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