07974d2a | 1658 | block = rs->rs_start + rs->rs_requested; |
ff7f4cb4 | 1659 | n = n->rb_right; |
5b924ae2 SW |
1660 | if (n == NULL) |
1661 | break; | |
1662 | rs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
1663 | } | |
1664 | } | |
1665 | ||
1666 | spin_unlock(&rgd->rd_rsspin); | |
1667 | return block; | |
1668 | } | |
1669 | ||
5b924ae2 SW |
1670 | /** |
1671 | * gfs2_reservation_check_and_update - Check for reservations during block alloc | |
1672 | * @rbm: The current position in the resource group | |
3d39fcd1 | 1673 | * @rs: Our own reservation |
ff7f4cb4 | 1674 | * @minext: The minimum extent length |
5ce13431 | 1675 | * @maxext: A pointer to the maximum extent structure |
5b924ae2 SW |
1676 | * |
1677 | * This checks the current position in the rgrp to see whether there is | |
1678 | * a reservation covering this block. If not then this function is a | |
1679 | * no-op. If there is, then the position is moved to the end of the | |
1680 | * contiguous reservation(s) so that we are pointing at the first | |
1681 | * non-reserved block. | |
1682 | * | |
1683 | * Returns: 0 if no reservation, 1 if @rbm has changed, otherwise an error | |
1684 | */ | |
1685 | ||
1686 | static int gfs2_reservation_check_and_update(struct gfs2_rbm *rbm, | |
3d39fcd1 | 1687 | struct gfs2_blkreserv *rs, |
5ce13431 BP |
1688 | u32 minext, |
1689 | struct gfs2_extent *maxext) | |
5b924ae2 SW |
1690 | { |
1691 | u64 block = gfs2_rbm_to_block(rbm); | |
ff7f4cb4 | 1692 | u32 extlen = 1; |
5b924ae2 | 1693 | u64 nblock; |
5b924ae2 | 1694 | |
ff7f4cb4 SW |
1695 | /* |
1696 | * If we have a minimum extent length, then skip over any extent | |
1697 | * which is less than the min extent length in size. | |
1698 | */ | |
f38e998f | 1699 | if (minext > 1) { |
ff7f4cb4 | 1700 | extlen = gfs2_free_extlen(rbm, minext); |
5ce13431 | 1701 | if (extlen <= maxext->len) |
ff7f4cb4 SW |
1702 | goto fail; |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * Check the extent which has been found against the reservations | |
1707 | * and skip if parts of it are already reserved | |
1708 | */ | |
3d39fcd1 | 1709 | nblock = gfs2_next_unreserved_block(rbm->rgd, block, extlen, rs); |
5ce13431 BP |
1710 | if (nblock == block) { |
1711 | if (!minext || extlen >= minext) | |
1712 | return 0; | |
1713 | ||
1714 | if (extlen > maxext->len) { | |
1715 | maxext->len = extlen; | |
1716 | maxext->rbm = *rbm; | |
1717 | } | |
0eacdd16 AG |
1718 | } else { |
1719 | u64 len = nblock - block; | |
1720 | if (len >= (u64)1 << 32) | |
1721 | return -E2BIG; | |
1722 | extlen = len; | |
5ce13431 | 1723 | } |
0eacdd16 AG |
1724 | fail: |
1725 | if (gfs2_rbm_add(rbm, extlen)) | |
1726 | return -E2BIG; | |
5b924ae2 SW |
1727 | return 1; |
1728 | } | |
1729 | ||
1730 | /** | |
1731 | * gfs2_rbm_find - Look for blocks of a particular state | |
1732 | * @rbm: Value/result starting position and final position | |
1733 | * @state: The state which we want to find | |
f38e998f | 1734 | * @minext: Pointer to the requested extent length |
5ce13431 | 1735 | * This is updated to be the actual reservation size. |
3d39fcd1 | 1736 | * @rs: Our own reservation (NULL to skip checking for reservations) |
5b924ae2 SW |
1737 | * @nowrap: Stop looking at the end of the rgrp, rather than wrapping |
1738 | * around until we've reached the starting point. | |
1739 | * | |
1740 | * Side effects: | |
1741 | * - If looking for free blocks, we set GBF_FULL on each bitmap which | |
1742 | * has no free blocks in it. | |
5ea5050c BP |
1743 | * - If looking for free blocks, we set rd_extfail_pt on each rgrp which |
1744 | * has come up short on a free block search. | |
5b924ae2 SW |
1745 | * |
1746 | * Returns: 0 on success, -ENOSPC if there is no block of the requested state | |
1747 | */ | |
1748 | ||
5ce13431 | 1749 | static int gfs2_rbm_find(struct gfs2_rbm *rbm, u8 state, u32 *minext, |
3d39fcd1 | 1750 | struct gfs2_blkreserv *rs, bool nowrap) |
5b924ae2 | 1751 | { |
71921ef8 | 1752 | bool scan_from_start = rbm->bii == 0 && rbm->offset == 0; |
5b924ae2 | 1753 | struct buffer_head *bh; |
71921ef8 | 1754 | int last_bii; |
5b924ae2 SW |
1755 | u32 offset; |
1756 | u8 *buffer; | |
71921ef8 | 1757 | bool wrapped = false; |
5b924ae2 | 1758 | int ret; |
e579ed4f | 1759 | struct gfs2_bitmap *bi; |
5ce13431 | 1760 | struct gfs2_extent maxext = { .rbm.rgd = rbm->rgd, }; |
5b924ae2 | 1761 | |
71921ef8 AG |
1762 | /* |
1763 | * Determine the last bitmap to search. If we're not starting at the | |
1764 | * beginning of a bitmap, we need to search that bitmap twice to scan | |
1765 | * the entire resource group. | |
5b924ae2 | 1766 | */ |
71921ef8 | 1767 | last_bii = rbm->bii - (rbm->offset == 0); |
5b924ae2 SW |
1768 | |
1769 | while(1) { | |
e579ed4f | 1770 | bi = rbm_bi(rbm); |
2fdc2fa2 | 1771 | if (test_bit(GBF_FULL, &bi->bi_flags) && |
5b924ae2 SW |
1772 | (state == GFS2_BLKST_FREE)) |
1773 | goto next_bitmap; | |
1774 | ||
e579ed4f BP |
1775 | bh = bi->bi_bh; |
1776 | buffer = bh->b_data + bi->bi_offset; | |
5b924ae2 | 1777 | WARN_ON(!buffer_uptodate(bh)); |
e579ed4f BP |
1778 | if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) |
1779 | buffer = bi->bi_clone + bi->bi_offset; | |
281b4952 | 1780 | offset = gfs2_bitfit(buffer, bi->bi_bytes, rbm->offset, state); |
71921ef8 AG |
1781 | if (offset == BFITNOENT) { |
1782 | if (state == GFS2_BLKST_FREE && rbm->offset == 0) | |
1783 | set_bit(GBF_FULL, &bi->bi_flags); | |
1784 | goto next_bitmap; | |
1785 | } | |
5b924ae2 | 1786 | rbm->offset = offset; |
bea906ee | 1787 | if (!rs || !minext) |
5b924ae2 SW |
1788 | return 0; |
1789 | ||
3d39fcd1 | 1790 | ret = gfs2_reservation_check_and_update(rbm, rs, *minext, |
5ce13431 | 1791 | &maxext); |
5b924ae2 SW |
1792 | if (ret == 0) |
1793 | return 0; | |
71921ef8 | 1794 | if (ret > 0) |
8d8b752a | 1795 | goto next_iter; |
5d50d532 | 1796 | if (ret == -E2BIG) { |
e579ed4f | 1797 | rbm->bii = 0; |
5d50d532 | 1798 | rbm->offset = 0; |
5d50d532 SW |
1799 | goto res_covered_end_of_rgrp; |
1800 | } | |
5b924ae2 SW |
1801 | return ret; |
1802 | ||
5b924ae2 SW |
1803 | next_bitmap: /* Find next bitmap in the rgrp */ |
1804 | rbm->offset = 0; | |
e579ed4f BP |
1805 | rbm->bii++; |
1806 | if (rbm->bii == rbm->rgd->rd_length) | |
1807 | rbm->bii = 0; | |
5d50d532 | 1808 | res_covered_end_of_rgrp: |
71921ef8 AG |
1809 | if (rbm->bii == 0) { |
1810 | if (wrapped) | |
1811 | break; | |
1812 | wrapped = true; | |
1813 | if (nowrap) | |
1814 | break; | |
1815 | } | |
8d8b752a | 1816 | next_iter: |
71921ef8 AG |
1817 | /* Have we scanned the entire resource group? */ |
1818 | if (wrapped && rbm->bii > last_bii) | |
5b924ae2 SW |
1819 | break; |
1820 | } | |
1821 | ||
f38e998f | 1822 | if (state != GFS2_BLKST_FREE) |
5ce13431 BP |
1823 | return -ENOSPC; |
1824 | ||
5ea5050c BP |
1825 | /* If the extent was too small, and it's smaller than the smallest |
1826 | to have failed before, remember for future reference that it's | |
1827 | useless to search this rgrp again for this amount or more. */ | |
71921ef8 AG |
1828 | if (wrapped && (scan_from_start || rbm->bii > last_bii) && |
1829 | *minext < rbm->rgd->rd_extfail_pt) | |
f38e998f | 1830 | rbm->rgd->rd_extfail_pt = *minext - 1; |
5ea5050c | 1831 | |
5ce13431 BP |
1832 | /* If the maximum extent we found is big enough to fulfill the |
1833 | minimum requirements, use it anyway. */ | |
1834 | if (maxext.len) { | |
1835 | *rbm = maxext.rbm; | |
1836 | *minext = maxext.len; | |
1837 | return 0; | |
1838 | } | |
1839 | ||
5b924ae2 SW |
1840 | return -ENOSPC; |
1841 | } | |
1842 | ||
c8cdf479 SW |
1843 | /** |
1844 | * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes | |
1845 | * @rgd: The rgrp | |
886b1416 BP |
1846 | * @last_unlinked: block address of the last dinode we unlinked |
1847 | * @skip: block address we should explicitly not unlink | |
c8cdf479 | 1848 | * |
1a0eae88 BP |
1849 | * Returns: 0 if no error |
1850 | * The inode, if one has been found, in inode. | |
c8cdf479 SW |
1851 | */ |
1852 | ||
044b9414 | 1853 | static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) |
c8cdf479 | 1854 | { |
5b924ae2 | 1855 | u64 block; |
5f3eae75 | 1856 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
044b9414 SW |
1857 | struct gfs2_glock *gl; |
1858 | struct gfs2_inode *ip; | |
1859 | int error; | |
1860 | int found = 0; | |
e579ed4f | 1861 | struct gfs2_rbm rbm = { .rgd = rgd, .bii = 0, .offset = 0 }; |
c8cdf479 | 1862 | |
5b924ae2 | 1863 | while (1) { |
5ce13431 | 1864 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_UNLINKED, NULL, NULL, |
8381e602 | 1865 | true); |
5b924ae2 SW |
1866 | if (error == -ENOSPC) |
1867 | break; | |
1868 | if (WARN_ON_ONCE(error)) | |
24c73873 | 1869 | break; |
b3e47ca0 | 1870 | |
5b924ae2 SW |
1871 | block = gfs2_rbm_to_block(&rbm); |
1872 | if (gfs2_rbm_from_block(&rbm, block + 1)) | |
1873 | break; | |
1874 | if (*last_unlinked != NO_BLOCK && block <= *last_unlinked) | |
c8cdf479 | 1875 | continue; |
5b924ae2 | 1876 | if (block == skip) |
1e19a195 | 1877 | continue; |
5b924ae2 | 1878 | *last_unlinked = block; |
044b9414 | 1879 | |
5ea31bc0 | 1880 | error = gfs2_glock_get(sdp, block, &gfs2_iopen_glops, CREATE, &gl); |
044b9414 SW |
1881 | if (error) |
1882 | continue; | |
1883 | ||
1884 | /* If the inode is already in cache, we can ignore it here | |
1885 | * because the existing inode disposal code will deal with | |
1886 | * it when all refs have gone away. Accessing gl_object like | |
1887 | * this is not safe in general. Here it is ok because we do | |
1888 | * not dereference the pointer, and we only need an approx | |
1889 | * answer to whether it is NULL or not. | |
1890 | */ | |
1891 | ip = gl->gl_object; | |
1892 | ||
a0e3cc65 | 1893 | if (ip || !gfs2_queue_delete_work(gl, 0)) |
044b9414 SW |
1894 | gfs2_glock_put(gl); |
1895 | else | |
1896 | found++; | |
1897 | ||
1898 | /* Limit reclaim to sensible number of tasks */ | |
44ad37d6 | 1899 | if (found > NR_CPUS) |
044b9414 | 1900 | return; |
c8cdf479 SW |
1901 | } |
1902 | ||
1903 | rgd->rd_flags &= ~GFS2_RDF_CHECK; | |
044b9414 | 1904 | return; |
c8cdf479 SW |
1905 | } |
1906 | ||
bcd97c06 SW |
1907 | /** |
1908 | * gfs2_rgrp_congested - Use stats to figure out whether an rgrp is congested | |
1909 | * @rgd: The rgrp in question | |
1910 | * @loops: An indication of how picky we can be (0=very, 1=less so) | |
1911 | * | |
1912 | * This function uses the recently added glock statistics in order to | |
1913 | * figure out whether a parciular resource group is suffering from | |
1914 | * contention from multiple nodes. This is done purely on the basis | |
1915 | * of timings, since this is the only data we have to work with and | |
1916 | * our aim here is to reject a resource group which is highly contended | |
1917 | * but (very important) not to do this too often in order to ensure that | |
1918 | * we do not land up introducing fragmentation by changing resource | |
1919 | * groups when not actually required. | |
1920 | * | |
1921 | * The calculation is fairly simple, we want to know whether the SRTTB | |
1922 | * (i.e. smoothed round trip time for blocking operations) to acquire | |
1923 | * the lock for this rgrp's glock is significantly greater than the | |
1924 | * time taken for resource groups on average. We introduce a margin in | |
1925 | * the form of the variable @var which is computed as the sum of the two | |
1926 | * respective variences, and multiplied by a factor depending on @loops | |
1927 | * and whether we have a lot of data to base the decision on. This is | |
1928 | * then tested against the square difference of the means in order to | |
1929 | * decide whether the result is statistically significant or not. | |
1930 | * | |
1931 | * Returns: A boolean verdict on the congestion status | |
1932 | */ | |
1933 | ||
1934 | static bool gfs2_rgrp_congested(const struct gfs2_rgrpd *rgd, int loops) | |
1935 | { | |
1936 | const struct gfs2_glock *gl = rgd->rd_gl; | |
15562c43 | 1937 | const struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
bcd97c06 | 1938 | struct gfs2_lkstats *st; |
4d207133 BH |
1939 | u64 r_dcount, l_dcount; |
1940 | u64 l_srttb, a_srttb = 0; | |
bcd97c06 | 1941 | s64 srttb_diff; |
4d207133 BH |
1942 | u64 sqr_diff; |
1943 | u64 var; | |
0166b197 | 1944 | int cpu, nonzero = 0; |
bcd97c06 SW |
1945 | |
1946 | preempt_disable(); | |
f4a3ae93 BP |
1947 | for_each_present_cpu(cpu) { |
1948 | st = &per_cpu_ptr(sdp->sd_lkstats, cpu)->lkstats[LM_TYPE_RGRP]; | |
0166b197 BP |
1949 | if (st->stats[GFS2_LKS_SRTTB]) { |
1950 | a_srttb += st->stats[GFS2_LKS_SRTTB]; | |
1951 | nonzero++; | |
1952 | } | |
f4a3ae93 | 1953 | } |
bcd97c06 | 1954 | st = &this_cpu_ptr(sdp->sd_lkstats)->lkstats[LM_TYPE_RGRP]; |
0166b197 BP |
1955 | if (nonzero) |
1956 | do_div(a_srttb, nonzero); | |
bcd97c06 SW |
1957 | r_dcount = st->stats[GFS2_LKS_DCOUNT]; |
1958 | var = st->stats[GFS2_LKS_SRTTVARB] + | |
1959 | gl->gl_stats.stats[GFS2_LKS_SRTTVARB]; | |
1960 | preempt_enable(); | |
1961 | ||
1962 | l_srttb = gl->gl_stats.stats[GFS2_LKS_SRTTB]; | |
1963 | l_dcount = gl->gl_stats.stats[GFS2_LKS_DCOUNT]; | |
1964 | ||
f4a3ae93 | 1965 | if ((l_dcount < 1) || (r_dcount < 1) || (a_srttb == 0)) |
bcd97c06 SW |
1966 | return false; |
1967 | ||
f4a3ae93 | 1968 | srttb_diff = a_srttb - l_srttb; |
bcd97c06 SW |
1969 | sqr_diff = srttb_diff * srttb_diff; |
1970 | ||
1971 | var *= 2; | |
1972 | if (l_dcount < 8 || r_dcount < 8) | |
1973 | var *= 2; | |
1974 | if (loops == 1) | |
1975 | var *= 2; | |
1976 | ||
1977 | return ((srttb_diff < 0) && (sqr_diff > var)); | |
1978 | } | |
1979 | ||
1980 | /** | |
1981 | * gfs2_rgrp_used_recently | |
1982 | * @rs: The block reservation with the rgrp to test | |
1983 | * @msecs: The time limit in milliseconds | |
1984 | * | |
1985 | * Returns: True if the rgrp glock has been used within the time limit | |
1986 | */ | |
1987 | static bool gfs2_rgrp_used_recently(const struct gfs2_blkreserv *rs, | |
1988 | u64 msecs) | |
1989 | { | |
1990 | u64 tdiff; | |
1991 | ||
1992 | tdiff = ktime_to_ns(ktime_sub(ktime_get_real(), | |
c65b76b8 | 1993 | rs->rs_rgd->rd_gl->gl_dstamp)); |
bcd97c06 SW |
1994 | |
1995 | return tdiff > (msecs * 1000 * 1000); | |
1996 | } | |
1997 | ||
9dbe9610 SW |
1998 | static u32 gfs2_orlov_skip(const struct gfs2_inode *ip) |
1999 | { | |
2000 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2001 | u32 skip; | |
2002 | ||
2003 | get_random_bytes(&skip, sizeof(skip)); | |
2004 | return skip % sdp->sd_rgrps; | |
2005 | } | |
2006 | ||
c743ffd0 SW |
2007 | static bool gfs2_select_rgrp(struct gfs2_rgrpd **pos, const struct gfs2_rgrpd *begin) |
2008 | { | |
2009 | struct gfs2_rgrpd *rgd = *pos; | |
aa8920c9 | 2010 | struct gfs2_sbd *sdp = rgd->rd_sbd; |
c743ffd0 SW |
2011 | |
2012 | rgd = gfs2_rgrpd_get_next(rgd); | |
2013 | if (rgd == NULL) | |
aa8920c9 | 2014 | rgd = gfs2_rgrpd_get_first(sdp); |
c743ffd0 SW |
2015 | *pos = rgd; |
2016 | if (rgd != begin) /* If we didn't wrap */ | |
2017 | return true; | |
2018 | return false; | |
2019 | } | |
2020 | ||
0e27c18c BP |
2021 | /** |
2022 | * fast_to_acquire - determine if a resource group will be fast to acquire | |
c551f66c | 2023 | * @rgd: The rgrp |
0e27c18c BP |
2024 | * |
2025 | * If this is one of our preferred rgrps, it should be quicker to acquire, | |
2026 | * because we tried to set ourselves up as dlm lock master. | |
2027 | */ | |
2028 | static inline int fast_to_acquire(struct gfs2_rgrpd *rgd) | |
2029 | { | |
2030 | struct gfs2_glock *gl = rgd->rd_gl; | |
2031 | ||
2032 | if (gl->gl_state != LM_ST_UNLOCKED && list_empty(&gl->gl_holders) && | |
2033 | !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) && | |
2034 | !test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
2035 | return 1; | |
2036 | if (rgd->rd_flags & GFS2_RDF_PREFERRED) | |
2037 | return 1; | |
2038 | return 0; | |
2039 | } | |
2040 | ||
b3b94faa | 2041 | /** |
666d1d8a | 2042 | * gfs2_inplace_reserve - Reserve space in the filesystem |
b3b94faa | 2043 | * @ip: the inode to reserve space for |
7b9cff46 | 2044 | * @ap: the allocation parameters |
b3b94faa | 2045 | * |
25435e5e AD |
2046 | * We try our best to find an rgrp that has at least ap->target blocks |
2047 | * available. After a couple of passes (loops == 2), the prospects of finding | |
2048 | * such an rgrp diminish. At this stage, we return the first rgrp that has | |
725d0e9d | 2049 | * at least ap->min_target blocks available. |
25435e5e AD |
2050 | * |
2051 | * Returns: 0 on success, | |
2052 | * -ENOMEM if a suitable rgrp can't be found | |
2053 | * errno otherwise | |
b3b94faa DT |
2054 | */ |
2055 | ||
25435e5e | 2056 | int gfs2_inplace_reserve(struct gfs2_inode *ip, struct gfs2_alloc_parms *ap) |
b3b94faa | 2057 | { |
feaa7bba | 2058 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
8e2e0047 | 2059 | struct gfs2_rgrpd *begin = NULL; |
a097dc7e | 2060 | struct gfs2_blkreserv *rs = &ip->i_res; |
4fc7ec31 | 2061 | int error = 0, flags = LM_FLAG_NODE_SCOPE; |
9e514605 | 2062 | bool rg_locked; |
666d1d8a | 2063 | u64 last_unlinked = NO_BLOCK; |
4272006d | 2064 | u32 target = ap->target; |
7c9ca621 | 2065 | int loops = 0; |
725d0e9d AG |
2066 | u32 free_blocks, blocks_available, skip = 0; |
2067 | ||
2068 | BUG_ON(rs->rs_reserved); | |
b3b94faa | 2069 | |
90306c41 BM |
2070 | if (sdp->sd_args.ar_rgrplvb) |
2071 | flags |= GL_SKIP; | |
4272006d | 2072 | if (gfs2_assert_warn(sdp, target)) |
c743ffd0 | 2073 | return -EINVAL; |
8e2e0047 | 2074 | if (gfs2_rs_active(rs)) { |
c65b76b8 AG |
2075 | begin = rs->rs_rgd; |
2076 | } else if (rs->rs_rgd && | |
2077 | rgrp_contains_block(rs->rs_rgd, ip->i_goal)) { | |
2078 | begin = rs->rs_rgd; | |
8e2e0047 | 2079 | } else { |
00a158be | 2080 | check_and_update_goal(ip); |
c65b76b8 | 2081 | rs->rs_rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); |
8e2e0047 | 2082 | } |
7b9cff46 | 2083 | if (S_ISDIR(ip->i_inode.i_mode) && (ap->aflags & GFS2_AF_ORLOV)) |
9dbe9610 | 2084 | skip = gfs2_orlov_skip(ip); |
c65b76b8 | 2085 | if (rs->rs_rgd == NULL) |
7c9ca621 BP |
2086 | return -EBADSLT; |
2087 | ||
2088 | while (loops < 3) { | |
725d0e9d AG |
2089 | struct gfs2_rgrpd *rgd; |
2090 | ||
9e514605 AG |
2091 | rg_locked = gfs2_glock_is_locked_by_me(rs->rs_rgd->rd_gl); |
2092 | if (rg_locked) { | |
2093 | rgrp_lock_local(rs->rs_rgd); | |
2094 | } else { | |
9dbe9610 SW |
2095 | if (skip && skip--) |
2096 | goto next_rgrp; | |
0e27c18c BP |
2097 | if (!gfs2_rs_active(rs)) { |
2098 | if (loops == 0 && | |
c65b76b8 | 2099 | !fast_to_acquire(rs->rs_rgd)) |
0e27c18c BP |
2100 | goto next_rgrp; |
2101 | if ((loops < 2) && | |
2102 | gfs2_rgrp_used_recently(rs, 1000) && | |
c65b76b8 | 2103 | gfs2_rgrp_congested(rs->rs_rgd, loops)) |
0e27c18c BP |
2104 | goto next_rgrp; |
2105 | } | |
c65b76b8 | 2106 | error = gfs2_glock_nq_init(rs->rs_rgd->rd_gl, |
8e2e0047 | 2107 | LM_ST_EXCLUSIVE, flags, |
21f09c43 | 2108 | &ip->i_rgd_gh); |
c743ffd0 SW |
2109 | if (unlikely(error)) |
2110 | return error; | |
9e514605 | 2111 | rgrp_lock_local(rs->rs_rgd); |
bcd97c06 | 2112 | if (!gfs2_rs_active(rs) && (loops < 2) && |
c65b76b8 | 2113 | gfs2_rgrp_congested(rs->rs_rgd, loops)) |
bcd97c06 | 2114 | goto skip_rgrp; |
c743ffd0 | 2115 | if (sdp->sd_args.ar_rgrplvb) { |
c65b76b8 | 2116 | error = update_rgrp_lvb(rs->rs_rgd); |
c743ffd0 | 2117 | if (unlikely(error)) { |
9e514605 | 2118 | rgrp_unlock_local(rs->rs_rgd); |
21f09c43 | 2119 | gfs2_glock_dq_uninit(&ip->i_rgd_gh); |
90306c41 BM |
2120 | return error; |
2121 | } | |
2122 | } | |
292c8c14 | 2123 | } |
666d1d8a | 2124 | |
243fea4d | 2125 | /* Skip unusable resource groups */ |
c65b76b8 | 2126 | if ((rs->rs_rgd->rd_flags & (GFS2_RGF_NOALLOC | |
5ea5050c | 2127 | GFS2_RDF_ERROR)) || |
4272006d | 2128 | (loops == 0 && target > rs->rs_rgd->rd_extfail_pt)) |
c743ffd0 SW |
2129 | goto skip_rgrp; |
2130 | ||
2131 | if (sdp->sd_args.ar_rgrplvb) | |
c65b76b8 | 2132 | gfs2_rgrp_bh_get(rs->rs_rgd); |
c743ffd0 SW |
2133 | |
2134 | /* Get a reservation if we don't already have one */ | |
2135 | if (!gfs2_rs_active(rs)) | |
c65b76b8 | 2136 | rg_mblk_search(rs->rs_rgd, ip, ap); |
c743ffd0 SW |
2137 | |
2138 | /* Skip rgrps when we can't get a reservation on first pass */ | |
2139 | if (!gfs2_rs_active(rs) && (loops < 1)) | |
2140 | goto check_rgrp; | |
2141 | ||
2142 | /* If rgrp has enough free space, use it */ | |
725d0e9d AG |
2143 | rgd = rs->rs_rgd; |
2144 | spin_lock(&rgd->rd_rsspin); | |
2145 | free_blocks = rgd_free(rgd, rs); | |
2146 | blocks_available = rgd->rd_free_clone - rgd->rd_reserved; | |
2147 | if (free_blocks < target || blocks_available < target) { | |
2148 | spin_unlock(&rgd->rd_rsspin); | |
2149 | goto check_rgrp; | |
b3b94faa | 2150 | } |
725d0e9d AG |
2151 | rs->rs_reserved = ap->target; |
2152 | if (rs->rs_reserved > blocks_available) | |
2153 | rs->rs_reserved = blocks_available; | |
2154 | rgd->rd_reserved += rs->rs_reserved; | |
2155 | spin_unlock(&rgd->rd_rsspin); | |
9e514605 | 2156 | rgrp_unlock_local(rs->rs_rgd); |
725d0e9d | 2157 | return 0; |
c743ffd0 SW |
2158 | check_rgrp: |
2159 | /* Check for unlinked inodes which can be reclaimed */ | |
c65b76b8 AG |
2160 | if (rs->rs_rgd->rd_flags & GFS2_RDF_CHECK) |
2161 | try_rgrp_unlink(rs->rs_rgd, &last_unlinked, | |
c743ffd0 SW |
2162 | ip->i_no_addr); |
2163 | skip_rgrp: | |
9e514605 AG |
2164 | rgrp_unlock_local(rs->rs_rgd); |
2165 | ||
1330edbe BP |
2166 | /* Drop reservation, if we couldn't use reserved rgrp */ |
2167 | if (gfs2_rs_active(rs)) | |
2168 | gfs2_rs_deltree(rs); | |
2169 | ||
c743ffd0 SW |
2170 | /* Unlock rgrp if required */ |
2171 | if (!rg_locked) | |
21f09c43 | 2172 | gfs2_glock_dq_uninit(&ip->i_rgd_gh); |
c743ffd0 SW |
2173 | next_rgrp: |
2174 | /* Find the next rgrp, and continue looking */ | |
c65b76b8 | 2175 | if (gfs2_select_rgrp(&rs->rs_rgd, begin)) |
c743ffd0 | 2176 | continue; |
9dbe9610 SW |
2177 | if (skip) |
2178 | continue; | |
c743ffd0 SW |
2179 | |
2180 | /* If we've scanned all the rgrps, but found no free blocks | |
2181 | * then this checks for some less likely conditions before | |
2182 | * trying again. | |
2183 | */ | |
c743ffd0 SW |
2184 | loops++; |
2185 | /* Check that fs hasn't grown if writing to rindex */ | |
2186 | if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { | |
2187 | error = gfs2_ri_update(ip); | |
2188 | if (error) | |
2189 | return error; | |
2190 | } | |
2191 | /* Flushing the log may release space */ | |
4272006d AG |
2192 | if (loops == 2) { |
2193 | if (ap->min_target) | |
2194 | target = ap->min_target; | |
805c0907 BP |
2195 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
2196 | GFS2_LFC_INPLACE_RESERVE); | |
4272006d | 2197 | } |
b3b94faa | 2198 | } |
b3b94faa | 2199 | |
c743ffd0 | 2200 | return -ENOSPC; |
b3b94faa DT |
2201 | } |
2202 | ||
2203 | /** | |
2204 | * gfs2_inplace_release - release an inplace reservation | |
2205 | * @ip: the inode the reservation was taken out on | |
2206 | * | |
2207 | * Release a reservation made by gfs2_inplace_reserve(). | |
2208 | */ | |
2209 | ||
2210 | void gfs2_inplace_release(struct gfs2_inode *ip) | |
2211 | { | |
725d0e9d AG |
2212 | struct gfs2_blkreserv *rs = &ip->i_res; |
2213 | ||
2214 | if (rs->rs_reserved) { | |
2215 | struct gfs2_rgrpd *rgd = rs->rs_rgd; | |
2216 | ||
2217 | spin_lock(&rgd->rd_rsspin); | |
2218 | BUG_ON(rgd->rd_reserved < rs->rs_reserved); | |
2219 | rgd->rd_reserved -= rs->rs_reserved; | |
2220 | spin_unlock(&rgd->rd_rsspin); | |
2221 | rs->rs_reserved = 0; | |
2222 | } | |
21f09c43 AG |
2223 | if (gfs2_holder_initialized(&ip->i_rgd_gh)) |
2224 | gfs2_glock_dq_uninit(&ip->i_rgd_gh); | |
b3b94faa DT |
2225 | } |
2226 | ||
b3e47ca0 BP |
2227 | /** |
2228 | * gfs2_alloc_extent - allocate an extent from a given bitmap | |
4a993fb1 | 2229 | * @rbm: the resource group information |
b3e47ca0 | 2230 | * @dinode: TRUE if the first block we allocate is for a dinode |
c04a2ef3 | 2231 | * @n: The extent length (value/result) |
b3e47ca0 | 2232 | * |
c04a2ef3 | 2233 | * Add the bitmap buffer to the transaction. |
b3e47ca0 | 2234 | * Set the found bits to @new_state to change block's allocation state. |
b3e47ca0 | 2235 | */ |
c04a2ef3 | 2236 | static void gfs2_alloc_extent(const struct gfs2_rbm *rbm, bool dinode, |
4a993fb1 | 2237 | unsigned int *n) |
b3e47ca0 | 2238 | { |
c04a2ef3 | 2239 | struct gfs2_rbm pos = { .rgd = rbm->rgd, }; |
b3e47ca0 | 2240 | const unsigned int elen = *n; |
c04a2ef3 SW |
2241 | u64 block; |
2242 | int ret; | |
b3e47ca0 | 2243 | |
c04a2ef3 SW |
2244 | *n = 1; |
2245 | block = gfs2_rbm_to_block(rbm); | |
e579ed4f | 2246 | gfs2_trans_add_meta(rbm->rgd->rd_gl, rbm_bi(rbm)->bi_bh); |
3e6339dd | 2247 | gfs2_setbit(rbm, true, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
c04a2ef3 | 2248 | block++; |
60a0b8f9 | 2249 | while (*n < elen) { |
c04a2ef3 | 2250 | ret = gfs2_rbm_from_block(&pos, block); |
dffe12a8 | 2251 | if (ret || gfs2_testbit(&pos, true) != GFS2_BLKST_FREE) |
60a0b8f9 | 2252 | break; |
e579ed4f | 2253 | gfs2_trans_add_meta(pos.rgd->rd_gl, rbm_bi(&pos)->bi_bh); |
3e6339dd | 2254 | gfs2_setbit(&pos, true, GFS2_BLKST_USED); |
60a0b8f9 | 2255 | (*n)++; |
c04a2ef3 | 2256 | block++; |
c8cdf479 | 2257 | } |
b3b94faa DT |
2258 | } |
2259 | ||
2260 | /** | |
2261 | * rgblk_free - Change alloc state of given block(s) | |
2262 | * @sdp: the filesystem | |
0ddeded4 | 2263 | * @rgd: the resource group the blocks are in |
b3b94faa DT |
2264 | * @bstart: the start of a run of blocks to free |
2265 | * @blen: the length of the block run (all must lie within ONE RG!) | |
2266 | * @new_state: GFS2_BLKST_XXX the after-allocation block state | |
b3b94faa DT |
2267 | */ |
2268 | ||
0ddeded4 AG |
2269 | static void rgblk_free(struct gfs2_sbd *sdp, struct gfs2_rgrpd *rgd, |
2270 | u64 bstart, u32 blen, unsigned char new_state) | |
b3b94faa | 2271 | { |
3b1d0b9d | 2272 | struct gfs2_rbm rbm; |
d24e0569 | 2273 | struct gfs2_bitmap *bi, *bi_prev = NULL; |
b3b94faa | 2274 | |
0ddeded4 | 2275 | rbm.rgd = rgd; |
f654683d | 2276 | if (WARN_ON_ONCE(gfs2_rbm_from_block(&rbm, bstart))) |
0ddeded4 | 2277 | return; |
b3b94faa | 2278 | while (blen--) { |
e579ed4f | 2279 | bi = rbm_bi(&rbm); |
d24e0569 BP |
2280 | if (bi != bi_prev) { |
2281 | if (!bi->bi_clone) { | |
2282 | bi->bi_clone = kmalloc(bi->bi_bh->b_size, | |
2283 | GFP_NOFS | __GFP_NOFAIL); | |
2284 | memcpy(bi->bi_clone + bi->bi_offset, | |
2285 | bi->bi_bh->b_data + bi->bi_offset, | |
281b4952 | 2286 | bi->bi_bytes); |
d24e0569 BP |
2287 | } |
2288 | gfs2_trans_add_meta(rbm.rgd->rd_gl, bi->bi_bh); | |
2289 | bi_prev = bi; | |
b3b94faa | 2290 | } |
3e6339dd | 2291 | gfs2_setbit(&rbm, false, new_state); |
0eacdd16 | 2292 | gfs2_rbm_add(&rbm, 1); |
b3b94faa | 2293 | } |
b3b94faa DT |
2294 | } |
2295 | ||
2296 | /** | |
09010978 SW |
2297 | * gfs2_rgrp_dump - print out an rgrp |
2298 | * @seq: The iterator | |
0e539ca1 | 2299 | * @rgd: The rgrp in question |
3792ce97 | 2300 | * @fs_id_buf: pointer to file system id (if requested) |
09010978 SW |
2301 | * |
2302 | */ | |
2303 | ||
0e539ca1 | 2304 | void gfs2_rgrp_dump(struct seq_file *seq, struct gfs2_rgrpd *rgd, |
3792ce97 | 2305 | const char *fs_id_buf) |
09010978 | 2306 | { |
8e2e0047 BP |
2307 | struct gfs2_blkreserv *trs; |
2308 | const struct rb_node *n; | |
2309 | ||
9e514605 | 2310 | spin_lock(&rgd->rd_rsspin); |
725d0e9d | 2311 | gfs2_print_dbg(seq, "%s R: n:%llu f:%02x b:%u/%u i:%u q:%u r:%u e:%u\n", |
3792ce97 | 2312 | fs_id_buf, |
09010978 | 2313 | (unsigned long long)rgd->rd_addr, rgd->rd_flags, |
8e2e0047 | 2314 | rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes, |
725d0e9d | 2315 | rgd->rd_requested, rgd->rd_reserved, rgd->rd_extfail_pt); |
72244b6b BP |
2316 | if (rgd->rd_sbd->sd_args.ar_rgrplvb) { |
2317 | struct gfs2_rgrp_lvb *rgl = rgd->rd_rgl; | |
2318 | ||
3792ce97 | 2319 | gfs2_print_dbg(seq, "%s L: f:%02x b:%u i:%u\n", fs_id_buf, |
72244b6b BP |
2320 | be32_to_cpu(rgl->rl_flags), |
2321 | be32_to_cpu(rgl->rl_free), | |
2322 | be32_to_cpu(rgl->rl_dinodes)); | |
2323 | } | |
8e2e0047 BP |
2324 | for (n = rb_first(&rgd->rd_rstree); n; n = rb_next(&trs->rs_node)) { |
2325 | trs = rb_entry(n, struct gfs2_blkreserv, rs_node); | |
3792ce97 | 2326 | dump_rs(seq, trs, fs_id_buf); |
8e2e0047 BP |
2327 | } |
2328 | spin_unlock(&rgd->rd_rsspin); | |
09010978 SW |
2329 | } |
2330 | ||
6050b9c7 SW |
2331 | static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) |
2332 | { | |
2333 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
98fb0574 | 2334 | char fs_id_buf[sizeof(sdp->sd_fsname) + 7]; |
3792ce97 | 2335 | |
6050b9c7 | 2336 | fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", |
86d00636 | 2337 | (unsigned long long)rgd->rd_addr); |
6050b9c7 | 2338 | fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); |
3792ce97 | 2339 | sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname); |
0e539ca1 | 2340 | gfs2_rgrp_dump(NULL, rgd, fs_id_buf); |
6050b9c7 SW |
2341 | rgd->rd_flags |= GFS2_RDF_ERROR; |
2342 | } | |
2343 | ||
8e2e0047 | 2344 | /** |
5b924ae2 SW |
2345 | * gfs2_adjust_reservation - Adjust (or remove) a reservation after allocation |
2346 | * @ip: The inode we have just allocated blocks for | |
2347 | * @rbm: The start of the allocated blocks | |
2348 | * @len: The extent length | |
8e2e0047 | 2349 | * |
5b924ae2 SW |
2350 | * Adjusts a reservation after an allocation has taken place. If the |
2351 | * reservation does not match the allocation, or if it is now empty | |
2352 | * then it is removed. | |
8e2e0047 | 2353 | */ |
5b924ae2 SW |
2354 | |
2355 | static void gfs2_adjust_reservation(struct gfs2_inode *ip, | |
2356 | const struct gfs2_rbm *rbm, unsigned len) | |
8e2e0047 | 2357 | { |
a097dc7e | 2358 | struct gfs2_blkreserv *rs = &ip->i_res; |
5b924ae2 | 2359 | struct gfs2_rgrpd *rgd = rbm->rgd; |
8e2e0047 | 2360 | |
725d0e9d AG |
2361 | BUG_ON(rs->rs_reserved < len); |
2362 | rs->rs_reserved -= len; | |
5b924ae2 | 2363 | if (gfs2_rs_active(rs)) { |
c65b76b8 AG |
2364 | u64 start = gfs2_rbm_to_block(rbm); |
2365 | ||
2366 | if (rs->rs_start == start) { | |
2367 | unsigned int rlen; | |
2368 | ||
2369 | rs->rs_start += len; | |
07974d2a AG |
2370 | rlen = min(rs->rs_requested, len); |
2371 | rs->rs_requested -= rlen; | |
2372 | rgd->rd_requested -= rlen; | |
9e733d39 | 2373 | trace_gfs2_rs(rs, TRACE_RS_CLAIM); |
c65b76b8 | 2374 | if (rs->rs_start < rgd->rd_data0 + rgd->rd_data && |
07974d2a | 2375 | rs->rs_requested) |
725d0e9d | 2376 | return; |
1a855033 BP |
2377 | /* We used up our block reservation, so we should |
2378 | reserve more blocks next time. */ | |
21f09c43 | 2379 | atomic_add(RGRP_RSRV_ADDBLKS, &ip->i_sizehint); |
5b924ae2 | 2380 | } |
20095218 | 2381 | __rs_deltree(rs); |
8e2e0047 | 2382 | } |
8e2e0047 BP |
2383 | } |
2384 | ||
9e07f2cb SW |
2385 | /** |
2386 | * gfs2_set_alloc_start - Set starting point for block allocation | |
2387 | * @rbm: The rbm which will be set to the required location | |
2388 | * @ip: The gfs2 inode | |
2389 | * @dinode: Flag to say if allocation includes a new inode | |
2390 | * | |
2391 | * This sets the starting point from the reservation if one is active | |
2392 | * otherwise it falls back to guessing a start point based on the | |
2393 | * inode's goal block or the last allocation point in the rgrp. | |
2394 | */ | |
2395 | ||
2396 | static void gfs2_set_alloc_start(struct gfs2_rbm *rbm, | |
2397 | const struct gfs2_inode *ip, bool dinode) | |
2398 | { | |
2399 | u64 goal; | |
2400 | ||
a097dc7e | 2401 | if (gfs2_rs_active(&ip->i_res)) { |
c65b76b8 AG |
2402 | goal = ip->i_res.rs_start; |
2403 | } else { | |
2404 | if (!dinode && rgrp_contains_block(rbm->rgd, ip->i_goal)) | |
2405 | goal = ip->i_goal; | |
2406 | else | |
2407 | goal = rbm->rgd->rd_last_alloc + rbm->rgd->rd_data0; | |
9e07f2cb | 2408 | } |
f654683d AG |
2409 | if (WARN_ON_ONCE(gfs2_rbm_from_block(rbm, goal))) { |
2410 | rbm->bii = 0; | |
2411 | rbm->offset = 0; | |
2412 | } | |
9e07f2cb SW |
2413 | } |
2414 | ||
09010978 | 2415 | /** |
6e87ed0f | 2416 | * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode |
1639431a | 2417 | * @ip: the inode to allocate the block for |
09010978 | 2418 | * @bn: Used to return the starting block number |
8e2e0047 | 2419 | * @nblocks: requested number of blocks/extent length (value/result) |
6e87ed0f | 2420 | * @dinode: 1 if we're allocating a dinode block, else 0 |
3c5d785a | 2421 | * @generation: the generation number of the inode |
b3b94faa | 2422 | * |
09010978 | 2423 | * Returns: 0 or error |
b3b94faa DT |
2424 | */ |
2425 | ||
6a8099ed | 2426 | int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, |
6e87ed0f | 2427 | bool dinode, u64 *generation) |
b3b94faa | 2428 | { |
feaa7bba | 2429 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
d9ba7615 | 2430 | struct buffer_head *dibh; |
c65b76b8 | 2431 | struct gfs2_rbm rbm = { .rgd = ip->i_res.rs_rgd, }; |
3c5d785a | 2432 | u64 block; /* block, within the file system scope */ |
f38e998f | 2433 | u32 minext = 1; |
b2598965 | 2434 | int error = -ENOSPC; |
62e252ee | 2435 | |
725d0e9d AG |
2436 | BUG_ON(ip->i_res.rs_reserved < *nblocks); |
2437 | ||
9e514605 | 2438 | rgrp_lock_local(rbm.rgd); |
b2598965 AG |
2439 | if (gfs2_rs_active(&ip->i_res)) { |
2440 | gfs2_set_alloc_start(&rbm, ip, dinode); | |
2441 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &minext, &ip->i_res, false); | |
2442 | } | |
137834a6 | 2443 | if (error == -ENOSPC) { |
9e07f2cb | 2444 | gfs2_set_alloc_start(&rbm, ip, dinode); |
f38e998f | 2445 | error = gfs2_rbm_find(&rbm, GFS2_BLKST_FREE, &minext, NULL, false); |
137834a6 SW |
2446 | } |
2447 | ||
62e252ee | 2448 | /* Since all blocks are reserved in advance, this shouldn't happen */ |
5b924ae2 | 2449 | if (error) { |
5ea5050c | 2450 | fs_warn(sdp, "inum=%llu error=%d, nblocks=%u, full=%d fail_pt=%d\n", |
9e733d39 | 2451 | (unsigned long long)ip->i_no_addr, error, *nblocks, |
5ea5050c BP |
2452 | test_bit(GBF_FULL, &rbm.rgd->rd_bits->bi_flags), |
2453 | rbm.rgd->rd_extfail_pt); | |
62e252ee | 2454 | goto rgrp_error; |
8e2e0047 | 2455 | } |
62e252ee | 2456 | |
c04a2ef3 SW |
2457 | gfs2_alloc_extent(&rbm, dinode, nblocks); |
2458 | block = gfs2_rbm_to_block(&rbm); | |
c743ffd0 | 2459 | rbm.rgd->rd_last_alloc = block - rbm.rgd->rd_data0; |
3c5d785a | 2460 | if (!dinode) { |
3ed08bef | 2461 | ip->i_goal = block + *nblocks - 1; |
3c5d785a BP |
2462 | error = gfs2_meta_inode_buffer(ip, &dibh); |
2463 | if (error == 0) { | |
2464 | struct gfs2_dinode *di = | |
2465 | (struct gfs2_dinode *)dibh->b_data; | |
350a9b0a | 2466 | gfs2_trans_add_meta(ip->i_gl, dibh); |
3c5d785a BP |
2467 | di->di_goal_meta = di->di_goal_data = |
2468 | cpu_to_be64(ip->i_goal); | |
2469 | brelse(dibh); | |
2470 | } | |
d9ba7615 | 2471 | } |
725d0e9d AG |
2472 | spin_lock(&rbm.rgd->rd_rsspin); |
2473 | gfs2_adjust_reservation(ip, &rbm, *nblocks); | |
2474 | if (rbm.rgd->rd_free < *nblocks || rbm.rgd->rd_reserved < *nblocks) { | |
e54c78a2 | 2475 | fs_warn(sdp, "nblocks=%u\n", *nblocks); |
725d0e9d | 2476 | spin_unlock(&rbm.rgd->rd_rsspin); |
09010978 | 2477 | goto rgrp_error; |
8e2e0047 | 2478 | } |
725d0e9d AG |
2479 | BUG_ON(rbm.rgd->rd_reserved < *nblocks); |
2480 | BUG_ON(rbm.rgd->rd_free_clone < *nblocks); | |
2481 | BUG_ON(rbm.rgd->rd_free < *nblocks); | |
2482 | rbm.rgd->rd_reserved -= *nblocks; | |
2483 | rbm.rgd->rd_free_clone -= *nblocks; | |
4a993fb1 | 2484 | rbm.rgd->rd_free -= *nblocks; |
725d0e9d | 2485 | spin_unlock(&rbm.rgd->rd_rsspin); |
3c5d785a | 2486 | if (dinode) { |
4a993fb1 SW |
2487 | rbm.rgd->rd_dinodes++; |
2488 | *generation = rbm.rgd->rd_igeneration++; | |
3c5d785a | 2489 | if (*generation == 0) |
4a993fb1 | 2490 | *generation = rbm.rgd->rd_igeneration++; |
3c5d785a | 2491 | } |
b3b94faa | 2492 | |
350a9b0a | 2493 | gfs2_trans_add_meta(rbm.rgd->rd_gl, rbm.rgd->rd_bits[0].bi_bh); |
4a993fb1 | 2494 | gfs2_rgrp_out(rbm.rgd, rbm.rgd->rd_bits[0].bi_bh->b_data); |
9e514605 | 2495 | rgrp_unlock_local(rbm.rgd); |
b3b94faa | 2496 | |
6a8099ed | 2497 | gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); |
3c5d785a | 2498 | if (dinode) |
fbb27873 | 2499 | gfs2_trans_remove_revoke(sdp, block, *nblocks); |
6a8099ed | 2500 | |
fd4b4e04 | 2501 | gfs2_quota_change(ip, *nblocks, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa | 2502 | |
4a993fb1 | 2503 | trace_gfs2_block_alloc(ip, rbm.rgd, block, *nblocks, |
6e87ed0f | 2504 | dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); |
6050b9c7 SW |
2505 | *bn = block; |
2506 | return 0; | |
2507 | ||
2508 | rgrp_error: | |
9e514605 | 2509 | rgrp_unlock_local(rbm.rgd); |
4a993fb1 | 2510 | gfs2_rgrp_error(rbm.rgd); |
6050b9c7 | 2511 | return -EIO; |
b3b94faa DT |
2512 | } |
2513 | ||
2514 | /** | |
46fcb2ed | 2515 | * __gfs2_free_blocks - free a contiguous run of block(s) |
b3b94faa | 2516 | * @ip: the inode these blocks are being freed from |
0ddeded4 | 2517 | * @rgd: the resource group the blocks are in |
b3b94faa DT |
2518 | * @bstart: first block of a run of contiguous blocks |
2519 | * @blen: the length of the block run | |
46fcb2ed | 2520 | * @meta: 1 if the blocks represent metadata |
b3b94faa DT |
2521 | * |
2522 | */ | |
2523 | ||
0ddeded4 AG |
2524 | void __gfs2_free_blocks(struct gfs2_inode *ip, struct gfs2_rgrpd *rgd, |
2525 | u64 bstart, u32 blen, int meta) | |
b3b94faa | 2526 | { |
feaa7bba | 2527 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa | 2528 | |
9e514605 | 2529 | rgrp_lock_local(rgd); |
0ddeded4 | 2530 | rgblk_free(sdp, rgd, bstart, blen, GFS2_BLKST_FREE); |
41db1ab9 | 2531 | trace_gfs2_block_alloc(ip, rgd, bstart, blen, GFS2_BLKST_FREE); |
cfc8b549 | 2532 | rgd->rd_free += blen; |
66fc061b | 2533 | rgd->rd_flags &= ~GFS2_RGF_TRIMMED; |
350a9b0a | 2534 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2535 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
9e514605 | 2536 | rgrp_unlock_local(rgd); |
b3b94faa | 2537 | |
6d3117b4 | 2538 | /* Directories keep their data in the metadata address space */ |
68942870 BP |
2539 | if (meta || ip->i_depth || gfs2_is_jdata(ip)) |
2540 | gfs2_journal_wipe(ip, bstart, blen); | |
4c16c36a | 2541 | } |
b3b94faa | 2542 | |
4c16c36a BP |
2543 | /** |
2544 | * gfs2_free_meta - free a contiguous run of data block(s) | |
2545 | * @ip: the inode these blocks are being freed from | |
0ddeded4 | 2546 | * @rgd: the resource group the blocks are in |
4c16c36a BP |
2547 | * @bstart: first block of a run of contiguous blocks |
2548 | * @blen: the length of the block run | |
2549 | * | |
2550 | */ | |
2551 | ||
0ddeded4 AG |
2552 | void gfs2_free_meta(struct gfs2_inode *ip, struct gfs2_rgrpd *rgd, |
2553 | u64 bstart, u32 blen) | |
4c16c36a BP |
2554 | { |
2555 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
2556 | ||
0ddeded4 | 2557 | __gfs2_free_blocks(ip, rgd, bstart, blen, 1); |
b3b94faa | 2558 | gfs2_statfs_change(sdp, 0, +blen, 0); |
2933f925 | 2559 | gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); |
b3b94faa DT |
2560 | } |
2561 | ||
feaa7bba SW |
2562 | void gfs2_unlink_di(struct inode *inode) |
2563 | { | |
2564 | struct gfs2_inode *ip = GFS2_I(inode); | |
2565 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
2566 | struct gfs2_rgrpd *rgd; | |
dbb7cae2 | 2567 | u64 blkno = ip->i_no_addr; |
feaa7bba | 2568 | |
0ddeded4 | 2569 | rgd = gfs2_blk2rgrpd(sdp, blkno, true); |
feaa7bba SW |
2570 | if (!rgd) |
2571 | return; | |
9e514605 | 2572 | rgrp_lock_local(rgd); |
0ddeded4 | 2573 | rgblk_free(sdp, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
41db1ab9 | 2574 | trace_gfs2_block_alloc(ip, rgd, blkno, 1, GFS2_BLKST_UNLINKED); |
350a9b0a | 2575 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2576 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
f5580d0f | 2577 | be32_add_cpu(&rgd->rd_rgl->rl_unlinked, 1); |
9e514605 | 2578 | rgrp_unlock_local(rgd); |
feaa7bba SW |
2579 | } |
2580 | ||
a18c78c5 | 2581 | void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) |
b3b94faa DT |
2582 | { |
2583 | struct gfs2_sbd *sdp = rgd->rd_sbd; | |
b3b94faa | 2584 | |
9e514605 | 2585 | rgrp_lock_local(rgd); |
0ddeded4 | 2586 | rgblk_free(sdp, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
73f74948 | 2587 | if (!rgd->rd_dinodes) |
b3b94faa | 2588 | gfs2_consist_rgrpd(rgd); |
73f74948 | 2589 | rgd->rd_dinodes--; |
cfc8b549 | 2590 | rgd->rd_free++; |
b3b94faa | 2591 | |
350a9b0a | 2592 | gfs2_trans_add_meta(rgd->rd_gl, rgd->rd_bits[0].bi_bh); |
42d52e38 | 2593 | gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); |
9e514605 | 2594 | rgrp_unlock_local(rgd); |
f5580d0f | 2595 | be32_add_cpu(&rgd->rd_rgl->rl_unlinked, -1); |
b3b94faa DT |
2596 | |
2597 | gfs2_statfs_change(sdp, 0, +1, -1); | |
41db1ab9 | 2598 | trace_gfs2_block_alloc(ip, rgd, ip->i_no_addr, 1, GFS2_BLKST_FREE); |
2933f925 | 2599 | gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); |
68942870 | 2600 | gfs2_journal_wipe(ip, ip->i_no_addr, 1); |
b3b94faa DT |
2601 | } |
2602 | ||
acf7e244 SW |
2603 | /** |
2604 | * gfs2_check_blk_type - Check the type of a block | |
2605 | * @sdp: The superblock | |
2606 | * @no_addr: The block number to check | |
2607 | * @type: The block type we are looking for | |
2608 | * | |
9e514605 AG |
2609 | * The inode glock of @no_addr must be held. The @type to check for is either |
2610 | * GFS2_BLKST_DINODE or GFS2_BLKST_UNLINKED; checking for type GFS2_BLKST_FREE | |
2611 | * or GFS2_BLKST_USED would make no sense. | |
2612 | * | |
acf7e244 SW |
2613 | * Returns: 0 if the block type matches the expected type |
2614 | * -ESTALE if it doesn't match | |
2615 | * or -ve errno if something went wrong while checking | |
2616 | */ | |
2617 | ||
2618 | int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) | |
2619 | { | |
2620 | struct gfs2_rgrpd *rgd; | |
8339ee54 | 2621 | struct gfs2_holder rgd_gh; |
dffe12a8 | 2622 | struct gfs2_rbm rbm; |
58884c4d | 2623 | int error = -EINVAL; |
acf7e244 | 2624 | |
66fc061b | 2625 | rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); |
acf7e244 | 2626 | if (!rgd) |
8339ee54 | 2627 | goto fail; |
acf7e244 SW |
2628 | |
2629 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); | |
2630 | if (error) | |
8339ee54 | 2631 | goto fail; |
acf7e244 | 2632 | |
dffe12a8 BP |
2633 | rbm.rgd = rgd; |
2634 | error = gfs2_rbm_from_block(&rbm, no_addr); | |
bc923818 | 2635 | if (!WARN_ON_ONCE(error)) { |
9e514605 AG |
2636 | /* |
2637 | * No need to take the local resource group lock here; the | |
2638 | * inode glock of @no_addr provides the necessary | |
2639 | * synchronization in case the block is an inode. (In case | |
2640 | * the block is not an inode, the block type will not match | |
2641 | * the @type we are looking for.) | |
2642 | */ | |
bc923818 ZQ |
2643 | if (gfs2_testbit(&rbm, false) != type) |
2644 | error = -ESTALE; | |
2645 | } | |
acf7e244 SW |
2646 | |
2647 | gfs2_glock_dq_uninit(&rgd_gh); | |
bc923818 | 2648 | |
acf7e244 SW |
2649 | fail: |
2650 | return error; | |
2651 | } | |
2652 | ||
b3b94faa DT |
2653 | /** |
2654 | * gfs2_rlist_add - add a RG to a list of RGs | |
70b0c365 | 2655 | * @ip: the inode |
b3b94faa DT |
2656 | * @rlist: the list of resource groups |
2657 | * @block: the block | |
2658 | * | |
2659 | * Figure out what RG a block belongs to and add that RG to the list | |
2660 | * | |
2661 | * FIXME: Don't use NOFAIL | |
2662 | * | |
2663 | */ | |
2664 | ||
70b0c365 | 2665 | void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, |
cd915493 | 2666 | u64 block) |
b3b94faa | 2667 | { |
70b0c365 | 2668 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
2669 | struct gfs2_rgrpd *rgd; |
2670 | struct gfs2_rgrpd **tmp; | |
2671 | unsigned int new_space; | |
2672 | unsigned int x; | |
2673 | ||
2674 | if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) | |
2675 | return; | |
2676 | ||
03f8c41c AG |
2677 | /* |
2678 | * The resource group last accessed is kept in the last position. | |
2679 | */ | |
2680 | ||
2681 | if (rlist->rl_rgrps) { | |
2682 | rgd = rlist->rl_rgd[rlist->rl_rgrps - 1]; | |
2683 | if (rgrp_contains_block(rgd, block)) | |
2684 | return; | |
66fc061b | 2685 | rgd = gfs2_blk2rgrpd(sdp, block, 1); |
03f8c41c | 2686 | } else { |
c65b76b8 | 2687 | rgd = ip->i_res.rs_rgd; |
03f8c41c AG |
2688 | if (!rgd || !rgrp_contains_block(rgd, block)) |
2689 | rgd = gfs2_blk2rgrpd(sdp, block, 1); | |
2690 | } | |
2691 | ||
b3b94faa | 2692 | if (!rgd) { |
03f8c41c AG |
2693 | fs_err(sdp, "rlist_add: no rgrp for block %llu\n", |
2694 | (unsigned long long)block); | |
b3b94faa DT |
2695 | return; |
2696 | } | |
2697 | ||
03f8c41c AG |
2698 | for (x = 0; x < rlist->rl_rgrps; x++) { |
2699 | if (rlist->rl_rgd[x] == rgd) { | |
2700 | swap(rlist->rl_rgd[x], | |
2701 | rlist->rl_rgd[rlist->rl_rgrps - 1]); | |
b3b94faa | 2702 | return; |
03f8c41c AG |
2703 | } |
2704 | } | |
b3b94faa DT |
2705 | |
2706 | if (rlist->rl_rgrps == rlist->rl_space) { | |
2707 | new_space = rlist->rl_space + 10; | |
2708 | ||
2709 | tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), | |
dd894be8 | 2710 | GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
2711 | |
2712 | if (rlist->rl_rgd) { | |
2713 | memcpy(tmp, rlist->rl_rgd, | |
2714 | rlist->rl_space * sizeof(struct gfs2_rgrpd *)); | |
2715 | kfree(rlist->rl_rgd); | |
2716 | } | |
2717 | ||
2718 | rlist->rl_space = new_space; | |
2719 | rlist->rl_rgd = tmp; | |
2720 | } | |
2721 | ||
2722 | rlist->rl_rgd[rlist->rl_rgrps++] = rgd; | |
2723 | } | |
2724 | ||
2725 | /** | |
2726 | * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate | |
2727 | * and initialize an array of glock holders for them | |
2728 | * @rlist: the list of resource groups | |
b3b94faa DT |
2729 | * |
2730 | * FIXME: Don't use NOFAIL | |
2731 | * | |
2732 | */ | |
2733 | ||
c3abc29e | 2734 | void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist) |
b3b94faa DT |
2735 | { |
2736 | unsigned int x; | |
2737 | ||
6da2ec56 KC |
2738 | rlist->rl_ghs = kmalloc_array(rlist->rl_rgrps, |
2739 | sizeof(struct gfs2_holder), | |
2740 | GFP_NOFS | __GFP_NOFAIL); | |
b3b94faa | 2741 | for (x = 0; x < rlist->rl_rgrps; x++) |
4fc7ec31 BP |
2742 | gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, LM_ST_EXCLUSIVE, |
2743 | LM_FLAG_NODE_SCOPE, &rlist->rl_ghs[x]); | |
b3b94faa DT |
2744 | } |
2745 | ||
2746 | /** | |
2747 | * gfs2_rlist_free - free a resource group list | |
27ff6a0f | 2748 | * @rlist: the list of resource groups |
b3b94faa DT |
2749 | * |
2750 | */ | |
2751 | ||
2752 | void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) | |
2753 | { | |
2754 | unsigned int x; | |
2755 | ||
2756 | kfree(rlist->rl_rgd); | |
2757 | ||
2758 | if (rlist->rl_ghs) { | |
2759 | for (x = 0; x < rlist->rl_rgrps; x++) | |
2760 | gfs2_holder_uninit(&rlist->rl_ghs[x]); | |
2761 | kfree(rlist->rl_ghs); | |
8e2e0047 | 2762 | rlist->rl_ghs = NULL; |
b3b94faa DT |
2763 | } |
2764 | } | |
2765 | ||
9e514605 AG |
2766 | void rgrp_lock_local(struct gfs2_rgrpd *rgd) |
2767 | { | |
2768 | BUG_ON(!gfs2_glock_is_held_excl(rgd->rd_gl) && | |
2769 | !test_bit(SDF_NORECOVERY, &rgd->rd_sbd->sd_flags)); | |
2770 | mutex_lock(&rgd->rd_mutex); | |
2771 | } | |
2772 | ||
2773 | void rgrp_unlock_local(struct gfs2_rgrpd *rgd) | |
2774 | { | |
2775 | mutex_unlock(&rgd->rd_mutex); | |
2776 | } |