1 /* 2 * Copyright (C) 2012 Alexander Block. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 #include <linux/bsearch.h> 20 #include <linux/fs.h> 21 #include <linux/file.h> 22 #include <linux/sort.h> 23 #include <linux/mount.h> 24 #include <linux/xattr.h> 25 #include <linux/posix_acl_xattr.h> 26 #include <linux/radix-tree.h> 27 #include <linux/vmalloc.h> 28 #include <linux/string.h> 29 #include <linux/compat.h> 30 31 #include "send.h" 32 #include "backref.h" 33 #include "hash.h" 34 #include "locking.h" 35 #include "disk-io.h" 36 #include "btrfs_inode.h" 37 #include "transaction.h" 38 #include "compression.h" 39 40 /* 41 * A fs_path is a helper to dynamically build path names with unknown size. 42 * It reallocates the internal buffer on demand. 43 * It allows fast adding of path elements on the right side (normal path) and 44 * fast adding to the left side (reversed path). A reversed path can also be 45 * unreversed if needed. 46 */ 47 struct fs_path { 48 union { 49 struct { 50 char *start; 51 char *end; 52 53 char *buf; 54 unsigned short buf_len:15; 55 unsigned short reversed:1; 56 char inline_buf[]; 57 }; 58 /* 59 * Average path length does not exceed 200 bytes, we'll have 60 * better packing in the slab and higher chance to satisfy 61 * a allocation later during send. 62 */ 63 char pad[256]; 64 }; 65 }; 66 #define FS_PATH_INLINE_SIZE \ 67 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) 68 69 70 /* reused for each extent */ 71 struct clone_root { 72 struct btrfs_root *root; 73 u64 ino; 74 u64 offset; 75 76 u64 found_refs; 77 }; 78 79 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128 80 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2) 81 82 struct send_ctx { 83 struct file *send_filp; 84 loff_t send_off; 85 char *send_buf; 86 u32 send_size; 87 u32 send_max_size; 88 u64 total_send_size; 89 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1]; 90 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ 91 92 struct btrfs_root *send_root; 93 struct btrfs_root *parent_root; 94 struct clone_root *clone_roots; 95 int clone_roots_cnt; 96 97 /* current state of the compare_tree call */ 98 struct btrfs_path *left_path; 99 struct btrfs_path *right_path; 100 struct btrfs_key *cmp_key; 101 102 /* 103 * infos of the currently processed inode. In case of deleted inodes, 104 * these are the values from the deleted inode. 105 */ 106 u64 cur_ino; 107 u64 cur_inode_gen; 108 int cur_inode_new; 109 int cur_inode_new_gen; 110 int cur_inode_deleted; 111 u64 cur_inode_size; 112 u64 cur_inode_mode; 113 u64 cur_inode_rdev; 114 u64 cur_inode_last_extent; 115 116 u64 send_progress; 117 118 struct list_head new_refs; 119 struct list_head deleted_refs; 120 121 struct radix_tree_root name_cache; 122 struct list_head name_cache_list; 123 int name_cache_size; 124 125 struct file_ra_state ra; 126 127 char *read_buf; 128 129 /* 130 * We process inodes by their increasing order, so if before an 131 * incremental send we reverse the parent/child relationship of 132 * directories such that a directory with a lower inode number was 133 * the parent of a directory with a higher inode number, and the one 134 * becoming the new parent got renamed too, we can't rename/move the 135 * directory with lower inode number when we finish processing it - we 136 * must process the directory with higher inode number first, then 137 * rename/move it and then rename/move the directory with lower inode 138 * number. Example follows. 139 * 140 * Tree state when the first send was performed: 141 * 142 * . 143 * |-- a (ino 257) 144 * |-- b (ino 258) 145 * | 146 * | 147 * |-- c (ino 259) 148 * | |-- d (ino 260) 149 * | 150 * |-- c2 (ino 261) 151 * 152 * Tree state when the second (incremental) send is performed: 153 * 154 * . 155 * |-- a (ino 257) 156 * |-- b (ino 258) 157 * |-- c2 (ino 261) 158 * |-- d2 (ino 260) 159 * |-- cc (ino 259) 160 * 161 * The sequence of steps that lead to the second state was: 162 * 163 * mv /a/b/c/d /a/b/c2/d2 164 * mv /a/b/c /a/b/c2/d2/cc 165 * 166 * "c" has lower inode number, but we can't move it (2nd mv operation) 167 * before we move "d", which has higher inode number. 168 * 169 * So we just memorize which move/rename operations must be performed 170 * later when their respective parent is processed and moved/renamed. 171 */ 172 173 /* Indexed by parent directory inode number. */ 174 struct rb_root pending_dir_moves; 175 176 /* 177 * Reverse index, indexed by the inode number of a directory that 178 * is waiting for the move/rename of its immediate parent before its 179 * own move/rename can be performed. 180 */ 181 struct rb_root waiting_dir_moves; 182 183 /* 184 * A directory that is going to be rm'ed might have a child directory 185 * which is in the pending directory moves index above. In this case, 186 * the directory can only be removed after the move/rename of its child 187 * is performed. Example: 188 * 189 * Parent snapshot: 190 * 191 * . (ino 256) 192 * |-- a/ (ino 257) 193 * |-- b/ (ino 258) 194 * |-- c/ (ino 259) 195 * | |-- x/ (ino 260) 196 * | 197 * |-- y/ (ino 261) 198 * 199 * Send snapshot: 200 * 201 * . (ino 256) 202 * |-- a/ (ino 257) 203 * |-- b/ (ino 258) 204 * |-- YY/ (ino 261) 205 * |-- x/ (ino 260) 206 * 207 * Sequence of steps that lead to the send snapshot: 208 * rm -f /a/b/c/foo.txt 209 * mv /a/b/y /a/b/YY 210 * mv /a/b/c/x /a/b/YY 211 * rmdir /a/b/c 212 * 213 * When the child is processed, its move/rename is delayed until its 214 * parent is processed (as explained above), but all other operations 215 * like update utimes, chown, chgrp, etc, are performed and the paths 216 * that it uses for those operations must use the orphanized name of 217 * its parent (the directory we're going to rm later), so we need to 218 * memorize that name. 219 * 220 * Indexed by the inode number of the directory to be deleted. 221 */ 222 struct rb_root orphan_dirs; 223 }; 224 225 struct pending_dir_move { 226 struct rb_node node; 227 struct list_head list; 228 u64 parent_ino; 229 u64 ino; 230 u64 gen; 231 struct list_head update_refs; 232 }; 233 234 struct waiting_dir_move { 235 struct rb_node node; 236 u64 ino; 237 /* 238 * There might be some directory that could not be removed because it 239 * was waiting for this directory inode to be moved first. Therefore 240 * after this directory is moved, we can try to rmdir the ino rmdir_ino. 241 */ 242 u64 rmdir_ino; 243 bool orphanized; 244 }; 245 246 struct orphan_dir_info { 247 struct rb_node node; 248 u64 ino; 249 u64 gen; 250 }; 251 252 struct name_cache_entry { 253 struct list_head list; 254 /* 255 * radix_tree has only 32bit entries but we need to handle 64bit inums. 256 * We use the lower 32bit of the 64bit inum to store it in the tree. If 257 * more then one inum would fall into the same entry, we use radix_list 258 * to store the additional entries. radix_list is also used to store 259 * entries where two entries have the same inum but different 260 * generations. 261 */ 262 struct list_head radix_list; 263 u64 ino; 264 u64 gen; 265 u64 parent_ino; 266 u64 parent_gen; 267 int ret; 268 int need_later_update; 269 int name_len; 270 char name[]; 271 }; 272 273 static void inconsistent_snapshot_error(struct send_ctx *sctx, 274 enum btrfs_compare_tree_result result, 275 const char *what) 276 { 277 const char *result_string; 278 279 switch (result) { 280 case BTRFS_COMPARE_TREE_NEW: 281 result_string = "new"; 282 break; 283 case BTRFS_COMPARE_TREE_DELETED: 284 result_string = "deleted"; 285 break; 286 case BTRFS_COMPARE_TREE_CHANGED: 287 result_string = "updated"; 288 break; 289 case BTRFS_COMPARE_TREE_SAME: 290 ASSERT(0); 291 result_string = "unchanged"; 292 break; 293 default: 294 ASSERT(0); 295 result_string = "unexpected"; 296 } 297 298 btrfs_err(sctx->send_root->fs_info, 299 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu", 300 result_string, what, sctx->cmp_key->objectid, 301 sctx->send_root->root_key.objectid, 302 (sctx->parent_root ? 303 sctx->parent_root->root_key.objectid : 0)); 304 } 305 306 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); 307 308 static struct waiting_dir_move * 309 get_waiting_dir_move(struct send_ctx *sctx, u64 ino); 310 311 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino); 312 313 static int need_send_hole(struct send_ctx *sctx) 314 { 315 return (sctx->parent_root && !sctx->cur_inode_new && 316 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && 317 S_ISREG(sctx->cur_inode_mode)); 318 } 319 320 static void fs_path_reset(struct fs_path *p) 321 { 322 if (p->reversed) { 323 p->start = p->buf + p->buf_len - 1; 324 p->end = p->start; 325 *p->start = 0; 326 } else { 327 p->start = p->buf; 328 p->end = p->start; 329 *p->start = 0; 330 } 331 } 332 333 static struct fs_path *fs_path_alloc(void) 334 { 335 struct fs_path *p; 336 337 p = kmalloc(sizeof(*p), GFP_KERNEL); 338 if (!p) 339 return NULL; 340 p->reversed = 0; 341 p->buf = p->inline_buf; 342 p->buf_len = FS_PATH_INLINE_SIZE; 343 fs_path_reset(p); 344 return p; 345 } 346 347 static struct fs_path *fs_path_alloc_reversed(void) 348 { 349 struct fs_path *p; 350 351 p = fs_path_alloc(); 352 if (!p) 353 return NULL; 354 p->reversed = 1; 355 fs_path_reset(p); 356 return p; 357 } 358 359 static void fs_path_free(struct fs_path *p) 360 { 361 if (!p) 362 return; 363 if (p->buf != p->inline_buf) 364 kfree(p->buf); 365 kfree(p); 366 } 367 368 static int fs_path_len(struct fs_path *p) 369 { 370 return p->end - p->start; 371 } 372 373 static int fs_path_ensure_buf(struct fs_path *p, int len) 374 { 375 char *tmp_buf; 376 int path_len; 377 int old_buf_len; 378 379 len++; 380 381 if (p->buf_len >= len) 382 return 0; 383 384 if (len > PATH_MAX) { 385 WARN_ON(1); 386 return -ENOMEM; 387 } 388 389 path_len = p->end - p->start; 390 old_buf_len = p->buf_len; 391 392 /* 393 * First time the inline_buf does not suffice 394 */ 395 if (p->buf == p->inline_buf) { 396 tmp_buf = kmalloc(len, GFP_KERNEL); 397 if (tmp_buf) 398 memcpy(tmp_buf, p->buf, old_buf_len); 399 } else { 400 tmp_buf = krealloc(p->buf, len, GFP_KERNEL); 401 } 402 if (!tmp_buf) 403 return -ENOMEM; 404 p->buf = tmp_buf; 405 /* 406 * The real size of the buffer is bigger, this will let the fast path 407 * happen most of the time 408 */ 409 p->buf_len = ksize(p->buf); 410 411 if (p->reversed) { 412 tmp_buf = p->buf + old_buf_len - path_len - 1; 413 p->end = p->buf + p->buf_len - 1; 414 p->start = p->end - path_len; 415 memmove(p->start, tmp_buf, path_len + 1); 416 } else { 417 p->start = p->buf; 418 p->end = p->start + path_len; 419 } 420 return 0; 421 } 422 423 static int fs_path_prepare_for_add(struct fs_path *p, int name_len, 424 char **prepared) 425 { 426 int ret; 427 int new_len; 428 429 new_len = p->end - p->start + name_len; 430 if (p->start != p->end) 431 new_len++; 432 ret = fs_path_ensure_buf(p, new_len); 433 if (ret < 0) 434 goto out; 435 436 if (p->reversed) { 437 if (p->start != p->end) 438 *--p->start = '/'; 439 p->start -= name_len; 440 *prepared = p->start; 441 } else { 442 if (p->start != p->end) 443 *p->end++ = '/'; 444 *prepared = p->end; 445 p->end += name_len; 446 *p->end = 0; 447 } 448 449 out: 450 return ret; 451 } 452 453 static int fs_path_add(struct fs_path *p, const char *name, int name_len) 454 { 455 int ret; 456 char *prepared; 457 458 ret = fs_path_prepare_for_add(p, name_len, &prepared); 459 if (ret < 0) 460 goto out; 461 memcpy(prepared, name, name_len); 462 463 out: 464 return ret; 465 } 466 467 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) 468 { 469 int ret; 470 char *prepared; 471 472 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); 473 if (ret < 0) 474 goto out; 475 memcpy(prepared, p2->start, p2->end - p2->start); 476 477 out: 478 return ret; 479 } 480 481 static int fs_path_add_from_extent_buffer(struct fs_path *p, 482 struct extent_buffer *eb, 483 unsigned long off, int len) 484 { 485 int ret; 486 char *prepared; 487 488 ret = fs_path_prepare_for_add(p, len, &prepared); 489 if (ret < 0) 490 goto out; 491 492 read_extent_buffer(eb, prepared, off, len); 493 494 out: 495 return ret; 496 } 497 498 static int fs_path_copy(struct fs_path *p, struct fs_path *from) 499 { 500 int ret; 501 502 p->reversed = from->reversed; 503 fs_path_reset(p); 504 505 ret = fs_path_add_path(p, from); 506 507 return ret; 508 } 509 510 511 static void fs_path_unreverse(struct fs_path *p) 512 { 513 char *tmp; 514 int len; 515 516 if (!p->reversed) 517 return; 518 519 tmp = p->start; 520 len = p->end - p->start; 521 p->start = p->buf; 522 p->end = p->start + len; 523 memmove(p->start, tmp, len + 1); 524 p->reversed = 0; 525 } 526 527 static struct btrfs_path *alloc_path_for_send(void) 528 { 529 struct btrfs_path *path; 530 531 path = btrfs_alloc_path(); 532 if (!path) 533 return NULL; 534 path->search_commit_root = 1; 535 path->skip_locking = 1; 536 path->need_commit_sem = 1; 537 return path; 538 } 539 540 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) 541 { 542 int ret; 543 u32 pos = 0; 544 545 while (pos < len) { 546 ret = kernel_write(filp, buf + pos, len - pos, off); 547 /* TODO handle that correctly */ 548 /*if (ret == -ERESTARTSYS) { 549 continue; 550 }*/ 551 if (ret < 0) 552 return ret; 553 if (ret == 0) { 554 return -EIO; 555 } 556 pos += ret; 557 } 558 559 return 0; 560 } 561 562 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) 563 { 564 struct btrfs_tlv_header *hdr; 565 int total_len = sizeof(*hdr) + len; 566 int left = sctx->send_max_size - sctx->send_size; 567 568 if (unlikely(left < total_len)) 569 return -EOVERFLOW; 570 571 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); 572 hdr->tlv_type = cpu_to_le16(attr); 573 hdr->tlv_len = cpu_to_le16(len); 574 memcpy(hdr + 1, data, len); 575 sctx->send_size += total_len; 576 577 return 0; 578 } 579 580 #define TLV_PUT_DEFINE_INT(bits) \ 581 static int tlv_put_u##bits(struct send_ctx *sctx, \ 582 u##bits attr, u##bits value) \ 583 { \ 584 __le##bits __tmp = cpu_to_le##bits(value); \ 585 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ 586 } 587 588 TLV_PUT_DEFINE_INT(64) 589 590 static int tlv_put_string(struct send_ctx *sctx, u16 attr, 591 const char *str, int len) 592 { 593 if (len == -1) 594 len = strlen(str); 595 return tlv_put(sctx, attr, str, len); 596 } 597 598 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, 599 const u8 *uuid) 600 { 601 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); 602 } 603 604 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, 605 struct extent_buffer *eb, 606 struct btrfs_timespec *ts) 607 { 608 struct btrfs_timespec bts; 609 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); 610 return tlv_put(sctx, attr, &bts, sizeof(bts)); 611 } 612 613 614 #define TLV_PUT(sctx, attrtype, attrlen, data) \ 615 do { \ 616 ret = tlv_put(sctx, attrtype, attrlen, data); \ 617 if (ret < 0) \ 618 goto tlv_put_failure; \ 619 } while (0) 620 621 #define TLV_PUT_INT(sctx, attrtype, bits, value) \ 622 do { \ 623 ret = tlv_put_u##bits(sctx, attrtype, value); \ 624 if (ret < 0) \ 625 goto tlv_put_failure; \ 626 } while (0) 627 628 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) 629 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) 630 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) 631 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) 632 #define TLV_PUT_STRING(sctx, attrtype, str, len) \ 633 do { \ 634 ret = tlv_put_string(sctx, attrtype, str, len); \ 635 if (ret < 0) \ 636 goto tlv_put_failure; \ 637 } while (0) 638 #define TLV_PUT_PATH(sctx, attrtype, p) \ 639 do { \ 640 ret = tlv_put_string(sctx, attrtype, p->start, \ 641 p->end - p->start); \ 642 if (ret < 0) \ 643 goto tlv_put_failure; \ 644 } while(0) 645 #define TLV_PUT_UUID(sctx, attrtype, uuid) \ 646 do { \ 647 ret = tlv_put_uuid(sctx, attrtype, uuid); \ 648 if (ret < 0) \ 649 goto tlv_put_failure; \ 650 } while (0) 651 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ 652 do { \ 653 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ 654 if (ret < 0) \ 655 goto tlv_put_failure; \ 656 } while (0) 657 658 static int send_header(struct send_ctx *sctx) 659 { 660 struct btrfs_stream_header hdr; 661 662 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); 663 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); 664 665 return write_buf(sctx->send_filp, &hdr, sizeof(hdr), 666 &sctx->send_off); 667 } 668 669 /* 670 * For each command/item we want to send to userspace, we call this function. 671 */ 672 static int begin_cmd(struct send_ctx *sctx, int cmd) 673 { 674 struct btrfs_cmd_header *hdr; 675 676 if (WARN_ON(!sctx->send_buf)) 677 return -EINVAL; 678 679 BUG_ON(sctx->send_size); 680 681 sctx->send_size += sizeof(*hdr); 682 hdr = (struct btrfs_cmd_header *)sctx->send_buf; 683 hdr->cmd = cpu_to_le16(cmd); 684 685 return 0; 686 } 687 688 static int send_cmd(struct send_ctx *sctx) 689 { 690 int ret; 691 struct btrfs_cmd_header *hdr; 692 u32 crc; 693 694 hdr = (struct btrfs_cmd_header *)sctx->send_buf; 695 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr)); 696 hdr->crc = 0; 697 698 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); 699 hdr->crc = cpu_to_le32(crc); 700 701 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, 702 &sctx->send_off); 703 704 sctx->total_send_size += sctx->send_size; 705 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size; 706 sctx->send_size = 0; 707 708 return ret; 709 } 710 711 /* 712 * Sends a move instruction to user space 713 */ 714 static int send_rename(struct send_ctx *sctx, 715 struct fs_path *from, struct fs_path *to) 716 { 717 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 718 int ret; 719 720 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start); 721 722 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); 723 if (ret < 0) 724 goto out; 725 726 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); 727 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); 728 729 ret = send_cmd(sctx); 730 731 tlv_put_failure: 732 out: 733 return ret; 734 } 735 736 /* 737 * Sends a link instruction to user space 738 */ 739 static int send_link(struct send_ctx *sctx, 740 struct fs_path *path, struct fs_path *lnk) 741 { 742 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 743 int ret; 744 745 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start); 746 747 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); 748 if (ret < 0) 749 goto out; 750 751 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); 752 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); 753 754 ret = send_cmd(sctx); 755 756 tlv_put_failure: 757 out: 758 return ret; 759 } 760 761 /* 762 * Sends an unlink instruction to user space 763 */ 764 static int send_unlink(struct send_ctx *sctx, struct fs_path *path) 765 { 766 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 767 int ret; 768 769 btrfs_debug(fs_info, "send_unlink %s", path->start); 770 771 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); 772 if (ret < 0) 773 goto out; 774 775 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); 776 777 ret = send_cmd(sctx); 778 779 tlv_put_failure: 780 out: 781 return ret; 782 } 783 784 /* 785 * Sends a rmdir instruction to user space 786 */ 787 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) 788 { 789 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 790 int ret; 791 792 btrfs_debug(fs_info, "send_rmdir %s", path->start); 793 794 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); 795 if (ret < 0) 796 goto out; 797 798 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); 799 800 ret = send_cmd(sctx); 801 802 tlv_put_failure: 803 out: 804 return ret; 805 } 806 807 /* 808 * Helper function to retrieve some fields from an inode item. 809 */ 810 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, 811 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, 812 u64 *gid, u64 *rdev) 813 { 814 int ret; 815 struct btrfs_inode_item *ii; 816 struct btrfs_key key; 817 818 key.objectid = ino; 819 key.type = BTRFS_INODE_ITEM_KEY; 820 key.offset = 0; 821 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 822 if (ret) { 823 if (ret > 0) 824 ret = -ENOENT; 825 return ret; 826 } 827 828 ii = btrfs_item_ptr(path->nodes[0], path->slots[0], 829 struct btrfs_inode_item); 830 if (size) 831 *size = btrfs_inode_size(path->nodes[0], ii); 832 if (gen) 833 *gen = btrfs_inode_generation(path->nodes[0], ii); 834 if (mode) 835 *mode = btrfs_inode_mode(path->nodes[0], ii); 836 if (uid) 837 *uid = btrfs_inode_uid(path->nodes[0], ii); 838 if (gid) 839 *gid = btrfs_inode_gid(path->nodes[0], ii); 840 if (rdev) 841 *rdev = btrfs_inode_rdev(path->nodes[0], ii); 842 843 return ret; 844 } 845 846 static int get_inode_info(struct btrfs_root *root, 847 u64 ino, u64 *size, u64 *gen, 848 u64 *mode, u64 *uid, u64 *gid, 849 u64 *rdev) 850 { 851 struct btrfs_path *path; 852 int ret; 853 854 path = alloc_path_for_send(); 855 if (!path) 856 return -ENOMEM; 857 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, 858 rdev); 859 btrfs_free_path(path); 860 return ret; 861 } 862 863 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, 864 struct fs_path *p, 865 void *ctx); 866 867 /* 868 * Helper function to iterate the entries in ONE btrfs_inode_ref or 869 * btrfs_inode_extref. 870 * The iterate callback may return a non zero value to stop iteration. This can 871 * be a negative value for error codes or 1 to simply stop it. 872 * 873 * path must point to the INODE_REF or INODE_EXTREF when called. 874 */ 875 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, 876 struct btrfs_key *found_key, int resolve, 877 iterate_inode_ref_t iterate, void *ctx) 878 { 879 struct extent_buffer *eb = path->nodes[0]; 880 struct btrfs_item *item; 881 struct btrfs_inode_ref *iref; 882 struct btrfs_inode_extref *extref; 883 struct btrfs_path *tmp_path; 884 struct fs_path *p; 885 u32 cur = 0; 886 u32 total; 887 int slot = path->slots[0]; 888 u32 name_len; 889 char *start; 890 int ret = 0; 891 int num = 0; 892 int index; 893 u64 dir; 894 unsigned long name_off; 895 unsigned long elem_size; 896 unsigned long ptr; 897 898 p = fs_path_alloc_reversed(); 899 if (!p) 900 return -ENOMEM; 901 902 tmp_path = alloc_path_for_send(); 903 if (!tmp_path) { 904 fs_path_free(p); 905 return -ENOMEM; 906 } 907 908 909 if (found_key->type == BTRFS_INODE_REF_KEY) { 910 ptr = (unsigned long)btrfs_item_ptr(eb, slot, 911 struct btrfs_inode_ref); 912 item = btrfs_item_nr(slot); 913 total = btrfs_item_size(eb, item); 914 elem_size = sizeof(*iref); 915 } else { 916 ptr = btrfs_item_ptr_offset(eb, slot); 917 total = btrfs_item_size_nr(eb, slot); 918 elem_size = sizeof(*extref); 919 } 920 921 while (cur < total) { 922 fs_path_reset(p); 923 924 if (found_key->type == BTRFS_INODE_REF_KEY) { 925 iref = (struct btrfs_inode_ref *)(ptr + cur); 926 name_len = btrfs_inode_ref_name_len(eb, iref); 927 name_off = (unsigned long)(iref + 1); 928 index = btrfs_inode_ref_index(eb, iref); 929 dir = found_key->offset; 930 } else { 931 extref = (struct btrfs_inode_extref *)(ptr + cur); 932 name_len = btrfs_inode_extref_name_len(eb, extref); 933 name_off = (unsigned long)&extref->name; 934 index = btrfs_inode_extref_index(eb, extref); 935 dir = btrfs_inode_extref_parent(eb, extref); 936 } 937 938 if (resolve) { 939 start = btrfs_ref_to_path(root, tmp_path, name_len, 940 name_off, eb, dir, 941 p->buf, p->buf_len); 942 if (IS_ERR(start)) { 943 ret = PTR_ERR(start); 944 goto out; 945 } 946 if (start < p->buf) { 947 /* overflow , try again with larger buffer */ 948 ret = fs_path_ensure_buf(p, 949 p->buf_len + p->buf - start); 950 if (ret < 0) 951 goto out; 952 start = btrfs_ref_to_path(root, tmp_path, 953 name_len, name_off, 954 eb, dir, 955 p->buf, p->buf_len); 956 if (IS_ERR(start)) { 957 ret = PTR_ERR(start); 958 goto out; 959 } 960 BUG_ON(start < p->buf); 961 } 962 p->start = start; 963 } else { 964 ret = fs_path_add_from_extent_buffer(p, eb, name_off, 965 name_len); 966 if (ret < 0) 967 goto out; 968 } 969 970 cur += elem_size + name_len; 971 ret = iterate(num, dir, index, p, ctx); 972 if (ret) 973 goto out; 974 num++; 975 } 976 977 out: 978 btrfs_free_path(tmp_path); 979 fs_path_free(p); 980 return ret; 981 } 982 983 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, 984 const char *name, int name_len, 985 const char *data, int data_len, 986 u8 type, void *ctx); 987 988 /* 989 * Helper function to iterate the entries in ONE btrfs_dir_item. 990 * The iterate callback may return a non zero value to stop iteration. This can 991 * be a negative value for error codes or 1 to simply stop it. 992 * 993 * path must point to the dir item when called. 994 */ 995 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, 996 iterate_dir_item_t iterate, void *ctx) 997 { 998 int ret = 0; 999 struct extent_buffer *eb; 1000 struct btrfs_item *item; 1001 struct btrfs_dir_item *di; 1002 struct btrfs_key di_key; 1003 char *buf = NULL; 1004 int buf_len; 1005 u32 name_len; 1006 u32 data_len; 1007 u32 cur; 1008 u32 len; 1009 u32 total; 1010 int slot; 1011 int num; 1012 u8 type; 1013 1014 /* 1015 * Start with a small buffer (1 page). If later we end up needing more 1016 * space, which can happen for xattrs on a fs with a leaf size greater 1017 * then the page size, attempt to increase the buffer. Typically xattr 1018 * values are small. 1019 */ 1020 buf_len = PATH_MAX; 1021 buf = kmalloc(buf_len, GFP_KERNEL); 1022 if (!buf) { 1023 ret = -ENOMEM; 1024 goto out; 1025 } 1026 1027 eb = path->nodes[0]; 1028 slot = path->slots[0]; 1029 item = btrfs_item_nr(slot); 1030 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); 1031 cur = 0; 1032 len = 0; 1033 total = btrfs_item_size(eb, item); 1034 1035 num = 0; 1036 while (cur < total) { 1037 name_len = btrfs_dir_name_len(eb, di); 1038 data_len = btrfs_dir_data_len(eb, di); 1039 type = btrfs_dir_type(eb, di); 1040 btrfs_dir_item_key_to_cpu(eb, di, &di_key); 1041 1042 if (type == BTRFS_FT_XATTR) { 1043 if (name_len > XATTR_NAME_MAX) { 1044 ret = -ENAMETOOLONG; 1045 goto out; 1046 } 1047 if (name_len + data_len > 1048 BTRFS_MAX_XATTR_SIZE(root->fs_info)) { 1049 ret = -E2BIG; 1050 goto out; 1051 } 1052 } else { 1053 /* 1054 * Path too long 1055 */ 1056 if (name_len + data_len > PATH_MAX) { 1057 ret = -ENAMETOOLONG; 1058 goto out; 1059 } 1060 } 1061 1062 ret = btrfs_is_name_len_valid(eb, path->slots[0], 1063 (unsigned long)(di + 1), name_len + data_len); 1064 if (!ret) { 1065 ret = -EIO; 1066 goto out; 1067 } 1068 if (name_len + data_len > buf_len) { 1069 buf_len = name_len + data_len; 1070 if (is_vmalloc_addr(buf)) { 1071 vfree(buf); 1072 buf = NULL; 1073 } else { 1074 char *tmp = krealloc(buf, buf_len, 1075 GFP_KERNEL | __GFP_NOWARN); 1076 1077 if (!tmp) 1078 kfree(buf); 1079 buf = tmp; 1080 } 1081 if (!buf) { 1082 buf = kvmalloc(buf_len, GFP_KERNEL); 1083 if (!buf) { 1084 ret = -ENOMEM; 1085 goto out; 1086 } 1087 } 1088 } 1089 1090 read_extent_buffer(eb, buf, (unsigned long)(di + 1), 1091 name_len + data_len); 1092 1093 len = sizeof(*di) + name_len + data_len; 1094 di = (struct btrfs_dir_item *)((char *)di + len); 1095 cur += len; 1096 1097 ret = iterate(num, &di_key, buf, name_len, buf + name_len, 1098 data_len, type, ctx); 1099 if (ret < 0) 1100 goto out; 1101 if (ret) { 1102 ret = 0; 1103 goto out; 1104 } 1105 1106 num++; 1107 } 1108 1109 out: 1110 kvfree(buf); 1111 return ret; 1112 } 1113 1114 static int __copy_first_ref(int num, u64 dir, int index, 1115 struct fs_path *p, void *ctx) 1116 { 1117 int ret; 1118 struct fs_path *pt = ctx; 1119 1120 ret = fs_path_copy(pt, p); 1121 if (ret < 0) 1122 return ret; 1123 1124 /* we want the first only */ 1125 return 1; 1126 } 1127 1128 /* 1129 * Retrieve the first path of an inode. If an inode has more then one 1130 * ref/hardlink, this is ignored. 1131 */ 1132 static int get_inode_path(struct btrfs_root *root, 1133 u64 ino, struct fs_path *path) 1134 { 1135 int ret; 1136 struct btrfs_key key, found_key; 1137 struct btrfs_path *p; 1138 1139 p = alloc_path_for_send(); 1140 if (!p) 1141 return -ENOMEM; 1142 1143 fs_path_reset(path); 1144 1145 key.objectid = ino; 1146 key.type = BTRFS_INODE_REF_KEY; 1147 key.offset = 0; 1148 1149 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); 1150 if (ret < 0) 1151 goto out; 1152 if (ret) { 1153 ret = 1; 1154 goto out; 1155 } 1156 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); 1157 if (found_key.objectid != ino || 1158 (found_key.type != BTRFS_INODE_REF_KEY && 1159 found_key.type != BTRFS_INODE_EXTREF_KEY)) { 1160 ret = -ENOENT; 1161 goto out; 1162 } 1163 1164 ret = iterate_inode_ref(root, p, &found_key, 1, 1165 __copy_first_ref, path); 1166 if (ret < 0) 1167 goto out; 1168 ret = 0; 1169 1170 out: 1171 btrfs_free_path(p); 1172 return ret; 1173 } 1174 1175 struct backref_ctx { 1176 struct send_ctx *sctx; 1177 1178 struct btrfs_path *path; 1179 /* number of total found references */ 1180 u64 found; 1181 1182 /* 1183 * used for clones found in send_root. clones found behind cur_objectid 1184 * and cur_offset are not considered as allowed clones. 1185 */ 1186 u64 cur_objectid; 1187 u64 cur_offset; 1188 1189 /* may be truncated in case it's the last extent in a file */ 1190 u64 extent_len; 1191 1192 /* data offset in the file extent item */ 1193 u64 data_offset; 1194 1195 /* Just to check for bugs in backref resolving */ 1196 int found_itself; 1197 }; 1198 1199 static int __clone_root_cmp_bsearch(const void *key, const void *elt) 1200 { 1201 u64 root = (u64)(uintptr_t)key; 1202 struct clone_root *cr = (struct clone_root *)elt; 1203 1204 if (root < cr->root->objectid) 1205 return -1; 1206 if (root > cr->root->objectid) 1207 return 1; 1208 return 0; 1209 } 1210 1211 static int __clone_root_cmp_sort(const void *e1, const void *e2) 1212 { 1213 struct clone_root *cr1 = (struct clone_root *)e1; 1214 struct clone_root *cr2 = (struct clone_root *)e2; 1215 1216 if (cr1->root->objectid < cr2->root->objectid) 1217 return -1; 1218 if (cr1->root->objectid > cr2->root->objectid) 1219 return 1; 1220 return 0; 1221 } 1222 1223 /* 1224 * Called for every backref that is found for the current extent. 1225 * Results are collected in sctx->clone_roots->ino/offset/found_refs 1226 */ 1227 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) 1228 { 1229 struct backref_ctx *bctx = ctx_; 1230 struct clone_root *found; 1231 int ret; 1232 u64 i_size; 1233 1234 /* First check if the root is in the list of accepted clone sources */ 1235 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, 1236 bctx->sctx->clone_roots_cnt, 1237 sizeof(struct clone_root), 1238 __clone_root_cmp_bsearch); 1239 if (!found) 1240 return 0; 1241 1242 if (found->root == bctx->sctx->send_root && 1243 ino == bctx->cur_objectid && 1244 offset == bctx->cur_offset) { 1245 bctx->found_itself = 1; 1246 } 1247 1248 /* 1249 * There are inodes that have extents that lie behind its i_size. Don't 1250 * accept clones from these extents. 1251 */ 1252 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL, 1253 NULL, NULL, NULL); 1254 btrfs_release_path(bctx->path); 1255 if (ret < 0) 1256 return ret; 1257 1258 if (offset + bctx->data_offset + bctx->extent_len > i_size) 1259 return 0; 1260 1261 /* 1262 * Make sure we don't consider clones from send_root that are 1263 * behind the current inode/offset. 1264 */ 1265 if (found->root == bctx->sctx->send_root) { 1266 /* 1267 * TODO for the moment we don't accept clones from the inode 1268 * that is currently send. We may change this when 1269 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same 1270 * file. 1271 */ 1272 if (ino >= bctx->cur_objectid) 1273 return 0; 1274 } 1275 1276 bctx->found++; 1277 found->found_refs++; 1278 if (ino < found->ino) { 1279 found->ino = ino; 1280 found->offset = offset; 1281 } else if (found->ino == ino) { 1282 /* 1283 * same extent found more then once in the same file. 1284 */ 1285 if (found->offset > offset + bctx->extent_len) 1286 found->offset = offset; 1287 } 1288 1289 return 0; 1290 } 1291 1292 /* 1293 * Given an inode, offset and extent item, it finds a good clone for a clone 1294 * instruction. Returns -ENOENT when none could be found. The function makes 1295 * sure that the returned clone is usable at the point where sending is at the 1296 * moment. This means, that no clones are accepted which lie behind the current 1297 * inode+offset. 1298 * 1299 * path must point to the extent item when called. 1300 */ 1301 static int find_extent_clone(struct send_ctx *sctx, 1302 struct btrfs_path *path, 1303 u64 ino, u64 data_offset, 1304 u64 ino_size, 1305 struct clone_root **found) 1306 { 1307 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 1308 int ret; 1309 int extent_type; 1310 u64 logical; 1311 u64 disk_byte; 1312 u64 num_bytes; 1313 u64 extent_item_pos; 1314 u64 flags = 0; 1315 struct btrfs_file_extent_item *fi; 1316 struct extent_buffer *eb = path->nodes[0]; 1317 struct backref_ctx *backref_ctx = NULL; 1318 struct clone_root *cur_clone_root; 1319 struct btrfs_key found_key; 1320 struct btrfs_path *tmp_path; 1321 int compressed; 1322 u32 i; 1323 1324 tmp_path = alloc_path_for_send(); 1325 if (!tmp_path) 1326 return -ENOMEM; 1327 1328 /* We only use this path under the commit sem */ 1329 tmp_path->need_commit_sem = 0; 1330 1331 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL); 1332 if (!backref_ctx) { 1333 ret = -ENOMEM; 1334 goto out; 1335 } 1336 1337 backref_ctx->path = tmp_path; 1338 1339 if (data_offset >= ino_size) { 1340 /* 1341 * There may be extents that lie behind the file's size. 1342 * I at least had this in combination with snapshotting while 1343 * writing large files. 1344 */ 1345 ret = 0; 1346 goto out; 1347 } 1348 1349 fi = btrfs_item_ptr(eb, path->slots[0], 1350 struct btrfs_file_extent_item); 1351 extent_type = btrfs_file_extent_type(eb, fi); 1352 if (extent_type == BTRFS_FILE_EXTENT_INLINE) { 1353 ret = -ENOENT; 1354 goto out; 1355 } 1356 compressed = btrfs_file_extent_compression(eb, fi); 1357 1358 num_bytes = btrfs_file_extent_num_bytes(eb, fi); 1359 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); 1360 if (disk_byte == 0) { 1361 ret = -ENOENT; 1362 goto out; 1363 } 1364 logical = disk_byte + btrfs_file_extent_offset(eb, fi); 1365 1366 down_read(&fs_info->commit_root_sem); 1367 ret = extent_from_logical(fs_info, disk_byte, tmp_path, 1368 &found_key, &flags); 1369 up_read(&fs_info->commit_root_sem); 1370 btrfs_release_path(tmp_path); 1371 1372 if (ret < 0) 1373 goto out; 1374 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 1375 ret = -EIO; 1376 goto out; 1377 } 1378 1379 /* 1380 * Setup the clone roots. 1381 */ 1382 for (i = 0; i < sctx->clone_roots_cnt; i++) { 1383 cur_clone_root = sctx->clone_roots + i; 1384 cur_clone_root->ino = (u64)-1; 1385 cur_clone_root->offset = 0; 1386 cur_clone_root->found_refs = 0; 1387 } 1388 1389 backref_ctx->sctx = sctx; 1390 backref_ctx->found = 0; 1391 backref_ctx->cur_objectid = ino; 1392 backref_ctx->cur_offset = data_offset; 1393 backref_ctx->found_itself = 0; 1394 backref_ctx->extent_len = num_bytes; 1395 /* 1396 * For non-compressed extents iterate_extent_inodes() gives us extent 1397 * offsets that already take into account the data offset, but not for 1398 * compressed extents, since the offset is logical and not relative to 1399 * the physical extent locations. We must take this into account to 1400 * avoid sending clone offsets that go beyond the source file's size, 1401 * which would result in the clone ioctl failing with -EINVAL on the 1402 * receiving end. 1403 */ 1404 if (compressed == BTRFS_COMPRESS_NONE) 1405 backref_ctx->data_offset = 0; 1406 else 1407 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi); 1408 1409 /* 1410 * The last extent of a file may be too large due to page alignment. 1411 * We need to adjust extent_len in this case so that the checks in 1412 * __iterate_backrefs work. 1413 */ 1414 if (data_offset + num_bytes >= ino_size) 1415 backref_ctx->extent_len = ino_size - data_offset; 1416 1417 /* 1418 * Now collect all backrefs. 1419 */ 1420 if (compressed == BTRFS_COMPRESS_NONE) 1421 extent_item_pos = logical - found_key.objectid; 1422 else 1423 extent_item_pos = 0; 1424 ret = iterate_extent_inodes(fs_info, found_key.objectid, 1425 extent_item_pos, 1, __iterate_backrefs, 1426 backref_ctx, false); 1427 1428 if (ret < 0) 1429 goto out; 1430 1431 if (!backref_ctx->found_itself) { 1432 /* found a bug in backref code? */ 1433 ret = -EIO; 1434 btrfs_err(fs_info, 1435 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu", 1436 ino, data_offset, disk_byte, found_key.objectid); 1437 goto out; 1438 } 1439 1440 btrfs_debug(fs_info, 1441 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu", 1442 data_offset, ino, num_bytes, logical); 1443 1444 if (!backref_ctx->found) 1445 btrfs_debug(fs_info, "no clones found"); 1446 1447 cur_clone_root = NULL; 1448 for (i = 0; i < sctx->clone_roots_cnt; i++) { 1449 if (sctx->clone_roots[i].found_refs) { 1450 if (!cur_clone_root) 1451 cur_clone_root = sctx->clone_roots + i; 1452 else if (sctx->clone_roots[i].root == sctx->send_root) 1453 /* prefer clones from send_root over others */ 1454 cur_clone_root = sctx->clone_roots + i; 1455 } 1456 1457 } 1458 1459 if (cur_clone_root) { 1460 *found = cur_clone_root; 1461 ret = 0; 1462 } else { 1463 ret = -ENOENT; 1464 } 1465 1466 out: 1467 btrfs_free_path(tmp_path); 1468 kfree(backref_ctx); 1469 return ret; 1470 } 1471 1472 static int read_symlink(struct btrfs_root *root, 1473 u64 ino, 1474 struct fs_path *dest) 1475 { 1476 int ret; 1477 struct btrfs_path *path; 1478 struct btrfs_key key; 1479 struct btrfs_file_extent_item *ei; 1480 u8 type; 1481 u8 compression; 1482 unsigned long off; 1483 int len; 1484 1485 path = alloc_path_for_send(); 1486 if (!path) 1487 return -ENOMEM; 1488 1489 key.objectid = ino; 1490 key.type = BTRFS_EXTENT_DATA_KEY; 1491 key.offset = 0; 1492 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 1493 if (ret < 0) 1494 goto out; 1495 if (ret) { 1496 /* 1497 * An empty symlink inode. Can happen in rare error paths when 1498 * creating a symlink (transaction committed before the inode 1499 * eviction handler removed the symlink inode items and a crash 1500 * happened in between or the subvol was snapshoted in between). 1501 * Print an informative message to dmesg/syslog so that the user 1502 * can delete the symlink. 1503 */ 1504 btrfs_err(root->fs_info, 1505 "Found empty symlink inode %llu at root %llu", 1506 ino, root->root_key.objectid); 1507 ret = -EIO; 1508 goto out; 1509 } 1510 1511 ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 1512 struct btrfs_file_extent_item); 1513 type = btrfs_file_extent_type(path->nodes[0], ei); 1514 compression = btrfs_file_extent_compression(path->nodes[0], ei); 1515 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); 1516 BUG_ON(compression); 1517 1518 off = btrfs_file_extent_inline_start(ei); 1519 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei); 1520 1521 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); 1522 1523 out: 1524 btrfs_free_path(path); 1525 return ret; 1526 } 1527 1528 /* 1529 * Helper function to generate a file name that is unique in the root of 1530 * send_root and parent_root. This is used to generate names for orphan inodes. 1531 */ 1532 static int gen_unique_name(struct send_ctx *sctx, 1533 u64 ino, u64 gen, 1534 struct fs_path *dest) 1535 { 1536 int ret = 0; 1537 struct btrfs_path *path; 1538 struct btrfs_dir_item *di; 1539 char tmp[64]; 1540 int len; 1541 u64 idx = 0; 1542 1543 path = alloc_path_for_send(); 1544 if (!path) 1545 return -ENOMEM; 1546 1547 while (1) { 1548 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", 1549 ino, gen, idx); 1550 ASSERT(len < sizeof(tmp)); 1551 1552 di = btrfs_lookup_dir_item(NULL, sctx->send_root, 1553 path, BTRFS_FIRST_FREE_OBJECTID, 1554 tmp, strlen(tmp), 0); 1555 btrfs_release_path(path); 1556 if (IS_ERR(di)) { 1557 ret = PTR_ERR(di); 1558 goto out; 1559 } 1560 if (di) { 1561 /* not unique, try again */ 1562 idx++; 1563 continue; 1564 } 1565 1566 if (!sctx->parent_root) { 1567 /* unique */ 1568 ret = 0; 1569 break; 1570 } 1571 1572 di = btrfs_lookup_dir_item(NULL, sctx->parent_root, 1573 path, BTRFS_FIRST_FREE_OBJECTID, 1574 tmp, strlen(tmp), 0); 1575 btrfs_release_path(path); 1576 if (IS_ERR(di)) { 1577 ret = PTR_ERR(di); 1578 goto out; 1579 } 1580 if (di) { 1581 /* not unique, try again */ 1582 idx++; 1583 continue; 1584 } 1585 /* unique */ 1586 break; 1587 } 1588 1589 ret = fs_path_add(dest, tmp, strlen(tmp)); 1590 1591 out: 1592 btrfs_free_path(path); 1593 return ret; 1594 } 1595 1596 enum inode_state { 1597 inode_state_no_change, 1598 inode_state_will_create, 1599 inode_state_did_create, 1600 inode_state_will_delete, 1601 inode_state_did_delete, 1602 }; 1603 1604 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) 1605 { 1606 int ret; 1607 int left_ret; 1608 int right_ret; 1609 u64 left_gen; 1610 u64 right_gen; 1611 1612 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, 1613 NULL, NULL); 1614 if (ret < 0 && ret != -ENOENT) 1615 goto out; 1616 left_ret = ret; 1617 1618 if (!sctx->parent_root) { 1619 right_ret = -ENOENT; 1620 } else { 1621 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, 1622 NULL, NULL, NULL, NULL); 1623 if (ret < 0 && ret != -ENOENT) 1624 goto out; 1625 right_ret = ret; 1626 } 1627 1628 if (!left_ret && !right_ret) { 1629 if (left_gen == gen && right_gen == gen) { 1630 ret = inode_state_no_change; 1631 } else if (left_gen == gen) { 1632 if (ino < sctx->send_progress) 1633 ret = inode_state_did_create; 1634 else 1635 ret = inode_state_will_create; 1636 } else if (right_gen == gen) { 1637 if (ino < sctx->send_progress) 1638 ret = inode_state_did_delete; 1639 else 1640 ret = inode_state_will_delete; 1641 } else { 1642 ret = -ENOENT; 1643 } 1644 } else if (!left_ret) { 1645 if (left_gen == gen) { 1646 if (ino < sctx->send_progress) 1647 ret = inode_state_did_create; 1648 else 1649 ret = inode_state_will_create; 1650 } else { 1651 ret = -ENOENT; 1652 } 1653 } else if (!right_ret) { 1654 if (right_gen == gen) { 1655 if (ino < sctx->send_progress) 1656 ret = inode_state_did_delete; 1657 else 1658 ret = inode_state_will_delete; 1659 } else { 1660 ret = -ENOENT; 1661 } 1662 } else { 1663 ret = -ENOENT; 1664 } 1665 1666 out: 1667 return ret; 1668 } 1669 1670 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) 1671 { 1672 int ret; 1673 1674 if (ino == BTRFS_FIRST_FREE_OBJECTID) 1675 return 1; 1676 1677 ret = get_cur_inode_state(sctx, ino, gen); 1678 if (ret < 0) 1679 goto out; 1680 1681 if (ret == inode_state_no_change || 1682 ret == inode_state_did_create || 1683 ret == inode_state_will_delete) 1684 ret = 1; 1685 else 1686 ret = 0; 1687 1688 out: 1689 return ret; 1690 } 1691 1692 /* 1693 * Helper function to lookup a dir item in a dir. 1694 */ 1695 static int lookup_dir_item_inode(struct btrfs_root *root, 1696 u64 dir, const char *name, int name_len, 1697 u64 *found_inode, 1698 u8 *found_type) 1699 { 1700 int ret = 0; 1701 struct btrfs_dir_item *di; 1702 struct btrfs_key key; 1703 struct btrfs_path *path; 1704 1705 path = alloc_path_for_send(); 1706 if (!path) 1707 return -ENOMEM; 1708 1709 di = btrfs_lookup_dir_item(NULL, root, path, 1710 dir, name, name_len, 0); 1711 if (!di) { 1712 ret = -ENOENT; 1713 goto out; 1714 } 1715 if (IS_ERR(di)) { 1716 ret = PTR_ERR(di); 1717 goto out; 1718 } 1719 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); 1720 if (key.type == BTRFS_ROOT_ITEM_KEY) { 1721 ret = -ENOENT; 1722 goto out; 1723 } 1724 *found_inode = key.objectid; 1725 *found_type = btrfs_dir_type(path->nodes[0], di); 1726 1727 out: 1728 btrfs_free_path(path); 1729 return ret; 1730 } 1731 1732 /* 1733 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, 1734 * generation of the parent dir and the name of the dir entry. 1735 */ 1736 static int get_first_ref(struct btrfs_root *root, u64 ino, 1737 u64 *dir, u64 *dir_gen, struct fs_path *name) 1738 { 1739 int ret; 1740 struct btrfs_key key; 1741 struct btrfs_key found_key; 1742 struct btrfs_path *path; 1743 int len; 1744 u64 parent_dir; 1745 1746 path = alloc_path_for_send(); 1747 if (!path) 1748 return -ENOMEM; 1749 1750 key.objectid = ino; 1751 key.type = BTRFS_INODE_REF_KEY; 1752 key.offset = 0; 1753 1754 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); 1755 if (ret < 0) 1756 goto out; 1757 if (!ret) 1758 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1759 path->slots[0]); 1760 if (ret || found_key.objectid != ino || 1761 (found_key.type != BTRFS_INODE_REF_KEY && 1762 found_key.type != BTRFS_INODE_EXTREF_KEY)) { 1763 ret = -ENOENT; 1764 goto out; 1765 } 1766 1767 if (found_key.type == BTRFS_INODE_REF_KEY) { 1768 struct btrfs_inode_ref *iref; 1769 iref = btrfs_item_ptr(path->nodes[0], path->slots[0], 1770 struct btrfs_inode_ref); 1771 len = btrfs_inode_ref_name_len(path->nodes[0], iref); 1772 ret = fs_path_add_from_extent_buffer(name, path->nodes[0], 1773 (unsigned long)(iref + 1), 1774 len); 1775 parent_dir = found_key.offset; 1776 } else { 1777 struct btrfs_inode_extref *extref; 1778 extref = btrfs_item_ptr(path->nodes[0], path->slots[0], 1779 struct btrfs_inode_extref); 1780 len = btrfs_inode_extref_name_len(path->nodes[0], extref); 1781 ret = fs_path_add_from_extent_buffer(name, path->nodes[0], 1782 (unsigned long)&extref->name, len); 1783 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); 1784 } 1785 if (ret < 0) 1786 goto out; 1787 btrfs_release_path(path); 1788 1789 if (dir_gen) { 1790 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, 1791 NULL, NULL, NULL); 1792 if (ret < 0) 1793 goto out; 1794 } 1795 1796 *dir = parent_dir; 1797 1798 out: 1799 btrfs_free_path(path); 1800 return ret; 1801 } 1802 1803 static int is_first_ref(struct btrfs_root *root, 1804 u64 ino, u64 dir, 1805 const char *name, int name_len) 1806 { 1807 int ret; 1808 struct fs_path *tmp_name; 1809 u64 tmp_dir; 1810 1811 tmp_name = fs_path_alloc(); 1812 if (!tmp_name) 1813 return -ENOMEM; 1814 1815 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name); 1816 if (ret < 0) 1817 goto out; 1818 1819 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { 1820 ret = 0; 1821 goto out; 1822 } 1823 1824 ret = !memcmp(tmp_name->start, name, name_len); 1825 1826 out: 1827 fs_path_free(tmp_name); 1828 return ret; 1829 } 1830 1831 /* 1832 * Used by process_recorded_refs to determine if a new ref would overwrite an 1833 * already existing ref. In case it detects an overwrite, it returns the 1834 * inode/gen in who_ino/who_gen. 1835 * When an overwrite is detected, process_recorded_refs does proper orphanizing 1836 * to make sure later references to the overwritten inode are possible. 1837 * Orphanizing is however only required for the first ref of an inode. 1838 * process_recorded_refs does an additional is_first_ref check to see if 1839 * orphanizing is really required. 1840 */ 1841 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, 1842 const char *name, int name_len, 1843 u64 *who_ino, u64 *who_gen, u64 *who_mode) 1844 { 1845 int ret = 0; 1846 u64 gen; 1847 u64 other_inode = 0; 1848 u8 other_type = 0; 1849 1850 if (!sctx->parent_root) 1851 goto out; 1852 1853 ret = is_inode_existent(sctx, dir, dir_gen); 1854 if (ret <= 0) 1855 goto out; 1856 1857 /* 1858 * If we have a parent root we need to verify that the parent dir was 1859 * not deleted and then re-created, if it was then we have no overwrite 1860 * and we can just unlink this entry. 1861 */ 1862 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) { 1863 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, 1864 NULL, NULL, NULL); 1865 if (ret < 0 && ret != -ENOENT) 1866 goto out; 1867 if (ret) { 1868 ret = 0; 1869 goto out; 1870 } 1871 if (gen != dir_gen) 1872 goto out; 1873 } 1874 1875 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, 1876 &other_inode, &other_type); 1877 if (ret < 0 && ret != -ENOENT) 1878 goto out; 1879 if (ret) { 1880 ret = 0; 1881 goto out; 1882 } 1883 1884 /* 1885 * Check if the overwritten ref was already processed. If yes, the ref 1886 * was already unlinked/moved, so we can safely assume that we will not 1887 * overwrite anything at this point in time. 1888 */ 1889 if (other_inode > sctx->send_progress || 1890 is_waiting_for_move(sctx, other_inode)) { 1891 ret = get_inode_info(sctx->parent_root, other_inode, NULL, 1892 who_gen, who_mode, NULL, NULL, NULL); 1893 if (ret < 0) 1894 goto out; 1895 1896 ret = 1; 1897 *who_ino = other_inode; 1898 } else { 1899 ret = 0; 1900 } 1901 1902 out: 1903 return ret; 1904 } 1905 1906 /* 1907 * Checks if the ref was overwritten by an already processed inode. This is 1908 * used by __get_cur_name_and_parent to find out if the ref was orphanized and 1909 * thus the orphan name needs be used. 1910 * process_recorded_refs also uses it to avoid unlinking of refs that were 1911 * overwritten. 1912 */ 1913 static int did_overwrite_ref(struct send_ctx *sctx, 1914 u64 dir, u64 dir_gen, 1915 u64 ino, u64 ino_gen, 1916 const char *name, int name_len) 1917 { 1918 int ret = 0; 1919 u64 gen; 1920 u64 ow_inode; 1921 u8 other_type; 1922 1923 if (!sctx->parent_root) 1924 goto out; 1925 1926 ret = is_inode_existent(sctx, dir, dir_gen); 1927 if (ret <= 0) 1928 goto out; 1929 1930 if (dir != BTRFS_FIRST_FREE_OBJECTID) { 1931 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, 1932 NULL, NULL, NULL); 1933 if (ret < 0 && ret != -ENOENT) 1934 goto out; 1935 if (ret) { 1936 ret = 0; 1937 goto out; 1938 } 1939 if (gen != dir_gen) 1940 goto out; 1941 } 1942 1943 /* check if the ref was overwritten by another ref */ 1944 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, 1945 &ow_inode, &other_type); 1946 if (ret < 0 && ret != -ENOENT) 1947 goto out; 1948 if (ret) { 1949 /* was never and will never be overwritten */ 1950 ret = 0; 1951 goto out; 1952 } 1953 1954 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, 1955 NULL, NULL); 1956 if (ret < 0) 1957 goto out; 1958 1959 if (ow_inode == ino && gen == ino_gen) { 1960 ret = 0; 1961 goto out; 1962 } 1963 1964 /* 1965 * We know that it is or will be overwritten. Check this now. 1966 * The current inode being processed might have been the one that caused 1967 * inode 'ino' to be orphanized, therefore check if ow_inode matches 1968 * the current inode being processed. 1969 */ 1970 if ((ow_inode < sctx->send_progress) || 1971 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino && 1972 gen == sctx->cur_inode_gen)) 1973 ret = 1; 1974 else 1975 ret = 0; 1976 1977 out: 1978 return ret; 1979 } 1980 1981 /* 1982 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode 1983 * that got overwritten. This is used by process_recorded_refs to determine 1984 * if it has to use the path as returned by get_cur_path or the orphan name. 1985 */ 1986 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) 1987 { 1988 int ret = 0; 1989 struct fs_path *name = NULL; 1990 u64 dir; 1991 u64 dir_gen; 1992 1993 if (!sctx->parent_root) 1994 goto out; 1995 1996 name = fs_path_alloc(); 1997 if (!name) 1998 return -ENOMEM; 1999 2000 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); 2001 if (ret < 0) 2002 goto out; 2003 2004 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, 2005 name->start, fs_path_len(name)); 2006 2007 out: 2008 fs_path_free(name); 2009 return ret; 2010 } 2011 2012 /* 2013 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, 2014 * so we need to do some special handling in case we have clashes. This function 2015 * takes care of this with the help of name_cache_entry::radix_list. 2016 * In case of error, nce is kfreed. 2017 */ 2018 static int name_cache_insert(struct send_ctx *sctx, 2019 struct name_cache_entry *nce) 2020 { 2021 int ret = 0; 2022 struct list_head *nce_head; 2023 2024 nce_head = radix_tree_lookup(&sctx->name_cache, 2025 (unsigned long)nce->ino); 2026 if (!nce_head) { 2027 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL); 2028 if (!nce_head) { 2029 kfree(nce); 2030 return -ENOMEM; 2031 } 2032 INIT_LIST_HEAD(nce_head); 2033 2034 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); 2035 if (ret < 0) { 2036 kfree(nce_head); 2037 kfree(nce); 2038 return ret; 2039 } 2040 } 2041 list_add_tail(&nce->radix_list, nce_head); 2042 list_add_tail(&nce->list, &sctx->name_cache_list); 2043 sctx->name_cache_size++; 2044 2045 return ret; 2046 } 2047 2048 static void name_cache_delete(struct send_ctx *sctx, 2049 struct name_cache_entry *nce) 2050 { 2051 struct list_head *nce_head; 2052 2053 nce_head = radix_tree_lookup(&sctx->name_cache, 2054 (unsigned long)nce->ino); 2055 if (!nce_head) { 2056 btrfs_err(sctx->send_root->fs_info, 2057 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", 2058 nce->ino, sctx->name_cache_size); 2059 } 2060 2061 list_del(&nce->radix_list); 2062 list_del(&nce->list); 2063 sctx->name_cache_size--; 2064 2065 /* 2066 * We may not get to the final release of nce_head if the lookup fails 2067 */ 2068 if (nce_head && list_empty(nce_head)) { 2069 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); 2070 kfree(nce_head); 2071 } 2072 } 2073 2074 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, 2075 u64 ino, u64 gen) 2076 { 2077 struct list_head *nce_head; 2078 struct name_cache_entry *cur; 2079 2080 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); 2081 if (!nce_head) 2082 return NULL; 2083 2084 list_for_each_entry(cur, nce_head, radix_list) { 2085 if (cur->ino == ino && cur->gen == gen) 2086 return cur; 2087 } 2088 return NULL; 2089 } 2090 2091 /* 2092 * Removes the entry from the list and adds it back to the end. This marks the 2093 * entry as recently used so that name_cache_clean_unused does not remove it. 2094 */ 2095 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) 2096 { 2097 list_del(&nce->list); 2098 list_add_tail(&nce->list, &sctx->name_cache_list); 2099 } 2100 2101 /* 2102 * Remove some entries from the beginning of name_cache_list. 2103 */ 2104 static void name_cache_clean_unused(struct send_ctx *sctx) 2105 { 2106 struct name_cache_entry *nce; 2107 2108 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) 2109 return; 2110 2111 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { 2112 nce = list_entry(sctx->name_cache_list.next, 2113 struct name_cache_entry, list); 2114 name_cache_delete(sctx, nce); 2115 kfree(nce); 2116 } 2117 } 2118 2119 static void name_cache_free(struct send_ctx *sctx) 2120 { 2121 struct name_cache_entry *nce; 2122 2123 while (!list_empty(&sctx->name_cache_list)) { 2124 nce = list_entry(sctx->name_cache_list.next, 2125 struct name_cache_entry, list); 2126 name_cache_delete(sctx, nce); 2127 kfree(nce); 2128 } 2129 } 2130 2131 /* 2132 * Used by get_cur_path for each ref up to the root. 2133 * Returns 0 if it succeeded. 2134 * Returns 1 if the inode is not existent or got overwritten. In that case, the 2135 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 2136 * is returned, parent_ino/parent_gen are not guaranteed to be valid. 2137 * Returns <0 in case of error. 2138 */ 2139 static int __get_cur_name_and_parent(struct send_ctx *sctx, 2140 u64 ino, u64 gen, 2141 u64 *parent_ino, 2142 u64 *parent_gen, 2143 struct fs_path *dest) 2144 { 2145 int ret; 2146 int nce_ret; 2147 struct name_cache_entry *nce = NULL; 2148 2149 /* 2150 * First check if we already did a call to this function with the same 2151 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes 2152 * return the cached result. 2153 */ 2154 nce = name_cache_search(sctx, ino, gen); 2155 if (nce) { 2156 if (ino < sctx->send_progress && nce->need_later_update) { 2157 name_cache_delete(sctx, nce); 2158 kfree(nce); 2159 nce = NULL; 2160 } else { 2161 name_cache_used(sctx, nce); 2162 *parent_ino = nce->parent_ino; 2163 *parent_gen = nce->parent_gen; 2164 ret = fs_path_add(dest, nce->name, nce->name_len); 2165 if (ret < 0) 2166 goto out; 2167 ret = nce->ret; 2168 goto out; 2169 } 2170 } 2171 2172 /* 2173 * If the inode is not existent yet, add the orphan name and return 1. 2174 * This should only happen for the parent dir that we determine in 2175 * __record_new_ref 2176 */ 2177 ret = is_inode_existent(sctx, ino, gen); 2178 if (ret < 0) 2179 goto out; 2180 2181 if (!ret) { 2182 ret = gen_unique_name(sctx, ino, gen, dest); 2183 if (ret < 0) 2184 goto out; 2185 ret = 1; 2186 goto out_cache; 2187 } 2188 2189 /* 2190 * Depending on whether the inode was already processed or not, use 2191 * send_root or parent_root for ref lookup. 2192 */ 2193 if (ino < sctx->send_progress) 2194 ret = get_first_ref(sctx->send_root, ino, 2195 parent_ino, parent_gen, dest); 2196 else 2197 ret = get_first_ref(sctx->parent_root, ino, 2198 parent_ino, parent_gen, dest); 2199 if (ret < 0) 2200 goto out; 2201 2202 /* 2203 * Check if the ref was overwritten by an inode's ref that was processed 2204 * earlier. If yes, treat as orphan and return 1. 2205 */ 2206 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, 2207 dest->start, dest->end - dest->start); 2208 if (ret < 0) 2209 goto out; 2210 if (ret) { 2211 fs_path_reset(dest); 2212 ret = gen_unique_name(sctx, ino, gen, dest); 2213 if (ret < 0) 2214 goto out; 2215 ret = 1; 2216 } 2217 2218 out_cache: 2219 /* 2220 * Store the result of the lookup in the name cache. 2221 */ 2222 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); 2223 if (!nce) { 2224 ret = -ENOMEM; 2225 goto out; 2226 } 2227 2228 nce->ino = ino; 2229 nce->gen = gen; 2230 nce->parent_ino = *parent_ino; 2231 nce->parent_gen = *parent_gen; 2232 nce->name_len = fs_path_len(dest); 2233 nce->ret = ret; 2234 strcpy(nce->name, dest->start); 2235 2236 if (ino < sctx->send_progress) 2237 nce->need_later_update = 0; 2238 else 2239 nce->need_later_update = 1; 2240 2241 nce_ret = name_cache_insert(sctx, nce); 2242 if (nce_ret < 0) 2243 ret = nce_ret; 2244 name_cache_clean_unused(sctx); 2245 2246 out: 2247 return ret; 2248 } 2249 2250 /* 2251 * Magic happens here. This function returns the first ref to an inode as it 2252 * would look like while receiving the stream at this point in time. 2253 * We walk the path up to the root. For every inode in between, we check if it 2254 * was already processed/sent. If yes, we continue with the parent as found 2255 * in send_root. If not, we continue with the parent as found in parent_root. 2256 * If we encounter an inode that was deleted at this point in time, we use the 2257 * inodes "orphan" name instead of the real name and stop. Same with new inodes 2258 * that were not created yet and overwritten inodes/refs. 2259 * 2260 * When do we have have orphan inodes: 2261 * 1. When an inode is freshly created and thus no valid refs are available yet 2262 * 2. When a directory lost all it's refs (deleted) but still has dir items 2263 * inside which were not processed yet (pending for move/delete). If anyone 2264 * tried to get the path to the dir items, it would get a path inside that 2265 * orphan directory. 2266 * 3. When an inode is moved around or gets new links, it may overwrite the ref 2267 * of an unprocessed inode. If in that case the first ref would be 2268 * overwritten, the overwritten inode gets "orphanized". Later when we 2269 * process this overwritten inode, it is restored at a new place by moving 2270 * the orphan inode. 2271 * 2272 * sctx->send_progress tells this function at which point in time receiving 2273 * would be. 2274 */ 2275 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, 2276 struct fs_path *dest) 2277 { 2278 int ret = 0; 2279 struct fs_path *name = NULL; 2280 u64 parent_inode = 0; 2281 u64 parent_gen = 0; 2282 int stop = 0; 2283 2284 name = fs_path_alloc(); 2285 if (!name) { 2286 ret = -ENOMEM; 2287 goto out; 2288 } 2289 2290 dest->reversed = 1; 2291 fs_path_reset(dest); 2292 2293 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { 2294 struct waiting_dir_move *wdm; 2295 2296 fs_path_reset(name); 2297 2298 if (is_waiting_for_rm(sctx, ino)) { 2299 ret = gen_unique_name(sctx, ino, gen, name); 2300 if (ret < 0) 2301 goto out; 2302 ret = fs_path_add_path(dest, name); 2303 break; 2304 } 2305 2306 wdm = get_waiting_dir_move(sctx, ino); 2307 if (wdm && wdm->orphanized) { 2308 ret = gen_unique_name(sctx, ino, gen, name); 2309 stop = 1; 2310 } else if (wdm) { 2311 ret = get_first_ref(sctx->parent_root, ino, 2312 &parent_inode, &parent_gen, name); 2313 } else { 2314 ret = __get_cur_name_and_parent(sctx, ino, gen, 2315 &parent_inode, 2316 &parent_gen, name); 2317 if (ret) 2318 stop = 1; 2319 } 2320 2321 if (ret < 0) 2322 goto out; 2323 2324 ret = fs_path_add_path(dest, name); 2325 if (ret < 0) 2326 goto out; 2327 2328 ino = parent_inode; 2329 gen = parent_gen; 2330 } 2331 2332 out: 2333 fs_path_free(name); 2334 if (!ret) 2335 fs_path_unreverse(dest); 2336 return ret; 2337 } 2338 2339 /* 2340 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace 2341 */ 2342 static int send_subvol_begin(struct send_ctx *sctx) 2343 { 2344 int ret; 2345 struct btrfs_root *send_root = sctx->send_root; 2346 struct btrfs_root *parent_root = sctx->parent_root; 2347 struct btrfs_path *path; 2348 struct btrfs_key key; 2349 struct btrfs_root_ref *ref; 2350 struct extent_buffer *leaf; 2351 char *name = NULL; 2352 int namelen; 2353 2354 path = btrfs_alloc_path(); 2355 if (!path) 2356 return -ENOMEM; 2357 2358 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); 2359 if (!name) { 2360 btrfs_free_path(path); 2361 return -ENOMEM; 2362 } 2363 2364 key.objectid = send_root->objectid; 2365 key.type = BTRFS_ROOT_BACKREF_KEY; 2366 key.offset = 0; 2367 2368 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, 2369 &key, path, 1, 0); 2370 if (ret < 0) 2371 goto out; 2372 if (ret) { 2373 ret = -ENOENT; 2374 goto out; 2375 } 2376 2377 leaf = path->nodes[0]; 2378 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2379 if (key.type != BTRFS_ROOT_BACKREF_KEY || 2380 key.objectid != send_root->objectid) { 2381 ret = -ENOENT; 2382 goto out; 2383 } 2384 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); 2385 namelen = btrfs_root_ref_name_len(leaf, ref); 2386 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); 2387 btrfs_release_path(path); 2388 2389 if (parent_root) { 2390 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); 2391 if (ret < 0) 2392 goto out; 2393 } else { 2394 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); 2395 if (ret < 0) 2396 goto out; 2397 } 2398 2399 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); 2400 2401 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid)) 2402 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, 2403 sctx->send_root->root_item.received_uuid); 2404 else 2405 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, 2406 sctx->send_root->root_item.uuid); 2407 2408 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, 2409 le64_to_cpu(sctx->send_root->root_item.ctransid)); 2410 if (parent_root) { 2411 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) 2412 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, 2413 parent_root->root_item.received_uuid); 2414 else 2415 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, 2416 parent_root->root_item.uuid); 2417 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, 2418 le64_to_cpu(sctx->parent_root->root_item.ctransid)); 2419 } 2420 2421 ret = send_cmd(sctx); 2422 2423 tlv_put_failure: 2424 out: 2425 btrfs_free_path(path); 2426 kfree(name); 2427 return ret; 2428 } 2429 2430 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) 2431 { 2432 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 2433 int ret = 0; 2434 struct fs_path *p; 2435 2436 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size); 2437 2438 p = fs_path_alloc(); 2439 if (!p) 2440 return -ENOMEM; 2441 2442 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); 2443 if (ret < 0) 2444 goto out; 2445 2446 ret = get_cur_path(sctx, ino, gen, p); 2447 if (ret < 0) 2448 goto out; 2449 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 2450 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); 2451 2452 ret = send_cmd(sctx); 2453 2454 tlv_put_failure: 2455 out: 2456 fs_path_free(p); 2457 return ret; 2458 } 2459 2460 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) 2461 { 2462 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 2463 int ret = 0; 2464 struct fs_path *p; 2465 2466 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode); 2467 2468 p = fs_path_alloc(); 2469 if (!p) 2470 return -ENOMEM; 2471 2472 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); 2473 if (ret < 0) 2474 goto out; 2475 2476 ret = get_cur_path(sctx, ino, gen, p); 2477 if (ret < 0) 2478 goto out; 2479 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 2480 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); 2481 2482 ret = send_cmd(sctx); 2483 2484 tlv_put_failure: 2485 out: 2486 fs_path_free(p); 2487 return ret; 2488 } 2489 2490 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) 2491 { 2492 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 2493 int ret = 0; 2494 struct fs_path *p; 2495 2496 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu", 2497 ino, uid, gid); 2498 2499 p = fs_path_alloc(); 2500 if (!p) 2501 return -ENOMEM; 2502 2503 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); 2504 if (ret < 0) 2505 goto out; 2506 2507 ret = get_cur_path(sctx, ino, gen, p); 2508 if (ret < 0) 2509 goto out; 2510 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 2511 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); 2512 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); 2513 2514 ret = send_cmd(sctx); 2515 2516 tlv_put_failure: 2517 out: 2518 fs_path_free(p); 2519 return ret; 2520 } 2521 2522 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) 2523 { 2524 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 2525 int ret = 0; 2526 struct fs_path *p = NULL; 2527 struct btrfs_inode_item *ii; 2528 struct btrfs_path *path = NULL; 2529 struct extent_buffer *eb; 2530 struct btrfs_key key; 2531 int slot; 2532 2533 btrfs_debug(fs_info, "send_utimes %llu", ino); 2534 2535 p = fs_path_alloc(); 2536 if (!p) 2537 return -ENOMEM; 2538 2539 path = alloc_path_for_send(); 2540 if (!path) { 2541 ret = -ENOMEM; 2542 goto out; 2543 } 2544 2545 key.objectid = ino; 2546 key.type = BTRFS_INODE_ITEM_KEY; 2547 key.offset = 0; 2548 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); 2549 if (ret > 0) 2550 ret = -ENOENT; 2551 if (ret < 0) 2552 goto out; 2553 2554 eb = path->nodes[0]; 2555 slot = path->slots[0]; 2556 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); 2557 2558 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); 2559 if (ret < 0) 2560 goto out; 2561 2562 ret = get_cur_path(sctx, ino, gen, p); 2563 if (ret < 0) 2564 goto out; 2565 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 2566 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); 2567 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); 2568 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); 2569 /* TODO Add otime support when the otime patches get into upstream */ 2570 2571 ret = send_cmd(sctx); 2572 2573 tlv_put_failure: 2574 out: 2575 fs_path_free(p); 2576 btrfs_free_path(path); 2577 return ret; 2578 } 2579 2580 /* 2581 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have 2582 * a valid path yet because we did not process the refs yet. So, the inode 2583 * is created as orphan. 2584 */ 2585 static int send_create_inode(struct send_ctx *sctx, u64 ino) 2586 { 2587 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 2588 int ret = 0; 2589 struct fs_path *p; 2590 int cmd; 2591 u64 gen; 2592 u64 mode; 2593 u64 rdev; 2594 2595 btrfs_debug(fs_info, "send_create_inode %llu", ino); 2596 2597 p = fs_path_alloc(); 2598 if (!p) 2599 return -ENOMEM; 2600 2601 if (ino != sctx->cur_ino) { 2602 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, 2603 NULL, NULL, &rdev); 2604 if (ret < 0) 2605 goto out; 2606 } else { 2607 gen = sctx->cur_inode_gen; 2608 mode = sctx->cur_inode_mode; 2609 rdev = sctx->cur_inode_rdev; 2610 } 2611 2612 if (S_ISREG(mode)) { 2613 cmd = BTRFS_SEND_C_MKFILE; 2614 } else if (S_ISDIR(mode)) { 2615 cmd = BTRFS_SEND_C_MKDIR; 2616 } else if (S_ISLNK(mode)) { 2617 cmd = BTRFS_SEND_C_SYMLINK; 2618 } else if (S_ISCHR(mode) || S_ISBLK(mode)) { 2619 cmd = BTRFS_SEND_C_MKNOD; 2620 } else if (S_ISFIFO(mode)) { 2621 cmd = BTRFS_SEND_C_MKFIFO; 2622 } else if (S_ISSOCK(mode)) { 2623 cmd = BTRFS_SEND_C_MKSOCK; 2624 } else { 2625 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o", 2626 (int)(mode & S_IFMT)); 2627 ret = -EOPNOTSUPP; 2628 goto out; 2629 } 2630 2631 ret = begin_cmd(sctx, cmd); 2632 if (ret < 0) 2633 goto out; 2634 2635 ret = gen_unique_name(sctx, ino, gen, p); 2636 if (ret < 0) 2637 goto out; 2638 2639 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 2640 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); 2641 2642 if (S_ISLNK(mode)) { 2643 fs_path_reset(p); 2644 ret = read_symlink(sctx->send_root, ino, p); 2645 if (ret < 0) 2646 goto out; 2647 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); 2648 } else if (S_ISCHR(mode) || S_ISBLK(mode) || 2649 S_ISFIFO(mode) || S_ISSOCK(mode)) { 2650 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); 2651 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); 2652 } 2653 2654 ret = send_cmd(sctx); 2655 if (ret < 0) 2656 goto out; 2657 2658 2659 tlv_put_failure: 2660 out: 2661 fs_path_free(p); 2662 return ret; 2663 } 2664 2665 /* 2666 * We need some special handling for inodes that get processed before the parent 2667 * directory got created. See process_recorded_refs for details. 2668 * This function does the check if we already created the dir out of order. 2669 */ 2670 static int did_create_dir(struct send_ctx *sctx, u64 dir) 2671 { 2672 int ret = 0; 2673 struct btrfs_path *path = NULL; 2674 struct btrfs_key key; 2675 struct btrfs_key found_key; 2676 struct btrfs_key di_key; 2677 struct extent_buffer *eb; 2678 struct btrfs_dir_item *di; 2679 int slot; 2680 2681 path = alloc_path_for_send(); 2682 if (!path) { 2683 ret = -ENOMEM; 2684 goto out; 2685 } 2686 2687 key.objectid = dir; 2688 key.type = BTRFS_DIR_INDEX_KEY; 2689 key.offset = 0; 2690 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); 2691 if (ret < 0) 2692 goto out; 2693 2694 while (1) { 2695 eb = path->nodes[0]; 2696 slot = path->slots[0]; 2697 if (slot >= btrfs_header_nritems(eb)) { 2698 ret = btrfs_next_leaf(sctx->send_root, path); 2699 if (ret < 0) { 2700 goto out; 2701 } else if (ret > 0) { 2702 ret = 0; 2703 break; 2704 } 2705 continue; 2706 } 2707 2708 btrfs_item_key_to_cpu(eb, &found_key, slot); 2709 if (found_key.objectid != key.objectid || 2710 found_key.type != key.type) { 2711 ret = 0; 2712 goto out; 2713 } 2714 2715 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); 2716 btrfs_dir_item_key_to_cpu(eb, di, &di_key); 2717 2718 if (di_key.type != BTRFS_ROOT_ITEM_KEY && 2719 di_key.objectid < sctx->send_progress) { 2720 ret = 1; 2721 goto out; 2722 } 2723 2724 path->slots[0]++; 2725 } 2726 2727 out: 2728 btrfs_free_path(path); 2729 return ret; 2730 } 2731 2732 /* 2733 * Only creates the inode if it is: 2734 * 1. Not a directory 2735 * 2. Or a directory which was not created already due to out of order 2736 * directories. See did_create_dir and process_recorded_refs for details. 2737 */ 2738 static int send_create_inode_if_needed(struct send_ctx *sctx) 2739 { 2740 int ret; 2741 2742 if (S_ISDIR(sctx->cur_inode_mode)) { 2743 ret = did_create_dir(sctx, sctx->cur_ino); 2744 if (ret < 0) 2745 goto out; 2746 if (ret) { 2747 ret = 0; 2748 goto out; 2749 } 2750 } 2751 2752 ret = send_create_inode(sctx, sctx->cur_ino); 2753 if (ret < 0) 2754 goto out; 2755 2756 out: 2757 return ret; 2758 } 2759 2760 struct recorded_ref { 2761 struct list_head list; 2762 char *name; 2763 struct fs_path *full_path; 2764 u64 dir; 2765 u64 dir_gen; 2766 int name_len; 2767 }; 2768 2769 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) 2770 { 2771 ref->full_path = path; 2772 ref->name = (char *)kbasename(ref->full_path->start); 2773 ref->name_len = ref->full_path->end - ref->name; 2774 } 2775 2776 /* 2777 * We need to process new refs before deleted refs, but compare_tree gives us 2778 * everything mixed. So we first record all refs and later process them. 2779 * This function is a helper to record one ref. 2780 */ 2781 static int __record_ref(struct list_head *head, u64 dir, 2782 u64 dir_gen, struct fs_path *path) 2783 { 2784 struct recorded_ref *ref; 2785 2786 ref = kmalloc(sizeof(*ref), GFP_KERNEL); 2787 if (!ref) 2788 return -ENOMEM; 2789 2790 ref->dir = dir; 2791 ref->dir_gen = dir_gen; 2792 set_ref_path(ref, path); 2793 list_add_tail(&ref->list, head); 2794 return 0; 2795 } 2796 2797 static int dup_ref(struct recorded_ref *ref, struct list_head *list) 2798 { 2799 struct recorded_ref *new; 2800 2801 new = kmalloc(sizeof(*ref), GFP_KERNEL); 2802 if (!new) 2803 return -ENOMEM; 2804 2805 new->dir = ref->dir; 2806 new->dir_gen = ref->dir_gen; 2807 new->full_path = NULL; 2808 INIT_LIST_HEAD(&new->list); 2809 list_add_tail(&new->list, list); 2810 return 0; 2811 } 2812 2813 static void __free_recorded_refs(struct list_head *head) 2814 { 2815 struct recorded_ref *cur; 2816 2817 while (!list_empty(head)) { 2818 cur = list_entry(head->next, struct recorded_ref, list); 2819 fs_path_free(cur->full_path); 2820 list_del(&cur->list); 2821 kfree(cur); 2822 } 2823 } 2824 2825 static void free_recorded_refs(struct send_ctx *sctx) 2826 { 2827 __free_recorded_refs(&sctx->new_refs); 2828 __free_recorded_refs(&sctx->deleted_refs); 2829 } 2830 2831 /* 2832 * Renames/moves a file/dir to its orphan name. Used when the first 2833 * ref of an unprocessed inode gets overwritten and for all non empty 2834 * directories. 2835 */ 2836 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, 2837 struct fs_path *path) 2838 { 2839 int ret; 2840 struct fs_path *orphan; 2841 2842 orphan = fs_path_alloc(); 2843 if (!orphan) 2844 return -ENOMEM; 2845 2846 ret = gen_unique_name(sctx, ino, gen, orphan); 2847 if (ret < 0) 2848 goto out; 2849 2850 ret = send_rename(sctx, path, orphan); 2851 2852 out: 2853 fs_path_free(orphan); 2854 return ret; 2855 } 2856 2857 static struct orphan_dir_info * 2858 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) 2859 { 2860 struct rb_node **p = &sctx->orphan_dirs.rb_node; 2861 struct rb_node *parent = NULL; 2862 struct orphan_dir_info *entry, *odi; 2863 2864 odi = kmalloc(sizeof(*odi), GFP_KERNEL); 2865 if (!odi) 2866 return ERR_PTR(-ENOMEM); 2867 odi->ino = dir_ino; 2868 odi->gen = 0; 2869 2870 while (*p) { 2871 parent = *p; 2872 entry = rb_entry(parent, struct orphan_dir_info, node); 2873 if (dir_ino < entry->ino) { 2874 p = &(*p)->rb_left; 2875 } else if (dir_ino > entry->ino) { 2876 p = &(*p)->rb_right; 2877 } else { 2878 kfree(odi); 2879 return entry; 2880 } 2881 } 2882 2883 rb_link_node(&odi->node, parent, p); 2884 rb_insert_color(&odi->node, &sctx->orphan_dirs); 2885 return odi; 2886 } 2887 2888 static struct orphan_dir_info * 2889 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) 2890 { 2891 struct rb_node *n = sctx->orphan_dirs.rb_node; 2892 struct orphan_dir_info *entry; 2893 2894 while (n) { 2895 entry = rb_entry(n, struct orphan_dir_info, node); 2896 if (dir_ino < entry->ino) 2897 n = n->rb_left; 2898 else if (dir_ino > entry->ino) 2899 n = n->rb_right; 2900 else 2901 return entry; 2902 } 2903 return NULL; 2904 } 2905 2906 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino) 2907 { 2908 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino); 2909 2910 return odi != NULL; 2911 } 2912 2913 static void free_orphan_dir_info(struct send_ctx *sctx, 2914 struct orphan_dir_info *odi) 2915 { 2916 if (!odi) 2917 return; 2918 rb_erase(&odi->node, &sctx->orphan_dirs); 2919 kfree(odi); 2920 } 2921 2922 /* 2923 * Returns 1 if a directory can be removed at this point in time. 2924 * We check this by iterating all dir items and checking if the inode behind 2925 * the dir item was already processed. 2926 */ 2927 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, 2928 u64 send_progress) 2929 { 2930 int ret = 0; 2931 struct btrfs_root *root = sctx->parent_root; 2932 struct btrfs_path *path; 2933 struct btrfs_key key; 2934 struct btrfs_key found_key; 2935 struct btrfs_key loc; 2936 struct btrfs_dir_item *di; 2937 2938 /* 2939 * Don't try to rmdir the top/root subvolume dir. 2940 */ 2941 if (dir == BTRFS_FIRST_FREE_OBJECTID) 2942 return 0; 2943 2944 path = alloc_path_for_send(); 2945 if (!path) 2946 return -ENOMEM; 2947 2948 key.objectid = dir; 2949 key.type = BTRFS_DIR_INDEX_KEY; 2950 key.offset = 0; 2951 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2952 if (ret < 0) 2953 goto out; 2954 2955 while (1) { 2956 struct waiting_dir_move *dm; 2957 2958 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2959 ret = btrfs_next_leaf(root, path); 2960 if (ret < 0) 2961 goto out; 2962 else if (ret > 0) 2963 break; 2964 continue; 2965 } 2966 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 2967 path->slots[0]); 2968 if (found_key.objectid != key.objectid || 2969 found_key.type != key.type) 2970 break; 2971 2972 di = btrfs_item_ptr(path->nodes[0], path->slots[0], 2973 struct btrfs_dir_item); 2974 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); 2975 2976 dm = get_waiting_dir_move(sctx, loc.objectid); 2977 if (dm) { 2978 struct orphan_dir_info *odi; 2979 2980 odi = add_orphan_dir_info(sctx, dir); 2981 if (IS_ERR(odi)) { 2982 ret = PTR_ERR(odi); 2983 goto out; 2984 } 2985 odi->gen = dir_gen; 2986 dm->rmdir_ino = dir; 2987 ret = 0; 2988 goto out; 2989 } 2990 2991 if (loc.objectid > send_progress) { 2992 struct orphan_dir_info *odi; 2993 2994 odi = get_orphan_dir_info(sctx, dir); 2995 free_orphan_dir_info(sctx, odi); 2996 ret = 0; 2997 goto out; 2998 } 2999 3000 path->slots[0]++; 3001 } 3002 3003 ret = 1; 3004 3005 out: 3006 btrfs_free_path(path); 3007 return ret; 3008 } 3009 3010 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) 3011 { 3012 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); 3013 3014 return entry != NULL; 3015 } 3016 3017 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) 3018 { 3019 struct rb_node **p = &sctx->waiting_dir_moves.rb_node; 3020 struct rb_node *parent = NULL; 3021 struct waiting_dir_move *entry, *dm; 3022 3023 dm = kmalloc(sizeof(*dm), GFP_KERNEL); 3024 if (!dm) 3025 return -ENOMEM; 3026 dm->ino = ino; 3027 dm->rmdir_ino = 0; 3028 dm->orphanized = orphanized; 3029 3030 while (*p) { 3031 parent = *p; 3032 entry = rb_entry(parent, struct waiting_dir_move, node); 3033 if (ino < entry->ino) { 3034 p = &(*p)->rb_left; 3035 } else if (ino > entry->ino) { 3036 p = &(*p)->rb_right; 3037 } else { 3038 kfree(dm); 3039 return -EEXIST; 3040 } 3041 } 3042 3043 rb_link_node(&dm->node, parent, p); 3044 rb_insert_color(&dm->node, &sctx->waiting_dir_moves); 3045 return 0; 3046 } 3047 3048 static struct waiting_dir_move * 3049 get_waiting_dir_move(struct send_ctx *sctx, u64 ino) 3050 { 3051 struct rb_node *n = sctx->waiting_dir_moves.rb_node; 3052 struct waiting_dir_move *entry; 3053 3054 while (n) { 3055 entry = rb_entry(n, struct waiting_dir_move, node); 3056 if (ino < entry->ino) 3057 n = n->rb_left; 3058 else if (ino > entry->ino) 3059 n = n->rb_right; 3060 else 3061 return entry; 3062 } 3063 return NULL; 3064 } 3065 3066 static void free_waiting_dir_move(struct send_ctx *sctx, 3067 struct waiting_dir_move *dm) 3068 { 3069 if (!dm) 3070 return; 3071 rb_erase(&dm->node, &sctx->waiting_dir_moves); 3072 kfree(dm); 3073 } 3074 3075 static int add_pending_dir_move(struct send_ctx *sctx, 3076 u64 ino, 3077 u64 ino_gen, 3078 u64 parent_ino, 3079 struct list_head *new_refs, 3080 struct list_head *deleted_refs, 3081 const bool is_orphan) 3082 { 3083 struct rb_node **p = &sctx->pending_dir_moves.rb_node; 3084 struct rb_node *parent = NULL; 3085 struct pending_dir_move *entry = NULL, *pm; 3086 struct recorded_ref *cur; 3087 int exists = 0; 3088 int ret; 3089 3090 pm = kmalloc(sizeof(*pm), GFP_KERNEL); 3091 if (!pm) 3092 return -ENOMEM; 3093 pm->parent_ino = parent_ino; 3094 pm->ino = ino; 3095 pm->gen = ino_gen; 3096 INIT_LIST_HEAD(&pm->list); 3097 INIT_LIST_HEAD(&pm->update_refs); 3098 RB_CLEAR_NODE(&pm->node); 3099 3100 while (*p) { 3101 parent = *p; 3102 entry = rb_entry(parent, struct pending_dir_move, node); 3103 if (parent_ino < entry->parent_ino) { 3104 p = &(*p)->rb_left; 3105 } else if (parent_ino > entry->parent_ino) { 3106 p = &(*p)->rb_right; 3107 } else { 3108 exists = 1; 3109 break; 3110 } 3111 } 3112 3113 list_for_each_entry(cur, deleted_refs, list) { 3114 ret = dup_ref(cur, &pm->update_refs); 3115 if (ret < 0) 3116 goto out; 3117 } 3118 list_for_each_entry(cur, new_refs, list) { 3119 ret = dup_ref(cur, &pm->update_refs); 3120 if (ret < 0) 3121 goto out; 3122 } 3123 3124 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan); 3125 if (ret) 3126 goto out; 3127 3128 if (exists) { 3129 list_add_tail(&pm->list, &entry->list); 3130 } else { 3131 rb_link_node(&pm->node, parent, p); 3132 rb_insert_color(&pm->node, &sctx->pending_dir_moves); 3133 } 3134 ret = 0; 3135 out: 3136 if (ret) { 3137 __free_recorded_refs(&pm->update_refs); 3138 kfree(pm); 3139 } 3140 return ret; 3141 } 3142 3143 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, 3144 u64 parent_ino) 3145 { 3146 struct rb_node *n = sctx->pending_dir_moves.rb_node; 3147 struct pending_dir_move *entry; 3148 3149 while (n) { 3150 entry = rb_entry(n, struct pending_dir_move, node); 3151 if (parent_ino < entry->parent_ino) 3152 n = n->rb_left; 3153 else if (parent_ino > entry->parent_ino) 3154 n = n->rb_right; 3155 else 3156 return entry; 3157 } 3158 return NULL; 3159 } 3160 3161 static int path_loop(struct send_ctx *sctx, struct fs_path *name, 3162 u64 ino, u64 gen, u64 *ancestor_ino) 3163 { 3164 int ret = 0; 3165 u64 parent_inode = 0; 3166 u64 parent_gen = 0; 3167 u64 start_ino = ino; 3168 3169 *ancestor_ino = 0; 3170 while (ino != BTRFS_FIRST_FREE_OBJECTID) { 3171 fs_path_reset(name); 3172 3173 if (is_waiting_for_rm(sctx, ino)) 3174 break; 3175 if (is_waiting_for_move(sctx, ino)) { 3176 if (*ancestor_ino == 0) 3177 *ancestor_ino = ino; 3178 ret = get_first_ref(sctx->parent_root, ino, 3179 &parent_inode, &parent_gen, name); 3180 } else { 3181 ret = __get_cur_name_and_parent(sctx, ino, gen, 3182 &parent_inode, 3183 &parent_gen, name); 3184 if (ret > 0) { 3185 ret = 0; 3186 break; 3187 } 3188 } 3189 if (ret < 0) 3190 break; 3191 if (parent_inode == start_ino) { 3192 ret = 1; 3193 if (*ancestor_ino == 0) 3194 *ancestor_ino = ino; 3195 break; 3196 } 3197 ino = parent_inode; 3198 gen = parent_gen; 3199 } 3200 return ret; 3201 } 3202 3203 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) 3204 { 3205 struct fs_path *from_path = NULL; 3206 struct fs_path *to_path = NULL; 3207 struct fs_path *name = NULL; 3208 u64 orig_progress = sctx->send_progress; 3209 struct recorded_ref *cur; 3210 u64 parent_ino, parent_gen; 3211 struct waiting_dir_move *dm = NULL; 3212 u64 rmdir_ino = 0; 3213 u64 ancestor; 3214 bool is_orphan; 3215 int ret; 3216 3217 name = fs_path_alloc(); 3218 from_path = fs_path_alloc(); 3219 if (!name || !from_path) { 3220 ret = -ENOMEM; 3221 goto out; 3222 } 3223 3224 dm = get_waiting_dir_move(sctx, pm->ino); 3225 ASSERT(dm); 3226 rmdir_ino = dm->rmdir_ino; 3227 is_orphan = dm->orphanized; 3228 free_waiting_dir_move(sctx, dm); 3229 3230 if (is_orphan) { 3231 ret = gen_unique_name(sctx, pm->ino, 3232 pm->gen, from_path); 3233 } else { 3234 ret = get_first_ref(sctx->parent_root, pm->ino, 3235 &parent_ino, &parent_gen, name); 3236 if (ret < 0) 3237 goto out; 3238 ret = get_cur_path(sctx, parent_ino, parent_gen, 3239 from_path); 3240 if (ret < 0) 3241 goto out; 3242 ret = fs_path_add_path(from_path, name); 3243 } 3244 if (ret < 0) 3245 goto out; 3246 3247 sctx->send_progress = sctx->cur_ino + 1; 3248 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor); 3249 if (ret < 0) 3250 goto out; 3251 if (ret) { 3252 LIST_HEAD(deleted_refs); 3253 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID); 3254 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor, 3255 &pm->update_refs, &deleted_refs, 3256 is_orphan); 3257 if (ret < 0) 3258 goto out; 3259 if (rmdir_ino) { 3260 dm = get_waiting_dir_move(sctx, pm->ino); 3261 ASSERT(dm); 3262 dm->rmdir_ino = rmdir_ino; 3263 } 3264 goto out; 3265 } 3266 fs_path_reset(name); 3267 to_path = name; 3268 name = NULL; 3269 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); 3270 if (ret < 0) 3271 goto out; 3272 3273 ret = send_rename(sctx, from_path, to_path); 3274 if (ret < 0) 3275 goto out; 3276 3277 if (rmdir_ino) { 3278 struct orphan_dir_info *odi; 3279 3280 odi = get_orphan_dir_info(sctx, rmdir_ino); 3281 if (!odi) { 3282 /* already deleted */ 3283 goto finish; 3284 } 3285 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino); 3286 if (ret < 0) 3287 goto out; 3288 if (!ret) 3289 goto finish; 3290 3291 name = fs_path_alloc(); 3292 if (!name) { 3293 ret = -ENOMEM; 3294 goto out; 3295 } 3296 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name); 3297 if (ret < 0) 3298 goto out; 3299 ret = send_rmdir(sctx, name); 3300 if (ret < 0) 3301 goto out; 3302 free_orphan_dir_info(sctx, odi); 3303 } 3304 3305 finish: 3306 ret = send_utimes(sctx, pm->ino, pm->gen); 3307 if (ret < 0) 3308 goto out; 3309 3310 /* 3311 * After rename/move, need to update the utimes of both new parent(s) 3312 * and old parent(s). 3313 */ 3314 list_for_each_entry(cur, &pm->update_refs, list) { 3315 /* 3316 * The parent inode might have been deleted in the send snapshot 3317 */ 3318 ret = get_inode_info(sctx->send_root, cur->dir, NULL, 3319 NULL, NULL, NULL, NULL, NULL); 3320 if (ret == -ENOENT) { 3321 ret = 0; 3322 continue; 3323 } 3324 if (ret < 0) 3325 goto out; 3326 3327 ret = send_utimes(sctx, cur->dir, cur->dir_gen); 3328 if (ret < 0) 3329 goto out; 3330 } 3331 3332 out: 3333 fs_path_free(name); 3334 fs_path_free(from_path); 3335 fs_path_free(to_path); 3336 sctx->send_progress = orig_progress; 3337 3338 return ret; 3339 } 3340 3341 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) 3342 { 3343 if (!list_empty(&m->list)) 3344 list_del(&m->list); 3345 if (!RB_EMPTY_NODE(&m->node)) 3346 rb_erase(&m->node, &sctx->pending_dir_moves); 3347 __free_recorded_refs(&m->update_refs); 3348 kfree(m); 3349 } 3350 3351 static void tail_append_pending_moves(struct pending_dir_move *moves, 3352 struct list_head *stack) 3353 { 3354 if (list_empty(&moves->list)) { 3355 list_add_tail(&moves->list, stack); 3356 } else { 3357 LIST_HEAD(list); 3358 list_splice_init(&moves->list, &list); 3359 list_add_tail(&moves->list, stack); 3360 list_splice_tail(&list, stack); 3361 } 3362 } 3363 3364 static int apply_children_dir_moves(struct send_ctx *sctx) 3365 { 3366 struct pending_dir_move *pm; 3367 struct list_head stack; 3368 u64 parent_ino = sctx->cur_ino; 3369 int ret = 0; 3370 3371 pm = get_pending_dir_moves(sctx, parent_ino); 3372 if (!pm) 3373 return 0; 3374 3375 INIT_LIST_HEAD(&stack); 3376 tail_append_pending_moves(pm, &stack); 3377 3378 while (!list_empty(&stack)) { 3379 pm = list_first_entry(&stack, struct pending_dir_move, list); 3380 parent_ino = pm->ino; 3381 ret = apply_dir_move(sctx, pm); 3382 free_pending_move(sctx, pm); 3383 if (ret) 3384 goto out; 3385 pm = get_pending_dir_moves(sctx, parent_ino); 3386 if (pm) 3387 tail_append_pending_moves(pm, &stack); 3388 } 3389 return 0; 3390 3391 out: 3392 while (!list_empty(&stack)) { 3393 pm = list_first_entry(&stack, struct pending_dir_move, list); 3394 free_pending_move(sctx, pm); 3395 } 3396 return ret; 3397 } 3398 3399 /* 3400 * We might need to delay a directory rename even when no ancestor directory 3401 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was 3402 * renamed. This happens when we rename a directory to the old name (the name 3403 * in the parent root) of some other unrelated directory that got its rename 3404 * delayed due to some ancestor with higher number that got renamed. 3405 * 3406 * Example: 3407 * 3408 * Parent snapshot: 3409 * . (ino 256) 3410 * |---- a/ (ino 257) 3411 * | |---- file (ino 260) 3412 * | 3413 * |---- b/ (ino 258) 3414 * |---- c/ (ino 259) 3415 * 3416 * Send snapshot: 3417 * . (ino 256) 3418 * |---- a/ (ino 258) 3419 * |---- x/ (ino 259) 3420 * |---- y/ (ino 257) 3421 * |----- file (ino 260) 3422 * 3423 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257 3424 * from 'a' to 'x/y' happening first, which in turn depends on the rename of 3425 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream 3426 * must issue is: 3427 * 3428 * 1 - rename 259 from 'c' to 'x' 3429 * 2 - rename 257 from 'a' to 'x/y' 3430 * 3 - rename 258 from 'b' to 'a' 3431 * 3432 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can 3433 * be done right away and < 0 on error. 3434 */ 3435 static int wait_for_dest_dir_move(struct send_ctx *sctx, 3436 struct recorded_ref *parent_ref, 3437 const bool is_orphan) 3438 { 3439 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info; 3440 struct btrfs_path *path; 3441 struct btrfs_key key; 3442 struct btrfs_key di_key; 3443 struct btrfs_dir_item *di; 3444 u64 left_gen; 3445 u64 right_gen; 3446 int ret = 0; 3447 struct waiting_dir_move *wdm; 3448 3449 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) 3450 return 0; 3451 3452 path = alloc_path_for_send(); 3453 if (!path) 3454 return -ENOMEM; 3455 3456 key.objectid = parent_ref->dir; 3457 key.type = BTRFS_DIR_ITEM_KEY; 3458 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); 3459 3460 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); 3461 if (ret < 0) { 3462 goto out; 3463 } else if (ret > 0) { 3464 ret = 0; 3465 goto out; 3466 } 3467 3468 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name, 3469 parent_ref->name_len); 3470 if (!di) { 3471 ret = 0; 3472 goto out; 3473 } 3474 /* 3475 * di_key.objectid has the number of the inode that has a dentry in the 3476 * parent directory with the same name that sctx->cur_ino is being 3477 * renamed to. We need to check if that inode is in the send root as 3478 * well and if it is currently marked as an inode with a pending rename, 3479 * if it is, we need to delay the rename of sctx->cur_ino as well, so 3480 * that it happens after that other inode is renamed. 3481 */ 3482 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); 3483 if (di_key.type != BTRFS_INODE_ITEM_KEY) { 3484 ret = 0; 3485 goto out; 3486 } 3487 3488 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL, 3489 &left_gen, NULL, NULL, NULL, NULL); 3490 if (ret < 0) 3491 goto out; 3492 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL, 3493 &right_gen, NULL, NULL, NULL, NULL); 3494 if (ret < 0) { 3495 if (ret == -ENOENT) 3496 ret = 0; 3497 goto out; 3498 } 3499 3500 /* Different inode, no need to delay the rename of sctx->cur_ino */ 3501 if (right_gen != left_gen) { 3502 ret = 0; 3503 goto out; 3504 } 3505 3506 wdm = get_waiting_dir_move(sctx, di_key.objectid); 3507 if (wdm && !wdm->orphanized) { 3508 ret = add_pending_dir_move(sctx, 3509 sctx->cur_ino, 3510 sctx->cur_inode_gen, 3511 di_key.objectid, 3512 &sctx->new_refs, 3513 &sctx->deleted_refs, 3514 is_orphan); 3515 if (!ret) 3516 ret = 1; 3517 } 3518 out: 3519 btrfs_free_path(path); 3520 return ret; 3521 } 3522 3523 /* 3524 * Check if ino ino1 is an ancestor of inode ino2 in the given root. 3525 * Return 1 if true, 0 if false and < 0 on error. 3526 */ 3527 static int is_ancestor(struct btrfs_root *root, 3528 const u64 ino1, 3529 const u64 ino1_gen, 3530 const u64 ino2, 3531 struct fs_path *fs_path) 3532 { 3533 u64 ino = ino2; 3534 bool free_path = false; 3535 int ret = 0; 3536 3537 if (!fs_path) { 3538 fs_path = fs_path_alloc(); 3539 if (!fs_path) 3540 return -ENOMEM; 3541 free_path = true; 3542 } 3543 3544 while (ino > BTRFS_FIRST_FREE_OBJECTID) { 3545 u64 parent; 3546 u64 parent_gen; 3547 3548 fs_path_reset(fs_path); 3549 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path); 3550 if (ret < 0) { 3551 if (ret == -ENOENT && ino == ino2) 3552 ret = 0; 3553 goto out; 3554 } 3555 if (parent == ino1) { 3556 ret = parent_gen == ino1_gen ? 1 : 0; 3557 goto out; 3558 } 3559 ino = parent; 3560 } 3561 out: 3562 if (free_path) 3563 fs_path_free(fs_path); 3564 return ret; 3565 } 3566 3567 static int wait_for_parent_move(struct send_ctx *sctx, 3568 struct recorded_ref *parent_ref, 3569 const bool is_orphan) 3570 { 3571 int ret = 0; 3572 u64 ino = parent_ref->dir; 3573 u64 ino_gen = parent_ref->dir_gen; 3574 u64 parent_ino_before, parent_ino_after; 3575 struct fs_path *path_before = NULL; 3576 struct fs_path *path_after = NULL; 3577 int len1, len2; 3578 3579 path_after = fs_path_alloc(); 3580 path_before = fs_path_alloc(); 3581 if (!path_after || !path_before) { 3582 ret = -ENOMEM; 3583 goto out; 3584 } 3585 3586 /* 3587 * Our current directory inode may not yet be renamed/moved because some 3588 * ancestor (immediate or not) has to be renamed/moved first. So find if 3589 * such ancestor exists and make sure our own rename/move happens after 3590 * that ancestor is processed to avoid path build infinite loops (done 3591 * at get_cur_path()). 3592 */ 3593 while (ino > BTRFS_FIRST_FREE_OBJECTID) { 3594 u64 parent_ino_after_gen; 3595 3596 if (is_waiting_for_move(sctx, ino)) { 3597 /* 3598 * If the current inode is an ancestor of ino in the 3599 * parent root, we need to delay the rename of the 3600 * current inode, otherwise don't delayed the rename 3601 * because we can end up with a circular dependency 3602 * of renames, resulting in some directories never 3603 * getting the respective rename operations issued in 3604 * the send stream or getting into infinite path build 3605 * loops. 3606 */ 3607 ret = is_ancestor(sctx->parent_root, 3608 sctx->cur_ino, sctx->cur_inode_gen, 3609 ino, path_before); 3610 if (ret) 3611 break; 3612 } 3613 3614 fs_path_reset(path_before); 3615 fs_path_reset(path_after); 3616 3617 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, 3618 &parent_ino_after_gen, path_after); 3619 if (ret < 0) 3620 goto out; 3621 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, 3622 NULL, path_before); 3623 if (ret < 0 && ret != -ENOENT) { 3624 goto out; 3625 } else if (ret == -ENOENT) { 3626 ret = 0; 3627 break; 3628 } 3629 3630 len1 = fs_path_len(path_before); 3631 len2 = fs_path_len(path_after); 3632 if (ino > sctx->cur_ino && 3633 (parent_ino_before != parent_ino_after || len1 != len2 || 3634 memcmp(path_before->start, path_after->start, len1))) { 3635 u64 parent_ino_gen; 3636 3637 ret = get_inode_info(sctx->parent_root, ino, NULL, 3638 &parent_ino_gen, NULL, NULL, NULL, 3639 NULL); 3640 if (ret < 0) 3641 goto out; 3642 if (ino_gen == parent_ino_gen) { 3643 ret = 1; 3644 break; 3645 } 3646 } 3647 ino = parent_ino_after; 3648 ino_gen = parent_ino_after_gen; 3649 } 3650 3651 out: 3652 fs_path_free(path_before); 3653 fs_path_free(path_after); 3654 3655 if (ret == 1) { 3656 ret = add_pending_dir_move(sctx, 3657 sctx->cur_ino, 3658 sctx->cur_inode_gen, 3659 ino, 3660 &sctx->new_refs, 3661 &sctx->deleted_refs, 3662 is_orphan); 3663 if (!ret) 3664 ret = 1; 3665 } 3666 3667 return ret; 3668 } 3669 3670 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) 3671 { 3672 int ret; 3673 struct fs_path *new_path; 3674 3675 /* 3676 * Our reference's name member points to its full_path member string, so 3677 * we use here a new path. 3678 */ 3679 new_path = fs_path_alloc(); 3680 if (!new_path) 3681 return -ENOMEM; 3682 3683 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path); 3684 if (ret < 0) { 3685 fs_path_free(new_path); 3686 return ret; 3687 } 3688 ret = fs_path_add(new_path, ref->name, ref->name_len); 3689 if (ret < 0) { 3690 fs_path_free(new_path); 3691 return ret; 3692 } 3693 3694 fs_path_free(ref->full_path); 3695 set_ref_path(ref, new_path); 3696 3697 return 0; 3698 } 3699 3700 /* 3701 * This does all the move/link/unlink/rmdir magic. 3702 */ 3703 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) 3704 { 3705 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 3706 int ret = 0; 3707 struct recorded_ref *cur; 3708 struct recorded_ref *cur2; 3709 struct list_head check_dirs; 3710 struct fs_path *valid_path = NULL; 3711 u64 ow_inode = 0; 3712 u64 ow_gen; 3713 u64 ow_mode; 3714 int did_overwrite = 0; 3715 int is_orphan = 0; 3716 u64 last_dir_ino_rm = 0; 3717 bool can_rename = true; 3718 bool orphanized_dir = false; 3719 bool orphanized_ancestor = false; 3720 3721 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino); 3722 3723 /* 3724 * This should never happen as the root dir always has the same ref 3725 * which is always '..' 3726 */ 3727 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); 3728 INIT_LIST_HEAD(&check_dirs); 3729 3730 valid_path = fs_path_alloc(); 3731 if (!valid_path) { 3732 ret = -ENOMEM; 3733 goto out; 3734 } 3735 3736 /* 3737 * First, check if the first ref of the current inode was overwritten 3738 * before. If yes, we know that the current inode was already orphanized 3739 * and thus use the orphan name. If not, we can use get_cur_path to 3740 * get the path of the first ref as it would like while receiving at 3741 * this point in time. 3742 * New inodes are always orphan at the beginning, so force to use the 3743 * orphan name in this case. 3744 * The first ref is stored in valid_path and will be updated if it 3745 * gets moved around. 3746 */ 3747 if (!sctx->cur_inode_new) { 3748 ret = did_overwrite_first_ref(sctx, sctx->cur_ino, 3749 sctx->cur_inode_gen); 3750 if (ret < 0) 3751 goto out; 3752 if (ret) 3753 did_overwrite = 1; 3754 } 3755 if (sctx->cur_inode_new || did_overwrite) { 3756 ret = gen_unique_name(sctx, sctx->cur_ino, 3757 sctx->cur_inode_gen, valid_path); 3758 if (ret < 0) 3759 goto out; 3760 is_orphan = 1; 3761 } else { 3762 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, 3763 valid_path); 3764 if (ret < 0) 3765 goto out; 3766 } 3767 3768 list_for_each_entry(cur, &sctx->new_refs, list) { 3769 /* 3770 * We may have refs where the parent directory does not exist 3771 * yet. This happens if the parent directories inum is higher 3772 * the the current inum. To handle this case, we create the 3773 * parent directory out of order. But we need to check if this 3774 * did already happen before due to other refs in the same dir. 3775 */ 3776 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); 3777 if (ret < 0) 3778 goto out; 3779 if (ret == inode_state_will_create) { 3780 ret = 0; 3781 /* 3782 * First check if any of the current inodes refs did 3783 * already create the dir. 3784 */ 3785 list_for_each_entry(cur2, &sctx->new_refs, list) { 3786 if (cur == cur2) 3787 break; 3788 if (cur2->dir == cur->dir) { 3789 ret = 1; 3790 break; 3791 } 3792 } 3793 3794 /* 3795 * If that did not happen, check if a previous inode 3796 * did already create the dir. 3797 */ 3798 if (!ret) 3799 ret = did_create_dir(sctx, cur->dir); 3800 if (ret < 0) 3801 goto out; 3802 if (!ret) { 3803 ret = send_create_inode(sctx, cur->dir); 3804 if (ret < 0) 3805 goto out; 3806 } 3807 } 3808 3809 /* 3810 * Check if this new ref would overwrite the first ref of 3811 * another unprocessed inode. If yes, orphanize the 3812 * overwritten inode. If we find an overwritten ref that is 3813 * not the first ref, simply unlink it. 3814 */ 3815 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, 3816 cur->name, cur->name_len, 3817 &ow_inode, &ow_gen, &ow_mode); 3818 if (ret < 0) 3819 goto out; 3820 if (ret) { 3821 ret = is_first_ref(sctx->parent_root, 3822 ow_inode, cur->dir, cur->name, 3823 cur->name_len); 3824 if (ret < 0) 3825 goto out; 3826 if (ret) { 3827 struct name_cache_entry *nce; 3828 struct waiting_dir_move *wdm; 3829 3830 ret = orphanize_inode(sctx, ow_inode, ow_gen, 3831 cur->full_path); 3832 if (ret < 0) 3833 goto out; 3834 if (S_ISDIR(ow_mode)) 3835 orphanized_dir = true; 3836 3837 /* 3838 * If ow_inode has its rename operation delayed 3839 * make sure that its orphanized name is used in 3840 * the source path when performing its rename 3841 * operation. 3842 */ 3843 if (is_waiting_for_move(sctx, ow_inode)) { 3844 wdm = get_waiting_dir_move(sctx, 3845 ow_inode); 3846 ASSERT(wdm); 3847 wdm->orphanized = true; 3848 } 3849 3850 /* 3851 * Make sure we clear our orphanized inode's 3852 * name from the name cache. This is because the 3853 * inode ow_inode might be an ancestor of some 3854 * other inode that will be orphanized as well 3855 * later and has an inode number greater than 3856 * sctx->send_progress. We need to prevent 3857 * future name lookups from using the old name 3858 * and get instead the orphan name. 3859 */ 3860 nce = name_cache_search(sctx, ow_inode, ow_gen); 3861 if (nce) { 3862 name_cache_delete(sctx, nce); 3863 kfree(nce); 3864 } 3865 3866 /* 3867 * ow_inode might currently be an ancestor of 3868 * cur_ino, therefore compute valid_path (the 3869 * current path of cur_ino) again because it 3870 * might contain the pre-orphanization name of 3871 * ow_inode, which is no longer valid. 3872 */ 3873 ret = is_ancestor(sctx->parent_root, 3874 ow_inode, ow_gen, 3875 sctx->cur_ino, NULL); 3876 if (ret > 0) { 3877 orphanized_ancestor = true; 3878 fs_path_reset(valid_path); 3879 ret = get_cur_path(sctx, sctx->cur_ino, 3880 sctx->cur_inode_gen, 3881 valid_path); 3882 } 3883 if (ret < 0) 3884 goto out; 3885 } else { 3886 ret = send_unlink(sctx, cur->full_path); 3887 if (ret < 0) 3888 goto out; 3889 } 3890 } 3891 3892 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { 3893 ret = wait_for_dest_dir_move(sctx, cur, is_orphan); 3894 if (ret < 0) 3895 goto out; 3896 if (ret == 1) { 3897 can_rename = false; 3898 *pending_move = 1; 3899 } 3900 } 3901 3902 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root && 3903 can_rename) { 3904 ret = wait_for_parent_move(sctx, cur, is_orphan); 3905 if (ret < 0) 3906 goto out; 3907 if (ret == 1) { 3908 can_rename = false; 3909 *pending_move = 1; 3910 } 3911 } 3912 3913 /* 3914 * link/move the ref to the new place. If we have an orphan 3915 * inode, move it and update valid_path. If not, link or move 3916 * it depending on the inode mode. 3917 */ 3918 if (is_orphan && can_rename) { 3919 ret = send_rename(sctx, valid_path, cur->full_path); 3920 if (ret < 0) 3921 goto out; 3922 is_orphan = 0; 3923 ret = fs_path_copy(valid_path, cur->full_path); 3924 if (ret < 0) 3925 goto out; 3926 } else if (can_rename) { 3927 if (S_ISDIR(sctx->cur_inode_mode)) { 3928 /* 3929 * Dirs can't be linked, so move it. For moved 3930 * dirs, we always have one new and one deleted 3931 * ref. The deleted ref is ignored later. 3932 */ 3933 ret = send_rename(sctx, valid_path, 3934 cur->full_path); 3935 if (!ret) 3936 ret = fs_path_copy(valid_path, 3937 cur->full_path); 3938 if (ret < 0) 3939 goto out; 3940 } else { 3941 /* 3942 * We might have previously orphanized an inode 3943 * which is an ancestor of our current inode, 3944 * so our reference's full path, which was 3945 * computed before any such orphanizations, must 3946 * be updated. 3947 */ 3948 if (orphanized_dir) { 3949 ret = update_ref_path(sctx, cur); 3950 if (ret < 0) 3951 goto out; 3952 } 3953 ret = send_link(sctx, cur->full_path, 3954 valid_path); 3955 if (ret < 0) 3956 goto out; 3957 } 3958 } 3959 ret = dup_ref(cur, &check_dirs); 3960 if (ret < 0) 3961 goto out; 3962 } 3963 3964 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { 3965 /* 3966 * Check if we can already rmdir the directory. If not, 3967 * orphanize it. For every dir item inside that gets deleted 3968 * later, we do this check again and rmdir it then if possible. 3969 * See the use of check_dirs for more details. 3970 */ 3971 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, 3972 sctx->cur_ino); 3973 if (ret < 0) 3974 goto out; 3975 if (ret) { 3976 ret = send_rmdir(sctx, valid_path); 3977 if (ret < 0) 3978 goto out; 3979 } else if (!is_orphan) { 3980 ret = orphanize_inode(sctx, sctx->cur_ino, 3981 sctx->cur_inode_gen, valid_path); 3982 if (ret < 0) 3983 goto out; 3984 is_orphan = 1; 3985 } 3986 3987 list_for_each_entry(cur, &sctx->deleted_refs, list) { 3988 ret = dup_ref(cur, &check_dirs); 3989 if (ret < 0) 3990 goto out; 3991 } 3992 } else if (S_ISDIR(sctx->cur_inode_mode) && 3993 !list_empty(&sctx->deleted_refs)) { 3994 /* 3995 * We have a moved dir. Add the old parent to check_dirs 3996 */ 3997 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, 3998 list); 3999 ret = dup_ref(cur, &check_dirs); 4000 if (ret < 0) 4001 goto out; 4002 } else if (!S_ISDIR(sctx->cur_inode_mode)) { 4003 /* 4004 * We have a non dir inode. Go through all deleted refs and 4005 * unlink them if they were not already overwritten by other 4006 * inodes. 4007 */ 4008 list_for_each_entry(cur, &sctx->deleted_refs, list) { 4009 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, 4010 sctx->cur_ino, sctx->cur_inode_gen, 4011 cur->name, cur->name_len); 4012 if (ret < 0) 4013 goto out; 4014 if (!ret) { 4015 /* 4016 * If we orphanized any ancestor before, we need 4017 * to recompute the full path for deleted names, 4018 * since any such path was computed before we 4019 * processed any references and orphanized any 4020 * ancestor inode. 4021 */ 4022 if (orphanized_ancestor) { 4023 ret = update_ref_path(sctx, cur); 4024 if (ret < 0) 4025 goto out; 4026 } 4027 ret = send_unlink(sctx, cur->full_path); 4028 if (ret < 0) 4029 goto out; 4030 } 4031 ret = dup_ref(cur, &check_dirs); 4032 if (ret < 0) 4033 goto out; 4034 } 4035 /* 4036 * If the inode is still orphan, unlink the orphan. This may 4037 * happen when a previous inode did overwrite the first ref 4038 * of this inode and no new refs were added for the current 4039 * inode. Unlinking does not mean that the inode is deleted in 4040 * all cases. There may still be links to this inode in other 4041 * places. 4042 */ 4043 if (is_orphan) { 4044 ret = send_unlink(sctx, valid_path); 4045 if (ret < 0) 4046 goto out; 4047 } 4048 } 4049 4050 /* 4051 * We did collect all parent dirs where cur_inode was once located. We 4052 * now go through all these dirs and check if they are pending for 4053 * deletion and if it's finally possible to perform the rmdir now. 4054 * We also update the inode stats of the parent dirs here. 4055 */ 4056 list_for_each_entry(cur, &check_dirs, list) { 4057 /* 4058 * In case we had refs into dirs that were not processed yet, 4059 * we don't need to do the utime and rmdir logic for these dirs. 4060 * The dir will be processed later. 4061 */ 4062 if (cur->dir > sctx->cur_ino) 4063 continue; 4064 4065 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); 4066 if (ret < 0) 4067 goto out; 4068 4069 if (ret == inode_state_did_create || 4070 ret == inode_state_no_change) { 4071 /* TODO delayed utimes */ 4072 ret = send_utimes(sctx, cur->dir, cur->dir_gen); 4073 if (ret < 0) 4074 goto out; 4075 } else if (ret == inode_state_did_delete && 4076 cur->dir != last_dir_ino_rm) { 4077 ret = can_rmdir(sctx, cur->dir, cur->dir_gen, 4078 sctx->cur_ino); 4079 if (ret < 0) 4080 goto out; 4081 if (ret) { 4082 ret = get_cur_path(sctx, cur->dir, 4083 cur->dir_gen, valid_path); 4084 if (ret < 0) 4085 goto out; 4086 ret = send_rmdir(sctx, valid_path); 4087 if (ret < 0) 4088 goto out; 4089 last_dir_ino_rm = cur->dir; 4090 } 4091 } 4092 } 4093 4094 ret = 0; 4095 4096 out: 4097 __free_recorded_refs(&check_dirs); 4098 free_recorded_refs(sctx); 4099 fs_path_free(valid_path); 4100 return ret; 4101 } 4102 4103 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name, 4104 void *ctx, struct list_head *refs) 4105 { 4106 int ret = 0; 4107 struct send_ctx *sctx = ctx; 4108 struct fs_path *p; 4109 u64 gen; 4110 4111 p = fs_path_alloc(); 4112 if (!p) 4113 return -ENOMEM; 4114 4115 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, 4116 NULL, NULL); 4117 if (ret < 0) 4118 goto out; 4119 4120 ret = get_cur_path(sctx, dir, gen, p); 4121 if (ret < 0) 4122 goto out; 4123 ret = fs_path_add_path(p, name); 4124 if (ret < 0) 4125 goto out; 4126 4127 ret = __record_ref(refs, dir, gen, p); 4128 4129 out: 4130 if (ret) 4131 fs_path_free(p); 4132 return ret; 4133 } 4134 4135 static int __record_new_ref(int num, u64 dir, int index, 4136 struct fs_path *name, 4137 void *ctx) 4138 { 4139 struct send_ctx *sctx = ctx; 4140 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs); 4141 } 4142 4143 4144 static int __record_deleted_ref(int num, u64 dir, int index, 4145 struct fs_path *name, 4146 void *ctx) 4147 { 4148 struct send_ctx *sctx = ctx; 4149 return record_ref(sctx->parent_root, dir, name, ctx, 4150 &sctx->deleted_refs); 4151 } 4152 4153 static int record_new_ref(struct send_ctx *sctx) 4154 { 4155 int ret; 4156 4157 ret = iterate_inode_ref(sctx->send_root, sctx->left_path, 4158 sctx->cmp_key, 0, __record_new_ref, sctx); 4159 if (ret < 0) 4160 goto out; 4161 ret = 0; 4162 4163 out: 4164 return ret; 4165 } 4166 4167 static int record_deleted_ref(struct send_ctx *sctx) 4168 { 4169 int ret; 4170 4171 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, 4172 sctx->cmp_key, 0, __record_deleted_ref, sctx); 4173 if (ret < 0) 4174 goto out; 4175 ret = 0; 4176 4177 out: 4178 return ret; 4179 } 4180 4181 struct find_ref_ctx { 4182 u64 dir; 4183 u64 dir_gen; 4184 struct btrfs_root *root; 4185 struct fs_path *name; 4186 int found_idx; 4187 }; 4188 4189 static int __find_iref(int num, u64 dir, int index, 4190 struct fs_path *name, 4191 void *ctx_) 4192 { 4193 struct find_ref_ctx *ctx = ctx_; 4194 u64 dir_gen; 4195 int ret; 4196 4197 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && 4198 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { 4199 /* 4200 * To avoid doing extra lookups we'll only do this if everything 4201 * else matches. 4202 */ 4203 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, 4204 NULL, NULL, NULL); 4205 if (ret) 4206 return ret; 4207 if (dir_gen != ctx->dir_gen) 4208 return 0; 4209 ctx->found_idx = num; 4210 return 1; 4211 } 4212 return 0; 4213 } 4214 4215 static int find_iref(struct btrfs_root *root, 4216 struct btrfs_path *path, 4217 struct btrfs_key *key, 4218 u64 dir, u64 dir_gen, struct fs_path *name) 4219 { 4220 int ret; 4221 struct find_ref_ctx ctx; 4222 4223 ctx.dir = dir; 4224 ctx.name = name; 4225 ctx.dir_gen = dir_gen; 4226 ctx.found_idx = -1; 4227 ctx.root = root; 4228 4229 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); 4230 if (ret < 0) 4231 return ret; 4232 4233 if (ctx.found_idx == -1) 4234 return -ENOENT; 4235 4236 return ctx.found_idx; 4237 } 4238 4239 static int __record_changed_new_ref(int num, u64 dir, int index, 4240 struct fs_path *name, 4241 void *ctx) 4242 { 4243 u64 dir_gen; 4244 int ret; 4245 struct send_ctx *sctx = ctx; 4246 4247 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, 4248 NULL, NULL, NULL); 4249 if (ret) 4250 return ret; 4251 4252 ret = find_iref(sctx->parent_root, sctx->right_path, 4253 sctx->cmp_key, dir, dir_gen, name); 4254 if (ret == -ENOENT) 4255 ret = __record_new_ref(num, dir, index, name, sctx); 4256 else if (ret > 0) 4257 ret = 0; 4258 4259 return ret; 4260 } 4261 4262 static int __record_changed_deleted_ref(int num, u64 dir, int index, 4263 struct fs_path *name, 4264 void *ctx) 4265 { 4266 u64 dir_gen; 4267 int ret; 4268 struct send_ctx *sctx = ctx; 4269 4270 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, 4271 NULL, NULL, NULL); 4272 if (ret) 4273 return ret; 4274 4275 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, 4276 dir, dir_gen, name); 4277 if (ret == -ENOENT) 4278 ret = __record_deleted_ref(num, dir, index, name, sctx); 4279 else if (ret > 0) 4280 ret = 0; 4281 4282 return ret; 4283 } 4284 4285 static int record_changed_ref(struct send_ctx *sctx) 4286 { 4287 int ret = 0; 4288 4289 ret = iterate_inode_ref(sctx->send_root, sctx->left_path, 4290 sctx->cmp_key, 0, __record_changed_new_ref, sctx); 4291 if (ret < 0) 4292 goto out; 4293 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, 4294 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); 4295 if (ret < 0) 4296 goto out; 4297 ret = 0; 4298 4299 out: 4300 return ret; 4301 } 4302 4303 /* 4304 * Record and process all refs at once. Needed when an inode changes the 4305 * generation number, which means that it was deleted and recreated. 4306 */ 4307 static int process_all_refs(struct send_ctx *sctx, 4308 enum btrfs_compare_tree_result cmd) 4309 { 4310 int ret; 4311 struct btrfs_root *root; 4312 struct btrfs_path *path; 4313 struct btrfs_key key; 4314 struct btrfs_key found_key; 4315 struct extent_buffer *eb; 4316 int slot; 4317 iterate_inode_ref_t cb; 4318 int pending_move = 0; 4319 4320 path = alloc_path_for_send(); 4321 if (!path) 4322 return -ENOMEM; 4323 4324 if (cmd == BTRFS_COMPARE_TREE_NEW) { 4325 root = sctx->send_root; 4326 cb = __record_new_ref; 4327 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { 4328 root = sctx->parent_root; 4329 cb = __record_deleted_ref; 4330 } else { 4331 btrfs_err(sctx->send_root->fs_info, 4332 "Wrong command %d in process_all_refs", cmd); 4333 ret = -EINVAL; 4334 goto out; 4335 } 4336 4337 key.objectid = sctx->cmp_key->objectid; 4338 key.type = BTRFS_INODE_REF_KEY; 4339 key.offset = 0; 4340 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4341 if (ret < 0) 4342 goto out; 4343 4344 while (1) { 4345 eb = path->nodes[0]; 4346 slot = path->slots[0]; 4347 if (slot >= btrfs_header_nritems(eb)) { 4348 ret = btrfs_next_leaf(root, path); 4349 if (ret < 0) 4350 goto out; 4351 else if (ret > 0) 4352 break; 4353 continue; 4354 } 4355 4356 btrfs_item_key_to_cpu(eb, &found_key, slot); 4357 4358 if (found_key.objectid != key.objectid || 4359 (found_key.type != BTRFS_INODE_REF_KEY && 4360 found_key.type != BTRFS_INODE_EXTREF_KEY)) 4361 break; 4362 4363 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); 4364 if (ret < 0) 4365 goto out; 4366 4367 path->slots[0]++; 4368 } 4369 btrfs_release_path(path); 4370 4371 /* 4372 * We don't actually care about pending_move as we are simply 4373 * re-creating this inode and will be rename'ing it into place once we 4374 * rename the parent directory. 4375 */ 4376 ret = process_recorded_refs(sctx, &pending_move); 4377 out: 4378 btrfs_free_path(path); 4379 return ret; 4380 } 4381 4382 static int send_set_xattr(struct send_ctx *sctx, 4383 struct fs_path *path, 4384 const char *name, int name_len, 4385 const char *data, int data_len) 4386 { 4387 int ret = 0; 4388 4389 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); 4390 if (ret < 0) 4391 goto out; 4392 4393 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); 4394 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); 4395 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); 4396 4397 ret = send_cmd(sctx); 4398 4399 tlv_put_failure: 4400 out: 4401 return ret; 4402 } 4403 4404 static int send_remove_xattr(struct send_ctx *sctx, 4405 struct fs_path *path, 4406 const char *name, int name_len) 4407 { 4408 int ret = 0; 4409 4410 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); 4411 if (ret < 0) 4412 goto out; 4413 4414 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); 4415 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); 4416 4417 ret = send_cmd(sctx); 4418 4419 tlv_put_failure: 4420 out: 4421 return ret; 4422 } 4423 4424 static int __process_new_xattr(int num, struct btrfs_key *di_key, 4425 const char *name, int name_len, 4426 const char *data, int data_len, 4427 u8 type, void *ctx) 4428 { 4429 int ret; 4430 struct send_ctx *sctx = ctx; 4431 struct fs_path *p; 4432 struct posix_acl_xattr_header dummy_acl; 4433 4434 p = fs_path_alloc(); 4435 if (!p) 4436 return -ENOMEM; 4437 4438 /* 4439 * This hack is needed because empty acls are stored as zero byte 4440 * data in xattrs. Problem with that is, that receiving these zero byte 4441 * acls will fail later. To fix this, we send a dummy acl list that 4442 * only contains the version number and no entries. 4443 */ 4444 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || 4445 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { 4446 if (data_len == 0) { 4447 dummy_acl.a_version = 4448 cpu_to_le32(POSIX_ACL_XATTR_VERSION); 4449 data = (char *)&dummy_acl; 4450 data_len = sizeof(dummy_acl); 4451 } 4452 } 4453 4454 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); 4455 if (ret < 0) 4456 goto out; 4457 4458 ret = send_set_xattr(sctx, p, name, name_len, data, data_len); 4459 4460 out: 4461 fs_path_free(p); 4462 return ret; 4463 } 4464 4465 static int __process_deleted_xattr(int num, struct btrfs_key *di_key, 4466 const char *name, int name_len, 4467 const char *data, int data_len, 4468 u8 type, void *ctx) 4469 { 4470 int ret; 4471 struct send_ctx *sctx = ctx; 4472 struct fs_path *p; 4473 4474 p = fs_path_alloc(); 4475 if (!p) 4476 return -ENOMEM; 4477 4478 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); 4479 if (ret < 0) 4480 goto out; 4481 4482 ret = send_remove_xattr(sctx, p, name, name_len); 4483 4484 out: 4485 fs_path_free(p); 4486 return ret; 4487 } 4488 4489 static int process_new_xattr(struct send_ctx *sctx) 4490 { 4491 int ret = 0; 4492 4493 ret = iterate_dir_item(sctx->send_root, sctx->left_path, 4494 __process_new_xattr, sctx); 4495 4496 return ret; 4497 } 4498 4499 static int process_deleted_xattr(struct send_ctx *sctx) 4500 { 4501 return iterate_dir_item(sctx->parent_root, sctx->right_path, 4502 __process_deleted_xattr, sctx); 4503 } 4504 4505 struct find_xattr_ctx { 4506 const char *name; 4507 int name_len; 4508 int found_idx; 4509 char *found_data; 4510 int found_data_len; 4511 }; 4512 4513 static int __find_xattr(int num, struct btrfs_key *di_key, 4514 const char *name, int name_len, 4515 const char *data, int data_len, 4516 u8 type, void *vctx) 4517 { 4518 struct find_xattr_ctx *ctx = vctx; 4519 4520 if (name_len == ctx->name_len && 4521 strncmp(name, ctx->name, name_len) == 0) { 4522 ctx->found_idx = num; 4523 ctx->found_data_len = data_len; 4524 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); 4525 if (!ctx->found_data) 4526 return -ENOMEM; 4527 return 1; 4528 } 4529 return 0; 4530 } 4531 4532 static int find_xattr(struct btrfs_root *root, 4533 struct btrfs_path *path, 4534 struct btrfs_key *key, 4535 const char *name, int name_len, 4536 char **data, int *data_len) 4537 { 4538 int ret; 4539 struct find_xattr_ctx ctx; 4540 4541 ctx.name = name; 4542 ctx.name_len = name_len; 4543 ctx.found_idx = -1; 4544 ctx.found_data = NULL; 4545 ctx.found_data_len = 0; 4546 4547 ret = iterate_dir_item(root, path, __find_xattr, &ctx); 4548 if (ret < 0) 4549 return ret; 4550 4551 if (ctx.found_idx == -1) 4552 return -ENOENT; 4553 if (data) { 4554 *data = ctx.found_data; 4555 *data_len = ctx.found_data_len; 4556 } else { 4557 kfree(ctx.found_data); 4558 } 4559 return ctx.found_idx; 4560 } 4561 4562 4563 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, 4564 const char *name, int name_len, 4565 const char *data, int data_len, 4566 u8 type, void *ctx) 4567 { 4568 int ret; 4569 struct send_ctx *sctx = ctx; 4570 char *found_data = NULL; 4571 int found_data_len = 0; 4572 4573 ret = find_xattr(sctx->parent_root, sctx->right_path, 4574 sctx->cmp_key, name, name_len, &found_data, 4575 &found_data_len); 4576 if (ret == -ENOENT) { 4577 ret = __process_new_xattr(num, di_key, name, name_len, data, 4578 data_len, type, ctx); 4579 } else if (ret >= 0) { 4580 if (data_len != found_data_len || 4581 memcmp(data, found_data, data_len)) { 4582 ret = __process_new_xattr(num, di_key, name, name_len, 4583 data, data_len, type, ctx); 4584 } else { 4585 ret = 0; 4586 } 4587 } 4588 4589 kfree(found_data); 4590 return ret; 4591 } 4592 4593 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, 4594 const char *name, int name_len, 4595 const char *data, int data_len, 4596 u8 type, void *ctx) 4597 { 4598 int ret; 4599 struct send_ctx *sctx = ctx; 4600 4601 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, 4602 name, name_len, NULL, NULL); 4603 if (ret == -ENOENT) 4604 ret = __process_deleted_xattr(num, di_key, name, name_len, data, 4605 data_len, type, ctx); 4606 else if (ret >= 0) 4607 ret = 0; 4608 4609 return ret; 4610 } 4611 4612 static int process_changed_xattr(struct send_ctx *sctx) 4613 { 4614 int ret = 0; 4615 4616 ret = iterate_dir_item(sctx->send_root, sctx->left_path, 4617 __process_changed_new_xattr, sctx); 4618 if (ret < 0) 4619 goto out; 4620 ret = iterate_dir_item(sctx->parent_root, sctx->right_path, 4621 __process_changed_deleted_xattr, sctx); 4622 4623 out: 4624 return ret; 4625 } 4626 4627 static int process_all_new_xattrs(struct send_ctx *sctx) 4628 { 4629 int ret; 4630 struct btrfs_root *root; 4631 struct btrfs_path *path; 4632 struct btrfs_key key; 4633 struct btrfs_key found_key; 4634 struct extent_buffer *eb; 4635 int slot; 4636 4637 path = alloc_path_for_send(); 4638 if (!path) 4639 return -ENOMEM; 4640 4641 root = sctx->send_root; 4642 4643 key.objectid = sctx->cmp_key->objectid; 4644 key.type = BTRFS_XATTR_ITEM_KEY; 4645 key.offset = 0; 4646 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4647 if (ret < 0) 4648 goto out; 4649 4650 while (1) { 4651 eb = path->nodes[0]; 4652 slot = path->slots[0]; 4653 if (slot >= btrfs_header_nritems(eb)) { 4654 ret = btrfs_next_leaf(root, path); 4655 if (ret < 0) { 4656 goto out; 4657 } else if (ret > 0) { 4658 ret = 0; 4659 break; 4660 } 4661 continue; 4662 } 4663 4664 btrfs_item_key_to_cpu(eb, &found_key, slot); 4665 if (found_key.objectid != key.objectid || 4666 found_key.type != key.type) { 4667 ret = 0; 4668 goto out; 4669 } 4670 4671 ret = iterate_dir_item(root, path, __process_new_xattr, sctx); 4672 if (ret < 0) 4673 goto out; 4674 4675 path->slots[0]++; 4676 } 4677 4678 out: 4679 btrfs_free_path(path); 4680 return ret; 4681 } 4682 4683 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len) 4684 { 4685 struct btrfs_root *root = sctx->send_root; 4686 struct btrfs_fs_info *fs_info = root->fs_info; 4687 struct inode *inode; 4688 struct page *page; 4689 char *addr; 4690 struct btrfs_key key; 4691 pgoff_t index = offset >> PAGE_SHIFT; 4692 pgoff_t last_index; 4693 unsigned pg_offset = offset & ~PAGE_MASK; 4694 ssize_t ret = 0; 4695 4696 key.objectid = sctx->cur_ino; 4697 key.type = BTRFS_INODE_ITEM_KEY; 4698 key.offset = 0; 4699 4700 inode = btrfs_iget(fs_info->sb, &key, root, NULL); 4701 if (IS_ERR(inode)) 4702 return PTR_ERR(inode); 4703 4704 if (offset + len > i_size_read(inode)) { 4705 if (offset > i_size_read(inode)) 4706 len = 0; 4707 else 4708 len = offset - i_size_read(inode); 4709 } 4710 if (len == 0) 4711 goto out; 4712 4713 last_index = (offset + len - 1) >> PAGE_SHIFT; 4714 4715 /* initial readahead */ 4716 memset(&sctx->ra, 0, sizeof(struct file_ra_state)); 4717 file_ra_state_init(&sctx->ra, inode->i_mapping); 4718 4719 while (index <= last_index) { 4720 unsigned cur_len = min_t(unsigned, len, 4721 PAGE_SIZE - pg_offset); 4722 4723 page = find_lock_page(inode->i_mapping, index); 4724 if (!page) { 4725 page_cache_sync_readahead(inode->i_mapping, &sctx->ra, 4726 NULL, index, last_index + 1 - index); 4727 4728 page = find_or_create_page(inode->i_mapping, index, 4729 GFP_KERNEL); 4730 if (!page) { 4731 ret = -ENOMEM; 4732 break; 4733 } 4734 } 4735 4736 if (PageReadahead(page)) { 4737 page_cache_async_readahead(inode->i_mapping, &sctx->ra, 4738 NULL, page, index, last_index + 1 - index); 4739 } 4740 4741 if (!PageUptodate(page)) { 4742 btrfs_readpage(NULL, page); 4743 lock_page(page); 4744 if (!PageUptodate(page)) { 4745 unlock_page(page); 4746 put_page(page); 4747 ret = -EIO; 4748 break; 4749 } 4750 } 4751 4752 addr = kmap(page); 4753 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len); 4754 kunmap(page); 4755 unlock_page(page); 4756 put_page(page); 4757 index++; 4758 pg_offset = 0; 4759 len -= cur_len; 4760 ret += cur_len; 4761 } 4762 out: 4763 iput(inode); 4764 return ret; 4765 } 4766 4767 /* 4768 * Read some bytes from the current inode/file and send a write command to 4769 * user space. 4770 */ 4771 static int send_write(struct send_ctx *sctx, u64 offset, u32 len) 4772 { 4773 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; 4774 int ret = 0; 4775 struct fs_path *p; 4776 ssize_t num_read = 0; 4777 4778 p = fs_path_alloc(); 4779 if (!p) 4780 return -ENOMEM; 4781 4782 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len); 4783 4784 num_read = fill_read_buf(sctx, offset, len); 4785 if (num_read <= 0) { 4786 if (num_read < 0) 4787 ret = num_read; 4788 goto out; 4789 } 4790 4791 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); 4792 if (ret < 0) 4793 goto out; 4794 4795 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); 4796 if (ret < 0) 4797 goto out; 4798 4799 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 4800 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); 4801 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read); 4802 4803 ret = send_cmd(sctx); 4804 4805 tlv_put_failure: 4806 out: 4807 fs_path_free(p); 4808 if (ret < 0) 4809 return ret; 4810 return num_read; 4811 } 4812 4813 /* 4814 * Send a clone command to user space. 4815 */ 4816 static int send_clone(struct send_ctx *sctx, 4817 u64 offset, u32 len, 4818 struct clone_root *clone_root) 4819 { 4820 int ret = 0; 4821 struct fs_path *p; 4822 u64 gen; 4823 4824 btrfs_debug(sctx->send_root->fs_info, 4825 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu", 4826 offset, len, clone_root->root->objectid, clone_root->ino, 4827 clone_root->offset); 4828 4829 p = fs_path_alloc(); 4830 if (!p) 4831 return -ENOMEM; 4832 4833 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); 4834 if (ret < 0) 4835 goto out; 4836 4837 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); 4838 if (ret < 0) 4839 goto out; 4840 4841 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); 4842 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); 4843 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 4844 4845 if (clone_root->root == sctx->send_root) { 4846 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, 4847 &gen, NULL, NULL, NULL, NULL); 4848 if (ret < 0) 4849 goto out; 4850 ret = get_cur_path(sctx, clone_root->ino, gen, p); 4851 } else { 4852 ret = get_inode_path(clone_root->root, clone_root->ino, p); 4853 } 4854 if (ret < 0) 4855 goto out; 4856 4857 /* 4858 * If the parent we're using has a received_uuid set then use that as 4859 * our clone source as that is what we will look for when doing a 4860 * receive. 4861 * 4862 * This covers the case that we create a snapshot off of a received 4863 * subvolume and then use that as the parent and try to receive on a 4864 * different host. 4865 */ 4866 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid)) 4867 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, 4868 clone_root->root->root_item.received_uuid); 4869 else 4870 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, 4871 clone_root->root->root_item.uuid); 4872 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, 4873 le64_to_cpu(clone_root->root->root_item.ctransid)); 4874 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); 4875 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, 4876 clone_root->offset); 4877 4878 ret = send_cmd(sctx); 4879 4880 tlv_put_failure: 4881 out: 4882 fs_path_free(p); 4883 return ret; 4884 } 4885 4886 /* 4887 * Send an update extent command to user space. 4888 */ 4889 static int send_update_extent(struct send_ctx *sctx, 4890 u64 offset, u32 len) 4891 { 4892 int ret = 0; 4893 struct fs_path *p; 4894 4895 p = fs_path_alloc(); 4896 if (!p) 4897 return -ENOMEM; 4898 4899 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); 4900 if (ret < 0) 4901 goto out; 4902 4903 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); 4904 if (ret < 0) 4905 goto out; 4906 4907 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 4908 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); 4909 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); 4910 4911 ret = send_cmd(sctx); 4912 4913 tlv_put_failure: 4914 out: 4915 fs_path_free(p); 4916 return ret; 4917 } 4918 4919 static int send_hole(struct send_ctx *sctx, u64 end) 4920 { 4921 struct fs_path *p = NULL; 4922 u64 offset = sctx->cur_inode_last_extent; 4923 u64 len; 4924 int ret = 0; 4925 4926 p = fs_path_alloc(); 4927 if (!p) 4928 return -ENOMEM; 4929 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); 4930 if (ret < 0) 4931 goto tlv_put_failure; 4932 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE); 4933 while (offset < end) { 4934 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE); 4935 4936 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); 4937 if (ret < 0) 4938 break; 4939 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); 4940 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); 4941 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len); 4942 ret = send_cmd(sctx); 4943 if (ret < 0) 4944 break; 4945 offset += len; 4946 } 4947 tlv_put_failure: 4948 fs_path_free(p); 4949 return ret; 4950 } 4951 4952 static int send_extent_data(struct send_ctx *sctx, 4953 const u64 offset, 4954 const u64 len) 4955 { 4956 u64 sent = 0; 4957 4958 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) 4959 return send_update_extent(sctx, offset, len); 4960 4961 while (sent < len) { 4962 u64 size = len - sent; 4963 int ret; 4964 4965 if (size > BTRFS_SEND_READ_SIZE) 4966 size = BTRFS_SEND_READ_SIZE; 4967 ret = send_write(sctx, offset + sent, size); 4968 if (ret < 0) 4969 return ret; 4970 if (!ret) 4971 break; 4972 sent += ret; 4973 } 4974 return 0; 4975 } 4976 4977 static int clone_range(struct send_ctx *sctx, 4978 struct clone_root *clone_root, 4979 const u64 disk_byte, 4980 u64 data_offset, 4981 u64 offset, 4982 u64 len) 4983 { 4984 struct btrfs_path *path; 4985 struct btrfs_key key; 4986 int ret; 4987 4988 /* 4989 * Prevent cloning from a zero offset with a length matching the sector 4990 * size because in some scenarios this will make the receiver fail. 4991 * 4992 * For example, if in the source filesystem the extent at offset 0 4993 * has a length of sectorsize and it was written using direct IO, then 4994 * it can never be an inline extent (even if compression is enabled). 4995 * Then this extent can be cloned in the original filesystem to a non 4996 * zero file offset, but it may not be possible to clone in the 4997 * destination filesystem because it can be inlined due to compression 4998 * on the destination filesystem (as the receiver's write operations are 4999 * always done using buffered IO). The same happens when the original 5000 * filesystem does not have compression enabled but the destination 5001 * filesystem has. 5002 */ 5003 if (clone_root->offset == 0 && 5004 len == sctx->send_root->fs_info->sectorsize) 5005 return send_extent_data(sctx, offset, len); 5006 5007 path = alloc_path_for_send(); 5008 if (!path) 5009 return -ENOMEM; 5010 5011 /* 5012 * We can't send a clone operation for the entire range if we find 5013 * extent items in the respective range in the source file that 5014 * refer to different extents or if we find holes. 5015 * So check for that and do a mix of clone and regular write/copy 5016 * operations if needed. 5017 * 5018 * Example: 5019 * 5020 * mkfs.btrfs -f /dev/sda 5021 * mount /dev/sda /mnt 5022 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo 5023 * cp --reflink=always /mnt/foo /mnt/bar 5024 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo 5025 * btrfs subvolume snapshot -r /mnt /mnt/snap 5026 * 5027 * If when we send the snapshot and we are processing file bar (which 5028 * has a higher inode number than foo) we blindly send a clone operation 5029 * for the [0, 100K[ range from foo to bar, the receiver ends up getting 5030 * a file bar that matches the content of file foo - iow, doesn't match 5031 * the content from bar in the original filesystem. 5032 */ 5033 key.objectid = clone_root->ino; 5034 key.type = BTRFS_EXTENT_DATA_KEY; 5035 key.offset = clone_root->offset; 5036 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); 5037 if (ret < 0) 5038 goto out; 5039 if (ret > 0 && path->slots[0] > 0) { 5040 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); 5041 if (key.objectid == clone_root->ino && 5042 key.type == BTRFS_EXTENT_DATA_KEY) 5043 path->slots[0]--; 5044 } 5045 5046 while (true) { 5047 struct extent_buffer *leaf = path->nodes[0]; 5048 int slot = path->slots[0]; 5049 struct btrfs_file_extent_item *ei; 5050 u8 type; 5051 u64 ext_len; 5052 u64 clone_len; 5053 5054 if (slot >= btrfs_header_nritems(leaf)) { 5055 ret = btrfs_next_leaf(clone_root->root, path); 5056 if (ret < 0) 5057 goto out; 5058 else if (ret > 0) 5059 break; 5060 continue; 5061 } 5062 5063 btrfs_item_key_to_cpu(leaf, &key, slot); 5064 5065 /* 5066 * We might have an implicit trailing hole (NO_HOLES feature 5067 * enabled). We deal with it after leaving this loop. 5068 */ 5069 if (key.objectid != clone_root->ino || 5070 key.type != BTRFS_EXTENT_DATA_KEY) 5071 break; 5072 5073 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 5074 type = btrfs_file_extent_type(leaf, ei); 5075 if (type == BTRFS_FILE_EXTENT_INLINE) { 5076 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei); 5077 ext_len = PAGE_ALIGN(ext_len); 5078 } else { 5079 ext_len = btrfs_file_extent_num_bytes(leaf, ei); 5080 } 5081 5082 if (key.offset + ext_len <= clone_root->offset) 5083 goto next; 5084 5085 if (key.offset > clone_root->offset) { 5086 /* Implicit hole, NO_HOLES feature enabled. */ 5087 u64 hole_len = key.offset - clone_root->offset; 5088 5089 if (hole_len > len) 5090 hole_len = len; 5091 ret = send_extent_data(sctx, offset, hole_len); 5092 if (ret < 0) 5093 goto out; 5094 5095 len -= hole_len; 5096 if (len == 0) 5097 break; 5098 offset += hole_len; 5099 clone_root->offset += hole_len; 5100 data_offset += hole_len; 5101 } 5102 5103 if (key.offset >= clone_root->offset + len) 5104 break; 5105 5106 clone_len = min_t(u64, ext_len, len); 5107 5108 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && 5109 btrfs_file_extent_offset(leaf, ei) == data_offset) 5110 ret = send_clone(sctx, offset, clone_len, clone_root); 5111 else 5112 ret = send_extent_data(sctx, offset, clone_len); 5113 5114 if (ret < 0) 5115 goto out; 5116 5117 len -= clone_len; 5118 if (len == 0) 5119 break; 5120 offset += clone_len; 5121 clone_root->offset += clone_len; 5122 data_offset += clone_len; 5123 next: 5124 path->slots[0]++; 5125 } 5126 5127 if (len > 0) 5128 ret = send_extent_data(sctx, offset, len); 5129 else 5130 ret = 0; 5131 out: 5132 btrfs_free_path(path); 5133 return ret; 5134 } 5135 5136 static int send_write_or_clone(struct send_ctx *sctx, 5137 struct btrfs_path *path, 5138 struct btrfs_key *key, 5139 struct clone_root *clone_root) 5140 { 5141 int ret = 0; 5142 struct btrfs_file_extent_item *ei; 5143 u64 offset = key->offset; 5144 u64 len; 5145 u8 type; 5146 u64 bs = sctx->send_root->fs_info->sb->s_blocksize; 5147 5148 ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 5149 struct btrfs_file_extent_item); 5150 type = btrfs_file_extent_type(path->nodes[0], ei); 5151 if (type == BTRFS_FILE_EXTENT_INLINE) { 5152 len = btrfs_file_extent_inline_len(path->nodes[0], 5153 path->slots[0], ei); 5154 /* 5155 * it is possible the inline item won't cover the whole page, 5156 * but there may be items after this page. Make 5157 * sure to send the whole thing 5158 */ 5159 len = PAGE_ALIGN(len); 5160 } else { 5161 len = btrfs_file_extent_num_bytes(path->nodes[0], ei); 5162 } 5163 5164 if (offset + len > sctx->cur_inode_size) 5165 len = sctx->cur_inode_size - offset; 5166 if (len == 0) { 5167 ret = 0; 5168 goto out; 5169 } 5170 5171 if (clone_root && IS_ALIGNED(offset + len, bs)) { 5172 u64 disk_byte; 5173 u64 data_offset; 5174 5175 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); 5176 data_offset = btrfs_file_extent_offset(path->nodes[0], ei); 5177 ret = clone_range(sctx, clone_root, disk_byte, data_offset, 5178 offset, len); 5179 } else { 5180 ret = send_extent_data(sctx, offset, len); 5181 } 5182 out: 5183 return ret; 5184 } 5185 5186 static int is_extent_unchanged(struct send_ctx *sctx, 5187 struct btrfs_path *left_path, 5188 struct btrfs_key *ekey) 5189 { 5190 int ret = 0; 5191 struct btrfs_key key; 5192 struct btrfs_path *path = NULL; 5193 struct extent_buffer *eb; 5194 int slot; 5195 struct btrfs_key found_key; 5196 struct btrfs_file_extent_item *ei; 5197 u64 left_disknr; 5198 u64 right_disknr; 5199 u64 left_offset; 5200 u64 right_offset; 5201 u64 left_offset_fixed; 5202 u64 left_len; 5203 u64 right_len; 5204 u64 left_gen; 5205 u64 right_gen; 5206 u8 left_type; 5207 u8 right_type; 5208 5209 path = alloc_path_for_send(); 5210 if (!path) 5211 return -ENOMEM; 5212 5213 eb = left_path->nodes[0]; 5214 slot = left_path->slots[0]; 5215 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); 5216 left_type = btrfs_file_extent_type(eb, ei); 5217 5218 if (left_type != BTRFS_FILE_EXTENT_REG) { 5219 ret = 0; 5220 goto out; 5221 } 5222 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); 5223 left_len = btrfs_file_extent_num_bytes(eb, ei); 5224 left_offset = btrfs_file_extent_offset(eb, ei); 5225 left_gen = btrfs_file_extent_generation(eb, ei); 5226 5227 /* 5228 * Following comments will refer to these graphics. L is the left 5229 * extents which we are checking at the moment. 1-8 are the right 5230 * extents that we iterate. 5231 * 5232 * |-----L-----| 5233 * |-1-|-2a-|-3-|-4-|-5-|-6-| 5234 * 5235 * |-----L-----| 5236 * |--1--|-2b-|...(same as above) 5237 * 5238 * Alternative situation. Happens on files where extents got split. 5239 * |-----L-----| 5240 * |-----------7-----------|-6-| 5241 * 5242 * Alternative situation. Happens on files which got larger. 5243 * |-----L-----| 5244 * |-8-| 5245 * Nothing follows after 8. 5246 */ 5247 5248 key.objectid = ekey->objectid; 5249 key.type = BTRFS_EXTENT_DATA_KEY; 5250 key.offset = ekey->offset; 5251 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); 5252 if (ret < 0) 5253 goto out; 5254 if (ret) { 5255 ret = 0; 5256 goto out; 5257 } 5258 5259 /* 5260 * Handle special case where the right side has no extents at all. 5261 */ 5262 eb = path->nodes[0]; 5263 slot = path->slots[0]; 5264 btrfs_item_key_to_cpu(eb, &found_key, slot); 5265 if (found_key.objectid != key.objectid || 5266 found_key.type != key.type) { 5267 /* If we're a hole then just pretend nothing changed */ 5268 ret = (left_disknr) ? 0 : 1; 5269 goto out; 5270 } 5271 5272 /* 5273 * We're now on 2a, 2b or 7. 5274 */ 5275 key = found_key; 5276 while (key.offset < ekey->offset + left_len) { 5277 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); 5278 right_type = btrfs_file_extent_type(eb, ei); 5279 if (right_type != BTRFS_FILE_EXTENT_REG && 5280 right_type != BTRFS_FILE_EXTENT_INLINE) { 5281 ret = 0; 5282 goto out; 5283 } 5284 5285 if (right_type == BTRFS_FILE_EXTENT_INLINE) { 5286 right_len = btrfs_file_extent_inline_len(eb, slot, ei); 5287 right_len = PAGE_ALIGN(right_len); 5288 } else { 5289 right_len = btrfs_file_extent_num_bytes(eb, ei); 5290 } 5291 5292 /* 5293 * Are we at extent 8? If yes, we know the extent is changed. 5294 * This may only happen on the first iteration. 5295 */ 5296 if (found_key.offset + right_len <= ekey->offset) { 5297 /* If we're a hole just pretend nothing changed */ 5298 ret = (left_disknr) ? 0 : 1; 5299 goto out; 5300 } 5301 5302 /* 5303 * We just wanted to see if when we have an inline extent, what 5304 * follows it is a regular extent (wanted to check the above 5305 * condition for inline extents too). This should normally not 5306 * happen but it's possible for example when we have an inline 5307 * compressed extent representing data with a size matching 5308 * the page size (currently the same as sector size). 5309 */ 5310 if (right_type == BTRFS_FILE_EXTENT_INLINE) { 5311 ret = 0; 5312 goto out; 5313 } 5314 5315 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); 5316 right_offset = btrfs_file_extent_offset(eb, ei); 5317 right_gen = btrfs_file_extent_generation(eb, ei); 5318 5319 left_offset_fixed = left_offset; 5320 if (key.offset < ekey->offset) { 5321 /* Fix the right offset for 2a and 7. */ 5322 right_offset += ekey->offset - key.offset; 5323 } else { 5324 /* Fix the left offset for all behind 2a and 2b */ 5325 left_offset_fixed += key.offset - ekey->offset; 5326 } 5327 5328 /* 5329 * Check if we have the same extent. 5330 */ 5331 if (left_disknr != right_disknr || 5332 left_offset_fixed != right_offset || 5333 left_gen != right_gen) { 5334 ret = 0; 5335 goto out; 5336 } 5337 5338 /* 5339 * Go to the next extent. 5340 */ 5341 ret = btrfs_next_item(sctx->parent_root, path); 5342 if (ret < 0) 5343 goto out; 5344 if (!ret) { 5345 eb = path->nodes[0]; 5346 slot = path->slots[0]; 5347 btrfs_item_key_to_cpu(eb, &found_key, slot); 5348 } 5349 if (ret || found_key.objectid != key.objectid || 5350 found_key.type != key.type) { 5351 key.offset += right_len; 5352 break; 5353 } 5354 if (found_key.offset != key.offset + right_len) { 5355 ret = 0; 5356 goto out; 5357 } 5358 key = found_key; 5359 } 5360 5361 /* 5362 * We're now behind the left extent (treat as unchanged) or at the end 5363 * of the right side (treat as changed). 5364 */ 5365 if (key.offset >= ekey->offset + left_len) 5366 ret = 1; 5367 else 5368 ret = 0; 5369 5370 5371 out: 5372 btrfs_free_path(path); 5373 return ret; 5374 } 5375 5376 static int get_last_extent(struct send_ctx *sctx, u64 offset) 5377 { 5378 struct btrfs_path *path; 5379 struct btrfs_root *root = sctx->send_root; 5380 struct btrfs_file_extent_item *fi; 5381 struct btrfs_key key; 5382 u64 extent_end; 5383 u8 type; 5384 int ret; 5385 5386 path = alloc_path_for_send(); 5387 if (!path) 5388 return -ENOMEM; 5389 5390 sctx->cur_inode_last_extent = 0; 5391 5392 key.objectid = sctx->cur_ino; 5393 key.type = BTRFS_EXTENT_DATA_KEY; 5394 key.offset = offset; 5395 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); 5396 if (ret < 0) 5397 goto out; 5398 ret = 0; 5399 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 5400 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) 5401 goto out; 5402 5403 fi = btrfs_item_ptr(path->nodes[0], path->slots[0], 5404 struct btrfs_file_extent_item); 5405 type = btrfs_file_extent_type(path->nodes[0], fi); 5406 if (type == BTRFS_FILE_EXTENT_INLINE) { 5407 u64 size = btrfs_file_extent_inline_len(path->nodes[0], 5408 path->slots[0], fi); 5409 extent_end = ALIGN(key.offset + size, 5410 sctx->send_root->fs_info->sectorsize); 5411 } else { 5412 extent_end = key.offset + 5413 btrfs_file_extent_num_bytes(path->nodes[0], fi); 5414 } 5415 sctx->cur_inode_last_extent = extent_end; 5416 out: 5417 btrfs_free_path(path); 5418 return ret; 5419 } 5420 5421 static int range_is_hole_in_parent(struct send_ctx *sctx, 5422 const u64 start, 5423 const u64 end) 5424 { 5425 struct btrfs_path *path; 5426 struct btrfs_key key; 5427 struct btrfs_root *root = sctx->parent_root; 5428 u64 search_start = start; 5429 int ret; 5430 5431 path = alloc_path_for_send(); 5432 if (!path) 5433 return -ENOMEM; 5434 5435 key.objectid = sctx->cur_ino; 5436 key.type = BTRFS_EXTENT_DATA_KEY; 5437 key.offset = search_start; 5438 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5439 if (ret < 0) 5440 goto out; 5441 if (ret > 0 && path->slots[0] > 0) 5442 path->slots[0]--; 5443 5444 while (search_start < end) { 5445 struct extent_buffer *leaf = path->nodes[0]; 5446 int slot = path->slots[0]; 5447 struct btrfs_file_extent_item *fi; 5448 u64 extent_end; 5449 5450 if (slot >= btrfs_header_nritems(leaf)) { 5451 ret = btrfs_next_leaf(root, path); 5452 if (ret < 0) 5453 goto out; 5454 else if (ret > 0) 5455 break; 5456 continue; 5457 } 5458 5459 btrfs_item_key_to_cpu(leaf, &key, slot); 5460 if (key.objectid < sctx->cur_ino || 5461 key.type < BTRFS_EXTENT_DATA_KEY) 5462 goto next; 5463 if (key.objectid > sctx->cur_ino || 5464 key.type > BTRFS_EXTENT_DATA_KEY || 5465 key.offset >= end) 5466 break; 5467 5468 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 5469 if (btrfs_file_extent_type(leaf, fi) == 5470 BTRFS_FILE_EXTENT_INLINE) { 5471 u64 size = btrfs_file_extent_inline_len(leaf, slot, fi); 5472 5473 extent_end = ALIGN(key.offset + size, 5474 root->fs_info->sectorsize); 5475 } else { 5476 extent_end = key.offset + 5477 btrfs_file_extent_num_bytes(leaf, fi); 5478 } 5479 if (extent_end <= start) 5480 goto next; 5481 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) { 5482 search_start = extent_end; 5483 goto next; 5484 } 5485 ret = 0; 5486 goto out; 5487 next: 5488 path->slots[0]++; 5489 } 5490 ret = 1; 5491 out: 5492 btrfs_free_path(path); 5493 return ret; 5494 } 5495 5496 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, 5497 struct btrfs_key *key) 5498 { 5499 struct btrfs_file_extent_item *fi; 5500 u64 extent_end; 5501 u8 type; 5502 int ret = 0; 5503 5504 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) 5505 return 0; 5506 5507 if (sctx->cur_inode_last_extent == (u64)-1) { 5508 ret = get_last_extent(sctx, key->offset - 1); 5509 if (ret) 5510 return ret; 5511 } 5512 5513 fi = btrfs_item_ptr(path->nodes[0], path->slots[0], 5514 struct btrfs_file_extent_item); 5515 type = btrfs_file_extent_type(path->nodes[0], fi); 5516 if (type == BTRFS_FILE_EXTENT_INLINE) { 5517 u64 size = btrfs_file_extent_inline_len(path->nodes[0], 5518 path->slots[0], fi); 5519 extent_end = ALIGN(key->offset + size, 5520 sctx->send_root->fs_info->sectorsize); 5521 } else { 5522 extent_end = key->offset + 5523 btrfs_file_extent_num_bytes(path->nodes[0], fi); 5524 } 5525 5526 if (path->slots[0] == 0 && 5527 sctx->cur_inode_last_extent < key->offset) { 5528 /* 5529 * We might have skipped entire leafs that contained only 5530 * file extent items for our current inode. These leafs have 5531 * a generation number smaller (older) than the one in the 5532 * current leaf and the leaf our last extent came from, and 5533 * are located between these 2 leafs. 5534 */ 5535 ret = get_last_extent(sctx, key->offset - 1); 5536 if (ret) 5537 return ret; 5538 } 5539 5540 if (sctx->cur_inode_last_extent < key->offset) { 5541 ret = range_is_hole_in_parent(sctx, 5542 sctx->cur_inode_last_extent, 5543 key->offset); 5544 if (ret < 0) 5545 return ret; 5546 else if (ret == 0) 5547 ret = send_hole(sctx, key->offset); 5548 else 5549 ret = 0; 5550 } 5551 sctx->cur_inode_last_extent = extent_end; 5552 return ret; 5553 } 5554 5555 static int process_extent(struct send_ctx *sctx, 5556 struct btrfs_path *path, 5557 struct btrfs_key *key) 5558 { 5559 struct clone_root *found_clone = NULL; 5560 int ret = 0; 5561 5562 if (S_ISLNK(sctx->cur_inode_mode)) 5563 return 0; 5564 5565 if (sctx->parent_root && !sctx->cur_inode_new) { 5566 ret = is_extent_unchanged(sctx, path, key); 5567 if (ret < 0) 5568 goto out; 5569 if (ret) { 5570 ret = 0; 5571 goto out_hole; 5572 } 5573 } else { 5574 struct btrfs_file_extent_item *ei; 5575 u8 type; 5576 5577 ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 5578 struct btrfs_file_extent_item); 5579 type = btrfs_file_extent_type(path->nodes[0], ei); 5580 if (type == BTRFS_FILE_EXTENT_PREALLOC || 5581 type == BTRFS_FILE_EXTENT_REG) { 5582 /* 5583 * The send spec does not have a prealloc command yet, 5584 * so just leave a hole for prealloc'ed extents until 5585 * we have enough commands queued up to justify rev'ing 5586 * the send spec. 5587 */ 5588 if (type == BTRFS_FILE_EXTENT_PREALLOC) { 5589 ret = 0; 5590 goto out; 5591 } 5592 5593 /* Have a hole, just skip it. */ 5594 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { 5595 ret = 0; 5596 goto out; 5597 } 5598 } 5599 } 5600 5601 ret = find_extent_clone(sctx, path, key->objectid, key->offset, 5602 sctx->cur_inode_size, &found_clone); 5603 if (ret != -ENOENT && ret < 0) 5604 goto out; 5605 5606 ret = send_write_or_clone(sctx, path, key, found_clone); 5607 if (ret) 5608 goto out; 5609 out_hole: 5610 ret = maybe_send_hole(sctx, path, key); 5611 out: 5612 return ret; 5613 } 5614 5615 static int process_all_extents(struct send_ctx *sctx) 5616 { 5617 int ret; 5618 struct btrfs_root *root; 5619 struct btrfs_path *path; 5620 struct btrfs_key key; 5621 struct btrfs_key found_key; 5622 struct extent_buffer *eb; 5623 int slot; 5624 5625 root = sctx->send_root; 5626 path = alloc_path_for_send(); 5627 if (!path) 5628 return -ENOMEM; 5629 5630 key.objectid = sctx->cmp_key->objectid; 5631 key.type = BTRFS_EXTENT_DATA_KEY; 5632 key.offset = 0; 5633 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5634 if (ret < 0) 5635 goto out; 5636 5637 while (1) { 5638 eb = path->nodes[0]; 5639 slot = path->slots[0]; 5640 5641 if (slot >= btrfs_header_nritems(eb)) { 5642 ret = btrfs_next_leaf(root, path); 5643 if (ret < 0) { 5644 goto out; 5645 } else if (ret > 0) { 5646 ret = 0; 5647 break; 5648 } 5649 continue; 5650 } 5651 5652 btrfs_item_key_to_cpu(eb, &found_key, slot); 5653 5654 if (found_key.objectid != key.objectid || 5655 found_key.type != key.type) { 5656 ret = 0; 5657 goto out; 5658 } 5659 5660 ret = process_extent(sctx, path, &found_key); 5661 if (ret < 0) 5662 goto out; 5663 5664 path->slots[0]++; 5665 } 5666 5667 out: 5668 btrfs_free_path(path); 5669 return ret; 5670 } 5671 5672 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, 5673 int *pending_move, 5674 int *refs_processed) 5675 { 5676 int ret = 0; 5677 5678 if (sctx->cur_ino == 0) 5679 goto out; 5680 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && 5681 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) 5682 goto out; 5683 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) 5684 goto out; 5685 5686 ret = process_recorded_refs(sctx, pending_move); 5687 if (ret < 0) 5688 goto out; 5689 5690 *refs_processed = 1; 5691 out: 5692 return ret; 5693 } 5694 5695 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) 5696 { 5697 int ret = 0; 5698 u64 left_mode; 5699 u64 left_uid; 5700 u64 left_gid; 5701 u64 right_mode; 5702 u64 right_uid; 5703 u64 right_gid; 5704 int need_chmod = 0; 5705 int need_chown = 0; 5706 int pending_move = 0; 5707 int refs_processed = 0; 5708 5709 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, 5710 &refs_processed); 5711 if (ret < 0) 5712 goto out; 5713 5714 /* 5715 * We have processed the refs and thus need to advance send_progress. 5716 * Now, calls to get_cur_xxx will take the updated refs of the current 5717 * inode into account. 5718 * 5719 * On the other hand, if our current inode is a directory and couldn't 5720 * be moved/renamed because its parent was renamed/moved too and it has 5721 * a higher inode number, we can only move/rename our current inode 5722 * after we moved/renamed its parent. Therefore in this case operate on 5723 * the old path (pre move/rename) of our current inode, and the 5724 * move/rename will be performed later. 5725 */ 5726 if (refs_processed && !pending_move) 5727 sctx->send_progress = sctx->cur_ino + 1; 5728 5729 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) 5730 goto out; 5731 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) 5732 goto out; 5733 5734 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, 5735 &left_mode, &left_uid, &left_gid, NULL); 5736 if (ret < 0) 5737 goto out; 5738 5739 if (!sctx->parent_root || sctx->cur_inode_new) { 5740 need_chown = 1; 5741 if (!S_ISLNK(sctx->cur_inode_mode)) 5742 need_chmod = 1; 5743 } else { 5744 ret = get_inode_info(sctx->parent_root, sctx->cur_ino, 5745 NULL, NULL, &right_mode, &right_uid, 5746 &right_gid, NULL); 5747 if (ret < 0) 5748 goto out; 5749 5750 if (left_uid != right_uid || left_gid != right_gid) 5751 need_chown = 1; 5752 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) 5753 need_chmod = 1; 5754 } 5755 5756 if (S_ISREG(sctx->cur_inode_mode)) { 5757 if (need_send_hole(sctx)) { 5758 if (sctx->cur_inode_last_extent == (u64)-1 || 5759 sctx->cur_inode_last_extent < 5760 sctx->cur_inode_size) { 5761 ret = get_last_extent(sctx, (u64)-1); 5762 if (ret) 5763 goto out; 5764 } 5765 if (sctx->cur_inode_last_extent < 5766 sctx->cur_inode_size) { 5767 ret = send_hole(sctx, sctx->cur_inode_size); 5768 if (ret) 5769 goto out; 5770 } 5771 } 5772 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen, 5773 sctx->cur_inode_size); 5774 if (ret < 0) 5775 goto out; 5776 } 5777 5778 if (need_chown) { 5779 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, 5780 left_uid, left_gid); 5781 if (ret < 0) 5782 goto out; 5783 } 5784 if (need_chmod) { 5785 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, 5786 left_mode); 5787 if (ret < 0) 5788 goto out; 5789 } 5790 5791 /* 5792 * If other directory inodes depended on our current directory 5793 * inode's move/rename, now do their move/rename operations. 5794 */ 5795 if (!is_waiting_for_move(sctx, sctx->cur_ino)) { 5796 ret = apply_children_dir_moves(sctx); 5797 if (ret) 5798 goto out; 5799 /* 5800 * Need to send that every time, no matter if it actually 5801 * changed between the two trees as we have done changes to 5802 * the inode before. If our inode is a directory and it's 5803 * waiting to be moved/renamed, we will send its utimes when 5804 * it's moved/renamed, therefore we don't need to do it here. 5805 */ 5806 sctx->send_progress = sctx->cur_ino + 1; 5807 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); 5808 if (ret < 0) 5809 goto out; 5810 } 5811 5812 out: 5813 return ret; 5814 } 5815 5816 static int changed_inode(struct send_ctx *sctx, 5817 enum btrfs_compare_tree_result result) 5818 { 5819 int ret = 0; 5820 struct btrfs_key *key = sctx->cmp_key; 5821 struct btrfs_inode_item *left_ii = NULL; 5822 struct btrfs_inode_item *right_ii = NULL; 5823 u64 left_gen = 0; 5824 u64 right_gen = 0; 5825 5826 sctx->cur_ino = key->objectid; 5827 sctx->cur_inode_new_gen = 0; 5828 sctx->cur_inode_last_extent = (u64)-1; 5829 5830 /* 5831 * Set send_progress to current inode. This will tell all get_cur_xxx 5832 * functions that the current inode's refs are not updated yet. Later, 5833 * when process_recorded_refs is finished, it is set to cur_ino + 1. 5834 */ 5835 sctx->send_progress = sctx->cur_ino; 5836 5837 if (result == BTRFS_COMPARE_TREE_NEW || 5838 result == BTRFS_COMPARE_TREE_CHANGED) { 5839 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], 5840 sctx->left_path->slots[0], 5841 struct btrfs_inode_item); 5842 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], 5843 left_ii); 5844 } else { 5845 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], 5846 sctx->right_path->slots[0], 5847 struct btrfs_inode_item); 5848 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], 5849 right_ii); 5850 } 5851 if (result == BTRFS_COMPARE_TREE_CHANGED) { 5852 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], 5853 sctx->right_path->slots[0], 5854 struct btrfs_inode_item); 5855 5856 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], 5857 right_ii); 5858 5859 /* 5860 * The cur_ino = root dir case is special here. We can't treat 5861 * the inode as deleted+reused because it would generate a 5862 * stream that tries to delete/mkdir the root dir. 5863 */ 5864 if (left_gen != right_gen && 5865 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) 5866 sctx->cur_inode_new_gen = 1; 5867 } 5868 5869 if (result == BTRFS_COMPARE_TREE_NEW) { 5870 sctx->cur_inode_gen = left_gen; 5871 sctx->cur_inode_new = 1; 5872 sctx->cur_inode_deleted = 0; 5873 sctx->cur_inode_size = btrfs_inode_size( 5874 sctx->left_path->nodes[0], left_ii); 5875 sctx->cur_inode_mode = btrfs_inode_mode( 5876 sctx->left_path->nodes[0], left_ii); 5877 sctx->cur_inode_rdev = btrfs_inode_rdev( 5878 sctx->left_path->nodes[0], left_ii); 5879 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) 5880 ret = send_create_inode_if_needed(sctx); 5881 } else if (result == BTRFS_COMPARE_TREE_DELETED) { 5882 sctx->cur_inode_gen = right_gen; 5883 sctx->cur_inode_new = 0; 5884 sctx->cur_inode_deleted = 1; 5885 sctx->cur_inode_size = btrfs_inode_size( 5886 sctx->right_path->nodes[0], right_ii); 5887 sctx->cur_inode_mode = btrfs_inode_mode( 5888 sctx->right_path->nodes[0], right_ii); 5889 } else if (result == BTRFS_COMPARE_TREE_CHANGED) { 5890 /* 5891 * We need to do some special handling in case the inode was 5892 * reported as changed with a changed generation number. This 5893 * means that the original inode was deleted and new inode 5894 * reused the same inum. So we have to treat the old inode as 5895 * deleted and the new one as new. 5896 */ 5897 if (sctx->cur_inode_new_gen) { 5898 /* 5899 * First, process the inode as if it was deleted. 5900 */ 5901 sctx->cur_inode_gen = right_gen; 5902 sctx->cur_inode_new = 0; 5903 sctx->cur_inode_deleted = 1; 5904 sctx->cur_inode_size = btrfs_inode_size( 5905 sctx->right_path->nodes[0], right_ii); 5906 sctx->cur_inode_mode = btrfs_inode_mode( 5907 sctx->right_path->nodes[0], right_ii); 5908 ret = process_all_refs(sctx, 5909 BTRFS_COMPARE_TREE_DELETED); 5910 if (ret < 0) 5911 goto out; 5912 5913 /* 5914 * Now process the inode as if it was new. 5915 */ 5916 sctx->cur_inode_gen = left_gen; 5917 sctx->cur_inode_new = 1; 5918 sctx->cur_inode_deleted = 0; 5919 sctx->cur_inode_size = btrfs_inode_size( 5920 sctx->left_path->nodes[0], left_ii); 5921 sctx->cur_inode_mode = btrfs_inode_mode( 5922 sctx->left_path->nodes[0], left_ii); 5923 sctx->cur_inode_rdev = btrfs_inode_rdev( 5924 sctx->left_path->nodes[0], left_ii); 5925 ret = send_create_inode_if_needed(sctx); 5926 if (ret < 0) 5927 goto out; 5928 5929 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); 5930 if (ret < 0) 5931 goto out; 5932 /* 5933 * Advance send_progress now as we did not get into 5934 * process_recorded_refs_if_needed in the new_gen case. 5935 */ 5936 sctx->send_progress = sctx->cur_ino + 1; 5937 5938 /* 5939 * Now process all extents and xattrs of the inode as if 5940 * they were all new. 5941 */ 5942 ret = process_all_extents(sctx); 5943 if (ret < 0) 5944 goto out; 5945 ret = process_all_new_xattrs(sctx); 5946 if (ret < 0) 5947 goto out; 5948 } else { 5949 sctx->cur_inode_gen = left_gen; 5950 sctx->cur_inode_new = 0; 5951 sctx->cur_inode_new_gen = 0; 5952 sctx->cur_inode_deleted = 0; 5953 sctx->cur_inode_size = btrfs_inode_size( 5954 sctx->left_path->nodes[0], left_ii); 5955 sctx->cur_inode_mode = btrfs_inode_mode( 5956 sctx->left_path->nodes[0], left_ii); 5957 } 5958 } 5959 5960 out: 5961 return ret; 5962 } 5963 5964 /* 5965 * We have to process new refs before deleted refs, but compare_trees gives us 5966 * the new and deleted refs mixed. To fix this, we record the new/deleted refs 5967 * first and later process them in process_recorded_refs. 5968 * For the cur_inode_new_gen case, we skip recording completely because 5969 * changed_inode did already initiate processing of refs. The reason for this is 5970 * that in this case, compare_tree actually compares the refs of 2 different 5971 * inodes. To fix this, process_all_refs is used in changed_inode to handle all 5972 * refs of the right tree as deleted and all refs of the left tree as new. 5973 */ 5974 static int changed_ref(struct send_ctx *sctx, 5975 enum btrfs_compare_tree_result result) 5976 { 5977 int ret = 0; 5978 5979 if (sctx->cur_ino != sctx->cmp_key->objectid) { 5980 inconsistent_snapshot_error(sctx, result, "reference"); 5981 return -EIO; 5982 } 5983 5984 if (!sctx->cur_inode_new_gen && 5985 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { 5986 if (result == BTRFS_COMPARE_TREE_NEW) 5987 ret = record_new_ref(sctx); 5988 else if (result == BTRFS_COMPARE_TREE_DELETED) 5989 ret = record_deleted_ref(sctx); 5990 else if (result == BTRFS_COMPARE_TREE_CHANGED) 5991 ret = record_changed_ref(sctx); 5992 } 5993 5994 return ret; 5995 } 5996 5997 /* 5998 * Process new/deleted/changed xattrs. We skip processing in the 5999 * cur_inode_new_gen case because changed_inode did already initiate processing 6000 * of xattrs. The reason is the same as in changed_ref 6001 */ 6002 static int changed_xattr(struct send_ctx *sctx, 6003 enum btrfs_compare_tree_result result) 6004 { 6005 int ret = 0; 6006 6007 if (sctx->cur_ino != sctx->cmp_key->objectid) { 6008 inconsistent_snapshot_error(sctx, result, "xattr"); 6009 return -EIO; 6010 } 6011 6012 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { 6013 if (result == BTRFS_COMPARE_TREE_NEW) 6014 ret = process_new_xattr(sctx); 6015 else if (result == BTRFS_COMPARE_TREE_DELETED) 6016 ret = process_deleted_xattr(sctx); 6017 else if (result == BTRFS_COMPARE_TREE_CHANGED) 6018 ret = process_changed_xattr(sctx); 6019 } 6020 6021 return ret; 6022 } 6023 6024 /* 6025 * Process new/deleted/changed extents. We skip processing in the 6026 * cur_inode_new_gen case because changed_inode did already initiate processing 6027 * of extents. The reason is the same as in changed_ref 6028 */ 6029 static int changed_extent(struct send_ctx *sctx, 6030 enum btrfs_compare_tree_result result) 6031 { 6032 int ret = 0; 6033 6034 if (sctx->cur_ino != sctx->cmp_key->objectid) { 6035 6036 if (result == BTRFS_COMPARE_TREE_CHANGED) { 6037 struct extent_buffer *leaf_l; 6038 struct extent_buffer *leaf_r; 6039 struct btrfs_file_extent_item *ei_l; 6040 struct btrfs_file_extent_item *ei_r; 6041 6042 leaf_l = sctx->left_path->nodes[0]; 6043 leaf_r = sctx->right_path->nodes[0]; 6044 ei_l = btrfs_item_ptr(leaf_l, 6045 sctx->left_path->slots[0], 6046 struct btrfs_file_extent_item); 6047 ei_r = btrfs_item_ptr(leaf_r, 6048 sctx->right_path->slots[0], 6049 struct btrfs_file_extent_item); 6050 6051 /* 6052 * We may have found an extent item that has changed 6053 * only its disk_bytenr field and the corresponding 6054 * inode item was not updated. This case happens due to 6055 * very specific timings during relocation when a leaf 6056 * that contains file extent items is COWed while 6057 * relocation is ongoing and its in the stage where it 6058 * updates data pointers. So when this happens we can 6059 * safely ignore it since we know it's the same extent, 6060 * but just at different logical and physical locations 6061 * (when an extent is fully replaced with a new one, we 6062 * know the generation number must have changed too, 6063 * since snapshot creation implies committing the current 6064 * transaction, and the inode item must have been updated 6065 * as well). 6066 * This replacement of the disk_bytenr happens at 6067 * relocation.c:replace_file_extents() through 6068 * relocation.c:btrfs_reloc_cow_block(). 6069 */ 6070 if (btrfs_file_extent_generation(leaf_l, ei_l) == 6071 btrfs_file_extent_generation(leaf_r, ei_r) && 6072 btrfs_file_extent_ram_bytes(leaf_l, ei_l) == 6073 btrfs_file_extent_ram_bytes(leaf_r, ei_r) && 6074 btrfs_file_extent_compression(leaf_l, ei_l) == 6075 btrfs_file_extent_compression(leaf_r, ei_r) && 6076 btrfs_file_extent_encryption(leaf_l, ei_l) == 6077 btrfs_file_extent_encryption(leaf_r, ei_r) && 6078 btrfs_file_extent_other_encoding(leaf_l, ei_l) == 6079 btrfs_file_extent_other_encoding(leaf_r, ei_r) && 6080 btrfs_file_extent_type(leaf_l, ei_l) == 6081 btrfs_file_extent_type(leaf_r, ei_r) && 6082 btrfs_file_extent_disk_bytenr(leaf_l, ei_l) != 6083 btrfs_file_extent_disk_bytenr(leaf_r, ei_r) && 6084 btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) == 6085 btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) && 6086 btrfs_file_extent_offset(leaf_l, ei_l) == 6087 btrfs_file_extent_offset(leaf_r, ei_r) && 6088 btrfs_file_extent_num_bytes(leaf_l, ei_l) == 6089 btrfs_file_extent_num_bytes(leaf_r, ei_r)) 6090 return 0; 6091 } 6092 6093 inconsistent_snapshot_error(sctx, result, "extent"); 6094 return -EIO; 6095 } 6096 6097 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { 6098 if (result != BTRFS_COMPARE_TREE_DELETED) 6099 ret = process_extent(sctx, sctx->left_path, 6100 sctx->cmp_key); 6101 } 6102 6103 return ret; 6104 } 6105 6106 static int dir_changed(struct send_ctx *sctx, u64 dir) 6107 { 6108 u64 orig_gen, new_gen; 6109 int ret; 6110 6111 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, 6112 NULL, NULL); 6113 if (ret) 6114 return ret; 6115 6116 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, 6117 NULL, NULL, NULL); 6118 if (ret) 6119 return ret; 6120 6121 return (orig_gen != new_gen) ? 1 : 0; 6122 } 6123 6124 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, 6125 struct btrfs_key *key) 6126 { 6127 struct btrfs_inode_extref *extref; 6128 struct extent_buffer *leaf; 6129 u64 dirid = 0, last_dirid = 0; 6130 unsigned long ptr; 6131 u32 item_size; 6132 u32 cur_offset = 0; 6133 int ref_name_len; 6134 int ret = 0; 6135 6136 /* Easy case, just check this one dirid */ 6137 if (key->type == BTRFS_INODE_REF_KEY) { 6138 dirid = key->offset; 6139 6140 ret = dir_changed(sctx, dirid); 6141 goto out; 6142 } 6143 6144 leaf = path->nodes[0]; 6145 item_size = btrfs_item_size_nr(leaf, path->slots[0]); 6146 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 6147 while (cur_offset < item_size) { 6148 extref = (struct btrfs_inode_extref *)(ptr + 6149 cur_offset); 6150 dirid = btrfs_inode_extref_parent(leaf, extref); 6151 ref_name_len = btrfs_inode_extref_name_len(leaf, extref); 6152 cur_offset += ref_name_len + sizeof(*extref); 6153 if (dirid == last_dirid) 6154 continue; 6155 ret = dir_changed(sctx, dirid); 6156 if (ret) 6157 break; 6158 last_dirid = dirid; 6159 } 6160 out: 6161 return ret; 6162 } 6163 6164 /* 6165 * Updates compare related fields in sctx and simply forwards to the actual 6166 * changed_xxx functions. 6167 */ 6168 static int changed_cb(struct btrfs_path *left_path, 6169 struct btrfs_path *right_path, 6170 struct btrfs_key *key, 6171 enum btrfs_compare_tree_result result, 6172 void *ctx) 6173 { 6174 int ret = 0; 6175 struct send_ctx *sctx = ctx; 6176 6177 if (result == BTRFS_COMPARE_TREE_SAME) { 6178 if (key->type == BTRFS_INODE_REF_KEY || 6179 key->type == BTRFS_INODE_EXTREF_KEY) { 6180 ret = compare_refs(sctx, left_path, key); 6181 if (!ret) 6182 return 0; 6183 if (ret < 0) 6184 return ret; 6185 } else if (key->type == BTRFS_EXTENT_DATA_KEY) { 6186 return maybe_send_hole(sctx, left_path, key); 6187 } else { 6188 return 0; 6189 } 6190 result = BTRFS_COMPARE_TREE_CHANGED; 6191 ret = 0; 6192 } 6193 6194 sctx->left_path = left_path; 6195 sctx->right_path = right_path; 6196 sctx->cmp_key = key; 6197 6198 ret = finish_inode_if_needed(sctx, 0); 6199 if (ret < 0) 6200 goto out; 6201 6202 /* Ignore non-FS objects */ 6203 if (key->objectid == BTRFS_FREE_INO_OBJECTID || 6204 key->objectid == BTRFS_FREE_SPACE_OBJECTID) 6205 goto out; 6206 6207 if (key->type == BTRFS_INODE_ITEM_KEY) 6208 ret = changed_inode(sctx, result); 6209 else if (key->type == BTRFS_INODE_REF_KEY || 6210 key->type == BTRFS_INODE_EXTREF_KEY) 6211 ret = changed_ref(sctx, result); 6212 else if (key->type == BTRFS_XATTR_ITEM_KEY) 6213 ret = changed_xattr(sctx, result); 6214 else if (key->type == BTRFS_EXTENT_DATA_KEY) 6215 ret = changed_extent(sctx, result); 6216 6217 out: 6218 return ret; 6219 } 6220 6221 static int full_send_tree(struct send_ctx *sctx) 6222 { 6223 int ret; 6224 struct btrfs_root *send_root = sctx->send_root; 6225 struct btrfs_key key; 6226 struct btrfs_key found_key; 6227 struct btrfs_path *path; 6228 struct extent_buffer *eb; 6229 int slot; 6230 6231 path = alloc_path_for_send(); 6232 if (!path) 6233 return -ENOMEM; 6234 6235 key.objectid = BTRFS_FIRST_FREE_OBJECTID; 6236 key.type = BTRFS_INODE_ITEM_KEY; 6237 key.offset = 0; 6238 6239 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); 6240 if (ret < 0) 6241 goto out; 6242 if (ret) 6243 goto out_finish; 6244 6245 while (1) { 6246 eb = path->nodes[0]; 6247 slot = path->slots[0]; 6248 btrfs_item_key_to_cpu(eb, &found_key, slot); 6249 6250 ret = changed_cb(path, NULL, &found_key, 6251 BTRFS_COMPARE_TREE_NEW, sctx); 6252 if (ret < 0) 6253 goto out; 6254 6255 key.objectid = found_key.objectid; 6256 key.type = found_key.type; 6257 key.offset = found_key.offset + 1; 6258 6259 ret = btrfs_next_item(send_root, path); 6260 if (ret < 0) 6261 goto out; 6262 if (ret) { 6263 ret = 0; 6264 break; 6265 } 6266 } 6267 6268 out_finish: 6269 ret = finish_inode_if_needed(sctx, 1); 6270 6271 out: 6272 btrfs_free_path(path); 6273 return ret; 6274 } 6275 6276 static int send_subvol(struct send_ctx *sctx) 6277 { 6278 int ret; 6279 6280 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { 6281 ret = send_header(sctx); 6282 if (ret < 0) 6283 goto out; 6284 } 6285 6286 ret = send_subvol_begin(sctx); 6287 if (ret < 0) 6288 goto out; 6289 6290 if (sctx->parent_root) { 6291 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, 6292 changed_cb, sctx); 6293 if (ret < 0) 6294 goto out; 6295 ret = finish_inode_if_needed(sctx, 1); 6296 if (ret < 0) 6297 goto out; 6298 } else { 6299 ret = full_send_tree(sctx); 6300 if (ret < 0) 6301 goto out; 6302 } 6303 6304 out: 6305 free_recorded_refs(sctx); 6306 return ret; 6307 } 6308 6309 /* 6310 * If orphan cleanup did remove any orphans from a root, it means the tree 6311 * was modified and therefore the commit root is not the same as the current 6312 * root anymore. This is a problem, because send uses the commit root and 6313 * therefore can see inode items that don't exist in the current root anymore, 6314 * and for example make calls to btrfs_iget, which will do tree lookups based 6315 * on the current root and not on the commit root. Those lookups will fail, 6316 * returning a -ESTALE error, and making send fail with that error. So make 6317 * sure a send does not see any orphans we have just removed, and that it will 6318 * see the same inodes regardless of whether a transaction commit happened 6319 * before it started (meaning that the commit root will be the same as the 6320 * current root) or not. 6321 */ 6322 static int ensure_commit_roots_uptodate(struct send_ctx *sctx) 6323 { 6324 int i; 6325 struct btrfs_trans_handle *trans = NULL; 6326 6327 again: 6328 if (sctx->parent_root && 6329 sctx->parent_root->node != sctx->parent_root->commit_root) 6330 goto commit_trans; 6331 6332 for (i = 0; i < sctx->clone_roots_cnt; i++) 6333 if (sctx->clone_roots[i].root->node != 6334 sctx->clone_roots[i].root->commit_root) 6335 goto commit_trans; 6336 6337 if (trans) 6338 return btrfs_end_transaction(trans); 6339 6340 return 0; 6341 6342 commit_trans: 6343 /* Use any root, all fs roots will get their commit roots updated. */ 6344 if (!trans) { 6345 trans = btrfs_join_transaction(sctx->send_root); 6346 if (IS_ERR(trans)) 6347 return PTR_ERR(trans); 6348 goto again; 6349 } 6350 6351 return btrfs_commit_transaction(trans); 6352 } 6353 6354 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) 6355 { 6356 spin_lock(&root->root_item_lock); 6357 root->send_in_progress--; 6358 /* 6359 * Not much left to do, we don't know why it's unbalanced and 6360 * can't blindly reset it to 0. 6361 */ 6362 if (root->send_in_progress < 0) 6363 btrfs_err(root->fs_info, 6364 "send_in_progres unbalanced %d root %llu", 6365 root->send_in_progress, root->root_key.objectid); 6366 spin_unlock(&root->root_item_lock); 6367 } 6368 6369 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) 6370 { 6371 int ret = 0; 6372 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root; 6373 struct btrfs_fs_info *fs_info = send_root->fs_info; 6374 struct btrfs_root *clone_root; 6375 struct btrfs_key key; 6376 struct send_ctx *sctx = NULL; 6377 u32 i; 6378 u64 *clone_sources_tmp = NULL; 6379 int clone_sources_to_rollback = 0; 6380 unsigned alloc_size; 6381 int sort_clone_roots = 0; 6382 int index; 6383 6384 if (!capable(CAP_SYS_ADMIN)) 6385 return -EPERM; 6386 6387 /* 6388 * The subvolume must remain read-only during send, protect against 6389 * making it RW. This also protects against deletion. 6390 */ 6391 spin_lock(&send_root->root_item_lock); 6392 send_root->send_in_progress++; 6393 spin_unlock(&send_root->root_item_lock); 6394 6395 /* 6396 * This is done when we lookup the root, it should already be complete 6397 * by the time we get here. 6398 */ 6399 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE); 6400 6401 /* 6402 * Userspace tools do the checks and warn the user if it's 6403 * not RO. 6404 */ 6405 if (!btrfs_root_readonly(send_root)) { 6406 ret = -EPERM; 6407 goto out; 6408 } 6409 6410 /* 6411 * Check that we don't overflow at later allocations, we request 6412 * clone_sources_count + 1 items, and compare to unsigned long inside 6413 * access_ok. 6414 */ 6415 if (arg->clone_sources_count > 6416 ULONG_MAX / sizeof(struct clone_root) - 1) { 6417 ret = -EINVAL; 6418 goto out; 6419 } 6420 6421 if (!access_ok(VERIFY_READ, arg->clone_sources, 6422 sizeof(*arg->clone_sources) * 6423 arg->clone_sources_count)) { 6424 ret = -EFAULT; 6425 goto out; 6426 } 6427 6428 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { 6429 ret = -EINVAL; 6430 goto out; 6431 } 6432 6433 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL); 6434 if (!sctx) { 6435 ret = -ENOMEM; 6436 goto out; 6437 } 6438 6439 INIT_LIST_HEAD(&sctx->new_refs); 6440 INIT_LIST_HEAD(&sctx->deleted_refs); 6441 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL); 6442 INIT_LIST_HEAD(&sctx->name_cache_list); 6443 6444 sctx->flags = arg->flags; 6445 6446 sctx->send_filp = fget(arg->send_fd); 6447 if (!sctx->send_filp) { 6448 ret = -EBADF; 6449 goto out; 6450 } 6451 6452 sctx->send_root = send_root; 6453 /* 6454 * Unlikely but possible, if the subvolume is marked for deletion but 6455 * is slow to remove the directory entry, send can still be started 6456 */ 6457 if (btrfs_root_dead(sctx->send_root)) { 6458 ret = -EPERM; 6459 goto out; 6460 } 6461 6462 sctx->clone_roots_cnt = arg->clone_sources_count; 6463 6464 sctx->send_max_size = BTRFS_SEND_BUF_SIZE; 6465 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); 6466 if (!sctx->send_buf) { 6467 ret = -ENOMEM; 6468 goto out; 6469 } 6470 6471 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL); 6472 if (!sctx->read_buf) { 6473 ret = -ENOMEM; 6474 goto out; 6475 } 6476 6477 sctx->pending_dir_moves = RB_ROOT; 6478 sctx->waiting_dir_moves = RB_ROOT; 6479 sctx->orphan_dirs = RB_ROOT; 6480 6481 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1); 6482 6483 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL); 6484 if (!sctx->clone_roots) { 6485 ret = -ENOMEM; 6486 goto out; 6487 } 6488 6489 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources); 6490 6491 if (arg->clone_sources_count) { 6492 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL); 6493 if (!clone_sources_tmp) { 6494 ret = -ENOMEM; 6495 goto out; 6496 } 6497 6498 ret = copy_from_user(clone_sources_tmp, arg->clone_sources, 6499 alloc_size); 6500 if (ret) { 6501 ret = -EFAULT; 6502 goto out; 6503 } 6504 6505 for (i = 0; i < arg->clone_sources_count; i++) { 6506 key.objectid = clone_sources_tmp[i]; 6507 key.type = BTRFS_ROOT_ITEM_KEY; 6508 key.offset = (u64)-1; 6509 6510 index = srcu_read_lock(&fs_info->subvol_srcu); 6511 6512 clone_root = btrfs_read_fs_root_no_name(fs_info, &key); 6513 if (IS_ERR(clone_root)) { 6514 srcu_read_unlock(&fs_info->subvol_srcu, index); 6515 ret = PTR_ERR(clone_root); 6516 goto out; 6517 } 6518 spin_lock(&clone_root->root_item_lock); 6519 if (!btrfs_root_readonly(clone_root) || 6520 btrfs_root_dead(clone_root)) { 6521 spin_unlock(&clone_root->root_item_lock); 6522 srcu_read_unlock(&fs_info->subvol_srcu, index); 6523 ret = -EPERM; 6524 goto out; 6525 } 6526 clone_root->send_in_progress++; 6527 spin_unlock(&clone_root->root_item_lock); 6528 srcu_read_unlock(&fs_info->subvol_srcu, index); 6529 6530 sctx->clone_roots[i].root = clone_root; 6531 clone_sources_to_rollback = i + 1; 6532 } 6533 kvfree(clone_sources_tmp); 6534 clone_sources_tmp = NULL; 6535 } 6536 6537 if (arg->parent_root) { 6538 key.objectid = arg->parent_root; 6539 key.type = BTRFS_ROOT_ITEM_KEY; 6540 key.offset = (u64)-1; 6541 6542 index = srcu_read_lock(&fs_info->subvol_srcu); 6543 6544 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key); 6545 if (IS_ERR(sctx->parent_root)) { 6546 srcu_read_unlock(&fs_info->subvol_srcu, index); 6547 ret = PTR_ERR(sctx->parent_root); 6548 goto out; 6549 } 6550 6551 spin_lock(&sctx->parent_root->root_item_lock); 6552 sctx->parent_root->send_in_progress++; 6553 if (!btrfs_root_readonly(sctx->parent_root) || 6554 btrfs_root_dead(sctx->parent_root)) { 6555 spin_unlock(&sctx->parent_root->root_item_lock); 6556 srcu_read_unlock(&fs_info->subvol_srcu, index); 6557 ret = -EPERM; 6558 goto out; 6559 } 6560 spin_unlock(&sctx->parent_root->root_item_lock); 6561 6562 srcu_read_unlock(&fs_info->subvol_srcu, index); 6563 } 6564 6565 /* 6566 * Clones from send_root are allowed, but only if the clone source 6567 * is behind the current send position. This is checked while searching 6568 * for possible clone sources. 6569 */ 6570 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root; 6571 6572 /* We do a bsearch later */ 6573 sort(sctx->clone_roots, sctx->clone_roots_cnt, 6574 sizeof(*sctx->clone_roots), __clone_root_cmp_sort, 6575 NULL); 6576 sort_clone_roots = 1; 6577 6578 ret = ensure_commit_roots_uptodate(sctx); 6579 if (ret) 6580 goto out; 6581 6582 current->journal_info = BTRFS_SEND_TRANS_STUB; 6583 ret = send_subvol(sctx); 6584 current->journal_info = NULL; 6585 if (ret < 0) 6586 goto out; 6587 6588 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { 6589 ret = begin_cmd(sctx, BTRFS_SEND_C_END); 6590 if (ret < 0) 6591 goto out; 6592 ret = send_cmd(sctx); 6593 if (ret < 0) 6594 goto out; 6595 } 6596 6597 out: 6598 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); 6599 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { 6600 struct rb_node *n; 6601 struct pending_dir_move *pm; 6602 6603 n = rb_first(&sctx->pending_dir_moves); 6604 pm = rb_entry(n, struct pending_dir_move, node); 6605 while (!list_empty(&pm->list)) { 6606 struct pending_dir_move *pm2; 6607 6608 pm2 = list_first_entry(&pm->list, 6609 struct pending_dir_move, list); 6610 free_pending_move(sctx, pm2); 6611 } 6612 free_pending_move(sctx, pm); 6613 } 6614 6615 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); 6616 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { 6617 struct rb_node *n; 6618 struct waiting_dir_move *dm; 6619 6620 n = rb_first(&sctx->waiting_dir_moves); 6621 dm = rb_entry(n, struct waiting_dir_move, node); 6622 rb_erase(&dm->node, &sctx->waiting_dir_moves); 6623 kfree(dm); 6624 } 6625 6626 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); 6627 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { 6628 struct rb_node *n; 6629 struct orphan_dir_info *odi; 6630 6631 n = rb_first(&sctx->orphan_dirs); 6632 odi = rb_entry(n, struct orphan_dir_info, node); 6633 free_orphan_dir_info(sctx, odi); 6634 } 6635 6636 if (sort_clone_roots) { 6637 for (i = 0; i < sctx->clone_roots_cnt; i++) 6638 btrfs_root_dec_send_in_progress( 6639 sctx->clone_roots[i].root); 6640 } else { 6641 for (i = 0; sctx && i < clone_sources_to_rollback; i++) 6642 btrfs_root_dec_send_in_progress( 6643 sctx->clone_roots[i].root); 6644 6645 btrfs_root_dec_send_in_progress(send_root); 6646 } 6647 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) 6648 btrfs_root_dec_send_in_progress(sctx->parent_root); 6649 6650 kvfree(clone_sources_tmp); 6651 6652 if (sctx) { 6653 if (sctx->send_filp) 6654 fput(sctx->send_filp); 6655 6656 kvfree(sctx->clone_roots); 6657 kvfree(sctx->send_buf); 6658 kvfree(sctx->read_buf); 6659 6660 name_cache_free(sctx); 6661 6662 kfree(sctx); 6663 } 6664 6665 return ret; 6666 } 6667