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 * Maximum number of references an extent can have in order for us to attempt to
29 * issue clone operations instead of write operations. This currently exists to
30 * avoid hitting limitations of the backreference walking code (taking a lot of
31 * time and using too much memory for extents with large number of references).
33 #define SEND_MAX_EXTENT_REFS 64
36 * A fs_path is a helper to dynamically build path names with unknown size.
37 * It reallocates the internal buffer on demand.
38 * It allows fast adding of path elements on the right side (normal path) and
39 * fast adding to the left side (reversed path). A reversed path can also be
40 * unreversed if needed.
49 unsigned short buf_len:15;
50 unsigned short reversed:1;
54 * Average path length does not exceed 200 bytes, we'll have
55 * better packing in the slab and higher chance to satisfy
56 * a allocation later during send.
61 #define FS_PATH_INLINE_SIZE \
62 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
65 /* reused for each extent */
67 struct btrfs_root *root;
74 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
75 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
78 struct file *send_filp;
84 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
85 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
87 struct btrfs_root *send_root;
88 struct btrfs_root *parent_root;
89 struct clone_root *clone_roots;
92 /* current state of the compare_tree call */
93 struct btrfs_path *left_path;
94 struct btrfs_path *right_path;
95 struct btrfs_key *cmp_key;
98 * infos of the currently processed inode. In case of deleted inodes,
99 * these are the values from the deleted inode.
104 int cur_inode_new_gen;
105 int cur_inode_deleted;
109 u64 cur_inode_last_extent;
110 u64 cur_inode_next_write_offset;
111 bool ignore_cur_inode;
115 struct list_head new_refs;
116 struct list_head deleted_refs;
118 struct radix_tree_root name_cache;
119 struct list_head name_cache_list;
122 struct file_ra_state ra;
127 * We process inodes by their increasing order, so if before an
128 * incremental send we reverse the parent/child relationship of
129 * directories such that a directory with a lower inode number was
130 * the parent of a directory with a higher inode number, and the one
131 * becoming the new parent got renamed too, we can't rename/move the
132 * directory with lower inode number when we finish processing it - we
133 * must process the directory with higher inode number first, then
134 * rename/move it and then rename/move the directory with lower inode
135 * number. Example follows.
137 * Tree state when the first send was performed:
149 * Tree state when the second (incremental) send is performed:
158 * The sequence of steps that lead to the second state was:
160 * mv /a/b/c/d /a/b/c2/d2
161 * mv /a/b/c /a/b/c2/d2/cc
163 * "c" has lower inode number, but we can't move it (2nd mv operation)
164 * before we move "d", which has higher inode number.
166 * So we just memorize which move/rename operations must be performed
167 * later when their respective parent is processed and moved/renamed.
170 /* Indexed by parent directory inode number. */
171 struct rb_root pending_dir_moves;
174 * Reverse index, indexed by the inode number of a directory that
175 * is waiting for the move/rename of its immediate parent before its
176 * own move/rename can be performed.
178 struct rb_root waiting_dir_moves;
181 * A directory that is going to be rm'ed might have a child directory
182 * which is in the pending directory moves index above. In this case,
183 * the directory can only be removed after the move/rename of its child
184 * is performed. Example:
204 * Sequence of steps that lead to the send snapshot:
205 * rm -f /a/b/c/foo.txt
207 * mv /a/b/c/x /a/b/YY
210 * When the child is processed, its move/rename is delayed until its
211 * parent is processed (as explained above), but all other operations
212 * like update utimes, chown, chgrp, etc, are performed and the paths
213 * that it uses for those operations must use the orphanized name of
214 * its parent (the directory we're going to rm later), so we need to
215 * memorize that name.
217 * Indexed by the inode number of the directory to be deleted.
219 struct rb_root orphan_dirs;
222 struct pending_dir_move {
224 struct list_head list;
228 struct list_head update_refs;
231 struct waiting_dir_move {
235 * There might be some directory that could not be removed because it
236 * was waiting for this directory inode to be moved first. Therefore
237 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
243 struct orphan_dir_info {
247 u64 last_dir_index_offset;
250 struct name_cache_entry {
251 struct list_head list;
253 * radix_tree has only 32bit entries but we need to handle 64bit inums.
254 * We use the lower 32bit of the 64bit inum to store it in the tree. If
255 * more then one inum would fall into the same entry, we use radix_list
256 * to store the additional entries. radix_list is also used to store
257 * entries where two entries have the same inum but different
260 struct list_head radix_list;
266 int need_later_update;
272 #define ADVANCE_ONLY_NEXT -1
274 enum btrfs_compare_tree_result {
275 BTRFS_COMPARE_TREE_NEW,
276 BTRFS_COMPARE_TREE_DELETED,
277 BTRFS_COMPARE_TREE_CHANGED,
278 BTRFS_COMPARE_TREE_SAME,
280 typedef int (*btrfs_changed_cb_t)(struct btrfs_path *left_path,
281 struct btrfs_path *right_path,
282 struct btrfs_key *key,
283 enum btrfs_compare_tree_result result,
287 static void inconsistent_snapshot_error(struct send_ctx *sctx,
288 enum btrfs_compare_tree_result result,
291 const char *result_string;
294 case BTRFS_COMPARE_TREE_NEW:
295 result_string = "new";
297 case BTRFS_COMPARE_TREE_DELETED:
298 result_string = "deleted";
300 case BTRFS_COMPARE_TREE_CHANGED:
301 result_string = "updated";
303 case BTRFS_COMPARE_TREE_SAME:
305 result_string = "unchanged";
309 result_string = "unexpected";
312 btrfs_err(sctx->send_root->fs_info,
313 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
314 result_string, what, sctx->cmp_key->objectid,
315 sctx->send_root->root_key.objectid,
317 sctx->parent_root->root_key.objectid : 0));
320 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
322 static struct waiting_dir_move *
323 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
325 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
327 static int need_send_hole(struct send_ctx *sctx)
329 return (sctx->parent_root && !sctx->cur_inode_new &&
330 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
331 S_ISREG(sctx->cur_inode_mode));
334 static void fs_path_reset(struct fs_path *p)
337 p->start = p->buf + p->buf_len - 1;
347 static struct fs_path *fs_path_alloc(void)
351 p = kmalloc(sizeof(*p), GFP_KERNEL);
355 p->buf = p->inline_buf;
356 p->buf_len = FS_PATH_INLINE_SIZE;
361 static struct fs_path *fs_path_alloc_reversed(void)
373 static void fs_path_free(struct fs_path *p)
377 if (p->buf != p->inline_buf)
382 static int fs_path_len(struct fs_path *p)
384 return p->end - p->start;
387 static int fs_path_ensure_buf(struct fs_path *p, int len)
395 if (p->buf_len >= len)
398 if (len > PATH_MAX) {
403 path_len = p->end - p->start;
404 old_buf_len = p->buf_len;
407 * First time the inline_buf does not suffice
409 if (p->buf == p->inline_buf) {
410 tmp_buf = kmalloc(len, GFP_KERNEL);
412 memcpy(tmp_buf, p->buf, old_buf_len);
414 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
420 * The real size of the buffer is bigger, this will let the fast path
421 * happen most of the time
423 p->buf_len = ksize(p->buf);
426 tmp_buf = p->buf + old_buf_len - path_len - 1;
427 p->end = p->buf + p->buf_len - 1;
428 p->start = p->end - path_len;
429 memmove(p->start, tmp_buf, path_len + 1);
432 p->end = p->start + path_len;
437 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
443 new_len = p->end - p->start + name_len;
444 if (p->start != p->end)
446 ret = fs_path_ensure_buf(p, new_len);
451 if (p->start != p->end)
453 p->start -= name_len;
454 *prepared = p->start;
456 if (p->start != p->end)
467 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
472 ret = fs_path_prepare_for_add(p, name_len, &prepared);
475 memcpy(prepared, name, name_len);
481 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
486 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
489 memcpy(prepared, p2->start, p2->end - p2->start);
495 static int fs_path_add_from_extent_buffer(struct fs_path *p,
496 struct extent_buffer *eb,
497 unsigned long off, int len)
502 ret = fs_path_prepare_for_add(p, len, &prepared);
506 read_extent_buffer(eb, prepared, off, len);
512 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
516 p->reversed = from->reversed;
519 ret = fs_path_add_path(p, from);
525 static void fs_path_unreverse(struct fs_path *p)
534 len = p->end - p->start;
536 p->end = p->start + len;
537 memmove(p->start, tmp, len + 1);
541 static struct btrfs_path *alloc_path_for_send(void)
543 struct btrfs_path *path;
545 path = btrfs_alloc_path();
548 path->search_commit_root = 1;
549 path->skip_locking = 1;
550 path->need_commit_sem = 1;
554 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
560 ret = kernel_write(filp, buf + pos, len - pos, off);
561 /* TODO handle that correctly */
562 /*if (ret == -ERESTARTSYS) {
576 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
578 struct btrfs_tlv_header *hdr;
579 int total_len = sizeof(*hdr) + len;
580 int left = sctx->send_max_size - sctx->send_size;
582 if (unlikely(left < total_len))
585 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
586 hdr->tlv_type = cpu_to_le16(attr);
587 hdr->tlv_len = cpu_to_le16(len);
588 memcpy(hdr + 1, data, len);
589 sctx->send_size += total_len;
594 #define TLV_PUT_DEFINE_INT(bits) \
595 static int tlv_put_u##bits(struct send_ctx *sctx, \
596 u##bits attr, u##bits value) \
598 __le##bits __tmp = cpu_to_le##bits(value); \
599 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
602 TLV_PUT_DEFINE_INT(64)
604 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
605 const char *str, int len)
609 return tlv_put(sctx, attr, str, len);
612 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
615 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
618 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
619 struct extent_buffer *eb,
620 struct btrfs_timespec *ts)
622 struct btrfs_timespec bts;
623 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
624 return tlv_put(sctx, attr, &bts, sizeof(bts));
628 #define TLV_PUT(sctx, attrtype, data, attrlen) \
630 ret = tlv_put(sctx, attrtype, data, attrlen); \
632 goto tlv_put_failure; \
635 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
637 ret = tlv_put_u##bits(sctx, attrtype, value); \
639 goto tlv_put_failure; \
642 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
643 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
644 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
645 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
646 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
648 ret = tlv_put_string(sctx, attrtype, str, len); \
650 goto tlv_put_failure; \
652 #define TLV_PUT_PATH(sctx, attrtype, p) \
654 ret = tlv_put_string(sctx, attrtype, p->start, \
655 p->end - p->start); \
657 goto tlv_put_failure; \
659 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
661 ret = tlv_put_uuid(sctx, attrtype, uuid); \
663 goto tlv_put_failure; \
665 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
667 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
669 goto tlv_put_failure; \
672 static int send_header(struct send_ctx *sctx)
674 struct btrfs_stream_header hdr;
676 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
677 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
679 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
684 * For each command/item we want to send to userspace, we call this function.
686 static int begin_cmd(struct send_ctx *sctx, int cmd)
688 struct btrfs_cmd_header *hdr;
690 if (WARN_ON(!sctx->send_buf))
693 BUG_ON(sctx->send_size);
695 sctx->send_size += sizeof(*hdr);
696 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
697 hdr->cmd = cpu_to_le16(cmd);
702 static int send_cmd(struct send_ctx *sctx)
705 struct btrfs_cmd_header *hdr;
708 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
709 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
712 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
713 hdr->crc = cpu_to_le32(crc);
715 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
718 sctx->total_send_size += sctx->send_size;
719 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
726 * Sends a move instruction to user space
728 static int send_rename(struct send_ctx *sctx,
729 struct fs_path *from, struct fs_path *to)
731 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
734 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
736 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
740 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
741 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
743 ret = send_cmd(sctx);
751 * Sends a link instruction to user space
753 static int send_link(struct send_ctx *sctx,
754 struct fs_path *path, struct fs_path *lnk)
756 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
759 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
761 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
765 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
766 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
768 ret = send_cmd(sctx);
776 * Sends an unlink instruction to user space
778 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
780 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
783 btrfs_debug(fs_info, "send_unlink %s", path->start);
785 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
789 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
791 ret = send_cmd(sctx);
799 * Sends a rmdir instruction to user space
801 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
803 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
806 btrfs_debug(fs_info, "send_rmdir %s", path->start);
808 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
812 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
814 ret = send_cmd(sctx);
822 * Helper function to retrieve some fields from an inode item.
824 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
825 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
829 struct btrfs_inode_item *ii;
830 struct btrfs_key key;
833 key.type = BTRFS_INODE_ITEM_KEY;
835 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
842 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
843 struct btrfs_inode_item);
845 *size = btrfs_inode_size(path->nodes[0], ii);
847 *gen = btrfs_inode_generation(path->nodes[0], ii);
849 *mode = btrfs_inode_mode(path->nodes[0], ii);
851 *uid = btrfs_inode_uid(path->nodes[0], ii);
853 *gid = btrfs_inode_gid(path->nodes[0], ii);
855 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
860 static int get_inode_info(struct btrfs_root *root,
861 u64 ino, u64 *size, u64 *gen,
862 u64 *mode, u64 *uid, u64 *gid,
865 struct btrfs_path *path;
868 path = alloc_path_for_send();
871 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
873 btrfs_free_path(path);
877 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
882 * Helper function to iterate the entries in ONE btrfs_inode_ref or
883 * btrfs_inode_extref.
884 * The iterate callback may return a non zero value to stop iteration. This can
885 * be a negative value for error codes or 1 to simply stop it.
887 * path must point to the INODE_REF or INODE_EXTREF when called.
889 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
890 struct btrfs_key *found_key, int resolve,
891 iterate_inode_ref_t iterate, void *ctx)
893 struct extent_buffer *eb = path->nodes[0];
894 struct btrfs_item *item;
895 struct btrfs_inode_ref *iref;
896 struct btrfs_inode_extref *extref;
897 struct btrfs_path *tmp_path;
901 int slot = path->slots[0];
908 unsigned long name_off;
909 unsigned long elem_size;
912 p = fs_path_alloc_reversed();
916 tmp_path = alloc_path_for_send();
923 if (found_key->type == BTRFS_INODE_REF_KEY) {
924 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
925 struct btrfs_inode_ref);
926 item = btrfs_item_nr(slot);
927 total = btrfs_item_size(eb, item);
928 elem_size = sizeof(*iref);
930 ptr = btrfs_item_ptr_offset(eb, slot);
931 total = btrfs_item_size_nr(eb, slot);
932 elem_size = sizeof(*extref);
935 while (cur < total) {
938 if (found_key->type == BTRFS_INODE_REF_KEY) {
939 iref = (struct btrfs_inode_ref *)(ptr + cur);
940 name_len = btrfs_inode_ref_name_len(eb, iref);
941 name_off = (unsigned long)(iref + 1);
942 index = btrfs_inode_ref_index(eb, iref);
943 dir = found_key->offset;
945 extref = (struct btrfs_inode_extref *)(ptr + cur);
946 name_len = btrfs_inode_extref_name_len(eb, extref);
947 name_off = (unsigned long)&extref->name;
948 index = btrfs_inode_extref_index(eb, extref);
949 dir = btrfs_inode_extref_parent(eb, extref);
953 start = btrfs_ref_to_path(root, tmp_path, name_len,
957 ret = PTR_ERR(start);
960 if (start < p->buf) {
961 /* overflow , try again with larger buffer */
962 ret = fs_path_ensure_buf(p,
963 p->buf_len + p->buf - start);
966 start = btrfs_ref_to_path(root, tmp_path,
971 ret = PTR_ERR(start);
974 BUG_ON(start < p->buf);
978 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
984 cur += elem_size + name_len;
985 ret = iterate(num, dir, index, p, ctx);
992 btrfs_free_path(tmp_path);
997 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
998 const char *name, int name_len,
999 const char *data, int data_len,
1000 u8 type, void *ctx);
1003 * Helper function to iterate the entries in ONE btrfs_dir_item.
1004 * The iterate callback may return a non zero value to stop iteration. This can
1005 * be a negative value for error codes or 1 to simply stop it.
1007 * path must point to the dir item when called.
1009 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
1010 iterate_dir_item_t iterate, void *ctx)
1013 struct extent_buffer *eb;
1014 struct btrfs_item *item;
1015 struct btrfs_dir_item *di;
1016 struct btrfs_key di_key;
1029 * Start with a small buffer (1 page). If later we end up needing more
1030 * space, which can happen for xattrs on a fs with a leaf size greater
1031 * then the page size, attempt to increase the buffer. Typically xattr
1035 buf = kmalloc(buf_len, GFP_KERNEL);
1041 eb = path->nodes[0];
1042 slot = path->slots[0];
1043 item = btrfs_item_nr(slot);
1044 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1047 total = btrfs_item_size(eb, item);
1050 while (cur < total) {
1051 name_len = btrfs_dir_name_len(eb, di);
1052 data_len = btrfs_dir_data_len(eb, di);
1053 type = btrfs_dir_type(eb, di);
1054 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1056 if (type == BTRFS_FT_XATTR) {
1057 if (name_len > XATTR_NAME_MAX) {
1058 ret = -ENAMETOOLONG;
1061 if (name_len + data_len >
1062 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
1070 if (name_len + data_len > PATH_MAX) {
1071 ret = -ENAMETOOLONG;
1076 if (name_len + data_len > buf_len) {
1077 buf_len = name_len + data_len;
1078 if (is_vmalloc_addr(buf)) {
1082 char *tmp = krealloc(buf, buf_len,
1083 GFP_KERNEL | __GFP_NOWARN);
1090 buf = kvmalloc(buf_len, GFP_KERNEL);
1098 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1099 name_len + data_len);
1101 len = sizeof(*di) + name_len + data_len;
1102 di = (struct btrfs_dir_item *)((char *)di + len);
1105 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1106 data_len, type, ctx);
1122 static int __copy_first_ref(int num, u64 dir, int index,
1123 struct fs_path *p, void *ctx)
1126 struct fs_path *pt = ctx;
1128 ret = fs_path_copy(pt, p);
1132 /* we want the first only */
1137 * Retrieve the first path of an inode. If an inode has more then one
1138 * ref/hardlink, this is ignored.
1140 static int get_inode_path(struct btrfs_root *root,
1141 u64 ino, struct fs_path *path)
1144 struct btrfs_key key, found_key;
1145 struct btrfs_path *p;
1147 p = alloc_path_for_send();
1151 fs_path_reset(path);
1154 key.type = BTRFS_INODE_REF_KEY;
1157 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1164 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1165 if (found_key.objectid != ino ||
1166 (found_key.type != BTRFS_INODE_REF_KEY &&
1167 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1172 ret = iterate_inode_ref(root, p, &found_key, 1,
1173 __copy_first_ref, path);
1183 struct backref_ctx {
1184 struct send_ctx *sctx;
1186 /* number of total found references */
1190 * used for clones found in send_root. clones found behind cur_objectid
1191 * and cur_offset are not considered as allowed clones.
1196 /* may be truncated in case it's the last extent in a file */
1199 /* data offset in the file extent item */
1202 /* Just to check for bugs in backref resolving */
1206 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1208 u64 root = (u64)(uintptr_t)key;
1209 struct clone_root *cr = (struct clone_root *)elt;
1211 if (root < cr->root->root_key.objectid)
1213 if (root > cr->root->root_key.objectid)
1218 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1220 struct clone_root *cr1 = (struct clone_root *)e1;
1221 struct clone_root *cr2 = (struct clone_root *)e2;
1223 if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
1225 if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
1231 * Called for every backref that is found for the current extent.
1232 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1234 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1236 struct backref_ctx *bctx = ctx_;
1237 struct clone_root *found;
1239 /* First check if the root is in the list of accepted clone sources */
1240 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1241 bctx->sctx->clone_roots_cnt,
1242 sizeof(struct clone_root),
1243 __clone_root_cmp_bsearch);
1247 if (found->root == bctx->sctx->send_root &&
1248 ino == bctx->cur_objectid &&
1249 offset == bctx->cur_offset) {
1250 bctx->found_itself = 1;
1254 * Make sure we don't consider clones from send_root that are
1255 * behind the current inode/offset.
1257 if (found->root == bctx->sctx->send_root) {
1259 * If the source inode was not yet processed we can't issue a
1260 * clone operation, as the source extent does not exist yet at
1261 * the destination of the stream.
1263 if (ino > bctx->cur_objectid)
1266 * We clone from the inode currently being sent as long as the
1267 * source extent is already processed, otherwise we could try
1268 * to clone from an extent that does not exist yet at the
1269 * destination of the stream.
1271 if (ino == bctx->cur_objectid &&
1272 offset + bctx->extent_len >
1273 bctx->sctx->cur_inode_next_write_offset)
1278 found->found_refs++;
1279 if (ino < found->ino) {
1281 found->offset = offset;
1282 } else if (found->ino == ino) {
1284 * same extent found more then once in the same file.
1286 if (found->offset > offset + bctx->extent_len)
1287 found->offset = offset;
1294 * Given an inode, offset and extent item, it finds a good clone for a clone
1295 * instruction. Returns -ENOENT when none could be found. The function makes
1296 * sure that the returned clone is usable at the point where sending is at the
1297 * moment. This means, that no clones are accepted which lie behind the current
1300 * path must point to the extent item when called.
1302 static int find_extent_clone(struct send_ctx *sctx,
1303 struct btrfs_path *path,
1304 u64 ino, u64 data_offset,
1306 struct clone_root **found)
1308 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1314 u64 extent_item_pos;
1316 struct btrfs_file_extent_item *fi;
1317 struct extent_buffer *eb = path->nodes[0];
1318 struct backref_ctx *backref_ctx = NULL;
1319 struct clone_root *cur_clone_root;
1320 struct btrfs_key found_key;
1321 struct btrfs_path *tmp_path;
1322 struct btrfs_extent_item *ei;
1326 tmp_path = alloc_path_for_send();
1330 /* We only use this path under the commit sem */
1331 tmp_path->need_commit_sem = 0;
1333 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1339 if (data_offset >= ino_size) {
1341 * There may be extents that lie behind the file's size.
1342 * I at least had this in combination with snapshotting while
1343 * writing large files.
1349 fi = btrfs_item_ptr(eb, path->slots[0],
1350 struct btrfs_file_extent_item);
1351 extent_type = btrfs_file_extent_type(eb, fi);
1352 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1356 compressed = btrfs_file_extent_compression(eb, fi);
1358 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1359 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1360 if (disk_byte == 0) {
1364 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1366 down_read(&fs_info->commit_root_sem);
1367 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1368 &found_key, &flags);
1369 up_read(&fs_info->commit_root_sem);
1373 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1378 ei = btrfs_item_ptr(tmp_path->nodes[0], tmp_path->slots[0],
1379 struct btrfs_extent_item);
1381 * Backreference walking (iterate_extent_inodes() below) is currently
1382 * too expensive when an extent has a large number of references, both
1383 * in time spent and used memory. So for now just fallback to write
1384 * operations instead of clone operations when an extent has more than
1385 * a certain amount of references.
1387 if (btrfs_extent_refs(tmp_path->nodes[0], ei) > SEND_MAX_EXTENT_REFS) {
1391 btrfs_release_path(tmp_path);
1394 * Setup the clone roots.
1396 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1397 cur_clone_root = sctx->clone_roots + i;
1398 cur_clone_root->ino = (u64)-1;
1399 cur_clone_root->offset = 0;
1400 cur_clone_root->found_refs = 0;
1403 backref_ctx->sctx = sctx;
1404 backref_ctx->found = 0;
1405 backref_ctx->cur_objectid = ino;
1406 backref_ctx->cur_offset = data_offset;
1407 backref_ctx->found_itself = 0;
1408 backref_ctx->extent_len = num_bytes;
1410 * For non-compressed extents iterate_extent_inodes() gives us extent
1411 * offsets that already take into account the data offset, but not for
1412 * compressed extents, since the offset is logical and not relative to
1413 * the physical extent locations. We must take this into account to
1414 * avoid sending clone offsets that go beyond the source file's size,
1415 * which would result in the clone ioctl failing with -EINVAL on the
1418 if (compressed == BTRFS_COMPRESS_NONE)
1419 backref_ctx->data_offset = 0;
1421 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1424 * The last extent of a file may be too large due to page alignment.
1425 * We need to adjust extent_len in this case so that the checks in
1426 * __iterate_backrefs work.
1428 if (data_offset + num_bytes >= ino_size)
1429 backref_ctx->extent_len = ino_size - data_offset;
1432 * Now collect all backrefs.
1434 if (compressed == BTRFS_COMPRESS_NONE)
1435 extent_item_pos = logical - found_key.objectid;
1437 extent_item_pos = 0;
1438 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1439 extent_item_pos, 1, __iterate_backrefs,
1440 backref_ctx, false);
1445 if (!backref_ctx->found_itself) {
1446 /* found a bug in backref code? */
1449 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1450 ino, data_offset, disk_byte, found_key.objectid);
1454 btrfs_debug(fs_info,
1455 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1456 data_offset, ino, num_bytes, logical);
1458 if (!backref_ctx->found)
1459 btrfs_debug(fs_info, "no clones found");
1461 cur_clone_root = NULL;
1462 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1463 if (sctx->clone_roots[i].found_refs) {
1464 if (!cur_clone_root)
1465 cur_clone_root = sctx->clone_roots + i;
1466 else if (sctx->clone_roots[i].root == sctx->send_root)
1467 /* prefer clones from send_root over others */
1468 cur_clone_root = sctx->clone_roots + i;
1473 if (cur_clone_root) {
1474 *found = cur_clone_root;
1481 btrfs_free_path(tmp_path);
1486 static int read_symlink(struct btrfs_root *root,
1488 struct fs_path *dest)
1491 struct btrfs_path *path;
1492 struct btrfs_key key;
1493 struct btrfs_file_extent_item *ei;
1499 path = alloc_path_for_send();
1504 key.type = BTRFS_EXTENT_DATA_KEY;
1506 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1511 * An empty symlink inode. Can happen in rare error paths when
1512 * creating a symlink (transaction committed before the inode
1513 * eviction handler removed the symlink inode items and a crash
1514 * happened in between or the subvol was snapshoted in between).
1515 * Print an informative message to dmesg/syslog so that the user
1516 * can delete the symlink.
1518 btrfs_err(root->fs_info,
1519 "Found empty symlink inode %llu at root %llu",
1520 ino, root->root_key.objectid);
1525 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1526 struct btrfs_file_extent_item);
1527 type = btrfs_file_extent_type(path->nodes[0], ei);
1528 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1529 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1530 BUG_ON(compression);
1532 off = btrfs_file_extent_inline_start(ei);
1533 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
1535 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1538 btrfs_free_path(path);
1543 * Helper function to generate a file name that is unique in the root of
1544 * send_root and parent_root. This is used to generate names for orphan inodes.
1546 static int gen_unique_name(struct send_ctx *sctx,
1548 struct fs_path *dest)
1551 struct btrfs_path *path;
1552 struct btrfs_dir_item *di;
1557 path = alloc_path_for_send();
1562 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1564 ASSERT(len < sizeof(tmp));
1566 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1567 path, BTRFS_FIRST_FREE_OBJECTID,
1568 tmp, strlen(tmp), 0);
1569 btrfs_release_path(path);
1575 /* not unique, try again */
1580 if (!sctx->parent_root) {
1586 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1587 path, BTRFS_FIRST_FREE_OBJECTID,
1588 tmp, strlen(tmp), 0);
1589 btrfs_release_path(path);
1595 /* not unique, try again */
1603 ret = fs_path_add(dest, tmp, strlen(tmp));
1606 btrfs_free_path(path);
1611 inode_state_no_change,
1612 inode_state_will_create,
1613 inode_state_did_create,
1614 inode_state_will_delete,
1615 inode_state_did_delete,
1618 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1626 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1628 if (ret < 0 && ret != -ENOENT)
1632 if (!sctx->parent_root) {
1633 right_ret = -ENOENT;
1635 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1636 NULL, NULL, NULL, NULL);
1637 if (ret < 0 && ret != -ENOENT)
1642 if (!left_ret && !right_ret) {
1643 if (left_gen == gen && right_gen == gen) {
1644 ret = inode_state_no_change;
1645 } else if (left_gen == gen) {
1646 if (ino < sctx->send_progress)
1647 ret = inode_state_did_create;
1649 ret = inode_state_will_create;
1650 } else if (right_gen == gen) {
1651 if (ino < sctx->send_progress)
1652 ret = inode_state_did_delete;
1654 ret = inode_state_will_delete;
1658 } else if (!left_ret) {
1659 if (left_gen == gen) {
1660 if (ino < sctx->send_progress)
1661 ret = inode_state_did_create;
1663 ret = inode_state_will_create;
1667 } else if (!right_ret) {
1668 if (right_gen == gen) {
1669 if (ino < sctx->send_progress)
1670 ret = inode_state_did_delete;
1672 ret = inode_state_will_delete;
1684 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1688 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1691 ret = get_cur_inode_state(sctx, ino, gen);
1695 if (ret == inode_state_no_change ||
1696 ret == inode_state_did_create ||
1697 ret == inode_state_will_delete)
1707 * Helper function to lookup a dir item in a dir.
1709 static int lookup_dir_item_inode(struct btrfs_root *root,
1710 u64 dir, const char *name, int name_len,
1715 struct btrfs_dir_item *di;
1716 struct btrfs_key key;
1717 struct btrfs_path *path;
1719 path = alloc_path_for_send();
1723 di = btrfs_lookup_dir_item(NULL, root, path,
1724 dir, name, name_len, 0);
1725 if (IS_ERR_OR_NULL(di)) {
1726 ret = di ? PTR_ERR(di) : -ENOENT;
1729 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1730 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1734 *found_inode = key.objectid;
1735 *found_type = btrfs_dir_type(path->nodes[0], di);
1738 btrfs_free_path(path);
1743 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1744 * generation of the parent dir and the name of the dir entry.
1746 static int get_first_ref(struct btrfs_root *root, u64 ino,
1747 u64 *dir, u64 *dir_gen, struct fs_path *name)
1750 struct btrfs_key key;
1751 struct btrfs_key found_key;
1752 struct btrfs_path *path;
1756 path = alloc_path_for_send();
1761 key.type = BTRFS_INODE_REF_KEY;
1764 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1768 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1770 if (ret || found_key.objectid != ino ||
1771 (found_key.type != BTRFS_INODE_REF_KEY &&
1772 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1777 if (found_key.type == BTRFS_INODE_REF_KEY) {
1778 struct btrfs_inode_ref *iref;
1779 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1780 struct btrfs_inode_ref);
1781 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1782 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1783 (unsigned long)(iref + 1),
1785 parent_dir = found_key.offset;
1787 struct btrfs_inode_extref *extref;
1788 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1789 struct btrfs_inode_extref);
1790 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1791 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1792 (unsigned long)&extref->name, len);
1793 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1797 btrfs_release_path(path);
1800 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1809 btrfs_free_path(path);
1813 static int is_first_ref(struct btrfs_root *root,
1815 const char *name, int name_len)
1818 struct fs_path *tmp_name;
1821 tmp_name = fs_path_alloc();
1825 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1829 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1834 ret = !memcmp(tmp_name->start, name, name_len);
1837 fs_path_free(tmp_name);
1842 * Used by process_recorded_refs to determine if a new ref would overwrite an
1843 * already existing ref. In case it detects an overwrite, it returns the
1844 * inode/gen in who_ino/who_gen.
1845 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1846 * to make sure later references to the overwritten inode are possible.
1847 * Orphanizing is however only required for the first ref of an inode.
1848 * process_recorded_refs does an additional is_first_ref check to see if
1849 * orphanizing is really required.
1851 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1852 const char *name, int name_len,
1853 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1857 u64 other_inode = 0;
1860 if (!sctx->parent_root)
1863 ret = is_inode_existent(sctx, dir, dir_gen);
1868 * If we have a parent root we need to verify that the parent dir was
1869 * not deleted and then re-created, if it was then we have no overwrite
1870 * and we can just unlink this entry.
1872 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1873 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1875 if (ret < 0 && ret != -ENOENT)
1885 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1886 &other_inode, &other_type);
1887 if (ret < 0 && ret != -ENOENT)
1895 * Check if the overwritten ref was already processed. If yes, the ref
1896 * was already unlinked/moved, so we can safely assume that we will not
1897 * overwrite anything at this point in time.
1899 if (other_inode > sctx->send_progress ||
1900 is_waiting_for_move(sctx, other_inode)) {
1901 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1902 who_gen, who_mode, NULL, NULL, NULL);
1907 *who_ino = other_inode;
1917 * Checks if the ref was overwritten by an already processed inode. This is
1918 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1919 * thus the orphan name needs be used.
1920 * process_recorded_refs also uses it to avoid unlinking of refs that were
1923 static int did_overwrite_ref(struct send_ctx *sctx,
1924 u64 dir, u64 dir_gen,
1925 u64 ino, u64 ino_gen,
1926 const char *name, int name_len)
1933 if (!sctx->parent_root)
1936 ret = is_inode_existent(sctx, dir, dir_gen);
1940 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1941 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1943 if (ret < 0 && ret != -ENOENT)
1953 /* check if the ref was overwritten by another ref */
1954 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1955 &ow_inode, &other_type);
1956 if (ret < 0 && ret != -ENOENT)
1959 /* was never and will never be overwritten */
1964 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1969 if (ow_inode == ino && gen == ino_gen) {
1975 * We know that it is or will be overwritten. Check this now.
1976 * The current inode being processed might have been the one that caused
1977 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1978 * the current inode being processed.
1980 if ((ow_inode < sctx->send_progress) ||
1981 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1982 gen == sctx->cur_inode_gen))
1992 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1993 * that got overwritten. This is used by process_recorded_refs to determine
1994 * if it has to use the path as returned by get_cur_path or the orphan name.
1996 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1999 struct fs_path *name = NULL;
2003 if (!sctx->parent_root)
2006 name = fs_path_alloc();
2010 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
2014 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
2015 name->start, fs_path_len(name));
2023 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2024 * so we need to do some special handling in case we have clashes. This function
2025 * takes care of this with the help of name_cache_entry::radix_list.
2026 * In case of error, nce is kfreed.
2028 static int name_cache_insert(struct send_ctx *sctx,
2029 struct name_cache_entry *nce)
2032 struct list_head *nce_head;
2034 nce_head = radix_tree_lookup(&sctx->name_cache,
2035 (unsigned long)nce->ino);
2037 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2042 INIT_LIST_HEAD(nce_head);
2044 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2051 list_add_tail(&nce->radix_list, nce_head);
2052 list_add_tail(&nce->list, &sctx->name_cache_list);
2053 sctx->name_cache_size++;
2058 static void name_cache_delete(struct send_ctx *sctx,
2059 struct name_cache_entry *nce)
2061 struct list_head *nce_head;
2063 nce_head = radix_tree_lookup(&sctx->name_cache,
2064 (unsigned long)nce->ino);
2066 btrfs_err(sctx->send_root->fs_info,
2067 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2068 nce->ino, sctx->name_cache_size);
2071 list_del(&nce->radix_list);
2072 list_del(&nce->list);
2073 sctx->name_cache_size--;
2076 * We may not get to the final release of nce_head if the lookup fails
2078 if (nce_head && list_empty(nce_head)) {
2079 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2084 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2087 struct list_head *nce_head;
2088 struct name_cache_entry *cur;
2090 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2094 list_for_each_entry(cur, nce_head, radix_list) {
2095 if (cur->ino == ino && cur->gen == gen)
2102 * Removes the entry from the list and adds it back to the end. This marks the
2103 * entry as recently used so that name_cache_clean_unused does not remove it.
2105 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2107 list_del(&nce->list);
2108 list_add_tail(&nce->list, &sctx->name_cache_list);
2112 * Remove some entries from the beginning of name_cache_list.
2114 static void name_cache_clean_unused(struct send_ctx *sctx)
2116 struct name_cache_entry *nce;
2118 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2121 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2122 nce = list_entry(sctx->name_cache_list.next,
2123 struct name_cache_entry, list);
2124 name_cache_delete(sctx, nce);
2129 static void name_cache_free(struct send_ctx *sctx)
2131 struct name_cache_entry *nce;
2133 while (!list_empty(&sctx->name_cache_list)) {
2134 nce = list_entry(sctx->name_cache_list.next,
2135 struct name_cache_entry, list);
2136 name_cache_delete(sctx, nce);
2142 * Used by get_cur_path for each ref up to the root.
2143 * Returns 0 if it succeeded.
2144 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2145 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2146 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2147 * Returns <0 in case of error.
2149 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2153 struct fs_path *dest)
2157 struct name_cache_entry *nce = NULL;
2160 * First check if we already did a call to this function with the same
2161 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2162 * return the cached result.
2164 nce = name_cache_search(sctx, ino, gen);
2166 if (ino < sctx->send_progress && nce->need_later_update) {
2167 name_cache_delete(sctx, nce);
2171 name_cache_used(sctx, nce);
2172 *parent_ino = nce->parent_ino;
2173 *parent_gen = nce->parent_gen;
2174 ret = fs_path_add(dest, nce->name, nce->name_len);
2183 * If the inode is not existent yet, add the orphan name and return 1.
2184 * This should only happen for the parent dir that we determine in
2187 ret = is_inode_existent(sctx, ino, gen);
2192 ret = gen_unique_name(sctx, ino, gen, dest);
2200 * Depending on whether the inode was already processed or not, use
2201 * send_root or parent_root for ref lookup.
2203 if (ino < sctx->send_progress)
2204 ret = get_first_ref(sctx->send_root, ino,
2205 parent_ino, parent_gen, dest);
2207 ret = get_first_ref(sctx->parent_root, ino,
2208 parent_ino, parent_gen, dest);
2213 * Check if the ref was overwritten by an inode's ref that was processed
2214 * earlier. If yes, treat as orphan and return 1.
2216 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2217 dest->start, dest->end - dest->start);
2221 fs_path_reset(dest);
2222 ret = gen_unique_name(sctx, ino, gen, dest);
2230 * Store the result of the lookup in the name cache.
2232 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2240 nce->parent_ino = *parent_ino;
2241 nce->parent_gen = *parent_gen;
2242 nce->name_len = fs_path_len(dest);
2244 strcpy(nce->name, dest->start);
2246 if (ino < sctx->send_progress)
2247 nce->need_later_update = 0;
2249 nce->need_later_update = 1;
2251 nce_ret = name_cache_insert(sctx, nce);
2254 name_cache_clean_unused(sctx);
2261 * Magic happens here. This function returns the first ref to an inode as it
2262 * would look like while receiving the stream at this point in time.
2263 * We walk the path up to the root. For every inode in between, we check if it
2264 * was already processed/sent. If yes, we continue with the parent as found
2265 * in send_root. If not, we continue with the parent as found in parent_root.
2266 * If we encounter an inode that was deleted at this point in time, we use the
2267 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2268 * that were not created yet and overwritten inodes/refs.
2270 * When do we have orphan inodes:
2271 * 1. When an inode is freshly created and thus no valid refs are available yet
2272 * 2. When a directory lost all it's refs (deleted) but still has dir items
2273 * inside which were not processed yet (pending for move/delete). If anyone
2274 * tried to get the path to the dir items, it would get a path inside that
2276 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2277 * of an unprocessed inode. If in that case the first ref would be
2278 * overwritten, the overwritten inode gets "orphanized". Later when we
2279 * process this overwritten inode, it is restored at a new place by moving
2282 * sctx->send_progress tells this function at which point in time receiving
2285 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2286 struct fs_path *dest)
2289 struct fs_path *name = NULL;
2290 u64 parent_inode = 0;
2294 name = fs_path_alloc();
2301 fs_path_reset(dest);
2303 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2304 struct waiting_dir_move *wdm;
2306 fs_path_reset(name);
2308 if (is_waiting_for_rm(sctx, ino)) {
2309 ret = gen_unique_name(sctx, ino, gen, name);
2312 ret = fs_path_add_path(dest, name);
2316 wdm = get_waiting_dir_move(sctx, ino);
2317 if (wdm && wdm->orphanized) {
2318 ret = gen_unique_name(sctx, ino, gen, name);
2321 ret = get_first_ref(sctx->parent_root, ino,
2322 &parent_inode, &parent_gen, name);
2324 ret = __get_cur_name_and_parent(sctx, ino, gen,
2334 ret = fs_path_add_path(dest, name);
2345 fs_path_unreverse(dest);
2350 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2352 static int send_subvol_begin(struct send_ctx *sctx)
2355 struct btrfs_root *send_root = sctx->send_root;
2356 struct btrfs_root *parent_root = sctx->parent_root;
2357 struct btrfs_path *path;
2358 struct btrfs_key key;
2359 struct btrfs_root_ref *ref;
2360 struct extent_buffer *leaf;
2364 path = btrfs_alloc_path();
2368 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2370 btrfs_free_path(path);
2374 key.objectid = send_root->root_key.objectid;
2375 key.type = BTRFS_ROOT_BACKREF_KEY;
2378 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2387 leaf = path->nodes[0];
2388 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2389 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2390 key.objectid != send_root->root_key.objectid) {
2394 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2395 namelen = btrfs_root_ref_name_len(leaf, ref);
2396 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2397 btrfs_release_path(path);
2400 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2404 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2409 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2411 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2412 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2413 sctx->send_root->root_item.received_uuid);
2415 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2416 sctx->send_root->root_item.uuid);
2418 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2419 le64_to_cpu(sctx->send_root->root_item.ctransid));
2421 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2422 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2423 parent_root->root_item.received_uuid);
2425 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2426 parent_root->root_item.uuid);
2427 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2428 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2431 ret = send_cmd(sctx);
2435 btrfs_free_path(path);
2440 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2442 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2446 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2448 p = fs_path_alloc();
2452 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2456 ret = get_cur_path(sctx, ino, gen, p);
2459 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2460 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2462 ret = send_cmd(sctx);
2470 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2472 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2476 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2478 p = fs_path_alloc();
2482 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2486 ret = get_cur_path(sctx, ino, gen, p);
2489 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2490 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2492 ret = send_cmd(sctx);
2500 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2502 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2506 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2509 p = fs_path_alloc();
2513 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2517 ret = get_cur_path(sctx, ino, gen, p);
2520 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2521 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2522 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2524 ret = send_cmd(sctx);
2532 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2534 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2536 struct fs_path *p = NULL;
2537 struct btrfs_inode_item *ii;
2538 struct btrfs_path *path = NULL;
2539 struct extent_buffer *eb;
2540 struct btrfs_key key;
2543 btrfs_debug(fs_info, "send_utimes %llu", ino);
2545 p = fs_path_alloc();
2549 path = alloc_path_for_send();
2556 key.type = BTRFS_INODE_ITEM_KEY;
2558 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2564 eb = path->nodes[0];
2565 slot = path->slots[0];
2566 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2568 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2572 ret = get_cur_path(sctx, ino, gen, p);
2575 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2576 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2577 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2578 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2579 /* TODO Add otime support when the otime patches get into upstream */
2581 ret = send_cmd(sctx);
2586 btrfs_free_path(path);
2591 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2592 * a valid path yet because we did not process the refs yet. So, the inode
2593 * is created as orphan.
2595 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2597 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2605 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2607 p = fs_path_alloc();
2611 if (ino != sctx->cur_ino) {
2612 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2617 gen = sctx->cur_inode_gen;
2618 mode = sctx->cur_inode_mode;
2619 rdev = sctx->cur_inode_rdev;
2622 if (S_ISREG(mode)) {
2623 cmd = BTRFS_SEND_C_MKFILE;
2624 } else if (S_ISDIR(mode)) {
2625 cmd = BTRFS_SEND_C_MKDIR;
2626 } else if (S_ISLNK(mode)) {
2627 cmd = BTRFS_SEND_C_SYMLINK;
2628 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2629 cmd = BTRFS_SEND_C_MKNOD;
2630 } else if (S_ISFIFO(mode)) {
2631 cmd = BTRFS_SEND_C_MKFIFO;
2632 } else if (S_ISSOCK(mode)) {
2633 cmd = BTRFS_SEND_C_MKSOCK;
2635 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2636 (int)(mode & S_IFMT));
2641 ret = begin_cmd(sctx, cmd);
2645 ret = gen_unique_name(sctx, ino, gen, p);
2649 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2650 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2652 if (S_ISLNK(mode)) {
2654 ret = read_symlink(sctx->send_root, ino, p);
2657 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2658 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2659 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2660 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2661 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2664 ret = send_cmd(sctx);
2676 * We need some special handling for inodes that get processed before the parent
2677 * directory got created. See process_recorded_refs for details.
2678 * This function does the check if we already created the dir out of order.
2680 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2683 struct btrfs_path *path = NULL;
2684 struct btrfs_key key;
2685 struct btrfs_key found_key;
2686 struct btrfs_key di_key;
2687 struct extent_buffer *eb;
2688 struct btrfs_dir_item *di;
2691 path = alloc_path_for_send();
2698 key.type = BTRFS_DIR_INDEX_KEY;
2700 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2705 eb = path->nodes[0];
2706 slot = path->slots[0];
2707 if (slot >= btrfs_header_nritems(eb)) {
2708 ret = btrfs_next_leaf(sctx->send_root, path);
2711 } else if (ret > 0) {
2718 btrfs_item_key_to_cpu(eb, &found_key, slot);
2719 if (found_key.objectid != key.objectid ||
2720 found_key.type != key.type) {
2725 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2726 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2728 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2729 di_key.objectid < sctx->send_progress) {
2738 btrfs_free_path(path);
2743 * Only creates the inode if it is:
2744 * 1. Not a directory
2745 * 2. Or a directory which was not created already due to out of order
2746 * directories. See did_create_dir and process_recorded_refs for details.
2748 static int send_create_inode_if_needed(struct send_ctx *sctx)
2752 if (S_ISDIR(sctx->cur_inode_mode)) {
2753 ret = did_create_dir(sctx, sctx->cur_ino);
2762 ret = send_create_inode(sctx, sctx->cur_ino);
2770 struct recorded_ref {
2771 struct list_head list;
2773 struct fs_path *full_path;
2779 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2781 ref->full_path = path;
2782 ref->name = (char *)kbasename(ref->full_path->start);
2783 ref->name_len = ref->full_path->end - ref->name;
2787 * We need to process new refs before deleted refs, but compare_tree gives us
2788 * everything mixed. So we first record all refs and later process them.
2789 * This function is a helper to record one ref.
2791 static int __record_ref(struct list_head *head, u64 dir,
2792 u64 dir_gen, struct fs_path *path)
2794 struct recorded_ref *ref;
2796 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2801 ref->dir_gen = dir_gen;
2802 set_ref_path(ref, path);
2803 list_add_tail(&ref->list, head);
2807 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2809 struct recorded_ref *new;
2811 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2815 new->dir = ref->dir;
2816 new->dir_gen = ref->dir_gen;
2817 new->full_path = NULL;
2818 INIT_LIST_HEAD(&new->list);
2819 list_add_tail(&new->list, list);
2823 static void __free_recorded_refs(struct list_head *head)
2825 struct recorded_ref *cur;
2827 while (!list_empty(head)) {
2828 cur = list_entry(head->next, struct recorded_ref, list);
2829 fs_path_free(cur->full_path);
2830 list_del(&cur->list);
2835 static void free_recorded_refs(struct send_ctx *sctx)
2837 __free_recorded_refs(&sctx->new_refs);
2838 __free_recorded_refs(&sctx->deleted_refs);
2842 * Renames/moves a file/dir to its orphan name. Used when the first
2843 * ref of an unprocessed inode gets overwritten and for all non empty
2846 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2847 struct fs_path *path)
2850 struct fs_path *orphan;
2852 orphan = fs_path_alloc();
2856 ret = gen_unique_name(sctx, ino, gen, orphan);
2860 ret = send_rename(sctx, path, orphan);
2863 fs_path_free(orphan);
2867 static struct orphan_dir_info *
2868 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2870 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2871 struct rb_node *parent = NULL;
2872 struct orphan_dir_info *entry, *odi;
2876 entry = rb_entry(parent, struct orphan_dir_info, node);
2877 if (dir_ino < entry->ino) {
2879 } else if (dir_ino > entry->ino) {
2880 p = &(*p)->rb_right;
2886 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2888 return ERR_PTR(-ENOMEM);
2891 odi->last_dir_index_offset = 0;
2893 rb_link_node(&odi->node, parent, p);
2894 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2898 static struct orphan_dir_info *
2899 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2901 struct rb_node *n = sctx->orphan_dirs.rb_node;
2902 struct orphan_dir_info *entry;
2905 entry = rb_entry(n, struct orphan_dir_info, node);
2906 if (dir_ino < entry->ino)
2908 else if (dir_ino > entry->ino)
2916 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2918 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2923 static void free_orphan_dir_info(struct send_ctx *sctx,
2924 struct orphan_dir_info *odi)
2928 rb_erase(&odi->node, &sctx->orphan_dirs);
2933 * Returns 1 if a directory can be removed at this point in time.
2934 * We check this by iterating all dir items and checking if the inode behind
2935 * the dir item was already processed.
2937 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2941 struct btrfs_root *root = sctx->parent_root;
2942 struct btrfs_path *path;
2943 struct btrfs_key key;
2944 struct btrfs_key found_key;
2945 struct btrfs_key loc;
2946 struct btrfs_dir_item *di;
2947 struct orphan_dir_info *odi = NULL;
2950 * Don't try to rmdir the top/root subvolume dir.
2952 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2955 path = alloc_path_for_send();
2960 key.type = BTRFS_DIR_INDEX_KEY;
2963 odi = get_orphan_dir_info(sctx, dir);
2965 key.offset = odi->last_dir_index_offset;
2967 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2972 struct waiting_dir_move *dm;
2974 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2975 ret = btrfs_next_leaf(root, path);
2982 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2984 if (found_key.objectid != key.objectid ||
2985 found_key.type != key.type)
2988 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2989 struct btrfs_dir_item);
2990 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2992 dm = get_waiting_dir_move(sctx, loc.objectid);
2994 odi = add_orphan_dir_info(sctx, dir);
3000 odi->last_dir_index_offset = found_key.offset;
3001 dm->rmdir_ino = dir;
3006 if (loc.objectid > send_progress) {
3007 odi = add_orphan_dir_info(sctx, dir);
3013 odi->last_dir_index_offset = found_key.offset;
3020 free_orphan_dir_info(sctx, odi);
3025 btrfs_free_path(path);
3029 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3031 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
3033 return entry != NULL;
3036 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3038 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3039 struct rb_node *parent = NULL;
3040 struct waiting_dir_move *entry, *dm;
3042 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3047 dm->orphanized = orphanized;
3051 entry = rb_entry(parent, struct waiting_dir_move, node);
3052 if (ino < entry->ino) {
3054 } else if (ino > entry->ino) {
3055 p = &(*p)->rb_right;
3062 rb_link_node(&dm->node, parent, p);
3063 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3067 static struct waiting_dir_move *
3068 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3070 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3071 struct waiting_dir_move *entry;
3074 entry = rb_entry(n, struct waiting_dir_move, node);
3075 if (ino < entry->ino)
3077 else if (ino > entry->ino)
3085 static void free_waiting_dir_move(struct send_ctx *sctx,
3086 struct waiting_dir_move *dm)
3090 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3094 static int add_pending_dir_move(struct send_ctx *sctx,
3098 struct list_head *new_refs,
3099 struct list_head *deleted_refs,
3100 const bool is_orphan)
3102 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3103 struct rb_node *parent = NULL;
3104 struct pending_dir_move *entry = NULL, *pm;
3105 struct recorded_ref *cur;
3109 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3112 pm->parent_ino = parent_ino;
3115 INIT_LIST_HEAD(&pm->list);
3116 INIT_LIST_HEAD(&pm->update_refs);
3117 RB_CLEAR_NODE(&pm->node);
3121 entry = rb_entry(parent, struct pending_dir_move, node);
3122 if (parent_ino < entry->parent_ino) {
3124 } else if (parent_ino > entry->parent_ino) {
3125 p = &(*p)->rb_right;
3132 list_for_each_entry(cur, deleted_refs, list) {
3133 ret = dup_ref(cur, &pm->update_refs);
3137 list_for_each_entry(cur, new_refs, list) {
3138 ret = dup_ref(cur, &pm->update_refs);
3143 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3148 list_add_tail(&pm->list, &entry->list);
3150 rb_link_node(&pm->node, parent, p);
3151 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3156 __free_recorded_refs(&pm->update_refs);
3162 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3165 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3166 struct pending_dir_move *entry;
3169 entry = rb_entry(n, struct pending_dir_move, node);
3170 if (parent_ino < entry->parent_ino)
3172 else if (parent_ino > entry->parent_ino)
3180 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3181 u64 ino, u64 gen, u64 *ancestor_ino)
3184 u64 parent_inode = 0;
3186 u64 start_ino = ino;
3189 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3190 fs_path_reset(name);
3192 if (is_waiting_for_rm(sctx, ino))
3194 if (is_waiting_for_move(sctx, ino)) {
3195 if (*ancestor_ino == 0)
3196 *ancestor_ino = ino;
3197 ret = get_first_ref(sctx->parent_root, ino,
3198 &parent_inode, &parent_gen, name);
3200 ret = __get_cur_name_and_parent(sctx, ino, gen,
3210 if (parent_inode == start_ino) {
3212 if (*ancestor_ino == 0)
3213 *ancestor_ino = ino;
3222 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3224 struct fs_path *from_path = NULL;
3225 struct fs_path *to_path = NULL;
3226 struct fs_path *name = NULL;
3227 u64 orig_progress = sctx->send_progress;
3228 struct recorded_ref *cur;
3229 u64 parent_ino, parent_gen;
3230 struct waiting_dir_move *dm = NULL;
3236 name = fs_path_alloc();
3237 from_path = fs_path_alloc();
3238 if (!name || !from_path) {
3243 dm = get_waiting_dir_move(sctx, pm->ino);
3245 rmdir_ino = dm->rmdir_ino;
3246 is_orphan = dm->orphanized;
3247 free_waiting_dir_move(sctx, dm);
3250 ret = gen_unique_name(sctx, pm->ino,
3251 pm->gen, from_path);
3253 ret = get_first_ref(sctx->parent_root, pm->ino,
3254 &parent_ino, &parent_gen, name);
3257 ret = get_cur_path(sctx, parent_ino, parent_gen,
3261 ret = fs_path_add_path(from_path, name);
3266 sctx->send_progress = sctx->cur_ino + 1;
3267 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3271 LIST_HEAD(deleted_refs);
3272 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3273 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3274 &pm->update_refs, &deleted_refs,
3279 dm = get_waiting_dir_move(sctx, pm->ino);
3281 dm->rmdir_ino = rmdir_ino;
3285 fs_path_reset(name);
3288 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3292 ret = send_rename(sctx, from_path, to_path);
3297 struct orphan_dir_info *odi;
3300 odi = get_orphan_dir_info(sctx, rmdir_ino);
3302 /* already deleted */
3307 ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino);
3313 name = fs_path_alloc();
3318 ret = get_cur_path(sctx, rmdir_ino, gen, name);
3321 ret = send_rmdir(sctx, name);
3327 ret = send_utimes(sctx, pm->ino, pm->gen);
3332 * After rename/move, need to update the utimes of both new parent(s)
3333 * and old parent(s).
3335 list_for_each_entry(cur, &pm->update_refs, list) {
3337 * The parent inode might have been deleted in the send snapshot
3339 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3340 NULL, NULL, NULL, NULL, NULL);
3341 if (ret == -ENOENT) {
3348 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3355 fs_path_free(from_path);
3356 fs_path_free(to_path);
3357 sctx->send_progress = orig_progress;
3362 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3364 if (!list_empty(&m->list))
3366 if (!RB_EMPTY_NODE(&m->node))
3367 rb_erase(&m->node, &sctx->pending_dir_moves);
3368 __free_recorded_refs(&m->update_refs);
3372 static void tail_append_pending_moves(struct send_ctx *sctx,
3373 struct pending_dir_move *moves,
3374 struct list_head *stack)
3376 if (list_empty(&moves->list)) {
3377 list_add_tail(&moves->list, stack);
3380 list_splice_init(&moves->list, &list);
3381 list_add_tail(&moves->list, stack);
3382 list_splice_tail(&list, stack);
3384 if (!RB_EMPTY_NODE(&moves->node)) {
3385 rb_erase(&moves->node, &sctx->pending_dir_moves);
3386 RB_CLEAR_NODE(&moves->node);
3390 static int apply_children_dir_moves(struct send_ctx *sctx)
3392 struct pending_dir_move *pm;
3393 struct list_head stack;
3394 u64 parent_ino = sctx->cur_ino;
3397 pm = get_pending_dir_moves(sctx, parent_ino);
3401 INIT_LIST_HEAD(&stack);
3402 tail_append_pending_moves(sctx, pm, &stack);
3404 while (!list_empty(&stack)) {
3405 pm = list_first_entry(&stack, struct pending_dir_move, list);
3406 parent_ino = pm->ino;
3407 ret = apply_dir_move(sctx, pm);
3408 free_pending_move(sctx, pm);
3411 pm = get_pending_dir_moves(sctx, parent_ino);
3413 tail_append_pending_moves(sctx, pm, &stack);
3418 while (!list_empty(&stack)) {
3419 pm = list_first_entry(&stack, struct pending_dir_move, list);
3420 free_pending_move(sctx, pm);
3426 * We might need to delay a directory rename even when no ancestor directory
3427 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3428 * renamed. This happens when we rename a directory to the old name (the name
3429 * in the parent root) of some other unrelated directory that got its rename
3430 * delayed due to some ancestor with higher number that got renamed.
3436 * |---- a/ (ino 257)
3437 * | |---- file (ino 260)
3439 * |---- b/ (ino 258)
3440 * |---- c/ (ino 259)
3444 * |---- a/ (ino 258)
3445 * |---- x/ (ino 259)
3446 * |---- y/ (ino 257)
3447 * |----- file (ino 260)
3449 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3450 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3451 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3454 * 1 - rename 259 from 'c' to 'x'
3455 * 2 - rename 257 from 'a' to 'x/y'
3456 * 3 - rename 258 from 'b' to 'a'
3458 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3459 * be done right away and < 0 on error.
3461 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3462 struct recorded_ref *parent_ref,
3463 const bool is_orphan)
3465 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3466 struct btrfs_path *path;
3467 struct btrfs_key key;
3468 struct btrfs_key di_key;
3469 struct btrfs_dir_item *di;
3473 struct waiting_dir_move *wdm;
3475 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3478 path = alloc_path_for_send();
3482 key.objectid = parent_ref->dir;
3483 key.type = BTRFS_DIR_ITEM_KEY;
3484 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3486 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3489 } else if (ret > 0) {
3494 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3495 parent_ref->name_len);
3501 * di_key.objectid has the number of the inode that has a dentry in the
3502 * parent directory with the same name that sctx->cur_ino is being
3503 * renamed to. We need to check if that inode is in the send root as
3504 * well and if it is currently marked as an inode with a pending rename,
3505 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3506 * that it happens after that other inode is renamed.
3508 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3509 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3514 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3515 &left_gen, NULL, NULL, NULL, NULL);
3518 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3519 &right_gen, NULL, NULL, NULL, NULL);
3526 /* Different inode, no need to delay the rename of sctx->cur_ino */
3527 if (right_gen != left_gen) {
3532 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3533 if (wdm && !wdm->orphanized) {
3534 ret = add_pending_dir_move(sctx,
3536 sctx->cur_inode_gen,
3539 &sctx->deleted_refs,
3545 btrfs_free_path(path);
3550 * Check if inode ino2, or any of its ancestors, is inode ino1.
3551 * Return 1 if true, 0 if false and < 0 on error.
3553 static int check_ino_in_path(struct btrfs_root *root,
3558 struct fs_path *fs_path)
3563 return ino1_gen == ino2_gen;
3565 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3570 fs_path_reset(fs_path);
3571 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3575 return parent_gen == ino1_gen;
3582 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3583 * possible path (in case ino2 is not a directory and has multiple hard links).
3584 * Return 1 if true, 0 if false and < 0 on error.
3586 static int is_ancestor(struct btrfs_root *root,
3590 struct fs_path *fs_path)
3592 bool free_fs_path = false;
3594 struct btrfs_path *path = NULL;
3595 struct btrfs_key key;
3598 fs_path = fs_path_alloc();
3601 free_fs_path = true;
3604 path = alloc_path_for_send();
3610 key.objectid = ino2;
3611 key.type = BTRFS_INODE_REF_KEY;
3614 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3619 struct extent_buffer *leaf = path->nodes[0];
3620 int slot = path->slots[0];
3624 if (slot >= btrfs_header_nritems(leaf)) {
3625 ret = btrfs_next_leaf(root, path);
3633 btrfs_item_key_to_cpu(leaf, &key, slot);
3634 if (key.objectid != ino2)
3636 if (key.type != BTRFS_INODE_REF_KEY &&
3637 key.type != BTRFS_INODE_EXTREF_KEY)
3640 item_size = btrfs_item_size_nr(leaf, slot);
3641 while (cur_offset < item_size) {
3645 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3647 struct btrfs_inode_extref *extref;
3649 ptr = btrfs_item_ptr_offset(leaf, slot);
3650 extref = (struct btrfs_inode_extref *)
3652 parent = btrfs_inode_extref_parent(leaf,
3654 cur_offset += sizeof(*extref);
3655 cur_offset += btrfs_inode_extref_name_len(leaf,
3658 parent = key.offset;
3659 cur_offset = item_size;
3662 ret = get_inode_info(root, parent, NULL, &parent_gen,
3663 NULL, NULL, NULL, NULL);
3666 ret = check_ino_in_path(root, ino1, ino1_gen,
3667 parent, parent_gen, fs_path);
3675 btrfs_free_path(path);
3677 fs_path_free(fs_path);
3681 static int wait_for_parent_move(struct send_ctx *sctx,
3682 struct recorded_ref *parent_ref,
3683 const bool is_orphan)
3686 u64 ino = parent_ref->dir;
3687 u64 ino_gen = parent_ref->dir_gen;
3688 u64 parent_ino_before, parent_ino_after;
3689 struct fs_path *path_before = NULL;
3690 struct fs_path *path_after = NULL;
3693 path_after = fs_path_alloc();
3694 path_before = fs_path_alloc();
3695 if (!path_after || !path_before) {
3701 * Our current directory inode may not yet be renamed/moved because some
3702 * ancestor (immediate or not) has to be renamed/moved first. So find if
3703 * such ancestor exists and make sure our own rename/move happens after
3704 * that ancestor is processed to avoid path build infinite loops (done
3705 * at get_cur_path()).
3707 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3708 u64 parent_ino_after_gen;
3710 if (is_waiting_for_move(sctx, ino)) {
3712 * If the current inode is an ancestor of ino in the
3713 * parent root, we need to delay the rename of the
3714 * current inode, otherwise don't delayed the rename
3715 * because we can end up with a circular dependency
3716 * of renames, resulting in some directories never
3717 * getting the respective rename operations issued in
3718 * the send stream or getting into infinite path build
3721 ret = is_ancestor(sctx->parent_root,
3722 sctx->cur_ino, sctx->cur_inode_gen,
3728 fs_path_reset(path_before);
3729 fs_path_reset(path_after);
3731 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3732 &parent_ino_after_gen, path_after);
3735 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3737 if (ret < 0 && ret != -ENOENT) {
3739 } else if (ret == -ENOENT) {
3744 len1 = fs_path_len(path_before);
3745 len2 = fs_path_len(path_after);
3746 if (ino > sctx->cur_ino &&
3747 (parent_ino_before != parent_ino_after || len1 != len2 ||
3748 memcmp(path_before->start, path_after->start, len1))) {
3751 ret = get_inode_info(sctx->parent_root, ino, NULL,
3752 &parent_ino_gen, NULL, NULL, NULL,
3756 if (ino_gen == parent_ino_gen) {
3761 ino = parent_ino_after;
3762 ino_gen = parent_ino_after_gen;
3766 fs_path_free(path_before);
3767 fs_path_free(path_after);
3770 ret = add_pending_dir_move(sctx,
3772 sctx->cur_inode_gen,
3775 &sctx->deleted_refs,
3784 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3787 struct fs_path *new_path;
3790 * Our reference's name member points to its full_path member string, so
3791 * we use here a new path.
3793 new_path = fs_path_alloc();
3797 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3799 fs_path_free(new_path);
3802 ret = fs_path_add(new_path, ref->name, ref->name_len);
3804 fs_path_free(new_path);
3808 fs_path_free(ref->full_path);
3809 set_ref_path(ref, new_path);
3815 * This does all the move/link/unlink/rmdir magic.
3817 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3819 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3821 struct recorded_ref *cur;
3822 struct recorded_ref *cur2;
3823 struct list_head check_dirs;
3824 struct fs_path *valid_path = NULL;
3828 int did_overwrite = 0;
3830 u64 last_dir_ino_rm = 0;
3831 bool can_rename = true;
3832 bool orphanized_dir = false;
3833 bool orphanized_ancestor = false;
3835 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3838 * This should never happen as the root dir always has the same ref
3839 * which is always '..'
3841 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3842 INIT_LIST_HEAD(&check_dirs);
3844 valid_path = fs_path_alloc();
3851 * First, check if the first ref of the current inode was overwritten
3852 * before. If yes, we know that the current inode was already orphanized
3853 * and thus use the orphan name. If not, we can use get_cur_path to
3854 * get the path of the first ref as it would like while receiving at
3855 * this point in time.
3856 * New inodes are always orphan at the beginning, so force to use the
3857 * orphan name in this case.
3858 * The first ref is stored in valid_path and will be updated if it
3859 * gets moved around.
3861 if (!sctx->cur_inode_new) {
3862 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3863 sctx->cur_inode_gen);
3869 if (sctx->cur_inode_new || did_overwrite) {
3870 ret = gen_unique_name(sctx, sctx->cur_ino,
3871 sctx->cur_inode_gen, valid_path);
3876 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3882 list_for_each_entry(cur, &sctx->new_refs, list) {
3884 * We may have refs where the parent directory does not exist
3885 * yet. This happens if the parent directories inum is higher
3886 * than the current inum. To handle this case, we create the
3887 * parent directory out of order. But we need to check if this
3888 * did already happen before due to other refs in the same dir.
3890 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3893 if (ret == inode_state_will_create) {
3896 * First check if any of the current inodes refs did
3897 * already create the dir.
3899 list_for_each_entry(cur2, &sctx->new_refs, list) {
3902 if (cur2->dir == cur->dir) {
3909 * If that did not happen, check if a previous inode
3910 * did already create the dir.
3913 ret = did_create_dir(sctx, cur->dir);
3917 ret = send_create_inode(sctx, cur->dir);
3924 * Check if this new ref would overwrite the first ref of
3925 * another unprocessed inode. If yes, orphanize the
3926 * overwritten inode. If we find an overwritten ref that is
3927 * not the first ref, simply unlink it.
3929 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3930 cur->name, cur->name_len,
3931 &ow_inode, &ow_gen, &ow_mode);
3935 ret = is_first_ref(sctx->parent_root,
3936 ow_inode, cur->dir, cur->name,
3941 struct name_cache_entry *nce;
3942 struct waiting_dir_move *wdm;
3944 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3948 if (S_ISDIR(ow_mode))
3949 orphanized_dir = true;
3952 * If ow_inode has its rename operation delayed
3953 * make sure that its orphanized name is used in
3954 * the source path when performing its rename
3957 if (is_waiting_for_move(sctx, ow_inode)) {
3958 wdm = get_waiting_dir_move(sctx,
3961 wdm->orphanized = true;
3965 * Make sure we clear our orphanized inode's
3966 * name from the name cache. This is because the
3967 * inode ow_inode might be an ancestor of some
3968 * other inode that will be orphanized as well
3969 * later and has an inode number greater than
3970 * sctx->send_progress. We need to prevent
3971 * future name lookups from using the old name
3972 * and get instead the orphan name.
3974 nce = name_cache_search(sctx, ow_inode, ow_gen);
3976 name_cache_delete(sctx, nce);
3981 * ow_inode might currently be an ancestor of
3982 * cur_ino, therefore compute valid_path (the
3983 * current path of cur_ino) again because it
3984 * might contain the pre-orphanization name of
3985 * ow_inode, which is no longer valid.
3987 ret = is_ancestor(sctx->parent_root,
3989 sctx->cur_ino, NULL);
3991 orphanized_ancestor = true;
3992 fs_path_reset(valid_path);
3993 ret = get_cur_path(sctx, sctx->cur_ino,
3994 sctx->cur_inode_gen,
4000 ret = send_unlink(sctx, cur->full_path);
4006 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
4007 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
4016 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
4018 ret = wait_for_parent_move(sctx, cur, is_orphan);
4028 * link/move the ref to the new place. If we have an orphan
4029 * inode, move it and update valid_path. If not, link or move
4030 * it depending on the inode mode.
4032 if (is_orphan && can_rename) {
4033 ret = send_rename(sctx, valid_path, cur->full_path);
4037 ret = fs_path_copy(valid_path, cur->full_path);
4040 } else if (can_rename) {
4041 if (S_ISDIR(sctx->cur_inode_mode)) {
4043 * Dirs can't be linked, so move it. For moved
4044 * dirs, we always have one new and one deleted
4045 * ref. The deleted ref is ignored later.
4047 ret = send_rename(sctx, valid_path,
4050 ret = fs_path_copy(valid_path,
4056 * We might have previously orphanized an inode
4057 * which is an ancestor of our current inode,
4058 * so our reference's full path, which was
4059 * computed before any such orphanizations, must
4062 if (orphanized_dir) {
4063 ret = update_ref_path(sctx, cur);
4067 ret = send_link(sctx, cur->full_path,
4073 ret = dup_ref(cur, &check_dirs);
4078 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4080 * Check if we can already rmdir the directory. If not,
4081 * orphanize it. For every dir item inside that gets deleted
4082 * later, we do this check again and rmdir it then if possible.
4083 * See the use of check_dirs for more details.
4085 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4090 ret = send_rmdir(sctx, valid_path);
4093 } else if (!is_orphan) {
4094 ret = orphanize_inode(sctx, sctx->cur_ino,
4095 sctx->cur_inode_gen, valid_path);
4101 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4102 ret = dup_ref(cur, &check_dirs);
4106 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4107 !list_empty(&sctx->deleted_refs)) {
4109 * We have a moved dir. Add the old parent to check_dirs
4111 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4113 ret = dup_ref(cur, &check_dirs);
4116 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4118 * We have a non dir inode. Go through all deleted refs and
4119 * unlink them if they were not already overwritten by other
4122 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4123 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4124 sctx->cur_ino, sctx->cur_inode_gen,
4125 cur->name, cur->name_len);
4130 * If we orphanized any ancestor before, we need
4131 * to recompute the full path for deleted names,
4132 * since any such path was computed before we
4133 * processed any references and orphanized any
4136 if (orphanized_ancestor) {
4137 ret = update_ref_path(sctx, cur);
4141 ret = send_unlink(sctx, cur->full_path);
4145 ret = dup_ref(cur, &check_dirs);
4150 * If the inode is still orphan, unlink the orphan. This may
4151 * happen when a previous inode did overwrite the first ref
4152 * of this inode and no new refs were added for the current
4153 * inode. Unlinking does not mean that the inode is deleted in
4154 * all cases. There may still be links to this inode in other
4158 ret = send_unlink(sctx, valid_path);
4165 * We did collect all parent dirs where cur_inode was once located. We
4166 * now go through all these dirs and check if they are pending for
4167 * deletion and if it's finally possible to perform the rmdir now.
4168 * We also update the inode stats of the parent dirs here.
4170 list_for_each_entry(cur, &check_dirs, list) {
4172 * In case we had refs into dirs that were not processed yet,
4173 * we don't need to do the utime and rmdir logic for these dirs.
4174 * The dir will be processed later.
4176 if (cur->dir > sctx->cur_ino)
4179 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4183 if (ret == inode_state_did_create ||
4184 ret == inode_state_no_change) {
4185 /* TODO delayed utimes */
4186 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4189 } else if (ret == inode_state_did_delete &&
4190 cur->dir != last_dir_ino_rm) {
4191 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4196 ret = get_cur_path(sctx, cur->dir,
4197 cur->dir_gen, valid_path);
4200 ret = send_rmdir(sctx, valid_path);
4203 last_dir_ino_rm = cur->dir;
4211 __free_recorded_refs(&check_dirs);
4212 free_recorded_refs(sctx);
4213 fs_path_free(valid_path);
4217 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4218 void *ctx, struct list_head *refs)
4221 struct send_ctx *sctx = ctx;
4225 p = fs_path_alloc();
4229 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4234 ret = get_cur_path(sctx, dir, gen, p);
4237 ret = fs_path_add_path(p, name);
4241 ret = __record_ref(refs, dir, gen, p);
4249 static int __record_new_ref(int num, u64 dir, int index,
4250 struct fs_path *name,
4253 struct send_ctx *sctx = ctx;
4254 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4258 static int __record_deleted_ref(int num, u64 dir, int index,
4259 struct fs_path *name,
4262 struct send_ctx *sctx = ctx;
4263 return record_ref(sctx->parent_root, dir, name, ctx,
4264 &sctx->deleted_refs);
4267 static int record_new_ref(struct send_ctx *sctx)
4271 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4272 sctx->cmp_key, 0, __record_new_ref, sctx);
4281 static int record_deleted_ref(struct send_ctx *sctx)
4285 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4286 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4295 struct find_ref_ctx {
4298 struct btrfs_root *root;
4299 struct fs_path *name;
4303 static int __find_iref(int num, u64 dir, int index,
4304 struct fs_path *name,
4307 struct find_ref_ctx *ctx = ctx_;
4311 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4312 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4314 * To avoid doing extra lookups we'll only do this if everything
4317 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4321 if (dir_gen != ctx->dir_gen)
4323 ctx->found_idx = num;
4329 static int find_iref(struct btrfs_root *root,
4330 struct btrfs_path *path,
4331 struct btrfs_key *key,
4332 u64 dir, u64 dir_gen, struct fs_path *name)
4335 struct find_ref_ctx ctx;
4339 ctx.dir_gen = dir_gen;
4343 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4347 if (ctx.found_idx == -1)
4350 return ctx.found_idx;
4353 static int __record_changed_new_ref(int num, u64 dir, int index,
4354 struct fs_path *name,
4359 struct send_ctx *sctx = ctx;
4361 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4366 ret = find_iref(sctx->parent_root, sctx->right_path,
4367 sctx->cmp_key, dir, dir_gen, name);
4369 ret = __record_new_ref(num, dir, index, name, sctx);
4376 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4377 struct fs_path *name,
4382 struct send_ctx *sctx = ctx;
4384 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4389 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4390 dir, dir_gen, name);
4392 ret = __record_deleted_ref(num, dir, index, name, sctx);
4399 static int record_changed_ref(struct send_ctx *sctx)
4403 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4404 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4407 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4408 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4418 * Record and process all refs at once. Needed when an inode changes the
4419 * generation number, which means that it was deleted and recreated.
4421 static int process_all_refs(struct send_ctx *sctx,
4422 enum btrfs_compare_tree_result cmd)
4425 struct btrfs_root *root;
4426 struct btrfs_path *path;
4427 struct btrfs_key key;
4428 struct btrfs_key found_key;
4429 struct extent_buffer *eb;
4431 iterate_inode_ref_t cb;
4432 int pending_move = 0;
4434 path = alloc_path_for_send();
4438 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4439 root = sctx->send_root;
4440 cb = __record_new_ref;
4441 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4442 root = sctx->parent_root;
4443 cb = __record_deleted_ref;
4445 btrfs_err(sctx->send_root->fs_info,
4446 "Wrong command %d in process_all_refs", cmd);
4451 key.objectid = sctx->cmp_key->objectid;
4452 key.type = BTRFS_INODE_REF_KEY;
4454 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4459 eb = path->nodes[0];
4460 slot = path->slots[0];
4461 if (slot >= btrfs_header_nritems(eb)) {
4462 ret = btrfs_next_leaf(root, path);
4470 btrfs_item_key_to_cpu(eb, &found_key, slot);
4472 if (found_key.objectid != key.objectid ||
4473 (found_key.type != BTRFS_INODE_REF_KEY &&
4474 found_key.type != BTRFS_INODE_EXTREF_KEY))
4477 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4483 btrfs_release_path(path);
4486 * We don't actually care about pending_move as we are simply
4487 * re-creating this inode and will be rename'ing it into place once we
4488 * rename the parent directory.
4490 ret = process_recorded_refs(sctx, &pending_move);
4492 btrfs_free_path(path);
4496 static int send_set_xattr(struct send_ctx *sctx,
4497 struct fs_path *path,
4498 const char *name, int name_len,
4499 const char *data, int data_len)
4503 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4507 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4508 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4509 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4511 ret = send_cmd(sctx);
4518 static int send_remove_xattr(struct send_ctx *sctx,
4519 struct fs_path *path,
4520 const char *name, int name_len)
4524 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4528 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4529 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4531 ret = send_cmd(sctx);
4538 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4539 const char *name, int name_len,
4540 const char *data, int data_len,
4544 struct send_ctx *sctx = ctx;
4546 struct posix_acl_xattr_header dummy_acl;
4548 p = fs_path_alloc();
4553 * This hack is needed because empty acls are stored as zero byte
4554 * data in xattrs. Problem with that is, that receiving these zero byte
4555 * acls will fail later. To fix this, we send a dummy acl list that
4556 * only contains the version number and no entries.
4558 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4559 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4560 if (data_len == 0) {
4561 dummy_acl.a_version =
4562 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4563 data = (char *)&dummy_acl;
4564 data_len = sizeof(dummy_acl);
4568 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4572 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4579 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4580 const char *name, int name_len,
4581 const char *data, int data_len,
4585 struct send_ctx *sctx = ctx;
4588 p = fs_path_alloc();
4592 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4596 ret = send_remove_xattr(sctx, p, name, name_len);
4603 static int process_new_xattr(struct send_ctx *sctx)
4607 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4608 __process_new_xattr, sctx);
4613 static int process_deleted_xattr(struct send_ctx *sctx)
4615 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4616 __process_deleted_xattr, sctx);
4619 struct find_xattr_ctx {
4627 static int __find_xattr(int num, struct btrfs_key *di_key,
4628 const char *name, int name_len,
4629 const char *data, int data_len,
4630 u8 type, void *vctx)
4632 struct find_xattr_ctx *ctx = vctx;
4634 if (name_len == ctx->name_len &&
4635 strncmp(name, ctx->name, name_len) == 0) {
4636 ctx->found_idx = num;
4637 ctx->found_data_len = data_len;
4638 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4639 if (!ctx->found_data)
4646 static int find_xattr(struct btrfs_root *root,
4647 struct btrfs_path *path,
4648 struct btrfs_key *key,
4649 const char *name, int name_len,
4650 char **data, int *data_len)
4653 struct find_xattr_ctx ctx;
4656 ctx.name_len = name_len;
4658 ctx.found_data = NULL;
4659 ctx.found_data_len = 0;
4661 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4665 if (ctx.found_idx == -1)
4668 *data = ctx.found_data;
4669 *data_len = ctx.found_data_len;
4671 kfree(ctx.found_data);
4673 return ctx.found_idx;
4677 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4678 const char *name, int name_len,
4679 const char *data, int data_len,
4683 struct send_ctx *sctx = ctx;
4684 char *found_data = NULL;
4685 int found_data_len = 0;
4687 ret = find_xattr(sctx->parent_root, sctx->right_path,
4688 sctx->cmp_key, name, name_len, &found_data,
4690 if (ret == -ENOENT) {
4691 ret = __process_new_xattr(num, di_key, name, name_len, data,
4692 data_len, type, ctx);
4693 } else if (ret >= 0) {
4694 if (data_len != found_data_len ||
4695 memcmp(data, found_data, data_len)) {
4696 ret = __process_new_xattr(num, di_key, name, name_len,
4697 data, data_len, type, ctx);
4707 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4708 const char *name, int name_len,
4709 const char *data, int data_len,
4713 struct send_ctx *sctx = ctx;
4715 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4716 name, name_len, NULL, NULL);
4718 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4719 data_len, type, ctx);
4726 static int process_changed_xattr(struct send_ctx *sctx)
4730 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4731 __process_changed_new_xattr, sctx);
4734 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4735 __process_changed_deleted_xattr, sctx);
4741 static int process_all_new_xattrs(struct send_ctx *sctx)
4744 struct btrfs_root *root;
4745 struct btrfs_path *path;
4746 struct btrfs_key key;
4747 struct btrfs_key found_key;
4748 struct extent_buffer *eb;
4751 path = alloc_path_for_send();
4755 root = sctx->send_root;
4757 key.objectid = sctx->cmp_key->objectid;
4758 key.type = BTRFS_XATTR_ITEM_KEY;
4760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4765 eb = path->nodes[0];
4766 slot = path->slots[0];
4767 if (slot >= btrfs_header_nritems(eb)) {
4768 ret = btrfs_next_leaf(root, path);
4771 } else if (ret > 0) {
4778 btrfs_item_key_to_cpu(eb, &found_key, slot);
4779 if (found_key.objectid != key.objectid ||
4780 found_key.type != key.type) {
4785 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4793 btrfs_free_path(path);
4797 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4799 struct btrfs_root *root = sctx->send_root;
4800 struct btrfs_fs_info *fs_info = root->fs_info;
4801 struct inode *inode;
4804 struct btrfs_key key;
4805 pgoff_t index = offset >> PAGE_SHIFT;
4807 unsigned pg_offset = offset_in_page(offset);
4810 key.objectid = sctx->cur_ino;
4811 key.type = BTRFS_INODE_ITEM_KEY;
4814 inode = btrfs_iget(fs_info->sb, &key, root);
4816 return PTR_ERR(inode);
4818 if (offset + len > i_size_read(inode)) {
4819 if (offset > i_size_read(inode))
4822 len = offset - i_size_read(inode);
4827 last_index = (offset + len - 1) >> PAGE_SHIFT;
4829 /* initial readahead */
4830 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4831 file_ra_state_init(&sctx->ra, inode->i_mapping);
4833 while (index <= last_index) {
4834 unsigned cur_len = min_t(unsigned, len,
4835 PAGE_SIZE - pg_offset);
4837 page = find_lock_page(inode->i_mapping, index);
4839 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4840 NULL, index, last_index + 1 - index);
4842 page = find_or_create_page(inode->i_mapping, index,
4850 if (PageReadahead(page)) {
4851 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4852 NULL, page, index, last_index + 1 - index);
4855 if (!PageUptodate(page)) {
4856 btrfs_readpage(NULL, page);
4858 if (!PageUptodate(page)) {
4867 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4882 * Read some bytes from the current inode/file and send a write command to
4885 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4887 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4890 ssize_t num_read = 0;
4892 p = fs_path_alloc();
4896 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4898 num_read = fill_read_buf(sctx, offset, len);
4899 if (num_read <= 0) {
4905 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4909 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4913 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4914 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4915 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4917 ret = send_cmd(sctx);
4928 * Send a clone command to user space.
4930 static int send_clone(struct send_ctx *sctx,
4931 u64 offset, u32 len,
4932 struct clone_root *clone_root)
4938 btrfs_debug(sctx->send_root->fs_info,
4939 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4940 offset, len, clone_root->root->root_key.objectid,
4941 clone_root->ino, clone_root->offset);
4943 p = fs_path_alloc();
4947 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4951 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4955 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4956 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4957 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4959 if (clone_root->root == sctx->send_root) {
4960 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4961 &gen, NULL, NULL, NULL, NULL);
4964 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4966 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4972 * If the parent we're using has a received_uuid set then use that as
4973 * our clone source as that is what we will look for when doing a
4976 * This covers the case that we create a snapshot off of a received
4977 * subvolume and then use that as the parent and try to receive on a
4980 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4981 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4982 clone_root->root->root_item.received_uuid);
4984 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4985 clone_root->root->root_item.uuid);
4986 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4987 le64_to_cpu(clone_root->root->root_item.ctransid));
4988 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4989 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4990 clone_root->offset);
4992 ret = send_cmd(sctx);
5001 * Send an update extent command to user space.
5003 static int send_update_extent(struct send_ctx *sctx,
5004 u64 offset, u32 len)
5009 p = fs_path_alloc();
5013 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
5017 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5021 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5022 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5023 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
5025 ret = send_cmd(sctx);
5033 static int send_hole(struct send_ctx *sctx, u64 end)
5035 struct fs_path *p = NULL;
5036 u64 offset = sctx->cur_inode_last_extent;
5041 * A hole that starts at EOF or beyond it. Since we do not yet support
5042 * fallocate (for extent preallocation and hole punching), sending a
5043 * write of zeroes starting at EOF or beyond would later require issuing
5044 * a truncate operation which would undo the write and achieve nothing.
5046 if (offset >= sctx->cur_inode_size)
5050 * Don't go beyond the inode's i_size due to prealloc extents that start
5053 end = min_t(u64, end, sctx->cur_inode_size);
5055 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5056 return send_update_extent(sctx, offset, end - offset);
5058 p = fs_path_alloc();
5061 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5063 goto tlv_put_failure;
5064 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5065 while (offset < end) {
5066 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5068 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5071 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5072 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5073 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5074 ret = send_cmd(sctx);
5079 sctx->cur_inode_next_write_offset = offset;
5085 static int send_extent_data(struct send_ctx *sctx,
5091 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5092 return send_update_extent(sctx, offset, len);
5094 while (sent < len) {
5095 u64 size = len - sent;
5098 if (size > BTRFS_SEND_READ_SIZE)
5099 size = BTRFS_SEND_READ_SIZE;
5100 ret = send_write(sctx, offset + sent, size);
5110 static int clone_range(struct send_ctx *sctx,
5111 struct clone_root *clone_root,
5112 const u64 disk_byte,
5117 struct btrfs_path *path;
5118 struct btrfs_key key;
5120 u64 clone_src_i_size = 0;
5123 * Prevent cloning from a zero offset with a length matching the sector
5124 * size because in some scenarios this will make the receiver fail.
5126 * For example, if in the source filesystem the extent at offset 0
5127 * has a length of sectorsize and it was written using direct IO, then
5128 * it can never be an inline extent (even if compression is enabled).
5129 * Then this extent can be cloned in the original filesystem to a non
5130 * zero file offset, but it may not be possible to clone in the
5131 * destination filesystem because it can be inlined due to compression
5132 * on the destination filesystem (as the receiver's write operations are
5133 * always done using buffered IO). The same happens when the original
5134 * filesystem does not have compression enabled but the destination
5137 if (clone_root->offset == 0 &&
5138 len == sctx->send_root->fs_info->sectorsize)
5139 return send_extent_data(sctx, offset, len);
5141 path = alloc_path_for_send();
5146 * There are inodes that have extents that lie behind its i_size. Don't
5147 * accept clones from these extents.
5149 ret = __get_inode_info(clone_root->root, path, clone_root->ino,
5150 &clone_src_i_size, NULL, NULL, NULL, NULL, NULL);
5151 btrfs_release_path(path);
5156 * We can't send a clone operation for the entire range if we find
5157 * extent items in the respective range in the source file that
5158 * refer to different extents or if we find holes.
5159 * So check for that and do a mix of clone and regular write/copy
5160 * operations if needed.
5164 * mkfs.btrfs -f /dev/sda
5165 * mount /dev/sda /mnt
5166 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5167 * cp --reflink=always /mnt/foo /mnt/bar
5168 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5169 * btrfs subvolume snapshot -r /mnt /mnt/snap
5171 * If when we send the snapshot and we are processing file bar (which
5172 * has a higher inode number than foo) we blindly send a clone operation
5173 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5174 * a file bar that matches the content of file foo - iow, doesn't match
5175 * the content from bar in the original filesystem.
5177 key.objectid = clone_root->ino;
5178 key.type = BTRFS_EXTENT_DATA_KEY;
5179 key.offset = clone_root->offset;
5180 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5183 if (ret > 0 && path->slots[0] > 0) {
5184 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5185 if (key.objectid == clone_root->ino &&
5186 key.type == BTRFS_EXTENT_DATA_KEY)
5191 struct extent_buffer *leaf = path->nodes[0];
5192 int slot = path->slots[0];
5193 struct btrfs_file_extent_item *ei;
5197 u64 clone_data_offset;
5199 if (slot >= btrfs_header_nritems(leaf)) {
5200 ret = btrfs_next_leaf(clone_root->root, path);
5208 btrfs_item_key_to_cpu(leaf, &key, slot);
5211 * We might have an implicit trailing hole (NO_HOLES feature
5212 * enabled). We deal with it after leaving this loop.
5214 if (key.objectid != clone_root->ino ||
5215 key.type != BTRFS_EXTENT_DATA_KEY)
5218 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5219 type = btrfs_file_extent_type(leaf, ei);
5220 if (type == BTRFS_FILE_EXTENT_INLINE) {
5221 ext_len = btrfs_file_extent_ram_bytes(leaf, ei);
5222 ext_len = PAGE_ALIGN(ext_len);
5224 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5227 if (key.offset + ext_len <= clone_root->offset)
5230 if (key.offset > clone_root->offset) {
5231 /* Implicit hole, NO_HOLES feature enabled. */
5232 u64 hole_len = key.offset - clone_root->offset;
5236 ret = send_extent_data(sctx, offset, hole_len);
5244 clone_root->offset += hole_len;
5245 data_offset += hole_len;
5248 if (key.offset >= clone_root->offset + len)
5251 if (key.offset >= clone_src_i_size)
5254 if (key.offset + ext_len > clone_src_i_size)
5255 ext_len = clone_src_i_size - key.offset;
5257 clone_data_offset = btrfs_file_extent_offset(leaf, ei);
5258 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
5259 clone_root->offset = key.offset;
5260 if (clone_data_offset < data_offset &&
5261 clone_data_offset + ext_len > data_offset) {
5264 extent_offset = data_offset - clone_data_offset;
5265 ext_len -= extent_offset;
5266 clone_data_offset += extent_offset;
5267 clone_root->offset += extent_offset;
5271 clone_len = min_t(u64, ext_len, len);
5273 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5274 clone_data_offset == data_offset) {
5275 const u64 src_end = clone_root->offset + clone_len;
5276 const u64 sectorsize = SZ_64K;
5279 * We can't clone the last block, when its size is not
5280 * sector size aligned, into the middle of a file. If we
5281 * do so, the receiver will get a failure (-EINVAL) when
5282 * trying to clone or will silently corrupt the data in
5283 * the destination file if it's on a kernel without the
5284 * fix introduced by commit ac765f83f1397646
5285 * ("Btrfs: fix data corruption due to cloning of eof
5288 * So issue a clone of the aligned down range plus a
5289 * regular write for the eof block, if we hit that case.
5291 * Also, we use the maximum possible sector size, 64K,
5292 * because we don't know what's the sector size of the
5293 * filesystem that receives the stream, so we have to
5294 * assume the largest possible sector size.
5296 if (src_end == clone_src_i_size &&
5297 !IS_ALIGNED(src_end, sectorsize) &&
5298 offset + clone_len < sctx->cur_inode_size) {
5301 slen = ALIGN_DOWN(src_end - clone_root->offset,
5304 ret = send_clone(sctx, offset, slen,
5309 ret = send_extent_data(sctx, offset + slen,
5312 ret = send_clone(sctx, offset, clone_len,
5316 ret = send_extent_data(sctx, offset, clone_len);
5325 offset += clone_len;
5326 clone_root->offset += clone_len;
5327 data_offset += clone_len;
5333 ret = send_extent_data(sctx, offset, len);
5337 btrfs_free_path(path);
5341 static int send_write_or_clone(struct send_ctx *sctx,
5342 struct btrfs_path *path,
5343 struct btrfs_key *key,
5344 struct clone_root *clone_root)
5347 struct btrfs_file_extent_item *ei;
5348 u64 offset = key->offset;
5351 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5353 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5354 struct btrfs_file_extent_item);
5355 type = btrfs_file_extent_type(path->nodes[0], ei);
5356 if (type == BTRFS_FILE_EXTENT_INLINE) {
5357 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
5359 * it is possible the inline item won't cover the whole page,
5360 * but there may be items after this page. Make
5361 * sure to send the whole thing
5363 len = PAGE_ALIGN(len);
5365 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5368 if (offset >= sctx->cur_inode_size) {
5372 if (offset + len > sctx->cur_inode_size)
5373 len = sctx->cur_inode_size - offset;
5379 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5383 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5384 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5385 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5388 ret = send_extent_data(sctx, offset, len);
5390 sctx->cur_inode_next_write_offset = offset + len;
5395 static int is_extent_unchanged(struct send_ctx *sctx,
5396 struct btrfs_path *left_path,
5397 struct btrfs_key *ekey)
5400 struct btrfs_key key;
5401 struct btrfs_path *path = NULL;
5402 struct extent_buffer *eb;
5404 struct btrfs_key found_key;
5405 struct btrfs_file_extent_item *ei;
5410 u64 left_offset_fixed;
5418 path = alloc_path_for_send();
5422 eb = left_path->nodes[0];
5423 slot = left_path->slots[0];
5424 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5425 left_type = btrfs_file_extent_type(eb, ei);
5427 if (left_type != BTRFS_FILE_EXTENT_REG) {
5431 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5432 left_len = btrfs_file_extent_num_bytes(eb, ei);
5433 left_offset = btrfs_file_extent_offset(eb, ei);
5434 left_gen = btrfs_file_extent_generation(eb, ei);
5437 * Following comments will refer to these graphics. L is the left
5438 * extents which we are checking at the moment. 1-8 are the right
5439 * extents that we iterate.
5442 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5445 * |--1--|-2b-|...(same as above)
5447 * Alternative situation. Happens on files where extents got split.
5449 * |-----------7-----------|-6-|
5451 * Alternative situation. Happens on files which got larger.
5454 * Nothing follows after 8.
5457 key.objectid = ekey->objectid;
5458 key.type = BTRFS_EXTENT_DATA_KEY;
5459 key.offset = ekey->offset;
5460 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5469 * Handle special case where the right side has no extents at all.
5471 eb = path->nodes[0];
5472 slot = path->slots[0];
5473 btrfs_item_key_to_cpu(eb, &found_key, slot);
5474 if (found_key.objectid != key.objectid ||
5475 found_key.type != key.type) {
5476 /* If we're a hole then just pretend nothing changed */
5477 ret = (left_disknr) ? 0 : 1;
5482 * We're now on 2a, 2b or 7.
5485 while (key.offset < ekey->offset + left_len) {
5486 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5487 right_type = btrfs_file_extent_type(eb, ei);
5488 if (right_type != BTRFS_FILE_EXTENT_REG &&
5489 right_type != BTRFS_FILE_EXTENT_INLINE) {
5494 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5495 right_len = btrfs_file_extent_ram_bytes(eb, ei);
5496 right_len = PAGE_ALIGN(right_len);
5498 right_len = btrfs_file_extent_num_bytes(eb, ei);
5502 * Are we at extent 8? If yes, we know the extent is changed.
5503 * This may only happen on the first iteration.
5505 if (found_key.offset + right_len <= ekey->offset) {
5506 /* If we're a hole just pretend nothing changed */
5507 ret = (left_disknr) ? 0 : 1;
5512 * We just wanted to see if when we have an inline extent, what
5513 * follows it is a regular extent (wanted to check the above
5514 * condition for inline extents too). This should normally not
5515 * happen but it's possible for example when we have an inline
5516 * compressed extent representing data with a size matching
5517 * the page size (currently the same as sector size).
5519 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5524 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5525 right_offset = btrfs_file_extent_offset(eb, ei);
5526 right_gen = btrfs_file_extent_generation(eb, ei);
5528 left_offset_fixed = left_offset;
5529 if (key.offset < ekey->offset) {
5530 /* Fix the right offset for 2a and 7. */
5531 right_offset += ekey->offset - key.offset;
5533 /* Fix the left offset for all behind 2a and 2b */
5534 left_offset_fixed += key.offset - ekey->offset;
5538 * Check if we have the same extent.
5540 if (left_disknr != right_disknr ||
5541 left_offset_fixed != right_offset ||
5542 left_gen != right_gen) {
5548 * Go to the next extent.
5550 ret = btrfs_next_item(sctx->parent_root, path);
5554 eb = path->nodes[0];
5555 slot = path->slots[0];
5556 btrfs_item_key_to_cpu(eb, &found_key, slot);
5558 if (ret || found_key.objectid != key.objectid ||
5559 found_key.type != key.type) {
5560 key.offset += right_len;
5563 if (found_key.offset != key.offset + right_len) {
5571 * We're now behind the left extent (treat as unchanged) or at the end
5572 * of the right side (treat as changed).
5574 if (key.offset >= ekey->offset + left_len)
5581 btrfs_free_path(path);
5585 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5587 struct btrfs_path *path;
5588 struct btrfs_root *root = sctx->send_root;
5589 struct btrfs_file_extent_item *fi;
5590 struct btrfs_key key;
5595 path = alloc_path_for_send();
5599 sctx->cur_inode_last_extent = 0;
5601 key.objectid = sctx->cur_ino;
5602 key.type = BTRFS_EXTENT_DATA_KEY;
5603 key.offset = offset;
5604 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5608 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5609 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5612 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5613 struct btrfs_file_extent_item);
5614 type = btrfs_file_extent_type(path->nodes[0], fi);
5615 if (type == BTRFS_FILE_EXTENT_INLINE) {
5616 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5617 extent_end = ALIGN(key.offset + size,
5618 sctx->send_root->fs_info->sectorsize);
5620 extent_end = key.offset +
5621 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5623 sctx->cur_inode_last_extent = extent_end;
5625 btrfs_free_path(path);
5629 static int range_is_hole_in_parent(struct send_ctx *sctx,
5633 struct btrfs_path *path;
5634 struct btrfs_key key;
5635 struct btrfs_root *root = sctx->parent_root;
5636 u64 search_start = start;
5639 path = alloc_path_for_send();
5643 key.objectid = sctx->cur_ino;
5644 key.type = BTRFS_EXTENT_DATA_KEY;
5645 key.offset = search_start;
5646 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5649 if (ret > 0 && path->slots[0] > 0)
5652 while (search_start < end) {
5653 struct extent_buffer *leaf = path->nodes[0];
5654 int slot = path->slots[0];
5655 struct btrfs_file_extent_item *fi;
5658 if (slot >= btrfs_header_nritems(leaf)) {
5659 ret = btrfs_next_leaf(root, path);
5667 btrfs_item_key_to_cpu(leaf, &key, slot);
5668 if (key.objectid < sctx->cur_ino ||
5669 key.type < BTRFS_EXTENT_DATA_KEY)
5671 if (key.objectid > sctx->cur_ino ||
5672 key.type > BTRFS_EXTENT_DATA_KEY ||
5676 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5677 if (btrfs_file_extent_type(leaf, fi) ==
5678 BTRFS_FILE_EXTENT_INLINE) {
5679 u64 size = btrfs_file_extent_ram_bytes(leaf, fi);
5681 extent_end = ALIGN(key.offset + size,
5682 root->fs_info->sectorsize);
5684 extent_end = key.offset +
5685 btrfs_file_extent_num_bytes(leaf, fi);
5687 if (extent_end <= start)
5689 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5690 search_start = extent_end;
5700 btrfs_free_path(path);
5704 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5705 struct btrfs_key *key)
5707 struct btrfs_file_extent_item *fi;
5712 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5715 if (sctx->cur_inode_last_extent == (u64)-1) {
5716 ret = get_last_extent(sctx, key->offset - 1);
5721 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5722 struct btrfs_file_extent_item);
5723 type = btrfs_file_extent_type(path->nodes[0], fi);
5724 if (type == BTRFS_FILE_EXTENT_INLINE) {
5725 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5726 extent_end = ALIGN(key->offset + size,
5727 sctx->send_root->fs_info->sectorsize);
5729 extent_end = key->offset +
5730 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5733 if (path->slots[0] == 0 &&
5734 sctx->cur_inode_last_extent < key->offset) {
5736 * We might have skipped entire leafs that contained only
5737 * file extent items for our current inode. These leafs have
5738 * a generation number smaller (older) than the one in the
5739 * current leaf and the leaf our last extent came from, and
5740 * are located between these 2 leafs.
5742 ret = get_last_extent(sctx, key->offset - 1);
5747 if (sctx->cur_inode_last_extent < key->offset) {
5748 ret = range_is_hole_in_parent(sctx,
5749 sctx->cur_inode_last_extent,
5754 ret = send_hole(sctx, key->offset);
5758 sctx->cur_inode_last_extent = extent_end;
5762 static int process_extent(struct send_ctx *sctx,
5763 struct btrfs_path *path,
5764 struct btrfs_key *key)
5766 struct clone_root *found_clone = NULL;
5769 if (S_ISLNK(sctx->cur_inode_mode))
5772 if (sctx->parent_root && !sctx->cur_inode_new) {
5773 ret = is_extent_unchanged(sctx, path, key);
5781 struct btrfs_file_extent_item *ei;
5784 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5785 struct btrfs_file_extent_item);
5786 type = btrfs_file_extent_type(path->nodes[0], ei);
5787 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5788 type == BTRFS_FILE_EXTENT_REG) {
5790 * The send spec does not have a prealloc command yet,
5791 * so just leave a hole for prealloc'ed extents until
5792 * we have enough commands queued up to justify rev'ing
5795 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5800 /* Have a hole, just skip it. */
5801 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5808 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5809 sctx->cur_inode_size, &found_clone);
5810 if (ret != -ENOENT && ret < 0)
5813 ret = send_write_or_clone(sctx, path, key, found_clone);
5817 ret = maybe_send_hole(sctx, path, key);
5822 static int process_all_extents(struct send_ctx *sctx)
5825 struct btrfs_root *root;
5826 struct btrfs_path *path;
5827 struct btrfs_key key;
5828 struct btrfs_key found_key;
5829 struct extent_buffer *eb;
5832 root = sctx->send_root;
5833 path = alloc_path_for_send();
5837 key.objectid = sctx->cmp_key->objectid;
5838 key.type = BTRFS_EXTENT_DATA_KEY;
5840 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5845 eb = path->nodes[0];
5846 slot = path->slots[0];
5848 if (slot >= btrfs_header_nritems(eb)) {
5849 ret = btrfs_next_leaf(root, path);
5852 } else if (ret > 0) {
5859 btrfs_item_key_to_cpu(eb, &found_key, slot);
5861 if (found_key.objectid != key.objectid ||
5862 found_key.type != key.type) {
5867 ret = process_extent(sctx, path, &found_key);
5875 btrfs_free_path(path);
5879 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5881 int *refs_processed)
5885 if (sctx->cur_ino == 0)
5887 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5888 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5890 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5893 ret = process_recorded_refs(sctx, pending_move);
5897 *refs_processed = 1;
5902 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5913 int need_truncate = 1;
5914 int pending_move = 0;
5915 int refs_processed = 0;
5917 if (sctx->ignore_cur_inode)
5920 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5926 * We have processed the refs and thus need to advance send_progress.
5927 * Now, calls to get_cur_xxx will take the updated refs of the current
5928 * inode into account.
5930 * On the other hand, if our current inode is a directory and couldn't
5931 * be moved/renamed because its parent was renamed/moved too and it has
5932 * a higher inode number, we can only move/rename our current inode
5933 * after we moved/renamed its parent. Therefore in this case operate on
5934 * the old path (pre move/rename) of our current inode, and the
5935 * move/rename will be performed later.
5937 if (refs_processed && !pending_move)
5938 sctx->send_progress = sctx->cur_ino + 1;
5940 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5942 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5945 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5946 &left_mode, &left_uid, &left_gid, NULL);
5950 if (!sctx->parent_root || sctx->cur_inode_new) {
5952 if (!S_ISLNK(sctx->cur_inode_mode))
5954 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5959 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5960 &old_size, NULL, &right_mode, &right_uid,
5965 if (left_uid != right_uid || left_gid != right_gid)
5967 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5969 if ((old_size == sctx->cur_inode_size) ||
5970 (sctx->cur_inode_size > old_size &&
5971 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5975 if (S_ISREG(sctx->cur_inode_mode)) {
5976 if (need_send_hole(sctx)) {
5977 if (sctx->cur_inode_last_extent == (u64)-1 ||
5978 sctx->cur_inode_last_extent <
5979 sctx->cur_inode_size) {
5980 ret = get_last_extent(sctx, (u64)-1);
5984 if (sctx->cur_inode_last_extent <
5985 sctx->cur_inode_size) {
5986 ret = send_hole(sctx, sctx->cur_inode_size);
5991 if (need_truncate) {
5992 ret = send_truncate(sctx, sctx->cur_ino,
5993 sctx->cur_inode_gen,
5994 sctx->cur_inode_size);
6001 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6002 left_uid, left_gid);
6007 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6014 * If other directory inodes depended on our current directory
6015 * inode's move/rename, now do their move/rename operations.
6017 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
6018 ret = apply_children_dir_moves(sctx);
6022 * Need to send that every time, no matter if it actually
6023 * changed between the two trees as we have done changes to
6024 * the inode before. If our inode is a directory and it's
6025 * waiting to be moved/renamed, we will send its utimes when
6026 * it's moved/renamed, therefore we don't need to do it here.
6028 sctx->send_progress = sctx->cur_ino + 1;
6029 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
6038 struct parent_paths_ctx {
6039 struct list_head *refs;
6040 struct send_ctx *sctx;
6043 static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name,
6046 struct parent_paths_ctx *ppctx = ctx;
6048 return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx,
6053 * Issue unlink operations for all paths of the current inode found in the
6056 static int btrfs_unlink_all_paths(struct send_ctx *sctx)
6058 LIST_HEAD(deleted_refs);
6059 struct btrfs_path *path;
6060 struct btrfs_key key;
6061 struct parent_paths_ctx ctx;
6064 path = alloc_path_for_send();
6068 key.objectid = sctx->cur_ino;
6069 key.type = BTRFS_INODE_REF_KEY;
6071 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
6075 ctx.refs = &deleted_refs;
6079 struct extent_buffer *eb = path->nodes[0];
6080 int slot = path->slots[0];
6082 if (slot >= btrfs_header_nritems(eb)) {
6083 ret = btrfs_next_leaf(sctx->parent_root, path);
6091 btrfs_item_key_to_cpu(eb, &key, slot);
6092 if (key.objectid != sctx->cur_ino)
6094 if (key.type != BTRFS_INODE_REF_KEY &&
6095 key.type != BTRFS_INODE_EXTREF_KEY)
6098 ret = iterate_inode_ref(sctx->parent_root, path, &key, 1,
6099 record_parent_ref, &ctx);
6106 while (!list_empty(&deleted_refs)) {
6107 struct recorded_ref *ref;
6109 ref = list_first_entry(&deleted_refs, struct recorded_ref, list);
6110 ret = send_unlink(sctx, ref->full_path);
6113 fs_path_free(ref->full_path);
6114 list_del(&ref->list);
6119 btrfs_free_path(path);
6121 __free_recorded_refs(&deleted_refs);
6125 static int changed_inode(struct send_ctx *sctx,
6126 enum btrfs_compare_tree_result result)
6129 struct btrfs_key *key = sctx->cmp_key;
6130 struct btrfs_inode_item *left_ii = NULL;
6131 struct btrfs_inode_item *right_ii = NULL;
6135 sctx->cur_ino = key->objectid;
6136 sctx->cur_inode_new_gen = 0;
6137 sctx->cur_inode_last_extent = (u64)-1;
6138 sctx->cur_inode_next_write_offset = 0;
6139 sctx->ignore_cur_inode = false;
6142 * Set send_progress to current inode. This will tell all get_cur_xxx
6143 * functions that the current inode's refs are not updated yet. Later,
6144 * when process_recorded_refs is finished, it is set to cur_ino + 1.
6146 sctx->send_progress = sctx->cur_ino;
6148 if (result == BTRFS_COMPARE_TREE_NEW ||
6149 result == BTRFS_COMPARE_TREE_CHANGED) {
6150 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6151 sctx->left_path->slots[0],
6152 struct btrfs_inode_item);
6153 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6156 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6157 sctx->right_path->slots[0],
6158 struct btrfs_inode_item);
6159 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6162 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6163 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6164 sctx->right_path->slots[0],
6165 struct btrfs_inode_item);
6167 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6171 * The cur_ino = root dir case is special here. We can't treat
6172 * the inode as deleted+reused because it would generate a
6173 * stream that tries to delete/mkdir the root dir.
6175 if (left_gen != right_gen &&
6176 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6177 sctx->cur_inode_new_gen = 1;
6181 * Normally we do not find inodes with a link count of zero (orphans)
6182 * because the most common case is to create a snapshot and use it
6183 * for a send operation. However other less common use cases involve
6184 * using a subvolume and send it after turning it to RO mode just
6185 * after deleting all hard links of a file while holding an open
6186 * file descriptor against it or turning a RO snapshot into RW mode,
6187 * keep an open file descriptor against a file, delete it and then
6188 * turn the snapshot back to RO mode before using it for a send
6189 * operation. So if we find such cases, ignore the inode and all its
6190 * items completely if it's a new inode, or if it's a changed inode
6191 * make sure all its previous paths (from the parent snapshot) are all
6192 * unlinked and all other the inode items are ignored.
6194 if (result == BTRFS_COMPARE_TREE_NEW ||
6195 result == BTRFS_COMPARE_TREE_CHANGED) {
6198 nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6200 sctx->ignore_cur_inode = true;
6201 if (result == BTRFS_COMPARE_TREE_CHANGED)
6202 ret = btrfs_unlink_all_paths(sctx);
6207 if (result == BTRFS_COMPARE_TREE_NEW) {
6208 sctx->cur_inode_gen = left_gen;
6209 sctx->cur_inode_new = 1;
6210 sctx->cur_inode_deleted = 0;
6211 sctx->cur_inode_size = btrfs_inode_size(
6212 sctx->left_path->nodes[0], left_ii);
6213 sctx->cur_inode_mode = btrfs_inode_mode(
6214 sctx->left_path->nodes[0], left_ii);
6215 sctx->cur_inode_rdev = btrfs_inode_rdev(
6216 sctx->left_path->nodes[0], left_ii);
6217 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6218 ret = send_create_inode_if_needed(sctx);
6219 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6220 sctx->cur_inode_gen = right_gen;
6221 sctx->cur_inode_new = 0;
6222 sctx->cur_inode_deleted = 1;
6223 sctx->cur_inode_size = btrfs_inode_size(
6224 sctx->right_path->nodes[0], right_ii);
6225 sctx->cur_inode_mode = btrfs_inode_mode(
6226 sctx->right_path->nodes[0], right_ii);
6227 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
6229 * We need to do some special handling in case the inode was
6230 * reported as changed with a changed generation number. This
6231 * means that the original inode was deleted and new inode
6232 * reused the same inum. So we have to treat the old inode as
6233 * deleted and the new one as new.
6235 if (sctx->cur_inode_new_gen) {
6237 * First, process the inode as if it was deleted.
6239 sctx->cur_inode_gen = right_gen;
6240 sctx->cur_inode_new = 0;
6241 sctx->cur_inode_deleted = 1;
6242 sctx->cur_inode_size = btrfs_inode_size(
6243 sctx->right_path->nodes[0], right_ii);
6244 sctx->cur_inode_mode = btrfs_inode_mode(
6245 sctx->right_path->nodes[0], right_ii);
6246 ret = process_all_refs(sctx,
6247 BTRFS_COMPARE_TREE_DELETED);
6252 * Now process the inode as if it was new.
6254 sctx->cur_inode_gen = left_gen;
6255 sctx->cur_inode_new = 1;
6256 sctx->cur_inode_deleted = 0;
6257 sctx->cur_inode_size = btrfs_inode_size(
6258 sctx->left_path->nodes[0], left_ii);
6259 sctx->cur_inode_mode = btrfs_inode_mode(
6260 sctx->left_path->nodes[0], left_ii);
6261 sctx->cur_inode_rdev = btrfs_inode_rdev(
6262 sctx->left_path->nodes[0], left_ii);
6263 ret = send_create_inode_if_needed(sctx);
6267 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6271 * Advance send_progress now as we did not get into
6272 * process_recorded_refs_if_needed in the new_gen case.
6274 sctx->send_progress = sctx->cur_ino + 1;
6277 * Now process all extents and xattrs of the inode as if
6278 * they were all new.
6280 ret = process_all_extents(sctx);
6283 ret = process_all_new_xattrs(sctx);
6287 sctx->cur_inode_gen = left_gen;
6288 sctx->cur_inode_new = 0;
6289 sctx->cur_inode_new_gen = 0;
6290 sctx->cur_inode_deleted = 0;
6291 sctx->cur_inode_size = btrfs_inode_size(
6292 sctx->left_path->nodes[0], left_ii);
6293 sctx->cur_inode_mode = btrfs_inode_mode(
6294 sctx->left_path->nodes[0], left_ii);
6303 * We have to process new refs before deleted refs, but compare_trees gives us
6304 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6305 * first and later process them in process_recorded_refs.
6306 * For the cur_inode_new_gen case, we skip recording completely because
6307 * changed_inode did already initiate processing of refs. The reason for this is
6308 * that in this case, compare_tree actually compares the refs of 2 different
6309 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6310 * refs of the right tree as deleted and all refs of the left tree as new.
6312 static int changed_ref(struct send_ctx *sctx,
6313 enum btrfs_compare_tree_result result)
6317 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6318 inconsistent_snapshot_error(sctx, result, "reference");
6322 if (!sctx->cur_inode_new_gen &&
6323 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6324 if (result == BTRFS_COMPARE_TREE_NEW)
6325 ret = record_new_ref(sctx);
6326 else if (result == BTRFS_COMPARE_TREE_DELETED)
6327 ret = record_deleted_ref(sctx);
6328 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6329 ret = record_changed_ref(sctx);
6336 * Process new/deleted/changed xattrs. We skip processing in the
6337 * cur_inode_new_gen case because changed_inode did already initiate processing
6338 * of xattrs. The reason is the same as in changed_ref
6340 static int changed_xattr(struct send_ctx *sctx,
6341 enum btrfs_compare_tree_result result)
6345 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6346 inconsistent_snapshot_error(sctx, result, "xattr");
6350 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6351 if (result == BTRFS_COMPARE_TREE_NEW)
6352 ret = process_new_xattr(sctx);
6353 else if (result == BTRFS_COMPARE_TREE_DELETED)
6354 ret = process_deleted_xattr(sctx);
6355 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6356 ret = process_changed_xattr(sctx);
6363 * Process new/deleted/changed extents. We skip processing in the
6364 * cur_inode_new_gen case because changed_inode did already initiate processing
6365 * of extents. The reason is the same as in changed_ref
6367 static int changed_extent(struct send_ctx *sctx,
6368 enum btrfs_compare_tree_result result)
6373 * We have found an extent item that changed without the inode item
6374 * having changed. This can happen either after relocation (where the
6375 * disk_bytenr of an extent item is replaced at
6376 * relocation.c:replace_file_extents()) or after deduplication into a
6377 * file in both the parent and send snapshots (where an extent item can
6378 * get modified or replaced with a new one). Note that deduplication
6379 * updates the inode item, but it only changes the iversion (sequence
6380 * field in the inode item) of the inode, so if a file is deduplicated
6381 * the same amount of times in both the parent and send snapshots, its
6382 * iversion becames the same in both snapshots, whence the inode item is
6383 * the same on both snapshots.
6385 if (sctx->cur_ino != sctx->cmp_key->objectid)
6388 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6389 if (result != BTRFS_COMPARE_TREE_DELETED)
6390 ret = process_extent(sctx, sctx->left_path,
6397 static int dir_changed(struct send_ctx *sctx, u64 dir)
6399 u64 orig_gen, new_gen;
6402 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6407 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6412 return (orig_gen != new_gen) ? 1 : 0;
6415 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6416 struct btrfs_key *key)
6418 struct btrfs_inode_extref *extref;
6419 struct extent_buffer *leaf;
6420 u64 dirid = 0, last_dirid = 0;
6427 /* Easy case, just check this one dirid */
6428 if (key->type == BTRFS_INODE_REF_KEY) {
6429 dirid = key->offset;
6431 ret = dir_changed(sctx, dirid);
6435 leaf = path->nodes[0];
6436 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6437 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6438 while (cur_offset < item_size) {
6439 extref = (struct btrfs_inode_extref *)(ptr +
6441 dirid = btrfs_inode_extref_parent(leaf, extref);
6442 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6443 cur_offset += ref_name_len + sizeof(*extref);
6444 if (dirid == last_dirid)
6446 ret = dir_changed(sctx, dirid);
6456 * Updates compare related fields in sctx and simply forwards to the actual
6457 * changed_xxx functions.
6459 static int changed_cb(struct btrfs_path *left_path,
6460 struct btrfs_path *right_path,
6461 struct btrfs_key *key,
6462 enum btrfs_compare_tree_result result,
6466 struct send_ctx *sctx = ctx;
6468 if (result == BTRFS_COMPARE_TREE_SAME) {
6469 if (key->type == BTRFS_INODE_REF_KEY ||
6470 key->type == BTRFS_INODE_EXTREF_KEY) {
6471 ret = compare_refs(sctx, left_path, key);
6476 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6477 return maybe_send_hole(sctx, left_path, key);
6481 result = BTRFS_COMPARE_TREE_CHANGED;
6485 sctx->left_path = left_path;
6486 sctx->right_path = right_path;
6487 sctx->cmp_key = key;
6489 ret = finish_inode_if_needed(sctx, 0);
6493 /* Ignore non-FS objects */
6494 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6495 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6498 if (key->type == BTRFS_INODE_ITEM_KEY) {
6499 ret = changed_inode(sctx, result);
6500 } else if (!sctx->ignore_cur_inode) {
6501 if (key->type == BTRFS_INODE_REF_KEY ||
6502 key->type == BTRFS_INODE_EXTREF_KEY)
6503 ret = changed_ref(sctx, result);
6504 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6505 ret = changed_xattr(sctx, result);
6506 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6507 ret = changed_extent(sctx, result);
6514 static int full_send_tree(struct send_ctx *sctx)
6517 struct btrfs_root *send_root = sctx->send_root;
6518 struct btrfs_key key;
6519 struct btrfs_path *path;
6520 struct extent_buffer *eb;
6523 path = alloc_path_for_send();
6527 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6528 key.type = BTRFS_INODE_ITEM_KEY;
6531 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6538 eb = path->nodes[0];
6539 slot = path->slots[0];
6540 btrfs_item_key_to_cpu(eb, &key, slot);
6542 ret = changed_cb(path, NULL, &key,
6543 BTRFS_COMPARE_TREE_NEW, sctx);
6547 ret = btrfs_next_item(send_root, path);
6557 ret = finish_inode_if_needed(sctx, 1);
6560 btrfs_free_path(path);
6564 static int tree_move_down(struct btrfs_path *path, int *level)
6566 struct extent_buffer *eb;
6568 BUG_ON(*level == 0);
6569 eb = btrfs_read_node_slot(path->nodes[*level], path->slots[*level]);
6573 path->nodes[*level - 1] = eb;
6574 path->slots[*level - 1] = 0;
6579 static int tree_move_next_or_upnext(struct btrfs_path *path,
6580 int *level, int root_level)
6584 nritems = btrfs_header_nritems(path->nodes[*level]);
6586 path->slots[*level]++;
6588 while (path->slots[*level] >= nritems) {
6589 if (*level == root_level)
6593 path->slots[*level] = 0;
6594 free_extent_buffer(path->nodes[*level]);
6595 path->nodes[*level] = NULL;
6597 path->slots[*level]++;
6599 nritems = btrfs_header_nritems(path->nodes[*level]);
6606 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
6609 static int tree_advance(struct btrfs_path *path,
6610 int *level, int root_level,
6612 struct btrfs_key *key)
6616 if (*level == 0 || !allow_down) {
6617 ret = tree_move_next_or_upnext(path, level, root_level);
6619 ret = tree_move_down(path, level);
6623 btrfs_item_key_to_cpu(path->nodes[*level], key,
6624 path->slots[*level]);
6626 btrfs_node_key_to_cpu(path->nodes[*level], key,
6627 path->slots[*level]);
6632 static int tree_compare_item(struct btrfs_path *left_path,
6633 struct btrfs_path *right_path,
6638 unsigned long off1, off2;
6640 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
6641 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
6645 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
6646 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
6647 right_path->slots[0]);
6649 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
6651 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
6658 * This function compares two trees and calls the provided callback for
6659 * every changed/new/deleted item it finds.
6660 * If shared tree blocks are encountered, whole subtrees are skipped, making
6661 * the compare pretty fast on snapshotted subvolumes.
6663 * This currently works on commit roots only. As commit roots are read only,
6664 * we don't do any locking. The commit roots are protected with transactions.
6665 * Transactions are ended and rejoined when a commit is tried in between.
6667 * This function checks for modifications done to the trees while comparing.
6668 * If it detects a change, it aborts immediately.
6670 static int btrfs_compare_trees(struct btrfs_root *left_root,
6671 struct btrfs_root *right_root,
6672 btrfs_changed_cb_t changed_cb, void *ctx)
6674 struct btrfs_fs_info *fs_info = left_root->fs_info;
6677 struct btrfs_path *left_path = NULL;
6678 struct btrfs_path *right_path = NULL;
6679 struct btrfs_key left_key;
6680 struct btrfs_key right_key;
6681 char *tmp_buf = NULL;
6682 int left_root_level;
6683 int right_root_level;
6686 int left_end_reached;
6687 int right_end_reached;
6695 left_path = btrfs_alloc_path();
6700 right_path = btrfs_alloc_path();
6706 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
6712 left_path->search_commit_root = 1;
6713 left_path->skip_locking = 1;
6714 right_path->search_commit_root = 1;
6715 right_path->skip_locking = 1;
6718 * Strategy: Go to the first items of both trees. Then do
6720 * If both trees are at level 0
6721 * Compare keys of current items
6722 * If left < right treat left item as new, advance left tree
6724 * If left > right treat right item as deleted, advance right tree
6726 * If left == right do deep compare of items, treat as changed if
6727 * needed, advance both trees and repeat
6728 * If both trees are at the same level but not at level 0
6729 * Compare keys of current nodes/leafs
6730 * If left < right advance left tree and repeat
6731 * If left > right advance right tree and repeat
6732 * If left == right compare blockptrs of the next nodes/leafs
6733 * If they match advance both trees but stay at the same level
6735 * If they don't match advance both trees while allowing to go
6737 * If tree levels are different
6738 * Advance the tree that needs it and repeat
6740 * Advancing a tree means:
6741 * If we are at level 0, try to go to the next slot. If that's not
6742 * possible, go one level up and repeat. Stop when we found a level
6743 * where we could go to the next slot. We may at this point be on a
6746 * If we are not at level 0 and not on shared tree blocks, go one
6749 * If we are not at level 0 and on shared tree blocks, go one slot to
6750 * the right if possible or go up and right.
6753 down_read(&fs_info->commit_root_sem);
6754 left_level = btrfs_header_level(left_root->commit_root);
6755 left_root_level = left_level;
6756 left_path->nodes[left_level] =
6757 btrfs_clone_extent_buffer(left_root->commit_root);
6758 if (!left_path->nodes[left_level]) {
6759 up_read(&fs_info->commit_root_sem);
6764 right_level = btrfs_header_level(right_root->commit_root);
6765 right_root_level = right_level;
6766 right_path->nodes[right_level] =
6767 btrfs_clone_extent_buffer(right_root->commit_root);
6768 if (!right_path->nodes[right_level]) {
6769 up_read(&fs_info->commit_root_sem);
6773 up_read(&fs_info->commit_root_sem);
6775 if (left_level == 0)
6776 btrfs_item_key_to_cpu(left_path->nodes[left_level],
6777 &left_key, left_path->slots[left_level]);
6779 btrfs_node_key_to_cpu(left_path->nodes[left_level],
6780 &left_key, left_path->slots[left_level]);
6781 if (right_level == 0)
6782 btrfs_item_key_to_cpu(right_path->nodes[right_level],
6783 &right_key, right_path->slots[right_level]);
6785 btrfs_node_key_to_cpu(right_path->nodes[right_level],
6786 &right_key, right_path->slots[right_level]);
6788 left_end_reached = right_end_reached = 0;
6789 advance_left = advance_right = 0;
6793 if (advance_left && !left_end_reached) {
6794 ret = tree_advance(left_path, &left_level,
6796 advance_left != ADVANCE_ONLY_NEXT,
6799 left_end_reached = ADVANCE;
6804 if (advance_right && !right_end_reached) {
6805 ret = tree_advance(right_path, &right_level,
6807 advance_right != ADVANCE_ONLY_NEXT,
6810 right_end_reached = ADVANCE;
6816 if (left_end_reached && right_end_reached) {
6819 } else if (left_end_reached) {
6820 if (right_level == 0) {
6821 ret = changed_cb(left_path, right_path,
6823 BTRFS_COMPARE_TREE_DELETED,
6828 advance_right = ADVANCE;
6830 } else if (right_end_reached) {
6831 if (left_level == 0) {
6832 ret = changed_cb(left_path, right_path,
6834 BTRFS_COMPARE_TREE_NEW,
6839 advance_left = ADVANCE;
6843 if (left_level == 0 && right_level == 0) {
6844 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6846 ret = changed_cb(left_path, right_path,
6848 BTRFS_COMPARE_TREE_NEW,
6852 advance_left = ADVANCE;
6853 } else if (cmp > 0) {
6854 ret = changed_cb(left_path, right_path,
6856 BTRFS_COMPARE_TREE_DELETED,
6860 advance_right = ADVANCE;
6862 enum btrfs_compare_tree_result result;
6864 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
6865 ret = tree_compare_item(left_path, right_path,
6868 result = BTRFS_COMPARE_TREE_CHANGED;
6870 result = BTRFS_COMPARE_TREE_SAME;
6871 ret = changed_cb(left_path, right_path,
6872 &left_key, result, ctx);
6875 advance_left = ADVANCE;
6876 advance_right = ADVANCE;
6878 } else if (left_level == right_level) {
6879 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6881 advance_left = ADVANCE;
6882 } else if (cmp > 0) {
6883 advance_right = ADVANCE;
6885 left_blockptr = btrfs_node_blockptr(
6886 left_path->nodes[left_level],
6887 left_path->slots[left_level]);
6888 right_blockptr = btrfs_node_blockptr(
6889 right_path->nodes[right_level],
6890 right_path->slots[right_level]);
6891 left_gen = btrfs_node_ptr_generation(
6892 left_path->nodes[left_level],
6893 left_path->slots[left_level]);
6894 right_gen = btrfs_node_ptr_generation(
6895 right_path->nodes[right_level],
6896 right_path->slots[right_level]);
6897 if (left_blockptr == right_blockptr &&
6898 left_gen == right_gen) {
6900 * As we're on a shared block, don't
6901 * allow to go deeper.
6903 advance_left = ADVANCE_ONLY_NEXT;
6904 advance_right = ADVANCE_ONLY_NEXT;
6906 advance_left = ADVANCE;
6907 advance_right = ADVANCE;
6910 } else if (left_level < right_level) {
6911 advance_right = ADVANCE;
6913 advance_left = ADVANCE;
6918 btrfs_free_path(left_path);
6919 btrfs_free_path(right_path);
6924 static int send_subvol(struct send_ctx *sctx)
6928 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6929 ret = send_header(sctx);
6934 ret = send_subvol_begin(sctx);
6938 if (sctx->parent_root) {
6939 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6943 ret = finish_inode_if_needed(sctx, 1);
6947 ret = full_send_tree(sctx);
6953 free_recorded_refs(sctx);
6958 * If orphan cleanup did remove any orphans from a root, it means the tree
6959 * was modified and therefore the commit root is not the same as the current
6960 * root anymore. This is a problem, because send uses the commit root and
6961 * therefore can see inode items that don't exist in the current root anymore,
6962 * and for example make calls to btrfs_iget, which will do tree lookups based
6963 * on the current root and not on the commit root. Those lookups will fail,
6964 * returning a -ESTALE error, and making send fail with that error. So make
6965 * sure a send does not see any orphans we have just removed, and that it will
6966 * see the same inodes regardless of whether a transaction commit happened
6967 * before it started (meaning that the commit root will be the same as the
6968 * current root) or not.
6970 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6973 struct btrfs_trans_handle *trans = NULL;
6976 if (sctx->parent_root &&
6977 sctx->parent_root->node != sctx->parent_root->commit_root)
6980 for (i = 0; i < sctx->clone_roots_cnt; i++)
6981 if (sctx->clone_roots[i].root->node !=
6982 sctx->clone_roots[i].root->commit_root)
6986 return btrfs_end_transaction(trans);
6991 /* Use any root, all fs roots will get their commit roots updated. */
6993 trans = btrfs_join_transaction(sctx->send_root);
6995 return PTR_ERR(trans);
6999 return btrfs_commit_transaction(trans);
7003 * Make sure any existing dellaloc is flushed for any root used by a send
7004 * operation so that we do not miss any data and we do not race with writeback
7005 * finishing and changing a tree while send is using the tree. This could
7006 * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and
7007 * a send operation then uses the subvolume.
7008 * After flushing delalloc ensure_commit_roots_uptodate() must be called.
7010 static int flush_delalloc_roots(struct send_ctx *sctx)
7012 struct btrfs_root *root = sctx->parent_root;
7017 ret = btrfs_start_delalloc_snapshot(root);
7020 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7023 for (i = 0; i < sctx->clone_roots_cnt; i++) {
7024 root = sctx->clone_roots[i].root;
7025 ret = btrfs_start_delalloc_snapshot(root);
7028 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7034 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
7036 spin_lock(&root->root_item_lock);
7037 root->send_in_progress--;
7039 * Not much left to do, we don't know why it's unbalanced and
7040 * can't blindly reset it to 0.
7042 if (root->send_in_progress < 0)
7043 btrfs_err(root->fs_info,
7044 "send_in_progress unbalanced %d root %llu",
7045 root->send_in_progress, root->root_key.objectid);
7046 spin_unlock(&root->root_item_lock);
7049 static void dedupe_in_progress_warn(const struct btrfs_root *root)
7051 btrfs_warn_rl(root->fs_info,
7052 "cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
7053 root->root_key.objectid, root->dedupe_in_progress);
7056 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
7059 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
7060 struct btrfs_fs_info *fs_info = send_root->fs_info;
7061 struct btrfs_root *clone_root;
7062 struct btrfs_key key;
7063 struct send_ctx *sctx = NULL;
7065 u64 *clone_sources_tmp = NULL;
7066 int clone_sources_to_rollback = 0;
7067 unsigned alloc_size;
7068 int sort_clone_roots = 0;
7071 if (!capable(CAP_SYS_ADMIN))
7075 * The subvolume must remain read-only during send, protect against
7076 * making it RW. This also protects against deletion.
7078 spin_lock(&send_root->root_item_lock);
7079 if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
7080 dedupe_in_progress_warn(send_root);
7081 spin_unlock(&send_root->root_item_lock);
7084 send_root->send_in_progress++;
7085 spin_unlock(&send_root->root_item_lock);
7088 * Userspace tools do the checks and warn the user if it's
7091 if (!btrfs_root_readonly(send_root)) {
7097 * Check that we don't overflow at later allocations, we request
7098 * clone_sources_count + 1 items, and compare to unsigned long inside
7101 if (arg->clone_sources_count >
7102 ULONG_MAX / sizeof(struct clone_root) - 1) {
7107 if (!access_ok(arg->clone_sources,
7108 sizeof(*arg->clone_sources) *
7109 arg->clone_sources_count)) {
7114 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
7119 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
7125 INIT_LIST_HEAD(&sctx->new_refs);
7126 INIT_LIST_HEAD(&sctx->deleted_refs);
7127 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
7128 INIT_LIST_HEAD(&sctx->name_cache_list);
7130 sctx->flags = arg->flags;
7132 sctx->send_filp = fget(arg->send_fd);
7133 if (!sctx->send_filp) {
7138 sctx->send_root = send_root;
7140 * Unlikely but possible, if the subvolume is marked for deletion but
7141 * is slow to remove the directory entry, send can still be started
7143 if (btrfs_root_dead(sctx->send_root)) {
7148 sctx->clone_roots_cnt = arg->clone_sources_count;
7150 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
7151 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
7152 if (!sctx->send_buf) {
7157 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
7158 if (!sctx->read_buf) {
7163 sctx->pending_dir_moves = RB_ROOT;
7164 sctx->waiting_dir_moves = RB_ROOT;
7165 sctx->orphan_dirs = RB_ROOT;
7167 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
7169 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
7170 if (!sctx->clone_roots) {
7175 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
7177 if (arg->clone_sources_count) {
7178 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
7179 if (!clone_sources_tmp) {
7184 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
7191 for (i = 0; i < arg->clone_sources_count; i++) {
7192 key.objectid = clone_sources_tmp[i];
7193 key.type = BTRFS_ROOT_ITEM_KEY;
7194 key.offset = (u64)-1;
7196 index = srcu_read_lock(&fs_info->subvol_srcu);
7198 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
7199 if (IS_ERR(clone_root)) {
7200 srcu_read_unlock(&fs_info->subvol_srcu, index);
7201 ret = PTR_ERR(clone_root);
7204 spin_lock(&clone_root->root_item_lock);
7205 if (!btrfs_root_readonly(clone_root) ||
7206 btrfs_root_dead(clone_root)) {
7207 spin_unlock(&clone_root->root_item_lock);
7208 srcu_read_unlock(&fs_info->subvol_srcu, index);
7212 if (clone_root->dedupe_in_progress) {
7213 dedupe_in_progress_warn(clone_root);
7214 spin_unlock(&clone_root->root_item_lock);
7215 srcu_read_unlock(&fs_info->subvol_srcu, index);
7219 clone_root->send_in_progress++;
7220 spin_unlock(&clone_root->root_item_lock);
7221 srcu_read_unlock(&fs_info->subvol_srcu, index);
7223 sctx->clone_roots[i].root = clone_root;
7224 clone_sources_to_rollback = i + 1;
7226 kvfree(clone_sources_tmp);
7227 clone_sources_tmp = NULL;
7230 if (arg->parent_root) {
7231 key.objectid = arg->parent_root;
7232 key.type = BTRFS_ROOT_ITEM_KEY;
7233 key.offset = (u64)-1;
7235 index = srcu_read_lock(&fs_info->subvol_srcu);
7237 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
7238 if (IS_ERR(sctx->parent_root)) {
7239 srcu_read_unlock(&fs_info->subvol_srcu, index);
7240 ret = PTR_ERR(sctx->parent_root);
7244 spin_lock(&sctx->parent_root->root_item_lock);
7245 sctx->parent_root->send_in_progress++;
7246 if (!btrfs_root_readonly(sctx->parent_root) ||
7247 btrfs_root_dead(sctx->parent_root)) {
7248 spin_unlock(&sctx->parent_root->root_item_lock);
7249 srcu_read_unlock(&fs_info->subvol_srcu, index);
7253 if (sctx->parent_root->dedupe_in_progress) {
7254 dedupe_in_progress_warn(sctx->parent_root);
7255 spin_unlock(&sctx->parent_root->root_item_lock);
7256 srcu_read_unlock(&fs_info->subvol_srcu, index);
7260 spin_unlock(&sctx->parent_root->root_item_lock);
7262 srcu_read_unlock(&fs_info->subvol_srcu, index);
7266 * Clones from send_root are allowed, but only if the clone source
7267 * is behind the current send position. This is checked while searching
7268 * for possible clone sources.
7270 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
7272 /* We do a bsearch later */
7273 sort(sctx->clone_roots, sctx->clone_roots_cnt,
7274 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
7276 sort_clone_roots = 1;
7278 ret = flush_delalloc_roots(sctx);
7282 ret = ensure_commit_roots_uptodate(sctx);
7286 mutex_lock(&fs_info->balance_mutex);
7287 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
7288 mutex_unlock(&fs_info->balance_mutex);
7289 btrfs_warn_rl(fs_info,
7290 "cannot run send because a balance operation is in progress");
7294 fs_info->send_in_progress++;
7295 mutex_unlock(&fs_info->balance_mutex);
7297 current->journal_info = BTRFS_SEND_TRANS_STUB;
7298 ret = send_subvol(sctx);
7299 current->journal_info = NULL;
7300 mutex_lock(&fs_info->balance_mutex);
7301 fs_info->send_in_progress--;
7302 mutex_unlock(&fs_info->balance_mutex);
7306 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
7307 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
7310 ret = send_cmd(sctx);
7316 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
7317 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
7319 struct pending_dir_move *pm;
7321 n = rb_first(&sctx->pending_dir_moves);
7322 pm = rb_entry(n, struct pending_dir_move, node);
7323 while (!list_empty(&pm->list)) {
7324 struct pending_dir_move *pm2;
7326 pm2 = list_first_entry(&pm->list,
7327 struct pending_dir_move, list);
7328 free_pending_move(sctx, pm2);
7330 free_pending_move(sctx, pm);
7333 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
7334 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
7336 struct waiting_dir_move *dm;
7338 n = rb_first(&sctx->waiting_dir_moves);
7339 dm = rb_entry(n, struct waiting_dir_move, node);
7340 rb_erase(&dm->node, &sctx->waiting_dir_moves);
7344 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
7345 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
7347 struct orphan_dir_info *odi;
7349 n = rb_first(&sctx->orphan_dirs);
7350 odi = rb_entry(n, struct orphan_dir_info, node);
7351 free_orphan_dir_info(sctx, odi);
7354 if (sort_clone_roots) {
7355 for (i = 0; i < sctx->clone_roots_cnt; i++)
7356 btrfs_root_dec_send_in_progress(
7357 sctx->clone_roots[i].root);
7359 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
7360 btrfs_root_dec_send_in_progress(
7361 sctx->clone_roots[i].root);
7363 btrfs_root_dec_send_in_progress(send_root);
7365 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
7366 btrfs_root_dec_send_in_progress(sctx->parent_root);
7368 kvfree(clone_sources_tmp);
7371 if (sctx->send_filp)
7372 fput(sctx->send_filp);
7374 kvfree(sctx->clone_roots);
7375 kvfree(sctx->send_buf);
7376 kvfree(sctx->read_buf);
7378 name_cache_free(sctx);