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 20 #include "gfs2.h" 21 #include "incore.h" 22 #include "glock.h" 23 #include "glops.h" 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" 30 #include "util.h" 31 #include "log.h" 32 #include "inode.h" 33 #include "trace_gfs2.h" 34 35 #define BFITNOENT ((u32)~0) 36 #define NO_BLOCK ((u64)~0) 37 38 #if BITS_PER_LONG == 32 39 #define LBITMASK (0x55555555UL) 40 #define LBITSKIP55 (0x55555555UL) 41 #define LBITSKIP00 (0x00000000UL) 42 #else 43 #define LBITMASK (0x5555555555555555UL) 44 #define LBITSKIP55 (0x5555555555555555UL) 45 #define LBITSKIP00 (0x0000000000000000UL) 46 #endif 47 48 /* 49 * These routines are used by the resource group routines (rgrp.c) 50 * to keep track of block allocation. Each block is represented by two 51 * bits. So, each byte represents GFS2_NBBY (i.e. 4) blocks. 52 * 53 * 0 = Free 54 * 1 = Used (not metadata) 55 * 2 = Unlinked (still in use) inode 56 * 3 = Used (metadata) 57 */ 58 59 static const char valid_change[16] = { 60 /* current */ 61 /* n */ 0, 1, 1, 1, 62 /* e */ 1, 0, 0, 0, 63 /* w */ 0, 0, 0, 1, 64 1, 0, 0, 0 65 }; 66 67 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, 68 unsigned char old_state, 69 struct gfs2_bitmap **rbi); 70 71 /** 72 * gfs2_setbit - Set a bit in the bitmaps 73 * @buffer: the buffer that holds the bitmaps 74 * @buflen: the length (in bytes) of the buffer 75 * @block: the block to set 76 * @new_state: the new state of the block 77 * 78 */ 79 80 static inline void gfs2_setbit(struct gfs2_rgrpd *rgd, unsigned char *buf1, 81 unsigned char *buf2, unsigned int offset, 82 struct gfs2_bitmap *bi, u32 block, 83 unsigned char new_state) 84 { 85 unsigned char *byte1, *byte2, *end, cur_state; 86 unsigned int buflen = bi->bi_len; 87 const unsigned int bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; 88 89 byte1 = buf1 + offset + (block / GFS2_NBBY); 90 end = buf1 + offset + buflen; 91 92 BUG_ON(byte1 >= end); 93 94 cur_state = (*byte1 >> bit) & GFS2_BIT_MASK; 95 96 if (unlikely(!valid_change[new_state * 4 + cur_state])) { 97 printk(KERN_WARNING "GFS2: buf_blk = 0x%llx old_state=%d, " 98 "new_state=%d\n", 99 (unsigned long long)block, cur_state, new_state); 100 printk(KERN_WARNING "GFS2: rgrp=0x%llx bi_start=0x%lx\n", 101 (unsigned long long)rgd->rd_addr, 102 (unsigned long)bi->bi_start); 103 printk(KERN_WARNING "GFS2: bi_offset=0x%lx bi_len=0x%lx\n", 104 (unsigned long)bi->bi_offset, 105 (unsigned long)bi->bi_len); 106 dump_stack(); 107 gfs2_consist_rgrpd(rgd); 108 return; 109 } 110 *byte1 ^= (cur_state ^ new_state) << bit; 111 112 if (buf2) { 113 byte2 = buf2 + offset + (block / GFS2_NBBY); 114 cur_state = (*byte2 >> bit) & GFS2_BIT_MASK; 115 *byte2 ^= (cur_state ^ new_state) << bit; 116 } 117 } 118 119 /** 120 * gfs2_testbit - test a bit in the bitmaps 121 * @buffer: the buffer that holds the bitmaps 122 * @buflen: the length (in bytes) of the buffer 123 * @block: the block to read 124 * 125 */ 126 127 static inline unsigned char gfs2_testbit(struct gfs2_rgrpd *rgd, 128 const unsigned char *buffer, 129 unsigned int buflen, u32 block) 130 { 131 const unsigned char *byte, *end; 132 unsigned char cur_state; 133 unsigned int bit; 134 135 byte = buffer + (block / GFS2_NBBY); 136 bit = (block % GFS2_NBBY) * GFS2_BIT_SIZE; 137 end = buffer + buflen; 138 139 gfs2_assert(rgd->rd_sbd, byte < end); 140 141 cur_state = (*byte >> bit) & GFS2_BIT_MASK; 142 143 return cur_state; 144 } 145 146 /** 147 * gfs2_bit_search 148 * @ptr: Pointer to bitmap data 149 * @mask: Mask to use (normally 0x55555.... but adjusted for search start) 150 * @state: The state we are searching for 151 * 152 * We xor the bitmap data with a patter which is the bitwise opposite 153 * of what we are looking for, this gives rise to a pattern of ones 154 * wherever there is a match. Since we have two bits per entry, we 155 * take this pattern, shift it down by one place and then and it with 156 * the original. All the even bit positions (0,2,4, etc) then represent 157 * successful matches, so we mask with 0x55555..... to remove the unwanted 158 * odd bit positions. 159 * 160 * This allows searching of a whole u64 at once (32 blocks) with a 161 * single test (on 64 bit arches). 162 */ 163 164 static inline u64 gfs2_bit_search(const __le64 *ptr, u64 mask, u8 state) 165 { 166 u64 tmp; 167 static const u64 search[] = { 168 [0] = 0xffffffffffffffffULL, 169 [1] = 0xaaaaaaaaaaaaaaaaULL, 170 [2] = 0x5555555555555555ULL, 171 [3] = 0x0000000000000000ULL, 172 }; 173 tmp = le64_to_cpu(*ptr) ^ search[state]; 174 tmp &= (tmp >> 1); 175 tmp &= mask; 176 return tmp; 177 } 178 179 /** 180 * gfs2_bitfit - Search an rgrp's bitmap buffer to find a bit-pair representing 181 * a block in a given allocation state. 182 * @buffer: the buffer that holds the bitmaps 183 * @len: the length (in bytes) of the buffer 184 * @goal: start search at this block's bit-pair (within @buffer) 185 * @state: GFS2_BLKST_XXX the state of the block we're looking for. 186 * 187 * Scope of @goal and returned block number is only within this bitmap buffer, 188 * not entire rgrp or filesystem. @buffer will be offset from the actual 189 * beginning of a bitmap block buffer, skipping any header structures, but 190 * headers are always a multiple of 64 bits long so that the buffer is 191 * always aligned to a 64 bit boundary. 192 * 193 * The size of the buffer is in bytes, but is it assumed that it is 194 * always ok to read a complete multiple of 64 bits at the end 195 * of the block in case the end is no aligned to a natural boundary. 196 * 197 * Return: the block number (bitmap buffer scope) that was found 198 */ 199 200 static u32 gfs2_bitfit(const u8 *buf, const unsigned int len, 201 u32 goal, u8 state) 202 { 203 u32 spoint = (goal << 1) & ((8*sizeof(u64)) - 1); 204 const __le64 *ptr = ((__le64 *)buf) + (goal >> 5); 205 const __le64 *end = (__le64 *)(buf + ALIGN(len, sizeof(u64))); 206 u64 tmp; 207 u64 mask = 0x5555555555555555ULL; 208 u32 bit; 209 210 BUG_ON(state > 3); 211 212 /* Mask off bits we don't care about at the start of the search */ 213 mask <<= spoint; 214 tmp = gfs2_bit_search(ptr, mask, state); 215 ptr++; 216 while(tmp == 0 && ptr < end) { 217 tmp = gfs2_bit_search(ptr, 0x5555555555555555ULL, state); 218 ptr++; 219 } 220 /* Mask off any bits which are more than len bytes from the start */ 221 if (ptr == end && (len & (sizeof(u64) - 1))) 222 tmp &= (((u64)~0) >> (64 - 8*(len & (sizeof(u64) - 1)))); 223 /* Didn't find anything, so return */ 224 if (tmp == 0) 225 return BFITNOENT; 226 ptr--; 227 bit = __ffs64(tmp); 228 bit /= 2; /* two bits per entry in the bitmap */ 229 return (((const unsigned char *)ptr - buf) * GFS2_NBBY) + bit; 230 } 231 232 /** 233 * gfs2_bitcount - count the number of bits in a certain state 234 * @buffer: the buffer that holds the bitmaps 235 * @buflen: the length (in bytes) of the buffer 236 * @state: the state of the block we're looking for 237 * 238 * Returns: The number of bits 239 */ 240 241 static u32 gfs2_bitcount(struct gfs2_rgrpd *rgd, const u8 *buffer, 242 unsigned int buflen, u8 state) 243 { 244 const u8 *byte = buffer; 245 const u8 *end = buffer + buflen; 246 const u8 state1 = state << 2; 247 const u8 state2 = state << 4; 248 const u8 state3 = state << 6; 249 u32 count = 0; 250 251 for (; byte < end; byte++) { 252 if (((*byte) & 0x03) == state) 253 count++; 254 if (((*byte) & 0x0C) == state1) 255 count++; 256 if (((*byte) & 0x30) == state2) 257 count++; 258 if (((*byte) & 0xC0) == state3) 259 count++; 260 } 261 262 return count; 263 } 264 265 /** 266 * gfs2_rgrp_verify - Verify that a resource group is consistent 267 * @sdp: the filesystem 268 * @rgd: the rgrp 269 * 270 */ 271 272 void gfs2_rgrp_verify(struct gfs2_rgrpd *rgd) 273 { 274 struct gfs2_sbd *sdp = rgd->rd_sbd; 275 struct gfs2_bitmap *bi = NULL; 276 u32 length = rgd->rd_length; 277 u32 count[4], tmp; 278 int buf, x; 279 280 memset(count, 0, 4 * sizeof(u32)); 281 282 /* Count # blocks in each of 4 possible allocation states */ 283 for (buf = 0; buf < length; buf++) { 284 bi = rgd->rd_bits + buf; 285 for (x = 0; x < 4; x++) 286 count[x] += gfs2_bitcount(rgd, 287 bi->bi_bh->b_data + 288 bi->bi_offset, 289 bi->bi_len, x); 290 } 291 292 if (count[0] != rgd->rd_free) { 293 if (gfs2_consist_rgrpd(rgd)) 294 fs_err(sdp, "free data mismatch: %u != %u\n", 295 count[0], rgd->rd_free); 296 return; 297 } 298 299 tmp = rgd->rd_data - rgd->rd_free - rgd->rd_dinodes; 300 if (count[1] != tmp) { 301 if (gfs2_consist_rgrpd(rgd)) 302 fs_err(sdp, "used data mismatch: %u != %u\n", 303 count[1], tmp); 304 return; 305 } 306 307 if (count[2] + count[3] != rgd->rd_dinodes) { 308 if (gfs2_consist_rgrpd(rgd)) 309 fs_err(sdp, "used metadata mismatch: %u != %u\n", 310 count[2] + count[3], rgd->rd_dinodes); 311 return; 312 } 313 } 314 315 static inline int rgrp_contains_block(struct gfs2_rgrpd *rgd, u64 block) 316 { 317 u64 first = rgd->rd_data0; 318 u64 last = first + rgd->rd_data; 319 return first <= block && block < last; 320 } 321 322 /** 323 * gfs2_blk2rgrpd - Find resource group for a given data/meta block number 324 * @sdp: The GFS2 superblock 325 * @n: The data block number 326 * 327 * Returns: The resource group, or NULL if not found 328 */ 329 330 struct gfs2_rgrpd *gfs2_blk2rgrpd(struct gfs2_sbd *sdp, u64 blk) 331 { 332 struct rb_node **newn; 333 struct gfs2_rgrpd *cur; 334 335 spin_lock(&sdp->sd_rindex_spin); 336 newn = &sdp->sd_rindex_tree.rb_node; 337 while (*newn) { 338 cur = rb_entry(*newn, struct gfs2_rgrpd, rd_node); 339 if (blk < cur->rd_addr) 340 newn = &((*newn)->rb_left); 341 else if (blk >= cur->rd_data0 + cur->rd_data) 342 newn = &((*newn)->rb_right); 343 else { 344 spin_unlock(&sdp->sd_rindex_spin); 345 return cur; 346 } 347 } 348 spin_unlock(&sdp->sd_rindex_spin); 349 350 return NULL; 351 } 352 353 /** 354 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem 355 * @sdp: The GFS2 superblock 356 * 357 * Returns: The first rgrp in the filesystem 358 */ 359 360 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) 361 { 362 const struct rb_node *n; 363 struct gfs2_rgrpd *rgd; 364 365 spin_lock(&sdp->sd_rindex_spin); 366 n = rb_first(&sdp->sd_rindex_tree); 367 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); 368 spin_unlock(&sdp->sd_rindex_spin); 369 370 return rgd; 371 } 372 373 /** 374 * gfs2_rgrpd_get_next - get the next RG 375 * @rgd: A RG 376 * 377 * Returns: The next rgrp 378 */ 379 380 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) 381 { 382 struct gfs2_sbd *sdp = rgd->rd_sbd; 383 const struct rb_node *n; 384 385 spin_lock(&sdp->sd_rindex_spin); 386 n = rb_next(&rgd->rd_node); 387 if (n == NULL) 388 n = rb_first(&sdp->sd_rindex_tree); 389 390 if (unlikely(&rgd->rd_node == n)) { 391 spin_unlock(&sdp->sd_rindex_spin); 392 return NULL; 393 } 394 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); 395 spin_unlock(&sdp->sd_rindex_spin); 396 return rgd; 397 } 398 399 void gfs2_free_clones(struct gfs2_rgrpd *rgd) 400 { 401 int x; 402 403 for (x = 0; x < rgd->rd_length; x++) { 404 struct gfs2_bitmap *bi = rgd->rd_bits + x; 405 kfree(bi->bi_clone); 406 bi->bi_clone = NULL; 407 } 408 } 409 410 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) 411 { 412 struct rb_node *n; 413 struct gfs2_rgrpd *rgd; 414 struct gfs2_glock *gl; 415 416 while ((n = rb_first(&sdp->sd_rindex_tree))) { 417 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); 418 gl = rgd->rd_gl; 419 420 rb_erase(n, &sdp->sd_rindex_tree); 421 422 if (gl) { 423 spin_lock(&gl->gl_spin); 424 gl->gl_object = NULL; 425 spin_unlock(&gl->gl_spin); 426 gfs2_glock_add_to_lru(gl); 427 gfs2_glock_put(gl); 428 } 429 430 gfs2_free_clones(rgd); 431 kfree(rgd->rd_bits); 432 kmem_cache_free(gfs2_rgrpd_cachep, rgd); 433 } 434 } 435 436 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) 437 { 438 printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); 439 printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); 440 printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); 441 printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); 442 printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); 443 } 444 445 /** 446 * gfs2_compute_bitstructs - Compute the bitmap sizes 447 * @rgd: The resource group descriptor 448 * 449 * Calculates bitmap descriptors, one for each block that contains bitmap data 450 * 451 * Returns: errno 452 */ 453 454 static int compute_bitstructs(struct gfs2_rgrpd *rgd) 455 { 456 struct gfs2_sbd *sdp = rgd->rd_sbd; 457 struct gfs2_bitmap *bi; 458 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ 459 u32 bytes_left, bytes; 460 int x; 461 462 if (!length) 463 return -EINVAL; 464 465 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); 466 if (!rgd->rd_bits) 467 return -ENOMEM; 468 469 bytes_left = rgd->rd_bitbytes; 470 471 for (x = 0; x < length; x++) { 472 bi = rgd->rd_bits + x; 473 474 bi->bi_flags = 0; 475 /* small rgrp; bitmap stored completely in header block */ 476 if (length == 1) { 477 bytes = bytes_left; 478 bi->bi_offset = sizeof(struct gfs2_rgrp); 479 bi->bi_start = 0; 480 bi->bi_len = bytes; 481 /* header block */ 482 } else if (x == 0) { 483 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); 484 bi->bi_offset = sizeof(struct gfs2_rgrp); 485 bi->bi_start = 0; 486 bi->bi_len = bytes; 487 /* last block */ 488 } else if (x + 1 == length) { 489 bytes = bytes_left; 490 bi->bi_offset = sizeof(struct gfs2_meta_header); 491 bi->bi_start = rgd->rd_bitbytes - bytes_left; 492 bi->bi_len = bytes; 493 /* other blocks */ 494 } else { 495 bytes = sdp->sd_sb.sb_bsize - 496 sizeof(struct gfs2_meta_header); 497 bi->bi_offset = sizeof(struct gfs2_meta_header); 498 bi->bi_start = rgd->rd_bitbytes - bytes_left; 499 bi->bi_len = bytes; 500 } 501 502 bytes_left -= bytes; 503 } 504 505 if (bytes_left) { 506 gfs2_consist_rgrpd(rgd); 507 return -EIO; 508 } 509 bi = rgd->rd_bits + (length - 1); 510 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { 511 if (gfs2_consist_rgrpd(rgd)) { 512 gfs2_rindex_print(rgd); 513 fs_err(sdp, "start=%u len=%u offset=%u\n", 514 bi->bi_start, bi->bi_len, bi->bi_offset); 515 } 516 return -EIO; 517 } 518 519 return 0; 520 } 521 522 /** 523 * gfs2_ri_total - Total up the file system space, according to the rindex. 524 * 525 */ 526 u64 gfs2_ri_total(struct gfs2_sbd *sdp) 527 { 528 u64 total_data = 0; 529 struct inode *inode = sdp->sd_rindex; 530 struct gfs2_inode *ip = GFS2_I(inode); 531 char buf[sizeof(struct gfs2_rindex)]; 532 struct file_ra_state ra_state; 533 int error, rgrps; 534 535 mutex_lock(&sdp->sd_rindex_mutex); 536 file_ra_state_init(&ra_state, inode->i_mapping); 537 for (rgrps = 0;; rgrps++) { 538 loff_t pos = rgrps * sizeof(struct gfs2_rindex); 539 540 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) 541 break; 542 error = gfs2_internal_read(ip, &ra_state, buf, &pos, 543 sizeof(struct gfs2_rindex)); 544 if (error != sizeof(struct gfs2_rindex)) 545 break; 546 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); 547 } 548 mutex_unlock(&sdp->sd_rindex_mutex); 549 return total_data; 550 } 551 552 static void rgd_insert(struct gfs2_rgrpd *rgd) 553 { 554 struct gfs2_sbd *sdp = rgd->rd_sbd; 555 struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; 556 557 /* Figure out where to put new node */ 558 while (*newn) { 559 struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, 560 rd_node); 561 562 parent = *newn; 563 if (rgd->rd_addr < cur->rd_addr) 564 newn = &((*newn)->rb_left); 565 else if (rgd->rd_addr > cur->rd_addr) 566 newn = &((*newn)->rb_right); 567 else 568 return; 569 } 570 571 rb_link_node(&rgd->rd_node, parent, newn); 572 rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); 573 } 574 575 /** 576 * read_rindex_entry - Pull in a new resource index entry from the disk 577 * @gl: The glock covering the rindex inode 578 * 579 * Returns: 0 on success, > 0 on EOF, error code otherwise 580 */ 581 582 static int read_rindex_entry(struct gfs2_inode *ip, 583 struct file_ra_state *ra_state) 584 { 585 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 586 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); 587 struct gfs2_rindex buf; 588 int error; 589 struct gfs2_rgrpd *rgd; 590 591 if (pos >= i_size_read(&ip->i_inode)) 592 return 1; 593 594 error = gfs2_internal_read(ip, ra_state, (char *)&buf, &pos, 595 sizeof(struct gfs2_rindex)); 596 597 if (error != sizeof(struct gfs2_rindex)) 598 return (error == 0) ? 1 : error; 599 600 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); 601 error = -ENOMEM; 602 if (!rgd) 603 return error; 604 605 rgd->rd_sbd = sdp; 606 rgd->rd_addr = be64_to_cpu(buf.ri_addr); 607 rgd->rd_length = be32_to_cpu(buf.ri_length); 608 rgd->rd_data0 = be64_to_cpu(buf.ri_data0); 609 rgd->rd_data = be32_to_cpu(buf.ri_data); 610 rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); 611 612 error = compute_bitstructs(rgd); 613 if (error) 614 goto fail; 615 616 error = gfs2_glock_get(sdp, rgd->rd_addr, 617 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); 618 if (error) 619 goto fail; 620 621 rgd->rd_gl->gl_object = rgd; 622 rgd->rd_flags &= ~GFS2_RDF_UPTODATE; 623 if (rgd->rd_data > sdp->sd_max_rg_data) 624 sdp->sd_max_rg_data = rgd->rd_data; 625 spin_lock(&sdp->sd_rindex_spin); 626 rgd_insert(rgd); 627 sdp->sd_rgrps++; 628 spin_unlock(&sdp->sd_rindex_spin); 629 return error; 630 631 fail: 632 kfree(rgd->rd_bits); 633 kmem_cache_free(gfs2_rgrpd_cachep, rgd); 634 return error; 635 } 636 637 /** 638 * gfs2_ri_update - Pull in a new resource index from the disk 639 * @ip: pointer to the rindex inode 640 * 641 * Returns: 0 on successful update, error code otherwise 642 */ 643 644 static int gfs2_ri_update(struct gfs2_inode *ip) 645 { 646 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 647 struct inode *inode = &ip->i_inode; 648 struct file_ra_state ra_state; 649 int error; 650 651 file_ra_state_init(&ra_state, inode->i_mapping); 652 do { 653 error = read_rindex_entry(ip, &ra_state); 654 } while (error == 0); 655 656 if (error < 0) 657 return error; 658 659 sdp->sd_rindex_uptodate = 1; 660 return 0; 661 } 662 663 /** 664 * gfs2_rindex_update - Update the rindex if required 665 * @sdp: The GFS2 superblock 666 * 667 * We grab a lock on the rindex inode to make sure that it doesn't 668 * change whilst we are performing an operation. We keep this lock 669 * for quite long periods of time compared to other locks. This 670 * doesn't matter, since it is shared and it is very, very rarely 671 * accessed in the exclusive mode (i.e. only when expanding the filesystem). 672 * 673 * This makes sure that we're using the latest copy of the resource index 674 * special file, which might have been updated if someone expanded the 675 * filesystem (via gfs2_grow utility), which adds new resource groups. 676 * 677 * Returns: 0 on succeess, error code otherwise 678 */ 679 680 int gfs2_rindex_update(struct gfs2_sbd *sdp) 681 { 682 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); 683 struct gfs2_glock *gl = ip->i_gl; 684 struct gfs2_holder ri_gh; 685 int error = 0; 686 int unlock_required = 0; 687 688 /* Read new copy from disk if we don't have the latest */ 689 if (!sdp->sd_rindex_uptodate) { 690 mutex_lock(&sdp->sd_rindex_mutex); 691 if (!gfs2_glock_is_locked_by_me(gl)) { 692 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); 693 if (error) 694 return error; 695 unlock_required = 1; 696 } 697 if (!sdp->sd_rindex_uptodate) 698 error = gfs2_ri_update(ip); 699 if (unlock_required) 700 gfs2_glock_dq_uninit(&ri_gh); 701 mutex_unlock(&sdp->sd_rindex_mutex); 702 } 703 704 705 return error; 706 } 707 708 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) 709 { 710 const struct gfs2_rgrp *str = buf; 711 u32 rg_flags; 712 713 rg_flags = be32_to_cpu(str->rg_flags); 714 rg_flags &= ~GFS2_RDF_MASK; 715 rgd->rd_flags &= GFS2_RDF_MASK; 716 rgd->rd_flags |= rg_flags; 717 rgd->rd_free = be32_to_cpu(str->rg_free); 718 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); 719 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); 720 } 721 722 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) 723 { 724 struct gfs2_rgrp *str = buf; 725 726 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); 727 str->rg_free = cpu_to_be32(rgd->rd_free); 728 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); 729 str->__pad = cpu_to_be32(0); 730 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); 731 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); 732 } 733 734 /** 735 * gfs2_rgrp_go_lock - Read in a RG's header and bitmaps 736 * @rgd: the struct gfs2_rgrpd describing the RG to read in 737 * 738 * Read in all of a Resource Group's header and bitmap blocks. 739 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. 740 * 741 * Returns: errno 742 */ 743 744 int gfs2_rgrp_go_lock(struct gfs2_holder *gh) 745 { 746 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; 747 struct gfs2_sbd *sdp = rgd->rd_sbd; 748 struct gfs2_glock *gl = rgd->rd_gl; 749 unsigned int length = rgd->rd_length; 750 struct gfs2_bitmap *bi; 751 unsigned int x, y; 752 int error; 753 754 for (x = 0; x < length; x++) { 755 bi = rgd->rd_bits + x; 756 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); 757 if (error) 758 goto fail; 759 } 760 761 for (y = length; y--;) { 762 bi = rgd->rd_bits + y; 763 error = gfs2_meta_wait(sdp, bi->bi_bh); 764 if (error) 765 goto fail; 766 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : 767 GFS2_METATYPE_RG)) { 768 error = -EIO; 769 goto fail; 770 } 771 } 772 773 if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { 774 for (x = 0; x < length; x++) 775 clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); 776 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); 777 rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); 778 rgd->rd_free_clone = rgd->rd_free; 779 } 780 781 return 0; 782 783 fail: 784 while (x--) { 785 bi = rgd->rd_bits + x; 786 brelse(bi->bi_bh); 787 bi->bi_bh = NULL; 788 gfs2_assert_warn(sdp, !bi->bi_clone); 789 } 790 791 return error; 792 } 793 794 /** 795 * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() 796 * @rgd: the struct gfs2_rgrpd describing the RG to read in 797 * 798 */ 799 800 void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) 801 { 802 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; 803 int x, length = rgd->rd_length; 804 805 for (x = 0; x < length; x++) { 806 struct gfs2_bitmap *bi = rgd->rd_bits + x; 807 brelse(bi->bi_bh); 808 bi->bi_bh = NULL; 809 } 810 811 } 812 813 void gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, 814 struct buffer_head *bh, 815 const struct gfs2_bitmap *bi) 816 { 817 struct super_block *sb = sdp->sd_vfs; 818 struct block_device *bdev = sb->s_bdev; 819 const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / 820 bdev_logical_block_size(sb->s_bdev); 821 u64 blk; 822 sector_t start = 0; 823 sector_t nr_sects = 0; 824 int rv; 825 unsigned int x; 826 827 for (x = 0; x < bi->bi_len; x++) { 828 const u8 *orig = bh->b_data + bi->bi_offset + x; 829 const u8 *clone = bi->bi_clone + bi->bi_offset + x; 830 u8 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); 831 diff &= 0x55; 832 if (diff == 0) 833 continue; 834 blk = offset + ((bi->bi_start + x) * GFS2_NBBY); 835 blk *= sects_per_blk; /* convert to sectors */ 836 while(diff) { 837 if (diff & 1) { 838 if (nr_sects == 0) 839 goto start_new_extent; 840 if ((start + nr_sects) != blk) { 841 rv = blkdev_issue_discard(bdev, start, 842 nr_sects, GFP_NOFS, 843 0); 844 if (rv) 845 goto fail; 846 nr_sects = 0; 847 start_new_extent: 848 start = blk; 849 } 850 nr_sects += sects_per_blk; 851 } 852 diff >>= 2; 853 blk += sects_per_blk; 854 } 855 } 856 if (nr_sects) { 857 rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0); 858 if (rv) 859 goto fail; 860 } 861 return; 862 fail: 863 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); 864 sdp->sd_args.ar_discard = 0; 865 } 866 867 /** 868 * gfs2_qadata_get - get the struct gfs2_qadata structure for an inode 869 * @ip: the incore GFS2 inode structure 870 * 871 * Returns: the struct gfs2_qadata 872 */ 873 874 struct gfs2_qadata *gfs2_qadata_get(struct gfs2_inode *ip) 875 { 876 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 877 int error; 878 BUG_ON(ip->i_qadata != NULL); 879 ip->i_qadata = kzalloc(sizeof(struct gfs2_qadata), GFP_NOFS); 880 error = gfs2_rindex_update(sdp); 881 if (error) 882 fs_warn(sdp, "rindex update returns %d\n", error); 883 return ip->i_qadata; 884 } 885 886 /** 887 * gfs2_blkrsv_get - get the struct gfs2_blkreserv structure for an inode 888 * @ip: the incore GFS2 inode structure 889 * 890 * Returns: the struct gfs2_qadata 891 */ 892 893 static struct gfs2_blkreserv *gfs2_blkrsv_get(struct gfs2_inode *ip) 894 { 895 BUG_ON(ip->i_res != NULL); 896 ip->i_res = kzalloc(sizeof(struct gfs2_blkreserv), GFP_NOFS); 897 return ip->i_res; 898 } 899 900 /** 901 * try_rgrp_fit - See if a given reservation will fit in a given RG 902 * @rgd: the RG data 903 * @ip: the inode 904 * 905 * If there's room for the requested blocks to be allocated from the RG: 906 * 907 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) 908 */ 909 910 static int try_rgrp_fit(const struct gfs2_rgrpd *rgd, const struct gfs2_inode *ip) 911 { 912 const struct gfs2_blkreserv *rs = ip->i_res; 913 914 if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) 915 return 0; 916 if (rgd->rd_free_clone >= rs->rs_requested) 917 return 1; 918 return 0; 919 } 920 921 static inline u32 gfs2_bi2rgd_blk(struct gfs2_bitmap *bi, u32 blk) 922 { 923 return (bi->bi_start * GFS2_NBBY) + blk; 924 } 925 926 /** 927 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes 928 * @rgd: The rgrp 929 * 930 * Returns: 0 if no error 931 * The inode, if one has been found, in inode. 932 */ 933 934 static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) 935 { 936 u32 goal = 0, block; 937 u64 no_addr; 938 struct gfs2_sbd *sdp = rgd->rd_sbd; 939 struct gfs2_glock *gl; 940 struct gfs2_inode *ip; 941 int error; 942 int found = 0; 943 struct gfs2_bitmap *bi; 944 945 while (goal < rgd->rd_data) { 946 down_write(&sdp->sd_log_flush_lock); 947 block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, &bi); 948 up_write(&sdp->sd_log_flush_lock); 949 if (block == BFITNOENT) 950 break; 951 952 block = gfs2_bi2rgd_blk(bi, block); 953 /* rgblk_search can return a block < goal, so we need to 954 keep it marching forward. */ 955 no_addr = block + rgd->rd_data0; 956 goal = max(block + 1, goal + 1); 957 if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked) 958 continue; 959 if (no_addr == skip) 960 continue; 961 *last_unlinked = no_addr; 962 963 error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &gl); 964 if (error) 965 continue; 966 967 /* If the inode is already in cache, we can ignore it here 968 * because the existing inode disposal code will deal with 969 * it when all refs have gone away. Accessing gl_object like 970 * this is not safe in general. Here it is ok because we do 971 * not dereference the pointer, and we only need an approx 972 * answer to whether it is NULL or not. 973 */ 974 ip = gl->gl_object; 975 976 if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) 977 gfs2_glock_put(gl); 978 else 979 found++; 980 981 /* Limit reclaim to sensible number of tasks */ 982 if (found > NR_CPUS) 983 return; 984 } 985 986 rgd->rd_flags &= ~GFS2_RDF_CHECK; 987 return; 988 } 989 990 /** 991 * get_local_rgrp - Choose and lock a rgrp for allocation 992 * @ip: the inode to reserve space for 993 * @rgp: the chosen and locked rgrp 994 * 995 * Try to acquire rgrp in way which avoids contending with others. 996 * 997 * Returns: errno 998 */ 999 1000 static int get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked) 1001 { 1002 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1003 struct gfs2_rgrpd *rgd, *begin = NULL; 1004 struct gfs2_blkreserv *rs = ip->i_res; 1005 int error, rg_locked, flags = LM_FLAG_TRY; 1006 int loops = 0; 1007 1008 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) 1009 rgd = begin = ip->i_rgd; 1010 else 1011 rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal); 1012 1013 if (rgd == NULL) 1014 return -EBADSLT; 1015 1016 while (loops < 3) { 1017 rg_locked = 0; 1018 1019 if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) { 1020 rg_locked = 1; 1021 error = 0; 1022 } else { 1023 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 1024 flags, &rs->rs_rgd_gh); 1025 } 1026 switch (error) { 1027 case 0: 1028 if (try_rgrp_fit(rgd, ip)) { 1029 ip->i_rgd = rgd; 1030 return 0; 1031 } 1032 if (rgd->rd_flags & GFS2_RDF_CHECK) 1033 try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr); 1034 if (!rg_locked) 1035 gfs2_glock_dq_uninit(&rs->rs_rgd_gh); 1036 /* fall through */ 1037 case GLR_TRYFAILED: 1038 rgd = gfs2_rgrpd_get_next(rgd); 1039 if (rgd == begin) { 1040 flags = 0; 1041 loops++; 1042 } 1043 break; 1044 default: 1045 return error; 1046 } 1047 } 1048 1049 return -ENOSPC; 1050 } 1051 1052 static void gfs2_blkrsv_put(struct gfs2_inode *ip) 1053 { 1054 BUG_ON(ip->i_res == NULL); 1055 kfree(ip->i_res); 1056 ip->i_res = NULL; 1057 } 1058 1059 /** 1060 * gfs2_inplace_reserve - Reserve space in the filesystem 1061 * @ip: the inode to reserve space for 1062 * 1063 * Returns: errno 1064 */ 1065 1066 int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested) 1067 { 1068 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1069 struct gfs2_blkreserv *rs; 1070 int error = 0; 1071 u64 last_unlinked = NO_BLOCK; 1072 int tries = 0; 1073 1074 rs = gfs2_blkrsv_get(ip); 1075 if (!rs) 1076 return -ENOMEM; 1077 1078 rs->rs_requested = requested; 1079 if (gfs2_assert_warn(sdp, requested)) { 1080 error = -EINVAL; 1081 goto out; 1082 } 1083 1084 do { 1085 error = get_local_rgrp(ip, &last_unlinked); 1086 if (error != -ENOSPC) 1087 break; 1088 /* Check that fs hasn't grown if writing to rindex */ 1089 if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { 1090 error = gfs2_ri_update(ip); 1091 if (error) 1092 break; 1093 continue; 1094 } 1095 /* Flushing the log may release space */ 1096 gfs2_log_flush(sdp, NULL); 1097 } while (tries++ < 3); 1098 1099 out: 1100 if (error) 1101 gfs2_blkrsv_put(ip); 1102 return error; 1103 } 1104 1105 /** 1106 * gfs2_inplace_release - release an inplace reservation 1107 * @ip: the inode the reservation was taken out on 1108 * 1109 * Release a reservation made by gfs2_inplace_reserve(). 1110 */ 1111 1112 void gfs2_inplace_release(struct gfs2_inode *ip) 1113 { 1114 struct gfs2_blkreserv *rs = ip->i_res; 1115 1116 if (rs->rs_rgd_gh.gh_gl) 1117 gfs2_glock_dq_uninit(&rs->rs_rgd_gh); 1118 gfs2_blkrsv_put(ip); 1119 } 1120 1121 /** 1122 * gfs2_get_block_type - Check a block in a RG is of given type 1123 * @rgd: the resource group holding the block 1124 * @block: the block number 1125 * 1126 * Returns: The block type (GFS2_BLKST_*) 1127 */ 1128 1129 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) 1130 { 1131 struct gfs2_bitmap *bi = NULL; 1132 u32 length, rgrp_block, buf_block; 1133 unsigned int buf; 1134 unsigned char type; 1135 1136 length = rgd->rd_length; 1137 rgrp_block = block - rgd->rd_data0; 1138 1139 for (buf = 0; buf < length; buf++) { 1140 bi = rgd->rd_bits + buf; 1141 if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY) 1142 break; 1143 } 1144 1145 gfs2_assert(rgd->rd_sbd, buf < length); 1146 buf_block = rgrp_block - bi->bi_start * GFS2_NBBY; 1147 1148 type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset, 1149 bi->bi_len, buf_block); 1150 1151 return type; 1152 } 1153 1154 /** 1155 * rgblk_search - find a block in @state 1156 * @rgd: the resource group descriptor 1157 * @goal: the goal block within the RG (start here to search for avail block) 1158 * @state: GFS2_BLKST_XXX the before-allocation state to find 1159 * @dinode: TRUE if the first block we allocate is for a dinode 1160 * @rbi: address of the pointer to the bitmap containing the block found 1161 * 1162 * Walk rgrp's bitmap to find bits that represent a block in @state. 1163 * 1164 * This function never fails, because we wouldn't call it unless we 1165 * know (from reservation results, etc.) that a block is available. 1166 * 1167 * Scope of @goal is just within rgrp, not the whole filesystem. 1168 * Scope of @returned block is just within bitmap, not the whole filesystem. 1169 * 1170 * Returns: the block number found relative to the bitmap rbi 1171 */ 1172 1173 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, 1174 unsigned char state, 1175 struct gfs2_bitmap **rbi) 1176 { 1177 struct gfs2_bitmap *bi = NULL; 1178 const u32 length = rgd->rd_length; 1179 u32 blk = BFITNOENT; 1180 unsigned int buf, x; 1181 const u8 *buffer = NULL; 1182 1183 *rbi = NULL; 1184 /* Find bitmap block that contains bits for goal block */ 1185 for (buf = 0; buf < length; buf++) { 1186 bi = rgd->rd_bits + buf; 1187 /* Convert scope of "goal" from rgrp-wide to within found bit block */ 1188 if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { 1189 goal -= bi->bi_start * GFS2_NBBY; 1190 goto do_search; 1191 } 1192 } 1193 buf = 0; 1194 goal = 0; 1195 1196 do_search: 1197 /* Search (up to entire) bitmap in this rgrp for allocatable block. 1198 "x <= length", instead of "x < length", because we typically start 1199 the search in the middle of a bit block, but if we can't find an 1200 allocatable block anywhere else, we want to be able wrap around and 1201 search in the first part of our first-searched bit block. */ 1202 for (x = 0; x <= length; x++) { 1203 bi = rgd->rd_bits + buf; 1204 1205 if (test_bit(GBF_FULL, &bi->bi_flags) && 1206 (state == GFS2_BLKST_FREE)) 1207 goto skip; 1208 1209 /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone 1210 bitmaps, so we must search the originals for that. */ 1211 buffer = bi->bi_bh->b_data + bi->bi_offset; 1212 WARN_ON(!buffer_uptodate(bi->bi_bh)); 1213 if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) 1214 buffer = bi->bi_clone + bi->bi_offset; 1215 1216 blk = gfs2_bitfit(buffer, bi->bi_len, goal, state); 1217 if (blk != BFITNOENT) 1218 break; 1219 1220 if ((goal == 0) && (state == GFS2_BLKST_FREE)) 1221 set_bit(GBF_FULL, &bi->bi_flags); 1222 1223 /* Try next bitmap block (wrap back to rgrp header if at end) */ 1224 skip: 1225 buf++; 1226 buf %= length; 1227 goal = 0; 1228 } 1229 1230 if (blk != BFITNOENT) 1231 *rbi = bi; 1232 1233 return blk; 1234 } 1235 1236 /** 1237 * gfs2_alloc_extent - allocate an extent from a given bitmap 1238 * @rgd: the resource group descriptor 1239 * @bi: the bitmap within the rgrp 1240 * @blk: the block within the bitmap 1241 * @dinode: TRUE if the first block we allocate is for a dinode 1242 * @n: The extent length 1243 * 1244 * Add the found bitmap buffer to the transaction. 1245 * Set the found bits to @new_state to change block's allocation state. 1246 * Returns: starting block number of the extent (fs scope) 1247 */ 1248 static u64 gfs2_alloc_extent(struct gfs2_rgrpd *rgd, struct gfs2_bitmap *bi, 1249 u32 blk, bool dinode, unsigned int *n) 1250 { 1251 const unsigned int elen = *n; 1252 u32 goal; 1253 const u8 *buffer = NULL; 1254 1255 *n = 0; 1256 buffer = bi->bi_bh->b_data + bi->bi_offset; 1257 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); 1258 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, 1259 bi, blk, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); 1260 (*n)++; 1261 goal = blk; 1262 while (*n < elen) { 1263 goal++; 1264 if (goal >= (bi->bi_len * GFS2_NBBY)) 1265 break; 1266 if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) != 1267 GFS2_BLKST_FREE) 1268 break; 1269 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, 1270 bi, goal, GFS2_BLKST_USED); 1271 (*n)++; 1272 } 1273 blk = gfs2_bi2rgd_blk(bi, blk); 1274 rgd->rd_last_alloc = blk + *n - 1; 1275 return rgd->rd_data0 + blk; 1276 } 1277 1278 /** 1279 * rgblk_free - Change alloc state of given block(s) 1280 * @sdp: the filesystem 1281 * @bstart: the start of a run of blocks to free 1282 * @blen: the length of the block run (all must lie within ONE RG!) 1283 * @new_state: GFS2_BLKST_XXX the after-allocation block state 1284 * 1285 * Returns: Resource group containing the block(s) 1286 */ 1287 1288 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, 1289 u32 blen, unsigned char new_state) 1290 { 1291 struct gfs2_rgrpd *rgd; 1292 struct gfs2_bitmap *bi = NULL; 1293 u32 length, rgrp_blk, buf_blk; 1294 unsigned int buf; 1295 1296 rgd = gfs2_blk2rgrpd(sdp, bstart); 1297 if (!rgd) { 1298 if (gfs2_consist(sdp)) 1299 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); 1300 return NULL; 1301 } 1302 1303 length = rgd->rd_length; 1304 1305 rgrp_blk = bstart - rgd->rd_data0; 1306 1307 while (blen--) { 1308 for (buf = 0; buf < length; buf++) { 1309 bi = rgd->rd_bits + buf; 1310 if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY) 1311 break; 1312 } 1313 1314 gfs2_assert(rgd->rd_sbd, buf < length); 1315 1316 buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY; 1317 rgrp_blk++; 1318 1319 if (!bi->bi_clone) { 1320 bi->bi_clone = kmalloc(bi->bi_bh->b_size, 1321 GFP_NOFS | __GFP_NOFAIL); 1322 memcpy(bi->bi_clone + bi->bi_offset, 1323 bi->bi_bh->b_data + bi->bi_offset, 1324 bi->bi_len); 1325 } 1326 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); 1327 gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset, 1328 bi, buf_blk, new_state); 1329 } 1330 1331 return rgd; 1332 } 1333 1334 /** 1335 * gfs2_rgrp_dump - print out an rgrp 1336 * @seq: The iterator 1337 * @gl: The glock in question 1338 * 1339 */ 1340 1341 int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) 1342 { 1343 const struct gfs2_rgrpd *rgd = gl->gl_object; 1344 if (rgd == NULL) 1345 return 0; 1346 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n", 1347 (unsigned long long)rgd->rd_addr, rgd->rd_flags, 1348 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes); 1349 return 0; 1350 } 1351 1352 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) 1353 { 1354 struct gfs2_sbd *sdp = rgd->rd_sbd; 1355 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", 1356 (unsigned long long)rgd->rd_addr); 1357 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); 1358 gfs2_rgrp_dump(NULL, rgd->rd_gl); 1359 rgd->rd_flags |= GFS2_RDF_ERROR; 1360 } 1361 1362 /** 1363 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode 1364 * @ip: the inode to allocate the block for 1365 * @bn: Used to return the starting block number 1366 * @ndata: requested number of blocks/extent length (value/result) 1367 * @dinode: 1 if we're allocating a dinode block, else 0 1368 * @generation: the generation number of the inode 1369 * 1370 * Returns: 0 or error 1371 */ 1372 1373 int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, 1374 bool dinode, u64 *generation) 1375 { 1376 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1377 struct buffer_head *dibh; 1378 struct gfs2_rgrpd *rgd; 1379 unsigned int ndata; 1380 u32 goal, blk; /* block, within the rgrp scope */ 1381 u64 block; /* block, within the file system scope */ 1382 int error; 1383 struct gfs2_bitmap *bi; 1384 1385 /* Only happens if there is a bug in gfs2, return something distinctive 1386 * to ensure that it is noticed. 1387 */ 1388 if (ip->i_res == NULL) 1389 return -ECANCELED; 1390 1391 rgd = ip->i_rgd; 1392 1393 if (!dinode && rgrp_contains_block(rgd, ip->i_goal)) 1394 goal = ip->i_goal - rgd->rd_data0; 1395 else 1396 goal = rgd->rd_last_alloc; 1397 1398 blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, &bi); 1399 1400 /* Since all blocks are reserved in advance, this shouldn't happen */ 1401 if (blk == BFITNOENT) 1402 goto rgrp_error; 1403 1404 block = gfs2_alloc_extent(rgd, bi, blk, dinode, nblocks); 1405 ndata = *nblocks; 1406 if (dinode) 1407 ndata--; 1408 1409 if (!dinode) { 1410 ip->i_goal = block + ndata - 1; 1411 error = gfs2_meta_inode_buffer(ip, &dibh); 1412 if (error == 0) { 1413 struct gfs2_dinode *di = 1414 (struct gfs2_dinode *)dibh->b_data; 1415 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1416 di->di_goal_meta = di->di_goal_data = 1417 cpu_to_be64(ip->i_goal); 1418 brelse(dibh); 1419 } 1420 } 1421 if (rgd->rd_free < *nblocks) 1422 goto rgrp_error; 1423 1424 rgd->rd_free -= *nblocks; 1425 if (dinode) { 1426 rgd->rd_dinodes++; 1427 *generation = rgd->rd_igeneration++; 1428 if (*generation == 0) 1429 *generation = rgd->rd_igeneration++; 1430 } 1431 1432 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1433 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1434 1435 gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); 1436 if (dinode) 1437 gfs2_trans_add_unrevoke(sdp, block, 1); 1438 1439 /* 1440 * This needs reviewing to see why we cannot do the quota change 1441 * at this point in the dinode case. 1442 */ 1443 if (ndata) 1444 gfs2_quota_change(ip, ndata, ip->i_inode.i_uid, 1445 ip->i_inode.i_gid); 1446 1447 rgd->rd_free_clone -= *nblocks; 1448 trace_gfs2_block_alloc(ip, block, *nblocks, 1449 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); 1450 *bn = block; 1451 return 0; 1452 1453 rgrp_error: 1454 gfs2_rgrp_error(rgd); 1455 return -EIO; 1456 } 1457 1458 /** 1459 * __gfs2_free_blocks - free a contiguous run of block(s) 1460 * @ip: the inode these blocks are being freed from 1461 * @bstart: first block of a run of contiguous blocks 1462 * @blen: the length of the block run 1463 * @meta: 1 if the blocks represent metadata 1464 * 1465 */ 1466 1467 void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) 1468 { 1469 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1470 struct gfs2_rgrpd *rgd; 1471 1472 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); 1473 if (!rgd) 1474 return; 1475 trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE); 1476 rgd->rd_free += blen; 1477 1478 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1479 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1480 1481 /* Directories keep their data in the metadata address space */ 1482 if (meta || ip->i_depth) 1483 gfs2_meta_wipe(ip, bstart, blen); 1484 } 1485 1486 /** 1487 * gfs2_free_meta - free a contiguous run of data block(s) 1488 * @ip: the inode these blocks are being freed from 1489 * @bstart: first block of a run of contiguous blocks 1490 * @blen: the length of the block run 1491 * 1492 */ 1493 1494 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) 1495 { 1496 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1497 1498 __gfs2_free_blocks(ip, bstart, blen, 1); 1499 gfs2_statfs_change(sdp, 0, +blen, 0); 1500 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); 1501 } 1502 1503 void gfs2_unlink_di(struct inode *inode) 1504 { 1505 struct gfs2_inode *ip = GFS2_I(inode); 1506 struct gfs2_sbd *sdp = GFS2_SB(inode); 1507 struct gfs2_rgrpd *rgd; 1508 u64 blkno = ip->i_no_addr; 1509 1510 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); 1511 if (!rgd) 1512 return; 1513 trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED); 1514 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1515 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1516 } 1517 1518 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) 1519 { 1520 struct gfs2_sbd *sdp = rgd->rd_sbd; 1521 struct gfs2_rgrpd *tmp_rgd; 1522 1523 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); 1524 if (!tmp_rgd) 1525 return; 1526 gfs2_assert_withdraw(sdp, rgd == tmp_rgd); 1527 1528 if (!rgd->rd_dinodes) 1529 gfs2_consist_rgrpd(rgd); 1530 rgd->rd_dinodes--; 1531 rgd->rd_free++; 1532 1533 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1534 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1535 1536 gfs2_statfs_change(sdp, 0, +1, -1); 1537 } 1538 1539 1540 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) 1541 { 1542 gfs2_free_uninit_di(rgd, ip->i_no_addr); 1543 trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE); 1544 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); 1545 gfs2_meta_wipe(ip, ip->i_no_addr, 1); 1546 } 1547 1548 /** 1549 * gfs2_check_blk_type - Check the type of a block 1550 * @sdp: The superblock 1551 * @no_addr: The block number to check 1552 * @type: The block type we are looking for 1553 * 1554 * Returns: 0 if the block type matches the expected type 1555 * -ESTALE if it doesn't match 1556 * or -ve errno if something went wrong while checking 1557 */ 1558 1559 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) 1560 { 1561 struct gfs2_rgrpd *rgd; 1562 struct gfs2_holder rgd_gh; 1563 int error; 1564 1565 error = gfs2_rindex_update(sdp); 1566 if (error) 1567 return error; 1568 1569 error = -EINVAL; 1570 rgd = gfs2_blk2rgrpd(sdp, no_addr); 1571 if (!rgd) 1572 goto fail; 1573 1574 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); 1575 if (error) 1576 goto fail; 1577 1578 if (gfs2_get_block_type(rgd, no_addr) != type) 1579 error = -ESTALE; 1580 1581 gfs2_glock_dq_uninit(&rgd_gh); 1582 fail: 1583 return error; 1584 } 1585 1586 /** 1587 * gfs2_rlist_add - add a RG to a list of RGs 1588 * @ip: the inode 1589 * @rlist: the list of resource groups 1590 * @block: the block 1591 * 1592 * Figure out what RG a block belongs to and add that RG to the list 1593 * 1594 * FIXME: Don't use NOFAIL 1595 * 1596 */ 1597 1598 void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, 1599 u64 block) 1600 { 1601 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1602 struct gfs2_rgrpd *rgd; 1603 struct gfs2_rgrpd **tmp; 1604 unsigned int new_space; 1605 unsigned int x; 1606 1607 if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) 1608 return; 1609 1610 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) 1611 rgd = ip->i_rgd; 1612 else 1613 rgd = gfs2_blk2rgrpd(sdp, block); 1614 if (!rgd) { 1615 fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); 1616 return; 1617 } 1618 ip->i_rgd = rgd; 1619 1620 for (x = 0; x < rlist->rl_rgrps; x++) 1621 if (rlist->rl_rgd[x] == rgd) 1622 return; 1623 1624 if (rlist->rl_rgrps == rlist->rl_space) { 1625 new_space = rlist->rl_space + 10; 1626 1627 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), 1628 GFP_NOFS | __GFP_NOFAIL); 1629 1630 if (rlist->rl_rgd) { 1631 memcpy(tmp, rlist->rl_rgd, 1632 rlist->rl_space * sizeof(struct gfs2_rgrpd *)); 1633 kfree(rlist->rl_rgd); 1634 } 1635 1636 rlist->rl_space = new_space; 1637 rlist->rl_rgd = tmp; 1638 } 1639 1640 rlist->rl_rgd[rlist->rl_rgrps++] = rgd; 1641 } 1642 1643 /** 1644 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate 1645 * and initialize an array of glock holders for them 1646 * @rlist: the list of resource groups 1647 * @state: the lock state to acquire the RG lock in 1648 * @flags: the modifier flags for the holder structures 1649 * 1650 * FIXME: Don't use NOFAIL 1651 * 1652 */ 1653 1654 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) 1655 { 1656 unsigned int x; 1657 1658 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), 1659 GFP_NOFS | __GFP_NOFAIL); 1660 for (x = 0; x < rlist->rl_rgrps; x++) 1661 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, 1662 state, 0, 1663 &rlist->rl_ghs[x]); 1664 } 1665 1666 /** 1667 * gfs2_rlist_free - free a resource group list 1668 * @list: the list of resource groups 1669 * 1670 */ 1671 1672 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) 1673 { 1674 unsigned int x; 1675 1676 kfree(rlist->rl_rgd); 1677 1678 if (rlist->rl_ghs) { 1679 for (x = 0; x < rlist->rl_rgrps; x++) 1680 gfs2_holder_uninit(&rlist->rl_ghs[x]); 1681 kfree(rlist->rl_ghs); 1682 } 1683 } 1684 1685