1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2012 Alexander Block. All rights reserved.
6 #include <linux/bsearch.h>
8 #include <linux/file.h>
9 #include <linux/sort.h>
10 #include <linux/mount.h>
11 #include <linux/xattr.h>
12 #include <linux/posix_acl_xattr.h>
13 #include <linux/radix-tree.h>
14 #include <linux/vmalloc.h>
15 #include <linux/string.h>
16 #include <linux/compat.h>
17 #include <linux/crc32c.h>
23 #include "btrfs_inode.h"
24 #include "transaction.h"
25 #include "compression.h"
28 * A fs_path is a helper to dynamically build path names with unknown size.
29 * It reallocates the internal buffer on demand.
30 * It allows fast adding of path elements on the right side (normal path) and
31 * fast adding to the left side (reversed path). A reversed path can also be
32 * unreversed if needed.
41 unsigned short buf_len:15;
42 unsigned short reversed:1;
46 * Average path length does not exceed 200 bytes, we'll have
47 * better packing in the slab and higher chance to satisfy
48 * a allocation later during send.
53 #define FS_PATH_INLINE_SIZE \
54 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
57 /* reused for each extent */
59 struct btrfs_root *root;
66 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
67 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
70 struct file *send_filp;
76 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
77 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
79 struct btrfs_root *send_root;
80 struct btrfs_root *parent_root;
81 struct clone_root *clone_roots;
84 /* current state of the compare_tree call */
85 struct btrfs_path *left_path;
86 struct btrfs_path *right_path;
87 struct btrfs_key *cmp_key;
90 * infos of the currently processed inode. In case of deleted inodes,
91 * these are the values from the deleted inode.
96 int cur_inode_new_gen;
97 int cur_inode_deleted;
101 u64 cur_inode_last_extent;
102 u64 cur_inode_next_write_offset;
106 struct list_head new_refs;
107 struct list_head deleted_refs;
109 struct radix_tree_root name_cache;
110 struct list_head name_cache_list;
113 struct file_ra_state ra;
118 * We process inodes by their increasing order, so if before an
119 * incremental send we reverse the parent/child relationship of
120 * directories such that a directory with a lower inode number was
121 * the parent of a directory with a higher inode number, and the one
122 * becoming the new parent got renamed too, we can't rename/move the
123 * directory with lower inode number when we finish processing it - we
124 * must process the directory with higher inode number first, then
125 * rename/move it and then rename/move the directory with lower inode
126 * number. Example follows.
128 * Tree state when the first send was performed:
140 * Tree state when the second (incremental) send is performed:
149 * The sequence of steps that lead to the second state was:
151 * mv /a/b/c/d /a/b/c2/d2
152 * mv /a/b/c /a/b/c2/d2/cc
154 * "c" has lower inode number, but we can't move it (2nd mv operation)
155 * before we move "d", which has higher inode number.
157 * So we just memorize which move/rename operations must be performed
158 * later when their respective parent is processed and moved/renamed.
161 /* Indexed by parent directory inode number. */
162 struct rb_root pending_dir_moves;
165 * Reverse index, indexed by the inode number of a directory that
166 * is waiting for the move/rename of its immediate parent before its
167 * own move/rename can be performed.
169 struct rb_root waiting_dir_moves;
172 * A directory that is going to be rm'ed might have a child directory
173 * which is in the pending directory moves index above. In this case,
174 * the directory can only be removed after the move/rename of its child
175 * is performed. Example:
195 * Sequence of steps that lead to the send snapshot:
196 * rm -f /a/b/c/foo.txt
198 * mv /a/b/c/x /a/b/YY
201 * When the child is processed, its move/rename is delayed until its
202 * parent is processed (as explained above), but all other operations
203 * like update utimes, chown, chgrp, etc, are performed and the paths
204 * that it uses for those operations must use the orphanized name of
205 * its parent (the directory we're going to rm later), so we need to
206 * memorize that name.
208 * Indexed by the inode number of the directory to be deleted.
210 struct rb_root orphan_dirs;
213 struct pending_dir_move {
215 struct list_head list;
219 struct list_head update_refs;
222 struct waiting_dir_move {
226 * There might be some directory that could not be removed because it
227 * was waiting for this directory inode to be moved first. Therefore
228 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
234 struct orphan_dir_info {
240 struct name_cache_entry {
241 struct list_head list;
243 * radix_tree has only 32bit entries but we need to handle 64bit inums.
244 * We use the lower 32bit of the 64bit inum to store it in the tree. If
245 * more then one inum would fall into the same entry, we use radix_list
246 * to store the additional entries. radix_list is also used to store
247 * entries where two entries have the same inum but different
250 struct list_head radix_list;
256 int need_later_update;
262 static void inconsistent_snapshot_error(struct send_ctx *sctx,
263 enum btrfs_compare_tree_result result,
266 const char *result_string;
269 case BTRFS_COMPARE_TREE_NEW:
270 result_string = "new";
272 case BTRFS_COMPARE_TREE_DELETED:
273 result_string = "deleted";
275 case BTRFS_COMPARE_TREE_CHANGED:
276 result_string = "updated";
278 case BTRFS_COMPARE_TREE_SAME:
280 result_string = "unchanged";
284 result_string = "unexpected";
287 btrfs_err(sctx->send_root->fs_info,
288 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
289 result_string, what, sctx->cmp_key->objectid,
290 sctx->send_root->root_key.objectid,
292 sctx->parent_root->root_key.objectid : 0));
295 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
297 static struct waiting_dir_move *
298 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
300 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
302 static int need_send_hole(struct send_ctx *sctx)
304 return (sctx->parent_root && !sctx->cur_inode_new &&
305 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
306 S_ISREG(sctx->cur_inode_mode));
309 static void fs_path_reset(struct fs_path *p)
312 p->start = p->buf + p->buf_len - 1;
322 static struct fs_path *fs_path_alloc(void)
326 p = kmalloc(sizeof(*p), GFP_KERNEL);
330 p->buf = p->inline_buf;
331 p->buf_len = FS_PATH_INLINE_SIZE;
336 static struct fs_path *fs_path_alloc_reversed(void)
348 static void fs_path_free(struct fs_path *p)
352 if (p->buf != p->inline_buf)
357 static int fs_path_len(struct fs_path *p)
359 return p->end - p->start;
362 static int fs_path_ensure_buf(struct fs_path *p, int len)
370 if (p->buf_len >= len)
373 if (len > PATH_MAX) {
378 path_len = p->end - p->start;
379 old_buf_len = p->buf_len;
382 * First time the inline_buf does not suffice
384 if (p->buf == p->inline_buf) {
385 tmp_buf = kmalloc(len, GFP_KERNEL);
387 memcpy(tmp_buf, p->buf, old_buf_len);
389 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
395 * The real size of the buffer is bigger, this will let the fast path
396 * happen most of the time
398 p->buf_len = ksize(p->buf);
401 tmp_buf = p->buf + old_buf_len - path_len - 1;
402 p->end = p->buf + p->buf_len - 1;
403 p->start = p->end - path_len;
404 memmove(p->start, tmp_buf, path_len + 1);
407 p->end = p->start + path_len;
412 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
418 new_len = p->end - p->start + name_len;
419 if (p->start != p->end)
421 ret = fs_path_ensure_buf(p, new_len);
426 if (p->start != p->end)
428 p->start -= name_len;
429 *prepared = p->start;
431 if (p->start != p->end)
442 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
447 ret = fs_path_prepare_for_add(p, name_len, &prepared);
450 memcpy(prepared, name, name_len);
456 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
461 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
464 memcpy(prepared, p2->start, p2->end - p2->start);
470 static int fs_path_add_from_extent_buffer(struct fs_path *p,
471 struct extent_buffer *eb,
472 unsigned long off, int len)
477 ret = fs_path_prepare_for_add(p, len, &prepared);
481 read_extent_buffer(eb, prepared, off, len);
487 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
491 p->reversed = from->reversed;
494 ret = fs_path_add_path(p, from);
500 static void fs_path_unreverse(struct fs_path *p)
509 len = p->end - p->start;
511 p->end = p->start + len;
512 memmove(p->start, tmp, len + 1);
516 static struct btrfs_path *alloc_path_for_send(void)
518 struct btrfs_path *path;
520 path = btrfs_alloc_path();
523 path->search_commit_root = 1;
524 path->skip_locking = 1;
525 path->need_commit_sem = 1;
529 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
535 ret = kernel_write(filp, buf + pos, len - pos, off);
536 /* TODO handle that correctly */
537 /*if (ret == -ERESTARTSYS) {
551 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
553 struct btrfs_tlv_header *hdr;
554 int total_len = sizeof(*hdr) + len;
555 int left = sctx->send_max_size - sctx->send_size;
557 if (unlikely(left < total_len))
560 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
561 hdr->tlv_type = cpu_to_le16(attr);
562 hdr->tlv_len = cpu_to_le16(len);
563 memcpy(hdr + 1, data, len);
564 sctx->send_size += total_len;
569 #define TLV_PUT_DEFINE_INT(bits) \
570 static int tlv_put_u##bits(struct send_ctx *sctx, \
571 u##bits attr, u##bits value) \
573 __le##bits __tmp = cpu_to_le##bits(value); \
574 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
577 TLV_PUT_DEFINE_INT(64)
579 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
580 const char *str, int len)
584 return tlv_put(sctx, attr, str, len);
587 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
590 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
593 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
594 struct extent_buffer *eb,
595 struct btrfs_timespec *ts)
597 struct btrfs_timespec bts;
598 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
599 return tlv_put(sctx, attr, &bts, sizeof(bts));
603 #define TLV_PUT(sctx, attrtype, data, attrlen) \
605 ret = tlv_put(sctx, attrtype, data, attrlen); \
607 goto tlv_put_failure; \
610 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
612 ret = tlv_put_u##bits(sctx, attrtype, value); \
614 goto tlv_put_failure; \
617 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
618 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
619 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
620 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
621 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
623 ret = tlv_put_string(sctx, attrtype, str, len); \
625 goto tlv_put_failure; \
627 #define TLV_PUT_PATH(sctx, attrtype, p) \
629 ret = tlv_put_string(sctx, attrtype, p->start, \
630 p->end - p->start); \
632 goto tlv_put_failure; \
634 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
636 ret = tlv_put_uuid(sctx, attrtype, uuid); \
638 goto tlv_put_failure; \
640 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
642 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
644 goto tlv_put_failure; \
647 static int send_header(struct send_ctx *sctx)
649 struct btrfs_stream_header hdr;
651 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
652 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
654 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
659 * For each command/item we want to send to userspace, we call this function.
661 static int begin_cmd(struct send_ctx *sctx, int cmd)
663 struct btrfs_cmd_header *hdr;
665 if (WARN_ON(!sctx->send_buf))
668 BUG_ON(sctx->send_size);
670 sctx->send_size += sizeof(*hdr);
671 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
672 hdr->cmd = cpu_to_le16(cmd);
677 static int send_cmd(struct send_ctx *sctx)
680 struct btrfs_cmd_header *hdr;
683 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
684 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
687 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
688 hdr->crc = cpu_to_le32(crc);
690 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
693 sctx->total_send_size += sctx->send_size;
694 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
701 * Sends a move instruction to user space
703 static int send_rename(struct send_ctx *sctx,
704 struct fs_path *from, struct fs_path *to)
706 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
709 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
711 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
715 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
716 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
718 ret = send_cmd(sctx);
726 * Sends a link instruction to user space
728 static int send_link(struct send_ctx *sctx,
729 struct fs_path *path, struct fs_path *lnk)
731 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
734 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
736 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
740 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
741 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
743 ret = send_cmd(sctx);
751 * Sends an unlink instruction to user space
753 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
755 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
758 btrfs_debug(fs_info, "send_unlink %s", path->start);
760 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
764 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
766 ret = send_cmd(sctx);
774 * Sends a rmdir instruction to user space
776 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
778 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
781 btrfs_debug(fs_info, "send_rmdir %s", path->start);
783 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
787 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
789 ret = send_cmd(sctx);
797 * Helper function to retrieve some fields from an inode item.
799 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
800 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
804 struct btrfs_inode_item *ii;
805 struct btrfs_key key;
808 key.type = BTRFS_INODE_ITEM_KEY;
810 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
817 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
818 struct btrfs_inode_item);
820 *size = btrfs_inode_size(path->nodes[0], ii);
822 *gen = btrfs_inode_generation(path->nodes[0], ii);
824 *mode = btrfs_inode_mode(path->nodes[0], ii);
826 *uid = btrfs_inode_uid(path->nodes[0], ii);
828 *gid = btrfs_inode_gid(path->nodes[0], ii);
830 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
835 static int get_inode_info(struct btrfs_root *root,
836 u64 ino, u64 *size, u64 *gen,
837 u64 *mode, u64 *uid, u64 *gid,
840 struct btrfs_path *path;
843 path = alloc_path_for_send();
846 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
848 btrfs_free_path(path);
852 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
857 * Helper function to iterate the entries in ONE btrfs_inode_ref or
858 * btrfs_inode_extref.
859 * The iterate callback may return a non zero value to stop iteration. This can
860 * be a negative value for error codes or 1 to simply stop it.
862 * path must point to the INODE_REF or INODE_EXTREF when called.
864 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
865 struct btrfs_key *found_key, int resolve,
866 iterate_inode_ref_t iterate, void *ctx)
868 struct extent_buffer *eb = path->nodes[0];
869 struct btrfs_item *item;
870 struct btrfs_inode_ref *iref;
871 struct btrfs_inode_extref *extref;
872 struct btrfs_path *tmp_path;
876 int slot = path->slots[0];
883 unsigned long name_off;
884 unsigned long elem_size;
887 p = fs_path_alloc_reversed();
891 tmp_path = alloc_path_for_send();
898 if (found_key->type == BTRFS_INODE_REF_KEY) {
899 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
900 struct btrfs_inode_ref);
901 item = btrfs_item_nr(slot);
902 total = btrfs_item_size(eb, item);
903 elem_size = sizeof(*iref);
905 ptr = btrfs_item_ptr_offset(eb, slot);
906 total = btrfs_item_size_nr(eb, slot);
907 elem_size = sizeof(*extref);
910 while (cur < total) {
913 if (found_key->type == BTRFS_INODE_REF_KEY) {
914 iref = (struct btrfs_inode_ref *)(ptr + cur);
915 name_len = btrfs_inode_ref_name_len(eb, iref);
916 name_off = (unsigned long)(iref + 1);
917 index = btrfs_inode_ref_index(eb, iref);
918 dir = found_key->offset;
920 extref = (struct btrfs_inode_extref *)(ptr + cur);
921 name_len = btrfs_inode_extref_name_len(eb, extref);
922 name_off = (unsigned long)&extref->name;
923 index = btrfs_inode_extref_index(eb, extref);
924 dir = btrfs_inode_extref_parent(eb, extref);
928 start = btrfs_ref_to_path(root, tmp_path, name_len,
932 ret = PTR_ERR(start);
935 if (start < p->buf) {
936 /* overflow , try again with larger buffer */
937 ret = fs_path_ensure_buf(p,
938 p->buf_len + p->buf - start);
941 start = btrfs_ref_to_path(root, tmp_path,
946 ret = PTR_ERR(start);
949 BUG_ON(start < p->buf);
953 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
959 cur += elem_size + name_len;
960 ret = iterate(num, dir, index, p, ctx);
967 btrfs_free_path(tmp_path);
972 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
973 const char *name, int name_len,
974 const char *data, int data_len,
978 * Helper function to iterate the entries in ONE btrfs_dir_item.
979 * The iterate callback may return a non zero value to stop iteration. This can
980 * be a negative value for error codes or 1 to simply stop it.
982 * path must point to the dir item when called.
984 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
985 iterate_dir_item_t iterate, void *ctx)
988 struct extent_buffer *eb;
989 struct btrfs_item *item;
990 struct btrfs_dir_item *di;
991 struct btrfs_key di_key;
1004 * Start with a small buffer (1 page). If later we end up needing more
1005 * space, which can happen for xattrs on a fs with a leaf size greater
1006 * then the page size, attempt to increase the buffer. Typically xattr
1010 buf = kmalloc(buf_len, GFP_KERNEL);
1016 eb = path->nodes[0];
1017 slot = path->slots[0];
1018 item = btrfs_item_nr(slot);
1019 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1022 total = btrfs_item_size(eb, item);
1025 while (cur < total) {
1026 name_len = btrfs_dir_name_len(eb, di);
1027 data_len = btrfs_dir_data_len(eb, di);
1028 type = btrfs_dir_type(eb, di);
1029 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1031 if (type == BTRFS_FT_XATTR) {
1032 if (name_len > XATTR_NAME_MAX) {
1033 ret = -ENAMETOOLONG;
1036 if (name_len + data_len >
1037 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
1045 if (name_len + data_len > PATH_MAX) {
1046 ret = -ENAMETOOLONG;
1051 if (name_len + data_len > buf_len) {
1052 buf_len = name_len + data_len;
1053 if (is_vmalloc_addr(buf)) {
1057 char *tmp = krealloc(buf, buf_len,
1058 GFP_KERNEL | __GFP_NOWARN);
1065 buf = kvmalloc(buf_len, GFP_KERNEL);
1073 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1074 name_len + data_len);
1076 len = sizeof(*di) + name_len + data_len;
1077 di = (struct btrfs_dir_item *)((char *)di + len);
1080 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1081 data_len, type, ctx);
1097 static int __copy_first_ref(int num, u64 dir, int index,
1098 struct fs_path *p, void *ctx)
1101 struct fs_path *pt = ctx;
1103 ret = fs_path_copy(pt, p);
1107 /* we want the first only */
1112 * Retrieve the first path of an inode. If an inode has more then one
1113 * ref/hardlink, this is ignored.
1115 static int get_inode_path(struct btrfs_root *root,
1116 u64 ino, struct fs_path *path)
1119 struct btrfs_key key, found_key;
1120 struct btrfs_path *p;
1122 p = alloc_path_for_send();
1126 fs_path_reset(path);
1129 key.type = BTRFS_INODE_REF_KEY;
1132 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1139 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1140 if (found_key.objectid != ino ||
1141 (found_key.type != BTRFS_INODE_REF_KEY &&
1142 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1147 ret = iterate_inode_ref(root, p, &found_key, 1,
1148 __copy_first_ref, path);
1158 struct backref_ctx {
1159 struct send_ctx *sctx;
1161 struct btrfs_path *path;
1162 /* number of total found references */
1166 * used for clones found in send_root. clones found behind cur_objectid
1167 * and cur_offset are not considered as allowed clones.
1172 /* may be truncated in case it's the last extent in a file */
1175 /* data offset in the file extent item */
1178 /* Just to check for bugs in backref resolving */
1182 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1184 u64 root = (u64)(uintptr_t)key;
1185 struct clone_root *cr = (struct clone_root *)elt;
1187 if (root < cr->root->objectid)
1189 if (root > cr->root->objectid)
1194 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1196 struct clone_root *cr1 = (struct clone_root *)e1;
1197 struct clone_root *cr2 = (struct clone_root *)e2;
1199 if (cr1->root->objectid < cr2->root->objectid)
1201 if (cr1->root->objectid > cr2->root->objectid)
1207 * Called for every backref that is found for the current extent.
1208 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1210 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1212 struct backref_ctx *bctx = ctx_;
1213 struct clone_root *found;
1217 /* First check if the root is in the list of accepted clone sources */
1218 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1219 bctx->sctx->clone_roots_cnt,
1220 sizeof(struct clone_root),
1221 __clone_root_cmp_bsearch);
1225 if (found->root == bctx->sctx->send_root &&
1226 ino == bctx->cur_objectid &&
1227 offset == bctx->cur_offset) {
1228 bctx->found_itself = 1;
1232 * There are inodes that have extents that lie behind its i_size. Don't
1233 * accept clones from these extents.
1235 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1237 btrfs_release_path(bctx->path);
1241 if (offset + bctx->data_offset + bctx->extent_len > i_size)
1245 * Make sure we don't consider clones from send_root that are
1246 * behind the current inode/offset.
1248 if (found->root == bctx->sctx->send_root) {
1250 * TODO for the moment we don't accept clones from the inode
1251 * that is currently send. We may change this when
1252 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1255 if (ino >= bctx->cur_objectid)
1260 found->found_refs++;
1261 if (ino < found->ino) {
1263 found->offset = offset;
1264 } else if (found->ino == ino) {
1266 * same extent found more then once in the same file.
1268 if (found->offset > offset + bctx->extent_len)
1269 found->offset = offset;
1276 * Given an inode, offset and extent item, it finds a good clone for a clone
1277 * instruction. Returns -ENOENT when none could be found. The function makes
1278 * sure that the returned clone is usable at the point where sending is at the
1279 * moment. This means, that no clones are accepted which lie behind the current
1282 * path must point to the extent item when called.
1284 static int find_extent_clone(struct send_ctx *sctx,
1285 struct btrfs_path *path,
1286 u64 ino, u64 data_offset,
1288 struct clone_root **found)
1290 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1296 u64 extent_item_pos;
1298 struct btrfs_file_extent_item *fi;
1299 struct extent_buffer *eb = path->nodes[0];
1300 struct backref_ctx *backref_ctx = NULL;
1301 struct clone_root *cur_clone_root;
1302 struct btrfs_key found_key;
1303 struct btrfs_path *tmp_path;
1307 tmp_path = alloc_path_for_send();
1311 /* We only use this path under the commit sem */
1312 tmp_path->need_commit_sem = 0;
1314 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1320 backref_ctx->path = tmp_path;
1322 if (data_offset >= ino_size) {
1324 * There may be extents that lie behind the file's size.
1325 * I at least had this in combination with snapshotting while
1326 * writing large files.
1332 fi = btrfs_item_ptr(eb, path->slots[0],
1333 struct btrfs_file_extent_item);
1334 extent_type = btrfs_file_extent_type(eb, fi);
1335 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1339 compressed = btrfs_file_extent_compression(eb, fi);
1341 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1342 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1343 if (disk_byte == 0) {
1347 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1349 down_read(&fs_info->commit_root_sem);
1350 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1351 &found_key, &flags);
1352 up_read(&fs_info->commit_root_sem);
1353 btrfs_release_path(tmp_path);
1357 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1363 * Setup the clone roots.
1365 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1366 cur_clone_root = sctx->clone_roots + i;
1367 cur_clone_root->ino = (u64)-1;
1368 cur_clone_root->offset = 0;
1369 cur_clone_root->found_refs = 0;
1372 backref_ctx->sctx = sctx;
1373 backref_ctx->found = 0;
1374 backref_ctx->cur_objectid = ino;
1375 backref_ctx->cur_offset = data_offset;
1376 backref_ctx->found_itself = 0;
1377 backref_ctx->extent_len = num_bytes;
1379 * For non-compressed extents iterate_extent_inodes() gives us extent
1380 * offsets that already take into account the data offset, but not for
1381 * compressed extents, since the offset is logical and not relative to
1382 * the physical extent locations. We must take this into account to
1383 * avoid sending clone offsets that go beyond the source file's size,
1384 * which would result in the clone ioctl failing with -EINVAL on the
1387 if (compressed == BTRFS_COMPRESS_NONE)
1388 backref_ctx->data_offset = 0;
1390 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1393 * The last extent of a file may be too large due to page alignment.
1394 * We need to adjust extent_len in this case so that the checks in
1395 * __iterate_backrefs work.
1397 if (data_offset + num_bytes >= ino_size)
1398 backref_ctx->extent_len = ino_size - data_offset;
1401 * Now collect all backrefs.
1403 if (compressed == BTRFS_COMPRESS_NONE)
1404 extent_item_pos = logical - found_key.objectid;
1406 extent_item_pos = 0;
1407 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1408 extent_item_pos, 1, __iterate_backrefs,
1409 backref_ctx, false);
1414 if (!backref_ctx->found_itself) {
1415 /* found a bug in backref code? */
1418 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1419 ino, data_offset, disk_byte, found_key.objectid);
1423 btrfs_debug(fs_info,
1424 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1425 data_offset, ino, num_bytes, logical);
1427 if (!backref_ctx->found)
1428 btrfs_debug(fs_info, "no clones found");
1430 cur_clone_root = NULL;
1431 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1432 if (sctx->clone_roots[i].found_refs) {
1433 if (!cur_clone_root)
1434 cur_clone_root = sctx->clone_roots + i;
1435 else if (sctx->clone_roots[i].root == sctx->send_root)
1436 /* prefer clones from send_root over others */
1437 cur_clone_root = sctx->clone_roots + i;
1442 if (cur_clone_root) {
1443 *found = cur_clone_root;
1450 btrfs_free_path(tmp_path);
1455 static int read_symlink(struct btrfs_root *root,
1457 struct fs_path *dest)
1460 struct btrfs_path *path;
1461 struct btrfs_key key;
1462 struct btrfs_file_extent_item *ei;
1468 path = alloc_path_for_send();
1473 key.type = BTRFS_EXTENT_DATA_KEY;
1475 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1480 * An empty symlink inode. Can happen in rare error paths when
1481 * creating a symlink (transaction committed before the inode
1482 * eviction handler removed the symlink inode items and a crash
1483 * happened in between or the subvol was snapshoted in between).
1484 * Print an informative message to dmesg/syslog so that the user
1485 * can delete the symlink.
1487 btrfs_err(root->fs_info,
1488 "Found empty symlink inode %llu at root %llu",
1489 ino, root->root_key.objectid);
1494 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1495 struct btrfs_file_extent_item);
1496 type = btrfs_file_extent_type(path->nodes[0], ei);
1497 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1498 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1499 BUG_ON(compression);
1501 off = btrfs_file_extent_inline_start(ei);
1502 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1504 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1507 btrfs_free_path(path);
1512 * Helper function to generate a file name that is unique in the root of
1513 * send_root and parent_root. This is used to generate names for orphan inodes.
1515 static int gen_unique_name(struct send_ctx *sctx,
1517 struct fs_path *dest)
1520 struct btrfs_path *path;
1521 struct btrfs_dir_item *di;
1526 path = alloc_path_for_send();
1531 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1533 ASSERT(len < sizeof(tmp));
1535 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1536 path, BTRFS_FIRST_FREE_OBJECTID,
1537 tmp, strlen(tmp), 0);
1538 btrfs_release_path(path);
1544 /* not unique, try again */
1549 if (!sctx->parent_root) {
1555 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1556 path, BTRFS_FIRST_FREE_OBJECTID,
1557 tmp, strlen(tmp), 0);
1558 btrfs_release_path(path);
1564 /* not unique, try again */
1572 ret = fs_path_add(dest, tmp, strlen(tmp));
1575 btrfs_free_path(path);
1580 inode_state_no_change,
1581 inode_state_will_create,
1582 inode_state_did_create,
1583 inode_state_will_delete,
1584 inode_state_did_delete,
1587 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1595 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1597 if (ret < 0 && ret != -ENOENT)
1601 if (!sctx->parent_root) {
1602 right_ret = -ENOENT;
1604 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1605 NULL, NULL, NULL, NULL);
1606 if (ret < 0 && ret != -ENOENT)
1611 if (!left_ret && !right_ret) {
1612 if (left_gen == gen && right_gen == gen) {
1613 ret = inode_state_no_change;
1614 } else if (left_gen == gen) {
1615 if (ino < sctx->send_progress)
1616 ret = inode_state_did_create;
1618 ret = inode_state_will_create;
1619 } else if (right_gen == gen) {
1620 if (ino < sctx->send_progress)
1621 ret = inode_state_did_delete;
1623 ret = inode_state_will_delete;
1627 } else if (!left_ret) {
1628 if (left_gen == gen) {
1629 if (ino < sctx->send_progress)
1630 ret = inode_state_did_create;
1632 ret = inode_state_will_create;
1636 } else if (!right_ret) {
1637 if (right_gen == gen) {
1638 if (ino < sctx->send_progress)
1639 ret = inode_state_did_delete;
1641 ret = inode_state_will_delete;
1653 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1657 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1660 ret = get_cur_inode_state(sctx, ino, gen);
1664 if (ret == inode_state_no_change ||
1665 ret == inode_state_did_create ||
1666 ret == inode_state_will_delete)
1676 * Helper function to lookup a dir item in a dir.
1678 static int lookup_dir_item_inode(struct btrfs_root *root,
1679 u64 dir, const char *name, int name_len,
1684 struct btrfs_dir_item *di;
1685 struct btrfs_key key;
1686 struct btrfs_path *path;
1688 path = alloc_path_for_send();
1692 di = btrfs_lookup_dir_item(NULL, root, path,
1693 dir, name, name_len, 0);
1702 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1703 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1707 *found_inode = key.objectid;
1708 *found_type = btrfs_dir_type(path->nodes[0], di);
1711 btrfs_free_path(path);
1716 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1717 * generation of the parent dir and the name of the dir entry.
1719 static int get_first_ref(struct btrfs_root *root, u64 ino,
1720 u64 *dir, u64 *dir_gen, struct fs_path *name)
1723 struct btrfs_key key;
1724 struct btrfs_key found_key;
1725 struct btrfs_path *path;
1729 path = alloc_path_for_send();
1734 key.type = BTRFS_INODE_REF_KEY;
1737 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1741 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1743 if (ret || found_key.objectid != ino ||
1744 (found_key.type != BTRFS_INODE_REF_KEY &&
1745 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1750 if (found_key.type == BTRFS_INODE_REF_KEY) {
1751 struct btrfs_inode_ref *iref;
1752 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1753 struct btrfs_inode_ref);
1754 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1755 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1756 (unsigned long)(iref + 1),
1758 parent_dir = found_key.offset;
1760 struct btrfs_inode_extref *extref;
1761 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1762 struct btrfs_inode_extref);
1763 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1764 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1765 (unsigned long)&extref->name, len);
1766 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1770 btrfs_release_path(path);
1773 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1782 btrfs_free_path(path);
1786 static int is_first_ref(struct btrfs_root *root,
1788 const char *name, int name_len)
1791 struct fs_path *tmp_name;
1794 tmp_name = fs_path_alloc();
1798 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1802 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1807 ret = !memcmp(tmp_name->start, name, name_len);
1810 fs_path_free(tmp_name);
1815 * Used by process_recorded_refs to determine if a new ref would overwrite an
1816 * already existing ref. In case it detects an overwrite, it returns the
1817 * inode/gen in who_ino/who_gen.
1818 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1819 * to make sure later references to the overwritten inode are possible.
1820 * Orphanizing is however only required for the first ref of an inode.
1821 * process_recorded_refs does an additional is_first_ref check to see if
1822 * orphanizing is really required.
1824 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1825 const char *name, int name_len,
1826 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1830 u64 other_inode = 0;
1833 if (!sctx->parent_root)
1836 ret = is_inode_existent(sctx, dir, dir_gen);
1841 * If we have a parent root we need to verify that the parent dir was
1842 * not deleted and then re-created, if it was then we have no overwrite
1843 * and we can just unlink this entry.
1845 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1846 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1848 if (ret < 0 && ret != -ENOENT)
1858 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1859 &other_inode, &other_type);
1860 if (ret < 0 && ret != -ENOENT)
1868 * Check if the overwritten ref was already processed. If yes, the ref
1869 * was already unlinked/moved, so we can safely assume that we will not
1870 * overwrite anything at this point in time.
1872 if (other_inode > sctx->send_progress ||
1873 is_waiting_for_move(sctx, other_inode)) {
1874 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1875 who_gen, who_mode, NULL, NULL, NULL);
1880 *who_ino = other_inode;
1890 * Checks if the ref was overwritten by an already processed inode. This is
1891 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1892 * thus the orphan name needs be used.
1893 * process_recorded_refs also uses it to avoid unlinking of refs that were
1896 static int did_overwrite_ref(struct send_ctx *sctx,
1897 u64 dir, u64 dir_gen,
1898 u64 ino, u64 ino_gen,
1899 const char *name, int name_len)
1906 if (!sctx->parent_root)
1909 ret = is_inode_existent(sctx, dir, dir_gen);
1913 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1914 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1916 if (ret < 0 && ret != -ENOENT)
1926 /* check if the ref was overwritten by another ref */
1927 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1928 &ow_inode, &other_type);
1929 if (ret < 0 && ret != -ENOENT)
1932 /* was never and will never be overwritten */
1937 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1942 if (ow_inode == ino && gen == ino_gen) {
1948 * We know that it is or will be overwritten. Check this now.
1949 * The current inode being processed might have been the one that caused
1950 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1951 * the current inode being processed.
1953 if ((ow_inode < sctx->send_progress) ||
1954 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1955 gen == sctx->cur_inode_gen))
1965 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1966 * that got overwritten. This is used by process_recorded_refs to determine
1967 * if it has to use the path as returned by get_cur_path or the orphan name.
1969 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1972 struct fs_path *name = NULL;
1976 if (!sctx->parent_root)
1979 name = fs_path_alloc();
1983 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1987 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1988 name->start, fs_path_len(name));
1996 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1997 * so we need to do some special handling in case we have clashes. This function
1998 * takes care of this with the help of name_cache_entry::radix_list.
1999 * In case of error, nce is kfreed.
2001 static int name_cache_insert(struct send_ctx *sctx,
2002 struct name_cache_entry *nce)
2005 struct list_head *nce_head;
2007 nce_head = radix_tree_lookup(&sctx->name_cache,
2008 (unsigned long)nce->ino);
2010 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2015 INIT_LIST_HEAD(nce_head);
2017 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2024 list_add_tail(&nce->radix_list, nce_head);
2025 list_add_tail(&nce->list, &sctx->name_cache_list);
2026 sctx->name_cache_size++;
2031 static void name_cache_delete(struct send_ctx *sctx,
2032 struct name_cache_entry *nce)
2034 struct list_head *nce_head;
2036 nce_head = radix_tree_lookup(&sctx->name_cache,
2037 (unsigned long)nce->ino);
2039 btrfs_err(sctx->send_root->fs_info,
2040 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2041 nce->ino, sctx->name_cache_size);
2044 list_del(&nce->radix_list);
2045 list_del(&nce->list);
2046 sctx->name_cache_size--;
2049 * We may not get to the final release of nce_head if the lookup fails
2051 if (nce_head && list_empty(nce_head)) {
2052 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2057 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2060 struct list_head *nce_head;
2061 struct name_cache_entry *cur;
2063 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2067 list_for_each_entry(cur, nce_head, radix_list) {
2068 if (cur->ino == ino && cur->gen == gen)
2075 * Removes the entry from the list and adds it back to the end. This marks the
2076 * entry as recently used so that name_cache_clean_unused does not remove it.
2078 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2080 list_del(&nce->list);
2081 list_add_tail(&nce->list, &sctx->name_cache_list);
2085 * Remove some entries from the beginning of name_cache_list.
2087 static void name_cache_clean_unused(struct send_ctx *sctx)
2089 struct name_cache_entry *nce;
2091 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2094 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2095 nce = list_entry(sctx->name_cache_list.next,
2096 struct name_cache_entry, list);
2097 name_cache_delete(sctx, nce);
2102 static void name_cache_free(struct send_ctx *sctx)
2104 struct name_cache_entry *nce;
2106 while (!list_empty(&sctx->name_cache_list)) {
2107 nce = list_entry(sctx->name_cache_list.next,
2108 struct name_cache_entry, list);
2109 name_cache_delete(sctx, nce);
2115 * Used by get_cur_path for each ref up to the root.
2116 * Returns 0 if it succeeded.
2117 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2118 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2119 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2120 * Returns <0 in case of error.
2122 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2126 struct fs_path *dest)
2130 struct name_cache_entry *nce = NULL;
2133 * First check if we already did a call to this function with the same
2134 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2135 * return the cached result.
2137 nce = name_cache_search(sctx, ino, gen);
2139 if (ino < sctx->send_progress && nce->need_later_update) {
2140 name_cache_delete(sctx, nce);
2144 name_cache_used(sctx, nce);
2145 *parent_ino = nce->parent_ino;
2146 *parent_gen = nce->parent_gen;
2147 ret = fs_path_add(dest, nce->name, nce->name_len);
2156 * If the inode is not existent yet, add the orphan name and return 1.
2157 * This should only happen for the parent dir that we determine in
2160 ret = is_inode_existent(sctx, ino, gen);
2165 ret = gen_unique_name(sctx, ino, gen, dest);
2173 * Depending on whether the inode was already processed or not, use
2174 * send_root or parent_root for ref lookup.
2176 if (ino < sctx->send_progress)
2177 ret = get_first_ref(sctx->send_root, ino,
2178 parent_ino, parent_gen, dest);
2180 ret = get_first_ref(sctx->parent_root, ino,
2181 parent_ino, parent_gen, dest);
2186 * Check if the ref was overwritten by an inode's ref that was processed
2187 * earlier. If yes, treat as orphan and return 1.
2189 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2190 dest->start, dest->end - dest->start);
2194 fs_path_reset(dest);
2195 ret = gen_unique_name(sctx, ino, gen, dest);
2203 * Store the result of the lookup in the name cache.
2205 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2213 nce->parent_ino = *parent_ino;
2214 nce->parent_gen = *parent_gen;
2215 nce->name_len = fs_path_len(dest);
2217 strcpy(nce->name, dest->start);
2219 if (ino < sctx->send_progress)
2220 nce->need_later_update = 0;
2222 nce->need_later_update = 1;
2224 nce_ret = name_cache_insert(sctx, nce);
2227 name_cache_clean_unused(sctx);
2234 * Magic happens here. This function returns the first ref to an inode as it
2235 * would look like while receiving the stream at this point in time.
2236 * We walk the path up to the root. For every inode in between, we check if it
2237 * was already processed/sent. If yes, we continue with the parent as found
2238 * in send_root. If not, we continue with the parent as found in parent_root.
2239 * If we encounter an inode that was deleted at this point in time, we use the
2240 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2241 * that were not created yet and overwritten inodes/refs.
2243 * When do we have have orphan inodes:
2244 * 1. When an inode is freshly created and thus no valid refs are available yet
2245 * 2. When a directory lost all it's refs (deleted) but still has dir items
2246 * inside which were not processed yet (pending for move/delete). If anyone
2247 * tried to get the path to the dir items, it would get a path inside that
2249 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2250 * of an unprocessed inode. If in that case the first ref would be
2251 * overwritten, the overwritten inode gets "orphanized". Later when we
2252 * process this overwritten inode, it is restored at a new place by moving
2255 * sctx->send_progress tells this function at which point in time receiving
2258 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2259 struct fs_path *dest)
2262 struct fs_path *name = NULL;
2263 u64 parent_inode = 0;
2267 name = fs_path_alloc();
2274 fs_path_reset(dest);
2276 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2277 struct waiting_dir_move *wdm;
2279 fs_path_reset(name);
2281 if (is_waiting_for_rm(sctx, ino)) {
2282 ret = gen_unique_name(sctx, ino, gen, name);
2285 ret = fs_path_add_path(dest, name);
2289 wdm = get_waiting_dir_move(sctx, ino);
2290 if (wdm && wdm->orphanized) {
2291 ret = gen_unique_name(sctx, ino, gen, name);
2294 ret = get_first_ref(sctx->parent_root, ino,
2295 &parent_inode, &parent_gen, name);
2297 ret = __get_cur_name_and_parent(sctx, ino, gen,
2307 ret = fs_path_add_path(dest, name);
2318 fs_path_unreverse(dest);
2323 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2325 static int send_subvol_begin(struct send_ctx *sctx)
2328 struct btrfs_root *send_root = sctx->send_root;
2329 struct btrfs_root *parent_root = sctx->parent_root;
2330 struct btrfs_path *path;
2331 struct btrfs_key key;
2332 struct btrfs_root_ref *ref;
2333 struct extent_buffer *leaf;
2337 path = btrfs_alloc_path();
2341 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2343 btrfs_free_path(path);
2347 key.objectid = send_root->objectid;
2348 key.type = BTRFS_ROOT_BACKREF_KEY;
2351 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2360 leaf = path->nodes[0];
2361 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2362 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2363 key.objectid != send_root->objectid) {
2367 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2368 namelen = btrfs_root_ref_name_len(leaf, ref);
2369 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2370 btrfs_release_path(path);
2373 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2377 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2382 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2384 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2385 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2386 sctx->send_root->root_item.received_uuid);
2388 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2389 sctx->send_root->root_item.uuid);
2391 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2392 le64_to_cpu(sctx->send_root->root_item.ctransid));
2394 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2395 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2396 parent_root->root_item.received_uuid);
2398 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2399 parent_root->root_item.uuid);
2400 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2401 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2404 ret = send_cmd(sctx);
2408 btrfs_free_path(path);
2413 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2415 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2419 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2421 p = fs_path_alloc();
2425 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2429 ret = get_cur_path(sctx, ino, gen, p);
2432 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2433 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2435 ret = send_cmd(sctx);
2443 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2445 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2449 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2451 p = fs_path_alloc();
2455 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2459 ret = get_cur_path(sctx, ino, gen, p);
2462 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2463 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2465 ret = send_cmd(sctx);
2473 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2475 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2479 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2482 p = fs_path_alloc();
2486 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2490 ret = get_cur_path(sctx, ino, gen, p);
2493 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2494 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2495 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2497 ret = send_cmd(sctx);
2505 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2507 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2509 struct fs_path *p = NULL;
2510 struct btrfs_inode_item *ii;
2511 struct btrfs_path *path = NULL;
2512 struct extent_buffer *eb;
2513 struct btrfs_key key;
2516 btrfs_debug(fs_info, "send_utimes %llu", ino);
2518 p = fs_path_alloc();
2522 path = alloc_path_for_send();
2529 key.type = BTRFS_INODE_ITEM_KEY;
2531 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2537 eb = path->nodes[0];
2538 slot = path->slots[0];
2539 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2541 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2545 ret = get_cur_path(sctx, ino, gen, p);
2548 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2549 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2550 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2551 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2552 /* TODO Add otime support when the otime patches get into upstream */
2554 ret = send_cmd(sctx);
2559 btrfs_free_path(path);
2564 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2565 * a valid path yet because we did not process the refs yet. So, the inode
2566 * is created as orphan.
2568 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2570 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2578 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2580 p = fs_path_alloc();
2584 if (ino != sctx->cur_ino) {
2585 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2590 gen = sctx->cur_inode_gen;
2591 mode = sctx->cur_inode_mode;
2592 rdev = sctx->cur_inode_rdev;
2595 if (S_ISREG(mode)) {
2596 cmd = BTRFS_SEND_C_MKFILE;
2597 } else if (S_ISDIR(mode)) {
2598 cmd = BTRFS_SEND_C_MKDIR;
2599 } else if (S_ISLNK(mode)) {
2600 cmd = BTRFS_SEND_C_SYMLINK;
2601 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2602 cmd = BTRFS_SEND_C_MKNOD;
2603 } else if (S_ISFIFO(mode)) {
2604 cmd = BTRFS_SEND_C_MKFIFO;
2605 } else if (S_ISSOCK(mode)) {
2606 cmd = BTRFS_SEND_C_MKSOCK;
2608 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2609 (int)(mode & S_IFMT));
2614 ret = begin_cmd(sctx, cmd);
2618 ret = gen_unique_name(sctx, ino, gen, p);
2622 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2623 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2625 if (S_ISLNK(mode)) {
2627 ret = read_symlink(sctx->send_root, ino, p);
2630 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2631 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2632 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2633 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2634 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2637 ret = send_cmd(sctx);
2649 * We need some special handling for inodes that get processed before the parent
2650 * directory got created. See process_recorded_refs for details.
2651 * This function does the check if we already created the dir out of order.
2653 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2656 struct btrfs_path *path = NULL;
2657 struct btrfs_key key;
2658 struct btrfs_key found_key;
2659 struct btrfs_key di_key;
2660 struct extent_buffer *eb;
2661 struct btrfs_dir_item *di;
2664 path = alloc_path_for_send();
2671 key.type = BTRFS_DIR_INDEX_KEY;
2673 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2678 eb = path->nodes[0];
2679 slot = path->slots[0];
2680 if (slot >= btrfs_header_nritems(eb)) {
2681 ret = btrfs_next_leaf(sctx->send_root, path);
2684 } else if (ret > 0) {
2691 btrfs_item_key_to_cpu(eb, &found_key, slot);
2692 if (found_key.objectid != key.objectid ||
2693 found_key.type != key.type) {
2698 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2699 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2701 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2702 di_key.objectid < sctx->send_progress) {
2711 btrfs_free_path(path);
2716 * Only creates the inode if it is:
2717 * 1. Not a directory
2718 * 2. Or a directory which was not created already due to out of order
2719 * directories. See did_create_dir and process_recorded_refs for details.
2721 static int send_create_inode_if_needed(struct send_ctx *sctx)
2725 if (S_ISDIR(sctx->cur_inode_mode)) {
2726 ret = did_create_dir(sctx, sctx->cur_ino);
2735 ret = send_create_inode(sctx, sctx->cur_ino);
2743 struct recorded_ref {
2744 struct list_head list;
2746 struct fs_path *full_path;
2752 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2754 ref->full_path = path;
2755 ref->name = (char *)kbasename(ref->full_path->start);
2756 ref->name_len = ref->full_path->end - ref->name;
2760 * We need to process new refs before deleted refs, but compare_tree gives us
2761 * everything mixed. So we first record all refs and later process them.
2762 * This function is a helper to record one ref.
2764 static int __record_ref(struct list_head *head, u64 dir,
2765 u64 dir_gen, struct fs_path *path)
2767 struct recorded_ref *ref;
2769 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2774 ref->dir_gen = dir_gen;
2775 set_ref_path(ref, path);
2776 list_add_tail(&ref->list, head);
2780 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2782 struct recorded_ref *new;
2784 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2788 new->dir = ref->dir;
2789 new->dir_gen = ref->dir_gen;
2790 new->full_path = NULL;
2791 INIT_LIST_HEAD(&new->list);
2792 list_add_tail(&new->list, list);
2796 static void __free_recorded_refs(struct list_head *head)
2798 struct recorded_ref *cur;
2800 while (!list_empty(head)) {
2801 cur = list_entry(head->next, struct recorded_ref, list);
2802 fs_path_free(cur->full_path);
2803 list_del(&cur->list);
2808 static void free_recorded_refs(struct send_ctx *sctx)
2810 __free_recorded_refs(&sctx->new_refs);
2811 __free_recorded_refs(&sctx->deleted_refs);
2815 * Renames/moves a file/dir to its orphan name. Used when the first
2816 * ref of an unprocessed inode gets overwritten and for all non empty
2819 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2820 struct fs_path *path)
2823 struct fs_path *orphan;
2825 orphan = fs_path_alloc();
2829 ret = gen_unique_name(sctx, ino, gen, orphan);
2833 ret = send_rename(sctx, path, orphan);
2836 fs_path_free(orphan);
2840 static struct orphan_dir_info *
2841 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2843 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2844 struct rb_node *parent = NULL;
2845 struct orphan_dir_info *entry, *odi;
2847 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2849 return ERR_PTR(-ENOMEM);
2855 entry = rb_entry(parent, struct orphan_dir_info, node);
2856 if (dir_ino < entry->ino) {
2858 } else if (dir_ino > entry->ino) {
2859 p = &(*p)->rb_right;
2866 rb_link_node(&odi->node, parent, p);
2867 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2871 static struct orphan_dir_info *
2872 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2874 struct rb_node *n = sctx->orphan_dirs.rb_node;
2875 struct orphan_dir_info *entry;
2878 entry = rb_entry(n, struct orphan_dir_info, node);
2879 if (dir_ino < entry->ino)
2881 else if (dir_ino > entry->ino)
2889 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2891 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2896 static void free_orphan_dir_info(struct send_ctx *sctx,
2897 struct orphan_dir_info *odi)
2901 rb_erase(&odi->node, &sctx->orphan_dirs);
2906 * Returns 1 if a directory can be removed at this point in time.
2907 * We check this by iterating all dir items and checking if the inode behind
2908 * the dir item was already processed.
2910 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2914 struct btrfs_root *root = sctx->parent_root;
2915 struct btrfs_path *path;
2916 struct btrfs_key key;
2917 struct btrfs_key found_key;
2918 struct btrfs_key loc;
2919 struct btrfs_dir_item *di;
2922 * Don't try to rmdir the top/root subvolume dir.
2924 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2927 path = alloc_path_for_send();
2932 key.type = BTRFS_DIR_INDEX_KEY;
2934 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2939 struct waiting_dir_move *dm;
2941 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2942 ret = btrfs_next_leaf(root, path);
2949 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2951 if (found_key.objectid != key.objectid ||
2952 found_key.type != key.type)
2955 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2956 struct btrfs_dir_item);
2957 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2959 dm = get_waiting_dir_move(sctx, loc.objectid);
2961 struct orphan_dir_info *odi;
2963 odi = add_orphan_dir_info(sctx, dir);
2969 dm->rmdir_ino = dir;
2974 if (loc.objectid > send_progress) {
2975 struct orphan_dir_info *odi;
2977 odi = get_orphan_dir_info(sctx, dir);
2978 free_orphan_dir_info(sctx, odi);
2989 btrfs_free_path(path);
2993 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2995 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2997 return entry != NULL;
3000 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3002 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3003 struct rb_node *parent = NULL;
3004 struct waiting_dir_move *entry, *dm;
3006 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3011 dm->orphanized = orphanized;
3015 entry = rb_entry(parent, struct waiting_dir_move, node);
3016 if (ino < entry->ino) {
3018 } else if (ino > entry->ino) {
3019 p = &(*p)->rb_right;
3026 rb_link_node(&dm->node, parent, p);
3027 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3031 static struct waiting_dir_move *
3032 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3034 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3035 struct waiting_dir_move *entry;
3038 entry = rb_entry(n, struct waiting_dir_move, node);
3039 if (ino < entry->ino)
3041 else if (ino > entry->ino)
3049 static void free_waiting_dir_move(struct send_ctx *sctx,
3050 struct waiting_dir_move *dm)
3054 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3058 static int add_pending_dir_move(struct send_ctx *sctx,
3062 struct list_head *new_refs,
3063 struct list_head *deleted_refs,
3064 const bool is_orphan)
3066 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3067 struct rb_node *parent = NULL;
3068 struct pending_dir_move *entry = NULL, *pm;
3069 struct recorded_ref *cur;
3073 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3076 pm->parent_ino = parent_ino;
3079 INIT_LIST_HEAD(&pm->list);
3080 INIT_LIST_HEAD(&pm->update_refs);
3081 RB_CLEAR_NODE(&pm->node);
3085 entry = rb_entry(parent, struct pending_dir_move, node);
3086 if (parent_ino < entry->parent_ino) {
3088 } else if (parent_ino > entry->parent_ino) {
3089 p = &(*p)->rb_right;
3096 list_for_each_entry(cur, deleted_refs, list) {
3097 ret = dup_ref(cur, &pm->update_refs);
3101 list_for_each_entry(cur, new_refs, list) {
3102 ret = dup_ref(cur, &pm->update_refs);
3107 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3112 list_add_tail(&pm->list, &entry->list);
3114 rb_link_node(&pm->node, parent, p);
3115 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3120 __free_recorded_refs(&pm->update_refs);
3126 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3129 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3130 struct pending_dir_move *entry;
3133 entry = rb_entry(n, struct pending_dir_move, node);
3134 if (parent_ino < entry->parent_ino)
3136 else if (parent_ino > entry->parent_ino)
3144 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3145 u64 ino, u64 gen, u64 *ancestor_ino)
3148 u64 parent_inode = 0;
3150 u64 start_ino = ino;
3153 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3154 fs_path_reset(name);
3156 if (is_waiting_for_rm(sctx, ino))
3158 if (is_waiting_for_move(sctx, ino)) {
3159 if (*ancestor_ino == 0)
3160 *ancestor_ino = ino;
3161 ret = get_first_ref(sctx->parent_root, ino,
3162 &parent_inode, &parent_gen, name);
3164 ret = __get_cur_name_and_parent(sctx, ino, gen,
3174 if (parent_inode == start_ino) {
3176 if (*ancestor_ino == 0)
3177 *ancestor_ino = ino;
3186 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3188 struct fs_path *from_path = NULL;
3189 struct fs_path *to_path = NULL;
3190 struct fs_path *name = NULL;
3191 u64 orig_progress = sctx->send_progress;
3192 struct recorded_ref *cur;
3193 u64 parent_ino, parent_gen;
3194 struct waiting_dir_move *dm = NULL;
3200 name = fs_path_alloc();
3201 from_path = fs_path_alloc();
3202 if (!name || !from_path) {
3207 dm = get_waiting_dir_move(sctx, pm->ino);
3209 rmdir_ino = dm->rmdir_ino;
3210 is_orphan = dm->orphanized;
3211 free_waiting_dir_move(sctx, dm);
3214 ret = gen_unique_name(sctx, pm->ino,
3215 pm->gen, from_path);
3217 ret = get_first_ref(sctx->parent_root, pm->ino,
3218 &parent_ino, &parent_gen, name);
3221 ret = get_cur_path(sctx, parent_ino, parent_gen,
3225 ret = fs_path_add_path(from_path, name);
3230 sctx->send_progress = sctx->cur_ino + 1;
3231 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3235 LIST_HEAD(deleted_refs);
3236 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3237 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3238 &pm->update_refs, &deleted_refs,
3243 dm = get_waiting_dir_move(sctx, pm->ino);
3245 dm->rmdir_ino = rmdir_ino;
3249 fs_path_reset(name);
3252 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3256 ret = send_rename(sctx, from_path, to_path);
3261 struct orphan_dir_info *odi;
3263 odi = get_orphan_dir_info(sctx, rmdir_ino);
3265 /* already deleted */
3268 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino);
3274 name = fs_path_alloc();
3279 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3282 ret = send_rmdir(sctx, name);
3285 free_orphan_dir_info(sctx, odi);
3289 ret = send_utimes(sctx, pm->ino, pm->gen);
3294 * After rename/move, need to update the utimes of both new parent(s)
3295 * and old parent(s).
3297 list_for_each_entry(cur, &pm->update_refs, list) {
3299 * The parent inode might have been deleted in the send snapshot
3301 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3302 NULL, NULL, NULL, NULL, NULL);
3303 if (ret == -ENOENT) {
3310 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3317 fs_path_free(from_path);
3318 fs_path_free(to_path);
3319 sctx->send_progress = orig_progress;
3324 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3326 if (!list_empty(&m->list))
3328 if (!RB_EMPTY_NODE(&m->node))
3329 rb_erase(&m->node, &sctx->pending_dir_moves);
3330 __free_recorded_refs(&m->update_refs);
3334 static void tail_append_pending_moves(struct pending_dir_move *moves,
3335 struct list_head *stack)
3337 if (list_empty(&moves->list)) {
3338 list_add_tail(&moves->list, stack);
3341 list_splice_init(&moves->list, &list);
3342 list_add_tail(&moves->list, stack);
3343 list_splice_tail(&list, stack);
3347 static int apply_children_dir_moves(struct send_ctx *sctx)
3349 struct pending_dir_move *pm;
3350 struct list_head stack;
3351 u64 parent_ino = sctx->cur_ino;
3354 pm = get_pending_dir_moves(sctx, parent_ino);
3358 INIT_LIST_HEAD(&stack);
3359 tail_append_pending_moves(pm, &stack);
3361 while (!list_empty(&stack)) {
3362 pm = list_first_entry(&stack, struct pending_dir_move, list);
3363 parent_ino = pm->ino;
3364 ret = apply_dir_move(sctx, pm);
3365 free_pending_move(sctx, pm);
3368 pm = get_pending_dir_moves(sctx, parent_ino);
3370 tail_append_pending_moves(pm, &stack);
3375 while (!list_empty(&stack)) {
3376 pm = list_first_entry(&stack, struct pending_dir_move, list);
3377 free_pending_move(sctx, pm);
3383 * We might need to delay a directory rename even when no ancestor directory
3384 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3385 * renamed. This happens when we rename a directory to the old name (the name
3386 * in the parent root) of some other unrelated directory that got its rename
3387 * delayed due to some ancestor with higher number that got renamed.
3393 * |---- a/ (ino 257)
3394 * | |---- file (ino 260)
3396 * |---- b/ (ino 258)
3397 * |---- c/ (ino 259)
3401 * |---- a/ (ino 258)
3402 * |---- x/ (ino 259)
3403 * |---- y/ (ino 257)
3404 * |----- file (ino 260)
3406 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3407 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3408 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3411 * 1 - rename 259 from 'c' to 'x'
3412 * 2 - rename 257 from 'a' to 'x/y'
3413 * 3 - rename 258 from 'b' to 'a'
3415 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3416 * be done right away and < 0 on error.
3418 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3419 struct recorded_ref *parent_ref,
3420 const bool is_orphan)
3422 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3423 struct btrfs_path *path;
3424 struct btrfs_key key;
3425 struct btrfs_key di_key;
3426 struct btrfs_dir_item *di;
3430 struct waiting_dir_move *wdm;
3432 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3435 path = alloc_path_for_send();
3439 key.objectid = parent_ref->dir;
3440 key.type = BTRFS_DIR_ITEM_KEY;
3441 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3443 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3446 } else if (ret > 0) {
3451 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3452 parent_ref->name_len);
3458 * di_key.objectid has the number of the inode that has a dentry in the
3459 * parent directory with the same name that sctx->cur_ino is being
3460 * renamed to. We need to check if that inode is in the send root as
3461 * well and if it is currently marked as an inode with a pending rename,
3462 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3463 * that it happens after that other inode is renamed.
3465 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3466 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3471 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3472 &left_gen, NULL, NULL, NULL, NULL);
3475 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3476 &right_gen, NULL, NULL, NULL, NULL);
3483 /* Different inode, no need to delay the rename of sctx->cur_ino */
3484 if (right_gen != left_gen) {
3489 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3490 if (wdm && !wdm->orphanized) {
3491 ret = add_pending_dir_move(sctx,
3493 sctx->cur_inode_gen,
3496 &sctx->deleted_refs,
3502 btrfs_free_path(path);
3507 * Check if inode ino2, or any of its ancestors, is inode ino1.
3508 * Return 1 if true, 0 if false and < 0 on error.
3510 static int check_ino_in_path(struct btrfs_root *root,
3515 struct fs_path *fs_path)
3520 return ino1_gen == ino2_gen;
3522 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3527 fs_path_reset(fs_path);
3528 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3532 return parent_gen == ino1_gen;
3539 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3540 * possible path (in case ino2 is not a directory and has multiple hard links).
3541 * Return 1 if true, 0 if false and < 0 on error.
3543 static int is_ancestor(struct btrfs_root *root,
3547 struct fs_path *fs_path)
3549 bool free_fs_path = false;
3551 struct btrfs_path *path = NULL;
3552 struct btrfs_key key;
3555 fs_path = fs_path_alloc();
3558 free_fs_path = true;
3561 path = alloc_path_for_send();
3567 key.objectid = ino2;
3568 key.type = BTRFS_INODE_REF_KEY;
3571 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3576 struct extent_buffer *leaf = path->nodes[0];
3577 int slot = path->slots[0];
3581 if (slot >= btrfs_header_nritems(leaf)) {
3582 ret = btrfs_next_leaf(root, path);
3590 btrfs_item_key_to_cpu(leaf, &key, slot);
3591 if (key.objectid != ino2)
3593 if (key.type != BTRFS_INODE_REF_KEY &&
3594 key.type != BTRFS_INODE_EXTREF_KEY)
3597 item_size = btrfs_item_size_nr(leaf, slot);
3598 while (cur_offset < item_size) {
3602 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3604 struct btrfs_inode_extref *extref;
3606 ptr = btrfs_item_ptr_offset(leaf, slot);
3607 extref = (struct btrfs_inode_extref *)
3609 parent = btrfs_inode_extref_parent(leaf,
3611 cur_offset += sizeof(*extref);
3612 cur_offset += btrfs_inode_extref_name_len(leaf,
3615 parent = key.offset;
3616 cur_offset = item_size;
3619 ret = get_inode_info(root, parent, NULL, &parent_gen,
3620 NULL, NULL, NULL, NULL);
3623 ret = check_ino_in_path(root, ino1, ino1_gen,
3624 parent, parent_gen, fs_path);
3632 btrfs_free_path(path);
3634 fs_path_free(fs_path);
3638 static int wait_for_parent_move(struct send_ctx *sctx,
3639 struct recorded_ref *parent_ref,
3640 const bool is_orphan)
3643 u64 ino = parent_ref->dir;
3644 u64 ino_gen = parent_ref->dir_gen;
3645 u64 parent_ino_before, parent_ino_after;
3646 struct fs_path *path_before = NULL;
3647 struct fs_path *path_after = NULL;
3650 path_after = fs_path_alloc();
3651 path_before = fs_path_alloc();
3652 if (!path_after || !path_before) {
3658 * Our current directory inode may not yet be renamed/moved because some
3659 * ancestor (immediate or not) has to be renamed/moved first. So find if
3660 * such ancestor exists and make sure our own rename/move happens after
3661 * that ancestor is processed to avoid path build infinite loops (done
3662 * at get_cur_path()).
3664 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3665 u64 parent_ino_after_gen;
3667 if (is_waiting_for_move(sctx, ino)) {
3669 * If the current inode is an ancestor of ino in the
3670 * parent root, we need to delay the rename of the
3671 * current inode, otherwise don't delayed the rename
3672 * because we can end up with a circular dependency
3673 * of renames, resulting in some directories never
3674 * getting the respective rename operations issued in
3675 * the send stream or getting into infinite path build
3678 ret = is_ancestor(sctx->parent_root,
3679 sctx->cur_ino, sctx->cur_inode_gen,
3685 fs_path_reset(path_before);
3686 fs_path_reset(path_after);
3688 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3689 &parent_ino_after_gen, path_after);
3692 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3694 if (ret < 0 && ret != -ENOENT) {
3696 } else if (ret == -ENOENT) {
3701 len1 = fs_path_len(path_before);
3702 len2 = fs_path_len(path_after);
3703 if (ino > sctx->cur_ino &&
3704 (parent_ino_before != parent_ino_after || len1 != len2 ||
3705 memcmp(path_before->start, path_after->start, len1))) {
3708 ret = get_inode_info(sctx->parent_root, ino, NULL,
3709 &parent_ino_gen, NULL, NULL, NULL,
3713 if (ino_gen == parent_ino_gen) {
3718 ino = parent_ino_after;
3719 ino_gen = parent_ino_after_gen;
3723 fs_path_free(path_before);
3724 fs_path_free(path_after);
3727 ret = add_pending_dir_move(sctx,
3729 sctx->cur_inode_gen,
3732 &sctx->deleted_refs,
3741 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3744 struct fs_path *new_path;
3747 * Our reference's name member points to its full_path member string, so
3748 * we use here a new path.
3750 new_path = fs_path_alloc();
3754 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3756 fs_path_free(new_path);
3759 ret = fs_path_add(new_path, ref->name, ref->name_len);
3761 fs_path_free(new_path);
3765 fs_path_free(ref->full_path);
3766 set_ref_path(ref, new_path);
3772 * This does all the move/link/unlink/rmdir magic.
3774 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3776 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3778 struct recorded_ref *cur;
3779 struct recorded_ref *cur2;
3780 struct list_head check_dirs;
3781 struct fs_path *valid_path = NULL;
3785 int did_overwrite = 0;
3787 u64 last_dir_ino_rm = 0;
3788 bool can_rename = true;
3789 bool orphanized_dir = false;
3790 bool orphanized_ancestor = false;
3792 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3795 * This should never happen as the root dir always has the same ref
3796 * which is always '..'
3798 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3799 INIT_LIST_HEAD(&check_dirs);
3801 valid_path = fs_path_alloc();
3808 * First, check if the first ref of the current inode was overwritten
3809 * before. If yes, we know that the current inode was already orphanized
3810 * and thus use the orphan name. If not, we can use get_cur_path to
3811 * get the path of the first ref as it would like while receiving at
3812 * this point in time.
3813 * New inodes are always orphan at the beginning, so force to use the
3814 * orphan name in this case.
3815 * The first ref is stored in valid_path and will be updated if it
3816 * gets moved around.
3818 if (!sctx->cur_inode_new) {
3819 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3820 sctx->cur_inode_gen);
3826 if (sctx->cur_inode_new || did_overwrite) {
3827 ret = gen_unique_name(sctx, sctx->cur_ino,
3828 sctx->cur_inode_gen, valid_path);
3833 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3839 list_for_each_entry(cur, &sctx->new_refs, list) {
3841 * We may have refs where the parent directory does not exist
3842 * yet. This happens if the parent directories inum is higher
3843 * the the current inum. To handle this case, we create the
3844 * parent directory out of order. But we need to check if this
3845 * did already happen before due to other refs in the same dir.
3847 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3850 if (ret == inode_state_will_create) {
3853 * First check if any of the current inodes refs did
3854 * already create the dir.
3856 list_for_each_entry(cur2, &sctx->new_refs, list) {
3859 if (cur2->dir == cur->dir) {
3866 * If that did not happen, check if a previous inode
3867 * did already create the dir.
3870 ret = did_create_dir(sctx, cur->dir);
3874 ret = send_create_inode(sctx, cur->dir);
3881 * Check if this new ref would overwrite the first ref of
3882 * another unprocessed inode. If yes, orphanize the
3883 * overwritten inode. If we find an overwritten ref that is
3884 * not the first ref, simply unlink it.
3886 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3887 cur->name, cur->name_len,
3888 &ow_inode, &ow_gen, &ow_mode);
3892 ret = is_first_ref(sctx->parent_root,
3893 ow_inode, cur->dir, cur->name,
3898 struct name_cache_entry *nce;
3899 struct waiting_dir_move *wdm;
3901 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3905 if (S_ISDIR(ow_mode))
3906 orphanized_dir = true;
3909 * If ow_inode has its rename operation delayed
3910 * make sure that its orphanized name is used in
3911 * the source path when performing its rename
3914 if (is_waiting_for_move(sctx, ow_inode)) {
3915 wdm = get_waiting_dir_move(sctx,
3918 wdm->orphanized = true;
3922 * Make sure we clear our orphanized inode's
3923 * name from the name cache. This is because the
3924 * inode ow_inode might be an ancestor of some
3925 * other inode that will be orphanized as well
3926 * later and has an inode number greater than
3927 * sctx->send_progress. We need to prevent
3928 * future name lookups from using the old name
3929 * and get instead the orphan name.
3931 nce = name_cache_search(sctx, ow_inode, ow_gen);
3933 name_cache_delete(sctx, nce);
3938 * ow_inode might currently be an ancestor of
3939 * cur_ino, therefore compute valid_path (the
3940 * current path of cur_ino) again because it
3941 * might contain the pre-orphanization name of
3942 * ow_inode, which is no longer valid.
3944 ret = is_ancestor(sctx->parent_root,
3946 sctx->cur_ino, NULL);
3948 orphanized_ancestor = true;
3949 fs_path_reset(valid_path);
3950 ret = get_cur_path(sctx, sctx->cur_ino,
3951 sctx->cur_inode_gen,
3957 ret = send_unlink(sctx, cur->full_path);
3963 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3964 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3973 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3975 ret = wait_for_parent_move(sctx, cur, is_orphan);
3985 * link/move the ref to the new place. If we have an orphan
3986 * inode, move it and update valid_path. If not, link or move
3987 * it depending on the inode mode.
3989 if (is_orphan && can_rename) {
3990 ret = send_rename(sctx, valid_path, cur->full_path);
3994 ret = fs_path_copy(valid_path, cur->full_path);
3997 } else if (can_rename) {
3998 if (S_ISDIR(sctx->cur_inode_mode)) {
4000 * Dirs can't be linked, so move it. For moved
4001 * dirs, we always have one new and one deleted
4002 * ref. The deleted ref is ignored later.
4004 ret = send_rename(sctx, valid_path,
4007 ret = fs_path_copy(valid_path,
4013 * We might have previously orphanized an inode
4014 * which is an ancestor of our current inode,
4015 * so our reference's full path, which was
4016 * computed before any such orphanizations, must
4019 if (orphanized_dir) {
4020 ret = update_ref_path(sctx, cur);
4024 ret = send_link(sctx, cur->full_path,
4030 ret = dup_ref(cur, &check_dirs);
4035 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4037 * Check if we can already rmdir the directory. If not,
4038 * orphanize it. For every dir item inside that gets deleted
4039 * later, we do this check again and rmdir it then if possible.
4040 * See the use of check_dirs for more details.
4042 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4047 ret = send_rmdir(sctx, valid_path);
4050 } else if (!is_orphan) {
4051 ret = orphanize_inode(sctx, sctx->cur_ino,
4052 sctx->cur_inode_gen, valid_path);
4058 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4059 ret = dup_ref(cur, &check_dirs);
4063 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4064 !list_empty(&sctx->deleted_refs)) {
4066 * We have a moved dir. Add the old parent to check_dirs
4068 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4070 ret = dup_ref(cur, &check_dirs);
4073 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4075 * We have a non dir inode. Go through all deleted refs and
4076 * unlink them if they were not already overwritten by other
4079 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4080 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4081 sctx->cur_ino, sctx->cur_inode_gen,
4082 cur->name, cur->name_len);
4087 * If we orphanized any ancestor before, we need
4088 * to recompute the full path for deleted names,
4089 * since any such path was computed before we
4090 * processed any references and orphanized any
4093 if (orphanized_ancestor) {
4094 ret = update_ref_path(sctx, cur);
4098 ret = send_unlink(sctx, cur->full_path);
4102 ret = dup_ref(cur, &check_dirs);
4107 * If the inode is still orphan, unlink the orphan. This may
4108 * happen when a previous inode did overwrite the first ref
4109 * of this inode and no new refs were added for the current
4110 * inode. Unlinking does not mean that the inode is deleted in
4111 * all cases. There may still be links to this inode in other
4115 ret = send_unlink(sctx, valid_path);
4122 * We did collect all parent dirs where cur_inode was once located. We
4123 * now go through all these dirs and check if they are pending for
4124 * deletion and if it's finally possible to perform the rmdir now.
4125 * We also update the inode stats of the parent dirs here.
4127 list_for_each_entry(cur, &check_dirs, list) {
4129 * In case we had refs into dirs that were not processed yet,
4130 * we don't need to do the utime and rmdir logic for these dirs.
4131 * The dir will be processed later.
4133 if (cur->dir > sctx->cur_ino)
4136 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4140 if (ret == inode_state_did_create ||
4141 ret == inode_state_no_change) {
4142 /* TODO delayed utimes */
4143 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4146 } else if (ret == inode_state_did_delete &&
4147 cur->dir != last_dir_ino_rm) {
4148 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4153 ret = get_cur_path(sctx, cur->dir,
4154 cur->dir_gen, valid_path);
4157 ret = send_rmdir(sctx, valid_path);
4160 last_dir_ino_rm = cur->dir;
4168 __free_recorded_refs(&check_dirs);
4169 free_recorded_refs(sctx);
4170 fs_path_free(valid_path);
4174 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4175 void *ctx, struct list_head *refs)
4178 struct send_ctx *sctx = ctx;
4182 p = fs_path_alloc();
4186 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4191 ret = get_cur_path(sctx, dir, gen, p);
4194 ret = fs_path_add_path(p, name);
4198 ret = __record_ref(refs, dir, gen, p);
4206 static int __record_new_ref(int num, u64 dir, int index,
4207 struct fs_path *name,
4210 struct send_ctx *sctx = ctx;
4211 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4215 static int __record_deleted_ref(int num, u64 dir, int index,
4216 struct fs_path *name,
4219 struct send_ctx *sctx = ctx;
4220 return record_ref(sctx->parent_root, dir, name, ctx,
4221 &sctx->deleted_refs);
4224 static int record_new_ref(struct send_ctx *sctx)
4228 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4229 sctx->cmp_key, 0, __record_new_ref, sctx);
4238 static int record_deleted_ref(struct send_ctx *sctx)
4242 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4243 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4252 struct find_ref_ctx {
4255 struct btrfs_root *root;
4256 struct fs_path *name;
4260 static int __find_iref(int num, u64 dir, int index,
4261 struct fs_path *name,
4264 struct find_ref_ctx *ctx = ctx_;
4268 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4269 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4271 * To avoid doing extra lookups we'll only do this if everything
4274 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4278 if (dir_gen != ctx->dir_gen)
4280 ctx->found_idx = num;
4286 static int find_iref(struct btrfs_root *root,
4287 struct btrfs_path *path,
4288 struct btrfs_key *key,
4289 u64 dir, u64 dir_gen, struct fs_path *name)
4292 struct find_ref_ctx ctx;
4296 ctx.dir_gen = dir_gen;
4300 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4304 if (ctx.found_idx == -1)
4307 return ctx.found_idx;
4310 static int __record_changed_new_ref(int num, u64 dir, int index,
4311 struct fs_path *name,
4316 struct send_ctx *sctx = ctx;
4318 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4323 ret = find_iref(sctx->parent_root, sctx->right_path,
4324 sctx->cmp_key, dir, dir_gen, name);
4326 ret = __record_new_ref(num, dir, index, name, sctx);
4333 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4334 struct fs_path *name,
4339 struct send_ctx *sctx = ctx;
4341 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4346 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4347 dir, dir_gen, name);
4349 ret = __record_deleted_ref(num, dir, index, name, sctx);
4356 static int record_changed_ref(struct send_ctx *sctx)
4360 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4361 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4364 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4365 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4375 * Record and process all refs at once. Needed when an inode changes the
4376 * generation number, which means that it was deleted and recreated.
4378 static int process_all_refs(struct send_ctx *sctx,
4379 enum btrfs_compare_tree_result cmd)
4382 struct btrfs_root *root;
4383 struct btrfs_path *path;
4384 struct btrfs_key key;
4385 struct btrfs_key found_key;
4386 struct extent_buffer *eb;
4388 iterate_inode_ref_t cb;
4389 int pending_move = 0;
4391 path = alloc_path_for_send();
4395 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4396 root = sctx->send_root;
4397 cb = __record_new_ref;
4398 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4399 root = sctx->parent_root;
4400 cb = __record_deleted_ref;
4402 btrfs_err(sctx->send_root->fs_info,
4403 "Wrong command %d in process_all_refs", cmd);
4408 key.objectid = sctx->cmp_key->objectid;
4409 key.type = BTRFS_INODE_REF_KEY;
4411 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4416 eb = path->nodes[0];
4417 slot = path->slots[0];
4418 if (slot >= btrfs_header_nritems(eb)) {
4419 ret = btrfs_next_leaf(root, path);
4427 btrfs_item_key_to_cpu(eb, &found_key, slot);
4429 if (found_key.objectid != key.objectid ||
4430 (found_key.type != BTRFS_INODE_REF_KEY &&
4431 found_key.type != BTRFS_INODE_EXTREF_KEY))
4434 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4440 btrfs_release_path(path);
4443 * We don't actually care about pending_move as we are simply
4444 * re-creating this inode and will be rename'ing it into place once we
4445 * rename the parent directory.
4447 ret = process_recorded_refs(sctx, &pending_move);
4449 btrfs_free_path(path);
4453 static int send_set_xattr(struct send_ctx *sctx,
4454 struct fs_path *path,
4455 const char *name, int name_len,
4456 const char *data, int data_len)
4460 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4464 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4465 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4466 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4468 ret = send_cmd(sctx);
4475 static int send_remove_xattr(struct send_ctx *sctx,
4476 struct fs_path *path,
4477 const char *name, int name_len)
4481 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4485 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4486 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4488 ret = send_cmd(sctx);
4495 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4496 const char *name, int name_len,
4497 const char *data, int data_len,
4501 struct send_ctx *sctx = ctx;
4503 struct posix_acl_xattr_header dummy_acl;
4505 p = fs_path_alloc();
4510 * This hack is needed because empty acls are stored as zero byte
4511 * data in xattrs. Problem with that is, that receiving these zero byte
4512 * acls will fail later. To fix this, we send a dummy acl list that
4513 * only contains the version number and no entries.
4515 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4516 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4517 if (data_len == 0) {
4518 dummy_acl.a_version =
4519 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4520 data = (char *)&dummy_acl;
4521 data_len = sizeof(dummy_acl);
4525 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4529 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4536 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4537 const char *name, int name_len,
4538 const char *data, int data_len,
4542 struct send_ctx *sctx = ctx;
4545 p = fs_path_alloc();
4549 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4553 ret = send_remove_xattr(sctx, p, name, name_len);
4560 static int process_new_xattr(struct send_ctx *sctx)
4564 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4565 __process_new_xattr, sctx);
4570 static int process_deleted_xattr(struct send_ctx *sctx)
4572 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4573 __process_deleted_xattr, sctx);
4576 struct find_xattr_ctx {
4584 static int __find_xattr(int num, struct btrfs_key *di_key,
4585 const char *name, int name_len,
4586 const char *data, int data_len,
4587 u8 type, void *vctx)
4589 struct find_xattr_ctx *ctx = vctx;
4591 if (name_len == ctx->name_len &&
4592 strncmp(name, ctx->name, name_len) == 0) {
4593 ctx->found_idx = num;
4594 ctx->found_data_len = data_len;
4595 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4596 if (!ctx->found_data)
4603 static int find_xattr(struct btrfs_root *root,
4604 struct btrfs_path *path,
4605 struct btrfs_key *key,
4606 const char *name, int name_len,
4607 char **data, int *data_len)
4610 struct find_xattr_ctx ctx;
4613 ctx.name_len = name_len;
4615 ctx.found_data = NULL;
4616 ctx.found_data_len = 0;
4618 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4622 if (ctx.found_idx == -1)
4625 *data = ctx.found_data;
4626 *data_len = ctx.found_data_len;
4628 kfree(ctx.found_data);
4630 return ctx.found_idx;
4634 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4635 const char *name, int name_len,
4636 const char *data, int data_len,
4640 struct send_ctx *sctx = ctx;
4641 char *found_data = NULL;
4642 int found_data_len = 0;
4644 ret = find_xattr(sctx->parent_root, sctx->right_path,
4645 sctx->cmp_key, name, name_len, &found_data,
4647 if (ret == -ENOENT) {
4648 ret = __process_new_xattr(num, di_key, name, name_len, data,
4649 data_len, type, ctx);
4650 } else if (ret >= 0) {
4651 if (data_len != found_data_len ||
4652 memcmp(data, found_data, data_len)) {
4653 ret = __process_new_xattr(num, di_key, name, name_len,
4654 data, data_len, type, ctx);
4664 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4665 const char *name, int name_len,
4666 const char *data, int data_len,
4670 struct send_ctx *sctx = ctx;
4672 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4673 name, name_len, NULL, NULL);
4675 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4676 data_len, type, ctx);
4683 static int process_changed_xattr(struct send_ctx *sctx)
4687 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4688 __process_changed_new_xattr, sctx);
4691 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4692 __process_changed_deleted_xattr, sctx);
4698 static int process_all_new_xattrs(struct send_ctx *sctx)
4701 struct btrfs_root *root;
4702 struct btrfs_path *path;
4703 struct btrfs_key key;
4704 struct btrfs_key found_key;
4705 struct extent_buffer *eb;
4708 path = alloc_path_for_send();
4712 root = sctx->send_root;
4714 key.objectid = sctx->cmp_key->objectid;
4715 key.type = BTRFS_XATTR_ITEM_KEY;
4717 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4722 eb = path->nodes[0];
4723 slot = path->slots[0];
4724 if (slot >= btrfs_header_nritems(eb)) {
4725 ret = btrfs_next_leaf(root, path);
4728 } else if (ret > 0) {
4735 btrfs_item_key_to_cpu(eb, &found_key, slot);
4736 if (found_key.objectid != key.objectid ||
4737 found_key.type != key.type) {
4742 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4750 btrfs_free_path(path);
4754 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4756 struct btrfs_root *root = sctx->send_root;
4757 struct btrfs_fs_info *fs_info = root->fs_info;
4758 struct inode *inode;
4761 struct btrfs_key key;
4762 pgoff_t index = offset >> PAGE_SHIFT;
4764 unsigned pg_offset = offset & ~PAGE_MASK;
4767 key.objectid = sctx->cur_ino;
4768 key.type = BTRFS_INODE_ITEM_KEY;
4771 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4773 return PTR_ERR(inode);
4775 if (offset + len > i_size_read(inode)) {
4776 if (offset > i_size_read(inode))
4779 len = offset - i_size_read(inode);
4784 last_index = (offset + len - 1) >> PAGE_SHIFT;
4786 /* initial readahead */
4787 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4788 file_ra_state_init(&sctx->ra, inode->i_mapping);
4790 while (index <= last_index) {
4791 unsigned cur_len = min_t(unsigned, len,
4792 PAGE_SIZE - pg_offset);
4794 page = find_lock_page(inode->i_mapping, index);
4796 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4797 NULL, index, last_index + 1 - index);
4799 page = find_or_create_page(inode->i_mapping, index,
4807 if (PageReadahead(page)) {
4808 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4809 NULL, page, index, last_index + 1 - index);
4812 if (!PageUptodate(page)) {
4813 btrfs_readpage(NULL, page);
4815 if (!PageUptodate(page)) {
4824 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4839 * Read some bytes from the current inode/file and send a write command to
4842 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4844 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4847 ssize_t num_read = 0;
4849 p = fs_path_alloc();
4853 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4855 num_read = fill_read_buf(sctx, offset, len);
4856 if (num_read <= 0) {
4862 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4866 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4870 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4871 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4872 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4874 ret = send_cmd(sctx);
4885 * Send a clone command to user space.
4887 static int send_clone(struct send_ctx *sctx,
4888 u64 offset, u32 len,
4889 struct clone_root *clone_root)
4895 btrfs_debug(sctx->send_root->fs_info,
4896 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4897 offset, len, clone_root->root->objectid, clone_root->ino,
4898 clone_root->offset);
4900 p = fs_path_alloc();
4904 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4908 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4912 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4913 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4914 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4916 if (clone_root->root == sctx->send_root) {
4917 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4918 &gen, NULL, NULL, NULL, NULL);
4921 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4923 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4929 * If the parent we're using has a received_uuid set then use that as
4930 * our clone source as that is what we will look for when doing a
4933 * This covers the case that we create a snapshot off of a received
4934 * subvolume and then use that as the parent and try to receive on a
4937 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4938 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4939 clone_root->root->root_item.received_uuid);
4941 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4942 clone_root->root->root_item.uuid);
4943 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4944 le64_to_cpu(clone_root->root->root_item.ctransid));
4945 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4946 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4947 clone_root->offset);
4949 ret = send_cmd(sctx);
4958 * Send an update extent command to user space.
4960 static int send_update_extent(struct send_ctx *sctx,
4961 u64 offset, u32 len)
4966 p = fs_path_alloc();
4970 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4974 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4978 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4979 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4980 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4982 ret = send_cmd(sctx);
4990 static int send_hole(struct send_ctx *sctx, u64 end)
4992 struct fs_path *p = NULL;
4993 u64 offset = sctx->cur_inode_last_extent;
4997 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4998 return send_update_extent(sctx, offset, end - offset);
5000 p = fs_path_alloc();
5003 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5005 goto tlv_put_failure;
5006 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5007 while (offset < end) {
5008 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5010 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5013 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5014 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5015 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5016 ret = send_cmd(sctx);
5021 sctx->cur_inode_next_write_offset = offset;
5027 static int send_extent_data(struct send_ctx *sctx,
5033 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5034 return send_update_extent(sctx, offset, len);
5036 while (sent < len) {
5037 u64 size = len - sent;
5040 if (size > BTRFS_SEND_READ_SIZE)
5041 size = BTRFS_SEND_READ_SIZE;
5042 ret = send_write(sctx, offset + sent, size);
5052 static int clone_range(struct send_ctx *sctx,
5053 struct clone_root *clone_root,
5054 const u64 disk_byte,
5059 struct btrfs_path *path;
5060 struct btrfs_key key;
5064 * Prevent cloning from a zero offset with a length matching the sector
5065 * size because in some scenarios this will make the receiver fail.
5067 * For example, if in the source filesystem the extent at offset 0
5068 * has a length of sectorsize and it was written using direct IO, then
5069 * it can never be an inline extent (even if compression is enabled).
5070 * Then this extent can be cloned in the original filesystem to a non
5071 * zero file offset, but it may not be possible to clone in the
5072 * destination filesystem because it can be inlined due to compression
5073 * on the destination filesystem (as the receiver's write operations are
5074 * always done using buffered IO). The same happens when the original
5075 * filesystem does not have compression enabled but the destination
5078 if (clone_root->offset == 0 &&
5079 len == sctx->send_root->fs_info->sectorsize)
5080 return send_extent_data(sctx, offset, len);
5082 path = alloc_path_for_send();
5087 * We can't send a clone operation for the entire range if we find
5088 * extent items in the respective range in the source file that
5089 * refer to different extents or if we find holes.
5090 * So check for that and do a mix of clone and regular write/copy
5091 * operations if needed.
5095 * mkfs.btrfs -f /dev/sda
5096 * mount /dev/sda /mnt
5097 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5098 * cp --reflink=always /mnt/foo /mnt/bar
5099 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5100 * btrfs subvolume snapshot -r /mnt /mnt/snap
5102 * If when we send the snapshot and we are processing file bar (which
5103 * has a higher inode number than foo) we blindly send a clone operation
5104 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5105 * a file bar that matches the content of file foo - iow, doesn't match
5106 * the content from bar in the original filesystem.
5108 key.objectid = clone_root->ino;
5109 key.type = BTRFS_EXTENT_DATA_KEY;
5110 key.offset = clone_root->offset;
5111 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5114 if (ret > 0 && path->slots[0] > 0) {
5115 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5116 if (key.objectid == clone_root->ino &&
5117 key.type == BTRFS_EXTENT_DATA_KEY)
5122 struct extent_buffer *leaf = path->nodes[0];
5123 int slot = path->slots[0];
5124 struct btrfs_file_extent_item *ei;
5129 if (slot >= btrfs_header_nritems(leaf)) {
5130 ret = btrfs_next_leaf(clone_root->root, path);
5138 btrfs_item_key_to_cpu(leaf, &key, slot);
5141 * We might have an implicit trailing hole (NO_HOLES feature
5142 * enabled). We deal with it after leaving this loop.
5144 if (key.objectid != clone_root->ino ||
5145 key.type != BTRFS_EXTENT_DATA_KEY)
5148 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5149 type = btrfs_file_extent_type(leaf, ei);
5150 if (type == BTRFS_FILE_EXTENT_INLINE) {
5151 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
5152 ext_len = PAGE_ALIGN(ext_len);
5154 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5157 if (key.offset + ext_len <= clone_root->offset)
5160 if (key.offset > clone_root->offset) {
5161 /* Implicit hole, NO_HOLES feature enabled. */
5162 u64 hole_len = key.offset - clone_root->offset;
5166 ret = send_extent_data(sctx, offset, hole_len);
5174 clone_root->offset += hole_len;
5175 data_offset += hole_len;
5178 if (key.offset >= clone_root->offset + len)
5181 clone_len = min_t(u64, ext_len, len);
5183 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5184 btrfs_file_extent_offset(leaf, ei) == data_offset)
5185 ret = send_clone(sctx, offset, clone_len, clone_root);
5187 ret = send_extent_data(sctx, offset, clone_len);
5195 offset += clone_len;
5196 clone_root->offset += clone_len;
5197 data_offset += clone_len;
5203 ret = send_extent_data(sctx, offset, len);
5207 btrfs_free_path(path);
5211 static int send_write_or_clone(struct send_ctx *sctx,
5212 struct btrfs_path *path,
5213 struct btrfs_key *key,
5214 struct clone_root *clone_root)
5217 struct btrfs_file_extent_item *ei;
5218 u64 offset = key->offset;
5221 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5223 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5224 struct btrfs_file_extent_item);
5225 type = btrfs_file_extent_type(path->nodes[0], ei);
5226 if (type == BTRFS_FILE_EXTENT_INLINE) {
5227 len = btrfs_file_extent_inline_len(path->nodes[0],
5228 path->slots[0], ei);
5230 * it is possible the inline item won't cover the whole page,
5231 * but there may be items after this page. Make
5232 * sure to send the whole thing
5234 len = PAGE_ALIGN(len);
5236 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5239 if (offset + len > sctx->cur_inode_size)
5240 len = sctx->cur_inode_size - offset;
5246 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5250 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5251 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5252 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5255 ret = send_extent_data(sctx, offset, len);
5257 sctx->cur_inode_next_write_offset = offset + len;
5262 static int is_extent_unchanged(struct send_ctx *sctx,
5263 struct btrfs_path *left_path,
5264 struct btrfs_key *ekey)
5267 struct btrfs_key key;
5268 struct btrfs_path *path = NULL;
5269 struct extent_buffer *eb;
5271 struct btrfs_key found_key;
5272 struct btrfs_file_extent_item *ei;
5277 u64 left_offset_fixed;
5285 path = alloc_path_for_send();
5289 eb = left_path->nodes[0];
5290 slot = left_path->slots[0];
5291 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5292 left_type = btrfs_file_extent_type(eb, ei);
5294 if (left_type != BTRFS_FILE_EXTENT_REG) {
5298 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5299 left_len = btrfs_file_extent_num_bytes(eb, ei);
5300 left_offset = btrfs_file_extent_offset(eb, ei);
5301 left_gen = btrfs_file_extent_generation(eb, ei);
5304 * Following comments will refer to these graphics. L is the left
5305 * extents which we are checking at the moment. 1-8 are the right
5306 * extents that we iterate.
5309 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5312 * |--1--|-2b-|...(same as above)
5314 * Alternative situation. Happens on files where extents got split.
5316 * |-----------7-----------|-6-|
5318 * Alternative situation. Happens on files which got larger.
5321 * Nothing follows after 8.
5324 key.objectid = ekey->objectid;
5325 key.type = BTRFS_EXTENT_DATA_KEY;
5326 key.offset = ekey->offset;
5327 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5336 * Handle special case where the right side has no extents at all.
5338 eb = path->nodes[0];
5339 slot = path->slots[0];
5340 btrfs_item_key_to_cpu(eb, &found_key, slot);
5341 if (found_key.objectid != key.objectid ||
5342 found_key.type != key.type) {
5343 /* If we're a hole then just pretend nothing changed */
5344 ret = (left_disknr) ? 0 : 1;
5349 * We're now on 2a, 2b or 7.
5352 while (key.offset < ekey->offset + left_len) {
5353 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5354 right_type = btrfs_file_extent_type(eb, ei);
5355 if (right_type != BTRFS_FILE_EXTENT_REG &&
5356 right_type != BTRFS_FILE_EXTENT_INLINE) {
5361 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5362 right_len = btrfs_file_extent_inline_len(eb, slot, ei);
5363 right_len = PAGE_ALIGN(right_len);
5365 right_len = btrfs_file_extent_num_bytes(eb, ei);
5369 * Are we at extent 8? If yes, we know the extent is changed.
5370 * This may only happen on the first iteration.
5372 if (found_key.offset + right_len <= ekey->offset) {
5373 /* If we're a hole just pretend nothing changed */
5374 ret = (left_disknr) ? 0 : 1;
5379 * We just wanted to see if when we have an inline extent, what
5380 * follows it is a regular extent (wanted to check the above
5381 * condition for inline extents too). This should normally not
5382 * happen but it's possible for example when we have an inline
5383 * compressed extent representing data with a size matching
5384 * the page size (currently the same as sector size).
5386 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5391 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5392 right_offset = btrfs_file_extent_offset(eb, ei);
5393 right_gen = btrfs_file_extent_generation(eb, ei);
5395 left_offset_fixed = left_offset;
5396 if (key.offset < ekey->offset) {
5397 /* Fix the right offset for 2a and 7. */
5398 right_offset += ekey->offset - key.offset;
5400 /* Fix the left offset for all behind 2a and 2b */
5401 left_offset_fixed += key.offset - ekey->offset;
5405 * Check if we have the same extent.
5407 if (left_disknr != right_disknr ||
5408 left_offset_fixed != right_offset ||
5409 left_gen != right_gen) {
5415 * Go to the next extent.
5417 ret = btrfs_next_item(sctx->parent_root, path);
5421 eb = path->nodes[0];
5422 slot = path->slots[0];
5423 btrfs_item_key_to_cpu(eb, &found_key, slot);
5425 if (ret || found_key.objectid != key.objectid ||
5426 found_key.type != key.type) {
5427 key.offset += right_len;
5430 if (found_key.offset != key.offset + right_len) {
5438 * We're now behind the left extent (treat as unchanged) or at the end
5439 * of the right side (treat as changed).
5441 if (key.offset >= ekey->offset + left_len)
5448 btrfs_free_path(path);
5452 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5454 struct btrfs_path *path;
5455 struct btrfs_root *root = sctx->send_root;
5456 struct btrfs_file_extent_item *fi;
5457 struct btrfs_key key;
5462 path = alloc_path_for_send();
5466 sctx->cur_inode_last_extent = 0;
5468 key.objectid = sctx->cur_ino;
5469 key.type = BTRFS_EXTENT_DATA_KEY;
5470 key.offset = offset;
5471 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5475 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5476 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5479 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5480 struct btrfs_file_extent_item);
5481 type = btrfs_file_extent_type(path->nodes[0], fi);
5482 if (type == BTRFS_FILE_EXTENT_INLINE) {
5483 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5484 path->slots[0], fi);
5485 extent_end = ALIGN(key.offset + size,
5486 sctx->send_root->fs_info->sectorsize);
5488 extent_end = key.offset +
5489 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5491 sctx->cur_inode_last_extent = extent_end;
5493 btrfs_free_path(path);
5497 static int range_is_hole_in_parent(struct send_ctx *sctx,
5501 struct btrfs_path *path;
5502 struct btrfs_key key;
5503 struct btrfs_root *root = sctx->parent_root;
5504 u64 search_start = start;
5507 path = alloc_path_for_send();
5511 key.objectid = sctx->cur_ino;
5512 key.type = BTRFS_EXTENT_DATA_KEY;
5513 key.offset = search_start;
5514 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5517 if (ret > 0 && path->slots[0] > 0)
5520 while (search_start < end) {
5521 struct extent_buffer *leaf = path->nodes[0];
5522 int slot = path->slots[0];
5523 struct btrfs_file_extent_item *fi;
5526 if (slot >= btrfs_header_nritems(leaf)) {
5527 ret = btrfs_next_leaf(root, path);
5535 btrfs_item_key_to_cpu(leaf, &key, slot);
5536 if (key.objectid < sctx->cur_ino ||
5537 key.type < BTRFS_EXTENT_DATA_KEY)
5539 if (key.objectid > sctx->cur_ino ||
5540 key.type > BTRFS_EXTENT_DATA_KEY ||
5544 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5545 if (btrfs_file_extent_type(leaf, fi) ==
5546 BTRFS_FILE_EXTENT_INLINE) {
5547 u64 size = btrfs_file_extent_inline_len(leaf, slot, fi);
5549 extent_end = ALIGN(key.offset + size,
5550 root->fs_info->sectorsize);
5552 extent_end = key.offset +
5553 btrfs_file_extent_num_bytes(leaf, fi);
5555 if (extent_end <= start)
5557 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5558 search_start = extent_end;
5568 btrfs_free_path(path);
5572 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5573 struct btrfs_key *key)
5575 struct btrfs_file_extent_item *fi;
5580 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5583 if (sctx->cur_inode_last_extent == (u64)-1) {
5584 ret = get_last_extent(sctx, key->offset - 1);
5589 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5590 struct btrfs_file_extent_item);
5591 type = btrfs_file_extent_type(path->nodes[0], fi);
5592 if (type == BTRFS_FILE_EXTENT_INLINE) {
5593 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5594 path->slots[0], fi);
5595 extent_end = ALIGN(key->offset + size,
5596 sctx->send_root->fs_info->sectorsize);
5598 extent_end = key->offset +
5599 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5602 if (path->slots[0] == 0 &&
5603 sctx->cur_inode_last_extent < key->offset) {
5605 * We might have skipped entire leafs that contained only
5606 * file extent items for our current inode. These leafs have
5607 * a generation number smaller (older) than the one in the
5608 * current leaf and the leaf our last extent came from, and
5609 * are located between these 2 leafs.
5611 ret = get_last_extent(sctx, key->offset - 1);
5616 if (sctx->cur_inode_last_extent < key->offset) {
5617 ret = range_is_hole_in_parent(sctx,
5618 sctx->cur_inode_last_extent,
5623 ret = send_hole(sctx, key->offset);
5627 sctx->cur_inode_last_extent = extent_end;
5631 static int process_extent(struct send_ctx *sctx,
5632 struct btrfs_path *path,
5633 struct btrfs_key *key)
5635 struct clone_root *found_clone = NULL;
5638 if (S_ISLNK(sctx->cur_inode_mode))
5641 if (sctx->parent_root && !sctx->cur_inode_new) {
5642 ret = is_extent_unchanged(sctx, path, key);
5650 struct btrfs_file_extent_item *ei;
5653 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5654 struct btrfs_file_extent_item);
5655 type = btrfs_file_extent_type(path->nodes[0], ei);
5656 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5657 type == BTRFS_FILE_EXTENT_REG) {
5659 * The send spec does not have a prealloc command yet,
5660 * so just leave a hole for prealloc'ed extents until
5661 * we have enough commands queued up to justify rev'ing
5664 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5669 /* Have a hole, just skip it. */
5670 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5677 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5678 sctx->cur_inode_size, &found_clone);
5679 if (ret != -ENOENT && ret < 0)
5682 ret = send_write_or_clone(sctx, path, key, found_clone);
5686 ret = maybe_send_hole(sctx, path, key);
5691 static int process_all_extents(struct send_ctx *sctx)
5694 struct btrfs_root *root;
5695 struct btrfs_path *path;
5696 struct btrfs_key key;
5697 struct btrfs_key found_key;
5698 struct extent_buffer *eb;
5701 root = sctx->send_root;
5702 path = alloc_path_for_send();
5706 key.objectid = sctx->cmp_key->objectid;
5707 key.type = BTRFS_EXTENT_DATA_KEY;
5709 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5714 eb = path->nodes[0];
5715 slot = path->slots[0];
5717 if (slot >= btrfs_header_nritems(eb)) {
5718 ret = btrfs_next_leaf(root, path);
5721 } else if (ret > 0) {
5728 btrfs_item_key_to_cpu(eb, &found_key, slot);
5730 if (found_key.objectid != key.objectid ||
5731 found_key.type != key.type) {
5736 ret = process_extent(sctx, path, &found_key);
5744 btrfs_free_path(path);
5748 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5750 int *refs_processed)
5754 if (sctx->cur_ino == 0)
5756 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5757 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5759 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5762 ret = process_recorded_refs(sctx, pending_move);
5766 *refs_processed = 1;
5771 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5782 int need_truncate = 1;
5783 int pending_move = 0;
5784 int refs_processed = 0;
5786 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5792 * We have processed the refs and thus need to advance send_progress.
5793 * Now, calls to get_cur_xxx will take the updated refs of the current
5794 * inode into account.
5796 * On the other hand, if our current inode is a directory and couldn't
5797 * be moved/renamed because its parent was renamed/moved too and it has
5798 * a higher inode number, we can only move/rename our current inode
5799 * after we moved/renamed its parent. Therefore in this case operate on
5800 * the old path (pre move/rename) of our current inode, and the
5801 * move/rename will be performed later.
5803 if (refs_processed && !pending_move)
5804 sctx->send_progress = sctx->cur_ino + 1;
5806 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5808 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5811 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5812 &left_mode, &left_uid, &left_gid, NULL);
5816 if (!sctx->parent_root || sctx->cur_inode_new) {
5818 if (!S_ISLNK(sctx->cur_inode_mode))
5820 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5825 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5826 &old_size, NULL, &right_mode, &right_uid,
5831 if (left_uid != right_uid || left_gid != right_gid)
5833 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5835 if ((old_size == sctx->cur_inode_size) ||
5836 (sctx->cur_inode_size > old_size &&
5837 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5841 if (S_ISREG(sctx->cur_inode_mode)) {
5842 if (need_send_hole(sctx)) {
5843 if (sctx->cur_inode_last_extent == (u64)-1 ||
5844 sctx->cur_inode_last_extent <
5845 sctx->cur_inode_size) {
5846 ret = get_last_extent(sctx, (u64)-1);
5850 if (sctx->cur_inode_last_extent <
5851 sctx->cur_inode_size) {
5852 ret = send_hole(sctx, sctx->cur_inode_size);
5857 if (need_truncate) {
5858 ret = send_truncate(sctx, sctx->cur_ino,
5859 sctx->cur_inode_gen,
5860 sctx->cur_inode_size);
5867 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5868 left_uid, left_gid);
5873 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5880 * If other directory inodes depended on our current directory
5881 * inode's move/rename, now do their move/rename operations.
5883 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5884 ret = apply_children_dir_moves(sctx);
5888 * Need to send that every time, no matter if it actually
5889 * changed between the two trees as we have done changes to
5890 * the inode before. If our inode is a directory and it's
5891 * waiting to be moved/renamed, we will send its utimes when
5892 * it's moved/renamed, therefore we don't need to do it here.
5894 sctx->send_progress = sctx->cur_ino + 1;
5895 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5904 static int changed_inode(struct send_ctx *sctx,
5905 enum btrfs_compare_tree_result result)
5908 struct btrfs_key *key = sctx->cmp_key;
5909 struct btrfs_inode_item *left_ii = NULL;
5910 struct btrfs_inode_item *right_ii = NULL;
5914 sctx->cur_ino = key->objectid;
5915 sctx->cur_inode_new_gen = 0;
5916 sctx->cur_inode_last_extent = (u64)-1;
5917 sctx->cur_inode_next_write_offset = 0;
5920 * Set send_progress to current inode. This will tell all get_cur_xxx
5921 * functions that the current inode's refs are not updated yet. Later,
5922 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5924 sctx->send_progress = sctx->cur_ino;
5926 if (result == BTRFS_COMPARE_TREE_NEW ||
5927 result == BTRFS_COMPARE_TREE_CHANGED) {
5928 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5929 sctx->left_path->slots[0],
5930 struct btrfs_inode_item);
5931 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5934 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5935 sctx->right_path->slots[0],
5936 struct btrfs_inode_item);
5937 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5940 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5941 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5942 sctx->right_path->slots[0],
5943 struct btrfs_inode_item);
5945 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5949 * The cur_ino = root dir case is special here. We can't treat
5950 * the inode as deleted+reused because it would generate a
5951 * stream that tries to delete/mkdir the root dir.
5953 if (left_gen != right_gen &&
5954 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5955 sctx->cur_inode_new_gen = 1;
5958 if (result == BTRFS_COMPARE_TREE_NEW) {
5959 sctx->cur_inode_gen = left_gen;
5960 sctx->cur_inode_new = 1;
5961 sctx->cur_inode_deleted = 0;
5962 sctx->cur_inode_size = btrfs_inode_size(
5963 sctx->left_path->nodes[0], left_ii);
5964 sctx->cur_inode_mode = btrfs_inode_mode(
5965 sctx->left_path->nodes[0], left_ii);
5966 sctx->cur_inode_rdev = btrfs_inode_rdev(
5967 sctx->left_path->nodes[0], left_ii);
5968 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5969 ret = send_create_inode_if_needed(sctx);
5970 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5971 sctx->cur_inode_gen = right_gen;
5972 sctx->cur_inode_new = 0;
5973 sctx->cur_inode_deleted = 1;
5974 sctx->cur_inode_size = btrfs_inode_size(
5975 sctx->right_path->nodes[0], right_ii);
5976 sctx->cur_inode_mode = btrfs_inode_mode(
5977 sctx->right_path->nodes[0], right_ii);
5978 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5980 * We need to do some special handling in case the inode was
5981 * reported as changed with a changed generation number. This
5982 * means that the original inode was deleted and new inode
5983 * reused the same inum. So we have to treat the old inode as
5984 * deleted and the new one as new.
5986 if (sctx->cur_inode_new_gen) {
5988 * First, process the inode as if it was deleted.
5990 sctx->cur_inode_gen = right_gen;
5991 sctx->cur_inode_new = 0;
5992 sctx->cur_inode_deleted = 1;
5993 sctx->cur_inode_size = btrfs_inode_size(
5994 sctx->right_path->nodes[0], right_ii);
5995 sctx->cur_inode_mode = btrfs_inode_mode(
5996 sctx->right_path->nodes[0], right_ii);
5997 ret = process_all_refs(sctx,
5998 BTRFS_COMPARE_TREE_DELETED);
6003 * Now process the inode as if it was new.
6005 sctx->cur_inode_gen = left_gen;
6006 sctx->cur_inode_new = 1;
6007 sctx->cur_inode_deleted = 0;
6008 sctx->cur_inode_size = btrfs_inode_size(
6009 sctx->left_path->nodes[0], left_ii);
6010 sctx->cur_inode_mode = btrfs_inode_mode(
6011 sctx->left_path->nodes[0], left_ii);
6012 sctx->cur_inode_rdev = btrfs_inode_rdev(
6013 sctx->left_path->nodes[0], left_ii);
6014 ret = send_create_inode_if_needed(sctx);
6018 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6022 * Advance send_progress now as we did not get into
6023 * process_recorded_refs_if_needed in the new_gen case.
6025 sctx->send_progress = sctx->cur_ino + 1;
6028 * Now process all extents and xattrs of the inode as if
6029 * they were all new.
6031 ret = process_all_extents(sctx);
6034 ret = process_all_new_xattrs(sctx);
6038 sctx->cur_inode_gen = left_gen;
6039 sctx->cur_inode_new = 0;
6040 sctx->cur_inode_new_gen = 0;
6041 sctx->cur_inode_deleted = 0;
6042 sctx->cur_inode_size = btrfs_inode_size(
6043 sctx->left_path->nodes[0], left_ii);
6044 sctx->cur_inode_mode = btrfs_inode_mode(
6045 sctx->left_path->nodes[0], left_ii);
6054 * We have to process new refs before deleted refs, but compare_trees gives us
6055 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6056 * first and later process them in process_recorded_refs.
6057 * For the cur_inode_new_gen case, we skip recording completely because
6058 * changed_inode did already initiate processing of refs. The reason for this is
6059 * that in this case, compare_tree actually compares the refs of 2 different
6060 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6061 * refs of the right tree as deleted and all refs of the left tree as new.
6063 static int changed_ref(struct send_ctx *sctx,
6064 enum btrfs_compare_tree_result result)
6068 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6069 inconsistent_snapshot_error(sctx, result, "reference");
6073 if (!sctx->cur_inode_new_gen &&
6074 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6075 if (result == BTRFS_COMPARE_TREE_NEW)
6076 ret = record_new_ref(sctx);
6077 else if (result == BTRFS_COMPARE_TREE_DELETED)
6078 ret = record_deleted_ref(sctx);
6079 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6080 ret = record_changed_ref(sctx);
6087 * Process new/deleted/changed xattrs. We skip processing in the
6088 * cur_inode_new_gen case because changed_inode did already initiate processing
6089 * of xattrs. The reason is the same as in changed_ref
6091 static int changed_xattr(struct send_ctx *sctx,
6092 enum btrfs_compare_tree_result result)
6096 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6097 inconsistent_snapshot_error(sctx, result, "xattr");
6101 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6102 if (result == BTRFS_COMPARE_TREE_NEW)
6103 ret = process_new_xattr(sctx);
6104 else if (result == BTRFS_COMPARE_TREE_DELETED)
6105 ret = process_deleted_xattr(sctx);
6106 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6107 ret = process_changed_xattr(sctx);
6114 * Process new/deleted/changed extents. We skip processing in the
6115 * cur_inode_new_gen case because changed_inode did already initiate processing
6116 * of extents. The reason is the same as in changed_ref
6118 static int changed_extent(struct send_ctx *sctx,
6119 enum btrfs_compare_tree_result result)
6123 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6125 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6126 struct extent_buffer *leaf_l;
6127 struct extent_buffer *leaf_r;
6128 struct btrfs_file_extent_item *ei_l;
6129 struct btrfs_file_extent_item *ei_r;
6131 leaf_l = sctx->left_path->nodes[0];
6132 leaf_r = sctx->right_path->nodes[0];
6133 ei_l = btrfs_item_ptr(leaf_l,
6134 sctx->left_path->slots[0],
6135 struct btrfs_file_extent_item);
6136 ei_r = btrfs_item_ptr(leaf_r,
6137 sctx->right_path->slots[0],
6138 struct btrfs_file_extent_item);
6141 * We may have found an extent item that has changed
6142 * only its disk_bytenr field and the corresponding
6143 * inode item was not updated. This case happens due to
6144 * very specific timings during relocation when a leaf
6145 * that contains file extent items is COWed while
6146 * relocation is ongoing and its in the stage where it
6147 * updates data pointers. So when this happens we can
6148 * safely ignore it since we know it's the same extent,
6149 * but just at different logical and physical locations
6150 * (when an extent is fully replaced with a new one, we
6151 * know the generation number must have changed too,
6152 * since snapshot creation implies committing the current
6153 * transaction, and the inode item must have been updated
6155 * This replacement of the disk_bytenr happens at
6156 * relocation.c:replace_file_extents() through
6157 * relocation.c:btrfs_reloc_cow_block().
6159 if (btrfs_file_extent_generation(leaf_l, ei_l) ==
6160 btrfs_file_extent_generation(leaf_r, ei_r) &&
6161 btrfs_file_extent_ram_bytes(leaf_l, ei_l) ==
6162 btrfs_file_extent_ram_bytes(leaf_r, ei_r) &&
6163 btrfs_file_extent_compression(leaf_l, ei_l) ==
6164 btrfs_file_extent_compression(leaf_r, ei_r) &&
6165 btrfs_file_extent_encryption(leaf_l, ei_l) ==
6166 btrfs_file_extent_encryption(leaf_r, ei_r) &&
6167 btrfs_file_extent_other_encoding(leaf_l, ei_l) ==
6168 btrfs_file_extent_other_encoding(leaf_r, ei_r) &&
6169 btrfs_file_extent_type(leaf_l, ei_l) ==
6170 btrfs_file_extent_type(leaf_r, ei_r) &&
6171 btrfs_file_extent_disk_bytenr(leaf_l, ei_l) !=
6172 btrfs_file_extent_disk_bytenr(leaf_r, ei_r) &&
6173 btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) ==
6174 btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) &&
6175 btrfs_file_extent_offset(leaf_l, ei_l) ==
6176 btrfs_file_extent_offset(leaf_r, ei_r) &&
6177 btrfs_file_extent_num_bytes(leaf_l, ei_l) ==
6178 btrfs_file_extent_num_bytes(leaf_r, ei_r))
6182 inconsistent_snapshot_error(sctx, result, "extent");
6186 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6187 if (result != BTRFS_COMPARE_TREE_DELETED)
6188 ret = process_extent(sctx, sctx->left_path,
6195 static int dir_changed(struct send_ctx *sctx, u64 dir)
6197 u64 orig_gen, new_gen;
6200 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6205 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6210 return (orig_gen != new_gen) ? 1 : 0;
6213 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6214 struct btrfs_key *key)
6216 struct btrfs_inode_extref *extref;
6217 struct extent_buffer *leaf;
6218 u64 dirid = 0, last_dirid = 0;
6225 /* Easy case, just check this one dirid */
6226 if (key->type == BTRFS_INODE_REF_KEY) {
6227 dirid = key->offset;
6229 ret = dir_changed(sctx, dirid);
6233 leaf = path->nodes[0];
6234 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6235 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6236 while (cur_offset < item_size) {
6237 extref = (struct btrfs_inode_extref *)(ptr +
6239 dirid = btrfs_inode_extref_parent(leaf, extref);
6240 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6241 cur_offset += ref_name_len + sizeof(*extref);
6242 if (dirid == last_dirid)
6244 ret = dir_changed(sctx, dirid);
6254 * Updates compare related fields in sctx and simply forwards to the actual
6255 * changed_xxx functions.
6257 static int changed_cb(struct btrfs_path *left_path,
6258 struct btrfs_path *right_path,
6259 struct btrfs_key *key,
6260 enum btrfs_compare_tree_result result,
6264 struct send_ctx *sctx = ctx;
6266 if (result == BTRFS_COMPARE_TREE_SAME) {
6267 if (key->type == BTRFS_INODE_REF_KEY ||
6268 key->type == BTRFS_INODE_EXTREF_KEY) {
6269 ret = compare_refs(sctx, left_path, key);
6274 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6275 return maybe_send_hole(sctx, left_path, key);
6279 result = BTRFS_COMPARE_TREE_CHANGED;
6283 sctx->left_path = left_path;
6284 sctx->right_path = right_path;
6285 sctx->cmp_key = key;
6287 ret = finish_inode_if_needed(sctx, 0);
6291 /* Ignore non-FS objects */
6292 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6293 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6296 if (key->type == BTRFS_INODE_ITEM_KEY)
6297 ret = changed_inode(sctx, result);
6298 else if (key->type == BTRFS_INODE_REF_KEY ||
6299 key->type == BTRFS_INODE_EXTREF_KEY)
6300 ret = changed_ref(sctx, result);
6301 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6302 ret = changed_xattr(sctx, result);
6303 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6304 ret = changed_extent(sctx, result);
6310 static int full_send_tree(struct send_ctx *sctx)
6313 struct btrfs_root *send_root = sctx->send_root;
6314 struct btrfs_key key;
6315 struct btrfs_key found_key;
6316 struct btrfs_path *path;
6317 struct extent_buffer *eb;
6320 path = alloc_path_for_send();
6324 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6325 key.type = BTRFS_INODE_ITEM_KEY;
6328 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6335 eb = path->nodes[0];
6336 slot = path->slots[0];
6337 btrfs_item_key_to_cpu(eb, &found_key, slot);
6339 ret = changed_cb(path, NULL, &found_key,
6340 BTRFS_COMPARE_TREE_NEW, sctx);
6344 key.objectid = found_key.objectid;
6345 key.type = found_key.type;
6346 key.offset = found_key.offset + 1;
6348 ret = btrfs_next_item(send_root, path);
6358 ret = finish_inode_if_needed(sctx, 1);
6361 btrfs_free_path(path);
6365 static int send_subvol(struct send_ctx *sctx)
6369 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6370 ret = send_header(sctx);
6375 ret = send_subvol_begin(sctx);
6379 if (sctx->parent_root) {
6380 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6384 ret = finish_inode_if_needed(sctx, 1);
6388 ret = full_send_tree(sctx);
6394 free_recorded_refs(sctx);
6399 * If orphan cleanup did remove any orphans from a root, it means the tree
6400 * was modified and therefore the commit root is not the same as the current
6401 * root anymore. This is a problem, because send uses the commit root and
6402 * therefore can see inode items that don't exist in the current root anymore,
6403 * and for example make calls to btrfs_iget, which will do tree lookups based
6404 * on the current root and not on the commit root. Those lookups will fail,
6405 * returning a -ESTALE error, and making send fail with that error. So make
6406 * sure a send does not see any orphans we have just removed, and that it will
6407 * see the same inodes regardless of whether a transaction commit happened
6408 * before it started (meaning that the commit root will be the same as the
6409 * current root) or not.
6411 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6414 struct btrfs_trans_handle *trans = NULL;
6417 if (sctx->parent_root &&
6418 sctx->parent_root->node != sctx->parent_root->commit_root)
6421 for (i = 0; i < sctx->clone_roots_cnt; i++)
6422 if (sctx->clone_roots[i].root->node !=
6423 sctx->clone_roots[i].root->commit_root)
6427 return btrfs_end_transaction(trans);
6432 /* Use any root, all fs roots will get their commit roots updated. */
6434 trans = btrfs_join_transaction(sctx->send_root);
6436 return PTR_ERR(trans);
6440 return btrfs_commit_transaction(trans);
6443 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6445 spin_lock(&root->root_item_lock);
6446 root->send_in_progress--;
6448 * Not much left to do, we don't know why it's unbalanced and
6449 * can't blindly reset it to 0.
6451 if (root->send_in_progress < 0)
6452 btrfs_err(root->fs_info,
6453 "send_in_progres unbalanced %d root %llu",
6454 root->send_in_progress, root->root_key.objectid);
6455 spin_unlock(&root->root_item_lock);
6458 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
6461 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
6462 struct btrfs_fs_info *fs_info = send_root->fs_info;
6463 struct btrfs_root *clone_root;
6464 struct btrfs_key key;
6465 struct send_ctx *sctx = NULL;
6467 u64 *clone_sources_tmp = NULL;
6468 int clone_sources_to_rollback = 0;
6469 unsigned alloc_size;
6470 int sort_clone_roots = 0;
6473 if (!capable(CAP_SYS_ADMIN))
6477 * The subvolume must remain read-only during send, protect against
6478 * making it RW. This also protects against deletion.
6480 spin_lock(&send_root->root_item_lock);
6481 send_root->send_in_progress++;
6482 spin_unlock(&send_root->root_item_lock);
6485 * This is done when we lookup the root, it should already be complete
6486 * by the time we get here.
6488 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6491 * Userspace tools do the checks and warn the user if it's
6494 if (!btrfs_root_readonly(send_root)) {
6500 * Check that we don't overflow at later allocations, we request
6501 * clone_sources_count + 1 items, and compare to unsigned long inside
6504 if (arg->clone_sources_count >
6505 ULONG_MAX / sizeof(struct clone_root) - 1) {
6510 if (!access_ok(VERIFY_READ, arg->clone_sources,
6511 sizeof(*arg->clone_sources) *
6512 arg->clone_sources_count)) {
6517 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
6522 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
6528 INIT_LIST_HEAD(&sctx->new_refs);
6529 INIT_LIST_HEAD(&sctx->deleted_refs);
6530 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
6531 INIT_LIST_HEAD(&sctx->name_cache_list);
6533 sctx->flags = arg->flags;
6535 sctx->send_filp = fget(arg->send_fd);
6536 if (!sctx->send_filp) {
6541 sctx->send_root = send_root;
6543 * Unlikely but possible, if the subvolume is marked for deletion but
6544 * is slow to remove the directory entry, send can still be started
6546 if (btrfs_root_dead(sctx->send_root)) {
6551 sctx->clone_roots_cnt = arg->clone_sources_count;
6553 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6554 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
6555 if (!sctx->send_buf) {
6560 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
6561 if (!sctx->read_buf) {
6566 sctx->pending_dir_moves = RB_ROOT;
6567 sctx->waiting_dir_moves = RB_ROOT;
6568 sctx->orphan_dirs = RB_ROOT;
6570 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6572 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
6573 if (!sctx->clone_roots) {
6578 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6580 if (arg->clone_sources_count) {
6581 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
6582 if (!clone_sources_tmp) {
6587 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
6594 for (i = 0; i < arg->clone_sources_count; i++) {
6595 key.objectid = clone_sources_tmp[i];
6596 key.type = BTRFS_ROOT_ITEM_KEY;
6597 key.offset = (u64)-1;
6599 index = srcu_read_lock(&fs_info->subvol_srcu);
6601 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
6602 if (IS_ERR(clone_root)) {
6603 srcu_read_unlock(&fs_info->subvol_srcu, index);
6604 ret = PTR_ERR(clone_root);
6607 spin_lock(&clone_root->root_item_lock);
6608 if (!btrfs_root_readonly(clone_root) ||
6609 btrfs_root_dead(clone_root)) {
6610 spin_unlock(&clone_root->root_item_lock);
6611 srcu_read_unlock(&fs_info->subvol_srcu, index);
6615 clone_root->send_in_progress++;
6616 spin_unlock(&clone_root->root_item_lock);
6617 srcu_read_unlock(&fs_info->subvol_srcu, index);
6619 sctx->clone_roots[i].root = clone_root;
6620 clone_sources_to_rollback = i + 1;
6622 kvfree(clone_sources_tmp);
6623 clone_sources_tmp = NULL;
6626 if (arg->parent_root) {
6627 key.objectid = arg->parent_root;
6628 key.type = BTRFS_ROOT_ITEM_KEY;
6629 key.offset = (u64)-1;
6631 index = srcu_read_lock(&fs_info->subvol_srcu);
6633 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
6634 if (IS_ERR(sctx->parent_root)) {
6635 srcu_read_unlock(&fs_info->subvol_srcu, index);
6636 ret = PTR_ERR(sctx->parent_root);
6640 spin_lock(&sctx->parent_root->root_item_lock);
6641 sctx->parent_root->send_in_progress++;
6642 if (!btrfs_root_readonly(sctx->parent_root) ||
6643 btrfs_root_dead(sctx->parent_root)) {
6644 spin_unlock(&sctx->parent_root->root_item_lock);
6645 srcu_read_unlock(&fs_info->subvol_srcu, index);
6649 spin_unlock(&sctx->parent_root->root_item_lock);
6651 srcu_read_unlock(&fs_info->subvol_srcu, index);
6655 * Clones from send_root are allowed, but only if the clone source
6656 * is behind the current send position. This is checked while searching
6657 * for possible clone sources.
6659 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6661 /* We do a bsearch later */
6662 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6663 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6665 sort_clone_roots = 1;
6667 ret = ensure_commit_roots_uptodate(sctx);
6671 current->journal_info = BTRFS_SEND_TRANS_STUB;
6672 ret = send_subvol(sctx);
6673 current->journal_info = NULL;
6677 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6678 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6681 ret = send_cmd(sctx);
6687 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6688 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6690 struct pending_dir_move *pm;
6692 n = rb_first(&sctx->pending_dir_moves);
6693 pm = rb_entry(n, struct pending_dir_move, node);
6694 while (!list_empty(&pm->list)) {
6695 struct pending_dir_move *pm2;
6697 pm2 = list_first_entry(&pm->list,
6698 struct pending_dir_move, list);
6699 free_pending_move(sctx, pm2);
6701 free_pending_move(sctx, pm);
6704 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6705 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6707 struct waiting_dir_move *dm;
6709 n = rb_first(&sctx->waiting_dir_moves);
6710 dm = rb_entry(n, struct waiting_dir_move, node);
6711 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6715 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6716 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6718 struct orphan_dir_info *odi;
6720 n = rb_first(&sctx->orphan_dirs);
6721 odi = rb_entry(n, struct orphan_dir_info, node);
6722 free_orphan_dir_info(sctx, odi);
6725 if (sort_clone_roots) {
6726 for (i = 0; i < sctx->clone_roots_cnt; i++)
6727 btrfs_root_dec_send_in_progress(
6728 sctx->clone_roots[i].root);
6730 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6731 btrfs_root_dec_send_in_progress(
6732 sctx->clone_roots[i].root);
6734 btrfs_root_dec_send_in_progress(send_root);
6736 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6737 btrfs_root_dec_send_in_progress(sctx->parent_root);
6739 kvfree(clone_sources_tmp);
6742 if (sctx->send_filp)
6743 fput(sctx->send_filp);
6745 kvfree(sctx->clone_roots);
6746 kvfree(sctx->send_buf);
6747 kvfree(sctx->read_buf);
6749 name_cache_free(sctx);