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, bool exact) 331 { 332 struct rb_node *n, *next; 333 struct gfs2_rgrpd *cur; 334 335 if (gfs2_rindex_update(sdp)) 336 return NULL; 337 338 spin_lock(&sdp->sd_rindex_spin); 339 n = sdp->sd_rindex_tree.rb_node; 340 while (n) { 341 cur = rb_entry(n, struct gfs2_rgrpd, rd_node); 342 next = NULL; 343 if (blk < cur->rd_addr) 344 next = n->rb_left; 345 else if (blk >= cur->rd_data0 + cur->rd_data) 346 next = n->rb_right; 347 if (next == NULL) { 348 spin_unlock(&sdp->sd_rindex_spin); 349 if (exact) { 350 if (blk < cur->rd_addr) 351 return NULL; 352 if (blk >= cur->rd_data0 + cur->rd_data) 353 return NULL; 354 } 355 return cur; 356 } 357 n = next; 358 } 359 spin_unlock(&sdp->sd_rindex_spin); 360 361 return NULL; 362 } 363 364 /** 365 * gfs2_rgrpd_get_first - get the first Resource Group in the filesystem 366 * @sdp: The GFS2 superblock 367 * 368 * Returns: The first rgrp in the filesystem 369 */ 370 371 struct gfs2_rgrpd *gfs2_rgrpd_get_first(struct gfs2_sbd *sdp) 372 { 373 const struct rb_node *n; 374 struct gfs2_rgrpd *rgd; 375 376 spin_lock(&sdp->sd_rindex_spin); 377 n = rb_first(&sdp->sd_rindex_tree); 378 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); 379 spin_unlock(&sdp->sd_rindex_spin); 380 381 return rgd; 382 } 383 384 /** 385 * gfs2_rgrpd_get_next - get the next RG 386 * @rgd: A RG 387 * 388 * Returns: The next rgrp 389 */ 390 391 struct gfs2_rgrpd *gfs2_rgrpd_get_next(struct gfs2_rgrpd *rgd) 392 { 393 struct gfs2_sbd *sdp = rgd->rd_sbd; 394 const struct rb_node *n; 395 396 spin_lock(&sdp->sd_rindex_spin); 397 n = rb_next(&rgd->rd_node); 398 if (n == NULL) 399 n = rb_first(&sdp->sd_rindex_tree); 400 401 if (unlikely(&rgd->rd_node == n)) { 402 spin_unlock(&sdp->sd_rindex_spin); 403 return NULL; 404 } 405 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); 406 spin_unlock(&sdp->sd_rindex_spin); 407 return rgd; 408 } 409 410 void gfs2_free_clones(struct gfs2_rgrpd *rgd) 411 { 412 int x; 413 414 for (x = 0; x < rgd->rd_length; x++) { 415 struct gfs2_bitmap *bi = rgd->rd_bits + x; 416 kfree(bi->bi_clone); 417 bi->bi_clone = NULL; 418 } 419 } 420 421 void gfs2_clear_rgrpd(struct gfs2_sbd *sdp) 422 { 423 struct rb_node *n; 424 struct gfs2_rgrpd *rgd; 425 struct gfs2_glock *gl; 426 427 while ((n = rb_first(&sdp->sd_rindex_tree))) { 428 rgd = rb_entry(n, struct gfs2_rgrpd, rd_node); 429 gl = rgd->rd_gl; 430 431 rb_erase(n, &sdp->sd_rindex_tree); 432 433 if (gl) { 434 spin_lock(&gl->gl_spin); 435 gl->gl_object = NULL; 436 spin_unlock(&gl->gl_spin); 437 gfs2_glock_add_to_lru(gl); 438 gfs2_glock_put(gl); 439 } 440 441 gfs2_free_clones(rgd); 442 kfree(rgd->rd_bits); 443 kmem_cache_free(gfs2_rgrpd_cachep, rgd); 444 } 445 } 446 447 static void gfs2_rindex_print(const struct gfs2_rgrpd *rgd) 448 { 449 printk(KERN_INFO " ri_addr = %llu\n", (unsigned long long)rgd->rd_addr); 450 printk(KERN_INFO " ri_length = %u\n", rgd->rd_length); 451 printk(KERN_INFO " ri_data0 = %llu\n", (unsigned long long)rgd->rd_data0); 452 printk(KERN_INFO " ri_data = %u\n", rgd->rd_data); 453 printk(KERN_INFO " ri_bitbytes = %u\n", rgd->rd_bitbytes); 454 } 455 456 /** 457 * gfs2_compute_bitstructs - Compute the bitmap sizes 458 * @rgd: The resource group descriptor 459 * 460 * Calculates bitmap descriptors, one for each block that contains bitmap data 461 * 462 * Returns: errno 463 */ 464 465 static int compute_bitstructs(struct gfs2_rgrpd *rgd) 466 { 467 struct gfs2_sbd *sdp = rgd->rd_sbd; 468 struct gfs2_bitmap *bi; 469 u32 length = rgd->rd_length; /* # blocks in hdr & bitmap */ 470 u32 bytes_left, bytes; 471 int x; 472 473 if (!length) 474 return -EINVAL; 475 476 rgd->rd_bits = kcalloc(length, sizeof(struct gfs2_bitmap), GFP_NOFS); 477 if (!rgd->rd_bits) 478 return -ENOMEM; 479 480 bytes_left = rgd->rd_bitbytes; 481 482 for (x = 0; x < length; x++) { 483 bi = rgd->rd_bits + x; 484 485 bi->bi_flags = 0; 486 /* small rgrp; bitmap stored completely in header block */ 487 if (length == 1) { 488 bytes = bytes_left; 489 bi->bi_offset = sizeof(struct gfs2_rgrp); 490 bi->bi_start = 0; 491 bi->bi_len = bytes; 492 /* header block */ 493 } else if (x == 0) { 494 bytes = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_rgrp); 495 bi->bi_offset = sizeof(struct gfs2_rgrp); 496 bi->bi_start = 0; 497 bi->bi_len = bytes; 498 /* last block */ 499 } else if (x + 1 == length) { 500 bytes = bytes_left; 501 bi->bi_offset = sizeof(struct gfs2_meta_header); 502 bi->bi_start = rgd->rd_bitbytes - bytes_left; 503 bi->bi_len = bytes; 504 /* other blocks */ 505 } else { 506 bytes = sdp->sd_sb.sb_bsize - 507 sizeof(struct gfs2_meta_header); 508 bi->bi_offset = sizeof(struct gfs2_meta_header); 509 bi->bi_start = rgd->rd_bitbytes - bytes_left; 510 bi->bi_len = bytes; 511 } 512 513 bytes_left -= bytes; 514 } 515 516 if (bytes_left) { 517 gfs2_consist_rgrpd(rgd); 518 return -EIO; 519 } 520 bi = rgd->rd_bits + (length - 1); 521 if ((bi->bi_start + bi->bi_len) * GFS2_NBBY != rgd->rd_data) { 522 if (gfs2_consist_rgrpd(rgd)) { 523 gfs2_rindex_print(rgd); 524 fs_err(sdp, "start=%u len=%u offset=%u\n", 525 bi->bi_start, bi->bi_len, bi->bi_offset); 526 } 527 return -EIO; 528 } 529 530 return 0; 531 } 532 533 /** 534 * gfs2_ri_total - Total up the file system space, according to the rindex. 535 * 536 */ 537 u64 gfs2_ri_total(struct gfs2_sbd *sdp) 538 { 539 u64 total_data = 0; 540 struct inode *inode = sdp->sd_rindex; 541 struct gfs2_inode *ip = GFS2_I(inode); 542 char buf[sizeof(struct gfs2_rindex)]; 543 struct file_ra_state ra_state; 544 int error, rgrps; 545 546 file_ra_state_init(&ra_state, inode->i_mapping); 547 for (rgrps = 0;; rgrps++) { 548 loff_t pos = rgrps * sizeof(struct gfs2_rindex); 549 550 if (pos + sizeof(struct gfs2_rindex) > i_size_read(inode)) 551 break; 552 error = gfs2_internal_read(ip, &ra_state, buf, &pos, 553 sizeof(struct gfs2_rindex)); 554 if (error != sizeof(struct gfs2_rindex)) 555 break; 556 total_data += be32_to_cpu(((struct gfs2_rindex *)buf)->ri_data); 557 } 558 return total_data; 559 } 560 561 static int rgd_insert(struct gfs2_rgrpd *rgd) 562 { 563 struct gfs2_sbd *sdp = rgd->rd_sbd; 564 struct rb_node **newn = &sdp->sd_rindex_tree.rb_node, *parent = NULL; 565 566 /* Figure out where to put new node */ 567 while (*newn) { 568 struct gfs2_rgrpd *cur = rb_entry(*newn, struct gfs2_rgrpd, 569 rd_node); 570 571 parent = *newn; 572 if (rgd->rd_addr < cur->rd_addr) 573 newn = &((*newn)->rb_left); 574 else if (rgd->rd_addr > cur->rd_addr) 575 newn = &((*newn)->rb_right); 576 else 577 return -EEXIST; 578 } 579 580 rb_link_node(&rgd->rd_node, parent, newn); 581 rb_insert_color(&rgd->rd_node, &sdp->sd_rindex_tree); 582 sdp->sd_rgrps++; 583 return 0; 584 } 585 586 /** 587 * read_rindex_entry - Pull in a new resource index entry from the disk 588 * @gl: The glock covering the rindex inode 589 * 590 * Returns: 0 on success, > 0 on EOF, error code otherwise 591 */ 592 593 static int read_rindex_entry(struct gfs2_inode *ip, 594 struct file_ra_state *ra_state) 595 { 596 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 597 loff_t pos = sdp->sd_rgrps * sizeof(struct gfs2_rindex); 598 struct gfs2_rindex buf; 599 int error; 600 struct gfs2_rgrpd *rgd; 601 602 if (pos >= i_size_read(&ip->i_inode)) 603 return 1; 604 605 error = gfs2_internal_read(ip, ra_state, (char *)&buf, &pos, 606 sizeof(struct gfs2_rindex)); 607 608 if (error != sizeof(struct gfs2_rindex)) 609 return (error == 0) ? 1 : error; 610 611 rgd = kmem_cache_zalloc(gfs2_rgrpd_cachep, GFP_NOFS); 612 error = -ENOMEM; 613 if (!rgd) 614 return error; 615 616 rgd->rd_sbd = sdp; 617 rgd->rd_addr = be64_to_cpu(buf.ri_addr); 618 rgd->rd_length = be32_to_cpu(buf.ri_length); 619 rgd->rd_data0 = be64_to_cpu(buf.ri_data0); 620 rgd->rd_data = be32_to_cpu(buf.ri_data); 621 rgd->rd_bitbytes = be32_to_cpu(buf.ri_bitbytes); 622 623 error = compute_bitstructs(rgd); 624 if (error) 625 goto fail; 626 627 error = gfs2_glock_get(sdp, rgd->rd_addr, 628 &gfs2_rgrp_glops, CREATE, &rgd->rd_gl); 629 if (error) 630 goto fail; 631 632 rgd->rd_gl->gl_object = rgd; 633 rgd->rd_flags &= ~GFS2_RDF_UPTODATE; 634 if (rgd->rd_data > sdp->sd_max_rg_data) 635 sdp->sd_max_rg_data = rgd->rd_data; 636 spin_lock(&sdp->sd_rindex_spin); 637 error = rgd_insert(rgd); 638 spin_unlock(&sdp->sd_rindex_spin); 639 if (!error) 640 return 0; 641 642 error = 0; /* someone else read in the rgrp; free it and ignore it */ 643 644 fail: 645 kfree(rgd->rd_bits); 646 kmem_cache_free(gfs2_rgrpd_cachep, rgd); 647 return error; 648 } 649 650 /** 651 * gfs2_ri_update - Pull in a new resource index from the disk 652 * @ip: pointer to the rindex inode 653 * 654 * Returns: 0 on successful update, error code otherwise 655 */ 656 657 static int gfs2_ri_update(struct gfs2_inode *ip) 658 { 659 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 660 struct inode *inode = &ip->i_inode; 661 struct file_ra_state ra_state; 662 int error; 663 664 file_ra_state_init(&ra_state, inode->i_mapping); 665 do { 666 error = read_rindex_entry(ip, &ra_state); 667 } while (error == 0); 668 669 if (error < 0) 670 return error; 671 672 sdp->sd_rindex_uptodate = 1; 673 return 0; 674 } 675 676 /** 677 * gfs2_rindex_update - Update the rindex if required 678 * @sdp: The GFS2 superblock 679 * 680 * We grab a lock on the rindex inode to make sure that it doesn't 681 * change whilst we are performing an operation. We keep this lock 682 * for quite long periods of time compared to other locks. This 683 * doesn't matter, since it is shared and it is very, very rarely 684 * accessed in the exclusive mode (i.e. only when expanding the filesystem). 685 * 686 * This makes sure that we're using the latest copy of the resource index 687 * special file, which might have been updated if someone expanded the 688 * filesystem (via gfs2_grow utility), which adds new resource groups. 689 * 690 * Returns: 0 on succeess, error code otherwise 691 */ 692 693 int gfs2_rindex_update(struct gfs2_sbd *sdp) 694 { 695 struct gfs2_inode *ip = GFS2_I(sdp->sd_rindex); 696 struct gfs2_glock *gl = ip->i_gl; 697 struct gfs2_holder ri_gh; 698 int error = 0; 699 int unlock_required = 0; 700 701 /* Read new copy from disk if we don't have the latest */ 702 if (!sdp->sd_rindex_uptodate) { 703 if (!gfs2_glock_is_locked_by_me(gl)) { 704 error = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &ri_gh); 705 if (error) 706 return error; 707 unlock_required = 1; 708 } 709 if (!sdp->sd_rindex_uptodate) 710 error = gfs2_ri_update(ip); 711 if (unlock_required) 712 gfs2_glock_dq_uninit(&ri_gh); 713 } 714 715 return error; 716 } 717 718 static void gfs2_rgrp_in(struct gfs2_rgrpd *rgd, const void *buf) 719 { 720 const struct gfs2_rgrp *str = buf; 721 u32 rg_flags; 722 723 rg_flags = be32_to_cpu(str->rg_flags); 724 rg_flags &= ~GFS2_RDF_MASK; 725 rgd->rd_flags &= GFS2_RDF_MASK; 726 rgd->rd_flags |= rg_flags; 727 rgd->rd_free = be32_to_cpu(str->rg_free); 728 rgd->rd_dinodes = be32_to_cpu(str->rg_dinodes); 729 rgd->rd_igeneration = be64_to_cpu(str->rg_igeneration); 730 } 731 732 static void gfs2_rgrp_out(struct gfs2_rgrpd *rgd, void *buf) 733 { 734 struct gfs2_rgrp *str = buf; 735 736 str->rg_flags = cpu_to_be32(rgd->rd_flags & ~GFS2_RDF_MASK); 737 str->rg_free = cpu_to_be32(rgd->rd_free); 738 str->rg_dinodes = cpu_to_be32(rgd->rd_dinodes); 739 str->__pad = cpu_to_be32(0); 740 str->rg_igeneration = cpu_to_be64(rgd->rd_igeneration); 741 memset(&str->rg_reserved, 0, sizeof(str->rg_reserved)); 742 } 743 744 /** 745 * gfs2_rgrp_go_lock - Read in a RG's header and bitmaps 746 * @rgd: the struct gfs2_rgrpd describing the RG to read in 747 * 748 * Read in all of a Resource Group's header and bitmap blocks. 749 * Caller must eventually call gfs2_rgrp_relse() to free the bitmaps. 750 * 751 * Returns: errno 752 */ 753 754 int gfs2_rgrp_go_lock(struct gfs2_holder *gh) 755 { 756 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; 757 struct gfs2_sbd *sdp = rgd->rd_sbd; 758 struct gfs2_glock *gl = rgd->rd_gl; 759 unsigned int length = rgd->rd_length; 760 struct gfs2_bitmap *bi; 761 unsigned int x, y; 762 int error; 763 764 for (x = 0; x < length; x++) { 765 bi = rgd->rd_bits + x; 766 error = gfs2_meta_read(gl, rgd->rd_addr + x, 0, &bi->bi_bh); 767 if (error) 768 goto fail; 769 } 770 771 for (y = length; y--;) { 772 bi = rgd->rd_bits + y; 773 error = gfs2_meta_wait(sdp, bi->bi_bh); 774 if (error) 775 goto fail; 776 if (gfs2_metatype_check(sdp, bi->bi_bh, y ? GFS2_METATYPE_RB : 777 GFS2_METATYPE_RG)) { 778 error = -EIO; 779 goto fail; 780 } 781 } 782 783 if (!(rgd->rd_flags & GFS2_RDF_UPTODATE)) { 784 for (x = 0; x < length; x++) 785 clear_bit(GBF_FULL, &rgd->rd_bits[x].bi_flags); 786 gfs2_rgrp_in(rgd, (rgd->rd_bits[0].bi_bh)->b_data); 787 rgd->rd_flags |= (GFS2_RDF_UPTODATE | GFS2_RDF_CHECK); 788 rgd->rd_free_clone = rgd->rd_free; 789 } 790 791 return 0; 792 793 fail: 794 while (x--) { 795 bi = rgd->rd_bits + x; 796 brelse(bi->bi_bh); 797 bi->bi_bh = NULL; 798 gfs2_assert_warn(sdp, !bi->bi_clone); 799 } 800 801 return error; 802 } 803 804 /** 805 * gfs2_rgrp_go_unlock - Release RG bitmaps read in with gfs2_rgrp_bh_get() 806 * @rgd: the struct gfs2_rgrpd describing the RG to read in 807 * 808 */ 809 810 void gfs2_rgrp_go_unlock(struct gfs2_holder *gh) 811 { 812 struct gfs2_rgrpd *rgd = gh->gh_gl->gl_object; 813 int x, length = rgd->rd_length; 814 815 for (x = 0; x < length; x++) { 816 struct gfs2_bitmap *bi = rgd->rd_bits + x; 817 brelse(bi->bi_bh); 818 bi->bi_bh = NULL; 819 } 820 821 } 822 823 int gfs2_rgrp_send_discards(struct gfs2_sbd *sdp, u64 offset, 824 struct buffer_head *bh, 825 const struct gfs2_bitmap *bi, unsigned minlen, u64 *ptrimmed) 826 { 827 struct super_block *sb = sdp->sd_vfs; 828 struct block_device *bdev = sb->s_bdev; 829 const unsigned int sects_per_blk = sdp->sd_sb.sb_bsize / 830 bdev_logical_block_size(sb->s_bdev); 831 u64 blk; 832 sector_t start = 0; 833 sector_t nr_sects = 0; 834 int rv; 835 unsigned int x; 836 u32 trimmed = 0; 837 u8 diff; 838 839 for (x = 0; x < bi->bi_len; x++) { 840 const u8 *clone = bi->bi_clone ? bi->bi_clone : bi->bi_bh->b_data; 841 clone += bi->bi_offset; 842 clone += x; 843 if (bh) { 844 const u8 *orig = bh->b_data + bi->bi_offset + x; 845 diff = ~(*orig | (*orig >> 1)) & (*clone | (*clone >> 1)); 846 } else { 847 diff = ~(*clone | (*clone >> 1)); 848 } 849 diff &= 0x55; 850 if (diff == 0) 851 continue; 852 blk = offset + ((bi->bi_start + x) * GFS2_NBBY); 853 blk *= sects_per_blk; /* convert to sectors */ 854 while(diff) { 855 if (diff & 1) { 856 if (nr_sects == 0) 857 goto start_new_extent; 858 if ((start + nr_sects) != blk) { 859 if (nr_sects >= minlen) { 860 rv = blkdev_issue_discard(bdev, 861 start, nr_sects, 862 GFP_NOFS, 0); 863 if (rv) 864 goto fail; 865 trimmed += nr_sects; 866 } 867 nr_sects = 0; 868 start_new_extent: 869 start = blk; 870 } 871 nr_sects += sects_per_blk; 872 } 873 diff >>= 2; 874 blk += sects_per_blk; 875 } 876 } 877 if (nr_sects >= minlen) { 878 rv = blkdev_issue_discard(bdev, start, nr_sects, GFP_NOFS, 0); 879 if (rv) 880 goto fail; 881 trimmed += nr_sects; 882 } 883 if (ptrimmed) 884 *ptrimmed = trimmed; 885 return 0; 886 887 fail: 888 if (sdp->sd_args.ar_discard) 889 fs_warn(sdp, "error %d on discard request, turning discards off for this filesystem", rv); 890 sdp->sd_args.ar_discard = 0; 891 return -EIO; 892 } 893 894 /** 895 * gfs2_fitrim - Generate discard requests for unused bits of the filesystem 896 * @filp: Any file on the filesystem 897 * @argp: Pointer to the arguments (also used to pass result) 898 * 899 * Returns: 0 on success, otherwise error code 900 */ 901 902 int gfs2_fitrim(struct file *filp, void __user *argp) 903 { 904 struct inode *inode = filp->f_dentry->d_inode; 905 struct gfs2_sbd *sdp = GFS2_SB(inode); 906 struct request_queue *q = bdev_get_queue(sdp->sd_vfs->s_bdev); 907 struct buffer_head *bh; 908 struct gfs2_rgrpd *rgd; 909 struct gfs2_rgrpd *rgd_end; 910 struct gfs2_holder gh; 911 struct fstrim_range r; 912 int ret = 0; 913 u64 amt; 914 u64 trimmed = 0; 915 unsigned int x; 916 917 if (!capable(CAP_SYS_ADMIN)) 918 return -EPERM; 919 920 if (!blk_queue_discard(q)) 921 return -EOPNOTSUPP; 922 923 if (argp == NULL) { 924 r.start = 0; 925 r.len = ULLONG_MAX; 926 r.minlen = 0; 927 } else if (copy_from_user(&r, argp, sizeof(r))) 928 return -EFAULT; 929 930 rgd = gfs2_blk2rgrpd(sdp, r.start, 0); 931 rgd_end = gfs2_blk2rgrpd(sdp, r.start + r.len, 0); 932 933 while (1) { 934 935 ret = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh); 936 if (ret) 937 goto out; 938 939 if (!(rgd->rd_flags & GFS2_RGF_TRIMMED)) { 940 /* Trim each bitmap in the rgrp */ 941 for (x = 0; x < rgd->rd_length; x++) { 942 struct gfs2_bitmap *bi = rgd->rd_bits + x; 943 ret = gfs2_rgrp_send_discards(sdp, rgd->rd_data0, NULL, bi, r.minlen, &amt); 944 if (ret) { 945 gfs2_glock_dq_uninit(&gh); 946 goto out; 947 } 948 trimmed += amt; 949 } 950 951 /* Mark rgrp as having been trimmed */ 952 ret = gfs2_trans_begin(sdp, RES_RG_HDR, 0); 953 if (ret == 0) { 954 bh = rgd->rd_bits[0].bi_bh; 955 rgd->rd_flags |= GFS2_RGF_TRIMMED; 956 gfs2_trans_add_bh(rgd->rd_gl, bh, 1); 957 gfs2_rgrp_out(rgd, bh->b_data); 958 gfs2_trans_end(sdp); 959 } 960 } 961 gfs2_glock_dq_uninit(&gh); 962 963 if (rgd == rgd_end) 964 break; 965 966 rgd = gfs2_rgrpd_get_next(rgd); 967 } 968 969 out: 970 r.len = trimmed << 9; 971 if (argp && copy_to_user(argp, &r, sizeof(r))) 972 return -EFAULT; 973 974 return ret; 975 } 976 977 /** 978 * gfs2_qadata_get - get the struct gfs2_qadata structure for an inode 979 * @ip: the incore GFS2 inode structure 980 * 981 * Returns: the struct gfs2_qadata 982 */ 983 984 struct gfs2_qadata *gfs2_qadata_get(struct gfs2_inode *ip) 985 { 986 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 987 int error; 988 BUG_ON(ip->i_qadata != NULL); 989 ip->i_qadata = kzalloc(sizeof(struct gfs2_qadata), GFP_NOFS); 990 error = gfs2_rindex_update(sdp); 991 if (error) 992 fs_warn(sdp, "rindex update returns %d\n", error); 993 return ip->i_qadata; 994 } 995 996 /** 997 * gfs2_blkrsv_get - get the struct gfs2_blkreserv structure for an inode 998 * @ip: the incore GFS2 inode structure 999 * 1000 * Returns: the struct gfs2_qadata 1001 */ 1002 1003 static struct gfs2_blkreserv *gfs2_blkrsv_get(struct gfs2_inode *ip) 1004 { 1005 BUG_ON(ip->i_res != NULL); 1006 ip->i_res = kzalloc(sizeof(struct gfs2_blkreserv), GFP_NOFS); 1007 return ip->i_res; 1008 } 1009 1010 /** 1011 * try_rgrp_fit - See if a given reservation will fit in a given RG 1012 * @rgd: the RG data 1013 * @ip: the inode 1014 * 1015 * If there's room for the requested blocks to be allocated from the RG: 1016 * 1017 * Returns: 1 on success (it fits), 0 on failure (it doesn't fit) 1018 */ 1019 1020 static int try_rgrp_fit(const struct gfs2_rgrpd *rgd, const struct gfs2_inode *ip) 1021 { 1022 const struct gfs2_blkreserv *rs = ip->i_res; 1023 1024 if (rgd->rd_flags & (GFS2_RGF_NOALLOC | GFS2_RDF_ERROR)) 1025 return 0; 1026 if (rgd->rd_free_clone >= rs->rs_requested) 1027 return 1; 1028 return 0; 1029 } 1030 1031 static inline u32 gfs2_bi2rgd_blk(struct gfs2_bitmap *bi, u32 blk) 1032 { 1033 return (bi->bi_start * GFS2_NBBY) + blk; 1034 } 1035 1036 /** 1037 * try_rgrp_unlink - Look for any unlinked, allocated, but unused inodes 1038 * @rgd: The rgrp 1039 * 1040 * Returns: 0 if no error 1041 * The inode, if one has been found, in inode. 1042 */ 1043 1044 static void try_rgrp_unlink(struct gfs2_rgrpd *rgd, u64 *last_unlinked, u64 skip) 1045 { 1046 u32 goal = 0, block; 1047 u64 no_addr; 1048 struct gfs2_sbd *sdp = rgd->rd_sbd; 1049 struct gfs2_glock *gl; 1050 struct gfs2_inode *ip; 1051 int error; 1052 int found = 0; 1053 struct gfs2_bitmap *bi; 1054 1055 while (goal < rgd->rd_data) { 1056 down_write(&sdp->sd_log_flush_lock); 1057 block = rgblk_search(rgd, goal, GFS2_BLKST_UNLINKED, &bi); 1058 up_write(&sdp->sd_log_flush_lock); 1059 if (block == BFITNOENT) 1060 break; 1061 1062 block = gfs2_bi2rgd_blk(bi, block); 1063 /* rgblk_search can return a block < goal, so we need to 1064 keep it marching forward. */ 1065 no_addr = block + rgd->rd_data0; 1066 goal = max(block + 1, goal + 1); 1067 if (*last_unlinked != NO_BLOCK && no_addr <= *last_unlinked) 1068 continue; 1069 if (no_addr == skip) 1070 continue; 1071 *last_unlinked = no_addr; 1072 1073 error = gfs2_glock_get(sdp, no_addr, &gfs2_inode_glops, CREATE, &gl); 1074 if (error) 1075 continue; 1076 1077 /* If the inode is already in cache, we can ignore it here 1078 * because the existing inode disposal code will deal with 1079 * it when all refs have gone away. Accessing gl_object like 1080 * this is not safe in general. Here it is ok because we do 1081 * not dereference the pointer, and we only need an approx 1082 * answer to whether it is NULL or not. 1083 */ 1084 ip = gl->gl_object; 1085 1086 if (ip || queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0) 1087 gfs2_glock_put(gl); 1088 else 1089 found++; 1090 1091 /* Limit reclaim to sensible number of tasks */ 1092 if (found > NR_CPUS) 1093 return; 1094 } 1095 1096 rgd->rd_flags &= ~GFS2_RDF_CHECK; 1097 return; 1098 } 1099 1100 /** 1101 * get_local_rgrp - Choose and lock a rgrp for allocation 1102 * @ip: the inode to reserve space for 1103 * @rgp: the chosen and locked rgrp 1104 * 1105 * Try to acquire rgrp in way which avoids contending with others. 1106 * 1107 * Returns: errno 1108 */ 1109 1110 static int get_local_rgrp(struct gfs2_inode *ip, u64 *last_unlinked) 1111 { 1112 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1113 struct gfs2_rgrpd *rgd, *begin = NULL; 1114 struct gfs2_blkreserv *rs = ip->i_res; 1115 int error, rg_locked, flags = LM_FLAG_TRY; 1116 int loops = 0; 1117 1118 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, ip->i_goal)) 1119 rgd = begin = ip->i_rgd; 1120 else 1121 rgd = begin = gfs2_blk2rgrpd(sdp, ip->i_goal, 1); 1122 1123 if (rgd == NULL) 1124 return -EBADSLT; 1125 1126 while (loops < 3) { 1127 rg_locked = 0; 1128 1129 if (gfs2_glock_is_locked_by_me(rgd->rd_gl)) { 1130 rg_locked = 1; 1131 error = 0; 1132 } else { 1133 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 1134 flags, &rs->rs_rgd_gh); 1135 } 1136 switch (error) { 1137 case 0: 1138 if (try_rgrp_fit(rgd, ip)) { 1139 ip->i_rgd = rgd; 1140 return 0; 1141 } 1142 if (rgd->rd_flags & GFS2_RDF_CHECK) 1143 try_rgrp_unlink(rgd, last_unlinked, ip->i_no_addr); 1144 if (!rg_locked) 1145 gfs2_glock_dq_uninit(&rs->rs_rgd_gh); 1146 /* fall through */ 1147 case GLR_TRYFAILED: 1148 rgd = gfs2_rgrpd_get_next(rgd); 1149 if (rgd == begin) { 1150 flags = 0; 1151 loops++; 1152 } 1153 break; 1154 default: 1155 return error; 1156 } 1157 } 1158 1159 return -ENOSPC; 1160 } 1161 1162 static void gfs2_blkrsv_put(struct gfs2_inode *ip) 1163 { 1164 BUG_ON(ip->i_res == NULL); 1165 kfree(ip->i_res); 1166 ip->i_res = NULL; 1167 } 1168 1169 /** 1170 * gfs2_inplace_reserve - Reserve space in the filesystem 1171 * @ip: the inode to reserve space for 1172 * 1173 * Returns: errno 1174 */ 1175 1176 int gfs2_inplace_reserve(struct gfs2_inode *ip, u32 requested) 1177 { 1178 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1179 struct gfs2_blkreserv *rs; 1180 int error = 0; 1181 u64 last_unlinked = NO_BLOCK; 1182 int tries = 0; 1183 1184 rs = gfs2_blkrsv_get(ip); 1185 if (!rs) 1186 return -ENOMEM; 1187 1188 rs->rs_requested = requested; 1189 if (gfs2_assert_warn(sdp, requested)) { 1190 error = -EINVAL; 1191 goto out; 1192 } 1193 1194 do { 1195 error = get_local_rgrp(ip, &last_unlinked); 1196 if (error != -ENOSPC) 1197 break; 1198 /* Check that fs hasn't grown if writing to rindex */ 1199 if (ip == GFS2_I(sdp->sd_rindex) && !sdp->sd_rindex_uptodate) { 1200 error = gfs2_ri_update(ip); 1201 if (error) 1202 break; 1203 continue; 1204 } 1205 /* Flushing the log may release space */ 1206 gfs2_log_flush(sdp, NULL); 1207 } while (tries++ < 3); 1208 1209 out: 1210 if (error) 1211 gfs2_blkrsv_put(ip); 1212 return error; 1213 } 1214 1215 /** 1216 * gfs2_inplace_release - release an inplace reservation 1217 * @ip: the inode the reservation was taken out on 1218 * 1219 * Release a reservation made by gfs2_inplace_reserve(). 1220 */ 1221 1222 void gfs2_inplace_release(struct gfs2_inode *ip) 1223 { 1224 struct gfs2_blkreserv *rs = ip->i_res; 1225 1226 if (rs->rs_rgd_gh.gh_gl) 1227 gfs2_glock_dq_uninit(&rs->rs_rgd_gh); 1228 gfs2_blkrsv_put(ip); 1229 } 1230 1231 /** 1232 * gfs2_get_block_type - Check a block in a RG is of given type 1233 * @rgd: the resource group holding the block 1234 * @block: the block number 1235 * 1236 * Returns: The block type (GFS2_BLKST_*) 1237 */ 1238 1239 static unsigned char gfs2_get_block_type(struct gfs2_rgrpd *rgd, u64 block) 1240 { 1241 struct gfs2_bitmap *bi = NULL; 1242 u32 length, rgrp_block, buf_block; 1243 unsigned int buf; 1244 unsigned char type; 1245 1246 length = rgd->rd_length; 1247 rgrp_block = block - rgd->rd_data0; 1248 1249 for (buf = 0; buf < length; buf++) { 1250 bi = rgd->rd_bits + buf; 1251 if (rgrp_block < (bi->bi_start + bi->bi_len) * GFS2_NBBY) 1252 break; 1253 } 1254 1255 gfs2_assert(rgd->rd_sbd, buf < length); 1256 buf_block = rgrp_block - bi->bi_start * GFS2_NBBY; 1257 1258 type = gfs2_testbit(rgd, bi->bi_bh->b_data + bi->bi_offset, 1259 bi->bi_len, buf_block); 1260 1261 return type; 1262 } 1263 1264 /** 1265 * rgblk_search - find a block in @state 1266 * @rgd: the resource group descriptor 1267 * @goal: the goal block within the RG (start here to search for avail block) 1268 * @state: GFS2_BLKST_XXX the before-allocation state to find 1269 * @dinode: TRUE if the first block we allocate is for a dinode 1270 * @rbi: address of the pointer to the bitmap containing the block found 1271 * 1272 * Walk rgrp's bitmap to find bits that represent a block in @state. 1273 * 1274 * This function never fails, because we wouldn't call it unless we 1275 * know (from reservation results, etc.) that a block is available. 1276 * 1277 * Scope of @goal is just within rgrp, not the whole filesystem. 1278 * Scope of @returned block is just within bitmap, not the whole filesystem. 1279 * 1280 * Returns: the block number found relative to the bitmap rbi 1281 */ 1282 1283 static u32 rgblk_search(struct gfs2_rgrpd *rgd, u32 goal, 1284 unsigned char state, 1285 struct gfs2_bitmap **rbi) 1286 { 1287 struct gfs2_bitmap *bi = NULL; 1288 const u32 length = rgd->rd_length; 1289 u32 blk = BFITNOENT; 1290 unsigned int buf, x; 1291 const u8 *buffer = NULL; 1292 1293 *rbi = NULL; 1294 /* Find bitmap block that contains bits for goal block */ 1295 for (buf = 0; buf < length; buf++) { 1296 bi = rgd->rd_bits + buf; 1297 /* Convert scope of "goal" from rgrp-wide to within found bit block */ 1298 if (goal < (bi->bi_start + bi->bi_len) * GFS2_NBBY) { 1299 goal -= bi->bi_start * GFS2_NBBY; 1300 goto do_search; 1301 } 1302 } 1303 buf = 0; 1304 goal = 0; 1305 1306 do_search: 1307 /* Search (up to entire) bitmap in this rgrp for allocatable block. 1308 "x <= length", instead of "x < length", because we typically start 1309 the search in the middle of a bit block, but if we can't find an 1310 allocatable block anywhere else, we want to be able wrap around and 1311 search in the first part of our first-searched bit block. */ 1312 for (x = 0; x <= length; x++) { 1313 bi = rgd->rd_bits + buf; 1314 1315 if (test_bit(GBF_FULL, &bi->bi_flags) && 1316 (state == GFS2_BLKST_FREE)) 1317 goto skip; 1318 1319 /* The GFS2_BLKST_UNLINKED state doesn't apply to the clone 1320 bitmaps, so we must search the originals for that. */ 1321 buffer = bi->bi_bh->b_data + bi->bi_offset; 1322 WARN_ON(!buffer_uptodate(bi->bi_bh)); 1323 if (state != GFS2_BLKST_UNLINKED && bi->bi_clone) 1324 buffer = bi->bi_clone + bi->bi_offset; 1325 1326 blk = gfs2_bitfit(buffer, bi->bi_len, goal, state); 1327 if (blk != BFITNOENT) 1328 break; 1329 1330 if ((goal == 0) && (state == GFS2_BLKST_FREE)) 1331 set_bit(GBF_FULL, &bi->bi_flags); 1332 1333 /* Try next bitmap block (wrap back to rgrp header if at end) */ 1334 skip: 1335 buf++; 1336 buf %= length; 1337 goal = 0; 1338 } 1339 1340 if (blk != BFITNOENT) 1341 *rbi = bi; 1342 1343 return blk; 1344 } 1345 1346 /** 1347 * gfs2_alloc_extent - allocate an extent from a given bitmap 1348 * @rgd: the resource group descriptor 1349 * @bi: the bitmap within the rgrp 1350 * @blk: the block within the bitmap 1351 * @dinode: TRUE if the first block we allocate is for a dinode 1352 * @n: The extent length 1353 * 1354 * Add the found bitmap buffer to the transaction. 1355 * Set the found bits to @new_state to change block's allocation state. 1356 * Returns: starting block number of the extent (fs scope) 1357 */ 1358 static u64 gfs2_alloc_extent(struct gfs2_rgrpd *rgd, struct gfs2_bitmap *bi, 1359 u32 blk, bool dinode, unsigned int *n) 1360 { 1361 const unsigned int elen = *n; 1362 u32 goal; 1363 const u8 *buffer = NULL; 1364 1365 *n = 0; 1366 buffer = bi->bi_bh->b_data + bi->bi_offset; 1367 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); 1368 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, 1369 bi, blk, dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); 1370 (*n)++; 1371 goal = blk; 1372 while (*n < elen) { 1373 goal++; 1374 if (goal >= (bi->bi_len * GFS2_NBBY)) 1375 break; 1376 if (gfs2_testbit(rgd, buffer, bi->bi_len, goal) != 1377 GFS2_BLKST_FREE) 1378 break; 1379 gfs2_setbit(rgd, bi->bi_bh->b_data, bi->bi_clone, bi->bi_offset, 1380 bi, goal, GFS2_BLKST_USED); 1381 (*n)++; 1382 } 1383 blk = gfs2_bi2rgd_blk(bi, blk); 1384 rgd->rd_last_alloc = blk + *n - 1; 1385 return rgd->rd_data0 + blk; 1386 } 1387 1388 /** 1389 * rgblk_free - Change alloc state of given block(s) 1390 * @sdp: the filesystem 1391 * @bstart: the start of a run of blocks to free 1392 * @blen: the length of the block run (all must lie within ONE RG!) 1393 * @new_state: GFS2_BLKST_XXX the after-allocation block state 1394 * 1395 * Returns: Resource group containing the block(s) 1396 */ 1397 1398 static struct gfs2_rgrpd *rgblk_free(struct gfs2_sbd *sdp, u64 bstart, 1399 u32 blen, unsigned char new_state) 1400 { 1401 struct gfs2_rgrpd *rgd; 1402 struct gfs2_bitmap *bi = NULL; 1403 u32 length, rgrp_blk, buf_blk; 1404 unsigned int buf; 1405 1406 rgd = gfs2_blk2rgrpd(sdp, bstart, 1); 1407 if (!rgd) { 1408 if (gfs2_consist(sdp)) 1409 fs_err(sdp, "block = %llu\n", (unsigned long long)bstart); 1410 return NULL; 1411 } 1412 1413 length = rgd->rd_length; 1414 1415 rgrp_blk = bstart - rgd->rd_data0; 1416 1417 while (blen--) { 1418 for (buf = 0; buf < length; buf++) { 1419 bi = rgd->rd_bits + buf; 1420 if (rgrp_blk < (bi->bi_start + bi->bi_len) * GFS2_NBBY) 1421 break; 1422 } 1423 1424 gfs2_assert(rgd->rd_sbd, buf < length); 1425 1426 buf_blk = rgrp_blk - bi->bi_start * GFS2_NBBY; 1427 rgrp_blk++; 1428 1429 if (!bi->bi_clone) { 1430 bi->bi_clone = kmalloc(bi->bi_bh->b_size, 1431 GFP_NOFS | __GFP_NOFAIL); 1432 memcpy(bi->bi_clone + bi->bi_offset, 1433 bi->bi_bh->b_data + bi->bi_offset, 1434 bi->bi_len); 1435 } 1436 gfs2_trans_add_bh(rgd->rd_gl, bi->bi_bh, 1); 1437 gfs2_setbit(rgd, bi->bi_bh->b_data, NULL, bi->bi_offset, 1438 bi, buf_blk, new_state); 1439 } 1440 1441 return rgd; 1442 } 1443 1444 /** 1445 * gfs2_rgrp_dump - print out an rgrp 1446 * @seq: The iterator 1447 * @gl: The glock in question 1448 * 1449 */ 1450 1451 int gfs2_rgrp_dump(struct seq_file *seq, const struct gfs2_glock *gl) 1452 { 1453 const struct gfs2_rgrpd *rgd = gl->gl_object; 1454 if (rgd == NULL) 1455 return 0; 1456 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n", 1457 (unsigned long long)rgd->rd_addr, rgd->rd_flags, 1458 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes); 1459 return 0; 1460 } 1461 1462 static void gfs2_rgrp_error(struct gfs2_rgrpd *rgd) 1463 { 1464 struct gfs2_sbd *sdp = rgd->rd_sbd; 1465 fs_warn(sdp, "rgrp %llu has an error, marking it readonly until umount\n", 1466 (unsigned long long)rgd->rd_addr); 1467 fs_warn(sdp, "umount on all nodes and run fsck.gfs2 to fix the error\n"); 1468 gfs2_rgrp_dump(NULL, rgd->rd_gl); 1469 rgd->rd_flags |= GFS2_RDF_ERROR; 1470 } 1471 1472 /** 1473 * gfs2_alloc_blocks - Allocate one or more blocks of data and/or a dinode 1474 * @ip: the inode to allocate the block for 1475 * @bn: Used to return the starting block number 1476 * @ndata: requested number of blocks/extent length (value/result) 1477 * @dinode: 1 if we're allocating a dinode block, else 0 1478 * @generation: the generation number of the inode 1479 * 1480 * Returns: 0 or error 1481 */ 1482 1483 int gfs2_alloc_blocks(struct gfs2_inode *ip, u64 *bn, unsigned int *nblocks, 1484 bool dinode, u64 *generation) 1485 { 1486 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1487 struct buffer_head *dibh; 1488 struct gfs2_rgrpd *rgd; 1489 unsigned int ndata; 1490 u32 goal, blk; /* block, within the rgrp scope */ 1491 u64 block; /* block, within the file system scope */ 1492 int error; 1493 struct gfs2_bitmap *bi; 1494 1495 /* Only happens if there is a bug in gfs2, return something distinctive 1496 * to ensure that it is noticed. 1497 */ 1498 if (ip->i_res == NULL) 1499 return -ECANCELED; 1500 1501 rgd = ip->i_rgd; 1502 1503 if (!dinode && rgrp_contains_block(rgd, ip->i_goal)) 1504 goal = ip->i_goal - rgd->rd_data0; 1505 else 1506 goal = rgd->rd_last_alloc; 1507 1508 blk = rgblk_search(rgd, goal, GFS2_BLKST_FREE, &bi); 1509 1510 /* Since all blocks are reserved in advance, this shouldn't happen */ 1511 if (blk == BFITNOENT) 1512 goto rgrp_error; 1513 1514 block = gfs2_alloc_extent(rgd, bi, blk, dinode, nblocks); 1515 ndata = *nblocks; 1516 if (dinode) 1517 ndata--; 1518 1519 if (!dinode) { 1520 ip->i_goal = block + ndata - 1; 1521 error = gfs2_meta_inode_buffer(ip, &dibh); 1522 if (error == 0) { 1523 struct gfs2_dinode *di = 1524 (struct gfs2_dinode *)dibh->b_data; 1525 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1526 di->di_goal_meta = di->di_goal_data = 1527 cpu_to_be64(ip->i_goal); 1528 brelse(dibh); 1529 } 1530 } 1531 if (rgd->rd_free < *nblocks) 1532 goto rgrp_error; 1533 1534 rgd->rd_free -= *nblocks; 1535 if (dinode) { 1536 rgd->rd_dinodes++; 1537 *generation = rgd->rd_igeneration++; 1538 if (*generation == 0) 1539 *generation = rgd->rd_igeneration++; 1540 } 1541 1542 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1543 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1544 1545 gfs2_statfs_change(sdp, 0, -(s64)*nblocks, dinode ? 1 : 0); 1546 if (dinode) 1547 gfs2_trans_add_unrevoke(sdp, block, 1); 1548 1549 /* 1550 * This needs reviewing to see why we cannot do the quota change 1551 * at this point in the dinode case. 1552 */ 1553 if (ndata) 1554 gfs2_quota_change(ip, ndata, ip->i_inode.i_uid, 1555 ip->i_inode.i_gid); 1556 1557 rgd->rd_free_clone -= *nblocks; 1558 trace_gfs2_block_alloc(ip, block, *nblocks, 1559 dinode ? GFS2_BLKST_DINODE : GFS2_BLKST_USED); 1560 *bn = block; 1561 return 0; 1562 1563 rgrp_error: 1564 gfs2_rgrp_error(rgd); 1565 return -EIO; 1566 } 1567 1568 /** 1569 * __gfs2_free_blocks - free a contiguous run of block(s) 1570 * @ip: the inode these blocks are being freed from 1571 * @bstart: first block of a run of contiguous blocks 1572 * @blen: the length of the block run 1573 * @meta: 1 if the blocks represent metadata 1574 * 1575 */ 1576 1577 void __gfs2_free_blocks(struct gfs2_inode *ip, u64 bstart, u32 blen, int meta) 1578 { 1579 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1580 struct gfs2_rgrpd *rgd; 1581 1582 rgd = rgblk_free(sdp, bstart, blen, GFS2_BLKST_FREE); 1583 if (!rgd) 1584 return; 1585 trace_gfs2_block_alloc(ip, bstart, blen, GFS2_BLKST_FREE); 1586 rgd->rd_free += blen; 1587 rgd->rd_flags &= ~GFS2_RGF_TRIMMED; 1588 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1589 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1590 1591 /* Directories keep their data in the metadata address space */ 1592 if (meta || ip->i_depth) 1593 gfs2_meta_wipe(ip, bstart, blen); 1594 } 1595 1596 /** 1597 * gfs2_free_meta - free a contiguous run of data block(s) 1598 * @ip: the inode these blocks are being freed from 1599 * @bstart: first block of a run of contiguous blocks 1600 * @blen: the length of the block run 1601 * 1602 */ 1603 1604 void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen) 1605 { 1606 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1607 1608 __gfs2_free_blocks(ip, bstart, blen, 1); 1609 gfs2_statfs_change(sdp, 0, +blen, 0); 1610 gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid); 1611 } 1612 1613 void gfs2_unlink_di(struct inode *inode) 1614 { 1615 struct gfs2_inode *ip = GFS2_I(inode); 1616 struct gfs2_sbd *sdp = GFS2_SB(inode); 1617 struct gfs2_rgrpd *rgd; 1618 u64 blkno = ip->i_no_addr; 1619 1620 rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_UNLINKED); 1621 if (!rgd) 1622 return; 1623 trace_gfs2_block_alloc(ip, blkno, 1, GFS2_BLKST_UNLINKED); 1624 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1625 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1626 } 1627 1628 static void gfs2_free_uninit_di(struct gfs2_rgrpd *rgd, u64 blkno) 1629 { 1630 struct gfs2_sbd *sdp = rgd->rd_sbd; 1631 struct gfs2_rgrpd *tmp_rgd; 1632 1633 tmp_rgd = rgblk_free(sdp, blkno, 1, GFS2_BLKST_FREE); 1634 if (!tmp_rgd) 1635 return; 1636 gfs2_assert_withdraw(sdp, rgd == tmp_rgd); 1637 1638 if (!rgd->rd_dinodes) 1639 gfs2_consist_rgrpd(rgd); 1640 rgd->rd_dinodes--; 1641 rgd->rd_free++; 1642 1643 gfs2_trans_add_bh(rgd->rd_gl, rgd->rd_bits[0].bi_bh, 1); 1644 gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data); 1645 1646 gfs2_statfs_change(sdp, 0, +1, -1); 1647 } 1648 1649 1650 void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip) 1651 { 1652 gfs2_free_uninit_di(rgd, ip->i_no_addr); 1653 trace_gfs2_block_alloc(ip, ip->i_no_addr, 1, GFS2_BLKST_FREE); 1654 gfs2_quota_change(ip, -1, ip->i_inode.i_uid, ip->i_inode.i_gid); 1655 gfs2_meta_wipe(ip, ip->i_no_addr, 1); 1656 } 1657 1658 /** 1659 * gfs2_check_blk_type - Check the type of a block 1660 * @sdp: The superblock 1661 * @no_addr: The block number to check 1662 * @type: The block type we are looking for 1663 * 1664 * Returns: 0 if the block type matches the expected type 1665 * -ESTALE if it doesn't match 1666 * or -ve errno if something went wrong while checking 1667 */ 1668 1669 int gfs2_check_blk_type(struct gfs2_sbd *sdp, u64 no_addr, unsigned int type) 1670 { 1671 struct gfs2_rgrpd *rgd; 1672 struct gfs2_holder rgd_gh; 1673 int error = -EINVAL; 1674 1675 rgd = gfs2_blk2rgrpd(sdp, no_addr, 1); 1676 if (!rgd) 1677 goto fail; 1678 1679 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_SHARED, 0, &rgd_gh); 1680 if (error) 1681 goto fail; 1682 1683 if (gfs2_get_block_type(rgd, no_addr) != type) 1684 error = -ESTALE; 1685 1686 gfs2_glock_dq_uninit(&rgd_gh); 1687 fail: 1688 return error; 1689 } 1690 1691 /** 1692 * gfs2_rlist_add - add a RG to a list of RGs 1693 * @ip: the inode 1694 * @rlist: the list of resource groups 1695 * @block: the block 1696 * 1697 * Figure out what RG a block belongs to and add that RG to the list 1698 * 1699 * FIXME: Don't use NOFAIL 1700 * 1701 */ 1702 1703 void gfs2_rlist_add(struct gfs2_inode *ip, struct gfs2_rgrp_list *rlist, 1704 u64 block) 1705 { 1706 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 1707 struct gfs2_rgrpd *rgd; 1708 struct gfs2_rgrpd **tmp; 1709 unsigned int new_space; 1710 unsigned int x; 1711 1712 if (gfs2_assert_warn(sdp, !rlist->rl_ghs)) 1713 return; 1714 1715 if (ip->i_rgd && rgrp_contains_block(ip->i_rgd, block)) 1716 rgd = ip->i_rgd; 1717 else 1718 rgd = gfs2_blk2rgrpd(sdp, block, 1); 1719 if (!rgd) { 1720 fs_err(sdp, "rlist_add: no rgrp for block %llu\n", (unsigned long long)block); 1721 return; 1722 } 1723 ip->i_rgd = rgd; 1724 1725 for (x = 0; x < rlist->rl_rgrps; x++) 1726 if (rlist->rl_rgd[x] == rgd) 1727 return; 1728 1729 if (rlist->rl_rgrps == rlist->rl_space) { 1730 new_space = rlist->rl_space + 10; 1731 1732 tmp = kcalloc(new_space, sizeof(struct gfs2_rgrpd *), 1733 GFP_NOFS | __GFP_NOFAIL); 1734 1735 if (rlist->rl_rgd) { 1736 memcpy(tmp, rlist->rl_rgd, 1737 rlist->rl_space * sizeof(struct gfs2_rgrpd *)); 1738 kfree(rlist->rl_rgd); 1739 } 1740 1741 rlist->rl_space = new_space; 1742 rlist->rl_rgd = tmp; 1743 } 1744 1745 rlist->rl_rgd[rlist->rl_rgrps++] = rgd; 1746 } 1747 1748 /** 1749 * gfs2_rlist_alloc - all RGs have been added to the rlist, now allocate 1750 * and initialize an array of glock holders for them 1751 * @rlist: the list of resource groups 1752 * @state: the lock state to acquire the RG lock in 1753 * @flags: the modifier flags for the holder structures 1754 * 1755 * FIXME: Don't use NOFAIL 1756 * 1757 */ 1758 1759 void gfs2_rlist_alloc(struct gfs2_rgrp_list *rlist, unsigned int state) 1760 { 1761 unsigned int x; 1762 1763 rlist->rl_ghs = kcalloc(rlist->rl_rgrps, sizeof(struct gfs2_holder), 1764 GFP_NOFS | __GFP_NOFAIL); 1765 for (x = 0; x < rlist->rl_rgrps; x++) 1766 gfs2_holder_init(rlist->rl_rgd[x]->rd_gl, 1767 state, 0, 1768 &rlist->rl_ghs[x]); 1769 } 1770 1771 /** 1772 * gfs2_rlist_free - free a resource group list 1773 * @list: the list of resource groups 1774 * 1775 */ 1776 1777 void gfs2_rlist_free(struct gfs2_rgrp_list *rlist) 1778 { 1779 unsigned int x; 1780 1781 kfree(rlist->rl_rgd); 1782 1783 if (rlist->rl_ghs) { 1784 for (x = 0; x < rlist->rl_rgrps; x++) 1785 gfs2_holder_uninit(&rlist->rl_ghs[x]); 1786 kfree(rlist->rl_ghs); 1787 } 1788 } 1789 1790