1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2006 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 /* 11 * Implements Extendible Hashing as described in: 12 * "Extendible Hashing" by Fagin, et al in 13 * __ACM Trans. on Database Systems__, Sept 1979. 14 * 15 * 16 * Here's the layout of dirents which is essentially the same as that of ext2 17 * within a single block. The field de_name_len is the number of bytes 18 * actually required for the name (no null terminator). The field de_rec_len 19 * is the number of bytes allocated to the dirent. The offset of the next 20 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is 21 * deleted, the preceding dirent inherits its allocated space, ie 22 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained 23 * by adding de_rec_len to the current dirent, this essentially causes the 24 * deleted dirent to get jumped over when iterating through all the dirents. 25 * 26 * When deleting the first dirent in a block, there is no previous dirent so 27 * the field de_ino is set to zero to designate it as deleted. When allocating 28 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the 29 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first 30 * dirent is allocated. Otherwise it must go through all the 'used' dirents 31 * searching for one in which the amount of total space minus the amount of 32 * used space will provide enough space for the new dirent. 33 * 34 * There are two types of blocks in which dirents reside. In a stuffed dinode, 35 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of 36 * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the 37 * beginning of the leaf block. The dirents reside in leaves when 38 * 39 * dip->i_diskflags & GFS2_DIF_EXHASH is true 40 * 41 * Otherwise, the dirents are "linear", within a single stuffed dinode block. 42 * 43 * When the dirents are in leaves, the actual contents of the directory file are 44 * used as an array of 64-bit block pointers pointing to the leaf blocks. The 45 * dirents are NOT in the directory file itself. There can be more than one 46 * block pointer in the array that points to the same leaf. In fact, when a 47 * directory is first converted from linear to exhash, all of the pointers 48 * point to the same leaf. 49 * 50 * When a leaf is completely full, the size of the hash table can be 51 * doubled unless it is already at the maximum size which is hard coded into 52 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list, 53 * but never before the maximum hash table size has been reached. 54 */ 55 56 #include <linux/slab.h> 57 #include <linux/spinlock.h> 58 #include <linux/buffer_head.h> 59 #include <linux/sort.h> 60 #include <linux/gfs2_ondisk.h> 61 #include <linux/crc32.h> 62 #include <linux/vmalloc.h> 63 64 #include "gfs2.h" 65 #include "incore.h" 66 #include "dir.h" 67 #include "glock.h" 68 #include "inode.h" 69 #include "meta_io.h" 70 #include "quota.h" 71 #include "rgrp.h" 72 #include "trans.h" 73 #include "bmap.h" 74 #include "util.h" 75 76 #define IS_LEAF 1 /* Hashed (leaf) directory */ 77 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */ 78 79 #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */ 80 81 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) 82 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) 83 84 struct qstr gfs2_qdot __read_mostly; 85 struct qstr gfs2_qdotdot __read_mostly; 86 87 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, 88 const struct qstr *name, void *opaque); 89 90 int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, 91 struct buffer_head **bhp) 92 { 93 struct buffer_head *bh; 94 95 bh = gfs2_meta_new(ip->i_gl, block); 96 gfs2_trans_add_meta(ip->i_gl, bh); 97 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); 98 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); 99 *bhp = bh; 100 return 0; 101 } 102 103 static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, 104 struct buffer_head **bhp) 105 { 106 struct buffer_head *bh; 107 int error; 108 109 error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh); 110 if (error) 111 return error; 112 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { 113 brelse(bh); 114 return -EIO; 115 } 116 *bhp = bh; 117 return 0; 118 } 119 120 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, 121 unsigned int offset, unsigned int size) 122 { 123 struct buffer_head *dibh; 124 int error; 125 126 error = gfs2_meta_inode_buffer(ip, &dibh); 127 if (error) 128 return error; 129 130 gfs2_trans_add_meta(ip->i_gl, dibh); 131 memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); 132 if (ip->i_inode.i_size < offset + size) 133 i_size_write(&ip->i_inode, offset + size); 134 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; 135 gfs2_dinode_out(ip, dibh->b_data); 136 137 brelse(dibh); 138 139 return size; 140 } 141 142 143 144 /** 145 * gfs2_dir_write_data - Write directory information to the inode 146 * @ip: The GFS2 inode 147 * @buf: The buffer containing information to be written 148 * @offset: The file offset to start writing at 149 * @size: The amount of data to write 150 * 151 * Returns: The number of bytes correctly written or error code 152 */ 153 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, 154 u64 offset, unsigned int size) 155 { 156 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 157 struct buffer_head *dibh; 158 u64 lblock, dblock; 159 u32 extlen = 0; 160 unsigned int o; 161 int copied = 0; 162 int error = 0; 163 int new = 0; 164 165 if (!size) 166 return 0; 167 168 if (gfs2_is_stuffed(ip) && 169 offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) 170 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, 171 size); 172 173 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) 174 return -EINVAL; 175 176 if (gfs2_is_stuffed(ip)) { 177 error = gfs2_unstuff_dinode(ip, NULL); 178 if (error) 179 return error; 180 } 181 182 lblock = offset; 183 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); 184 185 while (copied < size) { 186 unsigned int amount; 187 struct buffer_head *bh; 188 189 amount = size - copied; 190 if (amount > sdp->sd_sb.sb_bsize - o) 191 amount = sdp->sd_sb.sb_bsize - o; 192 193 if (!extlen) { 194 new = 1; 195 error = gfs2_extent_map(&ip->i_inode, lblock, &new, 196 &dblock, &extlen); 197 if (error) 198 goto fail; 199 error = -EIO; 200 if (gfs2_assert_withdraw(sdp, dblock)) 201 goto fail; 202 } 203 204 if (amount == sdp->sd_jbsize || new) 205 error = gfs2_dir_get_new_buffer(ip, dblock, &bh); 206 else 207 error = gfs2_dir_get_existing_buffer(ip, dblock, &bh); 208 209 if (error) 210 goto fail; 211 212 gfs2_trans_add_meta(ip->i_gl, bh); 213 memcpy(bh->b_data + o, buf, amount); 214 brelse(bh); 215 216 buf += amount; 217 copied += amount; 218 lblock++; 219 dblock++; 220 extlen--; 221 222 o = sizeof(struct gfs2_meta_header); 223 } 224 225 out: 226 error = gfs2_meta_inode_buffer(ip, &dibh); 227 if (error) 228 return error; 229 230 if (ip->i_inode.i_size < offset + copied) 231 i_size_write(&ip->i_inode, offset + copied); 232 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; 233 234 gfs2_trans_add_meta(ip->i_gl, dibh); 235 gfs2_dinode_out(ip, dibh->b_data); 236 brelse(dibh); 237 238 return copied; 239 fail: 240 if (copied) 241 goto out; 242 return error; 243 } 244 245 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf, 246 unsigned int size) 247 { 248 struct buffer_head *dibh; 249 int error; 250 251 error = gfs2_meta_inode_buffer(ip, &dibh); 252 if (!error) { 253 memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size); 254 brelse(dibh); 255 } 256 257 return (error) ? error : size; 258 } 259 260 261 /** 262 * gfs2_dir_read_data - Read a data from a directory inode 263 * @ip: The GFS2 Inode 264 * @buf: The buffer to place result into 265 * @size: Amount of data to transfer 266 * 267 * Returns: The amount of data actually copied or the error 268 */ 269 static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf, 270 unsigned int size) 271 { 272 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 273 u64 lblock, dblock; 274 u32 extlen = 0; 275 unsigned int o; 276 int copied = 0; 277 int error = 0; 278 279 if (gfs2_is_stuffed(ip)) 280 return gfs2_dir_read_stuffed(ip, buf, size); 281 282 if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) 283 return -EINVAL; 284 285 lblock = 0; 286 o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); 287 288 while (copied < size) { 289 unsigned int amount; 290 struct buffer_head *bh; 291 int new; 292 293 amount = size - copied; 294 if (amount > sdp->sd_sb.sb_bsize - o) 295 amount = sdp->sd_sb.sb_bsize - o; 296 297 if (!extlen) { 298 new = 0; 299 error = gfs2_extent_map(&ip->i_inode, lblock, &new, 300 &dblock, &extlen); 301 if (error || !dblock) 302 goto fail; 303 BUG_ON(extlen < 1); 304 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); 305 } else { 306 error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh); 307 if (error) 308 goto fail; 309 } 310 error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); 311 if (error) { 312 brelse(bh); 313 goto fail; 314 } 315 dblock++; 316 extlen--; 317 memcpy(buf, bh->b_data + o, amount); 318 brelse(bh); 319 buf += (amount/sizeof(__be64)); 320 copied += amount; 321 lblock++; 322 o = sizeof(struct gfs2_meta_header); 323 } 324 325 return copied; 326 fail: 327 return (copied) ? copied : error; 328 } 329 330 /** 331 * gfs2_dir_get_hash_table - Get pointer to the dir hash table 332 * @ip: The inode in question 333 * 334 * Returns: The hash table or an error 335 */ 336 337 static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip) 338 { 339 struct inode *inode = &ip->i_inode; 340 int ret; 341 u32 hsize; 342 __be64 *hc; 343 344 BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH)); 345 346 hc = ip->i_hash_cache; 347 if (hc) 348 return hc; 349 350 hsize = 1 << ip->i_depth; 351 hsize *= sizeof(__be64); 352 if (hsize != i_size_read(&ip->i_inode)) { 353 gfs2_consist_inode(ip); 354 return ERR_PTR(-EIO); 355 } 356 357 hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN); 358 if (hc == NULL) 359 hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL); 360 361 if (hc == NULL) 362 return ERR_PTR(-ENOMEM); 363 364 ret = gfs2_dir_read_data(ip, hc, hsize); 365 if (ret < 0) { 366 if (is_vmalloc_addr(hc)) 367 vfree(hc); 368 else 369 kfree(hc); 370 return ERR_PTR(ret); 371 } 372 373 spin_lock(&inode->i_lock); 374 if (ip->i_hash_cache) { 375 if (is_vmalloc_addr(hc)) 376 vfree(hc); 377 else 378 kfree(hc); 379 } else { 380 ip->i_hash_cache = hc; 381 } 382 spin_unlock(&inode->i_lock); 383 384 return ip->i_hash_cache; 385 } 386 387 /** 388 * gfs2_dir_hash_inval - Invalidate dir hash 389 * @ip: The directory inode 390 * 391 * Must be called with an exclusive glock, or during glock invalidation. 392 */ 393 void gfs2_dir_hash_inval(struct gfs2_inode *ip) 394 { 395 __be64 *hc = ip->i_hash_cache; 396 ip->i_hash_cache = NULL; 397 if (is_vmalloc_addr(hc)) 398 vfree(hc); 399 else 400 kfree(hc); 401 } 402 403 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent) 404 { 405 return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0; 406 } 407 408 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, 409 const struct qstr *name, int ret) 410 { 411 if (!gfs2_dirent_sentinel(dent) && 412 be32_to_cpu(dent->de_hash) == name->hash && 413 be16_to_cpu(dent->de_name_len) == name->len && 414 memcmp(dent+1, name->name, name->len) == 0) 415 return ret; 416 return 0; 417 } 418 419 static int gfs2_dirent_find(const struct gfs2_dirent *dent, 420 const struct qstr *name, 421 void *opaque) 422 { 423 return __gfs2_dirent_find(dent, name, 1); 424 } 425 426 static int gfs2_dirent_prev(const struct gfs2_dirent *dent, 427 const struct qstr *name, 428 void *opaque) 429 { 430 return __gfs2_dirent_find(dent, name, 2); 431 } 432 433 /* 434 * name->name holds ptr to start of block. 435 * name->len holds size of block. 436 */ 437 static int gfs2_dirent_last(const struct gfs2_dirent *dent, 438 const struct qstr *name, 439 void *opaque) 440 { 441 const char *start = name->name; 442 const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); 443 if (name->len == (end - start)) 444 return 1; 445 return 0; 446 } 447 448 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, 449 const struct qstr *name, 450 void *opaque) 451 { 452 unsigned required = GFS2_DIRENT_SIZE(name->len); 453 unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); 454 unsigned totlen = be16_to_cpu(dent->de_rec_len); 455 456 if (gfs2_dirent_sentinel(dent)) 457 actual = 0; 458 if (totlen - actual >= required) 459 return 1; 460 return 0; 461 } 462 463 struct dirent_gather { 464 const struct gfs2_dirent **pdent; 465 unsigned offset; 466 }; 467 468 static int gfs2_dirent_gather(const struct gfs2_dirent *dent, 469 const struct qstr *name, 470 void *opaque) 471 { 472 struct dirent_gather *g = opaque; 473 if (!gfs2_dirent_sentinel(dent)) { 474 g->pdent[g->offset++] = dent; 475 } 476 return 0; 477 } 478 479 /* 480 * Other possible things to check: 481 * - Inode located within filesystem size (and on valid block) 482 * - Valid directory entry type 483 * Not sure how heavy-weight we want to make this... could also check 484 * hash is correct for example, but that would take a lot of extra time. 485 * For now the most important thing is to check that the various sizes 486 * are correct. 487 */ 488 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, 489 unsigned int size, unsigned int len, int first) 490 { 491 const char *msg = "gfs2_dirent too small"; 492 if (unlikely(size < sizeof(struct gfs2_dirent))) 493 goto error; 494 msg = "gfs2_dirent misaligned"; 495 if (unlikely(offset & 0x7)) 496 goto error; 497 msg = "gfs2_dirent points beyond end of block"; 498 if (unlikely(offset + size > len)) 499 goto error; 500 msg = "zero inode number"; 501 if (unlikely(!first && gfs2_dirent_sentinel(dent))) 502 goto error; 503 msg = "name length is greater than space in dirent"; 504 if (!gfs2_dirent_sentinel(dent) && 505 unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > 506 size)) 507 goto error; 508 return 0; 509 error: 510 printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg, 511 first ? "first in block" : "not first in block"); 512 return -EIO; 513 } 514 515 static int gfs2_dirent_offset(const void *buf) 516 { 517 const struct gfs2_meta_header *h = buf; 518 int offset; 519 520 BUG_ON(buf == NULL); 521 522 switch(be32_to_cpu(h->mh_type)) { 523 case GFS2_METATYPE_LF: 524 offset = sizeof(struct gfs2_leaf); 525 break; 526 case GFS2_METATYPE_DI: 527 offset = sizeof(struct gfs2_dinode); 528 break; 529 default: 530 goto wrong_type; 531 } 532 return offset; 533 wrong_type: 534 printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n", 535 be32_to_cpu(h->mh_type)); 536 return -1; 537 } 538 539 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, 540 unsigned int len, gfs2_dscan_t scan, 541 const struct qstr *name, 542 void *opaque) 543 { 544 struct gfs2_dirent *dent, *prev; 545 unsigned offset; 546 unsigned size; 547 int ret = 0; 548 549 ret = gfs2_dirent_offset(buf); 550 if (ret < 0) 551 goto consist_inode; 552 553 offset = ret; 554 prev = NULL; 555 dent = buf + offset; 556 size = be16_to_cpu(dent->de_rec_len); 557 if (gfs2_check_dirent(dent, offset, size, len, 1)) 558 goto consist_inode; 559 do { 560 ret = scan(dent, name, opaque); 561 if (ret) 562 break; 563 offset += size; 564 if (offset == len) 565 break; 566 prev = dent; 567 dent = buf + offset; 568 size = be16_to_cpu(dent->de_rec_len); 569 if (gfs2_check_dirent(dent, offset, size, len, 0)) 570 goto consist_inode; 571 } while(1); 572 573 switch(ret) { 574 case 0: 575 return NULL; 576 case 1: 577 return dent; 578 case 2: 579 return prev ? prev : dent; 580 default: 581 BUG_ON(ret > 0); 582 return ERR_PTR(ret); 583 } 584 585 consist_inode: 586 gfs2_consist_inode(GFS2_I(inode)); 587 return ERR_PTR(-EIO); 588 } 589 590 static int dirent_check_reclen(struct gfs2_inode *dip, 591 const struct gfs2_dirent *d, const void *end_p) 592 { 593 const void *ptr = d; 594 u16 rec_len = be16_to_cpu(d->de_rec_len); 595 596 if (unlikely(rec_len < sizeof(struct gfs2_dirent))) 597 goto broken; 598 ptr += rec_len; 599 if (ptr < end_p) 600 return rec_len; 601 if (ptr == end_p) 602 return -ENOENT; 603 broken: 604 gfs2_consist_inode(dip); 605 return -EIO; 606 } 607 608 /** 609 * dirent_next - Next dirent 610 * @dip: the directory 611 * @bh: The buffer 612 * @dent: Pointer to list of dirents 613 * 614 * Returns: 0 on success, error code otherwise 615 */ 616 617 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, 618 struct gfs2_dirent **dent) 619 { 620 struct gfs2_dirent *cur = *dent, *tmp; 621 char *bh_end = bh->b_data + bh->b_size; 622 int ret; 623 624 ret = dirent_check_reclen(dip, cur, bh_end); 625 if (ret < 0) 626 return ret; 627 628 tmp = (void *)cur + ret; 629 ret = dirent_check_reclen(dip, tmp, bh_end); 630 if (ret == -EIO) 631 return ret; 632 633 /* Only the first dent could ever have de_inum.no_addr == 0 */ 634 if (gfs2_dirent_sentinel(tmp)) { 635 gfs2_consist_inode(dip); 636 return -EIO; 637 } 638 639 *dent = tmp; 640 return 0; 641 } 642 643 /** 644 * dirent_del - Delete a dirent 645 * @dip: The GFS2 inode 646 * @bh: The buffer 647 * @prev: The previous dirent 648 * @cur: The current dirent 649 * 650 */ 651 652 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, 653 struct gfs2_dirent *prev, struct gfs2_dirent *cur) 654 { 655 u16 cur_rec_len, prev_rec_len; 656 657 if (gfs2_dirent_sentinel(cur)) { 658 gfs2_consist_inode(dip); 659 return; 660 } 661 662 gfs2_trans_add_meta(dip->i_gl, bh); 663 664 /* If there is no prev entry, this is the first entry in the block. 665 The de_rec_len is already as big as it needs to be. Just zero 666 out the inode number and return. */ 667 668 if (!prev) { 669 cur->de_inum.no_addr = 0; 670 cur->de_inum.no_formal_ino = 0; 671 return; 672 } 673 674 /* Combine this dentry with the previous one. */ 675 676 prev_rec_len = be16_to_cpu(prev->de_rec_len); 677 cur_rec_len = be16_to_cpu(cur->de_rec_len); 678 679 if ((char *)prev + prev_rec_len != (char *)cur) 680 gfs2_consist_inode(dip); 681 if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) 682 gfs2_consist_inode(dip); 683 684 prev_rec_len += cur_rec_len; 685 prev->de_rec_len = cpu_to_be16(prev_rec_len); 686 } 687 688 /* 689 * Takes a dent from which to grab space as an argument. Returns the 690 * newly created dent. 691 */ 692 static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, 693 struct gfs2_dirent *dent, 694 const struct qstr *name, 695 struct buffer_head *bh) 696 { 697 struct gfs2_inode *ip = GFS2_I(inode); 698 struct gfs2_dirent *ndent; 699 unsigned offset = 0, totlen; 700 701 if (!gfs2_dirent_sentinel(dent)) 702 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); 703 totlen = be16_to_cpu(dent->de_rec_len); 704 BUG_ON(offset + name->len > totlen); 705 gfs2_trans_add_meta(ip->i_gl, bh); 706 ndent = (struct gfs2_dirent *)((char *)dent + offset); 707 dent->de_rec_len = cpu_to_be16(offset); 708 gfs2_qstr2dirent(name, totlen - offset, ndent); 709 return ndent; 710 } 711 712 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode, 713 struct buffer_head *bh, 714 const struct qstr *name) 715 { 716 struct gfs2_dirent *dent; 717 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, 718 gfs2_dirent_find_space, name, NULL); 719 if (!dent || IS_ERR(dent)) 720 return dent; 721 return gfs2_init_dirent(inode, dent, name, bh); 722 } 723 724 static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, 725 struct buffer_head **bhp) 726 { 727 int error; 728 729 error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp); 730 if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { 731 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */ 732 error = -EIO; 733 } 734 735 return error; 736 } 737 738 /** 739 * get_leaf_nr - Get a leaf number associated with the index 740 * @dip: The GFS2 inode 741 * @index: 742 * @leaf_out: 743 * 744 * Returns: 0 on success, error code otherwise 745 */ 746 747 static int get_leaf_nr(struct gfs2_inode *dip, u32 index, 748 u64 *leaf_out) 749 { 750 __be64 *hash; 751 752 hash = gfs2_dir_get_hash_table(dip); 753 if (IS_ERR(hash)) 754 return PTR_ERR(hash); 755 *leaf_out = be64_to_cpu(*(hash + index)); 756 return 0; 757 } 758 759 static int get_first_leaf(struct gfs2_inode *dip, u32 index, 760 struct buffer_head **bh_out) 761 { 762 u64 leaf_no; 763 int error; 764 765 error = get_leaf_nr(dip, index, &leaf_no); 766 if (!error) 767 error = get_leaf(dip, leaf_no, bh_out); 768 769 return error; 770 } 771 772 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, 773 const struct qstr *name, 774 gfs2_dscan_t scan, 775 struct buffer_head **pbh) 776 { 777 struct buffer_head *bh; 778 struct gfs2_dirent *dent; 779 struct gfs2_inode *ip = GFS2_I(inode); 780 int error; 781 782 if (ip->i_diskflags & GFS2_DIF_EXHASH) { 783 struct gfs2_leaf *leaf; 784 unsigned hsize = 1 << ip->i_depth; 785 unsigned index; 786 u64 ln; 787 if (hsize * sizeof(u64) != i_size_read(inode)) { 788 gfs2_consist_inode(ip); 789 return ERR_PTR(-EIO); 790 } 791 792 index = name->hash >> (32 - ip->i_depth); 793 error = get_first_leaf(ip, index, &bh); 794 if (error) 795 return ERR_PTR(error); 796 do { 797 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, 798 scan, name, NULL); 799 if (dent) 800 goto got_dent; 801 leaf = (struct gfs2_leaf *)bh->b_data; 802 ln = be64_to_cpu(leaf->lf_next); 803 brelse(bh); 804 if (!ln) 805 break; 806 807 error = get_leaf(ip, ln, &bh); 808 } while(!error); 809 810 return error ? ERR_PTR(error) : NULL; 811 } 812 813 814 error = gfs2_meta_inode_buffer(ip, &bh); 815 if (error) 816 return ERR_PTR(error); 817 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL); 818 got_dent: 819 if (unlikely(dent == NULL || IS_ERR(dent))) { 820 brelse(bh); 821 bh = NULL; 822 } 823 *pbh = bh; 824 return dent; 825 } 826 827 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) 828 { 829 struct gfs2_inode *ip = GFS2_I(inode); 830 unsigned int n = 1; 831 u64 bn; 832 int error; 833 struct buffer_head *bh; 834 struct gfs2_leaf *leaf; 835 struct gfs2_dirent *dent; 836 struct qstr name = { .name = "" }; 837 838 error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); 839 if (error) 840 return NULL; 841 bh = gfs2_meta_new(ip->i_gl, bn); 842 if (!bh) 843 return NULL; 844 845 gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1); 846 gfs2_trans_add_meta(ip->i_gl, bh); 847 gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); 848 leaf = (struct gfs2_leaf *)bh->b_data; 849 leaf->lf_depth = cpu_to_be16(depth); 850 leaf->lf_entries = 0; 851 leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); 852 leaf->lf_next = 0; 853 memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved)); 854 dent = (struct gfs2_dirent *)(leaf+1); 855 gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); 856 *pbh = bh; 857 return leaf; 858 } 859 860 /** 861 * dir_make_exhash - Convert a stuffed directory into an ExHash directory 862 * @dip: The GFS2 inode 863 * 864 * Returns: 0 on success, error code otherwise 865 */ 866 867 static int dir_make_exhash(struct inode *inode) 868 { 869 struct gfs2_inode *dip = GFS2_I(inode); 870 struct gfs2_sbd *sdp = GFS2_SB(inode); 871 struct gfs2_dirent *dent; 872 struct qstr args; 873 struct buffer_head *bh, *dibh; 874 struct gfs2_leaf *leaf; 875 int y; 876 u32 x; 877 __be64 *lp; 878 u64 bn; 879 int error; 880 881 error = gfs2_meta_inode_buffer(dip, &dibh); 882 if (error) 883 return error; 884 885 /* Turn over a new leaf */ 886 887 leaf = new_leaf(inode, &bh, 0); 888 if (!leaf) 889 return -ENOSPC; 890 bn = bh->b_blocknr; 891 892 gfs2_assert(sdp, dip->i_entries < (1 << 16)); 893 leaf->lf_entries = cpu_to_be16(dip->i_entries); 894 895 /* Copy dirents */ 896 897 gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, 898 sizeof(struct gfs2_dinode)); 899 900 /* Find last entry */ 901 902 x = 0; 903 args.len = bh->b_size - sizeof(struct gfs2_dinode) + 904 sizeof(struct gfs2_leaf); 905 args.name = bh->b_data; 906 dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, 907 gfs2_dirent_last, &args, NULL); 908 if (!dent) { 909 brelse(bh); 910 brelse(dibh); 911 return -EIO; 912 } 913 if (IS_ERR(dent)) { 914 brelse(bh); 915 brelse(dibh); 916 return PTR_ERR(dent); 917 } 918 919 /* Adjust the last dirent's record length 920 (Remember that dent still points to the last entry.) */ 921 922 dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + 923 sizeof(struct gfs2_dinode) - 924 sizeof(struct gfs2_leaf)); 925 926 brelse(bh); 927 928 /* We're done with the new leaf block, now setup the new 929 hash table. */ 930 931 gfs2_trans_add_meta(dip->i_gl, dibh); 932 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 933 934 lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); 935 936 for (x = sdp->sd_hash_ptrs; x--; lp++) 937 *lp = cpu_to_be64(bn); 938 939 i_size_write(inode, sdp->sd_sb.sb_bsize / 2); 940 gfs2_add_inode_blocks(&dip->i_inode, 1); 941 dip->i_diskflags |= GFS2_DIF_EXHASH; 942 943 for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; 944 dip->i_depth = y; 945 946 gfs2_dinode_out(dip, dibh->b_data); 947 948 brelse(dibh); 949 950 return 0; 951 } 952 953 /** 954 * dir_split_leaf - Split a leaf block into two 955 * @dip: The GFS2 inode 956 * @index: 957 * @leaf_no: 958 * 959 * Returns: 0 on success, error code on failure 960 */ 961 962 static int dir_split_leaf(struct inode *inode, const struct qstr *name) 963 { 964 struct gfs2_inode *dip = GFS2_I(inode); 965 struct buffer_head *nbh, *obh, *dibh; 966 struct gfs2_leaf *nleaf, *oleaf; 967 struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; 968 u32 start, len, half_len, divider; 969 u64 bn, leaf_no; 970 __be64 *lp; 971 u32 index; 972 int x, moved = 0; 973 int error; 974 975 index = name->hash >> (32 - dip->i_depth); 976 error = get_leaf_nr(dip, index, &leaf_no); 977 if (error) 978 return error; 979 980 /* Get the old leaf block */ 981 error = get_leaf(dip, leaf_no, &obh); 982 if (error) 983 return error; 984 985 oleaf = (struct gfs2_leaf *)obh->b_data; 986 if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { 987 brelse(obh); 988 return 1; /* can't split */ 989 } 990 991 gfs2_trans_add_meta(dip->i_gl, obh); 992 993 nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); 994 if (!nleaf) { 995 brelse(obh); 996 return -ENOSPC; 997 } 998 bn = nbh->b_blocknr; 999 1000 /* Compute the start and len of leaf pointers in the hash table. */ 1001 len = 1 << (dip->i_depth - be16_to_cpu(oleaf->lf_depth)); 1002 half_len = len >> 1; 1003 if (!half_len) { 1004 printk(KERN_WARNING "i_depth %u lf_depth %u index %u\n", dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); 1005 gfs2_consist_inode(dip); 1006 error = -EIO; 1007 goto fail_brelse; 1008 } 1009 1010 start = (index & ~(len - 1)); 1011 1012 /* Change the pointers. 1013 Don't bother distinguishing stuffed from non-stuffed. 1014 This code is complicated enough already. */ 1015 lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS); 1016 if (!lp) { 1017 error = -ENOMEM; 1018 goto fail_brelse; 1019 } 1020 1021 /* Change the pointers */ 1022 for (x = 0; x < half_len; x++) 1023 lp[x] = cpu_to_be64(bn); 1024 1025 gfs2_dir_hash_inval(dip); 1026 1027 error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), 1028 half_len * sizeof(u64)); 1029 if (error != half_len * sizeof(u64)) { 1030 if (error >= 0) 1031 error = -EIO; 1032 goto fail_lpfree; 1033 } 1034 1035 kfree(lp); 1036 1037 /* Compute the divider */ 1038 divider = (start + half_len) << (32 - dip->i_depth); 1039 1040 /* Copy the entries */ 1041 dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); 1042 1043 do { 1044 next = dent; 1045 if (dirent_next(dip, obh, &next)) 1046 next = NULL; 1047 1048 if (!gfs2_dirent_sentinel(dent) && 1049 be32_to_cpu(dent->de_hash) < divider) { 1050 struct qstr str; 1051 str.name = (char*)(dent+1); 1052 str.len = be16_to_cpu(dent->de_name_len); 1053 str.hash = be32_to_cpu(dent->de_hash); 1054 new = gfs2_dirent_alloc(inode, nbh, &str); 1055 if (IS_ERR(new)) { 1056 error = PTR_ERR(new); 1057 break; 1058 } 1059 1060 new->de_inum = dent->de_inum; /* No endian worries */ 1061 new->de_type = dent->de_type; /* No endian worries */ 1062 be16_add_cpu(&nleaf->lf_entries, 1); 1063 1064 dirent_del(dip, obh, prev, dent); 1065 1066 if (!oleaf->lf_entries) 1067 gfs2_consist_inode(dip); 1068 be16_add_cpu(&oleaf->lf_entries, -1); 1069 1070 if (!prev) 1071 prev = dent; 1072 1073 moved = 1; 1074 } else { 1075 prev = dent; 1076 } 1077 dent = next; 1078 } while (dent); 1079 1080 oleaf->lf_depth = nleaf->lf_depth; 1081 1082 error = gfs2_meta_inode_buffer(dip, &dibh); 1083 if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { 1084 gfs2_trans_add_meta(dip->i_gl, dibh); 1085 gfs2_add_inode_blocks(&dip->i_inode, 1); 1086 gfs2_dinode_out(dip, dibh->b_data); 1087 brelse(dibh); 1088 } 1089 1090 brelse(obh); 1091 brelse(nbh); 1092 1093 return error; 1094 1095 fail_lpfree: 1096 kfree(lp); 1097 1098 fail_brelse: 1099 brelse(obh); 1100 brelse(nbh); 1101 return error; 1102 } 1103 1104 /** 1105 * dir_double_exhash - Double size of ExHash table 1106 * @dip: The GFS2 dinode 1107 * 1108 * Returns: 0 on success, error code on failure 1109 */ 1110 1111 static int dir_double_exhash(struct gfs2_inode *dip) 1112 { 1113 struct buffer_head *dibh; 1114 u32 hsize; 1115 u32 hsize_bytes; 1116 __be64 *hc; 1117 __be64 *hc2, *h; 1118 int x; 1119 int error = 0; 1120 1121 hsize = 1 << dip->i_depth; 1122 hsize_bytes = hsize * sizeof(__be64); 1123 1124 hc = gfs2_dir_get_hash_table(dip); 1125 if (IS_ERR(hc)) 1126 return PTR_ERR(hc); 1127 1128 h = hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN); 1129 if (hc2 == NULL) 1130 hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL); 1131 1132 if (!hc2) 1133 return -ENOMEM; 1134 1135 error = gfs2_meta_inode_buffer(dip, &dibh); 1136 if (error) 1137 goto out_kfree; 1138 1139 for (x = 0; x < hsize; x++) { 1140 *h++ = *hc; 1141 *h++ = *hc; 1142 hc++; 1143 } 1144 1145 error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); 1146 if (error != (hsize_bytes * 2)) 1147 goto fail; 1148 1149 gfs2_dir_hash_inval(dip); 1150 dip->i_hash_cache = hc2; 1151 dip->i_depth++; 1152 gfs2_dinode_out(dip, dibh->b_data); 1153 brelse(dibh); 1154 return 0; 1155 1156 fail: 1157 /* Replace original hash table & size */ 1158 gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); 1159 i_size_write(&dip->i_inode, hsize_bytes); 1160 gfs2_dinode_out(dip, dibh->b_data); 1161 brelse(dibh); 1162 out_kfree: 1163 if (is_vmalloc_addr(hc2)) 1164 vfree(hc2); 1165 else 1166 kfree(hc2); 1167 return error; 1168 } 1169 1170 /** 1171 * compare_dents - compare directory entries by hash value 1172 * @a: first dent 1173 * @b: second dent 1174 * 1175 * When comparing the hash entries of @a to @b: 1176 * gt: returns 1 1177 * lt: returns -1 1178 * eq: returns 0 1179 */ 1180 1181 static int compare_dents(const void *a, const void *b) 1182 { 1183 const struct gfs2_dirent *dent_a, *dent_b; 1184 u32 hash_a, hash_b; 1185 int ret = 0; 1186 1187 dent_a = *(const struct gfs2_dirent **)a; 1188 hash_a = be32_to_cpu(dent_a->de_hash); 1189 1190 dent_b = *(const struct gfs2_dirent **)b; 1191 hash_b = be32_to_cpu(dent_b->de_hash); 1192 1193 if (hash_a > hash_b) 1194 ret = 1; 1195 else if (hash_a < hash_b) 1196 ret = -1; 1197 else { 1198 unsigned int len_a = be16_to_cpu(dent_a->de_name_len); 1199 unsigned int len_b = be16_to_cpu(dent_b->de_name_len); 1200 1201 if (len_a > len_b) 1202 ret = 1; 1203 else if (len_a < len_b) 1204 ret = -1; 1205 else 1206 ret = memcmp(dent_a + 1, dent_b + 1, len_a); 1207 } 1208 1209 return ret; 1210 } 1211 1212 /** 1213 * do_filldir_main - read out directory entries 1214 * @dip: The GFS2 inode 1215 * @offset: The offset in the file to read from 1216 * @opaque: opaque data to pass to filldir 1217 * @filldir: The function to pass entries to 1218 * @darr: an array of struct gfs2_dirent pointers to read 1219 * @entries: the number of entries in darr 1220 * @copied: pointer to int that's non-zero if a entry has been copied out 1221 * 1222 * Jump through some hoops to make sure that if there are hash collsions, 1223 * they are read out at the beginning of a buffer. We want to minimize 1224 * the possibility that they will fall into different readdir buffers or 1225 * that someone will want to seek to that location. 1226 * 1227 * Returns: errno, >0 on exception from filldir 1228 */ 1229 1230 static int do_filldir_main(struct gfs2_inode *dip, u64 *offset, 1231 void *opaque, filldir_t filldir, 1232 const struct gfs2_dirent **darr, u32 entries, 1233 int *copied) 1234 { 1235 const struct gfs2_dirent *dent, *dent_next; 1236 u64 off, off_next; 1237 unsigned int x, y; 1238 int run = 0; 1239 int error = 0; 1240 1241 sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL); 1242 1243 dent_next = darr[0]; 1244 off_next = be32_to_cpu(dent_next->de_hash); 1245 off_next = gfs2_disk_hash2offset(off_next); 1246 1247 for (x = 0, y = 1; x < entries; x++, y++) { 1248 dent = dent_next; 1249 off = off_next; 1250 1251 if (y < entries) { 1252 dent_next = darr[y]; 1253 off_next = be32_to_cpu(dent_next->de_hash); 1254 off_next = gfs2_disk_hash2offset(off_next); 1255 1256 if (off < *offset) 1257 continue; 1258 *offset = off; 1259 1260 if (off_next == off) { 1261 if (*copied && !run) 1262 return 1; 1263 run = 1; 1264 } else 1265 run = 0; 1266 } else { 1267 if (off < *offset) 1268 continue; 1269 *offset = off; 1270 } 1271 1272 error = filldir(opaque, (const char *)(dent + 1), 1273 be16_to_cpu(dent->de_name_len), 1274 off, be64_to_cpu(dent->de_inum.no_addr), 1275 be16_to_cpu(dent->de_type)); 1276 if (error) 1277 return 1; 1278 1279 *copied = 1; 1280 } 1281 1282 /* Increment the *offset by one, so the next time we come into the 1283 do_filldir fxn, we get the next entry instead of the last one in the 1284 current leaf */ 1285 1286 (*offset)++; 1287 1288 return 0; 1289 } 1290 1291 static void *gfs2_alloc_sort_buffer(unsigned size) 1292 { 1293 void *ptr = NULL; 1294 1295 if (size < KMALLOC_MAX_SIZE) 1296 ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); 1297 if (!ptr) 1298 ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL); 1299 return ptr; 1300 } 1301 1302 static void gfs2_free_sort_buffer(void *ptr) 1303 { 1304 if (is_vmalloc_addr(ptr)) 1305 vfree(ptr); 1306 else 1307 kfree(ptr); 1308 } 1309 1310 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque, 1311 filldir_t filldir, int *copied, unsigned *depth, 1312 u64 leaf_no) 1313 { 1314 struct gfs2_inode *ip = GFS2_I(inode); 1315 struct gfs2_sbd *sdp = GFS2_SB(inode); 1316 struct buffer_head *bh; 1317 struct gfs2_leaf *lf; 1318 unsigned entries = 0, entries2 = 0; 1319 unsigned leaves = 0; 1320 const struct gfs2_dirent **darr, *dent; 1321 struct dirent_gather g; 1322 struct buffer_head **larr; 1323 int leaf = 0; 1324 int error, i; 1325 u64 lfn = leaf_no; 1326 1327 do { 1328 error = get_leaf(ip, lfn, &bh); 1329 if (error) 1330 goto out; 1331 lf = (struct gfs2_leaf *)bh->b_data; 1332 if (leaves == 0) 1333 *depth = be16_to_cpu(lf->lf_depth); 1334 entries += be16_to_cpu(lf->lf_entries); 1335 leaves++; 1336 lfn = be64_to_cpu(lf->lf_next); 1337 brelse(bh); 1338 } while(lfn); 1339 1340 if (!entries) 1341 return 0; 1342 1343 error = -ENOMEM; 1344 /* 1345 * The extra 99 entries are not normally used, but are a buffer 1346 * zone in case the number of entries in the leaf is corrupt. 1347 * 99 is the maximum number of entries that can fit in a single 1348 * leaf block. 1349 */ 1350 larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); 1351 if (!larr) 1352 goto out; 1353 darr = (const struct gfs2_dirent **)(larr + leaves); 1354 g.pdent = darr; 1355 g.offset = 0; 1356 lfn = leaf_no; 1357 1358 do { 1359 error = get_leaf(ip, lfn, &bh); 1360 if (error) 1361 goto out_free; 1362 lf = (struct gfs2_leaf *)bh->b_data; 1363 lfn = be64_to_cpu(lf->lf_next); 1364 if (lf->lf_entries) { 1365 entries2 += be16_to_cpu(lf->lf_entries); 1366 dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, 1367 gfs2_dirent_gather, NULL, &g); 1368 error = PTR_ERR(dent); 1369 if (IS_ERR(dent)) 1370 goto out_free; 1371 if (entries2 != g.offset) { 1372 fs_warn(sdp, "Number of entries corrupt in dir " 1373 "leaf %llu, entries2 (%u) != " 1374 "g.offset (%u)\n", 1375 (unsigned long long)bh->b_blocknr, 1376 entries2, g.offset); 1377 1378 error = -EIO; 1379 goto out_free; 1380 } 1381 error = 0; 1382 larr[leaf++] = bh; 1383 } else { 1384 brelse(bh); 1385 } 1386 } while(lfn); 1387 1388 BUG_ON(entries2 != entries); 1389 error = do_filldir_main(ip, offset, opaque, filldir, darr, 1390 entries, copied); 1391 out_free: 1392 for(i = 0; i < leaf; i++) 1393 brelse(larr[i]); 1394 gfs2_free_sort_buffer(larr); 1395 out: 1396 return error; 1397 } 1398 1399 /** 1400 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks. 1401 * 1402 * Note: we can't calculate each index like dir_e_read can because we don't 1403 * have the leaf, and therefore we don't have the depth, and therefore we 1404 * don't have the length. So we have to just read enough ahead to make up 1405 * for the loss of information. 1406 */ 1407 static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index, 1408 struct file_ra_state *f_ra) 1409 { 1410 struct gfs2_inode *ip = GFS2_I(inode); 1411 struct gfs2_glock *gl = ip->i_gl; 1412 struct buffer_head *bh; 1413 u64 blocknr = 0, last; 1414 unsigned count; 1415 1416 /* First check if we've already read-ahead for the whole range. */ 1417 if (index + MAX_RA_BLOCKS < f_ra->start) 1418 return; 1419 1420 f_ra->start = max((pgoff_t)index, f_ra->start); 1421 for (count = 0; count < MAX_RA_BLOCKS; count++) { 1422 if (f_ra->start >= hsize) /* if exceeded the hash table */ 1423 break; 1424 1425 last = blocknr; 1426 blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]); 1427 f_ra->start++; 1428 if (blocknr == last) 1429 continue; 1430 1431 bh = gfs2_getbuf(gl, blocknr, 1); 1432 if (trylock_buffer(bh)) { 1433 if (buffer_uptodate(bh)) { 1434 unlock_buffer(bh); 1435 brelse(bh); 1436 continue; 1437 } 1438 bh->b_end_io = end_buffer_read_sync; 1439 submit_bh(READA | REQ_META, bh); 1440 continue; 1441 } 1442 brelse(bh); 1443 } 1444 } 1445 1446 /** 1447 * dir_e_read - Reads the entries from a directory into a filldir buffer 1448 * @dip: dinode pointer 1449 * @offset: the hash of the last entry read shifted to the right once 1450 * @opaque: buffer for the filldir function to fill 1451 * @filldir: points to the filldir function to use 1452 * 1453 * Returns: errno 1454 */ 1455 1456 static int dir_e_read(struct inode *inode, u64 *offset, void *opaque, 1457 filldir_t filldir, struct file_ra_state *f_ra) 1458 { 1459 struct gfs2_inode *dip = GFS2_I(inode); 1460 u32 hsize, len = 0; 1461 u32 hash, index; 1462 __be64 *lp; 1463 int copied = 0; 1464 int error = 0; 1465 unsigned depth = 0; 1466 1467 hsize = 1 << dip->i_depth; 1468 hash = gfs2_dir_offset2hash(*offset); 1469 index = hash >> (32 - dip->i_depth); 1470 1471 if (dip->i_hash_cache == NULL) 1472 f_ra->start = 0; 1473 lp = gfs2_dir_get_hash_table(dip); 1474 if (IS_ERR(lp)) 1475 return PTR_ERR(lp); 1476 1477 gfs2_dir_readahead(inode, hsize, index, f_ra); 1478 1479 while (index < hsize) { 1480 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir, 1481 &copied, &depth, 1482 be64_to_cpu(lp[index])); 1483 if (error) 1484 break; 1485 1486 len = 1 << (dip->i_depth - depth); 1487 index = (index & ~(len - 1)) + len; 1488 } 1489 1490 if (error > 0) 1491 error = 0; 1492 return error; 1493 } 1494 1495 int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque, 1496 filldir_t filldir, struct file_ra_state *f_ra) 1497 { 1498 struct gfs2_inode *dip = GFS2_I(inode); 1499 struct gfs2_sbd *sdp = GFS2_SB(inode); 1500 struct dirent_gather g; 1501 const struct gfs2_dirent **darr, *dent; 1502 struct buffer_head *dibh; 1503 int copied = 0; 1504 int error; 1505 1506 if (!dip->i_entries) 1507 return 0; 1508 1509 if (dip->i_diskflags & GFS2_DIF_EXHASH) 1510 return dir_e_read(inode, offset, opaque, filldir, f_ra); 1511 1512 if (!gfs2_is_stuffed(dip)) { 1513 gfs2_consist_inode(dip); 1514 return -EIO; 1515 } 1516 1517 error = gfs2_meta_inode_buffer(dip, &dibh); 1518 if (error) 1519 return error; 1520 1521 error = -ENOMEM; 1522 /* 96 is max number of dirents which can be stuffed into an inode */ 1523 darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS); 1524 if (darr) { 1525 g.pdent = darr; 1526 g.offset = 0; 1527 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, 1528 gfs2_dirent_gather, NULL, &g); 1529 if (IS_ERR(dent)) { 1530 error = PTR_ERR(dent); 1531 goto out; 1532 } 1533 if (dip->i_entries != g.offset) { 1534 fs_warn(sdp, "Number of entries corrupt in dir %llu, " 1535 "ip->i_entries (%u) != g.offset (%u)\n", 1536 (unsigned long long)dip->i_no_addr, 1537 dip->i_entries, 1538 g.offset); 1539 error = -EIO; 1540 goto out; 1541 } 1542 error = do_filldir_main(dip, offset, opaque, filldir, darr, 1543 dip->i_entries, &copied); 1544 out: 1545 kfree(darr); 1546 } 1547 1548 if (error > 0) 1549 error = 0; 1550 1551 brelse(dibh); 1552 1553 return error; 1554 } 1555 1556 /** 1557 * gfs2_dir_search - Search a directory 1558 * @dip: The GFS2 inode 1559 * @filename: 1560 * @inode: 1561 * 1562 * This routine searches a directory for a file or another directory. 1563 * Assumes a glock is held on dip. 1564 * 1565 * Returns: errno 1566 */ 1567 1568 struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name) 1569 { 1570 struct buffer_head *bh; 1571 struct gfs2_dirent *dent; 1572 struct inode *inode; 1573 1574 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); 1575 if (dent) { 1576 if (IS_ERR(dent)) 1577 return ERR_CAST(dent); 1578 inode = gfs2_inode_lookup(dir->i_sb, 1579 be16_to_cpu(dent->de_type), 1580 be64_to_cpu(dent->de_inum.no_addr), 1581 be64_to_cpu(dent->de_inum.no_formal_ino), 0); 1582 brelse(bh); 1583 return inode; 1584 } 1585 return ERR_PTR(-ENOENT); 1586 } 1587 1588 int gfs2_dir_check(struct inode *dir, const struct qstr *name, 1589 const struct gfs2_inode *ip) 1590 { 1591 struct buffer_head *bh; 1592 struct gfs2_dirent *dent; 1593 int ret = -ENOENT; 1594 1595 dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); 1596 if (dent) { 1597 if (IS_ERR(dent)) 1598 return PTR_ERR(dent); 1599 if (ip) { 1600 if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr) 1601 goto out; 1602 if (be64_to_cpu(dent->de_inum.no_formal_ino) != 1603 ip->i_no_formal_ino) 1604 goto out; 1605 if (unlikely(IF2DT(ip->i_inode.i_mode) != 1606 be16_to_cpu(dent->de_type))) { 1607 gfs2_consist_inode(GFS2_I(dir)); 1608 ret = -EIO; 1609 goto out; 1610 } 1611 } 1612 ret = 0; 1613 out: 1614 brelse(bh); 1615 } 1616 return ret; 1617 } 1618 1619 static int dir_new_leaf(struct inode *inode, const struct qstr *name) 1620 { 1621 struct buffer_head *bh, *obh; 1622 struct gfs2_inode *ip = GFS2_I(inode); 1623 struct gfs2_leaf *leaf, *oleaf; 1624 int error; 1625 u32 index; 1626 u64 bn; 1627 1628 index = name->hash >> (32 - ip->i_depth); 1629 error = get_first_leaf(ip, index, &obh); 1630 if (error) 1631 return error; 1632 do { 1633 oleaf = (struct gfs2_leaf *)obh->b_data; 1634 bn = be64_to_cpu(oleaf->lf_next); 1635 if (!bn) 1636 break; 1637 brelse(obh); 1638 error = get_leaf(ip, bn, &obh); 1639 if (error) 1640 return error; 1641 } while(1); 1642 1643 gfs2_trans_add_meta(ip->i_gl, obh); 1644 1645 leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); 1646 if (!leaf) { 1647 brelse(obh); 1648 return -ENOSPC; 1649 } 1650 oleaf->lf_next = cpu_to_be64(bh->b_blocknr); 1651 brelse(bh); 1652 brelse(obh); 1653 1654 error = gfs2_meta_inode_buffer(ip, &bh); 1655 if (error) 1656 return error; 1657 gfs2_trans_add_meta(ip->i_gl, bh); 1658 gfs2_add_inode_blocks(&ip->i_inode, 1); 1659 gfs2_dinode_out(ip, bh->b_data); 1660 brelse(bh); 1661 return 0; 1662 } 1663 1664 /** 1665 * gfs2_dir_add - Add new filename into directory 1666 * @dip: The GFS2 inode 1667 * @filename: The new name 1668 * @inode: The inode number of the entry 1669 * @type: The type of the entry 1670 * 1671 * Returns: 0 on success, error code on failure 1672 */ 1673 1674 int gfs2_dir_add(struct inode *inode, const struct qstr *name, 1675 const struct gfs2_inode *nip) 1676 { 1677 struct gfs2_inode *ip = GFS2_I(inode); 1678 struct buffer_head *bh; 1679 struct gfs2_dirent *dent; 1680 struct gfs2_leaf *leaf; 1681 int error; 1682 1683 while(1) { 1684 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, 1685 &bh); 1686 if (dent) { 1687 if (IS_ERR(dent)) 1688 return PTR_ERR(dent); 1689 dent = gfs2_init_dirent(inode, dent, name, bh); 1690 gfs2_inum_out(nip, dent); 1691 dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode)); 1692 if (ip->i_diskflags & GFS2_DIF_EXHASH) { 1693 leaf = (struct gfs2_leaf *)bh->b_data; 1694 be16_add_cpu(&leaf->lf_entries, 1); 1695 } 1696 brelse(bh); 1697 ip->i_entries++; 1698 ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME; 1699 if (S_ISDIR(nip->i_inode.i_mode)) 1700 inc_nlink(&ip->i_inode); 1701 mark_inode_dirty(inode); 1702 error = 0; 1703 break; 1704 } 1705 if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) { 1706 error = dir_make_exhash(inode); 1707 if (error) 1708 break; 1709 continue; 1710 } 1711 error = dir_split_leaf(inode, name); 1712 if (error == 0) 1713 continue; 1714 if (error < 0) 1715 break; 1716 if (ip->i_depth < GFS2_DIR_MAX_DEPTH) { 1717 error = dir_double_exhash(ip); 1718 if (error) 1719 break; 1720 error = dir_split_leaf(inode, name); 1721 if (error < 0) 1722 break; 1723 if (error == 0) 1724 continue; 1725 } 1726 error = dir_new_leaf(inode, name); 1727 if (!error) 1728 continue; 1729 error = -ENOSPC; 1730 break; 1731 } 1732 return error; 1733 } 1734 1735 1736 /** 1737 * gfs2_dir_del - Delete a directory entry 1738 * @dip: The GFS2 inode 1739 * @filename: The filename 1740 * 1741 * Returns: 0 on success, error code on failure 1742 */ 1743 1744 int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry) 1745 { 1746 const struct qstr *name = &dentry->d_name; 1747 struct gfs2_dirent *dent, *prev = NULL; 1748 struct buffer_head *bh; 1749 1750 /* Returns _either_ the entry (if its first in block) or the 1751 previous entry otherwise */ 1752 dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); 1753 if (!dent) { 1754 gfs2_consist_inode(dip); 1755 return -EIO; 1756 } 1757 if (IS_ERR(dent)) { 1758 gfs2_consist_inode(dip); 1759 return PTR_ERR(dent); 1760 } 1761 /* If not first in block, adjust pointers accordingly */ 1762 if (gfs2_dirent_find(dent, name, NULL) == 0) { 1763 prev = dent; 1764 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); 1765 } 1766 1767 dirent_del(dip, bh, prev, dent); 1768 if (dip->i_diskflags & GFS2_DIF_EXHASH) { 1769 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; 1770 u16 entries = be16_to_cpu(leaf->lf_entries); 1771 if (!entries) 1772 gfs2_consist_inode(dip); 1773 leaf->lf_entries = cpu_to_be16(--entries); 1774 } 1775 brelse(bh); 1776 1777 if (!dip->i_entries) 1778 gfs2_consist_inode(dip); 1779 dip->i_entries--; 1780 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; 1781 if (S_ISDIR(dentry->d_inode->i_mode)) 1782 drop_nlink(&dip->i_inode); 1783 mark_inode_dirty(&dip->i_inode); 1784 1785 return 0; 1786 } 1787 1788 /** 1789 * gfs2_dir_mvino - Change inode number of directory entry 1790 * @dip: The GFS2 inode 1791 * @filename: 1792 * @new_inode: 1793 * 1794 * This routine changes the inode number of a directory entry. It's used 1795 * by rename to change ".." when a directory is moved. 1796 * Assumes a glock is held on dvp. 1797 * 1798 * Returns: errno 1799 */ 1800 1801 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, 1802 const struct gfs2_inode *nip, unsigned int new_type) 1803 { 1804 struct buffer_head *bh; 1805 struct gfs2_dirent *dent; 1806 int error; 1807 1808 dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); 1809 if (!dent) { 1810 gfs2_consist_inode(dip); 1811 return -EIO; 1812 } 1813 if (IS_ERR(dent)) 1814 return PTR_ERR(dent); 1815 1816 gfs2_trans_add_meta(dip->i_gl, bh); 1817 gfs2_inum_out(nip, dent); 1818 dent->de_type = cpu_to_be16(new_type); 1819 1820 if (dip->i_diskflags & GFS2_DIF_EXHASH) { 1821 brelse(bh); 1822 error = gfs2_meta_inode_buffer(dip, &bh); 1823 if (error) 1824 return error; 1825 gfs2_trans_add_meta(dip->i_gl, bh); 1826 } 1827 1828 dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME; 1829 gfs2_dinode_out(dip, bh->b_data); 1830 brelse(bh); 1831 return 0; 1832 } 1833 1834 /** 1835 * leaf_dealloc - Deallocate a directory leaf 1836 * @dip: the directory 1837 * @index: the hash table offset in the directory 1838 * @len: the number of pointers to this leaf 1839 * @leaf_no: the leaf number 1840 * @leaf_bh: buffer_head for the starting leaf 1841 * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 1842 * 1843 * Returns: errno 1844 */ 1845 1846 static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, 1847 u64 leaf_no, struct buffer_head *leaf_bh, 1848 int last_dealloc) 1849 { 1850 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 1851 struct gfs2_leaf *tmp_leaf; 1852 struct gfs2_rgrp_list rlist; 1853 struct buffer_head *bh, *dibh; 1854 u64 blk, nblk; 1855 unsigned int rg_blocks = 0, l_blocks = 0; 1856 char *ht; 1857 unsigned int x, size = len * sizeof(u64); 1858 int error; 1859 1860 error = gfs2_rindex_update(sdp); 1861 if (error) 1862 return error; 1863 1864 memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); 1865 1866 ht = kzalloc(size, GFP_NOFS); 1867 if (ht == NULL) 1868 ht = vzalloc(size); 1869 if (!ht) 1870 return -ENOMEM; 1871 1872 error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); 1873 if (error) 1874 goto out; 1875 1876 /* Count the number of leaves */ 1877 bh = leaf_bh; 1878 1879 for (blk = leaf_no; blk; blk = nblk) { 1880 if (blk != leaf_no) { 1881 error = get_leaf(dip, blk, &bh); 1882 if (error) 1883 goto out_rlist; 1884 } 1885 tmp_leaf = (struct gfs2_leaf *)bh->b_data; 1886 nblk = be64_to_cpu(tmp_leaf->lf_next); 1887 if (blk != leaf_no) 1888 brelse(bh); 1889 1890 gfs2_rlist_add(dip, &rlist, blk); 1891 l_blocks++; 1892 } 1893 1894 gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); 1895 1896 for (x = 0; x < rlist.rl_rgrps; x++) { 1897 struct gfs2_rgrpd *rgd; 1898 rgd = rlist.rl_ghs[x].gh_gl->gl_object; 1899 rg_blocks += rgd->rd_length; 1900 } 1901 1902 error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); 1903 if (error) 1904 goto out_rlist; 1905 1906 error = gfs2_trans_begin(sdp, 1907 rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + 1908 RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); 1909 if (error) 1910 goto out_rg_gunlock; 1911 1912 bh = leaf_bh; 1913 1914 for (blk = leaf_no; blk; blk = nblk) { 1915 if (blk != leaf_no) { 1916 error = get_leaf(dip, blk, &bh); 1917 if (error) 1918 goto out_end_trans; 1919 } 1920 tmp_leaf = (struct gfs2_leaf *)bh->b_data; 1921 nblk = be64_to_cpu(tmp_leaf->lf_next); 1922 if (blk != leaf_no) 1923 brelse(bh); 1924 1925 gfs2_free_meta(dip, blk, 1); 1926 gfs2_add_inode_blocks(&dip->i_inode, -1); 1927 } 1928 1929 error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); 1930 if (error != size) { 1931 if (error >= 0) 1932 error = -EIO; 1933 goto out_end_trans; 1934 } 1935 1936 error = gfs2_meta_inode_buffer(dip, &dibh); 1937 if (error) 1938 goto out_end_trans; 1939 1940 gfs2_trans_add_meta(dip->i_gl, dibh); 1941 /* On the last dealloc, make this a regular file in case we crash. 1942 (We don't want to free these blocks a second time.) */ 1943 if (last_dealloc) 1944 dip->i_inode.i_mode = S_IFREG; 1945 gfs2_dinode_out(dip, dibh->b_data); 1946 brelse(dibh); 1947 1948 out_end_trans: 1949 gfs2_trans_end(sdp); 1950 out_rg_gunlock: 1951 gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); 1952 out_rlist: 1953 gfs2_rlist_free(&rlist); 1954 gfs2_quota_unhold(dip); 1955 out: 1956 if (is_vmalloc_addr(ht)) 1957 vfree(ht); 1958 else 1959 kfree(ht); 1960 return error; 1961 } 1962 1963 /** 1964 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory 1965 * @dip: the directory 1966 * 1967 * Dealloc all on-disk directory leaves to FREEMETA state 1968 * Change on-disk inode type to "regular file" 1969 * 1970 * Returns: errno 1971 */ 1972 1973 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) 1974 { 1975 struct buffer_head *bh; 1976 struct gfs2_leaf *leaf; 1977 u32 hsize, len; 1978 u32 index = 0, next_index; 1979 __be64 *lp; 1980 u64 leaf_no; 1981 int error = 0, last; 1982 1983 hsize = 1 << dip->i_depth; 1984 1985 lp = gfs2_dir_get_hash_table(dip); 1986 if (IS_ERR(lp)) 1987 return PTR_ERR(lp); 1988 1989 while (index < hsize) { 1990 leaf_no = be64_to_cpu(lp[index]); 1991 if (leaf_no) { 1992 error = get_leaf(dip, leaf_no, &bh); 1993 if (error) 1994 goto out; 1995 leaf = (struct gfs2_leaf *)bh->b_data; 1996 len = 1 << (dip->i_depth - be16_to_cpu(leaf->lf_depth)); 1997 1998 next_index = (index & ~(len - 1)) + len; 1999 last = ((next_index >= hsize) ? 1 : 0); 2000 error = leaf_dealloc(dip, index, len, leaf_no, bh, 2001 last); 2002 brelse(bh); 2003 if (error) 2004 goto out; 2005 index = next_index; 2006 } else 2007 index++; 2008 } 2009 2010 if (index != hsize) { 2011 gfs2_consist_inode(dip); 2012 error = -EIO; 2013 } 2014 2015 out: 2016 2017 return error; 2018 } 2019 2020 /** 2021 * gfs2_diradd_alloc_required - find if adding entry will require an allocation 2022 * @ip: the file being written to 2023 * @filname: the filename that's going to be added 2024 * 2025 * Returns: 1 if alloc required, 0 if not, -ve on error 2026 */ 2027 2028 int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name) 2029 { 2030 struct gfs2_dirent *dent; 2031 struct buffer_head *bh; 2032 2033 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); 2034 if (!dent) { 2035 return 1; 2036 } 2037 if (IS_ERR(dent)) 2038 return PTR_ERR(dent); 2039 brelse(bh); 2040 return 0; 2041 } 2042 2043