1 /* -*- mode: c; c-basic-offset: 8; -*- 2 * vim: noexpandtab sw=8 ts=8 sts=0: 3 * 4 * xattr.c 5 * 6 * Copyright (C) 2008 Oracle. All rights reserved. 7 * 8 * CREDITS: 9 * Lots of code in this file is taken from ext3. 10 * 11 * This program is free software; you can redistribute it and/or 12 * modify it under the terms of the GNU General Public 13 * License as published by the Free Software Foundation; either 14 * version 2 of the License, or (at your option) any later version. 15 * 16 * This program is distributed in the hope that it will be useful, 17 * but WITHOUT ANY WARRANTY; without even the implied warranty of 18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 19 * General Public License for more details. 20 * 21 * You should have received a copy of the GNU General Public 22 * License along with this program; if not, write to the 23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 24 * Boston, MA 021110-1307, USA. 25 */ 26 27 #include <linux/capability.h> 28 #include <linux/fs.h> 29 #include <linux/types.h> 30 #include <linux/slab.h> 31 #include <linux/highmem.h> 32 #include <linux/pagemap.h> 33 #include <linux/uio.h> 34 #include <linux/sched.h> 35 #include <linux/splice.h> 36 #include <linux/mount.h> 37 #include <linux/writeback.h> 38 #include <linux/falloc.h> 39 #include <linux/sort.h> 40 41 #define MLOG_MASK_PREFIX ML_XATTR 42 #include <cluster/masklog.h> 43 44 #include "ocfs2.h" 45 #include "alloc.h" 46 #include "dlmglue.h" 47 #include "file.h" 48 #include "symlink.h" 49 #include "sysfile.h" 50 #include "inode.h" 51 #include "journal.h" 52 #include "ocfs2_fs.h" 53 #include "suballoc.h" 54 #include "uptodate.h" 55 #include "buffer_head_io.h" 56 #include "super.h" 57 #include "xattr.h" 58 59 60 struct ocfs2_xattr_def_value_root { 61 struct ocfs2_xattr_value_root xv; 62 struct ocfs2_extent_rec er; 63 }; 64 65 struct ocfs2_xattr_bucket { 66 struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET]; 67 struct ocfs2_xattr_header *xh; 68 }; 69 70 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root)) 71 #define OCFS2_XATTR_INLINE_SIZE 80 72 73 static struct ocfs2_xattr_def_value_root def_xv = { 74 .xv.xr_list.l_count = cpu_to_le16(1), 75 }; 76 77 struct xattr_handler *ocfs2_xattr_handlers[] = { 78 &ocfs2_xattr_user_handler, 79 &ocfs2_xattr_trusted_handler, 80 NULL 81 }; 82 83 static struct xattr_handler *ocfs2_xattr_handler_map[] = { 84 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler, 85 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler, 86 }; 87 88 struct ocfs2_xattr_info { 89 int name_index; 90 const char *name; 91 const void *value; 92 size_t value_len; 93 }; 94 95 struct ocfs2_xattr_search { 96 struct buffer_head *inode_bh; 97 /* 98 * xattr_bh point to the block buffer head which has extended attribute 99 * when extended attribute in inode, xattr_bh is equal to inode_bh. 100 */ 101 struct buffer_head *xattr_bh; 102 struct ocfs2_xattr_header *header; 103 struct ocfs2_xattr_bucket bucket; 104 void *base; 105 void *end; 106 struct ocfs2_xattr_entry *here; 107 int not_found; 108 }; 109 110 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode, 111 struct ocfs2_xattr_header *xh, 112 int index, 113 int *block_off, 114 int *new_offset); 115 116 static int ocfs2_xattr_index_block_find(struct inode *inode, 117 struct buffer_head *root_bh, 118 int name_index, 119 const char *name, 120 struct ocfs2_xattr_search *xs); 121 122 static int ocfs2_xattr_tree_list_index_block(struct inode *inode, 123 struct ocfs2_xattr_tree_root *xt, 124 char *buffer, 125 size_t buffer_size); 126 127 static int ocfs2_xattr_create_index_block(struct inode *inode, 128 struct ocfs2_xattr_search *xs); 129 130 static int ocfs2_xattr_set_entry_index_block(struct inode *inode, 131 struct ocfs2_xattr_info *xi, 132 struct ocfs2_xattr_search *xs); 133 134 static int ocfs2_delete_xattr_index_block(struct inode *inode, 135 struct buffer_head *xb_bh); 136 137 static inline struct xattr_handler *ocfs2_xattr_handler(int name_index) 138 { 139 struct xattr_handler *handler = NULL; 140 141 if (name_index > 0 && name_index < OCFS2_XATTR_MAX) 142 handler = ocfs2_xattr_handler_map[name_index]; 143 144 return handler; 145 } 146 147 static u32 ocfs2_xattr_name_hash(struct inode *inode, 148 char *prefix, 149 int prefix_len, 150 char *name, 151 int name_len) 152 { 153 /* Get hash value of uuid from super block */ 154 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash; 155 int i; 156 157 /* hash extended attribute prefix */ 158 for (i = 0; i < prefix_len; i++) { 159 hash = (hash << OCFS2_HASH_SHIFT) ^ 160 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^ 161 *prefix++; 162 } 163 /* hash extended attribute name */ 164 for (i = 0; i < name_len; i++) { 165 hash = (hash << OCFS2_HASH_SHIFT) ^ 166 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^ 167 *name++; 168 } 169 170 return hash; 171 } 172 173 /* 174 * ocfs2_xattr_hash_entry() 175 * 176 * Compute the hash of an extended attribute. 177 */ 178 static void ocfs2_xattr_hash_entry(struct inode *inode, 179 struct ocfs2_xattr_header *header, 180 struct ocfs2_xattr_entry *entry) 181 { 182 u32 hash = 0; 183 struct xattr_handler *handler = 184 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry)); 185 char *prefix = handler->prefix; 186 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset); 187 int prefix_len = strlen(handler->prefix); 188 189 hash = ocfs2_xattr_name_hash(inode, prefix, prefix_len, name, 190 entry->xe_name_len); 191 entry->xe_name_hash = cpu_to_le32(hash); 192 193 return; 194 } 195 196 static int ocfs2_xattr_extend_allocation(struct inode *inode, 197 u32 clusters_to_add, 198 struct buffer_head *xattr_bh, 199 struct ocfs2_xattr_value_root *xv) 200 { 201 int status = 0; 202 int restart_func = 0; 203 int credits = 0; 204 handle_t *handle = NULL; 205 struct ocfs2_alloc_context *data_ac = NULL; 206 struct ocfs2_alloc_context *meta_ac = NULL; 207 enum ocfs2_alloc_restarted why; 208 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 209 u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters); 210 struct ocfs2_extent_tree et; 211 212 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add); 213 214 ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv); 215 216 restart_all: 217 218 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, 219 &data_ac, &meta_ac); 220 if (status) { 221 mlog_errno(status); 222 goto leave; 223 } 224 225 credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el, 226 clusters_to_add); 227 handle = ocfs2_start_trans(osb, credits); 228 if (IS_ERR(handle)) { 229 status = PTR_ERR(handle); 230 handle = NULL; 231 mlog_errno(status); 232 goto leave; 233 } 234 235 restarted_transaction: 236 status = ocfs2_journal_access(handle, inode, xattr_bh, 237 OCFS2_JOURNAL_ACCESS_WRITE); 238 if (status < 0) { 239 mlog_errno(status); 240 goto leave; 241 } 242 243 prev_clusters = le32_to_cpu(xv->xr_clusters); 244 status = ocfs2_add_clusters_in_btree(osb, 245 inode, 246 &logical_start, 247 clusters_to_add, 248 0, 249 &et, 250 handle, 251 data_ac, 252 meta_ac, 253 &why); 254 if ((status < 0) && (status != -EAGAIN)) { 255 if (status != -ENOSPC) 256 mlog_errno(status); 257 goto leave; 258 } 259 260 status = ocfs2_journal_dirty(handle, xattr_bh); 261 if (status < 0) { 262 mlog_errno(status); 263 goto leave; 264 } 265 266 clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters; 267 268 if (why != RESTART_NONE && clusters_to_add) { 269 if (why == RESTART_META) { 270 mlog(0, "restarting function.\n"); 271 restart_func = 1; 272 } else { 273 BUG_ON(why != RESTART_TRANS); 274 275 mlog(0, "restarting transaction.\n"); 276 /* TODO: This can be more intelligent. */ 277 credits = ocfs2_calc_extend_credits(osb->sb, 278 et.et_root_el, 279 clusters_to_add); 280 status = ocfs2_extend_trans(handle, credits); 281 if (status < 0) { 282 /* handle still has to be committed at 283 * this point. */ 284 status = -ENOMEM; 285 mlog_errno(status); 286 goto leave; 287 } 288 goto restarted_transaction; 289 } 290 } 291 292 leave: 293 if (handle) { 294 ocfs2_commit_trans(osb, handle); 295 handle = NULL; 296 } 297 if (data_ac) { 298 ocfs2_free_alloc_context(data_ac); 299 data_ac = NULL; 300 } 301 if (meta_ac) { 302 ocfs2_free_alloc_context(meta_ac); 303 meta_ac = NULL; 304 } 305 if ((!status) && restart_func) { 306 restart_func = 0; 307 goto restart_all; 308 } 309 310 return status; 311 } 312 313 static int __ocfs2_remove_xattr_range(struct inode *inode, 314 struct buffer_head *root_bh, 315 struct ocfs2_xattr_value_root *xv, 316 u32 cpos, u32 phys_cpos, u32 len, 317 struct ocfs2_cached_dealloc_ctxt *dealloc) 318 { 319 int ret; 320 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); 321 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 322 struct inode *tl_inode = osb->osb_tl_inode; 323 handle_t *handle; 324 struct ocfs2_alloc_context *meta_ac = NULL; 325 struct ocfs2_extent_tree et; 326 327 ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv); 328 329 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac); 330 if (ret) { 331 mlog_errno(ret); 332 return ret; 333 } 334 335 mutex_lock(&tl_inode->i_mutex); 336 337 if (ocfs2_truncate_log_needs_flush(osb)) { 338 ret = __ocfs2_flush_truncate_log(osb); 339 if (ret < 0) { 340 mlog_errno(ret); 341 goto out; 342 } 343 } 344 345 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); 346 if (IS_ERR(handle)) { 347 ret = PTR_ERR(handle); 348 mlog_errno(ret); 349 goto out; 350 } 351 352 ret = ocfs2_journal_access(handle, inode, root_bh, 353 OCFS2_JOURNAL_ACCESS_WRITE); 354 if (ret) { 355 mlog_errno(ret); 356 goto out_commit; 357 } 358 359 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac, 360 dealloc); 361 if (ret) { 362 mlog_errno(ret); 363 goto out_commit; 364 } 365 366 le32_add_cpu(&xv->xr_clusters, -len); 367 368 ret = ocfs2_journal_dirty(handle, root_bh); 369 if (ret) { 370 mlog_errno(ret); 371 goto out_commit; 372 } 373 374 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); 375 if (ret) 376 mlog_errno(ret); 377 378 out_commit: 379 ocfs2_commit_trans(osb, handle); 380 out: 381 mutex_unlock(&tl_inode->i_mutex); 382 383 if (meta_ac) 384 ocfs2_free_alloc_context(meta_ac); 385 386 return ret; 387 } 388 389 static int ocfs2_xattr_shrink_size(struct inode *inode, 390 u32 old_clusters, 391 u32 new_clusters, 392 struct buffer_head *root_bh, 393 struct ocfs2_xattr_value_root *xv) 394 { 395 int ret = 0; 396 u32 trunc_len, cpos, phys_cpos, alloc_size; 397 u64 block; 398 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 399 struct ocfs2_cached_dealloc_ctxt dealloc; 400 401 ocfs2_init_dealloc_ctxt(&dealloc); 402 403 if (old_clusters <= new_clusters) 404 return 0; 405 406 cpos = new_clusters; 407 trunc_len = old_clusters - new_clusters; 408 while (trunc_len) { 409 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos, 410 &alloc_size, &xv->xr_list); 411 if (ret) { 412 mlog_errno(ret); 413 goto out; 414 } 415 416 if (alloc_size > trunc_len) 417 alloc_size = trunc_len; 418 419 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos, 420 phys_cpos, alloc_size, 421 &dealloc); 422 if (ret) { 423 mlog_errno(ret); 424 goto out; 425 } 426 427 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); 428 ocfs2_remove_xattr_clusters_from_cache(inode, block, 429 alloc_size); 430 cpos += alloc_size; 431 trunc_len -= alloc_size; 432 } 433 434 out: 435 ocfs2_schedule_truncate_log_flush(osb, 1); 436 ocfs2_run_deallocs(osb, &dealloc); 437 438 return ret; 439 } 440 441 static int ocfs2_xattr_value_truncate(struct inode *inode, 442 struct buffer_head *root_bh, 443 struct ocfs2_xattr_value_root *xv, 444 int len) 445 { 446 int ret; 447 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len); 448 u32 old_clusters = le32_to_cpu(xv->xr_clusters); 449 450 if (new_clusters == old_clusters) 451 return 0; 452 453 if (new_clusters > old_clusters) 454 ret = ocfs2_xattr_extend_allocation(inode, 455 new_clusters - old_clusters, 456 root_bh, xv); 457 else 458 ret = ocfs2_xattr_shrink_size(inode, 459 old_clusters, new_clusters, 460 root_bh, xv); 461 462 return ret; 463 } 464 465 static int ocfs2_xattr_list_entries(struct inode *inode, 466 struct ocfs2_xattr_header *header, 467 char *buffer, size_t buffer_size) 468 { 469 size_t rest = buffer_size; 470 int i; 471 472 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) { 473 struct ocfs2_xattr_entry *entry = &header->xh_entries[i]; 474 struct xattr_handler *handler = 475 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry)); 476 477 if (handler) { 478 size_t size = handler->list(inode, buffer, rest, 479 ((char *)header + 480 le16_to_cpu(entry->xe_name_offset)), 481 entry->xe_name_len); 482 if (buffer) { 483 if (size > rest) 484 return -ERANGE; 485 buffer += size; 486 } 487 rest -= size; 488 } 489 } 490 491 return buffer_size - rest; 492 } 493 494 static int ocfs2_xattr_ibody_list(struct inode *inode, 495 struct ocfs2_dinode *di, 496 char *buffer, 497 size_t buffer_size) 498 { 499 struct ocfs2_xattr_header *header = NULL; 500 struct ocfs2_inode_info *oi = OCFS2_I(inode); 501 int ret = 0; 502 503 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) 504 return ret; 505 506 header = (struct ocfs2_xattr_header *) 507 ((void *)di + inode->i_sb->s_blocksize - 508 le16_to_cpu(di->i_xattr_inline_size)); 509 510 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size); 511 512 return ret; 513 } 514 515 static int ocfs2_xattr_block_list(struct inode *inode, 516 struct ocfs2_dinode *di, 517 char *buffer, 518 size_t buffer_size) 519 { 520 struct buffer_head *blk_bh = NULL; 521 struct ocfs2_xattr_block *xb; 522 int ret = 0; 523 524 if (!di->i_xattr_loc) 525 return ret; 526 527 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), 528 le64_to_cpu(di->i_xattr_loc), 529 &blk_bh, OCFS2_BH_CACHED, inode); 530 if (ret < 0) { 531 mlog_errno(ret); 532 return ret; 533 } 534 /*Verify the signature of xattr block*/ 535 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE, 536 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) { 537 ret = -EFAULT; 538 goto cleanup; 539 } 540 541 xb = (struct ocfs2_xattr_block *)blk_bh->b_data; 542 543 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) { 544 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header; 545 ret = ocfs2_xattr_list_entries(inode, header, 546 buffer, buffer_size); 547 } else { 548 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; 549 ret = ocfs2_xattr_tree_list_index_block(inode, xt, 550 buffer, buffer_size); 551 } 552 cleanup: 553 brelse(blk_bh); 554 555 return ret; 556 } 557 558 ssize_t ocfs2_listxattr(struct dentry *dentry, 559 char *buffer, 560 size_t size) 561 { 562 int ret = 0, i_ret = 0, b_ret = 0; 563 struct buffer_head *di_bh = NULL; 564 struct ocfs2_dinode *di = NULL; 565 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode); 566 567 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb))) 568 return -EOPNOTSUPP; 569 570 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) 571 return ret; 572 573 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0); 574 if (ret < 0) { 575 mlog_errno(ret); 576 return ret; 577 } 578 579 di = (struct ocfs2_dinode *)di_bh->b_data; 580 581 down_read(&oi->ip_xattr_sem); 582 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size); 583 if (i_ret < 0) 584 b_ret = 0; 585 else { 586 if (buffer) { 587 buffer += i_ret; 588 size -= i_ret; 589 } 590 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di, 591 buffer, size); 592 if (b_ret < 0) 593 i_ret = 0; 594 } 595 up_read(&oi->ip_xattr_sem); 596 ocfs2_inode_unlock(dentry->d_inode, 0); 597 598 brelse(di_bh); 599 600 return i_ret + b_ret; 601 } 602 603 static int ocfs2_xattr_find_entry(int name_index, 604 const char *name, 605 struct ocfs2_xattr_search *xs) 606 { 607 struct ocfs2_xattr_entry *entry; 608 size_t name_len; 609 int i, cmp = 1; 610 611 if (name == NULL) 612 return -EINVAL; 613 614 name_len = strlen(name); 615 entry = xs->here; 616 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) { 617 cmp = name_index - ocfs2_xattr_get_type(entry); 618 if (!cmp) 619 cmp = name_len - entry->xe_name_len; 620 if (!cmp) 621 cmp = memcmp(name, (xs->base + 622 le16_to_cpu(entry->xe_name_offset)), 623 name_len); 624 if (cmp == 0) 625 break; 626 entry += 1; 627 } 628 xs->here = entry; 629 630 return cmp ? -ENODATA : 0; 631 } 632 633 static int ocfs2_xattr_get_value_outside(struct inode *inode, 634 struct ocfs2_xattr_value_root *xv, 635 void *buffer, 636 size_t len) 637 { 638 u32 cpos, p_cluster, num_clusters, bpc, clusters; 639 u64 blkno; 640 int i, ret = 0; 641 size_t cplen, blocksize; 642 struct buffer_head *bh = NULL; 643 struct ocfs2_extent_list *el; 644 645 el = &xv->xr_list; 646 clusters = le32_to_cpu(xv->xr_clusters); 647 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); 648 blocksize = inode->i_sb->s_blocksize; 649 650 cpos = 0; 651 while (cpos < clusters) { 652 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster, 653 &num_clusters, el); 654 if (ret) { 655 mlog_errno(ret); 656 goto out; 657 } 658 659 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster); 660 /* Copy ocfs2_xattr_value */ 661 for (i = 0; i < num_clusters * bpc; i++, blkno++) { 662 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno, 663 &bh, OCFS2_BH_CACHED, inode); 664 if (ret) { 665 mlog_errno(ret); 666 goto out; 667 } 668 669 cplen = len >= blocksize ? blocksize : len; 670 memcpy(buffer, bh->b_data, cplen); 671 len -= cplen; 672 buffer += cplen; 673 674 brelse(bh); 675 bh = NULL; 676 if (len == 0) 677 break; 678 } 679 cpos += num_clusters; 680 } 681 out: 682 return ret; 683 } 684 685 static int ocfs2_xattr_ibody_get(struct inode *inode, 686 int name_index, 687 const char *name, 688 void *buffer, 689 size_t buffer_size, 690 struct ocfs2_xattr_search *xs) 691 { 692 struct ocfs2_inode_info *oi = OCFS2_I(inode); 693 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data; 694 struct ocfs2_xattr_value_root *xv; 695 size_t size; 696 int ret = 0; 697 698 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) 699 return -ENODATA; 700 701 xs->end = (void *)di + inode->i_sb->s_blocksize; 702 xs->header = (struct ocfs2_xattr_header *) 703 (xs->end - le16_to_cpu(di->i_xattr_inline_size)); 704 xs->base = (void *)xs->header; 705 xs->here = xs->header->xh_entries; 706 707 ret = ocfs2_xattr_find_entry(name_index, name, xs); 708 if (ret) 709 return ret; 710 size = le64_to_cpu(xs->here->xe_value_size); 711 if (buffer) { 712 if (size > buffer_size) 713 return -ERANGE; 714 if (ocfs2_xattr_is_local(xs->here)) { 715 memcpy(buffer, (void *)xs->base + 716 le16_to_cpu(xs->here->xe_name_offset) + 717 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size); 718 } else { 719 xv = (struct ocfs2_xattr_value_root *) 720 (xs->base + le16_to_cpu( 721 xs->here->xe_name_offset) + 722 OCFS2_XATTR_SIZE(xs->here->xe_name_len)); 723 ret = ocfs2_xattr_get_value_outside(inode, xv, 724 buffer, size); 725 if (ret < 0) { 726 mlog_errno(ret); 727 return ret; 728 } 729 } 730 } 731 732 return size; 733 } 734 735 static int ocfs2_xattr_block_get(struct inode *inode, 736 int name_index, 737 const char *name, 738 void *buffer, 739 size_t buffer_size, 740 struct ocfs2_xattr_search *xs) 741 { 742 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data; 743 struct buffer_head *blk_bh = NULL; 744 struct ocfs2_xattr_block *xb; 745 struct ocfs2_xattr_value_root *xv; 746 size_t size; 747 int ret = -ENODATA, name_offset, name_len, block_off, i; 748 749 if (!di->i_xattr_loc) 750 return ret; 751 752 memset(&xs->bucket, 0, sizeof(xs->bucket)); 753 754 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), 755 le64_to_cpu(di->i_xattr_loc), 756 &blk_bh, OCFS2_BH_CACHED, inode); 757 if (ret < 0) { 758 mlog_errno(ret); 759 return ret; 760 } 761 /*Verify the signature of xattr block*/ 762 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE, 763 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) { 764 ret = -EFAULT; 765 goto cleanup; 766 } 767 768 xs->xattr_bh = blk_bh; 769 xb = (struct ocfs2_xattr_block *)blk_bh->b_data; 770 771 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) { 772 xs->header = &xb->xb_attrs.xb_header; 773 xs->base = (void *)xs->header; 774 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size; 775 xs->here = xs->header->xh_entries; 776 777 ret = ocfs2_xattr_find_entry(name_index, name, xs); 778 } else 779 ret = ocfs2_xattr_index_block_find(inode, blk_bh, 780 name_index, 781 name, xs); 782 783 if (ret) 784 goto cleanup; 785 size = le64_to_cpu(xs->here->xe_value_size); 786 if (buffer) { 787 ret = -ERANGE; 788 if (size > buffer_size) 789 goto cleanup; 790 791 name_offset = le16_to_cpu(xs->here->xe_name_offset); 792 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len); 793 i = xs->here - xs->header->xh_entries; 794 795 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) { 796 ret = ocfs2_xattr_bucket_get_name_value(inode, 797 xs->bucket.xh, 798 i, 799 &block_off, 800 &name_offset); 801 xs->base = xs->bucket.bhs[block_off]->b_data; 802 } 803 if (ocfs2_xattr_is_local(xs->here)) { 804 memcpy(buffer, (void *)xs->base + 805 name_offset + name_len, size); 806 } else { 807 xv = (struct ocfs2_xattr_value_root *) 808 (xs->base + name_offset + name_len); 809 ret = ocfs2_xattr_get_value_outside(inode, xv, 810 buffer, size); 811 if (ret < 0) { 812 mlog_errno(ret); 813 goto cleanup; 814 } 815 } 816 } 817 ret = size; 818 cleanup: 819 for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++) 820 brelse(xs->bucket.bhs[i]); 821 memset(&xs->bucket, 0, sizeof(xs->bucket)); 822 823 brelse(blk_bh); 824 return ret; 825 } 826 827 /* ocfs2_xattr_get() 828 * 829 * Copy an extended attribute into the buffer provided. 830 * Buffer is NULL to compute the size of buffer required. 831 */ 832 int ocfs2_xattr_get(struct inode *inode, 833 int name_index, 834 const char *name, 835 void *buffer, 836 size_t buffer_size) 837 { 838 int ret; 839 struct ocfs2_dinode *di = NULL; 840 struct buffer_head *di_bh = NULL; 841 struct ocfs2_inode_info *oi = OCFS2_I(inode); 842 struct ocfs2_xattr_search xis = { 843 .not_found = -ENODATA, 844 }; 845 struct ocfs2_xattr_search xbs = { 846 .not_found = -ENODATA, 847 }; 848 849 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb))) 850 return -EOPNOTSUPP; 851 852 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) 853 ret = -ENODATA; 854 855 ret = ocfs2_inode_lock(inode, &di_bh, 0); 856 if (ret < 0) { 857 mlog_errno(ret); 858 return ret; 859 } 860 xis.inode_bh = xbs.inode_bh = di_bh; 861 di = (struct ocfs2_dinode *)di_bh->b_data; 862 863 down_read(&oi->ip_xattr_sem); 864 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer, 865 buffer_size, &xis); 866 if (ret == -ENODATA) 867 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer, 868 buffer_size, &xbs); 869 up_read(&oi->ip_xattr_sem); 870 ocfs2_inode_unlock(inode, 0); 871 872 brelse(di_bh); 873 874 return ret; 875 } 876 877 static int __ocfs2_xattr_set_value_outside(struct inode *inode, 878 struct ocfs2_xattr_value_root *xv, 879 const void *value, 880 int value_len) 881 { 882 int ret = 0, i, cp_len, credits; 883 u16 blocksize = inode->i_sb->s_blocksize; 884 u32 p_cluster, num_clusters; 885 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); 886 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len); 887 u64 blkno; 888 struct buffer_head *bh = NULL; 889 handle_t *handle; 890 891 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters)); 892 893 credits = clusters * bpc; 894 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits); 895 if (IS_ERR(handle)) { 896 ret = PTR_ERR(handle); 897 mlog_errno(ret); 898 goto out; 899 } 900 901 while (cpos < clusters) { 902 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster, 903 &num_clusters, &xv->xr_list); 904 if (ret) { 905 mlog_errno(ret); 906 goto out_commit; 907 } 908 909 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster); 910 911 for (i = 0; i < num_clusters * bpc; i++, blkno++) { 912 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno, 913 &bh, OCFS2_BH_CACHED, inode); 914 if (ret) { 915 mlog_errno(ret); 916 goto out_commit; 917 } 918 919 ret = ocfs2_journal_access(handle, 920 inode, 921 bh, 922 OCFS2_JOURNAL_ACCESS_WRITE); 923 if (ret < 0) { 924 mlog_errno(ret); 925 goto out_commit; 926 } 927 928 cp_len = value_len > blocksize ? blocksize : value_len; 929 memcpy(bh->b_data, value, cp_len); 930 value_len -= cp_len; 931 value += cp_len; 932 if (cp_len < blocksize) 933 memset(bh->b_data + cp_len, 0, 934 blocksize - cp_len); 935 936 ret = ocfs2_journal_dirty(handle, bh); 937 if (ret < 0) { 938 mlog_errno(ret); 939 goto out_commit; 940 } 941 brelse(bh); 942 bh = NULL; 943 944 /* 945 * XXX: do we need to empty all the following 946 * blocks in this cluster? 947 */ 948 if (!value_len) 949 break; 950 } 951 cpos += num_clusters; 952 } 953 out_commit: 954 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); 955 out: 956 brelse(bh); 957 958 return ret; 959 } 960 961 static int ocfs2_xattr_cleanup(struct inode *inode, 962 struct ocfs2_xattr_info *xi, 963 struct ocfs2_xattr_search *xs, 964 size_t offs) 965 { 966 handle_t *handle = NULL; 967 int ret = 0; 968 size_t name_len = strlen(xi->name); 969 void *val = xs->base + offs; 970 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE; 971 972 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 973 OCFS2_XATTR_BLOCK_UPDATE_CREDITS); 974 if (IS_ERR(handle)) { 975 ret = PTR_ERR(handle); 976 mlog_errno(ret); 977 goto out; 978 } 979 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh, 980 OCFS2_JOURNAL_ACCESS_WRITE); 981 if (ret) { 982 mlog_errno(ret); 983 goto out_commit; 984 } 985 /* Decrease xattr count */ 986 le16_add_cpu(&xs->header->xh_count, -1); 987 /* Remove the xattr entry and tree root which has already be set*/ 988 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry)); 989 memset(val, 0, size); 990 991 ret = ocfs2_journal_dirty(handle, xs->xattr_bh); 992 if (ret < 0) 993 mlog_errno(ret); 994 out_commit: 995 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); 996 out: 997 return ret; 998 } 999 1000 static int ocfs2_xattr_update_entry(struct inode *inode, 1001 struct ocfs2_xattr_info *xi, 1002 struct ocfs2_xattr_search *xs, 1003 size_t offs) 1004 { 1005 handle_t *handle = NULL; 1006 int ret = 0; 1007 1008 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1009 OCFS2_XATTR_BLOCK_UPDATE_CREDITS); 1010 if (IS_ERR(handle)) { 1011 ret = PTR_ERR(handle); 1012 mlog_errno(ret); 1013 goto out; 1014 } 1015 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh, 1016 OCFS2_JOURNAL_ACCESS_WRITE); 1017 if (ret) { 1018 mlog_errno(ret); 1019 goto out_commit; 1020 } 1021 1022 xs->here->xe_name_offset = cpu_to_le16(offs); 1023 xs->here->xe_value_size = cpu_to_le64(xi->value_len); 1024 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE) 1025 ocfs2_xattr_set_local(xs->here, 1); 1026 else 1027 ocfs2_xattr_set_local(xs->here, 0); 1028 ocfs2_xattr_hash_entry(inode, xs->header, xs->here); 1029 1030 ret = ocfs2_journal_dirty(handle, xs->xattr_bh); 1031 if (ret < 0) 1032 mlog_errno(ret); 1033 out_commit: 1034 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); 1035 out: 1036 return ret; 1037 } 1038 1039 /* 1040 * ocfs2_xattr_set_value_outside() 1041 * 1042 * Set large size value in B tree. 1043 */ 1044 static int ocfs2_xattr_set_value_outside(struct inode *inode, 1045 struct ocfs2_xattr_info *xi, 1046 struct ocfs2_xattr_search *xs, 1047 size_t offs) 1048 { 1049 size_t name_len = strlen(xi->name); 1050 void *val = xs->base + offs; 1051 struct ocfs2_xattr_value_root *xv = NULL; 1052 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE; 1053 int ret = 0; 1054 1055 memset(val, 0, size); 1056 memcpy(val, xi->name, name_len); 1057 xv = (struct ocfs2_xattr_value_root *) 1058 (val + OCFS2_XATTR_SIZE(name_len)); 1059 xv->xr_clusters = 0; 1060 xv->xr_last_eb_blk = 0; 1061 xv->xr_list.l_tree_depth = 0; 1062 xv->xr_list.l_count = cpu_to_le16(1); 1063 xv->xr_list.l_next_free_rec = 0; 1064 1065 ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv, 1066 xi->value_len); 1067 if (ret < 0) { 1068 mlog_errno(ret); 1069 return ret; 1070 } 1071 ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value, 1072 xi->value_len); 1073 if (ret < 0) { 1074 mlog_errno(ret); 1075 return ret; 1076 } 1077 ret = ocfs2_xattr_update_entry(inode, xi, xs, offs); 1078 if (ret < 0) 1079 mlog_errno(ret); 1080 1081 return ret; 1082 } 1083 1084 /* 1085 * ocfs2_xattr_set_entry_local() 1086 * 1087 * Set, replace or remove extended attribute in local. 1088 */ 1089 static void ocfs2_xattr_set_entry_local(struct inode *inode, 1090 struct ocfs2_xattr_info *xi, 1091 struct ocfs2_xattr_search *xs, 1092 struct ocfs2_xattr_entry *last, 1093 size_t min_offs) 1094 { 1095 size_t name_len = strlen(xi->name); 1096 int i; 1097 1098 if (xi->value && xs->not_found) { 1099 /* Insert the new xattr entry. */ 1100 le16_add_cpu(&xs->header->xh_count, 1); 1101 ocfs2_xattr_set_type(last, xi->name_index); 1102 ocfs2_xattr_set_local(last, 1); 1103 last->xe_name_len = name_len; 1104 } else { 1105 void *first_val; 1106 void *val; 1107 size_t offs, size; 1108 1109 first_val = xs->base + min_offs; 1110 offs = le16_to_cpu(xs->here->xe_name_offset); 1111 val = xs->base + offs; 1112 1113 if (le64_to_cpu(xs->here->xe_value_size) > 1114 OCFS2_XATTR_INLINE_SIZE) 1115 size = OCFS2_XATTR_SIZE(name_len) + 1116 OCFS2_XATTR_ROOT_SIZE; 1117 else 1118 size = OCFS2_XATTR_SIZE(name_len) + 1119 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size)); 1120 1121 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) + 1122 OCFS2_XATTR_SIZE(xi->value_len)) { 1123 /* The old and the new value have the 1124 same size. Just replace the value. */ 1125 ocfs2_xattr_set_local(xs->here, 1); 1126 xs->here->xe_value_size = cpu_to_le64(xi->value_len); 1127 /* Clear value bytes. */ 1128 memset(val + OCFS2_XATTR_SIZE(name_len), 1129 0, 1130 OCFS2_XATTR_SIZE(xi->value_len)); 1131 memcpy(val + OCFS2_XATTR_SIZE(name_len), 1132 xi->value, 1133 xi->value_len); 1134 return; 1135 } 1136 /* Remove the old name+value. */ 1137 memmove(first_val + size, first_val, val - first_val); 1138 memset(first_val, 0, size); 1139 xs->here->xe_name_hash = 0; 1140 xs->here->xe_name_offset = 0; 1141 ocfs2_xattr_set_local(xs->here, 1); 1142 xs->here->xe_value_size = 0; 1143 1144 min_offs += size; 1145 1146 /* Adjust all value offsets. */ 1147 last = xs->header->xh_entries; 1148 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) { 1149 size_t o = le16_to_cpu(last->xe_name_offset); 1150 1151 if (o < offs) 1152 last->xe_name_offset = cpu_to_le16(o + size); 1153 last += 1; 1154 } 1155 1156 if (!xi->value) { 1157 /* Remove the old entry. */ 1158 last -= 1; 1159 memmove(xs->here, xs->here + 1, 1160 (void *)last - (void *)xs->here); 1161 memset(last, 0, sizeof(struct ocfs2_xattr_entry)); 1162 le16_add_cpu(&xs->header->xh_count, -1); 1163 } 1164 } 1165 if (xi->value) { 1166 /* Insert the new name+value. */ 1167 size_t size = OCFS2_XATTR_SIZE(name_len) + 1168 OCFS2_XATTR_SIZE(xi->value_len); 1169 void *val = xs->base + min_offs - size; 1170 1171 xs->here->xe_name_offset = cpu_to_le16(min_offs - size); 1172 memset(val, 0, size); 1173 memcpy(val, xi->name, name_len); 1174 memcpy(val + OCFS2_XATTR_SIZE(name_len), 1175 xi->value, 1176 xi->value_len); 1177 xs->here->xe_value_size = cpu_to_le64(xi->value_len); 1178 ocfs2_xattr_set_local(xs->here, 1); 1179 ocfs2_xattr_hash_entry(inode, xs->header, xs->here); 1180 } 1181 1182 return; 1183 } 1184 1185 /* 1186 * ocfs2_xattr_set_entry() 1187 * 1188 * Set extended attribute entry into inode or block. 1189 * 1190 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE, 1191 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(), 1192 * then set value in B tree with set_value_outside(). 1193 */ 1194 static int ocfs2_xattr_set_entry(struct inode *inode, 1195 struct ocfs2_xattr_info *xi, 1196 struct ocfs2_xattr_search *xs, 1197 int flag) 1198 { 1199 struct ocfs2_xattr_entry *last; 1200 struct ocfs2_inode_info *oi = OCFS2_I(inode); 1201 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data; 1202 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name); 1203 size_t size_l = 0; 1204 handle_t *handle = NULL; 1205 int free, i, ret; 1206 struct ocfs2_xattr_info xi_l = { 1207 .name_index = xi->name_index, 1208 .name = xi->name, 1209 .value = xi->value, 1210 .value_len = xi->value_len, 1211 }; 1212 1213 /* Compute min_offs, last and free space. */ 1214 last = xs->header->xh_entries; 1215 1216 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) { 1217 size_t offs = le16_to_cpu(last->xe_name_offset); 1218 if (offs < min_offs) 1219 min_offs = offs; 1220 last += 1; 1221 } 1222 1223 free = min_offs - ((void *)last - xs->base) - sizeof(__u32); 1224 if (free < 0) 1225 return -EFAULT; 1226 1227 if (!xs->not_found) { 1228 size_t size = 0; 1229 if (ocfs2_xattr_is_local(xs->here)) 1230 size = OCFS2_XATTR_SIZE(name_len) + 1231 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size)); 1232 else 1233 size = OCFS2_XATTR_SIZE(name_len) + 1234 OCFS2_XATTR_ROOT_SIZE; 1235 free += (size + sizeof(struct ocfs2_xattr_entry)); 1236 } 1237 /* Check free space in inode or block */ 1238 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) { 1239 if (free < sizeof(struct ocfs2_xattr_entry) + 1240 OCFS2_XATTR_SIZE(name_len) + 1241 OCFS2_XATTR_ROOT_SIZE) { 1242 ret = -ENOSPC; 1243 goto out; 1244 } 1245 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE; 1246 xi_l.value = (void *)&def_xv; 1247 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE; 1248 } else if (xi->value) { 1249 if (free < sizeof(struct ocfs2_xattr_entry) + 1250 OCFS2_XATTR_SIZE(name_len) + 1251 OCFS2_XATTR_SIZE(xi->value_len)) { 1252 ret = -ENOSPC; 1253 goto out; 1254 } 1255 } 1256 1257 if (!xs->not_found) { 1258 /* For existing extended attribute */ 1259 size_t size = OCFS2_XATTR_SIZE(name_len) + 1260 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size)); 1261 size_t offs = le16_to_cpu(xs->here->xe_name_offset); 1262 void *val = xs->base + offs; 1263 1264 if (ocfs2_xattr_is_local(xs->here) && size == size_l) { 1265 /* Replace existing local xattr with tree root */ 1266 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, 1267 offs); 1268 if (ret < 0) 1269 mlog_errno(ret); 1270 goto out; 1271 } else if (!ocfs2_xattr_is_local(xs->here)) { 1272 /* For existing xattr which has value outside */ 1273 struct ocfs2_xattr_value_root *xv = NULL; 1274 xv = (struct ocfs2_xattr_value_root *)(val + 1275 OCFS2_XATTR_SIZE(name_len)); 1276 1277 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) { 1278 /* 1279 * If new value need set outside also, 1280 * first truncate old value to new value, 1281 * then set new value with set_value_outside(). 1282 */ 1283 ret = ocfs2_xattr_value_truncate(inode, 1284 xs->xattr_bh, 1285 xv, 1286 xi->value_len); 1287 if (ret < 0) { 1288 mlog_errno(ret); 1289 goto out; 1290 } 1291 1292 ret = __ocfs2_xattr_set_value_outside(inode, 1293 xv, 1294 xi->value, 1295 xi->value_len); 1296 if (ret < 0) { 1297 mlog_errno(ret); 1298 goto out; 1299 } 1300 1301 ret = ocfs2_xattr_update_entry(inode, 1302 xi, 1303 xs, 1304 offs); 1305 if (ret < 0) 1306 mlog_errno(ret); 1307 goto out; 1308 } else { 1309 /* 1310 * If new value need set in local, 1311 * just trucate old value to zero. 1312 */ 1313 ret = ocfs2_xattr_value_truncate(inode, 1314 xs->xattr_bh, 1315 xv, 1316 0); 1317 if (ret < 0) 1318 mlog_errno(ret); 1319 } 1320 } 1321 } 1322 1323 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1324 OCFS2_INODE_UPDATE_CREDITS); 1325 if (IS_ERR(handle)) { 1326 ret = PTR_ERR(handle); 1327 mlog_errno(ret); 1328 goto out; 1329 } 1330 1331 ret = ocfs2_journal_access(handle, inode, xs->inode_bh, 1332 OCFS2_JOURNAL_ACCESS_WRITE); 1333 if (ret) { 1334 mlog_errno(ret); 1335 goto out_commit; 1336 } 1337 1338 if (!(flag & OCFS2_INLINE_XATTR_FL)) { 1339 /* set extended attribute in external block. */ 1340 ret = ocfs2_extend_trans(handle, 1341 OCFS2_INODE_UPDATE_CREDITS + 1342 OCFS2_XATTR_BLOCK_UPDATE_CREDITS); 1343 if (ret) { 1344 mlog_errno(ret); 1345 goto out_commit; 1346 } 1347 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh, 1348 OCFS2_JOURNAL_ACCESS_WRITE); 1349 if (ret) { 1350 mlog_errno(ret); 1351 goto out_commit; 1352 } 1353 } 1354 1355 /* 1356 * Set value in local, include set tree root in local. 1357 * This is the first step for value size >INLINE_SIZE. 1358 */ 1359 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs); 1360 1361 if (!(flag & OCFS2_INLINE_XATTR_FL)) { 1362 ret = ocfs2_journal_dirty(handle, xs->xattr_bh); 1363 if (ret < 0) { 1364 mlog_errno(ret); 1365 goto out_commit; 1366 } 1367 } 1368 1369 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) && 1370 (flag & OCFS2_INLINE_XATTR_FL)) { 1371 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1372 unsigned int xattrsize = osb->s_xattr_inline_size; 1373 1374 /* 1375 * Adjust extent record count or inline data size 1376 * to reserve space for extended attribute. 1377 */ 1378 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { 1379 struct ocfs2_inline_data *idata = &di->id2.i_data; 1380 le16_add_cpu(&idata->id_count, -xattrsize); 1381 } else if (!(ocfs2_inode_is_fast_symlink(inode))) { 1382 struct ocfs2_extent_list *el = &di->id2.i_list; 1383 le16_add_cpu(&el->l_count, -(xattrsize / 1384 sizeof(struct ocfs2_extent_rec))); 1385 } 1386 di->i_xattr_inline_size = cpu_to_le16(xattrsize); 1387 } 1388 /* Update xattr flag */ 1389 spin_lock(&oi->ip_lock); 1390 oi->ip_dyn_features |= flag; 1391 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); 1392 spin_unlock(&oi->ip_lock); 1393 /* Update inode ctime */ 1394 inode->i_ctime = CURRENT_TIME; 1395 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); 1396 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); 1397 1398 ret = ocfs2_journal_dirty(handle, xs->inode_bh); 1399 if (ret < 0) 1400 mlog_errno(ret); 1401 1402 out_commit: 1403 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); 1404 1405 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) { 1406 /* 1407 * Set value outside in B tree. 1408 * This is the second step for value size > INLINE_SIZE. 1409 */ 1410 size_t offs = le16_to_cpu(xs->here->xe_name_offset); 1411 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs); 1412 if (ret < 0) { 1413 int ret2; 1414 1415 mlog_errno(ret); 1416 /* 1417 * If set value outside failed, we have to clean 1418 * the junk tree root we have already set in local. 1419 */ 1420 ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs); 1421 if (ret2 < 0) 1422 mlog_errno(ret2); 1423 } 1424 } 1425 out: 1426 return ret; 1427 1428 } 1429 1430 static int ocfs2_remove_value_outside(struct inode*inode, 1431 struct buffer_head *bh, 1432 struct ocfs2_xattr_header *header) 1433 { 1434 int ret = 0, i; 1435 1436 for (i = 0; i < le16_to_cpu(header->xh_count); i++) { 1437 struct ocfs2_xattr_entry *entry = &header->xh_entries[i]; 1438 1439 if (!ocfs2_xattr_is_local(entry)) { 1440 struct ocfs2_xattr_value_root *xv; 1441 void *val; 1442 1443 val = (void *)header + 1444 le16_to_cpu(entry->xe_name_offset); 1445 xv = (struct ocfs2_xattr_value_root *) 1446 (val + OCFS2_XATTR_SIZE(entry->xe_name_len)); 1447 ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0); 1448 if (ret < 0) { 1449 mlog_errno(ret); 1450 return ret; 1451 } 1452 } 1453 } 1454 1455 return ret; 1456 } 1457 1458 static int ocfs2_xattr_ibody_remove(struct inode *inode, 1459 struct buffer_head *di_bh) 1460 { 1461 1462 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; 1463 struct ocfs2_xattr_header *header; 1464 int ret; 1465 1466 header = (struct ocfs2_xattr_header *) 1467 ((void *)di + inode->i_sb->s_blocksize - 1468 le16_to_cpu(di->i_xattr_inline_size)); 1469 1470 ret = ocfs2_remove_value_outside(inode, di_bh, header); 1471 1472 return ret; 1473 } 1474 1475 static int ocfs2_xattr_block_remove(struct inode *inode, 1476 struct buffer_head *blk_bh) 1477 { 1478 struct ocfs2_xattr_block *xb; 1479 int ret = 0; 1480 1481 xb = (struct ocfs2_xattr_block *)blk_bh->b_data; 1482 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) { 1483 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header); 1484 ret = ocfs2_remove_value_outside(inode, blk_bh, header); 1485 } else 1486 ret = ocfs2_delete_xattr_index_block(inode, blk_bh); 1487 1488 return ret; 1489 } 1490 1491 static int ocfs2_xattr_free_block(struct inode *inode, 1492 u64 block) 1493 { 1494 struct inode *xb_alloc_inode; 1495 struct buffer_head *xb_alloc_bh = NULL; 1496 struct buffer_head *blk_bh = NULL; 1497 struct ocfs2_xattr_block *xb; 1498 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1499 handle_t *handle; 1500 int ret = 0; 1501 u64 blk, bg_blkno; 1502 u16 bit; 1503 1504 ret = ocfs2_read_block(osb, block, &blk_bh, 1505 OCFS2_BH_CACHED, inode); 1506 if (ret < 0) { 1507 mlog_errno(ret); 1508 goto out; 1509 } 1510 1511 /*Verify the signature of xattr block*/ 1512 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE, 1513 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) { 1514 ret = -EFAULT; 1515 goto out; 1516 } 1517 1518 ret = ocfs2_xattr_block_remove(inode, blk_bh); 1519 if (ret < 0) { 1520 mlog_errno(ret); 1521 goto out; 1522 } 1523 1524 xb = (struct ocfs2_xattr_block *)blk_bh->b_data; 1525 blk = le64_to_cpu(xb->xb_blkno); 1526 bit = le16_to_cpu(xb->xb_suballoc_bit); 1527 bg_blkno = ocfs2_which_suballoc_group(blk, bit); 1528 1529 xb_alloc_inode = ocfs2_get_system_file_inode(osb, 1530 EXTENT_ALLOC_SYSTEM_INODE, 1531 le16_to_cpu(xb->xb_suballoc_slot)); 1532 if (!xb_alloc_inode) { 1533 ret = -ENOMEM; 1534 mlog_errno(ret); 1535 goto out; 1536 } 1537 mutex_lock(&xb_alloc_inode->i_mutex); 1538 1539 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1); 1540 if (ret < 0) { 1541 mlog_errno(ret); 1542 goto out_mutex; 1543 } 1544 1545 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); 1546 if (IS_ERR(handle)) { 1547 ret = PTR_ERR(handle); 1548 mlog_errno(ret); 1549 goto out_unlock; 1550 } 1551 1552 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh, 1553 bit, bg_blkno, 1); 1554 if (ret < 0) 1555 mlog_errno(ret); 1556 1557 ocfs2_commit_trans(osb, handle); 1558 out_unlock: 1559 ocfs2_inode_unlock(xb_alloc_inode, 1); 1560 brelse(xb_alloc_bh); 1561 out_mutex: 1562 mutex_unlock(&xb_alloc_inode->i_mutex); 1563 iput(xb_alloc_inode); 1564 out: 1565 brelse(blk_bh); 1566 return ret; 1567 } 1568 1569 /* 1570 * ocfs2_xattr_remove() 1571 * 1572 * Free extended attribute resources associated with this inode. 1573 */ 1574 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh) 1575 { 1576 struct ocfs2_inode_info *oi = OCFS2_I(inode); 1577 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; 1578 handle_t *handle; 1579 int ret; 1580 1581 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb))) 1582 return 0; 1583 1584 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) 1585 return 0; 1586 1587 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) { 1588 ret = ocfs2_xattr_ibody_remove(inode, di_bh); 1589 if (ret < 0) { 1590 mlog_errno(ret); 1591 goto out; 1592 } 1593 } 1594 1595 if (di->i_xattr_loc) { 1596 ret = ocfs2_xattr_free_block(inode, 1597 le64_to_cpu(di->i_xattr_loc)); 1598 if (ret < 0) { 1599 mlog_errno(ret); 1600 goto out; 1601 } 1602 } 1603 1604 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1605 OCFS2_INODE_UPDATE_CREDITS); 1606 if (IS_ERR(handle)) { 1607 ret = PTR_ERR(handle); 1608 mlog_errno(ret); 1609 goto out; 1610 } 1611 ret = ocfs2_journal_access(handle, inode, di_bh, 1612 OCFS2_JOURNAL_ACCESS_WRITE); 1613 if (ret) { 1614 mlog_errno(ret); 1615 goto out_commit; 1616 } 1617 1618 di->i_xattr_loc = 0; 1619 1620 spin_lock(&oi->ip_lock); 1621 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL); 1622 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); 1623 spin_unlock(&oi->ip_lock); 1624 1625 ret = ocfs2_journal_dirty(handle, di_bh); 1626 if (ret < 0) 1627 mlog_errno(ret); 1628 out_commit: 1629 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); 1630 out: 1631 return ret; 1632 } 1633 1634 static int ocfs2_xattr_has_space_inline(struct inode *inode, 1635 struct ocfs2_dinode *di) 1636 { 1637 struct ocfs2_inode_info *oi = OCFS2_I(inode); 1638 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size; 1639 int free; 1640 1641 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE) 1642 return 0; 1643 1644 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { 1645 struct ocfs2_inline_data *idata = &di->id2.i_data; 1646 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size); 1647 } else if (ocfs2_inode_is_fast_symlink(inode)) { 1648 free = ocfs2_fast_symlink_chars(inode->i_sb) - 1649 le64_to_cpu(di->i_size); 1650 } else { 1651 struct ocfs2_extent_list *el = &di->id2.i_list; 1652 free = (le16_to_cpu(el->l_count) - 1653 le16_to_cpu(el->l_next_free_rec)) * 1654 sizeof(struct ocfs2_extent_rec); 1655 } 1656 if (free >= xattrsize) 1657 return 1; 1658 1659 return 0; 1660 } 1661 1662 /* 1663 * ocfs2_xattr_ibody_find() 1664 * 1665 * Find extended attribute in inode block and 1666 * fill search info into struct ocfs2_xattr_search. 1667 */ 1668 static int ocfs2_xattr_ibody_find(struct inode *inode, 1669 int name_index, 1670 const char *name, 1671 struct ocfs2_xattr_search *xs) 1672 { 1673 struct ocfs2_inode_info *oi = OCFS2_I(inode); 1674 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data; 1675 int ret; 1676 int has_space = 0; 1677 1678 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) 1679 return 0; 1680 1681 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) { 1682 down_read(&oi->ip_alloc_sem); 1683 has_space = ocfs2_xattr_has_space_inline(inode, di); 1684 up_read(&oi->ip_alloc_sem); 1685 if (!has_space) 1686 return 0; 1687 } 1688 1689 xs->xattr_bh = xs->inode_bh; 1690 xs->end = (void *)di + inode->i_sb->s_blocksize; 1691 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) 1692 xs->header = (struct ocfs2_xattr_header *) 1693 (xs->end - le16_to_cpu(di->i_xattr_inline_size)); 1694 else 1695 xs->header = (struct ocfs2_xattr_header *) 1696 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size); 1697 xs->base = (void *)xs->header; 1698 xs->here = xs->header->xh_entries; 1699 1700 /* Find the named attribute. */ 1701 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) { 1702 ret = ocfs2_xattr_find_entry(name_index, name, xs); 1703 if (ret && ret != -ENODATA) 1704 return ret; 1705 xs->not_found = ret; 1706 } 1707 1708 return 0; 1709 } 1710 1711 /* 1712 * ocfs2_xattr_ibody_set() 1713 * 1714 * Set, replace or remove an extended attribute into inode block. 1715 * 1716 */ 1717 static int ocfs2_xattr_ibody_set(struct inode *inode, 1718 struct ocfs2_xattr_info *xi, 1719 struct ocfs2_xattr_search *xs) 1720 { 1721 struct ocfs2_inode_info *oi = OCFS2_I(inode); 1722 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data; 1723 int ret; 1724 1725 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) 1726 return -ENOSPC; 1727 1728 down_write(&oi->ip_alloc_sem); 1729 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) { 1730 if (!ocfs2_xattr_has_space_inline(inode, di)) { 1731 ret = -ENOSPC; 1732 goto out; 1733 } 1734 } 1735 1736 ret = ocfs2_xattr_set_entry(inode, xi, xs, 1737 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL)); 1738 out: 1739 up_write(&oi->ip_alloc_sem); 1740 1741 return ret; 1742 } 1743 1744 /* 1745 * ocfs2_xattr_block_find() 1746 * 1747 * Find extended attribute in external block and 1748 * fill search info into struct ocfs2_xattr_search. 1749 */ 1750 static int ocfs2_xattr_block_find(struct inode *inode, 1751 int name_index, 1752 const char *name, 1753 struct ocfs2_xattr_search *xs) 1754 { 1755 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data; 1756 struct buffer_head *blk_bh = NULL; 1757 struct ocfs2_xattr_block *xb; 1758 int ret = 0; 1759 1760 if (!di->i_xattr_loc) 1761 return ret; 1762 1763 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), 1764 le64_to_cpu(di->i_xattr_loc), 1765 &blk_bh, OCFS2_BH_CACHED, inode); 1766 if (ret < 0) { 1767 mlog_errno(ret); 1768 return ret; 1769 } 1770 /*Verify the signature of xattr block*/ 1771 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE, 1772 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) { 1773 ret = -EFAULT; 1774 goto cleanup; 1775 } 1776 1777 xs->xattr_bh = blk_bh; 1778 xb = (struct ocfs2_xattr_block *)blk_bh->b_data; 1779 1780 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) { 1781 xs->header = &xb->xb_attrs.xb_header; 1782 xs->base = (void *)xs->header; 1783 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size; 1784 xs->here = xs->header->xh_entries; 1785 1786 ret = ocfs2_xattr_find_entry(name_index, name, xs); 1787 } else 1788 ret = ocfs2_xattr_index_block_find(inode, blk_bh, 1789 name_index, 1790 name, xs); 1791 1792 if (ret && ret != -ENODATA) { 1793 xs->xattr_bh = NULL; 1794 goto cleanup; 1795 } 1796 xs->not_found = ret; 1797 return 0; 1798 cleanup: 1799 brelse(blk_bh); 1800 1801 return ret; 1802 } 1803 1804 /* 1805 * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree 1806 * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header 1807 * re-initialized. 1808 */ 1809 static int ocfs2_restore_xattr_block(struct inode *inode, 1810 struct ocfs2_xattr_search *xs) 1811 { 1812 int ret; 1813 handle_t *handle; 1814 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1815 struct ocfs2_xattr_block *xb = 1816 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data; 1817 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list; 1818 u16 xb_flags = le16_to_cpu(xb->xb_flags); 1819 1820 BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) || 1821 le16_to_cpu(el->l_next_free_rec) != 0); 1822 1823 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS); 1824 if (IS_ERR(handle)) { 1825 ret = PTR_ERR(handle); 1826 handle = NULL; 1827 goto out; 1828 } 1829 1830 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh, 1831 OCFS2_JOURNAL_ACCESS_WRITE); 1832 if (ret < 0) { 1833 mlog_errno(ret); 1834 goto out_commit; 1835 } 1836 1837 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize - 1838 offsetof(struct ocfs2_xattr_block, xb_attrs)); 1839 1840 xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED); 1841 1842 ocfs2_journal_dirty(handle, xs->xattr_bh); 1843 1844 out_commit: 1845 ocfs2_commit_trans(osb, handle); 1846 out: 1847 return ret; 1848 } 1849 1850 /* 1851 * ocfs2_xattr_block_set() 1852 * 1853 * Set, replace or remove an extended attribute into external block. 1854 * 1855 */ 1856 static int ocfs2_xattr_block_set(struct inode *inode, 1857 struct ocfs2_xattr_info *xi, 1858 struct ocfs2_xattr_search *xs) 1859 { 1860 struct buffer_head *new_bh = NULL; 1861 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1862 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data; 1863 struct ocfs2_alloc_context *meta_ac = NULL; 1864 handle_t *handle = NULL; 1865 struct ocfs2_xattr_block *xblk = NULL; 1866 u16 suballoc_bit_start; 1867 u32 num_got; 1868 u64 first_blkno; 1869 int ret; 1870 1871 if (!xs->xattr_bh) { 1872 /* 1873 * Alloc one external block for extended attribute 1874 * outside of inode. 1875 */ 1876 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac); 1877 if (ret < 0) { 1878 mlog_errno(ret); 1879 goto out; 1880 } 1881 handle = ocfs2_start_trans(osb, 1882 OCFS2_XATTR_BLOCK_CREATE_CREDITS); 1883 if (IS_ERR(handle)) { 1884 ret = PTR_ERR(handle); 1885 mlog_errno(ret); 1886 goto out; 1887 } 1888 ret = ocfs2_journal_access(handle, inode, xs->inode_bh, 1889 OCFS2_JOURNAL_ACCESS_CREATE); 1890 if (ret < 0) { 1891 mlog_errno(ret); 1892 goto out_commit; 1893 } 1894 1895 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1, 1896 &suballoc_bit_start, &num_got, 1897 &first_blkno); 1898 if (ret < 0) { 1899 mlog_errno(ret); 1900 goto out_commit; 1901 } 1902 1903 new_bh = sb_getblk(inode->i_sb, first_blkno); 1904 ocfs2_set_new_buffer_uptodate(inode, new_bh); 1905 1906 ret = ocfs2_journal_access(handle, inode, new_bh, 1907 OCFS2_JOURNAL_ACCESS_CREATE); 1908 if (ret < 0) { 1909 mlog_errno(ret); 1910 goto out_commit; 1911 } 1912 1913 /* Initialize ocfs2_xattr_block */ 1914 xs->xattr_bh = new_bh; 1915 xblk = (struct ocfs2_xattr_block *)new_bh->b_data; 1916 memset(xblk, 0, inode->i_sb->s_blocksize); 1917 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE); 1918 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num); 1919 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start); 1920 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation); 1921 xblk->xb_blkno = cpu_to_le64(first_blkno); 1922 1923 xs->header = &xblk->xb_attrs.xb_header; 1924 xs->base = (void *)xs->header; 1925 xs->end = (void *)xblk + inode->i_sb->s_blocksize; 1926 xs->here = xs->header->xh_entries; 1927 1928 1929 ret = ocfs2_journal_dirty(handle, new_bh); 1930 if (ret < 0) { 1931 mlog_errno(ret); 1932 goto out_commit; 1933 } 1934 di->i_xattr_loc = cpu_to_le64(first_blkno); 1935 ret = ocfs2_journal_dirty(handle, xs->inode_bh); 1936 if (ret < 0) 1937 mlog_errno(ret); 1938 out_commit: 1939 ocfs2_commit_trans(osb, handle); 1940 out: 1941 if (meta_ac) 1942 ocfs2_free_alloc_context(meta_ac); 1943 if (ret < 0) 1944 return ret; 1945 } else 1946 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data; 1947 1948 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) { 1949 /* Set extended attribute into external block */ 1950 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL); 1951 if (!ret || ret != -ENOSPC) 1952 goto end; 1953 1954 ret = ocfs2_xattr_create_index_block(inode, xs); 1955 if (ret) 1956 goto end; 1957 } 1958 1959 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs); 1960 if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0) 1961 ret = ocfs2_restore_xattr_block(inode, xs); 1962 1963 end: 1964 1965 return ret; 1966 } 1967 1968 /* 1969 * ocfs2_xattr_set() 1970 * 1971 * Set, replace or remove an extended attribute for this inode. 1972 * value is NULL to remove an existing extended attribute, else either 1973 * create or replace an extended attribute. 1974 */ 1975 int ocfs2_xattr_set(struct inode *inode, 1976 int name_index, 1977 const char *name, 1978 const void *value, 1979 size_t value_len, 1980 int flags) 1981 { 1982 struct buffer_head *di_bh = NULL; 1983 struct ocfs2_dinode *di; 1984 int ret; 1985 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 1986 1987 struct ocfs2_xattr_info xi = { 1988 .name_index = name_index, 1989 .name = name, 1990 .value = value, 1991 .value_len = value_len, 1992 }; 1993 1994 struct ocfs2_xattr_search xis = { 1995 .not_found = -ENODATA, 1996 }; 1997 1998 struct ocfs2_xattr_search xbs = { 1999 .not_found = -ENODATA, 2000 }; 2001 2002 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb))) 2003 return -EOPNOTSUPP; 2004 2005 ret = ocfs2_inode_lock(inode, &di_bh, 1); 2006 if (ret < 0) { 2007 mlog_errno(ret); 2008 return ret; 2009 } 2010 xis.inode_bh = xbs.inode_bh = di_bh; 2011 di = (struct ocfs2_dinode *)di_bh->b_data; 2012 2013 down_write(&OCFS2_I(inode)->ip_xattr_sem); 2014 /* 2015 * Scan inode and external block to find the same name 2016 * extended attribute and collect search infomation. 2017 */ 2018 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis); 2019 if (ret) 2020 goto cleanup; 2021 if (xis.not_found) { 2022 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs); 2023 if (ret) 2024 goto cleanup; 2025 } 2026 2027 if (xis.not_found && xbs.not_found) { 2028 ret = -ENODATA; 2029 if (flags & XATTR_REPLACE) 2030 goto cleanup; 2031 ret = 0; 2032 if (!value) 2033 goto cleanup; 2034 } else { 2035 ret = -EEXIST; 2036 if (flags & XATTR_CREATE) 2037 goto cleanup; 2038 } 2039 2040 if (!value) { 2041 /* Remove existing extended attribute */ 2042 if (!xis.not_found) 2043 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis); 2044 else if (!xbs.not_found) 2045 ret = ocfs2_xattr_block_set(inode, &xi, &xbs); 2046 } else { 2047 /* We always try to set extended attribute into inode first*/ 2048 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis); 2049 if (!ret && !xbs.not_found) { 2050 /* 2051 * If succeed and that extended attribute existing in 2052 * external block, then we will remove it. 2053 */ 2054 xi.value = NULL; 2055 xi.value_len = 0; 2056 ret = ocfs2_xattr_block_set(inode, &xi, &xbs); 2057 } else if (ret == -ENOSPC) { 2058 if (di->i_xattr_loc && !xbs.xattr_bh) { 2059 ret = ocfs2_xattr_block_find(inode, name_index, 2060 name, &xbs); 2061 if (ret) 2062 goto cleanup; 2063 } 2064 /* 2065 * If no space in inode, we will set extended attribute 2066 * into external block. 2067 */ 2068 ret = ocfs2_xattr_block_set(inode, &xi, &xbs); 2069 if (ret) 2070 goto cleanup; 2071 if (!xis.not_found) { 2072 /* 2073 * If succeed and that extended attribute 2074 * existing in inode, we will remove it. 2075 */ 2076 xi.value = NULL; 2077 xi.value_len = 0; 2078 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis); 2079 } 2080 } 2081 } 2082 cleanup: 2083 up_write(&OCFS2_I(inode)->ip_xattr_sem); 2084 ocfs2_inode_unlock(inode, 1); 2085 brelse(di_bh); 2086 brelse(xbs.xattr_bh); 2087 for (i = 0; i < blk_per_bucket; i++) 2088 brelse(xbs.bucket.bhs[i]); 2089 2090 return ret; 2091 } 2092 2093 static inline u32 ocfs2_xattr_hash_by_name(struct inode *inode, 2094 int name_index, 2095 const char *suffix_name) 2096 { 2097 struct xattr_handler *handler = ocfs2_xattr_handler(name_index); 2098 char *prefix = handler->prefix; 2099 int prefix_len = strlen(handler->prefix); 2100 2101 return ocfs2_xattr_name_hash(inode, prefix, prefix_len, 2102 (char *)suffix_name, strlen(suffix_name)); 2103 } 2104 2105 /* 2106 * Find the xattr extent rec which may contains name_hash. 2107 * e_cpos will be the first name hash of the xattr rec. 2108 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list. 2109 */ 2110 static int ocfs2_xattr_get_rec(struct inode *inode, 2111 u32 name_hash, 2112 u64 *p_blkno, 2113 u32 *e_cpos, 2114 u32 *num_clusters, 2115 struct ocfs2_extent_list *el) 2116 { 2117 int ret = 0, i; 2118 struct buffer_head *eb_bh = NULL; 2119 struct ocfs2_extent_block *eb; 2120 struct ocfs2_extent_rec *rec = NULL; 2121 u64 e_blkno = 0; 2122 2123 if (el->l_tree_depth) { 2124 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh); 2125 if (ret) { 2126 mlog_errno(ret); 2127 goto out; 2128 } 2129 2130 eb = (struct ocfs2_extent_block *) eb_bh->b_data; 2131 el = &eb->h_list; 2132 2133 if (el->l_tree_depth) { 2134 ocfs2_error(inode->i_sb, 2135 "Inode %lu has non zero tree depth in " 2136 "xattr tree block %llu\n", inode->i_ino, 2137 (unsigned long long)eb_bh->b_blocknr); 2138 ret = -EROFS; 2139 goto out; 2140 } 2141 } 2142 2143 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) { 2144 rec = &el->l_recs[i]; 2145 2146 if (le32_to_cpu(rec->e_cpos) <= name_hash) { 2147 e_blkno = le64_to_cpu(rec->e_blkno); 2148 break; 2149 } 2150 } 2151 2152 if (!e_blkno) { 2153 ocfs2_error(inode->i_sb, "Inode %lu has bad extent " 2154 "record (%u, %u, 0) in xattr", inode->i_ino, 2155 le32_to_cpu(rec->e_cpos), 2156 ocfs2_rec_clusters(el, rec)); 2157 ret = -EROFS; 2158 goto out; 2159 } 2160 2161 *p_blkno = le64_to_cpu(rec->e_blkno); 2162 *num_clusters = le16_to_cpu(rec->e_leaf_clusters); 2163 if (e_cpos) 2164 *e_cpos = le32_to_cpu(rec->e_cpos); 2165 out: 2166 brelse(eb_bh); 2167 return ret; 2168 } 2169 2170 typedef int (xattr_bucket_func)(struct inode *inode, 2171 struct ocfs2_xattr_bucket *bucket, 2172 void *para); 2173 2174 static int ocfs2_find_xe_in_bucket(struct inode *inode, 2175 struct buffer_head *header_bh, 2176 int name_index, 2177 const char *name, 2178 u32 name_hash, 2179 u16 *xe_index, 2180 int *found) 2181 { 2182 int i, ret = 0, cmp = 1, block_off, new_offset; 2183 struct ocfs2_xattr_header *xh = 2184 (struct ocfs2_xattr_header *)header_bh->b_data; 2185 size_t name_len = strlen(name); 2186 struct ocfs2_xattr_entry *xe = NULL; 2187 struct buffer_head *name_bh = NULL; 2188 char *xe_name; 2189 2190 /* 2191 * We don't use binary search in the bucket because there 2192 * may be multiple entries with the same name hash. 2193 */ 2194 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) { 2195 xe = &xh->xh_entries[i]; 2196 2197 if (name_hash > le32_to_cpu(xe->xe_name_hash)) 2198 continue; 2199 else if (name_hash < le32_to_cpu(xe->xe_name_hash)) 2200 break; 2201 2202 cmp = name_index - ocfs2_xattr_get_type(xe); 2203 if (!cmp) 2204 cmp = name_len - xe->xe_name_len; 2205 if (cmp) 2206 continue; 2207 2208 ret = ocfs2_xattr_bucket_get_name_value(inode, 2209 xh, 2210 i, 2211 &block_off, 2212 &new_offset); 2213 if (ret) { 2214 mlog_errno(ret); 2215 break; 2216 } 2217 2218 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), 2219 header_bh->b_blocknr + block_off, 2220 &name_bh, OCFS2_BH_CACHED, inode); 2221 if (ret) { 2222 mlog_errno(ret); 2223 break; 2224 } 2225 xe_name = name_bh->b_data + new_offset; 2226 2227 cmp = memcmp(name, xe_name, name_len); 2228 brelse(name_bh); 2229 name_bh = NULL; 2230 2231 if (cmp == 0) { 2232 *xe_index = i; 2233 *found = 1; 2234 ret = 0; 2235 break; 2236 } 2237 } 2238 2239 return ret; 2240 } 2241 2242 /* 2243 * Find the specified xattr entry in a series of buckets. 2244 * This series start from p_blkno and last for num_clusters. 2245 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains 2246 * the num of the valid buckets. 2247 * 2248 * Return the buffer_head this xattr should reside in. And if the xattr's 2249 * hash is in the gap of 2 buckets, return the lower bucket. 2250 */ 2251 static int ocfs2_xattr_bucket_find(struct inode *inode, 2252 int name_index, 2253 const char *name, 2254 u32 name_hash, 2255 u64 p_blkno, 2256 u32 first_hash, 2257 u32 num_clusters, 2258 struct ocfs2_xattr_search *xs) 2259 { 2260 int ret, found = 0; 2261 struct buffer_head *bh = NULL; 2262 struct buffer_head *lower_bh = NULL; 2263 struct ocfs2_xattr_header *xh = NULL; 2264 struct ocfs2_xattr_entry *xe = NULL; 2265 u16 index = 0; 2266 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 2267 int low_bucket = 0, bucket, high_bucket; 2268 u32 last_hash; 2269 u64 blkno; 2270 2271 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, 2272 &bh, OCFS2_BH_CACHED, inode); 2273 if (ret) { 2274 mlog_errno(ret); 2275 goto out; 2276 } 2277 2278 xh = (struct ocfs2_xattr_header *)bh->b_data; 2279 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1; 2280 2281 while (low_bucket <= high_bucket) { 2282 brelse(bh); 2283 bh = NULL; 2284 bucket = (low_bucket + high_bucket) / 2; 2285 2286 blkno = p_blkno + bucket * blk_per_bucket; 2287 2288 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno, 2289 &bh, OCFS2_BH_CACHED, inode); 2290 if (ret) { 2291 mlog_errno(ret); 2292 goto out; 2293 } 2294 2295 xh = (struct ocfs2_xattr_header *)bh->b_data; 2296 xe = &xh->xh_entries[0]; 2297 if (name_hash < le32_to_cpu(xe->xe_name_hash)) { 2298 high_bucket = bucket - 1; 2299 continue; 2300 } 2301 2302 /* 2303 * Check whether the hash of the last entry in our 2304 * bucket is larger than the search one. for an empty 2305 * bucket, the last one is also the first one. 2306 */ 2307 if (xh->xh_count) 2308 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1]; 2309 2310 last_hash = le32_to_cpu(xe->xe_name_hash); 2311 2312 /* record lower_bh which may be the insert place. */ 2313 brelse(lower_bh); 2314 lower_bh = bh; 2315 bh = NULL; 2316 2317 if (name_hash > le32_to_cpu(xe->xe_name_hash)) { 2318 low_bucket = bucket + 1; 2319 continue; 2320 } 2321 2322 /* the searched xattr should reside in this bucket if exists. */ 2323 ret = ocfs2_find_xe_in_bucket(inode, lower_bh, 2324 name_index, name, name_hash, 2325 &index, &found); 2326 if (ret) { 2327 mlog_errno(ret); 2328 goto out; 2329 } 2330 break; 2331 } 2332 2333 /* 2334 * Record the bucket we have found. 2335 * When the xattr's hash value is in the gap of 2 buckets, we will 2336 * always set it to the previous bucket. 2337 */ 2338 if (!lower_bh) { 2339 /* 2340 * We can't find any bucket whose first name_hash is less 2341 * than the find name_hash. 2342 */ 2343 BUG_ON(bh->b_blocknr != p_blkno); 2344 lower_bh = bh; 2345 bh = NULL; 2346 } 2347 xs->bucket.bhs[0] = lower_bh; 2348 xs->bucket.xh = (struct ocfs2_xattr_header *) 2349 xs->bucket.bhs[0]->b_data; 2350 lower_bh = NULL; 2351 2352 xs->header = xs->bucket.xh; 2353 xs->base = xs->bucket.bhs[0]->b_data; 2354 xs->end = xs->base + inode->i_sb->s_blocksize; 2355 2356 if (found) { 2357 /* 2358 * If we have found the xattr enty, read all the blocks in 2359 * this bucket. 2360 */ 2361 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 2362 xs->bucket.bhs[0]->b_blocknr + 1, 2363 blk_per_bucket - 1, &xs->bucket.bhs[1], 2364 OCFS2_BH_CACHED, inode); 2365 if (ret) { 2366 mlog_errno(ret); 2367 goto out; 2368 } 2369 2370 xs->here = &xs->header->xh_entries[index]; 2371 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name, 2372 (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index); 2373 } else 2374 ret = -ENODATA; 2375 2376 out: 2377 brelse(bh); 2378 brelse(lower_bh); 2379 return ret; 2380 } 2381 2382 static int ocfs2_xattr_index_block_find(struct inode *inode, 2383 struct buffer_head *root_bh, 2384 int name_index, 2385 const char *name, 2386 struct ocfs2_xattr_search *xs) 2387 { 2388 int ret; 2389 struct ocfs2_xattr_block *xb = 2390 (struct ocfs2_xattr_block *)root_bh->b_data; 2391 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root; 2392 struct ocfs2_extent_list *el = &xb_root->xt_list; 2393 u64 p_blkno = 0; 2394 u32 first_hash, num_clusters = 0; 2395 u32 name_hash = ocfs2_xattr_hash_by_name(inode, name_index, name); 2396 2397 if (le16_to_cpu(el->l_next_free_rec) == 0) 2398 return -ENODATA; 2399 2400 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n", 2401 name, name_hash, name_index); 2402 2403 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash, 2404 &num_clusters, el); 2405 if (ret) { 2406 mlog_errno(ret); 2407 goto out; 2408 } 2409 2410 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash); 2411 2412 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash " 2413 "in the rec is %u\n", num_clusters, p_blkno, first_hash); 2414 2415 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash, 2416 p_blkno, first_hash, num_clusters, xs); 2417 2418 out: 2419 return ret; 2420 } 2421 2422 static int ocfs2_iterate_xattr_buckets(struct inode *inode, 2423 u64 blkno, 2424 u32 clusters, 2425 xattr_bucket_func *func, 2426 void *para) 2427 { 2428 int i, j, ret = 0; 2429 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 2430 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)); 2431 u32 num_buckets = clusters * bpc; 2432 struct ocfs2_xattr_bucket bucket; 2433 2434 memset(&bucket, 0, sizeof(bucket)); 2435 2436 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n", 2437 clusters, blkno); 2438 2439 for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) { 2440 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 2441 blkno, blk_per_bucket, 2442 bucket.bhs, OCFS2_BH_CACHED, inode); 2443 if (ret) { 2444 mlog_errno(ret); 2445 goto out; 2446 } 2447 2448 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data; 2449 /* 2450 * The real bucket num in this series of blocks is stored 2451 * in the 1st bucket. 2452 */ 2453 if (i == 0) 2454 num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets); 2455 2456 mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno, 2457 le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash)); 2458 if (func) { 2459 ret = func(inode, &bucket, para); 2460 if (ret) { 2461 mlog_errno(ret); 2462 break; 2463 } 2464 } 2465 2466 for (j = 0; j < blk_per_bucket; j++) 2467 brelse(bucket.bhs[j]); 2468 memset(&bucket, 0, sizeof(bucket)); 2469 } 2470 2471 out: 2472 for (j = 0; j < blk_per_bucket; j++) 2473 brelse(bucket.bhs[j]); 2474 2475 return ret; 2476 } 2477 2478 struct ocfs2_xattr_tree_list { 2479 char *buffer; 2480 size_t buffer_size; 2481 }; 2482 2483 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode, 2484 struct ocfs2_xattr_header *xh, 2485 int index, 2486 int *block_off, 2487 int *new_offset) 2488 { 2489 u16 name_offset; 2490 2491 if (index < 0 || index >= le16_to_cpu(xh->xh_count)) 2492 return -EINVAL; 2493 2494 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset); 2495 2496 *block_off = name_offset >> inode->i_sb->s_blocksize_bits; 2497 *new_offset = name_offset % inode->i_sb->s_blocksize; 2498 2499 return 0; 2500 } 2501 2502 static int ocfs2_list_xattr_bucket(struct inode *inode, 2503 struct ocfs2_xattr_bucket *bucket, 2504 void *para) 2505 { 2506 int ret = 0; 2507 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para; 2508 size_t size; 2509 int i, block_off, new_offset; 2510 2511 for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) { 2512 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i]; 2513 struct xattr_handler *handler = 2514 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry)); 2515 2516 if (handler) { 2517 ret = ocfs2_xattr_bucket_get_name_value(inode, 2518 bucket->xh, 2519 i, 2520 &block_off, 2521 &new_offset); 2522 if (ret) 2523 break; 2524 size = handler->list(inode, xl->buffer, xl->buffer_size, 2525 bucket->bhs[block_off]->b_data + 2526 new_offset, 2527 entry->xe_name_len); 2528 if (xl->buffer) { 2529 if (size > xl->buffer_size) 2530 return -ERANGE; 2531 xl->buffer += size; 2532 } 2533 xl->buffer_size -= size; 2534 } 2535 } 2536 2537 return ret; 2538 } 2539 2540 static int ocfs2_xattr_tree_list_index_block(struct inode *inode, 2541 struct ocfs2_xattr_tree_root *xt, 2542 char *buffer, 2543 size_t buffer_size) 2544 { 2545 struct ocfs2_extent_list *el = &xt->xt_list; 2546 int ret = 0; 2547 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0; 2548 u64 p_blkno = 0; 2549 struct ocfs2_xattr_tree_list xl = { 2550 .buffer = buffer, 2551 .buffer_size = buffer_size, 2552 }; 2553 2554 if (le16_to_cpu(el->l_next_free_rec) == 0) 2555 return 0; 2556 2557 while (name_hash > 0) { 2558 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, 2559 &e_cpos, &num_clusters, el); 2560 if (ret) { 2561 mlog_errno(ret); 2562 goto out; 2563 } 2564 2565 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters, 2566 ocfs2_list_xattr_bucket, 2567 &xl); 2568 if (ret) { 2569 mlog_errno(ret); 2570 goto out; 2571 } 2572 2573 if (e_cpos == 0) 2574 break; 2575 2576 name_hash = e_cpos - 1; 2577 } 2578 2579 ret = buffer_size - xl.buffer_size; 2580 out: 2581 return ret; 2582 } 2583 2584 static int cmp_xe(const void *a, const void *b) 2585 { 2586 const struct ocfs2_xattr_entry *l = a, *r = b; 2587 u32 l_hash = le32_to_cpu(l->xe_name_hash); 2588 u32 r_hash = le32_to_cpu(r->xe_name_hash); 2589 2590 if (l_hash > r_hash) 2591 return 1; 2592 if (l_hash < r_hash) 2593 return -1; 2594 return 0; 2595 } 2596 2597 static void swap_xe(void *a, void *b, int size) 2598 { 2599 struct ocfs2_xattr_entry *l = a, *r = b, tmp; 2600 2601 tmp = *l; 2602 memcpy(l, r, sizeof(struct ocfs2_xattr_entry)); 2603 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry)); 2604 } 2605 2606 /* 2607 * When the ocfs2_xattr_block is filled up, new bucket will be created 2608 * and all the xattr entries will be moved to the new bucket. 2609 * Note: we need to sort the entries since they are not saved in order 2610 * in the ocfs2_xattr_block. 2611 */ 2612 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode, 2613 struct buffer_head *xb_bh, 2614 struct buffer_head *xh_bh, 2615 struct buffer_head *data_bh) 2616 { 2617 int i, blocksize = inode->i_sb->s_blocksize; 2618 u16 offset, size, off_change; 2619 struct ocfs2_xattr_entry *xe; 2620 struct ocfs2_xattr_block *xb = 2621 (struct ocfs2_xattr_block *)xb_bh->b_data; 2622 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header; 2623 struct ocfs2_xattr_header *xh = 2624 (struct ocfs2_xattr_header *)xh_bh->b_data; 2625 u16 count = le16_to_cpu(xb_xh->xh_count); 2626 char *target = xh_bh->b_data, *src = xb_bh->b_data; 2627 2628 mlog(0, "cp xattr from block %llu to bucket %llu\n", 2629 (unsigned long long)xb_bh->b_blocknr, 2630 (unsigned long long)xh_bh->b_blocknr); 2631 2632 memset(xh_bh->b_data, 0, blocksize); 2633 if (data_bh) 2634 memset(data_bh->b_data, 0, blocksize); 2635 /* 2636 * Since the xe_name_offset is based on ocfs2_xattr_header, 2637 * there is a offset change corresponding to the change of 2638 * ocfs2_xattr_header's position. 2639 */ 2640 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header); 2641 xe = &xb_xh->xh_entries[count - 1]; 2642 offset = le16_to_cpu(xe->xe_name_offset) + off_change; 2643 size = blocksize - offset; 2644 2645 /* copy all the names and values. */ 2646 if (data_bh) 2647 target = data_bh->b_data; 2648 memcpy(target + offset, src + offset, size); 2649 2650 /* Init new header now. */ 2651 xh->xh_count = xb_xh->xh_count; 2652 xh->xh_num_buckets = cpu_to_le16(1); 2653 xh->xh_name_value_len = cpu_to_le16(size); 2654 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size); 2655 2656 /* copy all the entries. */ 2657 target = xh_bh->b_data; 2658 offset = offsetof(struct ocfs2_xattr_header, xh_entries); 2659 size = count * sizeof(struct ocfs2_xattr_entry); 2660 memcpy(target + offset, (char *)xb_xh + offset, size); 2661 2662 /* Change the xe offset for all the xe because of the move. */ 2663 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize + 2664 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header); 2665 for (i = 0; i < count; i++) 2666 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change); 2667 2668 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n", 2669 offset, size, off_change); 2670 2671 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry), 2672 cmp_xe, swap_xe); 2673 } 2674 2675 /* 2676 * After we move xattr from block to index btree, we have to 2677 * update ocfs2_xattr_search to the new xe and base. 2678 * 2679 * When the entry is in xattr block, xattr_bh indicates the storage place. 2680 * While if the entry is in index b-tree, "bucket" indicates the 2681 * real place of the xattr. 2682 */ 2683 static int ocfs2_xattr_update_xattr_search(struct inode *inode, 2684 struct ocfs2_xattr_search *xs, 2685 struct buffer_head *old_bh, 2686 struct buffer_head *new_bh) 2687 { 2688 int ret = 0; 2689 char *buf = old_bh->b_data; 2690 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf; 2691 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header; 2692 int i, blocksize = inode->i_sb->s_blocksize; 2693 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 2694 2695 xs->bucket.bhs[0] = new_bh; 2696 get_bh(new_bh); 2697 xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data; 2698 xs->header = xs->bucket.xh; 2699 2700 xs->base = new_bh->b_data; 2701 xs->end = xs->base + inode->i_sb->s_blocksize; 2702 2703 if (!xs->not_found) { 2704 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) { 2705 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 2706 xs->bucket.bhs[0]->b_blocknr + 1, 2707 blk_per_bucket - 1, &xs->bucket.bhs[1], 2708 OCFS2_BH_CACHED, inode); 2709 if (ret) { 2710 mlog_errno(ret); 2711 return ret; 2712 } 2713 2714 i = xs->here - old_xh->xh_entries; 2715 xs->here = &xs->header->xh_entries[i]; 2716 } 2717 } 2718 2719 return ret; 2720 } 2721 2722 static int ocfs2_xattr_create_index_block(struct inode *inode, 2723 struct ocfs2_xattr_search *xs) 2724 { 2725 int ret, credits = OCFS2_SUBALLOC_ALLOC; 2726 u32 bit_off, len; 2727 u64 blkno; 2728 handle_t *handle; 2729 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 2730 struct ocfs2_inode_info *oi = OCFS2_I(inode); 2731 struct ocfs2_alloc_context *data_ac; 2732 struct buffer_head *xh_bh = NULL, *data_bh = NULL; 2733 struct buffer_head *xb_bh = xs->xattr_bh; 2734 struct ocfs2_xattr_block *xb = 2735 (struct ocfs2_xattr_block *)xb_bh->b_data; 2736 struct ocfs2_xattr_tree_root *xr; 2737 u16 xb_flags = le16_to_cpu(xb->xb_flags); 2738 u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 2739 2740 mlog(0, "create xattr index block for %llu\n", 2741 (unsigned long long)xb_bh->b_blocknr); 2742 2743 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED); 2744 2745 ret = ocfs2_reserve_clusters(osb, 1, &data_ac); 2746 if (ret) { 2747 mlog_errno(ret); 2748 goto out; 2749 } 2750 2751 /* 2752 * XXX: 2753 * We can use this lock for now, and maybe move to a dedicated mutex 2754 * if performance becomes a problem later. 2755 */ 2756 down_write(&oi->ip_alloc_sem); 2757 2758 /* 2759 * 3 more credits, one for xattr block update, one for the 1st block 2760 * of the new xattr bucket and one for the value/data. 2761 */ 2762 credits += 3; 2763 handle = ocfs2_start_trans(osb, credits); 2764 if (IS_ERR(handle)) { 2765 ret = PTR_ERR(handle); 2766 mlog_errno(ret); 2767 goto out_sem; 2768 } 2769 2770 ret = ocfs2_journal_access(handle, inode, xb_bh, 2771 OCFS2_JOURNAL_ACCESS_WRITE); 2772 if (ret) { 2773 mlog_errno(ret); 2774 goto out_commit; 2775 } 2776 2777 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len); 2778 if (ret) { 2779 mlog_errno(ret); 2780 goto out_commit; 2781 } 2782 2783 /* 2784 * The bucket may spread in many blocks, and 2785 * we will only touch the 1st block and the last block 2786 * in the whole bucket(one for entry and one for data). 2787 */ 2788 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); 2789 2790 mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno); 2791 2792 xh_bh = sb_getblk(inode->i_sb, blkno); 2793 if (!xh_bh) { 2794 ret = -EIO; 2795 mlog_errno(ret); 2796 goto out_commit; 2797 } 2798 2799 ocfs2_set_new_buffer_uptodate(inode, xh_bh); 2800 2801 ret = ocfs2_journal_access(handle, inode, xh_bh, 2802 OCFS2_JOURNAL_ACCESS_CREATE); 2803 if (ret) { 2804 mlog_errno(ret); 2805 goto out_commit; 2806 } 2807 2808 if (bpb > 1) { 2809 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1); 2810 if (!data_bh) { 2811 ret = -EIO; 2812 mlog_errno(ret); 2813 goto out_commit; 2814 } 2815 2816 ocfs2_set_new_buffer_uptodate(inode, data_bh); 2817 2818 ret = ocfs2_journal_access(handle, inode, data_bh, 2819 OCFS2_JOURNAL_ACCESS_CREATE); 2820 if (ret) { 2821 mlog_errno(ret); 2822 goto out_commit; 2823 } 2824 } 2825 2826 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh); 2827 2828 ocfs2_journal_dirty(handle, xh_bh); 2829 if (data_bh) 2830 ocfs2_journal_dirty(handle, data_bh); 2831 2832 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh); 2833 2834 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */ 2835 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize - 2836 offsetof(struct ocfs2_xattr_block, xb_attrs)); 2837 2838 xr = &xb->xb_attrs.xb_root; 2839 xr->xt_clusters = cpu_to_le32(1); 2840 xr->xt_last_eb_blk = 0; 2841 xr->xt_list.l_tree_depth = 0; 2842 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb)); 2843 xr->xt_list.l_next_free_rec = cpu_to_le16(1); 2844 2845 xr->xt_list.l_recs[0].e_cpos = 0; 2846 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno); 2847 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1); 2848 2849 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED); 2850 2851 ret = ocfs2_journal_dirty(handle, xb_bh); 2852 if (ret) { 2853 mlog_errno(ret); 2854 goto out_commit; 2855 } 2856 2857 out_commit: 2858 ocfs2_commit_trans(osb, handle); 2859 2860 out_sem: 2861 up_write(&oi->ip_alloc_sem); 2862 2863 out: 2864 if (data_ac) 2865 ocfs2_free_alloc_context(data_ac); 2866 2867 brelse(xh_bh); 2868 brelse(data_bh); 2869 2870 return ret; 2871 } 2872 2873 static int cmp_xe_offset(const void *a, const void *b) 2874 { 2875 const struct ocfs2_xattr_entry *l = a, *r = b; 2876 u32 l_name_offset = le16_to_cpu(l->xe_name_offset); 2877 u32 r_name_offset = le16_to_cpu(r->xe_name_offset); 2878 2879 if (l_name_offset < r_name_offset) 2880 return 1; 2881 if (l_name_offset > r_name_offset) 2882 return -1; 2883 return 0; 2884 } 2885 2886 /* 2887 * defrag a xattr bucket if we find that the bucket has some 2888 * holes beteen name/value pairs. 2889 * We will move all the name/value pairs to the end of the bucket 2890 * so that we can spare some space for insertion. 2891 */ 2892 static int ocfs2_defrag_xattr_bucket(struct inode *inode, 2893 struct ocfs2_xattr_bucket *bucket) 2894 { 2895 int ret, i; 2896 size_t end, offset, len, value_len; 2897 struct ocfs2_xattr_header *xh; 2898 char *entries, *buf, *bucket_buf = NULL; 2899 u64 blkno = bucket->bhs[0]->b_blocknr; 2900 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 2901 u16 xh_free_start; 2902 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 2903 size_t blocksize = inode->i_sb->s_blocksize; 2904 handle_t *handle; 2905 struct buffer_head **bhs; 2906 struct ocfs2_xattr_entry *xe; 2907 2908 bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket, 2909 GFP_NOFS); 2910 if (!bhs) 2911 return -ENOMEM; 2912 2913 ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs, 2914 OCFS2_BH_CACHED, inode); 2915 if (ret) 2916 goto out; 2917 2918 /* 2919 * In order to make the operation more efficient and generic, 2920 * we copy all the blocks into a contiguous memory and do the 2921 * defragment there, so if anything is error, we will not touch 2922 * the real block. 2923 */ 2924 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS); 2925 if (!bucket_buf) { 2926 ret = -EIO; 2927 goto out; 2928 } 2929 2930 buf = bucket_buf; 2931 for (i = 0; i < blk_per_bucket; i++, buf += blocksize) 2932 memcpy(buf, bhs[i]->b_data, blocksize); 2933 2934 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket); 2935 if (IS_ERR(handle)) { 2936 ret = PTR_ERR(handle); 2937 handle = NULL; 2938 mlog_errno(ret); 2939 goto out; 2940 } 2941 2942 for (i = 0; i < blk_per_bucket; i++) { 2943 ret = ocfs2_journal_access(handle, inode, bhs[i], 2944 OCFS2_JOURNAL_ACCESS_WRITE); 2945 if (ret < 0) { 2946 mlog_errno(ret); 2947 goto commit; 2948 } 2949 } 2950 2951 xh = (struct ocfs2_xattr_header *)bucket_buf; 2952 entries = (char *)xh->xh_entries; 2953 xh_free_start = le16_to_cpu(xh->xh_free_start); 2954 2955 mlog(0, "adjust xattr bucket in %llu, count = %u, " 2956 "xh_free_start = %u, xh_name_value_len = %u.\n", 2957 blkno, le16_to_cpu(xh->xh_count), xh_free_start, 2958 le16_to_cpu(xh->xh_name_value_len)); 2959 2960 /* 2961 * sort all the entries by their offset. 2962 * the largest will be the first, so that we can 2963 * move them to the end one by one. 2964 */ 2965 sort(entries, le16_to_cpu(xh->xh_count), 2966 sizeof(struct ocfs2_xattr_entry), 2967 cmp_xe_offset, swap_xe); 2968 2969 /* Move all name/values to the end of the bucket. */ 2970 xe = xh->xh_entries; 2971 end = OCFS2_XATTR_BUCKET_SIZE; 2972 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) { 2973 offset = le16_to_cpu(xe->xe_name_offset); 2974 if (ocfs2_xattr_is_local(xe)) 2975 value_len = OCFS2_XATTR_SIZE( 2976 le64_to_cpu(xe->xe_value_size)); 2977 else 2978 value_len = OCFS2_XATTR_ROOT_SIZE; 2979 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len; 2980 2981 /* 2982 * We must make sure that the name/value pair 2983 * exist in the same block. So adjust end to 2984 * the previous block end if needed. 2985 */ 2986 if (((end - len) / blocksize != 2987 (end - 1) / blocksize)) 2988 end = end - end % blocksize; 2989 2990 if (end > offset + len) { 2991 memmove(bucket_buf + end - len, 2992 bucket_buf + offset, len); 2993 xe->xe_name_offset = cpu_to_le16(end - len); 2994 } 2995 2996 mlog_bug_on_msg(end < offset + len, "Defrag check failed for " 2997 "bucket %llu\n", (unsigned long long)blkno); 2998 2999 end -= len; 3000 } 3001 3002 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for " 3003 "bucket %llu\n", (unsigned long long)blkno); 3004 3005 if (xh_free_start == end) 3006 goto commit; 3007 3008 memset(bucket_buf + xh_free_start, 0, end - xh_free_start); 3009 xh->xh_free_start = cpu_to_le16(end); 3010 3011 /* sort the entries by their name_hash. */ 3012 sort(entries, le16_to_cpu(xh->xh_count), 3013 sizeof(struct ocfs2_xattr_entry), 3014 cmp_xe, swap_xe); 3015 3016 buf = bucket_buf; 3017 for (i = 0; i < blk_per_bucket; i++, buf += blocksize) { 3018 memcpy(bhs[i]->b_data, buf, blocksize); 3019 ocfs2_journal_dirty(handle, bhs[i]); 3020 } 3021 3022 commit: 3023 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); 3024 out: 3025 3026 if (bhs) { 3027 for (i = 0; i < blk_per_bucket; i++) 3028 brelse(bhs[i]); 3029 } 3030 kfree(bhs); 3031 3032 kfree(bucket_buf); 3033 return ret; 3034 } 3035 3036 /* 3037 * Move half nums of the xattr bucket in the previous cluster to this new 3038 * cluster. We only touch the last cluster of the previous extend record. 3039 * 3040 * first_bh is the first buffer_head of a series of bucket in the same 3041 * extent rec and header_bh is the header of one bucket in this cluster. 3042 * They will be updated if we move the data header_bh contains to the new 3043 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster. 3044 */ 3045 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode, 3046 handle_t *handle, 3047 struct buffer_head **first_bh, 3048 struct buffer_head **header_bh, 3049 u64 new_blkno, 3050 u64 prev_blkno, 3051 u32 num_clusters, 3052 u32 *first_hash) 3053 { 3054 int i, ret, credits; 3055 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 3056 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); 3057 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb); 3058 int blocksize = inode->i_sb->s_blocksize; 3059 struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL; 3060 struct ocfs2_xattr_header *new_xh; 3061 struct ocfs2_xattr_header *xh = 3062 (struct ocfs2_xattr_header *)((*first_bh)->b_data); 3063 3064 BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets); 3065 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize); 3066 3067 prev_bh = *first_bh; 3068 get_bh(prev_bh); 3069 xh = (struct ocfs2_xattr_header *)prev_bh->b_data; 3070 3071 prev_blkno += (num_clusters - 1) * bpc + bpc / 2; 3072 3073 mlog(0, "move half of xattrs in cluster %llu to %llu\n", 3074 prev_blkno, new_blkno); 3075 3076 /* 3077 * We need to update the 1st half of the new cluster and 3078 * 1 more for the update of the 1st bucket of the previous 3079 * extent record. 3080 */ 3081 credits = bpc / 2 + 1; 3082 ret = ocfs2_extend_trans(handle, credits); 3083 if (ret) { 3084 mlog_errno(ret); 3085 goto out; 3086 } 3087 3088 ret = ocfs2_journal_access(handle, inode, prev_bh, 3089 OCFS2_JOURNAL_ACCESS_WRITE); 3090 if (ret) { 3091 mlog_errno(ret); 3092 goto out; 3093 } 3094 3095 for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) { 3096 old_bh = new_bh = NULL; 3097 new_bh = sb_getblk(inode->i_sb, new_blkno); 3098 if (!new_bh) { 3099 ret = -EIO; 3100 mlog_errno(ret); 3101 goto out; 3102 } 3103 3104 ocfs2_set_new_buffer_uptodate(inode, new_bh); 3105 3106 ret = ocfs2_journal_access(handle, inode, new_bh, 3107 OCFS2_JOURNAL_ACCESS_CREATE); 3108 if (ret < 0) { 3109 mlog_errno(ret); 3110 brelse(new_bh); 3111 goto out; 3112 } 3113 3114 ret = ocfs2_read_block(osb, prev_blkno, 3115 &old_bh, OCFS2_BH_CACHED, inode); 3116 if (ret < 0) { 3117 mlog_errno(ret); 3118 brelse(new_bh); 3119 goto out; 3120 } 3121 3122 memcpy(new_bh->b_data, old_bh->b_data, blocksize); 3123 3124 if (i == 0) { 3125 new_xh = (struct ocfs2_xattr_header *)new_bh->b_data; 3126 new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2); 3127 3128 if (first_hash) 3129 *first_hash = le32_to_cpu( 3130 new_xh->xh_entries[0].xe_name_hash); 3131 new_first_bh = new_bh; 3132 get_bh(new_first_bh); 3133 } 3134 3135 ocfs2_journal_dirty(handle, new_bh); 3136 3137 if (*header_bh == old_bh) { 3138 brelse(*header_bh); 3139 *header_bh = new_bh; 3140 get_bh(*header_bh); 3141 3142 brelse(*first_bh); 3143 *first_bh = new_first_bh; 3144 get_bh(*first_bh); 3145 } 3146 brelse(new_bh); 3147 brelse(old_bh); 3148 } 3149 3150 le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2)); 3151 3152 ocfs2_journal_dirty(handle, prev_bh); 3153 out: 3154 brelse(prev_bh); 3155 brelse(new_first_bh); 3156 return ret; 3157 } 3158 3159 static int ocfs2_read_xattr_bucket(struct inode *inode, 3160 u64 blkno, 3161 struct buffer_head **bhs, 3162 int new) 3163 { 3164 int ret = 0; 3165 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 3166 3167 if (!new) 3168 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno, 3169 blk_per_bucket, bhs, 3170 OCFS2_BH_CACHED, inode); 3171 3172 for (i = 0; i < blk_per_bucket; i++) { 3173 bhs[i] = sb_getblk(inode->i_sb, blkno + i); 3174 if (bhs[i] == NULL) { 3175 ret = -EIO; 3176 mlog_errno(ret); 3177 break; 3178 } 3179 ocfs2_set_new_buffer_uptodate(inode, bhs[i]); 3180 } 3181 3182 return ret; 3183 } 3184 3185 /* 3186 * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk). 3187 * first_hash will record the 1st hash of the new bucket. 3188 */ 3189 static int ocfs2_half_xattr_bucket(struct inode *inode, 3190 handle_t *handle, 3191 u64 blk, 3192 u64 new_blk, 3193 u32 *first_hash, 3194 int new_bucket_head) 3195 { 3196 int ret, i; 3197 u16 count, start, len, name_value_len, xe_len, name_offset; 3198 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 3199 struct buffer_head **s_bhs, **t_bhs = NULL; 3200 struct ocfs2_xattr_header *xh; 3201 struct ocfs2_xattr_entry *xe; 3202 int blocksize = inode->i_sb->s_blocksize; 3203 3204 mlog(0, "move half of xattrs from bucket %llu to %llu\n", 3205 blk, new_blk); 3206 3207 s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS); 3208 if (!s_bhs) 3209 return -ENOMEM; 3210 3211 ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0); 3212 if (ret) { 3213 mlog_errno(ret); 3214 goto out; 3215 } 3216 3217 ret = ocfs2_journal_access(handle, inode, s_bhs[0], 3218 OCFS2_JOURNAL_ACCESS_WRITE); 3219 if (ret) { 3220 mlog_errno(ret); 3221 goto out; 3222 } 3223 3224 t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS); 3225 if (!t_bhs) { 3226 ret = -ENOMEM; 3227 goto out; 3228 } 3229 3230 ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head); 3231 if (ret) { 3232 mlog_errno(ret); 3233 goto out; 3234 } 3235 3236 for (i = 0; i < blk_per_bucket; i++) { 3237 ret = ocfs2_journal_access(handle, inode, t_bhs[i], 3238 OCFS2_JOURNAL_ACCESS_CREATE); 3239 if (ret) { 3240 mlog_errno(ret); 3241 goto out; 3242 } 3243 } 3244 3245 /* copy the whole bucket to the new first. */ 3246 for (i = 0; i < blk_per_bucket; i++) 3247 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize); 3248 3249 /* update the new bucket. */ 3250 xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data; 3251 count = le16_to_cpu(xh->xh_count); 3252 start = count / 2; 3253 3254 /* 3255 * Calculate the total name/value len and xh_free_start for 3256 * the old bucket first. 3257 */ 3258 name_offset = OCFS2_XATTR_BUCKET_SIZE; 3259 name_value_len = 0; 3260 for (i = 0; i < start; i++) { 3261 xe = &xh->xh_entries[i]; 3262 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len); 3263 if (ocfs2_xattr_is_local(xe)) 3264 xe_len += 3265 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); 3266 else 3267 xe_len += OCFS2_XATTR_ROOT_SIZE; 3268 name_value_len += xe_len; 3269 if (le16_to_cpu(xe->xe_name_offset) < name_offset) 3270 name_offset = le16_to_cpu(xe->xe_name_offset); 3271 } 3272 3273 /* 3274 * Now begin the modification to the new bucket. 3275 * 3276 * In the new bucket, We just move the xattr entry to the beginning 3277 * and don't touch the name/value. So there will be some holes in the 3278 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is 3279 * called. 3280 */ 3281 xe = &xh->xh_entries[start]; 3282 len = sizeof(struct ocfs2_xattr_entry) * (count - start); 3283 mlog(0, "mv xattr entry len %d from %d to %d\n", len, 3284 (int)((char *)xe - (char *)xh), 3285 (int)((char *)xh->xh_entries - (char *)xh)); 3286 memmove((char *)xh->xh_entries, (char *)xe, len); 3287 xe = &xh->xh_entries[count - start]; 3288 len = sizeof(struct ocfs2_xattr_entry) * start; 3289 memset((char *)xe, 0, len); 3290 3291 le16_add_cpu(&xh->xh_count, -start); 3292 le16_add_cpu(&xh->xh_name_value_len, -name_value_len); 3293 3294 /* Calculate xh_free_start for the new bucket. */ 3295 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE); 3296 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) { 3297 xe = &xh->xh_entries[i]; 3298 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len); 3299 if (ocfs2_xattr_is_local(xe)) 3300 xe_len += 3301 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); 3302 else 3303 xe_len += OCFS2_XATTR_ROOT_SIZE; 3304 if (le16_to_cpu(xe->xe_name_offset) < 3305 le16_to_cpu(xh->xh_free_start)) 3306 xh->xh_free_start = xe->xe_name_offset; 3307 } 3308 3309 /* set xh->xh_num_buckets for the new xh. */ 3310 if (new_bucket_head) 3311 xh->xh_num_buckets = cpu_to_le16(1); 3312 else 3313 xh->xh_num_buckets = 0; 3314 3315 for (i = 0; i < blk_per_bucket; i++) { 3316 ocfs2_journal_dirty(handle, t_bhs[i]); 3317 if (ret) 3318 mlog_errno(ret); 3319 } 3320 3321 /* store the first_hash of the new bucket. */ 3322 if (first_hash) 3323 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash); 3324 3325 /* 3326 * Now only update the 1st block of the old bucket. 3327 * Please note that the entry has been sorted already above. 3328 */ 3329 xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data; 3330 memset(&xh->xh_entries[start], 0, 3331 sizeof(struct ocfs2_xattr_entry) * (count - start)); 3332 xh->xh_count = cpu_to_le16(start); 3333 xh->xh_free_start = cpu_to_le16(name_offset); 3334 xh->xh_name_value_len = cpu_to_le16(name_value_len); 3335 3336 ocfs2_journal_dirty(handle, s_bhs[0]); 3337 if (ret) 3338 mlog_errno(ret); 3339 3340 out: 3341 if (s_bhs) { 3342 for (i = 0; i < blk_per_bucket; i++) 3343 brelse(s_bhs[i]); 3344 } 3345 kfree(s_bhs); 3346 3347 if (t_bhs) { 3348 for (i = 0; i < blk_per_bucket; i++) 3349 brelse(t_bhs[i]); 3350 } 3351 kfree(t_bhs); 3352 3353 return ret; 3354 } 3355 3356 /* 3357 * Copy xattr from one bucket to another bucket. 3358 * 3359 * The caller must make sure that the journal transaction 3360 * has enough space for journaling. 3361 */ 3362 static int ocfs2_cp_xattr_bucket(struct inode *inode, 3363 handle_t *handle, 3364 u64 s_blkno, 3365 u64 t_blkno, 3366 int t_is_new) 3367 { 3368 int ret, i; 3369 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 3370 int blocksize = inode->i_sb->s_blocksize; 3371 struct buffer_head **s_bhs, **t_bhs = NULL; 3372 3373 BUG_ON(s_blkno == t_blkno); 3374 3375 mlog(0, "cp bucket %llu to %llu, target is %d\n", 3376 s_blkno, t_blkno, t_is_new); 3377 3378 s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket, 3379 GFP_NOFS); 3380 if (!s_bhs) 3381 return -ENOMEM; 3382 3383 ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0); 3384 if (ret) 3385 goto out; 3386 3387 t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket, 3388 GFP_NOFS); 3389 if (!t_bhs) { 3390 ret = -ENOMEM; 3391 goto out; 3392 } 3393 3394 ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new); 3395 if (ret) 3396 goto out; 3397 3398 for (i = 0; i < blk_per_bucket; i++) { 3399 ret = ocfs2_journal_access(handle, inode, t_bhs[i], 3400 OCFS2_JOURNAL_ACCESS_WRITE); 3401 if (ret) 3402 goto out; 3403 } 3404 3405 for (i = 0; i < blk_per_bucket; i++) { 3406 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize); 3407 ocfs2_journal_dirty(handle, t_bhs[i]); 3408 } 3409 3410 out: 3411 if (s_bhs) { 3412 for (i = 0; i < blk_per_bucket; i++) 3413 brelse(s_bhs[i]); 3414 } 3415 kfree(s_bhs); 3416 3417 if (t_bhs) { 3418 for (i = 0; i < blk_per_bucket; i++) 3419 brelse(t_bhs[i]); 3420 } 3421 kfree(t_bhs); 3422 3423 return ret; 3424 } 3425 3426 /* 3427 * Copy one xattr cluster from src_blk to to_blk. 3428 * The to_blk will become the first bucket header of the cluster, so its 3429 * xh_num_buckets will be initialized as the bucket num in the cluster. 3430 */ 3431 static int ocfs2_cp_xattr_cluster(struct inode *inode, 3432 handle_t *handle, 3433 struct buffer_head *first_bh, 3434 u64 src_blk, 3435 u64 to_blk, 3436 u32 *first_hash) 3437 { 3438 int i, ret, credits; 3439 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 3440 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); 3441 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb); 3442 struct buffer_head *bh = NULL; 3443 struct ocfs2_xattr_header *xh; 3444 u64 to_blk_start = to_blk; 3445 3446 mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk); 3447 3448 /* 3449 * We need to update the new cluster and 1 more for the update of 3450 * the 1st bucket of the previous extent rec. 3451 */ 3452 credits = bpc + 1; 3453 ret = ocfs2_extend_trans(handle, credits); 3454 if (ret) { 3455 mlog_errno(ret); 3456 goto out; 3457 } 3458 3459 ret = ocfs2_journal_access(handle, inode, first_bh, 3460 OCFS2_JOURNAL_ACCESS_WRITE); 3461 if (ret) { 3462 mlog_errno(ret); 3463 goto out; 3464 } 3465 3466 for (i = 0; i < num_buckets; i++) { 3467 ret = ocfs2_cp_xattr_bucket(inode, handle, 3468 src_blk, to_blk, 1); 3469 if (ret) { 3470 mlog_errno(ret); 3471 goto out; 3472 } 3473 3474 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb); 3475 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb); 3476 } 3477 3478 /* update the old bucket header. */ 3479 xh = (struct ocfs2_xattr_header *)first_bh->b_data; 3480 le16_add_cpu(&xh->xh_num_buckets, -num_buckets); 3481 3482 ocfs2_journal_dirty(handle, first_bh); 3483 3484 /* update the new bucket header. */ 3485 ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode); 3486 if (ret < 0) { 3487 mlog_errno(ret); 3488 goto out; 3489 } 3490 3491 ret = ocfs2_journal_access(handle, inode, bh, 3492 OCFS2_JOURNAL_ACCESS_WRITE); 3493 if (ret) { 3494 mlog_errno(ret); 3495 goto out; 3496 } 3497 3498 xh = (struct ocfs2_xattr_header *)bh->b_data; 3499 xh->xh_num_buckets = cpu_to_le16(num_buckets); 3500 3501 ocfs2_journal_dirty(handle, bh); 3502 3503 if (first_hash) 3504 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash); 3505 out: 3506 brelse(bh); 3507 return ret; 3508 } 3509 3510 /* 3511 * Move half of the xattrs in this cluster to the new cluster. 3512 * This function should only be called when bucket size == cluster size. 3513 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead. 3514 */ 3515 static int ocfs2_half_xattr_cluster(struct inode *inode, 3516 handle_t *handle, 3517 u64 prev_blk, 3518 u64 new_blk, 3519 u32 *first_hash) 3520 { 3521 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 3522 int ret, credits = 2 * blk_per_bucket; 3523 3524 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize); 3525 3526 ret = ocfs2_extend_trans(handle, credits); 3527 if (ret) { 3528 mlog_errno(ret); 3529 return ret; 3530 } 3531 3532 /* Move half of the xattr in start_blk to the next bucket. */ 3533 return ocfs2_half_xattr_bucket(inode, handle, prev_blk, 3534 new_blk, first_hash, 1); 3535 } 3536 3537 /* 3538 * Move some xattrs from the old cluster to the new one since they are not 3539 * contiguous in ocfs2 xattr tree. 3540 * 3541 * new_blk starts a new separate cluster, and we will move some xattrs from 3542 * prev_blk to it. v_start will be set as the first name hash value in this 3543 * new cluster so that it can be used as e_cpos during tree insertion and 3544 * don't collide with our original b-tree operations. first_bh and header_bh 3545 * will also be updated since they will be used in ocfs2_extend_xattr_bucket 3546 * to extend the insert bucket. 3547 * 3548 * The problem is how much xattr should we move to the new one and when should 3549 * we update first_bh and header_bh? 3550 * 1. If cluster size > bucket size, that means the previous cluster has more 3551 * than 1 bucket, so just move half nums of bucket into the new cluster and 3552 * update the first_bh and header_bh if the insert bucket has been moved 3553 * to the new cluster. 3554 * 2. If cluster_size == bucket_size: 3555 * a) If the previous extent rec has more than one cluster and the insert 3556 * place isn't in the last cluster, copy the entire last cluster to the 3557 * new one. This time, we don't need to upate the first_bh and header_bh 3558 * since they will not be moved into the new cluster. 3559 * b) Otherwise, move the bottom half of the xattrs in the last cluster into 3560 * the new one. And we set the extend flag to zero if the insert place is 3561 * moved into the new allocated cluster since no extend is needed. 3562 */ 3563 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode, 3564 handle_t *handle, 3565 struct buffer_head **first_bh, 3566 struct buffer_head **header_bh, 3567 u64 new_blk, 3568 u64 prev_blk, 3569 u32 prev_clusters, 3570 u32 *v_start, 3571 int *extend) 3572 { 3573 int ret = 0; 3574 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); 3575 3576 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n", 3577 prev_blk, prev_clusters, new_blk); 3578 3579 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) 3580 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode, 3581 handle, 3582 first_bh, 3583 header_bh, 3584 new_blk, 3585 prev_blk, 3586 prev_clusters, 3587 v_start); 3588 else { 3589 u64 last_blk = prev_blk + bpc * (prev_clusters - 1); 3590 3591 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk) 3592 ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh, 3593 last_blk, new_blk, 3594 v_start); 3595 else { 3596 ret = ocfs2_half_xattr_cluster(inode, handle, 3597 last_blk, new_blk, 3598 v_start); 3599 3600 if ((*header_bh)->b_blocknr == last_blk && extend) 3601 *extend = 0; 3602 } 3603 } 3604 3605 return ret; 3606 } 3607 3608 /* 3609 * Add a new cluster for xattr storage. 3610 * 3611 * If the new cluster is contiguous with the previous one, it will be 3612 * appended to the same extent record, and num_clusters will be updated. 3613 * If not, we will insert a new extent for it and move some xattrs in 3614 * the last cluster into the new allocated one. 3615 * We also need to limit the maximum size of a btree leaf, otherwise we'll 3616 * lose the benefits of hashing because we'll have to search large leaves. 3617 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize, 3618 * if it's bigger). 3619 * 3620 * first_bh is the first block of the previous extent rec and header_bh 3621 * indicates the bucket we will insert the new xattrs. They will be updated 3622 * when the header_bh is moved into the new cluster. 3623 */ 3624 static int ocfs2_add_new_xattr_cluster(struct inode *inode, 3625 struct buffer_head *root_bh, 3626 struct buffer_head **first_bh, 3627 struct buffer_head **header_bh, 3628 u32 *num_clusters, 3629 u32 prev_cpos, 3630 u64 prev_blkno, 3631 int *extend) 3632 { 3633 int ret, credits; 3634 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); 3635 u32 prev_clusters = *num_clusters; 3636 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0; 3637 u64 block; 3638 handle_t *handle = NULL; 3639 struct ocfs2_alloc_context *data_ac = NULL; 3640 struct ocfs2_alloc_context *meta_ac = NULL; 3641 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 3642 struct ocfs2_extent_tree et; 3643 3644 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, " 3645 "previous xattr blkno = %llu\n", 3646 (unsigned long long)OCFS2_I(inode)->ip_blkno, 3647 prev_cpos, prev_blkno); 3648 3649 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh); 3650 3651 ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, 3652 &data_ac, &meta_ac); 3653 if (ret) { 3654 mlog_errno(ret); 3655 goto leave; 3656 } 3657 3658 credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el, 3659 clusters_to_add); 3660 handle = ocfs2_start_trans(osb, credits); 3661 if (IS_ERR(handle)) { 3662 ret = PTR_ERR(handle); 3663 handle = NULL; 3664 mlog_errno(ret); 3665 goto leave; 3666 } 3667 3668 ret = ocfs2_journal_access(handle, inode, root_bh, 3669 OCFS2_JOURNAL_ACCESS_WRITE); 3670 if (ret < 0) { 3671 mlog_errno(ret); 3672 goto leave; 3673 } 3674 3675 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1, 3676 clusters_to_add, &bit_off, &num_bits); 3677 if (ret < 0) { 3678 if (ret != -ENOSPC) 3679 mlog_errno(ret); 3680 goto leave; 3681 } 3682 3683 BUG_ON(num_bits > clusters_to_add); 3684 3685 block = ocfs2_clusters_to_blocks(osb->sb, bit_off); 3686 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n", 3687 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); 3688 3689 if (prev_blkno + prev_clusters * bpc == block && 3690 (prev_clusters + num_bits) << osb->s_clustersize_bits <= 3691 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) { 3692 /* 3693 * If this cluster is contiguous with the old one and 3694 * adding this new cluster, we don't surpass the limit of 3695 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be 3696 * initialized and used like other buckets in the previous 3697 * cluster. 3698 * So add it as a contiguous one. The caller will handle 3699 * its init process. 3700 */ 3701 v_start = prev_cpos + prev_clusters; 3702 *num_clusters = prev_clusters + num_bits; 3703 mlog(0, "Add contiguous %u clusters to previous extent rec.\n", 3704 num_bits); 3705 } else { 3706 ret = ocfs2_adjust_xattr_cross_cluster(inode, 3707 handle, 3708 first_bh, 3709 header_bh, 3710 block, 3711 prev_blkno, 3712 prev_clusters, 3713 &v_start, 3714 extend); 3715 if (ret) { 3716 mlog_errno(ret); 3717 goto leave; 3718 } 3719 } 3720 3721 if (handle->h_buffer_credits < credits) { 3722 /* 3723 * The journal has been restarted before, and don't 3724 * have enough space for the insertion, so extend it 3725 * here. 3726 */ 3727 ret = ocfs2_extend_trans(handle, credits); 3728 if (ret) { 3729 mlog_errno(ret); 3730 goto leave; 3731 } 3732 } 3733 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n", 3734 num_bits, block, v_start); 3735 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block, 3736 num_bits, 0, meta_ac); 3737 if (ret < 0) { 3738 mlog_errno(ret); 3739 goto leave; 3740 } 3741 3742 ret = ocfs2_journal_dirty(handle, root_bh); 3743 if (ret < 0) { 3744 mlog_errno(ret); 3745 goto leave; 3746 } 3747 3748 leave: 3749 if (handle) 3750 ocfs2_commit_trans(osb, handle); 3751 if (data_ac) 3752 ocfs2_free_alloc_context(data_ac); 3753 if (meta_ac) 3754 ocfs2_free_alloc_context(meta_ac); 3755 3756 return ret; 3757 } 3758 3759 /* 3760 * Extend a new xattr bucket and move xattrs to the end one by one until 3761 * We meet with start_bh. Only move half of the xattrs to the bucket after it. 3762 */ 3763 static int ocfs2_extend_xattr_bucket(struct inode *inode, 3764 struct buffer_head *first_bh, 3765 struct buffer_head *start_bh, 3766 u32 num_clusters) 3767 { 3768 int ret, credits; 3769 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 3770 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 3771 u64 start_blk = start_bh->b_blocknr, end_blk; 3772 u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb); 3773 handle_t *handle; 3774 struct ocfs2_xattr_header *first_xh = 3775 (struct ocfs2_xattr_header *)first_bh->b_data; 3776 u16 bucket = le16_to_cpu(first_xh->xh_num_buckets); 3777 3778 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting " 3779 "from %llu, len = %u\n", start_blk, 3780 (unsigned long long)first_bh->b_blocknr, num_clusters); 3781 3782 BUG_ON(bucket >= num_buckets); 3783 3784 end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket; 3785 3786 /* 3787 * We will touch all the buckets after the start_bh(include it). 3788 * Add one more bucket and modify the first_bh. 3789 */ 3790 credits = end_blk - start_blk + 2 * blk_per_bucket + 1; 3791 handle = ocfs2_start_trans(osb, credits); 3792 if (IS_ERR(handle)) { 3793 ret = PTR_ERR(handle); 3794 handle = NULL; 3795 mlog_errno(ret); 3796 goto out; 3797 } 3798 3799 ret = ocfs2_journal_access(handle, inode, first_bh, 3800 OCFS2_JOURNAL_ACCESS_WRITE); 3801 if (ret) { 3802 mlog_errno(ret); 3803 goto commit; 3804 } 3805 3806 while (end_blk != start_blk) { 3807 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk, 3808 end_blk + blk_per_bucket, 0); 3809 if (ret) 3810 goto commit; 3811 end_blk -= blk_per_bucket; 3812 } 3813 3814 /* Move half of the xattr in start_blk to the next bucket. */ 3815 ret = ocfs2_half_xattr_bucket(inode, handle, start_blk, 3816 start_blk + blk_per_bucket, NULL, 0); 3817 3818 le16_add_cpu(&first_xh->xh_num_buckets, 1); 3819 ocfs2_journal_dirty(handle, first_bh); 3820 3821 commit: 3822 ocfs2_commit_trans(osb, handle); 3823 out: 3824 return ret; 3825 } 3826 3827 /* 3828 * Add new xattr bucket in an extent record and adjust the buckets accordingly. 3829 * xb_bh is the ocfs2_xattr_block. 3830 * We will move all the buckets starting from header_bh to the next place. As 3831 * for this one, half num of its xattrs will be moved to the next one. 3832 * 3833 * We will allocate a new cluster if current cluster is full and adjust 3834 * header_bh and first_bh if the insert place is moved to the new cluster. 3835 */ 3836 static int ocfs2_add_new_xattr_bucket(struct inode *inode, 3837 struct buffer_head *xb_bh, 3838 struct buffer_head *header_bh) 3839 { 3840 struct ocfs2_xattr_header *first_xh = NULL; 3841 struct buffer_head *first_bh = NULL; 3842 struct ocfs2_xattr_block *xb = 3843 (struct ocfs2_xattr_block *)xb_bh->b_data; 3844 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root; 3845 struct ocfs2_extent_list *el = &xb_root->xt_list; 3846 struct ocfs2_xattr_header *xh = 3847 (struct ocfs2_xattr_header *)header_bh->b_data; 3848 u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash); 3849 struct super_block *sb = inode->i_sb; 3850 struct ocfs2_super *osb = OCFS2_SB(sb); 3851 int ret, num_buckets, extend = 1; 3852 u64 p_blkno; 3853 u32 e_cpos, num_clusters; 3854 3855 mlog(0, "Add new xattr bucket starting form %llu\n", 3856 (unsigned long long)header_bh->b_blocknr); 3857 3858 /* 3859 * Add refrence for header_bh here because it may be 3860 * changed in ocfs2_add_new_xattr_cluster and we need 3861 * to free it in the end. 3862 */ 3863 get_bh(header_bh); 3864 3865 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos, 3866 &num_clusters, el); 3867 if (ret) { 3868 mlog_errno(ret); 3869 goto out; 3870 } 3871 3872 ret = ocfs2_read_block(osb, p_blkno, 3873 &first_bh, OCFS2_BH_CACHED, inode); 3874 if (ret) { 3875 mlog_errno(ret); 3876 goto out; 3877 } 3878 3879 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters; 3880 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data; 3881 3882 if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) { 3883 ret = ocfs2_add_new_xattr_cluster(inode, 3884 xb_bh, 3885 &first_bh, 3886 &header_bh, 3887 &num_clusters, 3888 e_cpos, 3889 p_blkno, 3890 &extend); 3891 if (ret) { 3892 mlog_errno(ret); 3893 goto out; 3894 } 3895 } 3896 3897 if (extend) 3898 ret = ocfs2_extend_xattr_bucket(inode, 3899 first_bh, 3900 header_bh, 3901 num_clusters); 3902 if (ret) 3903 mlog_errno(ret); 3904 out: 3905 brelse(first_bh); 3906 brelse(header_bh); 3907 return ret; 3908 } 3909 3910 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode, 3911 struct ocfs2_xattr_bucket *bucket, 3912 int offs) 3913 { 3914 int block_off = offs >> inode->i_sb->s_blocksize_bits; 3915 3916 offs = offs % inode->i_sb->s_blocksize; 3917 return bucket->bhs[block_off]->b_data + offs; 3918 } 3919 3920 /* 3921 * Handle the normal xattr set, including replace, delete and new. 3922 * 3923 * Note: "local" indicates the real data's locality. So we can't 3924 * just its bucket locality by its length. 3925 */ 3926 static void ocfs2_xattr_set_entry_normal(struct inode *inode, 3927 struct ocfs2_xattr_info *xi, 3928 struct ocfs2_xattr_search *xs, 3929 u32 name_hash, 3930 int local) 3931 { 3932 struct ocfs2_xattr_entry *last, *xe; 3933 int name_len = strlen(xi->name); 3934 struct ocfs2_xattr_header *xh = xs->header; 3935 u16 count = le16_to_cpu(xh->xh_count), start; 3936 size_t blocksize = inode->i_sb->s_blocksize; 3937 char *val; 3938 size_t offs, size, new_size; 3939 3940 last = &xh->xh_entries[count]; 3941 if (!xs->not_found) { 3942 xe = xs->here; 3943 offs = le16_to_cpu(xe->xe_name_offset); 3944 if (ocfs2_xattr_is_local(xe)) 3945 size = OCFS2_XATTR_SIZE(name_len) + 3946 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); 3947 else 3948 size = OCFS2_XATTR_SIZE(name_len) + 3949 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE); 3950 3951 /* 3952 * If the new value will be stored outside, xi->value has been 3953 * initalized as an empty ocfs2_xattr_value_root, and the same 3954 * goes with xi->value_len, so we can set new_size safely here. 3955 * See ocfs2_xattr_set_in_bucket. 3956 */ 3957 new_size = OCFS2_XATTR_SIZE(name_len) + 3958 OCFS2_XATTR_SIZE(xi->value_len); 3959 3960 le16_add_cpu(&xh->xh_name_value_len, -size); 3961 if (xi->value) { 3962 if (new_size > size) 3963 goto set_new_name_value; 3964 3965 /* Now replace the old value with new one. */ 3966 if (local) 3967 xe->xe_value_size = cpu_to_le64(xi->value_len); 3968 else 3969 xe->xe_value_size = 0; 3970 3971 val = ocfs2_xattr_bucket_get_val(inode, 3972 &xs->bucket, offs); 3973 memset(val + OCFS2_XATTR_SIZE(name_len), 0, 3974 size - OCFS2_XATTR_SIZE(name_len)); 3975 if (OCFS2_XATTR_SIZE(xi->value_len) > 0) 3976 memcpy(val + OCFS2_XATTR_SIZE(name_len), 3977 xi->value, xi->value_len); 3978 3979 le16_add_cpu(&xh->xh_name_value_len, new_size); 3980 ocfs2_xattr_set_local(xe, local); 3981 return; 3982 } else { 3983 /* 3984 * Remove the old entry if there is more than one. 3985 * We don't remove the last entry so that we can 3986 * use it to indicate the hash value of the empty 3987 * bucket. 3988 */ 3989 last -= 1; 3990 le16_add_cpu(&xh->xh_count, -1); 3991 if (xh->xh_count) { 3992 memmove(xe, xe + 1, 3993 (void *)last - (void *)xe); 3994 memset(last, 0, 3995 sizeof(struct ocfs2_xattr_entry)); 3996 } else 3997 xh->xh_free_start = 3998 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE); 3999 4000 return; 4001 } 4002 } else { 4003 /* find a new entry for insert. */ 4004 int low = 0, high = count - 1, tmp; 4005 struct ocfs2_xattr_entry *tmp_xe; 4006 4007 while (low <= high && count) { 4008 tmp = (low + high) / 2; 4009 tmp_xe = &xh->xh_entries[tmp]; 4010 4011 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash)) 4012 low = tmp + 1; 4013 else if (name_hash < 4014 le32_to_cpu(tmp_xe->xe_name_hash)) 4015 high = tmp - 1; 4016 else { 4017 low = tmp; 4018 break; 4019 } 4020 } 4021 4022 xe = &xh->xh_entries[low]; 4023 if (low != count) 4024 memmove(xe + 1, xe, (void *)last - (void *)xe); 4025 4026 le16_add_cpu(&xh->xh_count, 1); 4027 memset(xe, 0, sizeof(struct ocfs2_xattr_entry)); 4028 xe->xe_name_hash = cpu_to_le32(name_hash); 4029 xe->xe_name_len = name_len; 4030 ocfs2_xattr_set_type(xe, xi->name_index); 4031 } 4032 4033 set_new_name_value: 4034 /* Insert the new name+value. */ 4035 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len); 4036 4037 /* 4038 * We must make sure that the name/value pair 4039 * exists in the same block. 4040 */ 4041 offs = le16_to_cpu(xh->xh_free_start); 4042 start = offs - size; 4043 4044 if (start >> inode->i_sb->s_blocksize_bits != 4045 (offs - 1) >> inode->i_sb->s_blocksize_bits) { 4046 offs = offs - offs % blocksize; 4047 xh->xh_free_start = cpu_to_le16(offs); 4048 } 4049 4050 val = ocfs2_xattr_bucket_get_val(inode, 4051 &xs->bucket, offs - size); 4052 xe->xe_name_offset = cpu_to_le16(offs - size); 4053 4054 memset(val, 0, size); 4055 memcpy(val, xi->name, name_len); 4056 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len); 4057 4058 xe->xe_value_size = cpu_to_le64(xi->value_len); 4059 ocfs2_xattr_set_local(xe, local); 4060 xs->here = xe; 4061 le16_add_cpu(&xh->xh_free_start, -size); 4062 le16_add_cpu(&xh->xh_name_value_len, size); 4063 4064 return; 4065 } 4066 4067 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode, 4068 handle_t *handle, 4069 struct ocfs2_xattr_search *xs, 4070 struct buffer_head **bhs, 4071 u16 bh_num) 4072 { 4073 int ret = 0, off, block_off; 4074 struct ocfs2_xattr_entry *xe = xs->here; 4075 4076 /* 4077 * First calculate all the blocks we should journal_access 4078 * and journal_dirty. The first block should always be touched. 4079 */ 4080 ret = ocfs2_journal_dirty(handle, bhs[0]); 4081 if (ret) 4082 mlog_errno(ret); 4083 4084 /* calc the data. */ 4085 off = le16_to_cpu(xe->xe_name_offset); 4086 block_off = off >> inode->i_sb->s_blocksize_bits; 4087 ret = ocfs2_journal_dirty(handle, bhs[block_off]); 4088 if (ret) 4089 mlog_errno(ret); 4090 4091 return ret; 4092 } 4093 4094 /* 4095 * Set the xattr entry in the specified bucket. 4096 * The bucket is indicated by xs->bucket and it should have the enough 4097 * space for the xattr insertion. 4098 */ 4099 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode, 4100 struct ocfs2_xattr_info *xi, 4101 struct ocfs2_xattr_search *xs, 4102 u32 name_hash, 4103 int local) 4104 { 4105 int i, ret; 4106 handle_t *handle = NULL; 4107 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 4108 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 4109 4110 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n", 4111 (unsigned long)xi->value_len, xi->name_index, 4112 (unsigned long long)xs->bucket.bhs[0]->b_blocknr); 4113 4114 if (!xs->bucket.bhs[1]) { 4115 ret = ocfs2_read_blocks(osb, 4116 xs->bucket.bhs[0]->b_blocknr + 1, 4117 blk_per_bucket - 1, &xs->bucket.bhs[1], 4118 OCFS2_BH_CACHED, inode); 4119 if (ret) { 4120 mlog_errno(ret); 4121 goto out; 4122 } 4123 } 4124 4125 handle = ocfs2_start_trans(osb, blk_per_bucket); 4126 if (IS_ERR(handle)) { 4127 ret = PTR_ERR(handle); 4128 handle = NULL; 4129 mlog_errno(ret); 4130 goto out; 4131 } 4132 4133 for (i = 0; i < blk_per_bucket; i++) { 4134 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i], 4135 OCFS2_JOURNAL_ACCESS_WRITE); 4136 if (ret < 0) { 4137 mlog_errno(ret); 4138 goto out; 4139 } 4140 } 4141 4142 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local); 4143 4144 /*Only dirty the blocks we have touched in set xattr. */ 4145 ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs, 4146 xs->bucket.bhs, blk_per_bucket); 4147 if (ret) 4148 mlog_errno(ret); 4149 out: 4150 ocfs2_commit_trans(osb, handle); 4151 4152 return ret; 4153 } 4154 4155 static int ocfs2_xattr_value_update_size(struct inode *inode, 4156 struct buffer_head *xe_bh, 4157 struct ocfs2_xattr_entry *xe, 4158 u64 new_size) 4159 { 4160 int ret; 4161 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 4162 handle_t *handle = NULL; 4163 4164 handle = ocfs2_start_trans(osb, 1); 4165 if (handle == NULL) { 4166 ret = -ENOMEM; 4167 mlog_errno(ret); 4168 goto out; 4169 } 4170 4171 ret = ocfs2_journal_access(handle, inode, xe_bh, 4172 OCFS2_JOURNAL_ACCESS_WRITE); 4173 if (ret < 0) { 4174 mlog_errno(ret); 4175 goto out_commit; 4176 } 4177 4178 xe->xe_value_size = cpu_to_le64(new_size); 4179 4180 ret = ocfs2_journal_dirty(handle, xe_bh); 4181 if (ret < 0) 4182 mlog_errno(ret); 4183 4184 out_commit: 4185 ocfs2_commit_trans(osb, handle); 4186 out: 4187 return ret; 4188 } 4189 4190 /* 4191 * Truncate the specified xe_off entry in xattr bucket. 4192 * bucket is indicated by header_bh and len is the new length. 4193 * Both the ocfs2_xattr_value_root and the entry will be updated here. 4194 * 4195 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed. 4196 */ 4197 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode, 4198 struct buffer_head *header_bh, 4199 int xe_off, 4200 int len) 4201 { 4202 int ret, offset; 4203 u64 value_blk; 4204 struct buffer_head *value_bh = NULL; 4205 struct ocfs2_xattr_value_root *xv; 4206 struct ocfs2_xattr_entry *xe; 4207 struct ocfs2_xattr_header *xh = 4208 (struct ocfs2_xattr_header *)header_bh->b_data; 4209 size_t blocksize = inode->i_sb->s_blocksize; 4210 4211 xe = &xh->xh_entries[xe_off]; 4212 4213 BUG_ON(!xe || ocfs2_xattr_is_local(xe)); 4214 4215 offset = le16_to_cpu(xe->xe_name_offset) + 4216 OCFS2_XATTR_SIZE(xe->xe_name_len); 4217 4218 value_blk = offset / blocksize; 4219 4220 /* We don't allow ocfs2_xattr_value to be stored in different block. */ 4221 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize); 4222 value_blk += header_bh->b_blocknr; 4223 4224 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk, 4225 &value_bh, OCFS2_BH_CACHED, inode); 4226 if (ret) { 4227 mlog_errno(ret); 4228 goto out; 4229 } 4230 4231 xv = (struct ocfs2_xattr_value_root *) 4232 (value_bh->b_data + offset % blocksize); 4233 4234 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n", 4235 xe_off, (unsigned long long)header_bh->b_blocknr, len); 4236 ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len); 4237 if (ret) { 4238 mlog_errno(ret); 4239 goto out; 4240 } 4241 4242 ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len); 4243 if (ret) { 4244 mlog_errno(ret); 4245 goto out; 4246 } 4247 4248 out: 4249 brelse(value_bh); 4250 return ret; 4251 } 4252 4253 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode, 4254 struct ocfs2_xattr_search *xs, 4255 int len) 4256 { 4257 int ret, offset; 4258 struct ocfs2_xattr_entry *xe = xs->here; 4259 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base; 4260 4261 BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe)); 4262 4263 offset = xe - xh->xh_entries; 4264 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0], 4265 offset, len); 4266 if (ret) 4267 mlog_errno(ret); 4268 4269 return ret; 4270 } 4271 4272 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode, 4273 struct ocfs2_xattr_search *xs, 4274 char *val, 4275 int value_len) 4276 { 4277 int offset; 4278 struct ocfs2_xattr_value_root *xv; 4279 struct ocfs2_xattr_entry *xe = xs->here; 4280 4281 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe)); 4282 4283 offset = le16_to_cpu(xe->xe_name_offset) + 4284 OCFS2_XATTR_SIZE(xe->xe_name_len); 4285 4286 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset); 4287 4288 return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len); 4289 } 4290 4291 static int ocfs2_rm_xattr_cluster(struct inode *inode, 4292 struct buffer_head *root_bh, 4293 u64 blkno, 4294 u32 cpos, 4295 u32 len) 4296 { 4297 int ret; 4298 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 4299 struct inode *tl_inode = osb->osb_tl_inode; 4300 handle_t *handle; 4301 struct ocfs2_xattr_block *xb = 4302 (struct ocfs2_xattr_block *)root_bh->b_data; 4303 struct ocfs2_alloc_context *meta_ac = NULL; 4304 struct ocfs2_cached_dealloc_ctxt dealloc; 4305 struct ocfs2_extent_tree et; 4306 4307 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh); 4308 4309 ocfs2_init_dealloc_ctxt(&dealloc); 4310 4311 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n", 4312 cpos, len, (unsigned long long)blkno); 4313 4314 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len); 4315 4316 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac); 4317 if (ret) { 4318 mlog_errno(ret); 4319 return ret; 4320 } 4321 4322 mutex_lock(&tl_inode->i_mutex); 4323 4324 if (ocfs2_truncate_log_needs_flush(osb)) { 4325 ret = __ocfs2_flush_truncate_log(osb); 4326 if (ret < 0) { 4327 mlog_errno(ret); 4328 goto out; 4329 } 4330 } 4331 4332 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); 4333 if (handle == NULL) { 4334 ret = -ENOMEM; 4335 mlog_errno(ret); 4336 goto out; 4337 } 4338 4339 ret = ocfs2_journal_access(handle, inode, root_bh, 4340 OCFS2_JOURNAL_ACCESS_WRITE); 4341 if (ret) { 4342 mlog_errno(ret); 4343 goto out_commit; 4344 } 4345 4346 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac, 4347 &dealloc); 4348 if (ret) { 4349 mlog_errno(ret); 4350 goto out_commit; 4351 } 4352 4353 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len); 4354 4355 ret = ocfs2_journal_dirty(handle, root_bh); 4356 if (ret) { 4357 mlog_errno(ret); 4358 goto out_commit; 4359 } 4360 4361 ret = ocfs2_truncate_log_append(osb, handle, blkno, len); 4362 if (ret) 4363 mlog_errno(ret); 4364 4365 out_commit: 4366 ocfs2_commit_trans(osb, handle); 4367 out: 4368 ocfs2_schedule_truncate_log_flush(osb, 1); 4369 4370 mutex_unlock(&tl_inode->i_mutex); 4371 4372 if (meta_ac) 4373 ocfs2_free_alloc_context(meta_ac); 4374 4375 ocfs2_run_deallocs(osb, &dealloc); 4376 4377 return ret; 4378 } 4379 4380 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode, 4381 struct ocfs2_xattr_search *xs) 4382 { 4383 handle_t *handle = NULL; 4384 struct ocfs2_xattr_header *xh = xs->bucket.xh; 4385 struct ocfs2_xattr_entry *last = &xh->xh_entries[ 4386 le16_to_cpu(xh->xh_count) - 1]; 4387 int ret = 0; 4388 4389 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1); 4390 if (IS_ERR(handle)) { 4391 ret = PTR_ERR(handle); 4392 mlog_errno(ret); 4393 return; 4394 } 4395 4396 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0], 4397 OCFS2_JOURNAL_ACCESS_WRITE); 4398 if (ret) { 4399 mlog_errno(ret); 4400 goto out_commit; 4401 } 4402 4403 /* Remove the old entry. */ 4404 memmove(xs->here, xs->here + 1, 4405 (void *)last - (void *)xs->here); 4406 memset(last, 0, sizeof(struct ocfs2_xattr_entry)); 4407 le16_add_cpu(&xh->xh_count, -1); 4408 4409 ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]); 4410 if (ret < 0) 4411 mlog_errno(ret); 4412 out_commit: 4413 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); 4414 } 4415 4416 /* 4417 * Set the xattr name/value in the bucket specified in xs. 4418 * 4419 * As the new value in xi may be stored in the bucket or in an outside cluster, 4420 * we divide the whole process into 3 steps: 4421 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket) 4422 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs) 4423 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside) 4424 * 4. If the clusters for the new outside value can't be allocated, we need 4425 * to free the xattr we allocated in set. 4426 */ 4427 static int ocfs2_xattr_set_in_bucket(struct inode *inode, 4428 struct ocfs2_xattr_info *xi, 4429 struct ocfs2_xattr_search *xs) 4430 { 4431 int ret, local = 1; 4432 size_t value_len; 4433 char *val = (char *)xi->value; 4434 struct ocfs2_xattr_entry *xe = xs->here; 4435 u32 name_hash = ocfs2_xattr_hash_by_name(inode, 4436 xi->name_index, xi->name); 4437 4438 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) { 4439 /* 4440 * We need to truncate the xattr storage first. 4441 * 4442 * If both the old and new value are stored to 4443 * outside block, we only need to truncate 4444 * the storage and then set the value outside. 4445 * 4446 * If the new value should be stored within block, 4447 * we should free all the outside block first and 4448 * the modification to the xattr block will be done 4449 * by following steps. 4450 */ 4451 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) 4452 value_len = xi->value_len; 4453 else 4454 value_len = 0; 4455 4456 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs, 4457 value_len); 4458 if (ret) 4459 goto out; 4460 4461 if (value_len) 4462 goto set_value_outside; 4463 } 4464 4465 value_len = xi->value_len; 4466 /* So we have to handle the inside block change now. */ 4467 if (value_len > OCFS2_XATTR_INLINE_SIZE) { 4468 /* 4469 * If the new value will be stored outside of block, 4470 * initalize a new empty value root and insert it first. 4471 */ 4472 local = 0; 4473 xi->value = &def_xv; 4474 xi->value_len = OCFS2_XATTR_ROOT_SIZE; 4475 } 4476 4477 ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local); 4478 if (ret) { 4479 mlog_errno(ret); 4480 goto out; 4481 } 4482 4483 if (value_len <= OCFS2_XATTR_INLINE_SIZE) 4484 goto out; 4485 4486 /* allocate the space now for the outside block storage. */ 4487 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs, 4488 value_len); 4489 if (ret) { 4490 mlog_errno(ret); 4491 4492 if (xs->not_found) { 4493 /* 4494 * We can't allocate enough clusters for outside 4495 * storage and we have allocated xattr already, 4496 * so need to remove it. 4497 */ 4498 ocfs2_xattr_bucket_remove_xs(inode, xs); 4499 } 4500 goto out; 4501 } 4502 4503 set_value_outside: 4504 ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len); 4505 out: 4506 return ret; 4507 } 4508 4509 /* check whether the xattr bucket is filled up with the same hash value. */ 4510 static int ocfs2_check_xattr_bucket_collision(struct inode *inode, 4511 struct ocfs2_xattr_bucket *bucket) 4512 { 4513 struct ocfs2_xattr_header *xh = bucket->xh; 4514 4515 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash == 4516 xh->xh_entries[0].xe_name_hash) { 4517 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, " 4518 "hash = %u\n", 4519 (unsigned long long)bucket->bhs[0]->b_blocknr, 4520 le32_to_cpu(xh->xh_entries[0].xe_name_hash)); 4521 return -ENOSPC; 4522 } 4523 4524 return 0; 4525 } 4526 4527 static int ocfs2_xattr_set_entry_index_block(struct inode *inode, 4528 struct ocfs2_xattr_info *xi, 4529 struct ocfs2_xattr_search *xs) 4530 { 4531 struct ocfs2_xattr_header *xh; 4532 struct ocfs2_xattr_entry *xe; 4533 u16 count, header_size, xh_free_start; 4534 int i, free, max_free, need, old; 4535 size_t value_size = 0, name_len = strlen(xi->name); 4536 size_t blocksize = inode->i_sb->s_blocksize; 4537 int ret, allocation = 0; 4538 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb); 4539 4540 mlog_entry("Set xattr %s in xattr index block\n", xi->name); 4541 4542 try_again: 4543 xh = xs->header; 4544 count = le16_to_cpu(xh->xh_count); 4545 xh_free_start = le16_to_cpu(xh->xh_free_start); 4546 header_size = sizeof(struct ocfs2_xattr_header) + 4547 count * sizeof(struct ocfs2_xattr_entry); 4548 max_free = OCFS2_XATTR_BUCKET_SIZE - 4549 le16_to_cpu(xh->xh_name_value_len) - header_size; 4550 4551 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size " 4552 "of %u which exceed block size\n", 4553 (unsigned long long)xs->bucket.bhs[0]->b_blocknr, 4554 header_size); 4555 4556 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) 4557 value_size = OCFS2_XATTR_ROOT_SIZE; 4558 else if (xi->value) 4559 value_size = OCFS2_XATTR_SIZE(xi->value_len); 4560 4561 if (xs->not_found) 4562 need = sizeof(struct ocfs2_xattr_entry) + 4563 OCFS2_XATTR_SIZE(name_len) + value_size; 4564 else { 4565 need = value_size + OCFS2_XATTR_SIZE(name_len); 4566 4567 /* 4568 * We only replace the old value if the new length is smaller 4569 * than the old one. Otherwise we will allocate new space in the 4570 * bucket to store it. 4571 */ 4572 xe = xs->here; 4573 if (ocfs2_xattr_is_local(xe)) 4574 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size)); 4575 else 4576 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE); 4577 4578 if (old >= value_size) 4579 need = 0; 4580 } 4581 4582 free = xh_free_start - header_size; 4583 /* 4584 * We need to make sure the new name/value pair 4585 * can exist in the same block. 4586 */ 4587 if (xh_free_start % blocksize < need) 4588 free -= xh_free_start % blocksize; 4589 4590 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, " 4591 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len =" 4592 " %u\n", xs->not_found, 4593 (unsigned long long)xs->bucket.bhs[0]->b_blocknr, 4594 free, need, max_free, le16_to_cpu(xh->xh_free_start), 4595 le16_to_cpu(xh->xh_name_value_len)); 4596 4597 if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) { 4598 if (need <= max_free && 4599 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) { 4600 /* 4601 * We can create the space by defragment. Since only the 4602 * name/value will be moved, the xe shouldn't be changed 4603 * in xs. 4604 */ 4605 ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket); 4606 if (ret) { 4607 mlog_errno(ret); 4608 goto out; 4609 } 4610 4611 xh_free_start = le16_to_cpu(xh->xh_free_start); 4612 free = xh_free_start - header_size; 4613 if (xh_free_start % blocksize < need) 4614 free -= xh_free_start % blocksize; 4615 4616 if (free >= need) 4617 goto xattr_set; 4618 4619 mlog(0, "Can't get enough space for xattr insert by " 4620 "defragment. Need %u bytes, but we have %d, so " 4621 "allocate new bucket for it.\n", need, free); 4622 } 4623 4624 /* 4625 * We have to add new buckets or clusters and one 4626 * allocation should leave us enough space for insert. 4627 */ 4628 BUG_ON(allocation); 4629 4630 /* 4631 * We do not allow for overlapping ranges between buckets. And 4632 * the maximum number of collisions we will allow for then is 4633 * one bucket's worth, so check it here whether we need to 4634 * add a new bucket for the insert. 4635 */ 4636 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket); 4637 if (ret) { 4638 mlog_errno(ret); 4639 goto out; 4640 } 4641 4642 ret = ocfs2_add_new_xattr_bucket(inode, 4643 xs->xattr_bh, 4644 xs->bucket.bhs[0]); 4645 if (ret) { 4646 mlog_errno(ret); 4647 goto out; 4648 } 4649 4650 for (i = 0; i < blk_per_bucket; i++) 4651 brelse(xs->bucket.bhs[i]); 4652 4653 memset(&xs->bucket, 0, sizeof(xs->bucket)); 4654 4655 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh, 4656 xi->name_index, 4657 xi->name, xs); 4658 if (ret && ret != -ENODATA) 4659 goto out; 4660 xs->not_found = ret; 4661 allocation = 1; 4662 goto try_again; 4663 } 4664 4665 xattr_set: 4666 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs); 4667 out: 4668 mlog_exit(ret); 4669 return ret; 4670 } 4671 4672 static int ocfs2_delete_xattr_in_bucket(struct inode *inode, 4673 struct ocfs2_xattr_bucket *bucket, 4674 void *para) 4675 { 4676 int ret = 0; 4677 struct ocfs2_xattr_header *xh = bucket->xh; 4678 u16 i; 4679 struct ocfs2_xattr_entry *xe; 4680 4681 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) { 4682 xe = &xh->xh_entries[i]; 4683 if (ocfs2_xattr_is_local(xe)) 4684 continue; 4685 4686 ret = ocfs2_xattr_bucket_value_truncate(inode, 4687 bucket->bhs[0], 4688 i, 0); 4689 if (ret) { 4690 mlog_errno(ret); 4691 break; 4692 } 4693 } 4694 4695 return ret; 4696 } 4697 4698 static int ocfs2_delete_xattr_index_block(struct inode *inode, 4699 struct buffer_head *xb_bh) 4700 { 4701 struct ocfs2_xattr_block *xb = 4702 (struct ocfs2_xattr_block *)xb_bh->b_data; 4703 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list; 4704 int ret = 0; 4705 u32 name_hash = UINT_MAX, e_cpos, num_clusters; 4706 u64 p_blkno; 4707 4708 if (le16_to_cpu(el->l_next_free_rec) == 0) 4709 return 0; 4710 4711 while (name_hash > 0) { 4712 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, 4713 &e_cpos, &num_clusters, el); 4714 if (ret) { 4715 mlog_errno(ret); 4716 goto out; 4717 } 4718 4719 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters, 4720 ocfs2_delete_xattr_in_bucket, 4721 NULL); 4722 if (ret) { 4723 mlog_errno(ret); 4724 goto out; 4725 } 4726 4727 ret = ocfs2_rm_xattr_cluster(inode, xb_bh, 4728 p_blkno, e_cpos, num_clusters); 4729 if (ret) { 4730 mlog_errno(ret); 4731 break; 4732 } 4733 4734 if (e_cpos == 0) 4735 break; 4736 4737 name_hash = e_cpos - 1; 4738 } 4739 4740 out: 4741 return ret; 4742 } 4743